Datasets:

timaeus
/

pile-arxiv

Dataset card Data Studio Files Files and versions Community

Split (1)

train · 100k rows

text stringlengths 4 2.78M	meta dict
--- abstract: 'We present a study of the hydrodynamics of an active particle—a model squirmer—in an environment with a broken rotational symmetry: a nematic liquid crystal. By combining simulations with analytic calculations, we show that the hydrodynamic coupling between the squirmer flow field and liquid crystalline director can lead to re-orientation of the swimmers. The preferred orientation depends on the exact details of the squirmer flow field. In a steady state, pushers are shown to swim parallel with the nematic director while pullers swim perpendicular to the nematic director. This behaviour arises solely from hydrodynamic coupling between the squirmer flow field and anisotropic viscosities of the host fluid. Our results suggest that an anisotropic swimming medium can be used to characterise and guide spherical microswimmers in the bulk.' author: - 'J. S. Lintuvuori$^1$, A. W[ü]{}rger$^1$ and K. Stratford$^2$' title: Hydrodynamics defines the stable swimming direction of spherical squirmers in a nematic liquid crystal --- Active materials use internal energy resources to propel themselves and have recently emerged as a topical research area within physics [@Ramaswamy2010; @Marchetti2013]. A natural example of an active systems is provided by swimming bacteria, while artificial microswimmers can be realised by self-propelling Janus particles [@brown14; @brown15; @ebbens12b; @sabass10; @ebbens12; @ebbens14; @wang15; @das15]. One big challenge is to control and direct the swimmers at the microscale. Success here could allow one to harness swimmers to do work, and it could lead to significant technological possibilities, for example, direct microengineering of new materials. Various possibilities have been explored in order to guide active particles. The most obvious one is to use confining walls, as both bacteria [@Lord63; @berke08] and artificial swimmers [@ishimoto13; @zottl14; @li14; @brown15; @das15; @lintuvuori16] are known to be attracted to surfaces, and swim near them. Motion along predefined pathways can be obtained by topographical patterns [@Sim16] or chemical functionalisation [@Usp16] of the surface. Force-free localization and steering of laser-powered Janus particles have been achieved by dynamical feedback [@Bre14] or by spatial modulation of the laser beam [@Loz16], which exerts a torque on the moving particle [@Bic14]. An alternative route to control the swimmers in the bulk is to use an anisotropic swimming media [@Arratia16], [e.g.]{} a liquid crystal [@Lavrento16]. Recent experiments of colloidal particles have demonstrated electrophoretic propulsion of spherical colloids in nematic LC [@Lavrento16; @Hernandez13]. Whereas rod-like bacteria are observed to swim along the direction set by the nematic director $\hat{\mathbf{n}}$ [@Smalyukh08; @Kumar13; @Aranson14]. Experimental applications include the self-assembly of bacteria dispersed in a nematic LC [@Abbott14], transport of colloidal cargo [@Weibel15] and accumulation of the bacteria to topological patterns [@lavrentovich1; @lavrentovich2]. Theoretical predictions include anomalous diffusion [@rik] and even backward swimming was predicted by theoretical calculations of Taylor-sheets in nematic LC [@powers1; @powers2]. In the case of rod-like swimmers ([e.g.]{} typical bacteria) the alignment is dominated by an elastic energy, which is minimised when the rods align their long-axis along $\hat{\mathbf{n}}$ [@Smalyukh08], thus rod-like swimmers are always expected to swim following the nematic director. For isotropic swimmers ([e.g.]{} spherical bacteria or artificial Janus swimmers) this is not the case: in the limit of spherical particles the elastic torque vanishes. In this letter, we study the dynamics of fully resolved [spherical]{} microswimmers in a nematic liquid crystal, by means of lattice Boltzmann simulations and analytical calculations, using a squirmer model [@lighthill52]. Our simulations show that the steady state swimming direction depends of the nature of the swimming mechanism. Spherical [pushers]{} undergo stable swimming following the direction set by the nematic director. Strikingly, a [puller]{} swims in steady state in a direction orthogonal to the far-field $\hat{\mathbf{n}}$. Using analytical calculations we show that the reorientation is due to a hydrodynamic torque, arising from the coupling between the squirmer flow field and anisotropy of the liquid crystalline viscosities [@miesowicz]. Further we show the reorientation rate scales linearly with the power of the squirmer flow field. Our results provide a robust and easy way to manipulate self-propelling organisms directly at the microscale, allowing for example sorting of the swimmers based on their hydrodynamic nature. [Squirmer model:]{} To simulate the dynamics of an active particle in a liquid crystal we employ a lattice Boltzmann (LB) method [@mikeLB]. We treat the self-propelling particle in the terms of a squirmer model [@lighthill52]. The tangential (slip) velocity profile at the particle surface leading to the squirmer motions is given by [@Magar03] $$%u(\theta)=2\sum_{n=1}^{\infty}\frac{\sin \theta}{n(n+1)}\frac{dP_n(\cos \theta)}{d\cos \theta}B_n u(\theta)= v_0 \sin (\theta)( 1+ \beta \cos \theta) \label{eq:veltangential}$$ where $v_0$ is a constant, $\beta$ the squirmer parameter, and $\theta$ the polar angle with respect to the particle axis [@ishimoto13]. In the LB method a no-slip boundary condition at the fluid/solid interface can be achieved by using a standard method of bounce-back on links (BBL) [@ladd1; @ladd2]. When the boundary is moving ([e.g.]{} a colloidal particle) the BBL condition needs to be modified to take into account particle motion [@ladd3]. These local rules can include additional terms, such as a surface slip velocity (Eq. \[eq:veltangential\]) leading to LB simulations of squirming motion [@ignacio1; @ignacio2]. [Liquid crystal model:]{} The nematic host fluid is described by a Landau – de Gennes free-energy whose density can be expressed in terms of a symmetric and traceless order parameter tensor $\mathbf{Q}$ as ${\cal F} = F(Q_{\alpha\beta}) + \tfrac{K}{2}(\partial_{\beta}Q_{\alpha \beta})^2$, with $$\label{eq:fed_bulk} F(Q_{\alpha\beta}) = A_0\left(1-\frac{\gamma}{3}\right)\frac{Q_{\alpha \beta}^2}{2}-\frac{\gamma}{3}Q_{\alpha \beta}Q_{\beta \gamma}Q_{\gamma \alpha} + \frac{\gamma}{4}(Q_{\alpha \beta}^2)^2$$ where Greek indices denote Cartesian coordinates and summation over repeated indices is implied. $A_0$ is a free energy scale, $\gamma$ is a temperature-like control parameter giving a order/disorder transition at $\gamma\sim 2.7$, and $K$ is an elastic constant. The anchoring at the particle surface is modeled by $f_s=W(Q_{\alpha\beta}-Q^{0}_{\alpha\beta})^2$, where $W$ is the anchoring strength and $Q^{0}_{\alpha\beta}$ is the preferred alignment of the nematic director at the particle surface. The hydrodynamic equation for the evolution of $\mathbf{Q}$ is [@berisedwards]: $(\partial_t + u_{\nu}\partial_{\nu})Q_{\alpha \beta} - S_{\alpha\beta}= \Gamma H_{\alpha \beta}$, where the first part describes the advection and $S_{\alpha\beta}$ describes the possible rotation/stretching of $\mathbf{Q}$ by the flow [@berisedwards]. $\Gamma$ is the rotational diffusion constant and the molecular field is $$H_{\alpha \beta}= -{\delta {\cal F} / \delta Q_{\alpha \beta}} + (\delta_{\alpha \beta}/3) {\mbox {\rm Tr}}({\delta {\cal F} / \delta Q_{\alpha \beta}}).$$ The fluid velocity obeys $\partial_\alpha u_\alpha = 0$, and the Navier-Stokes equation, which is coupled to the LC via a stress tensor. We employ a 3D lattice Boltzmann algorithm to solve the equations of motion (for further details see [e.g.]{} [@juho10; @Lintuvuori10]). [Simulation parameters:]{} We consider both pushers ($\beta < 0$) and pullers ($\beta > 0$). We fix the $v_0=0.0015$, giving the particle velocity $u_0\equiv \tfrac{2}{3}v_0=10^{-3}$ in lattice units (LU), but vary the squirmer parameter in the range $\beta \in [-5,+5]$. We fix the fluid viscosity $\eta = 0.1$ and the swimmer radius $R=4.0$ in LU (Fig. 1(a)). To model the nematic liquid crystal we use: $A_0 = 1.0$, $\gamma = 3.0$, $K = 0.01$, $\xi = 0.7$, $\Gamma = 0.3$ and a rotational viscosity $\gamma_1=\tfrac{2q^2}{\Gamma}=\tfrac{5}{3}$. The physics of our system is governed by the Reynolds (Re) and Ericksen (Er) numbers, which give the ratio of inertial and viscous forces, as well as the ratio of viscous and elastic force, respectively. Using the parameters above, we recover the following upper limits $\mathrm{Re} \equiv \tfrac{u_0R}{\eta}\approx 0.04$ and $\mathrm{Er}\equiv \tfrac{\gamma_1u_0R}{K}\approx 0.68$. Simulations are carried in a rectangular simulation box $64\times 64\times 64$, with periodic boundary conditions. ![(a) A cartoon showing the squirmer in a nematic liquid crystal, defining the angle $\phi$ used in the text, between the particle swimming direction $\mathbf{u}_0$ and the nematic director $\hat{\mathbf{n}}$. Examples of (b) the trajectory in $x-z$ plane and (c) the $\phi(t)$ observed in simulations of a puller ($\beta = +0.2$) and of a pusher ($\beta = -0.2$), with an initial orientation $\phi_0=45^{\circ}$.[]{data-label="Fig:AnglePM5"}](Figure1.eps){width="\columnwidth"} [Results:]{} First we consider the case where the particle surface does not impose an alignment of the the nematic director ($W=0$). We place a single swimmer into a nematic liquid crystal with an initial angle $\phi_0=45^{\circ}$ between the squirmer orientation and $\hat{\mathbf{n}}$ (Fig. 1(a)). For a puller ($\beta = +0.2$), the hydrodynamically induced torque rotates the particle away from the nematic director leading to a curved trajectory towards a direction perpendicular to $\hat{\mathbf{n}}$ (Fig. 1(b) and (c) solid line). A $\beta = -0.2$ pusher instead starts to turn in the opposite direction, leading to swimming in the direction set by the nematic director (Fig. 1(b) dashed line), reaching a a steady state orientation $\phi\approx 0$ (Fig. 1(c) dashed line). (See also [@Suppl] for additional movies of the puller and pusher.) The alignment of a pusher resembles the observation of bacterial swimmers in nematic LCs [@Smalyukh08; @Kumar13; @Aranson14; @Abbott14; @Weibel15], which are known to be rod-like pushers. However, for rod-like swimmers there exists an elastic energy penalty of re-alignment which depends on the orientation $\phi$ with respect to the $\hat{\mathbf n}$ and it is minimised when they align along $\hat{\mathbf n}$. Resulting elastic torque has been estimated as $T_{\mathrm{elastic}}\sim 4\pi K\phi L\ln(2L/R)\sim10^5\phi\mathrm{pN}\cdot\mathrm{nm}$ [@Smalyukh08], which is considerably larger than that typically generated by the bacteria themselves $\sim 10^3\mathrm{pN}\cdot\mathrm{nm}$ [@Berg07; @Smalyukh08]. Thus rod-like swimmers are expected to always align along $\hat{\mathbf n}$ independently of their swimming mechanism. On the contrary, for spherical swimmers $T_{\mathrm{elastic}} = 0$ thus any torque must arise solely from hydrodynamic interactions. To analyse the underlying physical mechanism, we discuss how the squirmer’s flow field $\mathbf{v}(\mathbf{r})$ interacts with a liquid crystal in terms of the nematohydrodynamic equations [@Suppl]. We study the torque exerted on the moving particle, $$\mathbf{T}=\oint\mathbf{r}\times\mathbf{\sigma}\cdot d\mathbf{S},$$ where the integral runs over the particle surface having oriented surface element $d\mathbf{S}$. A squirmer moving in an isotropic fluid with a viscosity $\eta_{\text{iso}}$ has a flow field $\mathbf{v}_{\text{iso}}(\mathbf{r})$. The viscous stress is defined as a linear function of velocity derivatives, $\mathbf{\sigma}_\text{iso}=\eta_\text{iso} \mathbf{\nabla}\mathbf{v}_\text{iso}$, and for a spherical particle one has $\mathbf{T}=0$. In a liquid crystal, the viscosity is an anisotropic fourth-rank tensor $\boldsymbol{\eta}$, and the stress is a rather intricate function of the strain $ \mathbf{\nabla}\mathbf{v}$ and the the order parameter $\mathbf{n}$. There is no analytical result for the squirmer velocity field $\mathbf{v}(\mathbf{r})$ in LC [@Stark01]. We resort to a simple approximation that consists in evaluating the stress with the anisotropic viscosity $\boldsymbol{\eta}$ (given by Leslie coefficients $\alpha_i$ for nematic LC [@deG93; @Suppl]) but using the velocity field $\mathbf{v}_{\text{iso}}$ [@Mon12], in the limit of small Reynolds and Ericksen numbers. From the velocity field of a moving squirmer $\mathbf{v}_{\text{iso}}$, we readily obtain the stress and the nemato-hydrodynamic torque exerted on the particle (for detailed calculation see supplementary material [@Suppl]). The anisotropic part of the viscous stress is dominated by $\mathbf{\sigma}-\mathbf{\sigma}_\text{iso}\propto\beta\boldsymbol{\eta}\hat{\mathbf{n}}(\hat{\mathbf{n}}\times\mathbf{\omega})$, where $\mathbf{\omega}=\mathbf{\nabla}\times\mathbf{v}_\text{iso}$ is the vorticity of the flow field and $\boldsymbol{\eta}$ the viscosity tensor. Inserting the known velocity field of a squirmer, we obtain the torque [@Suppl] $$\mathbf{T}_N = - 8\pi\beta\hat{\eta}v_0R^2 (\hat{\mathbf{n}}\cdot \hat{\mathbf{u}}) \hat{\mathbf{n}}\times \hat{\mathbf{u}}, \label{eq:torque}$$ where $\hat{\mathbf{u}}$ is the particle axis. The effective viscosity coefficient $\hat{\eta} = \tfrac{\alpha_1}{35} + \tfrac{\alpha_2 + \alpha_3}{2} + \tfrac{\alpha_5+\alpha_6}{20}$ is expressed in terms of the Leslie parameters $\alpha_i$ of a nematic liquid crystal [@Suppl] and is dominated by the coefficients $\alpha_{2,3}$ related to the rotational viscosities, while $\alpha_{1,5,6}$ corresponds to shear viscosities [@deG93; @Suppl]. When $\beta\hat{\eta}>0$, the torque aligns the particle axis on the order parameter. Throughout this paper we assume $\hat\eta<0$, which corresponds to measured values for common nematic LCs [e.g.]{} 5CB and MBBA and to the simulations [@Suppl]. Then Eq. (\[eq:torque\]) predicts that the stable orientation of pullers ($\beta>0$) is perpendicular to the nematic order, whereas pushers ($\beta<0$) move in the parallel direction. To test this predictions, we carried out simulations for a $\beta = +5$ puller and a $\beta = -5$ pusher, and initialised the system close to the unstable orientation. Fig. 2a shows for the evolution of $\phi(t)$ an S-shaped trajectory, towards the stable positions given by (eq. \[eq:torque\]). To determine the angular velocity $\Omega$, we match the torque $T_N$ with the friction induced by the particle’s rotation, $T_N - 8\pi\hat\eta_\Omega R^3 \Omega=0$, with the viscosity $\hat\eta_\Omega$ of rotational Stokes drag [@Suppl]. Noting that the scalar and vector products in (\[eq:torque\]) result in a factor $\cos\phi\sin\phi=\frac{1}{2}\sin(2\phi)$, we find $$\Omega= -\frac{1}{2}\beta\sin(2\phi)\frac{\hat\eta}{\hat\eta_\Omega}\frac{v_0}{R}. \label{eq:rotvel}$$ The scale is given by $v_0/R$, and $\Omega$ is proportional to the squirmer parameter $\beta$ and varies with the angle $\phi$. In Figure 2(b), this is compared with the numerical derivative $\Omega(\phi)=d\phi/dt$ from the simulation data for the $\beta=+5$ puller. The data shows very good agreement between theory and numerics. Starting from the initial position $\phi=0$, the simulated velocity increases linearly with $\phi$, then reaches a maximum at $\phi\approx 45^{\circ}$ and finally slows down when approaching the stable orientation $\phi=90^{\circ}$. ![(a) $\phi(t)$ exhibits an S-shaped evolution, with a stable configuration $\phi\approx 90^{\circ}$ ($\phi\approx 0^{\circ}$) for a puller (pusher). (b) The rotational velocity, $\Omega(\phi)$ is symmetric around $\phi = 45^{\circ}$ and vanishes for $\phi\rightarrow 0$ and $90^{\circ}$, in agreement with theoretical arguments (see text for details).[]{data-label="Fig:AnglePM5"}](Figure2.eps){width="\columnwidth"} Modifying the squirmer parameter $\beta$ keeps the swimming speed constant, but changes the power of the squirmer flow field [@brown15]. This far we have established that the sign of $\beta$ defines the stable swimming direction with respect to the nematic axis. To understand how the magnitude of the the hydrodynamically induced torques depend on the power of the squirmer flow field, we initialise the simulations with $\phi_0 = 45^{\circ}$, and systematically vary $\beta$ between -5 and +5. We evaluate the $\Omega(\beta)$ from a linear fit to early times on $\phi(t)$ data (see [e.g.]{} early times in Fig. 1(c)). The $\Omega(\beta)$ from simulations shows a linear dependence for all the values of $\beta$ considered (Fig. 3) and indeed the theory predicts $\Omega(\beta)\sim \beta$ for a fixed $\phi$ (see [e.g.]{} eq. (\[eq:rotvel\])). ![The rotational velocity $\Omega(\beta)$ shows a linear dependence of the squirming parameter $\beta$ for a fixed $\phi$. For $\beta = 0$, $\Omega$ takes a value $a\approx 8\times10^{-6}\tfrac{\text{rad}}{[t]}$ (see inset and text for details). []{data-label="Fig:RotVel"}](Figure3.eps){width="\columnwidth"} Interestingly our numerical simulation results show that the re-orientation dynamics for pullers is slightly more rapid than for pushers (see Fig. 1c and inset in Fig. 3 for $\beta =\pm 0.2$). Also the angular velocity shown in Fig. 3, does not vanish at $\beta = 0$ (inset Fig. 3) but in a steady state a neutral squirmer swims perpendicular to $\hat{\mathbf{n}}$ (See the supplement [@Suppl] for $\phi(t)$ for $\beta = 0$). This behaviour is not captured by our analytics. In our theoretical treatment we replace $\mathbf{v}(\mathbf{r})$ with the velocity field calculated in an isotropic liquid $\mathbf{v}_\text{iso}(\mathbf{r})$. The analytical results agree remarkably well with the (more precise) numerical simulations, concerning the dependencies of $\Omega$ on the squirmer parameter $\beta$ and the orientational angle $\phi$ (see [e.g.]{} Fig. 2(b) and Fig. 3). This is in line with a previous study of anisotropic diffusion of colloids, where this approximation was shown to compare favorably with numerically exact results [@Mon12]. The reorientation of $\beta =0$ swimmer could probably be reproduced when using the exact velocity field $\mathbf{v}(r)$, which depends itself on the viscosity anisotropy $\boldsymbol{\eta}$. ![The angle $\phi(t)$ for particles with (a) homeotropic and (b) planar anchoring of the nematic director at the particle surface ($WR/K\approx 4$) for both a puller ($\beta = +5$) and a pusher ($\beta = -5$). The insets show the defect structure around a passive particle: (a) Saturn ring defect for a homeotropic anchoring and (b) two boojums for a planar anchoring at the particle surface. (The arrow denotes the orientation of the far-field nematic director.)[]{data-label="Fig:AnglePMAnchoring"}](Figure4.eps){width="\columnwidth"} In all the examples above, we have considered a case where there is no anchoring at the nematic director at the surface of the colloidal particle ($W=0$). Typically in experiments the particle surface interacts with the nematic director ($W>0$). The case of homeotropic anchoring can lead to the formation of a Saturn ring defect near the particle surface (see e.g. inset in Fig. 4(a)). In the case of degenerate planar anchoring, two boojums are observed at both poles of the particle (inset of Fig. 4(b)). We still observe the reorientation of the squirmers when a reasonably strong surface anchoring is included ($WR/K = 4$), as shown in Fig. 4 for $\beta = + 5$ puller and $\beta = - 5$ pusher. This provides further evidence that the re-orientation is due to the hydrodynamic coupling between the squirmer flow field and the anisotropic viscosities of the LC, as opposed to short range elastic interactions. Our main finding is that nematic liquid crystal exerts a torque on a spherical microswimmer. This should be easily observable in experiments. Using typical values for the LC viscosities [@deG93; @Suppl], and for microswimmers ($R\sim 1\mu$m and $v_0\sim 1\ldots10\mu$m/s), we can estimate the magnitude of the torque (eq. \[eq:torque\]) $T\sim 4\beta\times(10^2\ldots 10^{3})\mathrm{pN}\cdot\mathrm{nm}$, and $\Omega\sim \beta \tfrac{\mathrm{rad}}{s}$, which is comparable to the recently observed rotation induced by a laser intensity gradient on a thermally powered Janus particle [@Loz16; @Bic14]. Further, the reorientation rate $\Omega$ is much faster than typical rotational diffusion. These, combined with the observation that the steady state behaviour is retained for $\tfrac{WR}{K}>0$, suggest that our prediction should be testable in the laboratory, for example by dispersing artificial swimmers [e.g.]{} [@brown14; @brown15; @ebbens12b; @sabass10; @ebbens12; @ebbens14; @wang15; @das15; @Loz16] into standard nematic liquid crystals [e.g.]{} 5CB or MBBA. [Conclusions:]{} We have presented a combined simulation and analytical calculation study of the steady state swimming of a spherical squirmer in a nematic liquid crystal. In a steady state a pusher will swim along the nematic director while a puller will be moving perpendicular to the direction set by the far-field $\hat{\mathbf{n}}$. We show via analytic calculations that the re-orientation of the swimmers arises from the hydrodynamic coupling between the squirmer flow field and the anisotropicity of the liquid crystalline viscosities. For a passive spherical colloidal particle moving slowly in a thermotropic nematic LC a ratio of viscosities parallel ($\|\|$) and perpendicular ($\perp$) to $\hat{\mathbf{n}}$ has been observed $\tfrac{\eta_{\perp}}{\eta_{\|\|}}\approx 2$ experimentally [@Loudet04; @Tiffany], and by both theoretical calculations and simulations [@Stark01; @Lintuvuori10]. Our calculations show that the anisotropy of the liquid crystal viscosities [@miesowicz] gives a rise to a hydrodynamic torque on the squirmer, leading to the observed steady state behaviour. Finally, the steady state behaviour persists even when a strong anchoring of the LC director at the particle surface is included, rendering it directly experimentally relevant. The predictions should be valid for spherical microswimmers. A good candidate for an experimental realisation of predictions would be to consider lyotropic nematic liquid crystal [@Mon12], for both artificial or bacterial swimmers. Here, recent experiments of a diffusion of colloidal particles showed a viscosity ratio $\eta_{\perp}/\eta_{\|\|}\sim 4$ [@Mon12], which is larger anisotropy than considered here. Using thermotropic (oil-based) LCs, would require particles capable swimming in oil. The predictions presented here could also be valid for a wider class of materials which exhibit anisotropic viscosities, [e.g.]{} lyotropic lamellar phases could an interesting host material for future studies. Our results suggest that anisotropic materials could offer an exciting, yet easy-to-use, platform to guide microswimmers in the bulk. This could allow for example directed transport, or sorting of swimmers based on their hydrodynamic signature, by simply dispersing them into an environment with a broken symmetry ([e.g]{} nematic liquid crystal.) [Acknowledgments:]{} JSL acknowledges funding from IdEx Bordeaux. AW acknowledges support by Agence Nationale de la Recherche through contract ANR-13-IS04-0003. [53]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{} , **, (). , , , , , , , , (). , , (). , , , , , , , , (). , , , , , , (). , , (). , , , , , (). , , , , , , , , (). , , , , , , (). , , , , , , , , , (). , , (). , , , , , (). , , (). , , (). , , (). , , , , , (). , , , , , , , (). , , , , , (). , , , , (). , , , , , (). , , , , (). , , , , (). , , (). , , , , , (). , , , , , , (). , , , , , (). , , , , , (). , , , , , , (). , , , , , , (). , , , , , , (). , , , , , (). , , , , (). , , , , (). , , , , (). , , (). , , (). , , , , , , , , (). , , , , (). , , (). , , (). , , (). , , (). , , (). , (, ). , , , , **, (). , , , , *, (). D. Marenduzzo, E. Orlandini, M. E. Cates and J. Yeomans, [Phys. Rev. E]{} [76]{}, 031921 (2007). H. Stark, Phys. Rep. 351, 387 (2001). , , , , *, (). , , (). , , , , , , , (). , (, ). , , , **, (). , , , , (). @\#1[@@@[(S\#1@italiccorr)]{}]{} [Supplementary information for squirmers in nematic liquid crystal*]{} Calculation of the Leslie viscosities ===================================== For the lattice Boltzmann model presented in the main text, the Leslie viscosities of a passive nematic liquid crystals are defined in the terms of, the rotation diffusion constant $~\Gamma$, the $\tfrac{3}{2}$ of the largest eigenvalue of the order parameter tensor $\mathbf{Q}$, $q$, flow alignment parameter $\xi$ and isotropic viscosity $\eta$. Using the values from the simulations in the main text $\Gamma = 0.3$, $q=\tfrac{1}{2}$, $\xi = 0.7$ and $\eta = 0.1$, we recover following Leslie viscosities [@deG93; @davide07] in simulation units: $$\begin{aligned} \alpha_1 &= -\frac{2}{3\Gamma}q^2(3 + 4q - 4q^2)\xi^2 &&\approx -1.09, \\ \alpha_2 &= \frac{1}{\Gamma}\left(-\frac{1}{3}q(2+q)\xi - q^2\right) &&\approx -1.81, \\ \alpha_3 &= \frac{1}{\Gamma}\left(-\frac{1}{3}q(2+q)\xi + q^2\right) &&\approx -0.14, \\ \alpha_4 &= \frac{4}{9\Gamma}(1-q)^2\xi^2 + \eta &&\approx 0.28, \\ \alpha_5 &= \frac{1}{3\Gamma}\left[q(4-q)\xi^2 + q(2+q)\xi\right] &&\approx 1.93, \\ \alpha_6 &= \frac{1}{3\Gamma}\left[q(4-q)\xi^2 - q(2+q)\xi\right] &&\approx -0.02. \label{alpha model}\end{aligned}$$ These give $\tfrac{\alpha_2 + \alpha_3}{2}\left(\approx -0.98\right) < 0$ as well as $\left\|\tfrac{\alpha_2 + \alpha_3}{2}\right\| >> \left\|\tfrac{\alpha_5 + \alpha_6}{20}\right\|\left(\approx 0.098\right)$ and $\left\|\tfrac{\alpha_2 + \alpha_3}{2}\right\| >> \left\|\tfrac{\alpha_1}{35}\right\|\left(\approx 0.03\right)$, as required in the main text. For comparison, we give measured values for liquid crystals 5CB (MBBA) of the coefficients $\alpha_2+\alpha_3=-107$ $(-79)$ mPa.s, $\alpha_5+\alpha_6=107$ $(81)$ mPa.s, $\alpha_1=-11$ $(7)$ mPa.s [@deG93]. The isotropic term takes the value $\alpha_4=75$ $(83)$ mPa.s. Neutral squirmer ================ ![The time evolution of the angle $\phi(t)$ between the squirmer orientation and nematic director $\hat{\mathbf{n}}$, shows that neutral squirmer ($\beta = 0$) orients perpendicular to the far-field nematic director.[]{data-label="Fig:AngleB0.0"}](AngleB0.eps){width="0.5\columnwidth"} Nematohydrodynamics =================== The action of a moving fluid on a dispersed particle is given by well-known relations for the force $$\mathbf{F}=\oint\mathbf{\sigma}\cdot d\mathbf{S} \label{32}$$ and the torque $$\mathbf{T}=\oint\mathbf{r}\times\mathbf{\sigma}\cdot d\mathbf{S}, \label{34}$$ where the integral is over the particle surface with position vector $\mathbf{r}$ and the oriented surface element $d\mathbf{S}=-dS\mathbf{\hat{r}}$. The stress tensor $\sigma_{ij}=\sigma_{ij}^{\prime}-P\delta_{ij}$ consists of a viscous part $\sigma_{ij}^{\prime}$ and the pressure $P$. In the framework of the stationary Leslie-Ericksen equations the viscous stress reads as [@deG93] $$\mathbf{\sigma}^{\prime} =\alpha_{1}\mathbf{nn\cdot E \cdot nn} + \alpha_{4} \mathbf{E} +\frac{\alpha_{5}+\alpha_{6}}{2}\left( \mathbf{nn\cdot E} + \mathbf{E \cdot nn}\right) +\frac{\alpha_{3}+\alpha_{2}}{2}\left( \mathbf{n N} + \mathbf{Nn}\right), \label{10}$$ with the strain tensor $$\mathbf{E}=\frac{1}{2} \left(\mathbf{\nabla v} + (\mathbf{\nabla v})^\dagger\right),$$ the vorticity $$\mathbf{\omega}= \mathbf{\nabla } \times \mathbf{v},$$ and the rate of change of the nematic order parameter $$\mathbf{N} = [ (\mathbf{v} - \mathbf{u})\cdot\mathbf{\nabla}] \hat{\mathbf{n}} + \frac{1}{2}\hat{\mathbf{n}}\times\mathbf{\omega}.$$ Approximations ============== With (\[10\]) it is not possible to solve Stokes’ equation ${\bf\nabla}\cdot{\bf\sigma}=0$. In order to obtain a problem that is tractable in a simple analytical approach, we resort to the following approximations. First, we neglect the deformation of the nematic order due to the interaction with the particle’s surface and with its velocity field. This corresponds to case of no surface anchoring ($W=0$) and the limit of a small Ericksen number, $\text{Er}\to 0$. In other words we assume that the order parameter is constant in space and time, $$\hat{\mathbf{n}}=\text{const.}$$ Then the rate of change of the order parameter simplifies as $$\mathbf{N}=\frac{1}{2}\hat{\mathbf{n}}\times\mathbf{\omega}.$$ Second, we calculate the viscous stress (\[10\]) with the velocity field of an active particle in an isotropic fluid $\mathbf{v}_{\text{iso}}(\mathbf{r})$. Writing the lowest terms of the well-known series as gradient and and rotational fields, we have $$\mathbf{v}_\text{iso} = \mathbf{\nabla}\Phi + \mathbf{\nabla} \times \mathbf{A}, \label{36}$$ with the scalar $$\Phi = - v_0 \left(\frac{R^3\hat{\mathbf{r}}\cdot\hat{\mathbf{u}}}{3 r^2} + \beta \frac{R^4P_2(\hat{\mathbf{r}}\cdot\hat{\mathbf{u}})}{3 r^3} \right) = - v_0 \left(\frac{R^3 \cos\theta}{3 r^2} + \beta \frac{R^4P_2( \cos\theta)}{3 r^3} \right)$$ and the vector field $$\mathbf{A} = \beta v_0\hat{\mathbf{u}}\times\hat{\mathbf{r}}\frac{R^2(\hat{\mathbf{r}}\cdot\hat{\mathbf{u}})}{2 r} = \beta v_0 \frac{R^2 \sin\theta \cos\theta}{2 r} \mathbf{e}_\varphi ,$$ where we have defined the unit vectors $\hat{\mathbf{r}}=\mathbf{r}/r$ and $\hat{\mathbf{u}}=\mathbf{u}/u$. The polar angle $\theta$ and the azimuthal unit vector $\mathbf{e}_\varphi$ are defined with respect to the particle axis $\hat{\mathbf{u}}$. The above relations give the flow field for an isotropic fluid, which is calculated with the viscosity parameter $\alpha_{4}$ only. Thus the viscosity anisotropy appears only in the prefactors in (\[10\]), and we neglect that the velocity field itself depends on the anisotropic viscosity parameters $\alpha_{1},\alpha_{2},\alpha_{3},\alpha_{5},\alpha_{6}$. Measured values for the anisotropy parameters are not small as compared to the isotropic viscosity $\alpha_4$ [@deG93]. In previous work, this approximation was used for evaluating the anisotropy of Brownian motion of spherical particles in zero anchoring conditions. Comparison with numerically exact results [@Sta01] revealed an error of less than ten percent [@Mon12], suggesting this approximation to be rather robust. Spelling out the about derivatives one readily obtains the explicit form of $\mathbf{v}_\text{iso}$. From the expression at the surface $r=R$, one finds the particle velocity $u_0=-\frac{2}{3}v_0$ and the slip velocity $$v_s = v_0 \sin\theta \left(1 + \beta \cos\theta \right) ,$$ with the squirmer parameter $\beta$. Note that $\mathbf{v}_\text{iso}$ is given in the laboratory frame and $v_{s}$ in the particle-fixed frame; both are related through $v_s=(\mathbf{v}_\text{iso}-\mathbf{u})\cdot(1-\hat{\mathbf{r}}\hat{\mathbf{r}})$. Torque exerted on the particle ============================== Calculating the strain tensor from (\[36\]) and inserting the symmetrized stress tensor (\[10\]), we evaluate the force and torque exerted on the particle. Not surprisingly, (\[32\]) vanishes, $\mathbf{F}=0$, since there is no external potential. The torque, on the contrary, takes a finite value. Inserting the velocity field in (\[10\]) and performing the integral in (\[34\]), one can evaluate the torque. For $\hat{\mathbf{n}}$ along $x$-axis and the particle moving in the $x-z$-plane, we find $$\mathbf{T}_N = -4\pi \beta\sin(2\phi)\hat{\eta} v_0 R^{2}\mathbf{\hat{y}} =- 8\pi\beta\hat{\eta}v_0R^2 (\hat{\mathbf{n}}\cdot \hat{\mathbf{u}}) \hat{\mathbf{n}}\times \hat{\mathbf{u}} , \label{42}$$ with $\phi$ denoting the angle between the particle axis and the order parameter, $\cos\phi=\hat{\mathbf{u}}\cdot\hat{\mathbf{n}}$, and the unit vector $\mathbf{\hat{y}}$ which is perpendicular to both $\mathbf{u}$ and $\hat{\mathbf{n}}$. The viscosity coefficient reads $$\hat{\eta}=\frac{\alpha_{1}}{35}+\frac{\alpha_{2}+\alpha_{3}}{2}+\frac{\alpha_{5}+\alpha_{6}}{20}.\label{44}$$ Not surprisingly, the isotropic viscosity $\alpha_{4}$ does not contribute; $\hat{\eta}$ is determined by the anisotropy parameters $\alpha_{1},\alpha_{2},\alpha_{3},\alpha_{5},\alpha_{6}$. The torque is perpendicular on the order parameter $\hat{\mathbf{n}}$ and the particle velocity $\mathbf{u}$; it vanishes both for parallel and perpendicular orientations, $\phi=0$ and $\phi=\frac{\pi}{2}$. With the viscosity parameters of commonly used LC, such as 5CB and MBBA, one finds that $\hat{\eta}$ is determined by the coefficient $\alpha_{2}+\alpha_{3}$. In other words, the interaction of an active particle and the nematic order is dominated by the last term of the viscous stress (\[10\]), which in a case of constant $\hat{\mathbf{n}}$ accounts for the vorticity of the fluid velocity field. For the mentioned systems $\alpha_{2}+\alpha_{3}$ is negative. In view of (\[42\]) this means that for $\beta>0$ the particle axis $\mathbf{\hat{u}}$ is turned away from the order parameter, and that the stable orientation is perpendicular to $\hat{\mathbf{n}}$. On the other hand, for a negative squirmer parameter $\beta<0$, we expect that the active particle aligns on the nematic order parameter. Angular velocity of the squirmer particle ========================================= The torque $\mathbf{T} $ exerted by the nematic order results in a angular velocity $\Omega$ of the squirmer particle, which in turn induces an additional flow field $$\mathbf{v}_\Omega = \frac{ R^3}{r^3} \mathbf{\Omega}\times \mathbf{r} = \frac{\Omega R^3}{r^2} \sin\theta \mathbf{e}_\varphi. \label{51}$$ The angular velocity adjusts such that the corresponding viscous torque exactly cancels that exerted by the nematic order, $\mathbf{T}_\Omega+ \mathbf{T}=0$. Here we use this relation in order to determine $\Omega$. Calculating the strain tensor $\mathbf{E}_\Omega$ and the vorticity vector $\mathbf{\omega}_\Omega$ we obtain the viscous stress (\[10\]). Upon performing the surface integral we find the viscous torque $$\mathbf{T}_\Omega= - 8\pi\hat{\eta}_\Omega \Omega R^3 \mathbf{e}_y, \label{52}$$ with the effective viscosity $$\hat{\eta}_\Omega= \frac{\alpha_4}{2} + \frac{\alpha_2+\alpha_3}{20} + \frac{9(\alpha_5+\alpha_6)}{40} + \frac{\alpha_1}{10}.$$ Identifying the elastic and viscous torques, we find the angular velocity of the squirmer $$\Omega = -\beta \frac{\sin(2\phi)}{2} \frac{\hat\alpha}{\hat\alpha_\Omega} \frac{v_0}{R} \equiv -\sin(2\phi)\Omega_0. \label{52}$$ Taking the ratio of the viscosity parameters equal to unity, $\hat\alpha\sim \hat\alpha_\Omega$, we find $\Omega_0=\beta v_0/2R$. Here a remark on the validity of our approximative evaluation of viscous stresses is in order: Since the viscous torque is necessarily opposite to the driving velocity $\Omega$, the viscosity parameter needs to be positive; in other words, the anisotropy parameters should be small as compared to the isotropic viscosity $\alpha_4$. This condition is not fulfilled by the model parameters (\[alpha model\]) nor by the values measured for common liquid crystals. As a consequence, the numerical values of the coefficients $\hat{\eta}$ and $\hat{\eta}_\Omega$ are probably subject to large uncertainties. [99]{} P. G. de Gennes and J. Prost, [The physics of Liquid crystals]{} 2nd ed. (Clarendon Press, Oxford, 1993). D. Marenduzzo, E. Orlandini, M. E. Cates and J. Yeomans, [Phys. Rev. E]{} [76]{}, 031921 (2007). H. Stark, Phys. Rep. 351, 387 (2001). F. Mondiot, J.-C. Loudet, O. Mondain-Moval, P. Snabre, A. Vilquin, A. Würger, Phys. Rev. E 86*, 010401(R) (2012).	{ "pile_set_name": "ArXiv" }
--- author: - \| Zhide Chen, Hong Zhu\ Department of Computer Science, Fudan University, Shanghai 200433, P.R.China.\ Key Laboratory of Intelligent Information Processing, Fudan University, Shanghai 200433, P.R.China.\ {02021091, hzhu}@fudan.edu.cn title: 'Quantum $m$-out-of-$n$ Oblivious Transfer[^1]' --- \[section\] \[section\] \[section\] \[section\] \[section\] \[section\] \[section\] \[section\] \[section\] normalsize[setsizexpt plus2pt minus5pt@ plus3pt plus3pt minus3ptlistilistI]{} startsection [section]{}[1]{}[@]{}[24pt plus 2pt minus 2pt]{} [12pt plus 2pt minus 2pt]{}[**]{} ================================================================================================= startsection [subsection]{}[2]{}[@]{}[12pt plus 2pt minus 2pt]{} [12pt plus 2pt minus 2pt]{}[*]{} ---------------------------------------------------------------------------------------------------- Abstract {#abstract .unnumbered} -------- [In the $m$-out-of-$n$ Oblivious Transfer ($OT$) model, one party $Alice$ sends $n$ bits to another party $Bob$, $Bob$ can get only $m$ bits from the $n$ bits. However, $Alice$ cannot know which $m$ bits $Bob$ received. Y.Mu and Naor presented classical $m$-out-of-$n$ Oblivious Transfer based on discrete logarithm. As the work of Shor, the discrete logarithm can be solved in polynomial time by quantum computers, so such $OT$s are unsecure to the quantum computer. In this paper, we construct a quantum $m$-out-of-$n$ $OT$ ($QOT$) scheme based on the transmission of polarized light and show that the scheme is robust to general attacks, i.e. the $QOT$ scheme satisfies statistical correctness and statistical privacy.\ \ Keywords. Quantum, Oblivious Transfer.* ]{} Introduction ============ A number of recent papers have provided compelling evidence that certain computational, cryptographic, and information theoretic tasks can be performed more efficiently by models based on quantum physics than those based on classical physics [@[Shor97]]. Oblivious Transfer (OT) is used as a key component in many applications of cryptography [@[WIE]; @[EGL85]; @[R81]]. Informally speaking in an Oblivious Transfer, $Alice$ sends a bit to $Bob$ that he receives half the time (this fact is out of their control), $Alice$ does not find out what happened, $Bob$ knows if he get the bit or nothing. Similarly, in a 1-out-of-2 Oblivious Transfer, $Alice$ has two bits $b_0,b_1$ that she sends to $Bob$ in such a way that he can decide to get either of them at his choosing but not both. $Alice$ never finds out which bit $Bob$ received. In 2001, Naor presented a 1-out-of-n Oblivious Transfer [@[Naor01]], Y.Mu showed that $m$-out-of-$n$ Oblivious Transfer could also be realized based on the discrete logarithm. In the $m$-out-of-$n$ Oblivious Transfer($1\leq m<n$) , $Alice$ sends $n$ bits to $Bob$, $Bob$ can get only $m$ of them. In the case of quantum, Claude Crépeau provided a 1-out-of-2 quantum Oblivious Transfer based on the transmission of polarized light in 1994. The protocol of Crépeau’s can be used directly to implement a one-out-of-three Oblivious Transfer. The organization of this paper is as following: in section 2, we give the definitions of the correctness and privacy of the $m$-out-of-$n$ OT protocol. In section 3, we review the 1-out-of-2 OT of Claude Crépeau and its intuition. In section 4, we construct an $m$-out-of-$n$ OT, and in section 5 we show that this scheme satisfies statistical correctness and statistical privacy . Definitions =========== The natural constraints(see below) of correctness and privacy of a $m$-out-of-$n$ OT($1\leq m<n$) is showed below. Perfect Correctness: It should be that when $Alice$ and $Bob$ follow the protocol and start with $Alice's$ input bits $b_{1},b_{2},\cdots,b_{n}$ and $Bob's$ input $c_1,c_2,\dots,c_m\in \{1,2,\cdots,n\}$, they finish with $Bob$ getting $b_{c_1},b_{c_2},\cdots,b_{c_m} \in $ $\{$ $b_{1}$, $b_{2}$, $\cdots$, $b_{n}\}$. Perfect Privacy: It should be that, $Alice$ can not find out about $c_1,c_2,\dots,c_m$, and $Bob$ can not find out more than $m$ of $b_1,b_2,\dots,b_n$. The protocol we describe in the next section is of probabilistic nature. We cannot show that this protocol perfectly satisfies the above constraints but satisfies in a statistical sense: after an amount of work in $O(N)$ time the protocol will satisfy for some positive constant $\epsilon <1$. Statistical Correctness: It should be that , except with probability at most $\varepsilon^{N}$, when $Alice$ and $Bob$ follow the protocol and start with $Alice's$ input bits $b_1,b_2,\dots,b_n$ and $Bob's$ input $c_1,c_2,\dots,c_m\in \{1,2,\cdots,n\}$ they finish with $Bob$ getting $b_{c_1},b_{c_2},\cdots,b_{c_m}\in \{b_{1},b_{2},\cdots,b_{n}\}$. Statistical Privacy: It should be that, except with probability at most $\epsilon^{N}$, $Alice$ can not find out $c_1,c_2,\dots,c_m$, and $Bob$ can not find out more than $m$ of $b_1,b_2,\dots,b_n$. Quantum 1-out-of-2 Oblivious Transfer ===================================== In this section, we introduce the quantum 1-out-of-2 OT provided by Claude Crépeau [@[C94]]. Let $\copyright\!\!\!\!\|$ denote the random variable that takes the binary value 0 with probability 1/2 and 1 with probability 1/2. Also, denote by $[\quad]_{i}$ the selection function such that $[a_{0},a_{1},\cdots,a_{k}]_{i}=a_{i}$. Let $\leftrightarrow \!\!\!\!\updownarrow\ =(\|\!\!\!\leftrightarrow\rangle,\|\!\!\uparrow\!\!\!\downarrow\rangle)$ and $\nwarrow\!\!\!\!\!\!\searrow\!\!\!\!\!\!\nearrow\!\!\!\!\!\!\swarrow\ =(\|\!\!\nwarrow\!\!\!\!\!\!\searrow\rangle,\|\!\nearrow\!\!\!\!\!\!\swarrow \rangle)$ denote respectively the bases of rectilinear and diagonal polarization in the quantum state space of a photon. The quantum 1-out-of-2 OT is as follows: Quantum 1-out-of-2 OT --------------------- 1-out-of-2 OT$(b_{0},b_{1})(c)$ 1. $DO_{i=1}^{2n}$ - $Alice$ picks a random bit $r_{i} \leftarrow \copyright\!\!\!\!\|$ - $Alice$ picks a random bit $\beta_{i}\leftarrow \copyright\!\!\!\!\|$ and defines her emission basis $(\|\varphi_{i}\rangle, \|\varphi_{i}^{\perp}\rangle)\leftarrow [\leftrightarrow \!\!\!\!\updownarrow,\nwarrow\!\!\!\!\!\!\searrow\!\!\!\!\!\!\nearrow\!\!\!\!\!\!\swarrow]_{\beta_{i}} $ - $Alice$ sends to Bob a photon $\pi_{i}$ with polarization $[\|\varphi_{i}\rangle, \|\varphi_{i}^{\perp}\rangle]_{r_i}$ - $Bob$ picks a random bit $\beta'_i \leftarrow \copyright\!\!\!\!\|$ and measures $\pi_{i}$ in basis $(\|\theta_{i}\rangle, \|\theta_{i}^{\perp}\rangle)\leftarrow[\leftrightarrow \!\!\!\!\updownarrow,\nwarrow\!\!\!\!\!\!\searrow\!\!\!\!\!\!\nearrow\!\!\!\!\!\!\swarrow]_{\beta_{i}'} $ - $Bob$ sets $r_{i}' \leftarrow \left\{% \begin{array}{ll} 0, & \hbox{if $\pi_{i}$ is observed as $\|\theta_{i}\rangle$} \\ 1, & \hbox{if $\pi_{i}$ is observed as $\|\theta_{i}^{\bot}\rangle$} \\ \end{array}% \right.$ 2. $DO_{i=1}^{n}$ - Bob runs $commit(r'_{i})$, $commit(\beta'_{i})$, $commit(r'_{n+i})$, $commit(\beta'_{n+i})$ with $Alice$ - $Alice$ picks $c_{i} \leftarrow \copyright\!\!\!\!\|$ and announces it to $Bob$ - Bob runs $unveil(r'_{nc_{i}+i}), unveil(\beta'_{nc_{i}+i})$ - $Alice $ checks that $\beta_{nc_{i}+i}=\beta'_{nc_{i}+i}\rightarrow r_{nc_{i}+i}=r'_{nc_{i}+i}$ - if $c_{i}=0$ then $Alice$ sets $\beta_{i}\leftarrow\beta_{n+i}$ and $r_{i}\leftarrow r_{n+i}$ and $Bob$ set $\beta'_{i}\leftarrow\beta'_{n+i}$ and $r'_{i}\leftarrow r'_{n+i}$ 3. $Alice$ announces her choices $\beta_{1}\beta_{2}\cdots\beta_{n}$ to $Bob$ 4. $Bob$ randomly selects two subsets $I_{0},I_{1}\subset \{1,2,\cdots,n\}$ subject to $\|I_{0}\|=\|I_{1}\|=n/3$, $I_{0} \cap I_{1}=\emptyset$ and $\forall i \in I_{c}, \beta_{i}=\beta_{i}'$, and he announces $\langle I_{0},I_{1} \rangle$ to $Alice$ 5. $Alice$ receives $\langle J_{0},J_{1} \rangle$=$\langle I_{0},I_{1} \rangle$, computes and sends $\widehat{b}_{0}\leftarrow b_{0} \oplus \bigoplus_{j \in J_{0}}r_{j}$ and $\widehat{b}_{1}\leftarrow b_{1} \oplus \bigoplus_{j \in J_{1}}r_{j} $ 6. $Bob$ receives $\langle \widehat b_{0},\widehat b_{1} \rangle$ and computes $b_{c}\leftarrow \widehat {b}_{c} \oplus \bigoplus_{j \in J_{c}} r'_{j} $ Intuition behind 1-out-of-2 OT ------------------------------ In this 1-out-of-2 QOT, $Alice$ must prevent $Bob$ from storing the photons and waiting until she discloses the bases before measuring them, which would allow him to obtain both of $Alice's$ bits with certainty. To realize this, $Alice$ gets $Bob$ to $commit$ to the bits that he received and the bases that he used to measure them. Before going ahead with $r_i$, say, $Alice$ checks that $Bob$ had committed properly to $r_{n+i}$ when he read that bit in the basis that she used to encode it. If at any stage $Alice$ observes a mistake ($\beta_{n+i}=\beta'_{n+i}$ but $r_{n+i}\neq r'_{n+i}$), she stops further interaction with $Bob$ who is definitely not performing his legal protocol (this should never happen if $Bob$ follows his protocol). In this protocol, $r_{1}r_{2} \cdots r_{n}$ are chosen by $Alice$ in step 1 and are sent to $Bob$ via an ambiguous coding referred to as the BB84 coding [@[BB84]]: when $Alice$ and $Bob$ choose the same emission and reception basis, the bit received is the same as what was sent and uncorrelated otherwise. $Bob$ builds two subsets: one $I_{c}$ that will allow him to get $b_{c}$, and one $I_{\overline{c}}$ that will spoil $b_{\overline{c}}$. The calculations of steps 5-6 are much that all the bits in a subset must be known by $Bob$ in order for him to be able to obtain the output bit connected to that subset. Protocol for Quantum $m$-out-of-$n$ Oblivious Transfer ====================================================== Weak Bit Commitment ------------------- In 1993, Gilles Brassard, etc provided a quantum bit commitment scheme provably unbreakable by both parties [@[BCJL]]. However, unconditionally quantum bit commitment was showed impossible [@Mayers]. In [@[DAUA]], Aharonov provided a weak bit commitment. [@[DAUA]] In the weak bit commitment protocol, the following requirements should hold. - If both Alice and Bob are honest, then both Alice and Bob accept. - (Binding) If Alice tries to change her mind about the value of $b$, then there is non zero probability that an honest Bob would reject. - (Sealing) If Bob attempts to learn information about the deposited bit $b$, then there is non zero probability that an honest Alice would reject. In the following scheme, $Bob$ will use this weak quantum bit commitment to commit. Intuition for $m$-out-of-$n$ OT ------------------------------- In the $m$-out-of-$n$ OT, $Bob$ should build $n$ subsets $I_{1},I_{2},\dots,I_{n}\subseteq \{1,2,\cdots,n\}$, $m$ of that will allow him to get $b_{c_1},b_{c_2},\dots,b_{c_m}$ ($c_1,c_2,\dots,c_m \in \{ 1,2,\dots,n \}$), and the other $I$’s will spoil the remnant $b$’s. In $I_1\cup I_2\cup\cdots\cup I_n$, the rate of the $i$’s satisfying $\beta_{i}'=\beta_{i}$ would be more than $\frac{m}{n}$ and less than $\frac{m+1}{n}$. i.e. $$\frac{m}{n}\leq \frac{\# \{i\| \beta _{i}= \beta _{i}', i\in I_1\cup\cdots\cup I_n\}}{\|I_1\cup\cdots\cup I_n\|}<\frac{m+1}{n}$$ In our scheme, we let the rate to be $\frac{\frac{m}{n}+\frac{m+1}{n}}{2}=\frac{2m+1}{2n}$. As $\beta$’s and $\beta'$’s are choice randomly, we have $$\lim_{N\rightarrow \infty} \frac{\#\{ \beta_i=\beta_i' \}}{N}=\frac{1}{2}.$$ For a large $N$, the rate of $i$’s in $\{1,2,\cdots,N \}$ that satisfy $\beta_{i}'=\beta_{i}$ would be approximately $\frac{1}{2}$, then $Bob$ should remove some $i$’s from the $\{1,2,\cdots,N \}$. The number of $i$’s that should be removed can be calculated as following:\ If $\frac{2m+1}{2n}<\frac{1}{2}$, there are more $i$’s that satisfy $\beta_{i}'=\beta_{i}$ than required, so $Bob$ should remove $x$ $i$’s that satisfying $\beta_{i}'= \beta_{i}$ from $\{1,2,\cdots,N \}$. $x$ can be calculated as follows: $$\begin{aligned} \frac{\frac{N}{2}-x}{N-x}&=&\frac{2m+1}{2n}\\ x&=&\frac{n-(2m+1)}{2n-(2m+1)}N\end{aligned}$$ If $\frac{2m+1}{2n}\geq\frac{1}{2}$, there are more $i$’s that satisfy $\beta_{i}'\neq\beta_{i}$ than required, so $Bob$ should remove $x$ $i$’s that satisfying $\beta_{i}'\neq \beta_{i}$ from $\{1,2,\cdots,N \}$. $x$ can be calculated as follows: $$\begin{aligned} \frac{\frac{N}{2}}{N-x}&=&\frac{2m+1}{2n}\\ x&=&\frac{(2m+1)-n}{2m+1}N\end{aligned}$$ $N$ must satisfy $(2n-(2m+1))(2m+1)\|((2m+1)-n)N$ so that $x$ would be an interger. we let the $i$’s that was removed from $\{1,2,\cdots,N\}$ be $u_1,u_2,\cdots,u_x$. Quantum $m$-out-of-$n$ OT ------------------------- In the $m$-out-of-$n$ $QOT$, $Alice$ has input $b_1,b_2,\cdots,b_n$, $Bob$ has input $c_1,c_2,\cdots,c_m$. The output of the scheme is $b_{c_1},b_{c_2},\cdots,b_{c_m}$. $m$-out-of-$n$ QOT$(b_1,b_2,\dots,b_n)(c_1,c_2,\dots,c_m)$ 1. $DO_{i=1}^{2N}$ - $Alice$ picks a random bit $r_{i} \leftarrow \copyright\!\!\!\!\|$ - $Alice$ picks a random bit $\beta_{i}\leftarrow \copyright\!\!\!\!\|$ and defines her emission basis $(\|\varphi_{i}\rangle, \|\varphi_{i}^{\perp}\rangle)\leftarrow[\leftrightarrow \!\!\!\!\updownarrow,\nwarrow\!\!\!\!\!\!\searrow\!\!\!\!\!\!\nearrow\!\!\!\!\!\!\swarrow]_{\beta_{i}} $ - $Alice$ sends to Bob a photon $\pi_{i}$ with polarization $[\|\varphi_{i}\rangle, \|\varphi_{i}^{\perp}\rangle]_{r_i}$ - $Bob$ picks a random bit $\beta'_i \leftarrow \copyright\!\!\!\!\|$ and measures $\pi_{i}$ in basis $(\|\theta_{i}\rangle, \|\theta_{i}^{\perp}\rangle)\leftarrow[\leftrightarrow \!\!\!\!\updownarrow,\nwarrow\!\!\!\!\!\!\searrow\!\!\!\!\!\!\nearrow\!\!\!\!\!\!\swarrow]_{\beta_{i}'} $ - $Bob$ sets $r_{i}' \leftarrow \left\{% \begin{array}{ll} 0, & \hbox{if $\pi_{i}$ is observed as $\|\theta_{i}\rangle$} \\ 1, & \hbox{if $\pi_{i}$ is observed as $\|\theta_{i}^{\bot}\rangle$} \\ \end{array}% \right. $ 2. $DO_{i=1}^{N}$ - Bob runs $commit(r'_{i})$, $commit(\beta'_{i})$, $commit(r'_{N+i})$, $commit(\beta'_{N+i})$ with $Alice$ - $Alice$ picks $d_{i} \leftarrow \copyright\!\!\!\!\|$ and announces it to $Bob$ - Bob runs $unveil(r'_{Nd_{i}+i}), unveil(\beta'_{Nd_{i}+i})$ - $Alice $ checks that $\beta_{Nd_{i}+i}=\beta'_{Nd_{i}+i}\rightarrow r_{Nd_{i}+i}=r'_{Nd_{i}+i}$ - if $d_{i}=0$ then $Alice$ sets $\beta_{i}\leftarrow\beta_{N+i}$ and $r_{i}\leftarrow r_{N+i}$ and $Bob$ set $\beta'_{i}\leftarrow\beta'_{N+i}$ and $r'_{i}\leftarrow r'_{N+i}$ 3. $Alice$ announces her choices $\beta_{1}\beta_{2}\cdots\beta_{N}$ to $Bob$ 4. $DO_{j=1}^{x}$ - If $\frac{2m+1}{2n}<\frac{1}{2}$ Bob runs $unveil(r'_{u_j})$, $ unveil(\beta'_{u_j})$ that satisfying $\beta_{u_j}= \beta'_{u_j}$, $Alice $ checks that $\beta_{u_j}=\beta'_{u_j}\rightarrow r_{u_j}=r'_{u_j}$ - If $\frac{2m+1}{2n}\geq \frac{1}{2}$ Bob runs $unveil(r'_{u_j})$, $unveil(\beta'_{u_j})$ that satisfying $\beta_{u_j}\neq \beta'_{u_j}$ 5. $Bob$ randomly selects n subsets $I_{1},I_{2},\cdots,I_{n} \subset \{1,2,\cdots,N\}-\{ u_1,u_2,\dots,u_x \}$ subject to $\|I_{1}\|=\|I_{2}\|=\cdots=\|I_{n}\|=(N-x)/n$, $\forall j\neq k$, $ I_{j} \cap I_{k}=\emptyset$ and $\forall j$ $\in I_{c_1}\cup I_{c_2}\cup \cdots\cup I_{c_m}$, $\beta_{j}=\beta_{j}'$, and he announces $\langle I_{1},I_{2},\cdots,I_{n} \rangle$ to $Alice$ 6. $Alice$ receives $\langle J_{1},J_{2},\cdots ,J_{n} \rangle$=$\langle I_{1},I_{2},\cdots,I_{n} \rangle$, computes and sends $\widehat{b}_{1}\leftarrow b_{1} \oplus \bigoplus_{j \in J_{1}}r_{j}$, $\widehat{b}_{2}\leftarrow b_{2} \oplus \bigoplus_{j \in J_{2}}r_{j}$, $\cdots$, $\widehat{b}_{n}\leftarrow b_{n} \oplus \bigoplus_{j \in J_{n}}r_{j}$ to $Bob$ 7. $Bob$ receives $\langle \widehat b_{1},\widehat b_{2},\cdots, \widehat b_{n} \rangle$ and computes $b_{c_i}\leftarrow \widehat {b}_{c_i} \oplus \bigoplus_{j \in J_{c_i}} r'_{c_j} , $ $i=1,2,\cdots,m$ Analysis ======== In the $m$-out-of-$n$ $QOT$, $Bob$ must read the photons sent by $Alice$ as they come: he cannot wait and read them later, individually or together. We assume that the channel used for the quantum transmission is free of errors, so that it is guaranteed that $r_{i}'=r_{i}$ whenever $\beta_{i}'=\beta_{i}$. we now show that under the assumption this protocol satisfies the statistical version of the above constraints. Correctness ----------- $\mathbf{Hoefding \quad inequality}$ [@[HO]] Let $X_1, X_2,\cdots, X_n$ be total independent random variables with identical probability distribution so that $E(X_i)=\mu$ and the range of $X_i$ is in $[a,b]$. Let the simple average $Y=(X_1+X_2+\cdots+X_n)/n$ and $\delta>0$, then $$Pr[\|Y-\mu\|\geq \delta]\leq 2\cdot e^{\frac{-2n\cdot \delta^2}{b-a}}$$ So, if $Pr[X_i=0]=Pr[X_i=1]=\frac{1}{2}$, then $\mu=\frac{1}{2}$ and $a=0,b=1$, we have the following inequality $$Pr[\|\sum _{i=1}^{n}\frac{X_i}{n}- \frac{1}{2} \|\geq \delta]\leq 2\cdot e^{-2\cdot n \delta^2}$$ We show that most of the time the output is correct if the parties abide to their prescribed protocol. In a given run of the protocol, $Bob$ will succeed in computing $b_{c_1},b_{c_2},\dots,b_{c_m}$ properly provided satisfying the following conditions :\ when $\frac{2m+1}{2n}< \frac{1}{2}$ $$\# \{i\| \beta _{i}= \beta _{i}'\}-x \geq (N-x)m/n$$ or when $\frac{2m+1}{2n}\geq \frac{1}{2}$ $$\# \{i\| \beta _{i}= \beta _{i}'\} \geq (N-x)m/n$$ Because in that case he can form $I_{c_1},I_{c_2},\dots,I_{c_m}$ as prescribed and then he can compute the output bit as $\widehat{b}_{c_i}\oplus \bigoplus _{j\in I_{c_i} }r_{j}'$ which is $$\widehat{b}_{c_i}\oplus \bigoplus _{j\in I_{c_i} }r_{j}' =b_{c_i}\oplus \bigoplus _{j\in J_{c_i} }r_{j} \bigoplus _{j\in I_{c_i} }r_{j}'=b_{c_i}\oplus \bigoplus _{j\in I_{c_i} }r_{j} \oplus r_{j}'$$ because $J_{c_i}$ is $I_{c_i}$. Since $\beta_{i}=\beta_{i}'\rightarrow r_{j} \oplus r_{j}'=0 $ makes all the right terms vanish, we end up with $$\widehat{b}_{c_i}\oplus \bigoplus _{j\in I_{c_i} }r_{j}' =b_{c_i}$$ Therefore the protocol gives the correct output unless satisfying the following conditions :\ when $\frac{2m+1}{2n}< \frac{1}{2}$ $$\# \{i\| \beta _{i}= \beta _{i}'\}-x < (N-x)m/n$$ or when $\frac{2m+1}{2n}\geq \frac{1}{2}$ $$\# \{i\| \beta _{i}= \beta _{i}'\} < (N-x)m/n$$ in which case $Bob$ is unable to form the set $I_{c_1},I_{c_2},\dots,I_{c_m}$ as prescribed. Now, we can calculate the probability that $Bob$ can not form $I_{c_1},I_{c_2},\dots,I_{c_m}$\ If $\frac{2m+1}{2n}< \frac{1}{2}$ (i.e. $2m+1< n$, $x=\frac{n-(2m+1)}{2n-(2m+1)}N$), then the probability that $Bob$ can get less than $m$ bits is given by $$\begin{aligned} & & P[\# \{i\| \beta _{i}= \beta _{i}'\}-x < (N-x)m/n]\\ &=& P[\# \{i\| \beta _{i}= \beta _{i}'\} < (N-x)m/n+x]\\ &=& P[\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' > N-((N-\frac{n-(2m+1)}{2n-(2m+1)}N)m/n\\ && +\frac{n-(2m+1)}{2n-(2m+1)}N)]\\ &=& P[\frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' > 1-\frac{n-(m+1)}{2n-(2m+1)}]\\ &=& P[\frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' > \frac{n-m}{2n-(2m+1)}]\\ &\leq & P[ \| \frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}'-\frac{1}{2}\|> \frac{n-m}{2n-(2m+1)}-\frac{1}{2} ]\end{aligned}$$ It is easy to check that $\frac{n-m}{2n-(2m+1)}-\frac{1}{2}>0$.\ Given that $P[\beta _{i} \oplus \beta _{i}'=1]=1/2$, let $N>\frac{\ln 2}{(\frac{n-m}{2n-(2m+1)}-\frac{1}{2})^2}$, this probability can be easily bounded by $$\begin{aligned} &<& 2\cdot e^{-2\cdot N(\frac{n-m}{2n-(2m+1)}-\frac{1}{2})^2}\\ &=& 2\cdot e^{- N(\frac{n-m}{2n-(2m+1)}-\frac{1}{2}) ^2} \cdot e^{- N(\frac{n-m}{2n-(2m+1)}-\frac{1}{2}) ^2} \\ &<& e^{- N(\frac{n-m}{2n-(2m+1)}-\frac{1}{2}) ^2}\\ &=&\varepsilon^N\end{aligned}$$ ($\varepsilon= e^{-(\frac{n-m}{2n-(2m+1)}-\frac{1}{2})^2}<1$) using Hoefding’s inequality.\ If $\frac{2m+1}{2n} \geq \frac{1}{2}$ (i.e. $2m+1\geq n$, $x=\frac{(2m+1)-n}{2m+1}$), then the probability that $Bob$ can get less than $m$ bits is given by $$\begin{aligned} & & P[\# \{i\| \beta _{i}= \beta _{i}'\} < (N-x)m/n]\\ &=& P[\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' > N-(N-\frac{(2m+1)-n}{2m+1}N)m/n]\\ &=& P[\frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' > 1-\frac{m}{2m+1}]\\ &\leq & P[ \| \frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}'-\frac{1}{2}\|>\frac{1}{2}- \frac{m}{2m+1} ]\end{aligned}$$ It is easy to check that $\frac{1}{2}- \frac{m}{2m+1}>0$.\ Given that $P[\beta _{i} \oplus \beta _{i}'=1]=1/2$, let $N>\frac{\ln 2}{(\frac{1}{2}-\frac{m}{2m+1}) ^2}$, this probability can be easily bounded by $$\begin{aligned} &<& 2\cdot e^{-2\cdot N ( \frac{1}{2}-\frac{m}{2m+1}) ^2 }\\ &=& 2\cdot e^{- N ( \frac{1}{2}-\frac{m}{2m+1}) ^2 }\cdot e^{- N ( \frac{1}{2}-\frac{m}{2m+1}) ^2 }\\ &<& e^{- N ( \frac{1}{2}-\frac{m}{2m+1}) ^2 } \\ &=&\varepsilon^N\end{aligned}$$ ($\varepsilon=e^{-(\frac{1}{2}-\frac{m}{2m+1})^2}<1$) using Hoefding’s inequality.\ So, $Bob$ can get less than $m$ bits that sent from $Alice$ with probability less than $\varepsilon^N$. Privacy ------- We analyse the privacy of each party individually as if he or she is facing a malicious opponent. ### Privacy for $Bob$ $Alice$ can not find out much about $c_1,c_2,\dots,c_m$, Proof. The only things $Alice$ gets though the protocol are the sets $J_1,J_2,\dots,J_n$. $\beta_{i}$’s and $\beta_{i}'$’s are independent from each other. $J_1,J_2,\dots,J_n$ will have uniform distribution over all possible pairs of disjoint subsets of size $\frac{N-x}{n}$ for $i=1,i=2,\dots$ as well as for $i=n$. Therefore $Alice$ learns nothing about the $c_1,c_2,\dots,c_m$. $\hfill \Box$ ### Privacy for $Alice$ Except with probability at most $\epsilon^{n}$, $Bob$ can not find out much information about more that $m$ of $b_1,b_2,\dots,b_n$. Proof. The probability of that $Bob$ gets more than $m$ bits (i.e. get at least $m+1$ bits). So\ \ If $\frac{2m+1}{2n}<\frac{1}{2}$ (i.e. $2m+1<n$, $x=\frac{n-(2m+1)}{2n-(2m+1)}N$), the probability that $Bob$ can get more than $m+1$ bits is given by $$\begin{aligned} && P[\# \{i\| \beta _{i}= \beta _{i}'\} - x\geq (N-x)(m+1)/n]\\ &=& P[\# \{i\| \beta _{i}= \beta _{i}'\} \geq (N-x)(m+1)/n+x]\\ &=& P[\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' \leq N-((N-\frac{n-(2m+1)}{2n-(2m+1)}N)(m\\ && +1)/n+\frac{n-(2m+1)}{2n-(2m+1)}N)]\\ &=& P[\frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' \leq 1-\frac{n-m}{2n-(2m+1)}]\\ &\leq & P[ \| \frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}'-\frac{1}{2}\|>\frac{1}{2}- \frac{n-m}{2n-(2m+1)} ]\end{aligned}$$ It is easy to check that $\frac{1}{2}- \frac{n-m}{2n-(2m+1)}>0$.\ Given that $P[\beta _{i} \oplus \beta _{i}'=1]=1/2$, let $N>\frac{\ln 2}{(\frac{1}{2}- \frac{n-m}{2n-(2m+1)}) ^2}$, this probability can be easily bounded by $$\begin{aligned} &<& 2\cdot e^{-2\cdot N ( \frac{1}{2}- \frac{n-m}{2n-(2m+1)}) ^2 }\\ &=& 2\cdot e^{-N ( \frac{1}{2}- \frac{n-m}{2n-(2m+1)}) ^2 }\cdot e^{- N ( \frac{1}{2}- \frac{n-m}{2n-(2m+1)}) ^2 }\\ &<& e^{- N ( \frac{1}{2}- \frac{n-m}{2n-(2m+1)}) ^2}\\ &=&\varepsilon^N\end{aligned}$$ ($\varepsilon=e^{-(\frac{1}{2}- \frac{n-m}{2n-(2m+1)})^2}<1$) using Hoefding’s inequality.\ \ If $\frac{2m+1}{2n}\geq \frac{1}{2}$ (i.e. $2m+1\geq n$, $x=\frac{(2m+1)-n}{2m+1}$), then the probability that $Bob$ can get more than $m+1$ bits is given by $$\begin{aligned} && P[\# \{i\| \beta _{i}= \beta _{i}'\} \geq (N-x)(m+1)/n]\\ &=& P[\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' \leq N-(N-\frac{(2m+1)-n}{2m+1}N)(m\\ &&+1)/n]\\ &=& P[\frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}' \leq 1-\frac{m+1}{2m+1}]\\ &\leq & P[ \| \frac{1}{N}\sum_{i=1}^{N} \beta _{i} \oplus \beta _{i}'-\frac{1}{2}\|> \frac{m+1}{2m+1}-\frac{1}{2} ]\end{aligned}$$ It is easy to check that $\frac{m+1}{2m+1}-\frac{1}{2}>0$.\ Given that $P[\beta _{i} \oplus \beta _{i}'=1]=1/2$, let $N>\frac{\ln 2}{(\frac{m+1}{2m+1}-\frac{1}{2}) ^2}$, the probability can be easily bounded by $$\begin{aligned} &<& 2\cdot e^{-2\cdot N ( \frac{m+1}{2m+1}-\frac{1}{2}) ^2 }\\ &=& 2\cdot e^{-N ( \frac{m+1}{2m+1}-\frac{1}{2}) ^2 }\cdot e^{- N ( \frac{m+1}{2m+1}-\frac{1}{2}) ^2 }\\ &<& e^{- N ( \frac{m+1}{2m+1}-\frac{1}{2}) ^2 }\\ &=&\varepsilon^N\end{aligned}$$ ($\varepsilon=e^{-(\frac{m+1}{2m+1}-\frac{1}{2})^2}<1$) using Hoefding’s inequality. Finally, we show that $Bob$ cannot get more than $m$ bits by attacking the weak quantum bit commitment. Let the probability that he can cheat $Alice$ in the weak QBC be $p$ ($0<p<1$), the probability that he can get one more bit is $p^{\frac{N-x}{n}}<\epsilon^N$ ($\epsilon=p^{\frac{1}{2n}}$). So, $Bob$ can get more than $m$ bits that sent from $Alice$ with probability less than $\varepsilon^N$. $\hfill \Box$ In the 1-out-of-2 OT scheme, $n=2$ and $m=1$, $\frac{2m+1}{2n}=\frac{3}{4}>\frac{1}{2}$, then the probability is less than $$2\cdot e^{-N \cdot 2( \frac{m+1}{2m+1}-\frac{1}{2}) ^2 }=2\cdot e^{-N \cdot 2( \frac{2}{3}-\frac{1}{2}) ^2 }=2\cdot e^{-\frac{N}{18} }$$ Conclusions and Future Work =========================== In this paper, we construct an quantum $m$-out-of-$n$ OT based on the transmission of polarized light, which is an extension of the quantum 1-out-f-2 OT, and prove that this scheme satisfies statistical correctness and statistical privacy, i.e. except with a small probability $\epsilon^N$, $Bob$ can get the correct $m$ bits, and cannot get one more bit than required. We think the following points is interesting for further research: 1. Implement and apply the QOT in the real world. 2. Find a QOT satisfies perfect correctness and perfect privacy. [99]{} Bennett, C.H. and Brassard, G., “Quantum Cryptography: Public-key Distribution and Coin Tossing“, In Proceedings of the International Conference on Computers, Systems and Signal Processing, Bangalore, India, December 1984, pp. 175-179. Brassard, G., Crépeau, C. Jozsa, R. and Langlois, D., “A Quantum Bit Commitment Scheme Probably unbreakable by both parties“, In Proceedings of the 34th Annual IEEE Symposium on Foundations of Computer Science, November 1993, pp.362-371 Claude Crépeau. “Quantum Oblivious Transfer“. Journal of Modern Optics, 41(12):2455¨C2466, 1994. Dorit Aharonov, Amnon Ta-Shma, Umesh V. Vazirani, Andrew Chi-Chih Yao. “Quantum bit escrow“. Proceedings of the 32nd Annual ACM Symposium on Theory of Computing(STOC’00), 2000. Even, S., Goldreich, O. and Lempel, A., “A Randomized Protocol for Signing Contracts“, Communications of the ACM, vol. 28, pp. 637-647, 1985. W. Hoefding, “Probability Inequalities for Sums of Bounded Random Variables“, Journal of the American Statistical Association, Vol.58, 1936, pp.13-30 Mayers, D. “Unconditionally Secure Quantum Bit Commitment is Impossible“. Physical Review Letters 78 . pp 3414-3417 (28 April 1997). Moni Naor, Benny Pinkas. “Efficient Oblivious Transfer Protocols“. SODA, 2001 P. W. Shor, “Polynomial-Time Algorithms for Prime Factorization and Discrete Logarithms on a Quantum Computer“, SIAM Journal on Computing, V.26:(5), 1997. Rabin, M.O., “How to exchange secrets by Oblivious Transfer“, technical report TR-81, Aiken Computation Laboratory, Harvard University, 1981. Wiesner, S., “Conjugate coding“, Sigact News, vol.15, no. 1, 1983, pp.78-88; Manuscript written circa 1970, unpublished until it appeared in SIGACT News. Yi Mu, Junqi Zhang, Vijay Varadharajan, “m out of n oblivious transfer,” ACISP 2002, Lecture Notes in Computer Science 2384, Springer Verlag, 2002. pp. 395-405 [^1]: This work is partially supported by a grant from the Ministry of Science and Technology (\#2001CCA03000), National Natural Science Fund (\#60273045) and Shanghai Science and Technology Development Fund (\#03JC14014).	{ "pile_set_name": "ArXiv" }
--- abstract: 'We prove distance bounds for graphs possessing positive Bakry-Émery curvature apart from an exceptional set, where the curvature is allowed to be non-positive. If the set of non-positively curved vertices is finite, then the graph admits an explicit upper bound for the diameter. Otherwise, the graph is a subset of the tubular neighborhood with an explicit radius around the non-positively curved vertices. Those results seem to be the first assuming non-constant Bakry-Émery curvature assumptions on graphs.' author: - 'Shiping Liu, Florentin Münch, Norbert Peyerimhoff, Christian Rose' title: 'Distance bounds for graphs with some negative Bakry-Émery curvature' --- Introduction ============ In Riemannian geometry, diameter bounds for complete connected Riemannian manifolds are well established under several curvature assumptions. The well known Bonnet-Myers theorem states that if the Ricci curvature of a manifold is larger than a positive threshold, the diameter of the manifold is finite, and, therefore, the manifold itself compact, see [@Jost-08; @Petersen-16] and the references therein. In particular, the classical Jacobi field technique used there provides also a sharp upper estimate for the diameter. Later on, this result was generalized in [@PetersenSprouse-98]. There, the authors assumed that the amount of the Ricci curvature of the manifold $M$ below a positive level is locally uniformly $L^p$-small for some $p>\dim M/2$, and obtain indeed a diameter bound depending on this kind of smallness of the curvature.\ The concept of Ricci curvature was transferred into various settings. Let us provide a brief summary of the history. Already in 1985, Bakry and Émery introduced Ricci curvature on diffusion semigroups via the highly generalizable $\Gamma$-calculus [@bakry1985diffusions] derived from the Bochner formula. This approach has first been applied to a discrete setting in [@elworthy91] and diversely used in [@schmuckenschlager1998curvature; @bauer2015li; @horn2014volume; @munch2017remarks; @munch2014li; @lin2010ricci; @fathi2015curvature; @gong2015properties; @hua2017stochastic; @LiuMuenchP-16; @chung2014harnack]. The theory of local metric measure spaces has also benefitted from the Bakry-Émery approach. For more information about Ricci-curvature on metric measure spaces, see [@ambrosio2014metric; @erbar2015equivalence; @lott2009ricci; @sturm2006geometry]. A concept of Ricci curvature on graphs via optimal transport has been introduced by Ollivier [@ollivier2009ricci] and applied in [@lin2010ricci; @lin2011ricci; @bauer2011ollivier; @jost2014ollivier]. Recently, Erbar, Maas and Mielke introduced a Ricci curvature on graphs via convexity of the entropy [@erbar2012ricci; @fathi2016entropic; @erbar2017ricci; @mielke2013geodesic]. In a highly celebrated paper, Erbar, Kuwada and Sturm proved that on metric measure spaces, the concepts of Ricci curvature via $\Gamma$-calculus (Bakry-Émery) and optimal transport and entropy (Lott-Sturm-Villani) coincide [@erbar2015equivalence]. On the other hand, in the setting of graphs, Bakry-[É]{}mery Ricci curvature and Ollivier Ricci curvature are often quite different and there are many open questions about the relations between these curvature notions. It is now natural to ask for analogues and generalizations of the diameter bounds for manifolds above to contexts in which concepts of Ricci curvature exist. For metric measure spaces, there have been attempts to generalize the Bonnet-Myers theorem to variable Ricci curvature bounds in an integral sense, see [@Ketterer-15] and the references therein. For connected graphs $G=(V,E)$, the authors of [@LiuMuenchP-16] show a sharp diameter bound assuming positive Bakry-Émery curvature in the $CD(K,N)$-setting for $N\in (0,\infty]$, notions of curvature we will introduce below. For convenience, we recall the result for further reference. \[thm:LMP\] Let $G=(V,w,m)$ be a graph. 1. Assume that $CD(K,\infty)$ holds for $K>0$ and the graph admits an upper bound $\operatorname{Deg}_{\max}$ for the weighted vertex degree. Then, we have $$\operatorname{\mathop{diam}}_d(G)\leq \frac{2\operatorname{Deg}_{\max}}K,$$ where $\operatorname{\mathop{diam}}_d$ is the diameter of $G$ with respect to the combinatorial distance. 2. Assume that $CD(K,N)$ holds for $K,N>0$. Suppose that $G$ is complete in the sense of [@hua2017stochastic] and satisfies $\inf_{x\in V} m(x) >0$. Then, we have $$\operatorname{\mathop{diam}}_\sigma(G)\leq \pi\sqrt{\frac{N}{K}},$$ where $\sigma$ is the resistance metric defined below. In this article, we generalize the above discrete Bonnet-Myers theorem to the situation where the graph is positively curved except on a vertex set $V_0$, where the curvature is allowed to be non-positive. The main result below states that a graph is always covered by the tubular neighborhood around the negatively curved vertices of an explicit radius depending on local curvature dimension assumptions, which are given pointwise by the Bochner formula shown below. This description of the curvature involves the Laplacian of the space considered. The idea is to compare the different curvature values on the sets $V_0$ and $V\setminus V_0$ via the semigroups associated to different Laplacians. On one hand, we have a graph of constant positive curvature, the lower curvature bound of $V\setminus V_0$, and a graph of constant negative curvature, the lower curvature bound of $V_0$. Those lead to different Laplacians and therefore to different semigroups, which have to be controlled in a manner such that the diameter of the whole graph can be bounded above. After we introduced the neccessary framework and the main result in the section below, we show several preparatory estimates of the semigroup depending on the set of negatively curved vertices and refine the analysis of the techniques developed in [@LiuMuenchP-16]. Setting and main result ======================= Let $G=(V,w,m)$ be a weighted, connected, locally finite graph. That is, on the vertex set $V\not=\emptyset$, we introduce a symmetric map $$w\colon V\times V\to [0,\infty),\quad (x,y)\mapsto w_{xy}=w_{yx}$$ and $$m\colon V\to (0,\infty),\quad x\mapsto m_x.$$ If $x,y$ are two vertices with $w_{xy}>0$, we say they are neighbors, or they are connected by an edge, and write $x\sim y$. We say $G$ is locally finite if each vertex has finitely many neighbors. The maps $w$ and $m$ introduced above represent the edge measure and the vertex measure of $G$, respectively. For any two vertices of a connected graph, there is a path connecting them. The graph distance $d\colon V\times V\to [0,\infty]$ is given by the number of edges in a shortest path between two vertices. The diameter $\operatorname{\mathop{diam}}(V')$ of a set $V'\subset V$ is the maximum graph distance between any two vertices in $V'$. By $T_r(V')$ we denote the tubular neighborhood of $V'\subset V$ of radius $r>0$. If $V'=\{x\}$ for some $x\in V$, then $B_r(x):=T_r(V')$, the ball around $x$ with radius $r>0$. As usual, the weighted degree of a vertex $x\in V$ is given by $$\operatorname{Deg}(x)=\frac 1{m(x)}\sum_{y\in V}w_{xy}.$$ We say that $G$ has bounded vertex degree if there exists $\operatorname{Deg}_{\max} < \infty$ with $\operatorname{Deg}(x) \le \operatorname{Deg}_{\max}$ for all vertices $x \in V$. Denote by ${\mathcal{C}}_c(V)$ the set of finitely supported functions on $V$ and $\Vert\cdot\Vert_\infty$ the maximum norm. The Laplacian on functions $f\in{\mathcal{C}}_c(V)$ is defined by $$\Delta f(x)=\frac{1}{m(x)}\sum_{y\in V}w_{xy}(f(y)-f(x)),\quad x\in V.$$ If $m(x)=\sum_{y\in V}w_{xy}$ for any $x\in V$, the associated Laplacian is called the normalized Laplacian. If $m(x)=1$ for any $x\in V$, the Laplacian is called combinatorial or physical. The definition of the Laplacian leads to the so-called carré du champ operator $\Gamma$: for all $f,g\in {\mathcal{C}}_c(V)$, $x \in V$: $$\begin{aligned} \Gamma(f,g)(x)&=&\frac 12(\Delta(fg)-f\Delta g-g\Delta f)(x) \\ &=& \frac{1}{2m(x)} \sum_{y \in V} \omega_{xy} (f(y)-f(x))(g(y)-g(x)).\end{aligned}$$ For simplicity, we always write $\Gamma(f):=\Gamma(f,f)$. Iterating $\Gamma$, we can define another form $\Gamma_2$, which is given by $$\Gamma_2(f,g)(x)=\frac 12(\Delta\Gamma(f,g)-\Gamma(f,\Delta g)-\Gamma(g,\Delta f))(x), \quad x\in V, f,g\in {\mathcal{C}}_c(V).$$ We abbreviate $\Gamma_2(f)=\Gamma_2(f,f)$.\ As mentioned before, the graph distance is defined by shortest paths between two points. In contrast, we can define another kind of metric coming from the operator $\Gamma$. Let $G=(V,w,m)$ be a graph. (i) A metric $\rho$ on $V$ is called intrinsic if for all $x \in V$, $$\begin{aligned} \left\\|\Gamma \rho(x,\cdot) \right\\|_\infty \leq 1. \end{aligned}$$ (ii) For an intrinsic metric $\rho$, the jump size of $\rho$ is given by $$R_\rho:=\sup\{\rho(x,y)\mid x\sim y\}.$$ (iii) The resistance metric $\sigma$ on $V$ is given by $$\sigma\colon V\times V\to[0.\infty],\quad (x,y)\mapsto\sup\{f(y)-f(x)\mid\Vert\Gamma f\Vert_\infty\leq 1\}.$$ As in the case of the graph distance, if $r$ is an intrinsic or the resistance metric, we define $\operatorname{\mathop{diam}}_r(V')$ for a subset $V'\subset V$ to be the diameter of $V'$ with respect to $r$, and $T^r_R(V')$ denotes the tubular neighborhood of $V'$ of radius $R$ with respect to $r$, etc. Intrinsic metrics have already been used to solve various problems on graphs, see, e.g., [@BHK-13; @BKW-15; @Fol-11; @Fol-14; @GHM-12; @HKMW-13; @HK-14]. A natural intrinsic metric on a graph was introduced in [@Huang-11 Definition 1.6.4] (see also [@hua2017stochastic Example 2.9]): $$\rho(x,y)=\inf_{\gamma} \sum_{i=0}^{n-1}(\operatorname{Deg}(x_i)\vee\operatorname{Deg}(x_{i+1}))^{-1/2},\quad x,y\in V,$$ where $\gamma$ is a path $x=x_0\sim x_1\sim\ldots\sim x_{n-1}\sim x_n=y$, and $a\vee b:=\max\{a,b\}$ for $a,b\in \mathbb{R}$. \[remark:intrinsic\] (i) All metrics smaller than an intrinsic metric are intrinsic, too. In general, the resistance metric $\sigma$ is not intrinsic, but is greater than all intrinsic metrics. (ii) The properties of the resistance metric rely on the properties of the underlying Laplacian. It is shown in Proposition \[prop:DistanceAndDimensionBoundedDegree\] that if $\operatorname{Deg}(x)\leq \operatorname{Deg}_{\max}<\infty$ for all $x\in V$, we have that $\rho$ is intrinsic with $$\rho(x,y) := \sqrt{\frac 2 {\operatorname{Deg}_{\max}}} d(x,y).$$ (iii) If $\rho$ is intrinsic and all $\rho$-balls are finite, then $G$ is complete [@hua2017stochastic Theorem 2.8]. For the reader’s convenience, we recall that a graph $G=(V,w,m)$ is complete in the sense of [@hua2017stochastic] if there exists a nondecreasing sequence of finitely supported functions $\{\eta_k\}_{k=1}^\infty$ such that $\lim_{k\to \infty}\eta_k=1_V$ and $\Gamma(\eta_k)\leq 1/k$, where $1_V$ is the constant function $1$ on $V$. The operator $\Gamma$ not only leads to a definition of a metric, but also to the curvature conditions in the sense of Bakry-Émery. Let $K\in {\mathbb{R}}$ and $N\in (0,\infty]$. (i) Define the pointwise curvature dimension condition $CD(K,N,x)$ for $x\in V$ by $$\Gamma_2(f)(x)\geq K\Gamma(f)(x)+\frac 1N(\Delta f)^2(x),\,\,\,\text{for any }\,\,\,f: V\to \mathbb{R}.$$ (ii) The curvature dimension condition $CD(K,N)$ holds iff $CD(K,N,x)$ holds for any $x\in V$. (iii) For any $x\in V$, we define $$K_{G,x}(N):=\sup\{K\in{\mathbb{R}}\mid CD(K,N,x)\}.$$ We will need different assumptions to guarantee the semigroup characterization of Bakry-Émery curvature (see [@hua2017stochastic; @gong2015properties]). These assumptions are satisfied whenever the vertex measure $m$ is non-degenerate, that is, $$\inf_{x\in V}m(x)>0,$$ and all balls with respect to an intrinsic metric are finite. In the case of bounded vertex degree $\operatorname{Deg}(x) \le \operatorname{Deg}_{\max}$ for all $x \in V$, the non-degenerate vertex measure condition can usually be dropped. In case of bounded vertex degree, the combinatorial distance is intrinsic up to a constant. Furthermore, we have a uniform control of the dimension in terms of the curvature. \[prop:DistanceAndDimensionBoundedDegree\] Let $G=(V,w,m)$ be a graph with bounded degree $\operatorname{Deg}(x) \leq \operatorname{Deg}_{\max}$. Then, $\rho:=d \sqrt{\frac {2} {\operatorname{Deg}_{\max}}}$ is an intrinsic metric. Furthermore if $G$ satisfies $CD(K,\infty)$, it also satisfies $CD(K-s, \frac{2\operatorname{Deg}_{\max}}s)$ for all $s>0$. Let $x,x_0 \in V$ and let $g:=\rho(x_0,\cdot)$ We have $$\begin{aligned} \Gamma g(x) &= \frac 1 {2m(x)}\sum_{y} w(x,y) (g(x)-g(y))^2\\ &\leq \frac 1 {2m(x)}\sum_{y} w(x,y) \sqrt{\frac {2} {\operatorname{Deg}_{\max}}}^2\\ &\leq\frac {2} {\operatorname{Deg}_{\max}} \cdot \frac {\operatorname{Deg}_{\max}}{2}\\ &=1 \end{aligned}$$ which shows that $\rho$ is an intrinsic metric. Furthermore, $CD(K,\infty)$ implies for all $f$, $$\begin{aligned} \Gamma_2 f \geq (K-s) \Gamma f + s \Gamma f \geq (K-s) \Gamma f + \frac s {2 \operatorname{Deg}_{\max}} (\Delta f)^2 \end{aligned}$$ where the latter inequality follows from Cauchy-Schwarz. Hence, $G$ satisfies the condition $CD(K-s, \frac{2\operatorname{Deg}_{\max}}s)$ as claimed. The main theorem stated below extends Theorem \[thm:LMP\] to the case of negatively curved vertices. \[thm:main\] Let $G=(V,w,m)$ be a weighted, complete graph with non-degenerate vertex measure $m$, let $N\in(0,\infty]$, and $\rho$ an intrinsic metric on $G$. Define $$V_0:=\{x\in V\mid K_{G,x}(N)\leq 0\}.$$ (i) If $N=\infty$, $V_0=\emptyset$, $\operatorname{Deg}(x)\leq \operatorname{Deg}_{\max}<\infty$ for any $x\in V$, and $CD(K_0,N)$ for $K_0>0$, then $$\operatorname{\mathop{diam}}_\rho(G)\leq \frac {2\sqrt{2\operatorname{Deg}_{\max}}}{K_0}.$$ (ii) If $N<\infty$, $V_0=\emptyset$, and $CD(K_0,N)$ for $K_0>0$, then $$\operatorname{\mathop{diam}}_\rho(G)\leq \pi\sqrt{\frac N{K_0}}.$$ (iii) If $N=\infty$, $V_0\not=\emptyset$, $\operatorname{Deg}(x)\leq \operatorname{Deg}_{\mathrm{max}}<\infty$ for all $x\in V$, and assuming, for $K,K_0>0$, $$\begin{aligned} CD(-K_0,N,x)\quad\forall\, x\in V\text{ and }CD(K,N,x)\quad \forall \, x\in V\setminus V_0, \end{aligned}$$ then $$V\subset T_{R}(V_0),\quad R:= 1 + 18.2\sqrt{2} \, e^{4K_0/K}\,\frac{ \operatorname{Deg}_{\max}}{\sqrt{KK_0}}.$$ (iv) If $N<\infty$, $V_0\not=\emptyset$, and assuming, for $K,K_0>0$, $$\begin{aligned} CD(-K_0,N,x)\quad\forall\, x\in V\text{ and }CD(K,N,x)\quad \forall \, x\in V\setminus V_0, \end{aligned}$$ then $$V\subset T^\rho_{R}(V_0),\quad R:=R_\rho+18.2{\mathrm{e}}^{2K_0/K}\sqrt{\frac{N}{K+K_0}}.$$ Note that (i) and (ii) in the above theorem are included in [@LiuMuenchP-16] since every intrinsic metric is dominated by the resistance metric and, therefore, $$\operatorname{\mathop{diam}}_\rho(G)\leq \operatorname{\mathop{diam}}_\sigma(G).$$ We also point out that any locally finite graph with $\operatorname{Deg}_{\max}<\infty$ is complete by Proposition \[prop:DistanceAndDimensionBoundedDegree\] and Remark \[remark:intrinsic\] (iii). CD conditions and semigroups ============================ By the spectral calculus, we can associate to $\Delta$ the heat semigroup $(P_t)_{t\geq 0}$. Using a standard argument, we derive a commutation formula for the semigroup and the gradient depending on the set of negatively curved vertices. \[gradientsg\] Let $G=(V,w,m)$ be a weighted, complete graph with non-degenerate vertex measure $m$, and $N\in(0,\infty]$. Define $$V_0:=\{x\in V\mid K_{G,x}(N)\leq 0\}.$$ Let $K>0,K_0\geq 0$ such that $$CD(-K_0,N,x)\quad \forall \, x\in V\quad \text{and}\quad CD(K,N,x)\quad\forall \, x\in V\setminus V_0.$$ Then for any bounded function $f\colon V\to {\mathbb{R}}$ with bounded $\Gamma f$, $$\begin{aligned} \label{prop:sgc} \Gamma(P_Tf)(x)\leq &{\mathrm{e}}^{-2KT}P_T(\Gamma f)(x)-\frac{1-e^{-2KT}}{KN}(\Delta P_Tf)^2(x)\notag\\ &+2(K_0+K)\Vert\Gamma f\Vert_\infty{\mathrm{e}}^{2K_0T}\int_0^T{\mathrm{e}}^{-2(K+K_0)s}P_s1_{V_0}(x){\mathrm{d}}s. \end{aligned}$$ Let $G=(V,w,m)$ be a weighted, complete graph with non-degenerate vertex measure $m$. If $G$ satisfies $CD(K,N)$, $K\in {\mathbb{R}}, N\in (0,\infty]$, it was shown in [@LiuMuenchP-16 Lemma 2.3] that for any bounded function $f\colon V\to {\mathbb{R}}$ with bounded $\Gamma f$, $$\label{eq:LMP16} \Gamma P_tf(x)\leq {\mathrm{e}}^{-2Kt}P_t\Gamma f(x)-\frac{1-{\mathrm{e}}^{-2Kt}}{KN}(\Delta P_tf)^2(x).$$ So the estimate (\[prop:sgc\]) is a refinement of (\[eq:LMP16\]) in the setting of Proposition \[gradientsg\]. As in the classical Ledoux-ansatz, for a bounded function $f\colon V\to {\mathbb{R}}$ and $T>0$, let $$F(s)={\mathrm{e}}^{-2Ks}P_s\Gamma P_{T-s}f(x)$$ and compute $$\begin{aligned} F'(s)&={\mathrm{e}}^{-2Ks}\left(-2KP_s\Gamma P_{T-s}f(x)+\Delta P_s\Gamma P_{T-s}f(x)+2P_s\Gamma(P_{T-s}f,-\Delta P_{T-s}f)(x)\right)\\ &={\mathrm{e}}^{-2Ks}P_s\left(-2K\Gamma P_{T-s}f+\Delta \Gamma P_{T-s}f-2\Gamma(P_{T-s}f,\Delta P_{T-s}f)\right)(x)\\ &={\mathrm{e}}^{-2Ks}P_s\left(2\Gamma_2(P_{T-s}f)-2K\Gamma(P_{T-s}f)\right)(x). \end{aligned}$$ It is well known that the heat semigroup is generated by a smooth integral kernel, which is called the heat kernel. In particular, it can be proved that there is a pointwise minimal version, called $p: (0,\infty) \times V \times V \to {\mathbb{R}}$, obtained via an exhaustion procedure by Dirichlet heat kernels on compact ($=$ finite) subsets of $V$ (see, e.g., [@linliu2015; @aweber2010]).Therefore, we get $$\begin{aligned} F'(s) &=2{\mathrm{e}}^{-2Ks}\sum_{y\in V}p(s,x,y)\left[\Gamma_2(P_{T-s}f)(y)-K\Gamma(P_{T-s}f)(y)\right]\\ &=2{\mathrm{e}}^{-2Ks}\left[\sum_{y\in V\setminus V_0}p(s,x,y)\left[\Gamma_2(P_{T-s}f)(y)-K\Gamma(P_{T-s}f)(y)\right]\right.\\ &\quad\left.+\sum_{y\in V_0}p(s,x,y)\left[\Gamma_2(P_{T-s}f)(y)+K_0\Gamma(P_{T-s}f)(y)\right]\right.\\ &\quad\left.-(K+K_0)\sum_{y\in V_0}p(s,x,y)\Gamma(P_{T-s}f)(y)\right]. \end{aligned}$$ Applying (\[eq:LMP16\]) to the last term above and applying the pointwise curvature dimension conditions to the first two terms, we have $$\begin{aligned} F'(s) &\geq 2{\mathrm{e}}^{-2Ks}\left[\frac 1NP_s(\Delta P_{T-s}f)^2(x)\right.\\ &\quad\left.-(K_0+K)\sum_{y\in V_0}p(s,x,y)\left({\mathrm{e}}^{2K_0(T-s)}P_{T-s}\Gamma f(y) +\frac{1-{\mathrm{e}}^{2K_0(T-s)}}{K_0N}(\Delta P_{T-s}f)^2(y)\right)\right]\\ &= -2(K+K_0){\mathrm{e}}^{2K_0T-2(K+K_0)s}\sum_{y\in V_0}P_{T-s}(\Gamma f)(y)p(s,x,y)\\ &\quad +\frac{2{\mathrm{e}}^{-2Ks}}{N}\left(P_s(\Delta P_{T-s}f)^2(x)+(K+K_0)\sum_{y\in V_0}p(s,x,y)\frac{{\mathrm{e}}^{2K_0(T-s)}-1}{K_0}(\Delta P_{T-s}f)^2(y)\right).\end{aligned}$$ Jensen’s inequality gives $$P_s(\Delta P_{T-s}f)^2(x)\geq (P_s\Delta P_{T-s}f)^2(x)=(\Delta P_Tf)^2(x).$$ Hence we have $$\begin{aligned} F'(s) &\geq -2(K+K_0){\mathrm{e}}^{2K_0T-2(K+K_0)s}\sum_{y\in V_0}P_{T-s}(\Gamma f)(y)p(s,x,y)+\frac{2{\mathrm{e}}^{-2Ks}}{N}(\Delta P_Tf)^2(x)\\ &\geq -2(K+K_0){\mathrm{e}}^{2K_0T-2(K+K_0)s}\Vert\Gamma f\Vert_\infty P_s1_{V_0}(x)+\frac{2{\mathrm{e}}^{-2Ks}}{N}(\Delta P_Tf)^2(x).\end{aligned}$$ Therefore, we have $$\begin{aligned} F(T)-F(0)&={\mathrm{e}}^{-2KT}P_T(\Gamma f)(x)-\Gamma(P_Tf)(x)\\ &=\int_0^TF'(s){\mathrm{d}}s \\ &\geq -2(K+K_0){\mathrm{e}}^{2K_0T}\Vert\Gamma f\Vert_\infty\int_0^T{\mathrm{e}}^{-2(K+K_0)s} P_s1_{V_0}(x){\mathrm{d}}s\\ &\quad +\int_0^T\frac{2{\mathrm{e}}^{-2Ks}}{N}{\mathrm{d}}s(\Delta P_Tf)^2(x).\end{aligned}$$ Rearranging yields the claim. To control the distance to the negatively curved part $V_0$, we need to estimate $P_t 1_{V_0} (x)$ in terms of $\rho(x,V_0)$. This is given by the following theorem. \[thm:Pt1\] Let $G=(V,w,m)$ be a weighted, complete graph with non-degenerate vertex measure satisfying $CD(-K_0,N)$ for some $K_0 \geq 0$ and some $N>0$, let $\rho$ be an intrinsic metric, $x \in V$, and $W \subset V$ with $\rho(x,W) \geq R$. Then, $$\begin{aligned} P_t 1_W (x) \leq \frac{\sqrt N}{R} \left(t\sqrt {K_0} + \sqrt{2t} \right) \label{eq:Pt1W leq}.\end{aligned}$$ From (\[eq:LMP16\]), we have for any bounded function $g$ with bounded $\Gamma g$, $$\frac{e^{2{K_0}s} - 1}{{K_0}N}\left(\Delta P_s g \right) ^2 \leq e^{2{K_0}s} P_s \Gamma g.$$ Hence, $$\begin{aligned} \left\|\Delta P_s g \right\| &\leq \sqrt{\frac{{K_0}N}{1 - e^{-2{K_0}s}}} \sqrt{P_s\Gamma g} \leq \sqrt{\frac{{K_0}N}{1 - \frac{1}{1+2{K_0}s}}} \sqrt{P_s\Gamma g} = \sqrt{{K_0} + \frac{1}{2s}} \sqrt{N P_s\Gamma g} \nonumber\\&\leq \left(\sqrt {K_0} + \frac{1}{\sqrt{2s}} \right)\sqrt{N P_s\Gamma g}. \label{eq: LPsg leq PsGammag}\end{aligned}$$ Let $g:= \left(1- \frac{\rho(x,\cdot)}{R} \right)_+$. Then, $\Gamma g \leq 1/R^2$ and $g + 1_W \leq 1$ and thus by (\[eq: LPsg leq PsGammag\]), $$\begin{aligned} P_t 1_W (x) &\leq 1-P_t g(x) \leq \int_0^t \|\Delta P_s g(x)\| ds \\ &\leq \int_0^t \left(\sqrt {K_0} +\frac{1}{\sqrt{2s}} \right)\sqrt{N P_s\Gamma g(x)} ds \\& \leq \frac{\sqrt N}{R} \left(t\sqrt {K_0} + \sqrt{2t} \right).\end{aligned}$$ This finishes the proof. We show that a Bonnet-Myers type diameter bound still holds if one allows some negative curvature. In contrast to Bonnet-Myers, we will bound the distance to the negatively curved part $V_0$ of the graph from above, which proves part (iv) of Theorem \[thm:main\]. \[thm:TubeNegativeCurvatureFiniteDimension\] Let $G=(V,w,m)$ be a connected graph with non-degenerate vertex measure $m$, $K,K_0 >0$. Let $\emptyset \neq V_0 \subset V$. Suppose $G$ satisfies $$CD(K,N,x)\quad\forall \, x\in V \setminus V_0\quad\text{and}\quad CD(-K_0,N,x)\quad \forall \,x\in V_0.$$ Let $\rho$ be an intrinsic metric with finite jump size $R_\rho>0$. Suppose $G$ is complete. Then for all $x_0 \in V$, one has $$\begin{aligned} \rho(x_0,V_0) \leq R_\rho + 18.2e^{2K_0/K} \sqrt{\frac{N}{K_0+K}}. \end{aligned}$$ In case of bounded vertex degree, we can drop the non-degeneracy assumption of $m$. By Theorem \[thm:Pt1\], we have $P_s1_{V_0} \leq \frac{\sqrt N}{\rho(V_0,\cdot)} \left(s\sqrt {K_0} + \sqrt{2s} \right)$. Thus, Proposition \[gradientsg\] implies that we have for any bounded function $f: V\to {\mathbb{R}}$ with bounded $\Gamma f$, $$\begin{aligned} e^{-2KT} P_T \Gamma f - \Gamma P_T f &\geq (\Delta P_T f)^2 \frac{1-e^{-2KT}}{KN} - \frac {\sqrt{N}}{\rho(V_0,\cdot)} H(K,K_0,T)\Vert\Gamma f\Vert_\infty \label{eq:integrate F'} \end{aligned}$$ with $$H(K,K_0,T) := 2(K_0+K)e^{2K_0 T} \int_0^T e^{-2(K_0+K)s}\left(s\sqrt {K_0} + \sqrt{2s} \right) ds.$$ On $V \setminus T^\rho_R(V_0) $, we have $\rho(V_0,\cdot) \geq R$. Activating (\[eq:integrate F’\]) towards $\|\Delta P_T f\|=\|\partial_T P_T f\|$ and throwing away the nonnegative term $\Gamma P_T f$ yields $$\begin{aligned} \|\partial_T P_T f\| \leq \sqrt{\left(e^{-2KT} + \frac{\sqrt{N}}{R}H(K,K_0,T)\right)\Vert\Gamma f\Vert_\infty} \sqrt{\frac{KN}{1-e^{-2KT}}}\label{eq: dtPtf leq sqrt} \end{aligned}$$ on $V\setminus T^\rho_R(V_0)$. On the other hand, activating (\[eq:integrate F’\]) towards $\Gamma P_T f$ and throwing away the nonnegative term $(\Delta P_T f)^2 \frac{1-e^{-2KT}}{Kn}$ yields $$\begin{aligned} \Gamma P_T f \leq \left({e^{-2KT} + \frac{\sqrt{N}}{R}H(K,K_0,T)}\right)\Vert\Gamma f\Vert_\infty \label{eq: GammaPTf leq Gamma plus H} \end{aligned}$$ on $V\setminus T^\rho_R(V_0)$. This gives good control on the time derivative and gradient of the semigroup. We estimate $$\begin{aligned} H(K,K_0,T)&\leq 2(K_0+K)e^{2K_0 T} \int_0^\infty e^{-2(K_0+K)s}\left(s\sqrt {K_0} + \sqrt{2s} \right) ds\notag \\& =2(K_0+K)e^{2K_0 T}\cdot \frac 1 4 \left(\frac{\sqrt{K_0}}{(K_0+K)^2} + \frac{\sqrt{\pi}}{(K_0+K)^{3/2}} \right)\notag\\ &= e^{2K_0 T}\cdot \frac 1 2 \left(\frac{\sqrt{K_0}}{K_0+K} + \sqrt{\frac{\pi}{K_0+K}} \right).\label{eq:Hesti} \end{aligned}$$ Moreover, for $t<T$, one has $$\begin{aligned} H(K,K_0,t) &\leq 2(K_0+K) e^{2K_0 t} \int_0^T e^{-2(K_0+K)s} \left(s\sqrt {K_0} + \sqrt{2s} \right) ds \nonumber\\& = e^{2K_0(t-T)}H(K,K_0,T) \label{eq: Ht HT}. \end{aligned}$$ By assumption, one has $\rho(x,y)\leq R_\rho$ whenever $x \sim y$. We fix $T, R,r>0$ and $x_0 \in V$. We suppose $\rho(x_0,V_0)= R+R_\rho + r$. Our aim is to show that $$\rho(V_0,x_0) \leq R_\rho + 18.2 e^{2K_0/K} \sqrt{\frac{N}{K_0+K}}.$$ Let us explain the strategy of the remaining proof first. We will consider functions $f$ with $\Gamma f \leq 1$ and being constant outside of $B_r(x_0)$. We need the additional distance $R$ to have reasonable estimates for $\partial P_t f$ and $\Gamma P_t f$ for all vertices in $B_{r+R_\rho}(x_0)$. The $R_\rho$ is needed to separate $B_r(x_0)$ and $T^\rho_R(V_0)$, i.e., to guarantee that there are no edges connecting two vertices from the two sets respectively. The distance $R$ will be chosen later to ensure that the term $\frac {\sqrt{N}}{R} H(K,K_0,T)$ is small enough to obtain good estimates for $r$. Let us denote $$\mathcal F:= \{f \in C(V):f(x_0)=0\mbox{ and } \Gamma f \leq 1 \mbox{ and } f(y) = \sup f<\infty,\,\, \forall\, y \in V\setminus B_r(x_0)\}.$$ We write $C_{\max}:=\sup\{f(y):f \in \mathcal F, y \in V\}< \infty$ due to connectedness. The idea is to take $f\in \mathcal F$ such that $\sup f$ is (close to be) maximal. Then, we take $P_t f$ and cut it off appropriately such that its cut-off version $h $ belongs to $\mathcal F$. By the estimate (\[eq: dtPtf leq sqrt\]) for $\|\partial_t P_t f\|$, we can upper bound $\|P_t f - f\|$ outside the tube $T^\rho_R(V_0)$. On the other hand for $y_0 \notin B_r(x_0)$, we can upper bound $\|P_t f(x_0) - P_t f (y_0)\|$ by the estimate (\[eq: GammaPTf leq Gamma plus H\]) for $\Gamma P_t f$. By triangle inequality, we can thus upper bound (the cut-off version of) $$\|f(y_0)-P_tf(y_0)+P_tf(y_0)-P_tf(x_0)+P_tf(x_0)-f(x_0)\|=\|f(y_0)-f(x_0)\|.$$ Notice that $ \|f(y_0)-f(x_0)\|\approx C_{\max} \geq r$, this leads to an upper estimate for $r$ when choosing $R$ and $T$ appropriately. The reason, why we take $C_{\max}$ as a substitute for the distance $\rho$, is that we need to forth- and back estimate between the distance and the gradient. The problem is that for $\rho$, we do not always have $f(x)- f(y) \leq \rho(x,y)$ when only assuming $\Gamma f \leq 1$. To avoid this problem, we take a certain resistance metric between $x_0$ and $V \setminus B_r(x_0)$ given by $C_{\max}$. We now give the details. Let $\varepsilon>0$. We choose $f \in \mathcal F$ such that $C:=\sup f \geq C_{\max} - \varepsilon$. W.l.o.g., we can assume that $f \geq 0$. We have $C_{\max} \geq r$ since the function $\tilde{f}:=\min(\rho(x_0,\cdot),r) \in \mathcal F$ and $\sup \tilde{f} = r$ . Now, we set $g_{\max} :=\inf \{P_T f(y):y \in B_{r+R_\rho}(x_0) \setminus B_r(x_0)\}$ and $$g(x):= \begin{cases} P_T f(x) \wedge g_{\max} &:x \in B_r(x_0) \\ g_{\max} &: else \end{cases},$$where $a\wedge b:=\min\{a,b\}$ for $a,b\in {\mathbb{R}}$. We remark that $B_{r+R_\rho}(x_0) \setminus B_r(x_0) \neq \emptyset$ due to the $R_\rho$ assumption (that is, $\rho(x,y)\leq R_\rho$ for $x\sim y$) and since $V_0 \neq \emptyset$ and since there is a path from $x_0$ to $V_0$ due to connectedness. The reason why we take the infimum over $B_{r+R_\rho}(x_0) \setminus B_r(x_0)$ and not over $V \setminus B_r(x_0)$ is that we want to control $\|P_T f(y_0) - f(y_0)\|$ at $y_0$ where the infimum of $P_Tf(\cdot)$ is almost attained. But this only works if $y_0$ is far away from the negatively curved set $V_0$. In fact, we have $$\label{eq: g} (g(y)-g(z))^2\leq (P_Tf(y)-P_Tf(z))^2,\,\,\,\,\text{for all }\,\,y,z \in V.$$ We can check (\[eq: g\]) as follows. When $y,z\in B_r(x_0)$, we have $$(g(y)-g(z))^2= (P_Tf(y)\wedge g_{\max}-P_Tf(z)\wedge g_{\max})^2\leq (P_Tf(y)-P_Tf(z))^2.$$ When one of the two vertices $y$ and $z$ lies in $B_r(x_0)$ and the other one lies outside $B_r(x_0)$, say $y\in B_r(x_0)$ and $z\in V\setminus B_r(x_0)$, we have $$(g(y)-g(z))^2=(P_Tf(y)\wedge g_{\max}-g_{\max})^2.$$ In case that $P_Tf(y)\leq g_{\max}$, we have by the definition of $g_{\max}$ that $$(P_Tf(y)\wedge g_{\max}-g_{\max})^2=(P_Tf(y)-g_{\max})^2\leq (P_Tf(y)-P_Tf(z))^2.$$ Otherwise when $P_Tf(y)>g_{\max}$, we have $$(P_Tf(y)\wedge g_{\max}-g_{\max})^2=0\leq (P_Tf(y)-P_Tf(z))^2.$$ When $y,z\in V\setminus B_r(x_0)$, we have $$(g(y)-g(z))^2=(g_{\max}-g_{\max})^2=0\leq (P_Tf(y)-P_Tf(z))^2.$$ This finishes the verification of (\[eq: g\]). We obtain directly from (\[eq: g\]) that $$\Gamma g\leq \Gamma P_Tf.$$ We observe that for all $y \in V \setminus B_{r+R_\rho}(x_0)$, there is no neighbor of $y$ in $B_r(x_0)$ due to the $R_\rho$ assumption. Since $g$ is constant on $V \setminus B_r(x_0)$, we obtain $\Gamma g = 0$ on $V \setminus B_{r+R_\rho}(x_0)$. Using $(\ref{eq: GammaPTf leq Gamma plus H})$, we derive further that $$\Gamma g \leq {e^{-2KT} + \frac{\sqrt{N}}{R}H(K,K_0,T)},$$ where we used the property $\Gamma f\leq 1$. We will later choose $T$ and $R$ such that this bound of $\Gamma g$ is significantly smaller than one. Setting the function $h: V\to {\mathbb{R}}$ to be $$h:= g \bigg/\sqrt{e^{-2KT} + \frac{\sqrt{N}}{R}H(K,K_0,T)},$$ we have $\Gamma h\leq 1$. Therefore $h - h(x_0) \in \mathcal F$ and thus, $\sup h - h(x_0) \leq C_{\max}$. Hence, $$\begin{aligned} g_{\max}-g(x_0) \leq C_{\max} \sqrt{e^{-2KT} + \frac{\sqrt{N}}{R}H(K,K_0,T)}. \label{eq: gy0 minus gx0} \end{aligned}$$ Let $y_0$ be a vertex in $B_{r+R_\rho}(x_0) \setminus B_r(x_0)$ such that $P_T f(y_0)-g(y_0) < \varepsilon$. We obtain by (\[eq: dtPtf leq sqrt\]) $$\begin{aligned} \|f(y_0) - g(y_0)\| &\leq \|f(y_0) - P_T f(y_0)\| + \varepsilon \leq \varepsilon + \int_0^T \|\partial_t P_t f(y_0)\| dt \nonumber \\ &\leq \varepsilon + \int_0^T \sqrt{e^{-2Kt} + \frac{\sqrt{N}}{R}H(K,K_0,t)} \sqrt{\frac{KN}{1-e^{-2Kt}}} dt. \label{eq:fy0 minus gy0} \end{aligned}$$ Analogously, we have $$\begin{aligned} \|g(x_0) - f(x_0)\| \leq \int_0^T \sqrt{e^{-2Kt} + \frac{\sqrt{N}}{R}H(K,K_0,t)} \sqrt{\frac{KN}{1-e^{-2Kt}}} dt. \label{eq:fx0 minus gx0} \end{aligned}$$ Noticing that $f(y_0)=\sup f=C$ and putting together (\[eq: gy0 minus gx0\]), (\[eq:fy0 minus gy0\]) and (\[eq:fx0 minus gx0\]) yield $$\begin{aligned} C_{\max}-\varepsilon &\leq C = f(y_0)-f(x_0) \\&\leq \|f(y_0) - g(y_0)\| + \|g(y_0)-g(x_0)\| + \|g(x_0)-f(x_0)\| \\ & \leq \varepsilon + 2\int_0^T \sqrt{e^{-2Kt} + \frac{\sqrt{N}}{R}H(K,K_0,t)} \sqrt{\frac{KN}{1-e^{-2Kt}}} dt \\ & \qquad + C_{\max} \sqrt{e^{-2KT} + \frac{\sqrt{N}}{R}H(K,K_0,T)}. \end{aligned}$$ Letting $\varepsilon$ tend to zero yields $$r \leq C_{\max} \leq \frac{2\int_0^T \sqrt{e^{-2Kt} + \frac{\sqrt{N}}{R}H(K,K_0,t)} \sqrt{\frac{KN}{1-e^{-2Kt}}} dt}{1-\sqrt{e^{-2KT} + \frac{\sqrt{N}}{R}H(K,K_0,T)} } \label{eq: r upper estimate}$$ whenever the denominator is positive. We set $T:=1/K$ and $R:=4\sqrt{N} H(K,K_0,1/K)$. Then the denominator of the RHS of (\[eq: r upper estimate\]) is $1-\sqrt{e^{-2}+1/4}>0$. Observe that (\[eq:Hesti\]) implies $$\begin{aligned} R\leq 2 \sqrt N e^{2K_0/K} \left(\frac{\sqrt{K_0}}{K_0+K} + \sqrt{\frac{\pi}{K_0+K}} \right). \label{eq: R def and estimate} \end{aligned}$$ Next, we estimate the numerator. By (\[eq: Ht HT\]), we have for $t \leq T=1/K$ that $$\begin{aligned} \frac{\sqrt{N}}{R}H(K,K_0,t) = \frac{H(K,K_0,t)}{4H(K,K_0,T)} \leq \frac 1 4 \exp \left[-2K_0\left(\frac 1 K - t \right) \right] \leq \frac 1 4. \label{eq: nRH leq 1 over 4} \end{aligned}$$ Therefore, we obtain $$\begin{aligned} &\int_0^T \sqrt{e^{-2Kt} + \frac{\sqrt{N}}{R}H(K,K_0,t)} \sqrt{\frac{KN}{1-e^{-2Kt}}} dt \\ \leq & \int_0^T \left[\sqrt{e^{-2Kt}} + \sqrt{\frac{\sqrt{N}}{R}H(K,K_0,t)} \right] \sqrt{\frac{KN}{1-e^{-2Kt}}} dt \\ \leq &\int_0^\infty \sqrt{e^{-2Kt}}\sqrt{\frac{KN}{1-e^{-2Kt}}} dt + \int_0^T \frac 1 2 \sqrt{\frac{KN}{1-e^{-2Kt}}} dt\\ =& \frac \pi 2 \sqrt{\frac{N}{K}} + \int_0^T \frac 1 2\sqrt{\frac{KN}{1-e^{-2Kt}}} dt\\ =& \frac \pi 2 \sqrt{\frac{N}{K}} +\frac{1}{2} \sqrt{\frac{N}{K}} \int_0^1 \frac{d\tau}{\sqrt{1-e^{-2\tau}}}\\ \leq& \frac {\pi+\operatorname{arctanh}(\sqrt{1-e^{-2}})} 2 \sqrt{\frac{N}{K}}. \end{aligned}$$ Thus, (\[eq: r upper estimate\]) implies that $$\begin{aligned} r \leq \frac{ (\pi+\operatorname{arctanh}(\sqrt{1-e^{-2}}))}{1 - \sqrt{e^{-2} + \frac 1 4}} \sqrt{\frac{N}{K}}. \end{aligned}$$ Using this and (\[eq: R def and estimate\]) yields $$\begin{aligned} \label{eq: rho first estimate} \rho(x_0,V_0) =r + R_\rho + R \hspace{0.7\textwidth} \nonumber & \\ \leq \frac{ (\pi+\operatorname{arctanh}(\sqrt{1-e^{-2}}))}{1 - \sqrt{e^{-2} + \frac 1 4}} \sqrt{\frac{N}{K}} + R_\rho + 2 \sqrt N e^{2K_0/K} \left(\frac{\sqrt{K_0}}{K_0+K} + \sqrt{\frac{\pi}{K_0+K}} \right). & \end{aligned}$$ It is left to show that $e^{2K_0/K}\sqrt{\frac{N}{K_0 +K}}$ is the dominating term in the sum and to give the corresponding coefficient. We start with comparing the addends in the brackets of (\[eq: rho first estimate\]): we have $$\frac{\sqrt{K_0}}{K_0+K} \leq \frac 1 {\sqrt{K_0+K}},$$ and, hence, $$R \leq 2(1+\sqrt{\pi})e^{2K_0/K}\sqrt{\frac{N}{K_0+K}}. $$ Notice that one has for $s\geq 0$, $$e^{2s} \geq \sqrt{1+s}.$$ Thus via $s:=K_0/K$, we obtain $$e^{2K_0/K} \sqrt{\frac{N}{K_0+K}} = \sqrt{\frac{N}{K}} \cdot e^{2K_0/K} \sqrt{\frac{1}{1+K_0/K}} \geq \sqrt{\frac{N}{K}}.$$ Hence, $$r \leq \frac{ (\pi+\operatorname{arctanh}(\sqrt{1-e^{-2}}))}{1 - \sqrt{e^{-2} + \frac 1 4}} e^{2K_0/K} \sqrt{\frac{N}{K_0+K}}.$$ We infer that $$\rho(x_0,V_0) \leq R_\rho+r+R \leq R_\rho + 18.2e^{2K_0/K} \sqrt{\frac{N}{K_0+K}}.$$ This finishes the proof. Combining Theorem \[thm:TubeNegativeCurvatureFiniteDimension\] and Proposition \[prop:DistanceAndDimensionBoundedDegree\], we obtain a distance bound for bounded vertex degree and infinite dimension, what proves part (iii) of Theorem \[thm:main\]. Let $G=(V,w,m)$ be a graph with finite maximal vertex degree $\operatorname{Deg}_{\max}$, and let $K,K_0>0$. Let $\emptyset \neq V_0\subset V$ and suppose that $G$ satisfies $$CD(K,\infty,x)\quad\forall \, x\in V \setminus V_0\quad\text{and}\quad CD(-K_0,\infty,x)\quad \forall \,x\in V_0.$$ Then, for all $x \in V$, one has $$\begin{aligned} d(x,V_0) \leq 1 + 26 e^{4K_0/K} \frac{\operatorname{Deg}_{\max}}{\sqrt{KK_0}}. \end{aligned}$$ Proposition \[prop:DistanceAndDimensionBoundedDegree\] yields $CD(-\widetilde{K_0}, N)$ on $V_0$ and $CD(\widetilde{K}, N)$ on $V\setminus V_0$ with $$\begin{aligned} \widetilde{K_0} &= K_0+s, \\ \widetilde{K} &= K-s,\\ N&= \frac{2\operatorname{Deg}_{\max}}s. \end{aligned}$$ Applying Theorem \[thm:TubeNegativeCurvatureFiniteDimension\] with $\rho:=d \sqrt{\frac {2} {\operatorname{Deg}_{\max}}}$ which is intrinsic due to Proposition \[prop:DistanceAndDimensionBoundedDegree\] yields $$\begin{aligned} \rho(x,V_0) \leq R_\rho + 18.2 e^{2(K_0+s)/(K-s)} \sqrt{\frac{2\operatorname{Deg}_{\max}}{(K_0+K)s}} \end{aligned}$$ with $R_\rho = \sqrt{\frac {2} {\operatorname{Deg}_{\max}}}.$ By choosing $s:= \frac{K K_0}{2(K+K_0)}$, we see $$\begin{aligned} \frac{K_0+s}{K-s} = \frac{2K_0(K+K_0) + KK_0}{2K(K+K_0)-KK_0} \leq \frac{2K_0(2K+K_0)}{K(2K+K_0)} = \frac{2K_0}{K} \end{aligned}$$ and $$\begin{aligned} \sqrt{\frac{2\operatorname{Deg}_{\max}}{(K_0+K)s}} = 2\sqrt{\frac{\operatorname{Deg}_{\max}}{KK_0}}. \end{aligned}$$ Therefore, $$\begin{aligned} \rho(x,V_0) \leq R_\rho + 36.4 e^{4K_0/K} \sqrt{\frac{\operatorname{Deg}_{\max}}{KK_0}} \end{aligned}$$ which implies $$\begin{aligned} d(x,V_0) \leq \rho(x,V_0)\sqrt{\frac{\operatorname{Deg}_{\max}}{2}} \leq 1 + 18.2 \sqrt{2} e^{4K_0/K} \frac{\operatorname{Deg}_{\max}}{\sqrt{KK_0}}\le 1+26 e^{4K_0/K} \frac{\operatorname{Deg}_{\max}}{\sqrt{KK_0}}. \end{aligned}$$ This finishes the proof. Acknowledgements {#acknowledgements .unnumbered} ================ C. R. is grateful for the hospitality of Durham University, where parts of this work have been carried out during his visits. Moreover, the research was supported by the EPRSC Grant EP/K016687/1 Topology, Geometry and Laplacians of Simplicial Complexes. F. M. wants to thank the German Research Foundation (DFG) and the German National Merit Foundation for financial support, and the Harvard University Center of Mathematical Sciences and Applications for their hospitality. [HKMW13]{} Luigi Ambrosio, Nicola Gigli, and Giuseppe Savar[é]{}. Metric measure spaces with [R]{}iemannian [R]{}icci curvature bounded from below. , 163(7): 1405–1490, 2014. Dominique Bakry and Michel [É]{}mery. Diffusions hypercontractives. In [S[é]{}minaire de Probabilit[é]{}s XIX 1983/84]{}, pages 177–206. Springer, 1985. Frank Bauer, Bobo Hua, and Matthias Keller. On the [$l^p$]{} spectrum of [L]{}aplacians on graphs. , 248:717–735, 2013. Frank Bauer, Paul Horn, Yong Lin, Gabor Lippner, Dan Mangoubi, Shing-Tung Yau, et al. Li-[Y]{}au inequality on graphs. , 99(3): 359–405, 2015. Frank Bauer, J[ü]{}rgen Jost, and Shiping Liu. Ollivier-[R]{}icci curvature and the spectrum of the normalized graph [L]{}aplace operator. , 19(6): 1185–1205, 2012. Frank Bauer, Matthias Keller, and Radosław K. Wojciechowski. Cheeger inequalities for unbounded graph [L]{}aplacians. , 17(2): 259–271, 2015. Fan Chung, Yong Lin, and Shing-Tung Yau. Harnack inequalities for graphs with non-negative [R]{}icci curvature. , 415(1): 25–32, 2014. K. D. Elworthy, Manifolds and graphs with mostly positive curvatures. , 96–110, [Progress in Probability]{}, 26, Birkhäuser Boston, Boston, MA, 1991. Ricci curvature bounds for weakly interacting [M]{}arkov chains. , 22: no. 40, 2017. Matthias Erbar, Kazumasa Kuwada, and Karl-Theodor Sturm. On the equivalence of the entropic curvature-dimension condition and [B]{}ochner’s inequality on metric measure spaces. , 201(3): 993–1071, 2015. Matthias Erbar and Jan Maas. Ricci curvature of finite [M]{}arkov chains via convexity of the entropy. , 206(3): 997–1038, 2012. Max Fathi and Jan Maas. Entropic [R]{}icci curvature bounds for discrete interacting systems. , 26(3): 1774–1806, 2016. Matthew Folz. Gaussian upper bounds for heat kernels of continuous time simple random walks. , 16: no. 62, 1693–1722, 2011. Matthew Folz. Volume growth and stochastic completeness of graphs. , 366(4): 2089–2119, 2014. Max Fathi and Yan Shu. Curvature and transport inequalities for [M]{}arkov chains in discrete spaces. , 24(1): 672–698, 2018. Alexander Grigor’yan, Xueping Huang, and Jun Masamune. On stochastic completeness of jump processes. , 271(3-4): 1211–1239, 2012. Chao Gong and Yong Lin. Equivalent properties for CD inequalities with unbounded Laplacians. , 38(5): 1059–1070, 2017. Bobo Hua and Matthias Keller. Harmonic functions of general graph [L]{}aplacians. , 51(1-2): 343–362, 2014. Xueping Huang, Matthias Keller, Jun Masamune, and Radoslaw K. Wojciechowski. A note on self-adjoint extensions of the [L]{}aplacian on weighted graphs. , 265(8): 1556–1578, 2013. Bobo Hua and Yong Lin. Stochastic completeness for graphs with curvature dimension conditions. , 306: 279–302, 2017. Paul Horn, Yong Lin, Shuang Liu, and Shing-Tung Yau. Volume doubling, [P]{}oincaré inequality and [G]{}aussian heat kernel estimate for nonnegative curvature graphs. , ahead of print, (2017-10-17). Xueping Huang. On stochastic completeness of weighted graphs. , 2011. J[ü]{}rgen Jost and Shiping Liu. Ollivier’s [R]{}icci curvature, local clustering and curvature-dimension inequalities on graphs. , 51(2): 300–322, 2014. J[ü]{}rgen Jost. . Universitext. Springer-Verlag, Berlin, fifth edition, 2008. Christian Ketterer. On the geometry of metric measure spaces with variable curvature bounds. , 27(3): 1951-1994, 2017. Yong Lin and Shuang Liu. Equivalent properties of CD inequality on graph. , 61(3): 431-440, 2018. https://arxiv.org/abs/1512.02677. Yong Lin, Linyuan Lu, and Shing-Tung Yau. Ricci curvature of graphs. , 63(4): 605–627, 2011. Shiping Liu, Florentin Münch, and Norbert Peyerimhoff. Bakry-[É]{}mery curvature and diameter bounds on graphs. , 57(2): no.67, 2018. John Lott and C[é]{}dric Villani. Ricci curvature for metric-measure spaces via optimal transport. , 169: 903–991, 2009. Yong Lin and Shing-Tung Yau. Ricci curvature and eigenvalue estimate on locally finite graphs. , 17(2): 343–356, 2010. Alexander Mielke. Geodesic convexity of the relative entropy in reversible [M]{}arkov chains. , 48(1-2): 1-31, 2013. Florentin M[ü]{}nch. Li-[Y]{}au inequality on finite graphs via non-linear curvature dimension conditions. , 120: 130-164, 2018. Florentin M[ü]{}nch. Remarks on curvature dimension conditions on graphs. , 56(1): no.11, 2017. Yann Ollivier. Ricci curvature of [M]{}arkov chains on metric spaces. , 256(3): 810–864, 2009. Peter Petersen. , volume 171 of [Graduate Texts in Mathematics*]{}. Springer, Cham, third edition, 2016. Peter Petersen and Chadwick Sprouse. Integral curvature bounds, distance estimates and applications. , 50(2): 269–298, 1998. Michael Schmuckenschl[ä]{}ger. Curvature of nonlocal markov generators. , 34: 189–197, 1998. Karl-Theodor Sturm. On the geometry of metric measure spaces, I. , 196(1): 65–131, 2006. Andreas Weber. Analysis of the physical Laplacian and the heat flow on a locally finite graph. , 370(1): 146–158, 2010. Shiping Liu,\ School of Mathematical Sciences, University of Science and Technology of China, Hefei 230026, Anhui Province, China\ `[email protected]`\ \ Florentin Münch,\ Max-Planck-Institut für Mathematik in den Naturwissenschaften, Leipzig, Inselstra[ß]{}e 22, 04103 Leipzig, Germany\ `[email protected]`\ \ Norbert Peyerimhoff,\ Department of Mathematical Sciences, Durham University, Durham DH1, 3LE, United Kingdom\ `[email protected]`\ \ Christian Rose,\ Max-Planck-Institut für Mathematik in den Naturwissenschaften, Leipzig, Inselstra[ß]{}e 22, 04103 Leipzig, Germany\ `[email protected]`	{ "pile_set_name": "ArXiv" }
--- abstract: 'We present [Emerge]{}, an Empirical ModEl for the foRmation of GalaxiEs, describing the evolution of individual galaxies in large volumes from $z \sim 10$ to the present day. We assign a star formation rate to each dark matter halo based on its growth rate, which specifies how much baryonic material becomes available, and the instantaneous baryon conversion efficiency, which determines how efficiently this material is converted to stars, thereby capturing the baryonic physics. Satellites are quenched following the delayed-then-rapid model, and they are tidally disrupted once their subhalo has lost a significant fraction of its mass. The model is constrained with observed data extending out to high redshift. The empirical relations are very flexible, and the model complexity is increased only if required by the data, assessed by several model selection statistics. We find that for the same final halo mass galaxies can have very different star formation histories. Nevertheless, the average star formation and accretion rates are in good agreement with models following an abundance matching strategy. Galaxies that are quenched at $z=0$ typically have a higher peak star formation rate compared to their star-forming counterparts. The accretion of stars can dominate the total mass of massive galaxies, but is insignificant for low-mass systems, independent of star-formation activity. [Emerge]{} predicts stellar-to-halo mass ratios for individual galaxies and introduces scatter self-consistently. We find that at fixed halo mass, passive galaxies have a higher stellar mass on average. The intra-cluster-mass in massive haloes can be up to 8 times larger than the mass of the central galaxy. Clustering for star-forming and quenched galaxies is in good agreement with observational constraints, indicating a realistic assignment of galaxies to haloes.' author: - \| Benjamin P. Moster$^{1,2,3,}$[^1], Thorsten Naab$^2$, Simon D. M. White$^2$\ $^1$Universitäts-Sternwarte, Ludwig-Maximilians-Universität München, Scheinerstr. 1, 81679 München, Germany\ $^2$Max-Planck Institut für Astrophysik, Karl-Schwarzschild Straße 1, 85748 Garching, Germany\ $^3$Kavli Institute for Cosmology and Institute of Astronomy, University of Cambridge, Madingley Rd, Cambridge CB3 0HA, UK\ bibliography: - 'astro.bib' date: 'Last updated 2017 January 1; in original form 2017 January 1' title: '[Emerge]{} – An empirical model for the formation of galaxies since $z\sim10$' --- \[firstpage\] cosmology: dark matter, theory – galaxies: evolution, formation, statistics, stellar content Introduction {#sec:intro} ============ In the standard model of cosmology, only a small fraction of the present energy density of the Universe is in the form of baryonic matter. The remaining dark components are the dynamically cold and collisionless dark matter [@Zwicky:1933aa; @Davis:1985aa], and a near-uniform dark energy field which can be described by a cosmological constant [@Riess:1998aa; @Perlmutter:1999aa; @Perlmutter:1999ab]. Together they form the foundation of the [ $\Lambda{\rm CDM}$ ]{}theory in which structure formation proceeds through gravitationally driven hierarchical collapse and merging. In the standard picture, galaxies form by the cooling and condensation of gas in the centres of virialised dark matter halos [@White:1978aa; @Fall:1980aa; @Blumenthal:1984aa], which results in a tight correlation between the properties of haloes and those of the galaxies they host. The formation and evolution of dark matter haloes has been studied extensively with large cosmological $N$-body simulations (@Springel:2005ab [@Boylan-Kolchin:2009aa]; @Klypin:2011aa; ). As this process only depends on gravity and the initial conditions have been measured very accurately [@Planck-Collaboration:2016aa], these simulations have converged and make accurate and definite predictions for the properties of dark matter haloes at all cosmic epochs [@Frenk:2012aa; @Knebe:2013aa]. The halo mass function (HMF) found in the simulations is very steep and the dark matter is distributed over many orders of magnitude. If galaxies were forming with the same efficiency in haloes of different masses, we would expect the galaxy stellar mass function (SMF) to have the same shape as the HMF. However, the observed local SMF has a very different shape, with a much shallower slope at the low-mass end and an exponential cut-off at much smaller masses [@Li:2009aa; @Bernardi:2013aa]. This tension indicates the complexity of the baryonic physics regulating galaxy formation, such as gas cooling, star formation, and feedback processes. There are several pathways to learn about the formation and evolution of galaxies. The most popular method are [‘ab initio’ models]{}, where an initial distribution of gas and dark matter is evolved according to a specified set of relevant physical processes, including all the various baryonic physics that one thinks is important (see the reviews by @Somerville:2015aa and @Naab:2016aa for more details). In [hydrodynamical simulations]{} the baryonic component is discretised and evolved hydrodynamically [@Hirschmann:2014aa; @Vogelsberger:2014aa; @Dubois:2014aa; @Hopkins:2014aa; @Khandai:2015aa; @Schaye:2015aa], while in [semi-analytic models]{} (SAMs) it is separated from the dark-matter-dominated growth of structure by post-processing halo merger trees with a series of physically motivated recipes (@White:1991aa; @Kauffmann:1993aa; @Cole:1994aa; @Somerville:1999aa; @Kang:2005aa; @Monaco:2007aa; @Benson:2012aa). The advantage of ab initio models is that they track galaxies and haloes self-consistently through cosmic time, and can test the impact of different physical processes on galaxy properties. If the models disagree with observations the model is changed either by implementing the physical processes differently, or by including new physical processes. However, these methods can only achieve a limited resolution, so that simplified and highly uncertain ‘sub-grid’ models have to be used to treat the unresolved physical processes, such as star and black hole formation and the related feedback. The effects of these processes then become tunable via free parameters, such that the models typically need to be calibrated with observations and are effectively phenomenological. Due to the complex interaction of different physical prescriptions, the model parameters can be degenerate and difficult to interpret if the model is not constrained well by data [@Lu:2012aa]. Moreover, there is considerable uncertainty about whether the physics of galaxy formation is reliably represented. A different option has emerged with the advent of data sets from large galaxy surveys [@York:2000aa; @Colless:2001aa; @Lilly:2007aa; @Driver:2011aa; @Grogin:2011aa; @McCracken:2012aa] and avoids explicitly modelling the baryonic physics. Instead, [empirical models]{} of galaxy formation use relations with adjustable parameters to statistically link observed galaxy properties to simulated dark matter haloes. In this way they can model galaxy formation unbiased by assumptions on poorly understood baryonic physics, instead ‘marginalising’ over these uncertainties. Predictions by empirical models are therefore very useful for planning future surveys and for the interpretation of the observations. Moreover, they provide a framework for ab initio models, and can thus help to constrain the relevant physical processes. Empirical results have been widely used to fix unconstrained parameters in the sub-grid models of hydrodynamic simulations, e.g. by requiring that simulated galaxies reproduce the empirically determined stellar-to-halo mass (SHM) ratio. Hence empirical and ab initio models are complementary methods that can be applied to study the physics that drives galaxy formation. In the [halo occupation distribution]{} and the related [conditional luminosity function]{} formalisms, the distribution of galaxies having specified intrinsic properties within main haloes of a given mass is constrained using galaxy abundance and clustering statistics (@Peacock:2000aa [@Seljak:2000aa; @White:2001aa; @Berlind:2002aa; @Yang:2003aa; @Zehavi:2004aa; @Zheng:2005aa]; @Tinker:2005aa [@Brown:2008aa; @Leauthaud:2012aa]). These approaches have typically been used at low redshift, as reliable galaxy clustering measurements are not available at high redshift. This problem can be circumvent by directly connecting galaxies to the underlying substructure. The [subhalo abundance matching]{} method links the luminosity or stellar mass of a galaxy to the dark matter halo mass by matching the cumulative abundance of galaxies to those of haloes and subhaloes (@Mo:1996aa [@Wechsler:1998aa; @Vale:2004aa]; @Conroy:2006aa; @Wang:2006aa [@Wang:2007aa]; @Moster:2010aa; @Behroozi:2010aa; @Guo:2010aa [@Trujillo-Gomez:2011aa]). The clustering statistics of galaxies can then be readily derived from the clustering of haloes in the simulation resulting in a remarkable agreement with observed correlation functions. There are two ways to derive the relation between stellar and halo mass in this context. In the backward modelling approach, observed galaxies and simulated haloes in an equal volume are rank ordered by mass and then matched one by one. The SHM relation can then be described by a fitting function. In the forward modelling approach, simulated dark matter haloes are populated with galaxies using a parameterised relation between stellar and halo mass. The free parameters are then constrained by requiring that the observed SMF be reproduced. This method has the advantage that scatter in the relation can easily be added to account for possible differences in the formation histories of haloes at a fixed mass. ------------------------------------ ------------------------------- ![image](boxDM){height="70mm"} ![image](boxG){height="70mm"} ------------------------------------ ------------------------------- As the evolution of dark matter haloes is determined by the cosmological model, the link between galaxies and haloes can be employed to infer the evolution of galaxy properties from the growth histories of haloes through cosmic time (@Conroy:2007aa; @White:2007aa; @Zheng:2007ab; @Firmani:2010aa [@Wang:2013aa]). In particular @Conroy:2009aa show how this method can be used to empirically constrain the average star formation histories (SFHs) and stellar mass growth of galaxies in haloes with a given mass since $z=2$. This approach has been extended by @Moster:2013aa and @Behroozi:2013aa to $z\sim8$ using halo merger trees that have been extracted from cosmological simulations. These multi-epoch abundance matching models have been very successful in describing the average evolution of galaxy properties in dark matter haloes. However, they do not self-consistently track the growth history of individual galaxies. Inferred galaxy properties such as clustering then only depend on halo mass, although it is well established that the spatial distribution of dark matter haloes depends on their formation time [e.g. @Gao:2005aa]. Hence galaxy properties should also depend on the formation history of the halo. A simple way to design an empirical model for the evolution of individual galaxies is presented by @Mutch:2013aa. Instead of linking integrated properties such as the present stellar mass to the halo mass, they connect instantaneous properties. In this model the star formation rate (SFR) of a galaxy is given by the product of the halo growth rate which determines how much material becomes available for star formation, and a parameterised baryon conversion efficiency which only depends on halo mass and determines how effectively this material is converted into stars. The stellar mass of a galaxy is then computed by integrating the SFHs of each galaxy through cosmic time taking into account galaxy mergers. In this way a complete formation history for every galaxy is provided. The connection between star formation and halo growth is also found observationally [@Tinker:2012aa] and in hydrodynamical simulations [@Feldmann:2016aa]. An alternative method is to directly relate the SFH to the mass of the halo at any given time [@Lu:2014aa; @Lu:2015aa]. In this work we present the novel empirical model [Emerge]{}[^2] that describes the formation of individual galaxies in dark matter haloes. We follow @Mutch:2013aa and compute the SFR from the halo growth rate and the instantaneous conversion efficiency. However, we use a more realistic parameterisation that depends on redshift and allows for different slopes at the low and high-mass ends. Moreover, instead of setting the SFR to zero for all satellite galaxies, we include empirical treatments for star formation after infall based on the ‘delayed-then-rapid’ mechanism found by @Wetzel:2012aa, and for tidal stripping. To construct this new empirical model we follow the philosophy laid out in @Lu:2014aa, and attempt to let the data speak for themselves in a way that is as independent as possible of any model assumptions. Specifically, we assume that structure formation is determined by a [ $\Lambda{\rm CDM}$ ]{}cosmology, that the SFR of a galaxy depends on its halo’s mass and growth rate and the redshift, and that once a galaxy becomes a satellite it continues to form stars until it is rapidly quenched, its stars are stripped to the background when its halo has lost a significant fraction of its mass, and it merges with the central galaxy on a dynamical friction timescale otherwise. We constrain each of these processes with a specific data set minimising the correlation between parameters. The complexity of the model is increased stepwise if the data require it, which we assess with a number of different model selection statistics. In this way we aim to construct the simplest model that is in agreement with the data. This paper is organised as follows. In Section \[sec:simobs\] we describe the cosmological dark matter simulations and the observational data sets we use to constrain our model. The methodology of our new empirical model is presented in Section \[sec:method\]. We further discuss how the model is constrained and compare our best-fit model to the data. In Section \[sec:results\] we present our main results for the growth of the stellar component and the integrated conversion efficiency. We discuss the model in Section \[sec:disc\] and provide a summary and an outlook in Section \[sec:sum\]. In Appendix \[sec:bayes\] the model selection process is explained in detail, and in Appendix \[sec:cov\] we discuss the correlation between model parameters. Throughout this work we assume a Planck [ $\Lambda{\rm CDM}$ ]{}cosmology with ([\_[m]{} $\Omega_{\rm m}$]{}, [\_ $\Omega_{\rm \Lambda}$]{}, [\_[b]{} $\Omega_{\rm b}$]{}, $h$, $n_\mathrm{s}$, $\sigma_8$) = (0.308, 0.692, 0.0484, 0.6781, 0.9677, 0.8149). We employ a @Chabrier:2003aa initial mass function (IMF) and we convert all stellar masses and SFRs to this IMF. All virial masses are computed according to the overdensity criterion by @Bryan:1998aa. In order to simplify the notation, we will use the capital $M$ to denote dark matter halo masses and the lower case $m$ to denote galaxy stellar masses. Simulations and Observations {#sec:simobs} ============================ Empirical models connect observed galaxy properties to simulated dark matter haloes. The main pillars of each empirical model are therefore a cosmological $N$-body simulation from which dark matter haloes and merger trees are extracted, and observed data. In this work we used two dark matter simulations with side lengths of $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ and $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$, respectively. We further used five different observational constraints taken from the literature. In this section we provide more details on the simulations and observations. Dark matter simulations {#sec:sims} ----------------------- Publication $z$ Area IMF $\lambda$ ---------------------- ----------- ------------ ----- ----------- @Robotham:2011aa 0.0 - 0.1 833.13 S UV @Salim:2007aa 0.0 - 0.2 741 C UV @Gunawardhana:2015aa 0.0 - 0.4 144 C H$\alpha$ @Ly:2011aa 0.8 0.82 S H$\alpha$ @Zheng:2007aa 0.2 - 1.0 0.458 C UV/IR @Rujopakarn:2010aa 0.0 - 1.2 $\le9$ S FIR @Smolcic:2009aa 0.1 - 1.3 2 S 1.4 GHz @Shim:2009aa 0.7 - 1.9 0.029 S H$\alpha$ @Tadaki:2011aa 0.0 - 0.2 0.016 S H$\alpha$ @Sobral:2013aa 2.2 $\le1.68$ S H$\alpha$ @Magnelli:2011aa 1.3 - 2.3 0.079 S IR @Hayes:2010aa 2.2 0.016 S H$\alpha$ @Karim:2011aa 0.2 - 3.0 1.72 C 1.4 GHz @Ly:2011ab 1 - 3 0.242 S UV @Kajisawa:2010aa 0.5 - 3.5 0.029 S UV/IR @Reddy:2009aa 1.9 - 3.4 0.906 K UV @Burgarella:2013aa 0 - 4 $\le0.6$ S UV/IR @Cucciati:2012aa 0 - 4.5 0.611 S UV @Dunne:2009aa 0 - 5 0.8 S 1.4 GHz @Le-Borgne:2009aa 1 - 5 0.07 S IR/mm @van-der-Burg:2010aa 3 - 5 4 S UV @Bourne:2016aa 0.5 - 6 0.064 C UV/IR @Duncan:2014aa 4 - 7 $\le0.017$ C UV @Oesch:2013aa 3.8 - 11 0.045 S UV @McLure:2013aa 6 - 10 $\le0.05$ S UV : Observed cosmic star formation rate densities[]{data-label="tab:obscsfrd"} \ Notes: Columns are publication (1), redshift range $z$ (2), survey area in deg$^2$ (3), IMF (4): C [@Chabrier:2003aa], S [@Salpeter:1955aa], K [@Kroupa:2001aa], and spectral range used to convert fluxes into SFRs (5). The empirical model presented in this paper follows the growth of dark matter haloes and assigns a SFR to the galaxy at its centre. For this, we have extracted the halo merger trees from two $N$-body simulations. The first simulation has a smaller volume and fewer particles, and consequently results in fewer halo merger trees. It is therefore well suited to run the empirical model multiple times, as done for parameter space exploration. The second simulation has a larger volume and more particles and hence more merger trees. We only use this simulation in single runs with previously determined parameters to cover a larger mass range. As we will show, the simulations lead to identical results for the galaxy populations. For both simulations we adopted cosmological parameters consistent with the latest results by the @Planck-Collaboration:2016aa. Specifically, we chose ${\ifmmode \Omega_{\rm m} \else $\Omega_{\rm m}$\fi}=0.308$, ${\ifmmode \Omega_{\rm \Lambda} \else $\Omega_{\rm \Lambda}$\fi}=0.692$, ${\ifmmode \Omega_{\rm b} \else $\Omega_{\rm b}$\fi}=0.0484$, $H_0=67.81{\ifmmode \,\rm km\,s^{-1}\,Mpc^{-1} \else $\,\rm km\,s^{-1}\,Mpc^{-1} $ \fi }$, $n_\mathrm{s}=0.9677$, and $\sigma_8=0.8149$. The initial conditions for both simulations were generated with the [Music]{} code [@Hahn:2011aa] using a power spectrum computed with the `CAMB` code [@Lewis:2000aa]. The first simulation has a side length of $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ and contains $512^3$ dark matter particles corresponding to a particle mass of $9.88\times10^8{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$. The second simulation has a side length of $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ and contains $1024^3$ dark matter particles corresponding to a particle mass of $2.92\times10^8{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$. Both simulations were run with periodic boundary conditions from redshift $z=63$ to 0 using the TreePM code [Gadget3]{} [@Springel:2005aa] creating 94 snapshots, equally spaced in scale factor ($\Delta a=0.01$). The gravitational softening was $6{\ifmmode {\rm~kpc} \else ${\rm~kpc}$\fi}$ for the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ simulation, and $3.3{\ifmmode {\rm~kpc} \else ${\rm~kpc}$\fi}$ for the $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ simulation. In the left panel of Figure \[fig:box\] we show a density map of the first simulation, remapped into a sheet with $6{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ thickness and $750{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ width using the method presented by @Carlson:2010aa. Dark matter haloes and subhaloes in the simulations were identified with the seven-dimensional halo finder [Rockstar]{} [@Behroozi:2013ac] for each snapshot. Halo masses were calculated using spherical overdensities, according to the criterion for a spherical collapse model of a tophat perturbation in a [ $\Lambda{\rm CDM}$ ]{}cosmology by @Bryan:1998aa. Given a minimal particle number of 100 for each halo the minimally resolved halo mass is $M_\mathrm{min}=10^{11}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ for the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box, and $M_\mathrm{min}=10^{10.5}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ for the $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box. Merger trees were generated using the [ConsistentTrees]{} algorithm [@Behroozi:2013ad], providing a physically consistent evolution of halo properties across time steps. We use the term ‘main halo’ to refer to distinct haloes that are not located within a larger halo, while the term ‘subhalo’ refers to all other haloes. We further assume that both main haloes and subhaloes host galaxies at their centres. The galaxy at the centre of a main halo is referred to as ‘central galaxy’, and all galaxies within subhaloes are referred to as ‘satellite galaxy’. Publication $z$ Area IMF $\lambda$ ------------------ ----------- ------- ----- ------------ @Salim:2007aa 0.0 - 0.2 741 C UV @Zheng:2007aa 0.2 - 1.0 0.458 C UV/IR @Twite:2012aa 1.0 1.4 C H$\alpha$ @Noeske:2007aa 0.2 - 1.1 0.5 K UV/IR @Tadaki:2011aa 2.2 0.016 S H$\alpha$ @Whitaker:2012aa 0.0 - 2.5 0.4 C UV/IR @Daddi:2007aa 1.4 - 2.5 0.06 S UV-1.4 GHz @Salmi:2012aa 0.9 - 1.3 0.06 C UV @Karim:2011aa 0.2 - 3.0 1.72 C 1.4 GHz @Kajisawa:2010aa 0.5 - 3.5 0.029 S UV/IR @Reddy:2012aa 1.4 - 3.7 0.44 S UV/IR @Feulner:2005aa 0.4 - 5.0 0.014 S UV/IR @Lee:2012aa 3.3 - 4.3 5.3 C UV/IR @Gonzalez:2011aa 4 - 6 0.015 S UV/IR @Schaerer:2010aa 6 - 8 2 S UV @Labbe:2013aa 8 0.04 S UV/IR @McLure:2011aa 6 - 8.7 0.013 C UV @Duncan:2014aa 4 - 7 0.017 C UV : Observed specific star formation rates[]{data-label="tab:obsssfr"} \ Notes: Columns are publication (1), redshift range $z$ (2), survey area in deg$^2$ (3), IMF (4), and spectral range used to convert fluxes into SFRs (5). Observations {#sec:observations} ------------ The empirical model assigns a SFR to each galaxy based on the growth rate of its halo, and the stellar masses are computed by integrating these. To constrain the star formation we use five different observational constraints: SMFs, cosmic SFR densities (CSFRDs), specific SFRs (sSFRs), fractions of quenched galaxies, and projected galaxy correlation functions. We convert all units to physical units using $h=0.6781$. All stellar masses and SFRs are converted to be consistent with a @Chabrier:2003aa IMF. An evolving or non-universal IMF is not considered. We specifically do not take into consideration any other systematic effects, such as different stellar populations synthesis models, dust models, spectral energy distribution fitting methods, assumed SFHs, metallicities, photometry, redshift measurements, and cosmic variance. Instead, we consider these effects to be sources of error for the model SFRs and stellar masses. As observational studies typically do not take these sources into account when quoting the errors, we calculate the variance between different observations and add the result quadratically to each data point to estimate the true systematic errors. In this way, the uncertainties in the observations, i.e. scatter between different data sets, will get translated into model uncertainties. The resulting confidence levels in modelled galaxy properties will thus reflect our lack of knowledge on systematic observational effects. We do not combine different measurements into an average sample (e.g. at various redshifts), but compute a corresponding model prediction for each measured data point. Publication Abb. $z$ Area IMF ------------------------ ------- ----------- -------- ----- @Baldry:2012aa Bal12 0.0 - 0.1 143 C @Li:2009aa LW09 0.0 - 0.2 6,437 C @Bernardi:2013aa Ber13 0.0 - 0.2 4,681 C @Perez-Gonzalez:2008aa Per08 0.0 - 4.0 0.184 S @Ilbert:2010aa Ilb10 0.0 - 2.0 2 C @Pozzetti:2010aa Poz10 0.1 - 1.0 1.4 C @Ilbert:2013aa Ilb13 0.2 - 4.0 1.52 C @Moustakas:2013aa Mou13 0.2 - 1.0 9 C @Muzzin:2013aa Muz13 0.2 - 4.0 1.62 K @Kajisawa:2009aa Kaj09 0.5 - 3.5 0.0364 S @Santini:2012aa San12 0.6 - 4.5 0.0092 S @Mortlock:2011aa Mor11 1.0 - 3.5 0.0121 S @Marchesini:2009aa Mar09 1.3 - 4.0 0.1620 K @Caputi:2011aa Cap11 3.0 - 5.0 0.6 S @Grazian:2015aa Gra15 3.5 - 7.5 0.1025 S @Lee:2012aa Lee12 3.7 - 5.0 0.0889 C @Gonzalez:2011aa Gon11 4.0 - 7.0 0.0150 S @Duncan:2014aa Dun14 4.0 - 7.0 0.0167 C @Song:2016aa Son16 4.0 - 8.0 0.0778 S : Observed stellar mass functions[]{data-label="tab:obssmf"} \ Notes: Columns are publication (1), abbreviation (2), redshift range $z$ (3), survey area in deg$^2$(4), and IMF (5). The normalisation of the instantaneous baryon conversion efficiency can be constrained with the observed CSFRD. In Table \[tab:obscsfrd\] and in Figure \[fig:csfrd\] we summarise all data sets that we have used in this work. A broad range of techniques has been used to convert fluxes into SFRs using narrowband (H$\alpha$), broadband (UV-IR), and radio (1.4 GHz) surveys up to $z=11$. Recent observations find that the CSFRD falls off rather steeply for $z>3$. Although there is generally a very good agreement between different data sets we find a variance of 0.1 dex which we include in the errors. To constrain the low and high-mass ends of the instantaneous conversion efficiency we use observed sSFRs for different stellar masses and redshifts up to $z=8$. The data sets are summarised in Table \[tab:obsssfr\] and in Figure \[fig:ssfr\]. As the data sets are not fully consistent with each other, we find a variance of 0.15 dex independent of stellar mass and redshift, and correct the errors to account for this. To constrain the overall evolution of galaxies and specifically contribution of galaxy mergers, we use the SMFs from $z=0$ to $z=8$ presented in Table \[tab:obssmf\] and in Figure \[fig:smf\]. This includes data from wide surveys that capture many massive galaxies, and from deep surveys that can well constrain the low-mass tail. Most recent observations find rather steep low-mass slopes, reconciling the tension between the integrated CSFRD and the integrated SMFs. At low redshift there is generally a good agreement between different data sets. However, at higher redshift some of the data sets are not consistent with each other showing that the systematic errors have not been fully considered. We find a variance of 0.015 dex at $z=0$ and 0.3 dex beyond $z=2$ and modify the errors accordingly. Additionally, the limited photometric information means that stellar populations cannot be fully constrained, leading to intrinsic scatter in the mass estimates relative to the true mass. As this effects becomes stronger at high redshift, we estimate the scatter as $\sigma(z)=0.08+0.06z$, based on the results of @Li:2009aa at $z\sim0.1$ and @Perez-Gonzalez:2008aa at $z\lesssim4$. Thus, when comparing to observed stellar masses we draw the stellar mass from a lognormal distribution with a mean value given by the model mass and a scatter of $\sigma(z)$. Publication $z$ Area IMF A/P ------------------- ----------- ------ ----- --------------- -- @Wetzel:2012aa 0.0 - 0.1 7,97 C sSFR @Drory:2009aa 0.2 - 1.0 1.73 C NUV-$R$-$J$ @Ilbert:2013aa 0.2 - 4.0 1.52 C NUV-$r^+$-$J$ @Moustakas:2013aa 0.2 - 1.0 9 C sSFR @Lin:2014aa 0.2 - 0.8 70 S sSFR @Muzzin:2013aa 0.2 - 4.0 1.62 K $U$-$V$-$J$ : Observed quenched fractions[]{data-label="tab:obsfq"} \ Notes: Columns are publication (1), redshift range $z$ (2), total survey area in deg$^2$ (3), IMF (4), and method to separate active from passive galaxies (5): active-to-passive cut based on the sSFR, or based on location in rest-frame colour-colour diagram (e.g. $U$-$V$-$J$: $U-V$ vs. $V-J$). Publication Survey Stellar mass IMF -------------- ---------------- -------------- --------- @Li:2006aa SDSS/DR2 K03 C @Guo:2011aa SDSS/DR7 K03 C @Yang:2012aa SDSS/DR7 B03 C : Observed projected correlation functions[]{data-label="tab:obswp"} \ Notes: Columns are publication (1), survey data (2), method to derive stellar mass (3): K03 [@Kauffmann:2003aa], B03 [@Bell:2003aa], and IMF (4). As in our model satellite galaxies keep forming stars for a specific timescale after their halo reaches its peak mass, we need an observational constraint to determine if it is still forming stars or has been quenched. Several authors have measured the SMF for both active and passive galaxies up to $z=0$, and can thus provide the fraction of quenched galaxies as a function of stellar mass. These measurements are summarised in Table \[tab:obsfq\] and in Figure \[fig:fq\]. To distinguish quenched from star forming galaxies observational studies use two different techniques. One is to also measure the SFR of each object and then applying a cut at a certain sSFR. If the SFR cannot be obtained for all galaxies because of limited depth, the classification can also be based on a colour-colour diagram that is easier to obtain. Although it is clear that these techniques do not correspond perfectly for each galaxy, it has been shown that they correlate well [@Williams:2009aa]. The variance between the different data sets is 10 per cent and we update the errors accordingly. In our model satellite galaxies get tidally stripped once their halo has fallen below a specific fraction of its peak mass. This affects galaxy clustering on small scales, so we use the projected galaxy auto-correlation functions summarised in Table \[tab:obswp\] and Figure \[fig:wp\] to constrain this fraction. All data sets are are based on the Sloan Digital Sky Survery [@York:2000aa], but while @Li:2006aa use the Data Release 2 [DR2; @Abazajian:2004aa], @Guo:2011aa and @Yang:2012aa use the Data Release 7 [DR7; @Abazajian:2009aa]. Stellar masses in @Li:2006aa and @Guo:2011aa are computed following @Kauffmann:2003aa, while @Yang:2012aa follow @Bell:2003aa. Although the data sets are in good agreement we find a variance of 0.15 dex which we add to the errors. Connecting galaxies and haloes {#sec:method} ============================== In this section we describe how [Emerge]{} relates galaxies to their dark matter haloes. To populate the dark matter halo merger trees that have been extracted from the simulations presented in section \[sec:sims\] with galaxies, we first compute the SFR based on the growth rate of the halo, and then integrate this to derive the stellar mass of each galaxy. We further consider the relevant effects for satellite galaxies, i.e. quenching, stripping and merging. For any set of model parameters we then compute a number of mock observations and compare them to the data presented in section \[sec:observations\]. Finally, the model parameters are fitted to reproduce the observations using a Markov chain Monte Carlo method. Star formation in dark matter haloes ------------------------------------ While in previous empirical models the stellar mass of a galaxy has been linked to its halo mass, [Emerge]{} connects the time derivatives of these two quantities, i.e. the SFR of a galaxy is linked to growth rate of its halo. We assume that the baryonic mass in every dark matter halo is given by the universal fraction $f_\mathrm{b} = {\ifmmode \Omega_{\rm b} \else $\Omega_{\rm b}$\fi}/{\ifmmode \Omega_{\rm m} \else $\Omega_{\rm m}$\fi}= 0.156$, and that the rate of infalling baryonic mass $\dot m_\mathrm{b}$ is proportional to the halo growth rate $\dot M$. The SFR is then given by the product of the baryonic growth rate and the instantaneous baryon conversion efficiency $\epsilon$ which may depend on halo mass and redshift: $$\label{eqn:sfrcen} \frac{{\rm d}m_* (M,z)}{{\rm d}t} = \frac{{\rm d}m_\mathrm{b}}{{\rm d}t} \cdot \, \epsilon(M,z) = f_\mathrm{b} \frac{{\rm d}M}{{\rm d}t} \cdot \, \epsilon(M,z)\;.$$ The baryonic growth rate at a given redshift $z$ thus determines the amount of material that is available for galaxy formation, while the conversion efficiency denotes how effectively this material is converted into stars. In the following sections we show how we compute these two quantities. ### The baryonic growth rate {#sec:mdotbary} The baryonic growth rate can be extracted from the dark matter simulations by assuming that the fraction of infalling baryons is equal to the universal fraction: $$\frac{{\rm d}m_\mathrm{b}}{{\rm d}t} = f_\mathrm{b} \frac{{\rm d}M}{{\rm d}t}\;.$$ All baryonic material that falls into the dark matter halo becomes available for star formation within one dynamical time of the halo $t_\mathrm{dyn} = (R_\mathrm{v}^3/\mathrm{G}M)^{1/2}$, where $\mathrm{G}$ is the gravitational constant and $R_\mathrm{v}$ is the virial radius. We therefore let the baryonic growth rate only depend on the halo growth rate averaged over the dynamical time, defined as $$\left\langle\frac{{\rm d}M}{{\rm d}t}\right\rangle_\mathrm{dyn} = \frac{M(t) - M(t-t_\mathrm{dyn})}{t_\mathrm{dyn}} \; ,$$ Consequently, the model becomes independent of the time resolution of the simulation as long as the evaluation timesteps are smaller than $t_\mathrm{dyn}$. This growth rate can be split into two contributions [@Diemer:2013aa]: the physical growth of the halo due to accretion (i.e. the mass growth inside a fixed radius), and pseudo evolution due to the growing virial radius over cosmic time as the background density decreases. The growth rate due to accretion can be combuted as $$\frac{{\rm d}M}{{\rm d}t} = \left\langle\frac{{\rm d}M}{{\rm d}t}\right\rangle_\mathrm{dyn} - 4\pi \, R_\mathrm{v}^2 \, \rho(R_\mathrm{v}) \, \left\langle\frac{{\rm d}R_\mathrm{v}}{{\rm d}t}\right\rangle_\mathrm{dyn} \; ,$$ where $R_\mathrm{v}$ is the virial radius of the main progenitor and $\rho(R_\mathrm{v})$ is its density at the virial radius [c.f. eqn. 9 in @Diemer:2013aa]. The density for a @Navarro:1997aa profile can be computed using the halo scale length extracted from the simulation. Here we are only interested in the infalling baryons that will be available for galaxy formation. However, unlike dark matter, baryons do not splash back beyond the virial radius, but typically remain inside. The rate of baryons becoming available for galaxy formation at the virial radius can thus be slightly higher than $f_\mathrm{b} \dot M$ in massive systems, more closely following the halo growth rate without the pseudo evolution correction. We have tested our model for standard growth rates and growth rates corrected for pseudo evolution and find that the results are identical except for a small shift (10 per cent) in the normalisation of the conversion efficiency. All results reported in this work are based on growth rates corrected for pseudo evolution. The baryonic growth rate describes how much material is becoming available for star formation, so we require it to be non-negative. Negative $\dot M$ will occur when haloes are stripped. In this case we set the growth rate to zero. ### The instantaneous baryon conversion efficiency {#sec:epsilon} The instantaneous baryon conversion efficiency $\epsilon$ describes how efficiently gas that is becoming available according to the baryonic growth function are converted into stars. This efficiency parametrises the effects of physical processes that determine how gas is converted into stars, i.e. gas cooling, star formation, and various feedback processes. Therefore, it can be a function of many parameters and can become arbitrarily complex. Following the philosophy of our approach to find the simplest model that is able to reproduce a large number of observations, we assume that the efficiency only depends on halo mass $M$ and redshift $z$. Comparing the observed SMF and the computed HMF, simple empirical models find that the integrated baryon conversion efficiency (i.e. the ratio between all stellar mass and baryonic mass in a halo $m/m_\mathrm{b}$) is a strong function of halo mass at any redshift. While peaking at a halo mass of $\log_{10}(M/{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi})\approx 12$, the integrated efficiency declines towards both lower and higher halo masses. In the currently favoured picture, feedback from supernovae can drive massive galactic winds that are able to eject gas from the shallow potential wells of low mass haloes, reducing the availability of fuel for star formation. In massive haloes it is believed that feedback from active galactic nuclei (AGN) can heat the gas in the halo and prevent it from cooling. Also gravitational heating may play a role in further preventing the gas from cooling. The halo mass where galaxy formation is most efficient can then be understood as the mass where the combination of the various physical processes that reduce the efficiency is minimal integrated over time. However, the integrated efficiency is still several steps away from the instantaneous efficiency as it includes the effects of stellar mass loss and galaxy mergers, so it may have a different dependence on halo mass and redshift. ![Illustration of the dependence of the instantaneous baryon conversion efficiency $\epsilon$ on halo mass $M$. The efficiency peaks close to the characteristic halo mass $M_1$ where it has the normalisation $\epsilon_\mathrm{N}$. Towards low masses the efficiency decreases with the slope $\beta$ while towards high masses it decreases with the slope $\gamma$. []{data-label="fig:epsilon"}](sketch.pdf){width="48.00000%"} To find a suitable parameterisation of the instantaneous baryon conversion efficiency, we used the predictions by @Moster:2013aa and find that at every redshift the dependence of the instantaneous efficiency on halo mass can be described by a double power law as is typically used for the integrated efficiency. This is consistent with the results of @Behroozi:2013ab [c.f. their figure 2]. Therefore we adopt the parameterisation used for the integrated efficiency introduced in @Moster:2010aa: $$\label{eqn:epsilon} \epsilon(M,z) = 2 \;\epsilon_\mathrm{N} \left[ \left(\frac{M}{M_1}\right)^{-\beta} + \left(\frac{M}{M_1}\right)^{\gamma}\right]^{-1} \; .$$ It is governed by four free parameters: the normalisation $\epsilon_\mathrm{N}$, the characteristic mass $M_1$ where the efficiency is equal to its normalisation, and the two slopes $\beta$ and $\gamma$ that determine how the efficiency decreases at low and high mass, respectively. An illustration of the dependence of the efficiency on halo mass is shown in Figure \[fig:epsilon\]. Baryon conversion is most efficient at a halo mass close to the characteristic one: $$\label{eqn:effmax} M_\mathrm{max}=M_1 \left(\frac{\beta}{\gamma}\right)^{1/(\beta+\gamma)} \; .$$ We typically expect the values of $\beta$ and $\gamma$ to be positive, i.e. at low masses the efficiency increases with increasing mass, while at high masses the efficiency decreases with increasing mass. However, we impose no a-priori restrictions on the values of the slopes, such that the parameterisation is very flexible and can adopt to a wide range of observational constraints. As the integrated baryon conversion efficiency depends on redshift [e.g. @Moster:2013aa], we allow the parameters of the instantaneous baryon conversion efficiency to vary with redshift. As before we try to find a parameterisation that is as simple as possible. For each parameter we adopt a linear dependence on the scale factor $a=(z+1)^{-1}$, and use model selection statistics to determine if this constitutes an improvement over a redshift-independent parameterisation (see Appendix \[sec:bayes\]). We find that the simplest model that can reproduce the data has redshift-dependence in $M1$, $\epsilon_\mathrm{N}$, and $\beta$ while the high mass slope $\gamma$ is constant over cosmic time. We hence adopt the following parameterisations: $$\begin{aligned} \log_{10} M_1(z)& = M_0 + M_\mathrm{z}(1-a) = M_0 + M_\mathrm{z}\frac{z}{z+1} \; ,\\ \epsilon_\mathrm{N}(z)& = \epsilon_0 + \epsilon_\mathrm{z}(1-a) = \epsilon_0 + \epsilon_\mathrm{z}\frac{z}{z+1} \; ,\\ \beta(z)& = \beta_0 + \beta_\mathrm{z}(1-a) = \beta_0 + \beta_\mathrm{z}\frac{z}{z+1} \; ,\\ \gamma(z)& = \gamma_0 \; .\end{aligned}$$ We do not impose any a priori constraints on these parameters, except for physical boundary conditions, e.g. $\epsilon_o+\epsilon_\mathrm{z}>0$. The instantaneous baryon conversion efficiency of any halo thus only depends on its mass and redshift, and no artificial scatter is added to it. This means that two haloes with the same mass at a given redshift will have the same instantaneous efficiency. However, the integrated efficiency and hence the stellar mass will depend on the full formation history of the halo, such that scatter is introduced self-consistently. ### The build-up of stellar mass {#sec:buildup} Having specified how we compute the baryonic growth rate $\dot m_{\mathrm b}$ and the instantaneous baryon conversion efficiency $\epsilon(M,z)$, we can now calculate the SFR of the central galaxy $\dot m_(M,\dot M,z)$ in each halo using equation ($\ref{eqn:sfrcen}$). The stellar mass $m_$ of each galaxy will then grow by both star formation given by $\dot m_$ (in-situ) and by the assembly of stars that formed outside the galaxy and are being accreted given by $\dot m_\mathrm{acc}$ (ex-situ). Moreover, we have to take into account the stellar mass that is being lost as a consequence of dying stars $\dot m_\mathrm{loss}$. The stellar mass at any time $t$ can then be calculated as $$\begin{aligned} \label{eqn:mstar} m_(t)& = \int_0^t \mathrm{d}t^\prime \;\left[ \dot m_(t^\prime) - \dot m_\mathrm{loss}(t^\prime) + \dot m_\mathrm{acc}(t^\prime)\right] \nonumber\\ & = \int_0^t \mathrm{d}t^\prime \;\left[ \dot m_(t^\prime) \cdot [1-f_\mathrm{loss}(t-t^\prime)] + \dot m_\mathrm{acc}(t^\prime)\right] \nonumber\\ & = m_\mathrm{SF} (t)+ m_\mathrm{acc} (t) \, ,\end{aligned}$$ where $f_\mathrm{loss}(t)$ is the fraction of mass lost by a single stellar population with an age t. Using the FSPS package [@Conroy:2010aa; @Conroy:2010ab] to calculate the rate of stellar mass loss for a @Chabrier:2003aa IMF, the fraction of lost stars is well-fit by $$\label{eqn:massloss} f_\mathrm{loss}(t) = 0.05 \ln \left(1+ \frac{t}{1.4~\mathrm{Myr}} \right) \; .$$ Note that we do not employ the instantaneous recycling approximation (where $f_\mathrm{loss}$ is constant), but compute the mass loss as a function of time. The first term in equation (\[eqn:mstar\]) can be integrated as $m_\mathrm{SF}(t) = \sum_{t_i<t} \dot m_{,i} \, \Delta t_i \, [1-f_\mathrm{loss} (t-t_i)]$. The second term depend on how accreted stars are added to the central galaxy during a merger. We discuss the details of merging galaxies in section \[sec:merging\]. ### Star formation in growing haloes {#sec:modelcentral} The model specified so far can be used to explain the growth of isolated central galaxies living in monotonically growing dark matter haloes. Figure \[fig:model\] illustrates this for four dark matter haloes with idealised average growth histories for different $z=0$ virial masses. For this illustration we have used the best-fit parameters that will be derived later, but a departure from their exact values will not change the qualitative behaviour of the model. In the top panel the redshift evolution of the halo mass is indicated by the lines of different colours. The background colour gives the instantaneous baryon conversion efficiency $\epsilon(M_{\rm h},z)$ for a given halo mass and redshift. The most massive halo with a $z=0$ virial mass of $10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ (solid black line) has a path that crosses the peak of the conversion efficiency already at $z\approx 4$, while less massive haloes will reach their peak efficiency later, e.g. a halo with a $z=0$ virial mass of $10^{11}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ (dashed blue line) reaches its peak efficiency only around $z\approx 0$. ![Illustration of the growth of haloes and their galaxies. The first panel shows the evolution of the halo mass for four idealised haloes with $z=0$ masses of $\log_{10}(M/{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi})=11,12,13$ and $14$. The background colour denotes the instantaneous baryon conversion efficiency $\epsilon(M,z)$ for a given halo mass and redshift. The second panel shows the corresponding growth rates. In the third panel the star formation rate for the galaxies within the four haloes is shown as computed with equation (\[eqn:sfrcen\]). It peaks close to the redshift where the halo mass passes through the maximum efficiency. The fourth panel gives the time integrated stellar mass, assuming a mass loss rate of 40 per cent. []{data-label="fig:model"}](model.pdf){width="47.00000%"} The second panel gives the halo growth rates $\dot M_\mathrm{h}$ for each system. Typically the growth rates of massive haloes peak late, i.e. they assemble most of their mass at low redshift, while haloes with lower mass have growth rates that peak earlier. However, the growth histories are rather flat once the peak has been reached, i.e for $z\lesssim5$. The third panel shows the SFR of each halo’s central galaxy $\dot m_\mathrm{}$, as computed with equation (\[eqn:sfrcen\]), i.e. the product of the universal baryon fraction $f_\mathrm{b}$, the halo’s growth rates $\dot M_\mathrm{h}$ (from panel 2), and the instantaneous baryon conversion efficiency $\epsilon(M_{\rm h},z)$ at that halo mass and redshift (from top panel). Since the variation in the halo growth rate is typically much smaller than the variation in the conversion efficiency, the change in the SFR is mostly dominated by the change in the conversion efficiency. At high redshift, when all haloes still have low mass, their conversion efficiency is rather low, leading to a low SFR for the central galaxies. As the halo mass grows, the conversion efficiency grows as well leading to higher SFRs. Once the halo mass has reached the value where the conversion efficiency peaks, the SFR reaches its maximum (with a slight modulation due the non-constant growth rate). As the halo becomes more massive, the conversion efficiency decreases again, leading to lower SFRs, so that the central galaxies stop forming stars. As massive haloes pass through the peak of the conversion efficiency earlier, their central galaxies’ SFR reaches its maximum at a higher redshift than systems with a lower mass. As a result, massive galaxies typically stop star formation earlier than low mass galaxies. Finally, the bottom panel gives the stellar mass of each halo’s central galaxy $m_\mathrm{}$, which has been computed by integrating the SFH. For this illustration only, mergers have been neglected, and a constant stellar mass loss of 40 per cent has been assumed. Typically, central galaxies in massive haloes have formed most of their in-situ mass at high redshift (e.g. $z\approx 3$ for $M_\mathrm{h}=10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$), and can then only grow by mergers. On the other side, central galaxies in low mass haloes have formed most of their stellar mass relatively recently. Satellite galaxies {#sec:modelsatellites} ------------------ Having specified the model for isolated, monotonously growing haloes (i.e. for central galaxies), we now have to consider situations where the halo is losing mass, and eventually merging with a larger halo (i.e. satellite galaxies). In [Emerge]{}, we make no formal distinction between haloes and subhaloes (or central and satellite galaxies) based any specific halo radius. Instead, the only input in our model is whether a halo grows in mass or not. We consider three effects that can impact the growth of galaxies in haloes that have stopped growing: quenching after the halo has not grown for some time, stripping once the halo has lost a significant fraction of its mass, and merging once a subhalo has lost its orbital energy. As with the model for growing haloes, we have chosen the simplest model that is able to reproduce all observational data. However, it would be straight forward to increase the model complexity if new observational data require it. ### Quenching {#sec:quenching} When a halo starts being accreted by a larger halo its own growth rate begins to decline. Eventually the halo reaches its peak mass, typically still far outside the virial radius of the larger halo, after which tidal forces strip mass from the halo. The galaxy in such a halo then experiences a reduced gas infall rate, such that the cold gas reservoir will eventually be used up by star formation, and the galaxy will be quenched. Using a galaxy group catalog from the Sloan Digital Sky Survey, together with cosmological simulation, @Wetzel:2013aa study the SFHs and quenching timescales of satellite galaxies. They find a ‘delayed-then-rapid’ quenching scenario in which the SFRs evolve unaffected for a few Gyr after infall, after which star formation quenches rapidly. In this way, satellites can grow significantly in stellar mass after infall, nearly identical to central galaxies. Moreover, they find that quenching time-scales are shorter for more massive satellites but do not depend on host halo mass. Using these empirical results, we construct our quenching model as follows: at each time step in the formation history of a halo, we record its previous maximum virial mass $M_\mathrm{peak}(t)$, and the time at which this mass was reached $t_\mathrm{peak}(t)$. If the current mass is lower than the peak mass ($M<M_\mathrm{peak}$), we keep the SFR of its galaxy constant, i.e. we use $\dot m (t_\mathrm{peak}$). After a time $\tau$ has elapsed and the halo mass is still below its previous peak mass, the SFR of the galaxy is set to 0. We parameterise this quenching time with respect to the halo’s dynamical time $t_\mathrm{dyn}$ (i.e. proportional to the current Hubble time), and allow for longer quenching times for low mass satellites: $$\label{eqn:satquenching} \tau = t_\mathrm{dyn} \cdot \tau_0 \left(\frac{m_}{10^{10}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}}\right)^{-\tau_\mathrm{s}} \; .$$ We further require a minimum quenching time of $\tau_0 \cdot t_\mathrm{dyn}$. The satellite quenching model thus has two free parameters, the normalisation $\tau_0$ describing the quenching time for massive galaxies with a stellar mass of $m_* \ge 10^{10}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$, and the slope $\tau_\mathrm{s}$ that describes how the quenching time of low mass galaxies changes with stellar mass. Both parameters are mostly constrained by the observed fraction of quenched galaxies as a function of stellar mass at different redshifts. For a positive $\tau_\mathrm{s}$, we get longer quenching times for less massive satellites [as found by @Wetzel:2013aa]. Using a satellite mass dependent parameterisation leads to a significantly better fit to the quenched fractions of low mass galaxies (see Appendix \[sec:bayes\] for more details). We also tried a model in which we do not keep the SFR constant after the peak halo mass has been reached, but let it decay exponentially until the quenching time has elapsed (when it is set to 0). However, we did not find any significant improvement over our standard model indicating that the data do not require this decay (see Appendix \[sec:bayes\]). If the halo mass becomes larger than the previous peak mass, the SFR is again determined by the halo growth rate and the instantaneous conversion efficiency (eqn. $\ref{eqn:sfrcen}$). ### Stripping {#sec:stripping} While a subhalo orbits within its host halo, strong gravitational tidal forces strip mass from the outer regions of the subhalo, lowering its own gravitational potential. As the stars in its galaxy are centrally concentrated and more tightly bound than the dark matter, the stellar mass of the satellite changes only slightly until most of the dark matter has been stripped off. However, at some point, the halo mass has been lowered enough that it becomes comparable to the mass of the galaxy, and the gravitational potential of the halo is not strong enough to protect the galaxy from becoming stripped as well. Consequently, once the halo has lost enough mass, the stars in its centre get tidally stripped to the host halo, where they are added to the stellar halo, or equivalently to the intra-cluster mass (ICM). We implement this process by comparing the halo mass at each time to the peak mass of the halo through its history. If the current halo mass $M$ is smaller than a fraction $f_\mathrm{s}$ of the peak mass $M_\mathrm{peak}$, $$\label{eqn:satstripping} M < f_\mathrm{s} \cdot M_\mathrm{peak}$$ we move all stellar mass $m$ within this halo to the ICM, $m_\mathrm{ICM,new}=m_\mathrm{ICM,old}+m$, and then set $m=\dot m=0$. The stripping parameter is mostly constrained by small-scale clustering, as early (late) stripping leads to less (more) satellites close to the central, such that the one-halo term of the galaxy two-point correlation function is lowered (enhanced). We also tried an alternative stripping model in which the galaxy is stripped, once the halo mass has been reduced below a given factor of the stellar mass of the galaxy in the centre of the halo ($M < f_\mathrm{s} \cdot m$), but this model resulted in a worse fit to the data (see Appendix \[sec:bayes\]). This is one of two channels by which the ICM can grow (the other being merging). The gravitational potential of a subhalo is too weak to support its own stellar halo. Thus, once a halo becomes a subhalo, we move all the mass in its ICM to the ICM of its host halo. ### Merging {#sec:merging} Eventually a subhalo will have lost all of its orbital energy due to dynamical friction and the satellite galaxy will merge with the central galaxy. We assume that during such a merger a fraction of satellite stars $f_\mathrm{esc}$ can escape from the central galaxy and ends up in the halo as diffuse stellar material not detected in standard surveys. The merger remnant will then have stellar mass of $$\label{eqn:satmerging} m_\mathrm{rem} = m_\mathrm{cen} + m_\mathrm{sat} \cdot (1-f_\mathrm{esc}) \; ,$$ while the ICM of the host halo will grow by $m_\mathrm{ICM,new}=m_\mathrm{ICM,old}+f_\mathrm{esc} m_\mathrm{sat}$. We treat the escape fraction $f_\mathrm{esc}$ as a free parameter. It is mostly constrained by the evolution of the massive end of the SMF at low redshift. A low escape fraction will lead to a strong evolution, while a high escape fraction will lead to little growth in the massive end. Note that the exact value of $f_\mathrm{esc}$ will depend on how the stellar mass function has been derived. If stellar masses are computed from Petrosian magnitudes, most of the mass in the outskirts of the galaxy will be seen as belonging to the ICM, resulting in a higher value for the escape fraction. If instead the stellar masses are computed from fits to the light profiles, less mass will be classified as ICM, so that the escape fraction is slower (see section \[sec:smfz0\] for more discussion). Due to the finite mass resolution of the simulations, sub-haloes can no longer be identified once tidally stripped below the resolution limit. Since mass loss can be substantial, this is important even for fairly massive subhaloes, and a special treatment of these so-called ‘orphans’ becomes necessary. We determine the orbital parameters at the last moment when a subhalo is identified in the simulation and use them in the dynamical friction estimate given by @Boylan-Kolchin:2008aa, which is applicable at radii $R<R_\mathrm{v}$. We keep the disrupted subhalo and the associated satellite until the dynamical friction time $t_\mathrm{df}$ has elapsed, and assume that only then does it truly merge with the main halo. While orbiting, the mean distance between the satellite and the central decays proportional to $f_\mathrm{dec}=\sqrt{1-\Delta t/t_\mathrm{df}}$, where $\Delta t$ is the time since the subhalo was last identified [see section 8.1.1 of @Binney:1987aa]. We therefore place the orphan galaxy randomly on a sphere around the central galaxy with a radius of $r=r_0f_\mathrm{dec}$, where $r_0$ is the distance when the subhalo was last identified. To compute if the galaxy gets tidally stripped, we extrapolate the halo mass linearly. Finally, if a host halo of an orphan merges with a larger halo and becomes a subhalo itself, we reset the merger clock, i.e. we recompute the dynamical friction time with respect to the new central galaxy, and assume the orphan will merge with it once this new time has elapsed. Obtaining mock observations {#sec:mocks} --------------------------- Parameter Best-fit Upper 1$\sigma$ Lower 1$\sigma$ ------------------- ---------- ----------------- ----------------- $M_0$ 11.339 +0.005 -0.008 $M_z$ 0.692 +0.010 -0.009 $\epsilon_0$ 0.005 +0.001 -0.001 $\epsilon_z$ 0.689 +0.003 -0.003 $\beta_0$ 3.344 +0.084 -0.101 $\beta_z$ -2.079 +0.127 -0.134 $\gamma_0$ 0.966 +0.002 -0.003 $f_\mathrm{esc}$ 0.388 +0.002 -0.002 $f_\mathrm{s}$ 0.122 +0.001 -0.001 $\tau_0$ 4.282 +0.015 -0.020 $\tau_\mathrm{s}$ 0.363 +0.014 -0.014 : Fitting results from MCMC[]{data-label="tab:bestfit"} \ Notes: All masses are in units of [\_ ${\rm M}_{\odot}$ ]{}. Having specified the model and its free parameters, we are now able to populate all merger trees of the simulation with galaxies. To determine the values and uncertainties of our parameters, we need to compare the predictions of the model to observations. As we aim to avoid degeneracies between model parameters, we choose a set of observations that have a different constraining power for each parameter. The idea here is to have a subset of observations that mostly constrains one given parameter, while another subset of observations constrains another parameter, and so forth. To constrain the overall evolution of the instantaneous conversion efficiency $\epsilon$ and specifically its characteristic mass ($M_0$ and $M_z$), we use the SMFs $\Phi$ up to $z=8$. The CSFRD [\^\* $\dot \rho^$ ]{}up to $z=11$ mostly constrains the evolution of the normalisation of $\epsilon$ ($\epsilon_0$ and $\epsilon_z$). The sSFRs $\Psi$ for galaxies of different stellar masses up to $z=9$ can constrain the evolution of the slopes of $\epsilon$ ($\beta_0$, $\beta_z$, and $\gamma_0$). The low redshift evolution of the massive end of the SMF together with the sSFR of massive galaxies at low redshift put strong constraints on the escape fraction $f_\mathrm{esc}$. The quenching parameters $\tau_0$ and $\tau_\mathrm{s}$ can be constrained with the fraction of quenched galaxies [ ${f_\mathrm{q}}$ ]{}as a function of stellar mass up to $z=4$. Finally, galaxy clustering on small scales, i.e. the projected two-point correlation function [ ${w_\mathrm{p}}$ ]{}for different stellar masses at small radii, put strong constraints on the stripping parameter $f_\mathrm{s}$. We construct our mock SMFs at the same redshift and mass bins as the observations, so that we can directly compare each individual data point $\Phi_i(m_i,z_i)$ . The CSFRD is computed at each simulation redshift. We interpolate between these redshifts using cubic splines to get the data points ${\ifmmode \dot \rho^ \else $\dot \rho^$ \fi}_i(z_i)$ which we compare to an observation at a specific redshift $z_i$. To compute the sSFR, we bin in stellar mass and redshift, and calculate the mean sSFR for each grid point. We use a 2d cubic spline interpolation on this grid to get $\Psi_i(m_i,z_i)$, which can be compared to a specific observed data point at a given redshift and stellar mass. We compute the fraction of quenched galaxies ${\ifmmode {f_\mathrm{q}} \else ${f_\mathrm{q}}$ \fi}_i(m_i,z)$ directly for the same stellar mass and redshift bins as the used observations. For that we divide the number of all quenched galaxies that lie in a specific bin by the total number of galaxies in the same bin. We define a galaxy to be quenched if its sSFR is below a redshift dependent threshold given by $\Psi<0.3t_\mathrm{H}^{-1}$, where $t_\mathrm{H}$ is the Hubble time at that redshift [see e.g. @Franx:2008aa]. Finally, we calculate the $z=0$ galaxy two-point correlation function in a given stellar mass bin using kd-trees following @Moore:2001aa, which we found to be very effective. Once the tree is constructed, we count the number of pairs in a distance bin $dd(r)$, and compute the average number of pairs for a random distribution $N_\mathrm{p}(r) = 2 \pi N^2 r^2 \Delta r L_\mathrm{box}^{-3}$, where $N$ is the total number of galaxies in a stellar mass bin. The real space correlation function is then given by $\xi(r)=dd(r)/N_\mathrm{p}(r)-1$, and we can compute the projected correlation function ${\ifmmode {w_\mathrm{p}} \else ${w_\mathrm{p}}$ \fi}_i({\ifmmode {r_\mathrm{p}} \else ${r_\mathrm{p}}$ \fi}_i,m_i)$ at the same projected radii [ ${r_\mathrm{p}}$ ]{}as the observations, by integrating $\xi(r)$ along the line of sight: $$\label{eqn:xi2wp} {\ifmmode {w_\mathrm{p}} \else ${w_\mathrm{p}}$ \fi}({\ifmmode {r_\mathrm{p}} \else ${r_\mathrm{p}}$ \fi},m) = 2 \int_{{\ifmmode {r_\mathrm{p}} \else ${r_\mathrm{p}}$ \fi}}^\infty {\rm d}r \, r \,\xi(r,m) \, \left(r^2-{\ifmmode {r_\mathrm{p}} \else ${r_\mathrm{p}}$ \fi}^2\right)^{-1/2} \;.$$ Fitting the model parameters {#sec:modelfitting} ---------------------------- ![Cosmic star formation rate density as a function of redshift. The symbols of different colours denote observational estimates derived with different methods (c.f. Table \[tab:obscsfrd\]). The black solid line shows the model prediction for the best-fit parameters for the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box, while the green dashed line shows the model with the same parameters for the $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box. The red/blue solid lines show the contribution from galaxies and their progenitors that are quenched/star-forming at $z=0$. []{data-label="fig:csfrd"}](csfrd.pdf){width="48.00000%"} ![image](ssfr.pdf){width="99.00000%"} For a specific model, i.e. a given set of parameters $\vec \theta$, we can compute the mock observations $\vec \mu (\vec \theta)$ and compare them to the observations $\vec \omega$ to get the difference $\vec \Delta$: $$\begin{aligned} \vec \mu (\vec \theta)& = (\log_{10}\Phi_\mathrm{m},\log_{10}{\ifmmode \dot \rho^ \else $\dot \rho^$ \fi}_\mathrm{m},\log_{10}\Psi_\mathrm{m},{\ifmmode {f_\mathrm{q}} \else ${f_\mathrm{q}}$ \fi}_\mathrm{m},{\ifmmode {w_\mathrm{p}} \else ${w_\mathrm{p}}$ \fi}_\mathrm{m})\\ \vec \omega& = (\log_{10}\Phi_\mathrm{o},\log_{10}{\ifmmode \dot \rho^ \else $\dot \rho^$ \fi}_\mathrm{o},\log_{10}\Psi_\mathrm{o},{\ifmmode {f_\mathrm{q}} \else ${f_\mathrm{q}}$ \fi}_\mathrm{o},{\ifmmode {w_\mathrm{p}} \else ${w_\mathrm{p}}$ \fi}_\mathrm{o})\\ \vec \Delta& = \vec \omega - \vec \mu (\vec \theta) \; .\end{aligned}$$ We can then compute $$\label{eqn:chi2} \chi^2 = \vec\Delta^T \, C^{-1} \, \vec\Delta \; ,$$ where $C$ is the covariance matrix of the observed data, and assign a likelihood to the model $$\label{eqn:likelihood} \mathcal{L} = \exp \left( -\chi^2/2\right) \; .$$ If available, we use the full covariance matrix, otherwise we calculate it as $C = \mathrm{diag}(\sigma^2_1,...,\sigma^2_N)$, where $\sigma_i$ is the uncertainty of the $i$th data point. Having assigned a probability to each possible model, we can now try to find the best-fit model, i.e. the model that maximises the likelihood (and minimises $\chi^2$), and derive the model uncertainty, i.e. the 1$\sigma$ errors of the parameters. Before sampling the posterior probability distribution, we found it very effective to first find the best-fit model using a dedicated method. We employed the [Hybrid]{} method presented in @Elson:2007aa, which combines elements of simulated annealing, Markov-Chain Monte Carlo, and particle-swarm methods. In [Hybrid]{} a set of chains is run according to the Metropolis-Hastings algorithm, but the step size of each walker is adjusted based on the ratio of the $\chi^2$ of a walker and the average $\chi^2$ at this step (i.e. how well the walker does compared to the others). Furthermore, the step size of all walkers is adjusted based on the ratio of the average $\chi^2$ at this step and the initial average $\chi^2$ (i.e. how well the walkers do compared to their starting position). To find the maximum likelihood, we ran 14 sets of chains starting from different regions in parameter space, each with 30 walkers and 2,000 steps (so 60,000 models each). This was sufficient, so that all sets of chains found the same minimum. Once we have identified the global minimum, we are interested in the posterior probability distribution around this point to derive the parameter uncertainties. We sample the posterior using the affine invariant ensemble sampler for MCMC presented by @Goodman:2010aa. In this method an ensemble of walkers is evolved by proposing a new position for a walker $k$ stretching along the line to another random walker $j$: $\vec\theta_{k,\mathrm{new}} = \vec\theta_j + Z(\vec\theta_k-\vec\theta_j)$, with the random number $Z$ drawn from the distribution $g(z)=1/\sqrt{z}$. The new position is accepted with the probability $q=Z^{N-1}\mathcal{L}(\vec\theta_{k,\mathrm{new}})/\mathcal{L}(\vec\theta_k)$, where $N$ is the number of parameters. After a few autocorrelation times the current walker positions represent an independent sample of the posterior distribution. We use 10 ensembles of 100 walkers each, which we initialise in a tight sphere around the previously found best-fit model, and evolve each walker for 1,000 steps, i.e. 1,000,000 models are computed. At this point the 1$\sigma$ parameter errors derived from the walker distribution have converged. We extract the errors from the final ensembles. Fitting results {#sec:fittingresults} --------------- The best-fit parameters and their 1$\sigma$ uncertainties are presented in Table \[tab:bestfit\]. The characteristic halo mass of the instantaneous conversion efficiency $M_1$ decreases from $1.1\times10^{12}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ at high redshift to $2.2\times10^{11}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ at $z=0$. The peak efficiency decreases from 70 per cent to lower than one per cent at $z=0$. The low mass slope of the efficiency steepens drastically from 1.3 to 3.3, leading to much lower efficiencies for low mass galaxies at low redshift. The constant high mass slope is found to be about 1. The escape fraction from mergers is constrained at just under 40 per cent, which is considerably higher than the 20 per cent we assumed in @Moster:2013aa. The stripping parameter $f_\mathrm{s}$ is slightly larger than 10 per cent, which means that satellite galaxies get stripped to the ICM once its subhalo’s mass has reached a tenth of its peak value. The satellite quenching times are found to be about 4 dynamical times for massive galaxies with $m\ge10^{10}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$, and considerably longer for galaxies with lower mass (e.g. about 10 dynamical times for galaxies with $m=10^{9}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$). We discuss the correlations between the model parameters in Appendix \[sec:cov\]. In the right panel of Figure \[fig:box\] we show the distribution of galaxies in the simulation box with $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ side length, calculated with our best-fit model. As in the left panel, the volume has been remapped into a a sheet with $6{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ thickness and $750{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ width. Each dot corresponds to a galaxy, while the size of the dots corresponds to the stellar mass and the colour indicates the specific star formation rate as given by the colour bar. Massive quenched (red) galaxies are clustered at the knots of the cosmic web, while active (blue) galaxies are preferentially located along the filaments. ![image](smf.pdf){width="99.00000%"} We first check how well the best-fit model is able to reproduce the data that has been used to fit the parameters to judge if the empirical relations we employed are sensible. In Figure \[fig:csfrd\] we compare the resulting CSFRD (lines) to the data that have been used to fit the model (symbols). Symbols of different colours denote estimates derived with different methods. These data strongly constrain the overall normalisation of the instantaneous conversion efficiency and its redshift evolution. The black solid line shows the model results obtained for the box with $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ side length which has been used to fit the parameters. The green dashed line has been obtained by running the model with the previously fitted parameters on the box with $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ side length, which is not only bigger but also better resolved. This indicates that the model is converged and reproduces the same SFRs independent of the simulation size and mass resolution. The red/blue solid lines show the contribution to the CSFRD from galaxies and their progenitors that are quenched/star-forming at $z=0$. This demonstrates that the progenitors of local star-forming galaxies played a very minor role in the peak of cosmic star formation at $z\sim2-3$, while the main contribution to cosmic star formation beyond $z\ge0.5$ is from galaxies that are now quenched. The SFRs of galaxies strongly depends on their stellar mass and on redshift. Figure \[fig:ssfr\] compares the sSFRs as a function of redshift of observed galaxies (symbols) to the model results (lines) for four stellar mass bins. For this, only galaxies have been selected that have the quoted stellar mass at the quoted redshift. Again, symbols of different colours have been derived with different methods. As these data give the SFR of low-mass and high-mass galaxies individually, they strongly constrain the slopes and the characteristic mass of the instantaneous conversion efficiency and their redshift evolutions. The black solid lines show the best-fit model obtained with the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box, while the green dashed lines show the same model run on the $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box. Also here, the results do not depend on box size and resolution. The model fits the observed data very well, including the plateau at high redshift. The SFRs for massive galaxies at high redshift are considerably lower than the radio observations, but the uncertainty between different observational methods is relatively large in this range as well. Integrating the SFRs, taking into account stellar mass loss and merging, provides us with stellar masses for all galaxies, so that we can compute the stellar mass functions up to high redshift. We compare the observed SMFs (symbols) to the model results (lines) in Figure \[fig:smf\]. These data give strong constraints on the characteristic mass of the instantaneous conversion efficiency and its redshift evolution. The solid black lines show the best-fit model for the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box and the green dashed lines show the same model run on the $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box, indicating that also the stellar masses are converged. The local SMF is reproduced very accurately, while at intermediate redshift ($1<z<3$) the model is on the upper end of the observations around the knee. At high redshift the SMFs are fitted very well. Thus it is possible to fit both observed SFRs and stellar masses if stellar mass loss and merging are taken into account. We therefore conclude that there is no tension for the stellar mass density between integrated SMFs and the integrated CSFRD, in agreement with previous findings [@Moster:2013aa; @Behroozi:2013aa]. ![image](fq.pdf){width="99.00000%"} ![Projected galaxy correlation function for five bins of increasing stellar mass (from top to bottom). The symbols indicate observational estimates [@Li:2006aa; @Guo:2011aa; @Yang:2012aa], and the black and green lines show the model prediction for the best fit parameters for the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ and $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ simulations, respectively. The shaded area indicates the $1\sigma$-confidence levels for the model. []{data-label="fig:wp"}](wp.pdf){width="45.00000%"} To constrain the quenching timescale for satellites we have used the fraction of quenched galaxies as function of stellar mass and redshift. For very short quenching times the quenched fraction will be too high, while for very long quenching timescales they will be too low. We compare the observed data (symbols) that have been used in the fit, to the results of the best-fit model (lines) in Figure \[fig:fq\]. The solid black lines show the best-fit model for the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box and the green dashed lines show the same model run on the $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box. Also for the fraction of quenched galaxies we find that the model has converged. At low redshift the model provides an excellent fit to the data. At intermediate redshift (around $z\sim1$) the fraction of quenched galaxies is slightly underpredicted for intermediate mass galaxies ($10^{10}<m_/{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}<10^{11}$). This corresponds to the stellar mass and redshift range, where the model SMF is on the higher side of the data. As the model gives a very good fit to the observed SFR in this range, we conclude that the observed SMFs and quenched fractions are well consistent with each other, but they are in slight disagreement with the observed SFRs. At higher redshift, the model agrees very well with the observed quenched fractions. We have used the projected correlation function to constrain the stripping parameter of our model. If satellite galaxies are stripped very early (i.e. when their subhalo still has a large fraction of its peak mass), the model correlation function on small scales will be too low compared to the data, while for late stripping (i.e. the satellite exists until its subhalo has lost a large fraction of its peak mass) it will be too high. On large scales, beyond the typical host halo radius for galaxies of a given stellar mass, the projected correlation function is not affected. In Figure \[fig:wp\] we show the observed data that have been used in the fit (symbols), and the results of the best-fit model (lines). The best-fit model for the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box is given by the solid black lines, while the green dashed lines show the same model run on the $200{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box. For both simulations the model agrees remarkably well with the data. On small scales this is the consequence of fitting the stripping parameter. However, as on large scales the correlation function is not impacted by this, the agreement between data and model is a direct result of assigning galaxies of a given stellar mass to haloes of the right mass. Galaxy clustering thus directly follows from halo clustering, which has been found before with simple empirical models, such as subhalo abundance matching. ![image](satpar.pdf){width="99.00000%"} Before we focus on the predictions of the new model and the comparison to previously published results, we demonstrate how the different observational data constrain the individual parameters. While the instantaneous conversion efficiency and its parameters are fixed by the evolution of the CSFRD (normalisation), the sSFRs (slopes) and the SMFs (characteristic halo mass), the satellite parameters are fixed by the fraction of quenched galaxies (quenching), small-scale clustering (stripping), and the low-redshift evolution of the massive end of the SMF (merging). In Figure \[fig:satpar\], we illustrate how each satellite parameter is constrained by one specific observation. In all panels the black lines show the best-fit model and the coloured lines give the results when the relevant parameter is chosen to have a larger and smaller value while keeping all other parameters fixed. The top left panel shows the $z=0$ SMF given different values for the merging parameter $f_\mathrm{esc}$. A smaller value of $f_\mathrm{esc}$ (red line) leads to more stellar mass ending up in the central galaxy after a merger (and less mass going into the ICM), so that massive galaxies, where accretion is a significant growth channel, have even higher stellar masses. This leads to an over-abundance of massive galaxies compared to the best-fit model and the observations. On the other hand, a larger value of $f_\mathrm{esc}$ (blue line), leads to less stellar mass going to the central galaxies and more mass being expelled into the ICM, such that the galaxies at the massive end grow less strongly, leading to fewer massive galaxies. As the SFRs of massive galaxies are directly constrained by observed SFRs, the growth of the massive end of the SMF directly constrains the merging parameter leading to $f_\mathrm{esc}=0.388$ such that almost half of the mass of a satellite escapes into the stellar halo. The bottom left panel of Figure \[fig:satpar\] shows the effects on the fraction of galaxies as function of stellar mass when the quenching parameters are varied. Keeping all parameters fixed (including the quenching slope $\tau_\mathrm{s}$), a higher normalisation $\tau_0$ (blue line) leads to longer quenching times for all satellites. As they can retain star formation for a longer time, this results in lower quenched fractions compared to the best-fit model (and the data) for all stellar masses. For a lower normalisation (red line), the satellites become quenched shortly after their halo reaches peak mass (i.e. when the galaxy becomes a satellite). Consequently this leads to higher fractions of quenched galaxies for all stellar masses. The normalisation is thus directly constrained by the quenched fractions, and we find a best-fit value of $\tau_0=4.282$ which means that in our model, massive satellites become quenched at about 4 dynamical halo times after their halo reached its peak mass. At $z=1$ and $z=0$ this corresponds to $\sim4{\ifmmode {\rm~Gyr} \else ${\rm~Gyr}$\fi}$ and $\sim8{\ifmmode {\rm~Gyr} \else ${\rm~Gyr}$\fi}$, respectively. A higher value for the quenching slope (cyan line) while keeping all other parameters fixed (including the normalisation $\tau_0$), leads to longer quenching times for low-mass galaxies. The fraction of low-mass quenched galaxies therefore is reduced compared to the best-fit model and the observations. At the massive end this does not change the quenched fraction, as we have fixed a minimal quenching time of $\tau_0\cdot t_\mathrm{dyn}$ such that the quenched fraction of massive satellites is only set by the normalisation $\tau_0$. A lower value for the quenching slope (magenta line) results in shorter quenching times for low-mass galaxies, increasing the fraction of low-mass quenched galaxies compared to the best-fit model. Consequently, a flat slope ($\tau_\mathrm{s}=0$), where the quenching time does not depend on the stellar mass of a satellite, leads to an upturn of the quenched fraction below a stellar mass of $m_\approx 10^{10}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$, and to significantly more quenched galaxies than observed. The slope is thus directly constrained by the quenched fractions at low stellar masses, and we find a best-fit value of $\tau_\mathrm{s}=0.363$. Thus, a low-mass satellite with $m_\approx 10^{8.5}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ keeps forming stars 3.5 times longer than massive satellites, corresponding to 14 dynamical halo times after their halo reached its peak mass. While this equals $\sim30{\ifmmode {\rm~Gyr} \else ${\rm~Gyr}$\fi}$ for satellites of this mass falling in at $z=0$, it only equals $\sim7.8{\ifmmode {\rm~Gyr} \else ${\rm~Gyr}$\fi}$ at $z=2$ so that enough time passes to quench them before $z=0$. Finally, the right-hand-side panels of Figure \[fig:satpar\] show the effects on the projected galaxy correlation function $w_\mathrm{p}(r_\mathrm{p})$ when the stripping parameter is varied. Each panel shows a different stellar mass bin. A smaller value of $f_\mathrm{s}$ (red line) means that satellite galaxies are stripped to the main halo once their subhalo has lost a larger fraction of its peak mass compared to the best-fit case. Therefore, the satellite has more time to sink closer to the central galaxy due to dynamical friction. This results in more close pairs and more power on small scales boosting the one-halo term of the auto-correlation function. Consequently, the projected correlation function is higher on small scales. On the other hand, a larger value of $f_\mathrm{s}$ (blue line) means that satellites are stripped when their subhalo has lost a smaller fraction of its peak mass. They are therefore stripped faster and do not have as much time to sink closer to the central galaxy resulting in less close pairs and reduced small-scale power. This leads to a lower projected correlation function on small scales. The stripping parameter is thus directly constrained by small-scale clustering leading to $f_\mathrm{s}=0.122$ such that satellites get stripped to the ICM once their subhalo has been stripped to 12 per cent of its peak mass. In summary we note that the empirical model fulfils its main purpose: to follow the stellar content of dark matter haloes over cosmic time such that all relevant observational data is reproduced. The model has been specifically designed to do this; for fewer parameters we are not able to fit all observations simultaneously, while for more parameters the agreement with the data does not improve significantly (see appendix \[sec:bayes\] for more details on the model selection process). The presented model is therefore the most simple model that is able to explain the observations, or phrased differently, the currently available statistical data does not provide any stronger constraints on galaxy formation than what has been implemented in this empirical model. When new data becomes available, we will be able to test if the model is consistent with the new data. If the model cannot explain these, we can increase the complexity of the model and in this way use the data to increase our knowledge about the formation of galaxies. We can also use the empirical model as a constraint for more detailed hydrodynamical simulations and in this way test whether our current understanding of the physical processes that drive galaxy formation is sufficient to explain the observed data. Results of the model {#sec:results} ==================== The new empirical method that has been presented in the previous section was designed to follow the assembly of galaxies based on the assembly of their dark matter haloes. Its free parameters have been fitted by requiring that a number of statistical observations be reproduced: SMFs, CSFRDs, sSFRs, quenched fractions, and small-scale clustering. We now focus on how the stellar content builds up in detail within different dark matter haloes, and the resulting SHM ratio for quenched and star-forming galaxies. The evolution of the stellar content {#sec:growth} ------------------------------------ Since the SFR of a galaxy in [Emerge]{} depends on both the current mass of its halo (through the efficiency) and its current growth rate, each galaxy has an individual SFH based on the growth history of its halo. This means that in haloes that experience most of their growth early-on, the SFHs also follow this behaviour with higher SFRs at high redshift compared to average systems. In the top panels of Figure \[fig:histA\] we present the resulting SFHs of central galaxies. Each panel shows the SFR as a function of redshift for galaxies in haloes of different $z=0$ mass from $M=10^{11}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ to $10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$. The thin lines are individual tracks for 20 randomly chosen systems for each panel. This shows that although for any given panel the final halo mass of each system is identical, the galaxies at the centre of these haloes can have very different SFHs, especially at high redshift and for low masses. Even though individual SFRs can differ considerably between galaxies in haloes with the same $z=0$ mass, we can compute the average SFHs as function of halo mass. The thick black lines show the median SFRs at each redshift for all central galaxies. This confirms the results of @Moster:2013aa for the dependence of the average SFHs on halo mass. While for low halo masses the SFRs peak late at relatively low values, the SFRs of galaxies in massive haloes peak earlier and at higher values. Typical Milky Way-like galaxies with $M(z=0)=10^{12}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ peak around $z=1$ with SFRs of a few solar masses per year, while massive systems with $M(z=0)=10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ peak around $z=4$ with SFRs of more than $150{\ifmmode M_{\odot}{\rm~yr}^{-1} \else $M_{\odot}{\rm~yr}^{-1}$ \fi}$. Individual galaxies however, can peak significantly earlier or later with much higher or lower SFRs. ![image](histA.pdf){width="98.00000%"} While each galaxy in [Emerge]{} has an individual SFR, we can still group them into galaxies that are star-forming and quenched at $z=0$, and investigate how the SFHs differ. The thick blue and red lines in indicate the median SFHs for central galaxies that are star-forming and quenched at $z=0$, respectively. For all halo masses, galaxies that are quenched at $z=0$ had on average a higher peak SFR than galaxies that are still star forming at $z=0$. This can be explained by the fact that the SFR is directly connected to the halo growth rate. Galaxies that are quenched today live in haloes that have low growth rates at low redshift. As by $z=0$ they need to reach the same virial mass as haloes that have high growth rates at late times, they need to form more mass at early times resulting in high growth rates at high redshift. Consequently, at early times the SFRs are higher for the galaxies in these haloes compared to galaxies that have high SFRs at late times (but low SFRs early-on). Since the peak SFR is reached at high redshift galaxies with low SFRs at late times had a higher peak SFR. Integrating the SFHs over cosmic time while taking into account the effects of stellar mass loss and mergers yields the stellar mass growth histories (eqn. \[eqn:mstar\]). In the bottom panels of Figure \[fig:histA\] we show the resulting growth histories of central galaxies. Each panel shows the stellar mass of the haloes’ central galaxies as a function of redshift for different $z=0$ virial masses. The thin lines again give individual tracks for the same 20 systems that were presented in the top panels, i.e. lines of the same colour represent the same halo. In each panel the final halo mass is the same for each halo, but the stellar mass of their central galaxies can vary significantly between individual systems. This shows that in [Emerge]{}, scatter in stellar mass at fixed halo mass is introduced automatically, without the need to add a random scatter to the average stellar mass as done in standard subhalo abundance matching. This has several reasons. Integrating the SFR of a halo over time shows that the final stellar mass just depends on the integrated efficiency. From equation (\[eqn:sfrcen\]) we get for the final stellar mass: $$m_(M) = f_\mathrm{b} \int_0^M \epsilon(M^\prime,z) \, \mathrm{d}M^\prime \, .$$ If the instantaneous efficiency is independent of redshift, the integral has the same value for each halo with the same final mass $M$. However, as the efficiency is redshift dependent, the final stellar mass depends on which track has been taken through $\epsilon-M$ space (c.f. top panel of Figure \[fig:model\]). If the efficiency is higher at high redshift for example, a halo with a lot of early growth will have an enhanced SFR at high redshift, resulting in a higher total stellar mass compared to a halo that growth late. Moreover, a different amount of merging satellite galaxies will lead to different final stellar masses of central galaxies even if the final halo masses are identical. ![image](histB.pdf){width="98.00000%"} The median stellar mass growth histories at each redshift for all central galaxies at a given $z=0$ halo mass are indicated by the thick black lines. Galaxies in massive haloes thus typically grow by orders of magnitude at early times and then have relatively little growth while galaxies in low mass halo tend to grow most of their stellar mass late. Still, individual galaxies can oppose this trend, e.g. there are low-mass galaxy that have most of their mass already in place at $z>4$ and thus have old stellar populations. Grouping centrals galaxies into systems that are star-forming and quenched at $z=0$ (thick blue and red lines), we find that the final stellar mass at $z=0$ is always larger for quenched systems compared to systems that are still forming stars. Since passive galaxies formed most of their stars at high redshift where the conversion efficiency is higher they were able to obtain higher stellar masses than systems that have formed most of their stars late at lower efficiency. ![image](efficiencyA.jpg){width="98.00000%"} As we can trace the evolution of every galaxy, we can investigate how much stellar mass was formed in-situ, i.e. within each galaxy, and how much was formed in other galaxies that have been accreted. In the top panels of Figure \[fig:histB\] we show the stellar mass as a function of redshift for mass that has formed in-situ (solid lines), and ex-situ (dashed lines), in haloes of different $z=0$ mass. The thin lines are ex-situ tracks for 20 randomly chosen individual galaxies for each panel. As the accreted material is added instantly to the central galaxy during mergers the ex-situ formed mass increases incrementally. The average amount of accreted stellar mass is given by the thick solid lines, and the average stellar mass that has formed within the galaxy is given by the thick dashed lines. The amount of in-situ formed stellar mass is generally much larger than the accreted mass. Only in very massive haloes with $M(z=0)\gtrsim10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ the final accreted mass exceeds the mass formed within the galaxy. We also note that at late times the in-situ formed mass decreases with time due to stellar mass loss from dying stars and low SFRs. Dividing these two components between star-forming and quenched central galaxies, we do not notice any significant variation. Both groups show very similar amounts of in-situ and ex-situ formed stellar mass. [@crrrrrr@]{} $\log_{10} M$ & $\Psi_1$ & $\Psi_2$ & $\Psi_3$ & $\Psi_4$ & $f_1$ & $f_2$\ \ 11.0 & 73.29 & 0.28 & 1.07 & -1.87 & 0.76 & 1.29\ 12.0 & 3.87 & 5.94 & 1.46 & 4.94 & 1.08 & 0.06\ 13.0 & 2.24 & 3.98 & 1.86 & 11.69 & 1.10 & 0.58\ 14.0 & 1.44 & 3.58 & 1.99 &16.26 & 0.94 & 1.43\ \ 11.0 & 147.50 & 2.50 & 0.89 & -1.99 & 1.87 & 0.05\ 12.0 & 7.13 & 7.43 & 1.43 & 4.88 & 1.12 & 0.06\ 13.0 & 2.58 & 4.14 & 1.86 & 11.72 & 1.09 & 0.55\ 14.0 & 1.44 & 3.58 & 1.99 &16.26 & 0.94 & 1.43\ \ 11.0 & 63.00 & -0.04 & 1.04 & -2.02 & 1.06 & 0.01\ 12.0 & 0.98 & 3.02 & 1.54 & 5.03 & 1.09 & 0.07\ 13.0 & 0.51 & 1.96 & 2.45 & 15.76 & 1.05 & 0.61\ \ Notes:* Columns are halo mass at $z=0$ in [\_ ${\rm M}_{\odot}$ ]{}(1), fitting parameters for the star formation histories (2-5), and fitting parameters for the accreted fractions (6-7). To compare the contributions of stellar mass formed in-situ and ex-situ further we present the fraction of accreted stellar mass as a function of redshift for the same $z=0$ halo masses in the bottom panels of Figure \[fig:histB\]. The thin lines show the tracks for the same 20 randomly selected individual galaxies as in the top panels. As mergers are discrete events, the accreted fraction increases instantaneously and then decreases again smoothly due to in-situ star formation. This process can be repeated several times, especially in massive haloes. The average fraction of accreted stellar mass is given by the thick black lines. We find that in low mass haloes the amount of accreted stellar mass is negligible. For a halo with $M(z=0)=10^{11}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ only $\sim 1$ per cent of the total stellar mass at $z=0$ has been accreted. This fraction increases with halo mass but is still just $\sim 2$ per cent at $z=0$ for a typical Milky Way-like galaxy with $M(z=0)=10^{12}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$. Only in massive haloes and at late times, the fraction of accreted stellar mass becomes significant. At $z=0$ an average halo with a mass of $M(z=0)=10^{13}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ has an accreted fraction of $\sim 20$ per cent, and an average halo with $M(z=0)=10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ even has accreted more than 50 per cent of its total mass. Interestingly though, individual galaxies can have drastically different accreted fractions. There are several galaxies in low mass haloes with accreted fractions of more than 10 per cent and up to 50 per cent, while in massive haloes we find a few galaxies that have accreted less than1 per cent of their total stellar mass. We also group the galaxies into star-forming and quenched systems (blue and red lines), but we find no significant difference between them. We find that the SFHs $\Psi(z)$ and the accreted fractions $f_\mathrm{acc}(z)$ for central galaxies in haloes with a given $z=0$ virial mass can be well approximated by the following fitting functions: $$\begin{aligned} \log \Psi(z) &=& -\log\left[ \Psi_1 (z+1)^{-\Psi_2} + \mathrm{e}^{\Psi_3(z+1)-\Psi_4} \right] ,\\ f_\mathrm{acc}(z) &=& f_2 \exp\left[-f_1 (z+1)\right] \, .\end{aligned}$$ The fitting parameters for all, star-forming, and quenched centrals as function of their $z=0$ virial mass are presented in Table \[tab:fittingfunctions\]. The integrated conversion efficiency {#sec:shm} ------------------------------------ The new empirical model describing the growth of galaxies in dark matter haloes is based on the instantaneous baryon conversion efficiency $\epsilon(M,z)$. Together with the growth rate of dark matter haloes it was used to derive SFRs for galaxies at each time, which were then integrated to get stellar masses. As we have shown that this leads to a range of stellar masses at a fixed halo mass, this will result in scatter in the SHM relation $m(M)$ and consequently in the integrated baryon conversion efficiency, which is defined as $\epsilon_\mathrm{int}(M) = m/m_\mathrm{b} = m/(f_\mathrm{b}M)$, where $m_\mathrm{b}$ is the total baryonic mass in a halo and $f_\mathrm{b}$ is the universal baryon fraction. In Figure \[fig:efficiencyA\] we plot the integrated conversion efficiency as a function of peak halo mass for each individual central galaxy in the simulation box. Each panel corresponds to a different redshift from $z=0.1$ (top left) to $z=8$ (bottom right). The colour of each point gives the sSFR of each galaxy ranging from star-forming (blue) to quenched (red) as indicated by the colour bar. The solid lines show the average conversion efficiency and the dashed lines indicate the standard deviation ($1\sigma$ scatter). The resulting average integrated conversion efficiencies for centrals are in very good agreement with previous results for the SHM ratio from subhalo abundance matching. At $z=0.1$ the maximum integrated conversion efficiency is 17 per cent – slightly lower than what has been found in previous empirical models – and it increases to 20 per cent at $z=4$. The corresponding halo mass at the peak is $\log_{10}(M/{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi})=12$ at $z=0.1$, increasing to $\log_{10}(M/{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi})=12.25$ at $z=4$. The low-mass slope is quite steep at low redshift and becomes shallower at high redshift. The high-mass slope is shallower and does not depend on redshift. We find that the majority of galaxies in massive haloes at $z=0.1$ are quenched, while at low halo masses most galaxies are star-forming. However, there are also active galaxies in massive haloes if the halo has a large growth rate at late times, and passive galaxies in low-mass haloes, typically in haloes that have stopped growing. At high redshift, all central galaxies are actively forming stars. ![Probability density function of the integrated baryon conversion efficiency $m/m_\mathrm{b}$ at $z=0.1$. Each line is for a different fixed peak halo mass: $M=10^{11}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ (black), $10^{12}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ (blue), $10^{13}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ (green), $10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ (red). []{data-label="fig:effslice"}](effslice.pdf){width="48.00000%"} [@rrrrrrrr@]{} $z$ & $M_1$ & $\epsilon_\mathrm{N}$ & $\beta$ & $\gamma$ & $M_\sigma$ & $\sigma_0$ & $\alpha$\ \ 0.1 & 11.80 & 0.14 & 1.75 & 0.57 & 10.80 & 0.16 & 1.00\ 0.5 & 11.85 & 0.16 & 1.70 & 0.58 & 10.70 & 0.14 & 0.90\ 1.0 & 11.95 & 0.18 & 1.60 & 0.60 & 10.60 & 0.12 & 0.75\ 2.0 & 12.00 & 0.18 & 1.55 & 0.62 & 10.50 & 0.10 & 0.50\ 4.0 & 12.05 & 0.19 & 1.50 & 0.64 & 10.40 & 0.08 & 0.40\ 8.0 & 12.10 & 0.24 & 1.30 & 0.64 & 10.30 & 0.02 & 0.10\ \ 0.1 & 11.65 & 0.17 & 1.80 & 0.57 & 10.00 & 0.14 & 0.55\ 0.5 & 11.75 & 0.19 & 1.75 & 0.58 & 9.90 & 0.12 & 0.45\ 1.0 & 11.85 & 0.21 & 1.65 & 0.60 & 9.80 & 0.08 & 0.40\ 2.0 & 11.90 & 0.21 & 1.60 & 0.62 & 9.70 & 0.07 & 0.35\ 4.0 & 12.00 & 0.21 & 1.55 & 0.64 & 9.60 & 0.06 & 0.30\ 8.0 & 12.10 & 0.28 & 1.30 & 0.64 & 9.50 & 0.04 & 0.20\ \ 0.1 & 11.75 & 0.12 & 1.75 & 0.57 & 10.35 & 0.20 & 1.10\ 0.5 & 11.80 & 0.14 & 1.70 & 0.58 & 10.25 & 0.10 & 0.50\ 1.0 & 11.90 & 0.15 & 1.60 & 0.60 & 10.15 & 0.08 & 0.45\ 2.0 & 11.95 & 0.16 & 1.55 & 0.62 & 10.05 & 0.05 & 0.35\ 4.0 & 12.05 & 0.18 & 1.50 & 0.64 & 9.95 & 0.03 & 0.30\ 8.0 & 12.10 & 0.24 & 1.30 & 1.64 & 9.85 & 0.02 & 0.10\ \ 0.1 & 11.78 & 0.15 & 1.78 & 0.57 & 10.85 & 0.16 & 1.00\ 0.5 & 11.86 & 0.18 & 1.67 & 0.58 & 10.80 & 0.14 & 0.75\ 1.0 & 11.98 & 0.19 & 1.53 & 0.59 & 10.75 & 0.12 & 0.60\ 2.0 & 11.99 & 0.19 & 1.46 & 0.59 & 10.70 & 0.10 & 0.45\ 4.0 & 12.07 & 0.20 & 1.36 & 0.60 & 10.60 & 0.06 & 0.35\ 8.0 & 12.10 & 0.24 & 1.30 & 0.60 & 10.40 & 0.02 & 0.30\ \ Notes: Masses are in [\_ ${\rm M}_{\odot}$ ]{}. Parameters are given for the relations as function of peak halo mass through history. The symbols in each panel of Figure \[fig:efficiencyA\] give the conversion efficiency of 8 individual systems that have been selected from the upper and lower $1\sigma$ contours for 4 halo masses at $z=0.1$. Identical symbols in different panels show these systems at higher redshift and the colour indicates their sSFR. Interestingly, galaxies that reside at the upper or lower $1\sigma$ level at $z=0.1$ have not been there throughout their evolution, but have moved there from different, typically more average, efficiencies. Massive systems were even at opposite $1\sigma$ level at high redshift: the high efficiency system with $\log_{10}(M/{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi})=14$ and low sSFR at $z=0.1$ (circle) started as a low efficiency system with high sSFR at $z\gtrsim4$, while the low efficiency system with the same halo mass and higher sSFR at $z=0.1$ (square) had a very high efficiency and lower sSFR during its early evolution. This can be understood as a result of the connection between SFR and halo growth rate. A system with a high halo growth rate early-on has a high SFR as well, but in order to reach the same halo mass at low redshift as other systems, the growth rate at late times and consequently the SFR must be low. Since the instantaneous conversion efficiency is higher at high redshift, systems with a high growth and SFRs at early times form more stars than systems with the same final halo mass but low growth and SFRs at high redshift. Separating central galaxies into star-forming and quenched systems at any given redshift shows that the average integrated baryon conversion efficiency of quenched galaxies (red lines) is higher than that of star-forming galaxies (blue lines). Consequently, this means that at fixed halo mass, passive galaxies have a higher stellar mass than active galaxies. This trend is most pronounced at low redshift. We can understand this behaviour as a result of linking SFR and halo growth rate, as well. Passive galaxies live in haloes that have little growth now but experienced high growth at early times, and thus had high SFRs then. As the instantaneous conversion efficiency was higher at high redshift, these galaxies were able to form more stars compared to other galaxies with the same final halo mass. Naively, this result seems to contradict the finding of weak lensing studies [e.g. @Mandelbaum:2006aa; @Mandelbaum:2016aa] that at fixed stellar mass the average mass of haloes harbouring passive galaxies is higher than that of haloes hosting active galaxies. However, one cannot simply invert the average $m(M)$ relation to obtain the $M(m)$ relation, because of the scatter. As a consequence of the Eddington bias and the larger fraction of active galaxies in low-mass systems, the average halo mass at fixed stellar mass is higher for quenched systems. We will explore this in more detail in future work. Another interesting feature of the relation between halo mass and the integrated conversion efficiency is the scatter at fixed halo mass, which arises naturally as a consequence of different haloes taking different paths through the $\epsilon(M,z)$ diagram. From the distribution of dots in Figure \[fig:efficiencyA\] one can get the impression that there is an large amount of scatter for low halo masses, especially at low redshift. However, the standard deviation (dashed black lines) is only a little larger at low masses than at high masses. To investigate this further we show the probability density function of the integrated conversion efficiency for galaxies in four different halo masses at $z=0.1$ in Figure \[fig:effslice\]. The distributions of the efficiency are quite narrow in massive haloes and somewhat broader in low mass haloes. Moreover, while in massive haloes the distributions are cut off at $\gtrsim2\sigma$, there is a longer tail in the low mass haloes, so that a few of these haloes host massive galaxies. This explains the slightly lower efficiencies compared to the one derived with subhalo abundance matching. Due to the form of the SMF, this increases the number of galaxies with higher stellar mass (Eddington bias). Consequently, the overall integrated efficiency needs to be slightly lower to accommodate for the higher number of more massive galaxies. To confirm this, we have performed a simple abundance matching experiment: both the halo and the stellar masses of our model catalogue including also subhaloes and satellite galaxies have been rank ordered independently and then matched one-to-one. The resulting efficiency is higher at all halo masses, and we find a peak efficiency at $z=0.1$ of 19 per cent, in good agreement with previous results from abundance matching. However, we stress that if the self-consistent scatter is taken into account this is reduced to a peak efficiency of 17 per cent. ![Comparison of the average conversion efficiency of all galaxies at $z=0.1$ to previously published results. This includes empirical models [@Moster:2010aa; @Behroozi:2010aa; @Guo:2010aa; @Moster:2013aa; @Behroozi:2013aa; @Reddick:2013aa; @Lu:2015aa; @Rodriguez-Puebla:2015aa], a galaxy group catalogue [@Yang:2008aa], X-ray observations of clusters [@Kravtsov:2014aa], and results for active and passive galaxies (blue and red symbols) from weak lensing [@Mandelbaum:2016aa; @Hudson:2015aa] and from satellite kinematics [@Wojtak:2013aa]. The shaded region corresponds to the $1\sigma$ confidence levels of our model. []{data-label="fig:shm"}](shm.pdf){width="48.00000%"} We find that at a given redshift the integrated baryon conversion efficiency can be well approximated by a double-power-law (eqn. \[eqn:epsilon\]). In Table \[tab:shm\] we present the values of the parameters for all centrals, quenched and star-forming centrals, and all galaxies at 6 different redshifts. Note that the parameters have been obtained with respect to the peak mass a halo had up to the given redshift. Furthermore we find that the logarithmic scatter (in dex) can be well approximated by $$\sigma = \sigma_0 + \log_{10}\left[\left(\frac{M}{M_\sigma}\right)^{-\alpha}+1\right] \, . \label{eqn:sigma}$$ The fitted values for these parameters are also shown in Table \[tab:shm\]. We show a comparison of our model result for the integrated baryon conversion efficiency at $z=0.1$ to previously published stuff in Figure \[fig:shm\]. All stellar and halo masses have been converted to our definitions. Overall, there is a good agreement between the different methods. However, there are some notable differences. The conversion efficiency of our model at the low-mass end is lower than the results that have been obtained with subhalo abundance matching [@Moster:2010aa; @Behroozi:2010aa; @Guo:2010aa; @Moster:2013aa; @Behroozi:2010aa]. This can be understood as the consequence of the scatter in the relation. While in abundance matching a constant log-normal scatter for all halo masses is typically assumed, the scatter in our new model results from different formation histories of the haloes, which leads to an increased scatter with a tail towards higher stellar masses in low-mass haloes (c.f. Figure \[fig:effslice\]). As explained above, this results in a lower efficiency, especially for low-mass haloes. Interestingly, the empirical model by @Lu:2015aa, which does not add scatter artificially, also predicts lower conversion efficiencies for low halo masses. At the massive end, our model agrees very well with other empirical models, though it predicts comparably high conversion efficiencies. Still, compared to direct methods, all empirical models predict rather low efficiencies. One reason for this discrepancy is that weak lensing [@Hudson:2015aa; @Mandelbaum:2016aa] and satellite kinematics studies [@Wojtak:2013aa] measure halo mass for galaxy populations with fixed stellar mass. Because of the scatter in the relation it cannot simply be inverted. Another reason is the derivation of the stellar mass. Studies specifically targeting massive systems [e.g. @Kravtsov:2014aa] integrate the surface brightness up to large radii or use a fitting function, while for the SMFs used in the empirical models typically Petrosian magnitudes are used. At the massive end this can lead to a discrepancy as a significant fraction of the light can be outside the aperture. However, it is unclear how far the surface brightness should be integrated and how much light should be included to derive a galaxy’s stellar mass. To some degree it is ambiguous if the light (or mass) belongs to the galaxy or if it is part of the ICM. It is more important that if a comparison is made, the definitions should be the same, which is unfortunately not always easy to achieve. We refer to section \[sec:smfz0\] for more discussion on this topic. ![The average conversion efficiency of main haloes from $z=0.1$ to $z=8$. The dashed lines only include the stellar mass of the central galaxy ($m_=m_\mathrm{c}$), while the solid lines include the total stellar mass of the central galaxy and the ICM ($m_=m_\mathrm{c}+m_\mathrm{ICM}$). []{data-label="fig:icm"}](icm.pdf){width="48.00000%"} Are more unambiguous quantity is the total amount of stellar mass that can be associated with a main halo, i.e. the stellar mass in the central galaxy $m_\mathrm{c}$ plus the ICM $m_\mathrm{ICM}$ (but without the mass in satellites). In our model the central galaxy can grow through star formation and mergers, and the ICM can grow through ejected stars in mergers, tidal stripping of satellites, and the infall of a subhalo with its own ICM (which is then transferred to the main halo). In Figure \[fig:icm\], we plot the conversion efficiency including only the mass of the central galaxy (dashed lines), and including the mass of the central galaxy plus the ICM (solid lines) from $z=0.1$ to $z=8$. We note that the lines can cross, because the halo mass at the given redshift has been used to calculate the conversion efficiency (and not the $z=0$ mass). While the ICM adds very little to the total mass at high redshift and for low halo masses, it dominates the stellar mass budget of massive main haloes at low redshift. For a halo with $M=10^{15}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ the ICM is larger than the mass of the central galaxy by a factor of almost 8. We note that over 80 per cent of the ICM in these haloes forms from the tidal disruption of satellites and is distributed throughout the halo. The contribution of ejected stars from merging satellites is therefore rather small and even if all those stars would be associated with the central galaxy ($f_\mathrm{esc}=0$), this would boost its mass only by a factor up to 2. Clustering of star-forming and quenched galaxies {#sec:wpc} ------------------------------------------------ ![Projected galaxy correlation function for star-forming (blue) and quenched (red) galaxies in five bins of increasing stellar mass (from top to bottom). The symbols represent the observational estimates [@Guo:2011aa], and the lines show the model prediction for the best fit parameters for the $150{\ifmmode {\rm~Mpc} \else ${\rm~Mpc}$\fi}$ box. []{data-label="fig:wpc"}](wpc.pdf){width="45.00000%"} We have used the projected auto-correlation functions of galaxies in different stellar mass bins to constrain our model. As the tidal stripping of satellite galaxies reduces the number of pairs at small scales, the clustering provides a measure for how effective the stripping can be. On large scales, galaxy clustering is only a consequence of halo clustering, so that it can be used to test if galaxies with given properties form in the right dark matter haloes. We have shown that on large scales the projected correlation function computed with our model agrees very well with the observed data (Figure \[fig:wp\]). Here, the correlation results from pairs that reside in different haloes and are predominantly centrals. This indicates that the relation between central galaxies and main haloes is described well by our model. Additional insight into the connection between galaxies and dark matter haloes can be obtained by studying the clustering of star-forming and quenched galaxies. This can reveal whether the modelled quenching processes lead to the correct distribution of galaxies in haloes. As the SFR depends on the halo growth rate and the instantaneous conversion efficiency, a galaxy can be quenched because it runs out of fuel (environmental quenching), or because the efficiency of the feedback becomes very high as the halo grows (mass quenching). As halo clustering at fixed mass is stronger for haloes with low growth rates, and passive galaxies are located in these haloes because of environmental quenching, the clustering of quenched galaxies is stronger. On small scales the correlation function is dominated by pairs in which at least one galaxy is a satellite, so the clustering of active and passive galaxies is strongly affected by how satellites are quenched. Because of mass quenching, active and passive central galaxies of a given stellar mass live in haloes with different masses, and consequently have a different correlation function at large scales where the signal is dominated by pairs of centrals. The large-scale clustering of star-forming and quenched galaxies therefore provides a strong constraint on mass quenching and consequently on the instantaneous conversion efficiency. We can asses if the effects of the quenching processes are captured well in the model by comparing the clustering of star-forming and quenched galaxies in the model to observational data. In Figure \[fig:wpc\], we plot the projected galaxy correlation function computed with our model (lines) and the observations by [@Guo:2011aa] (symbols) in five stellar mass bins for star-forming (blue) and quenched (red) galaxies. The observed galaxies have been divided with a $g-r$ colour cut, while the model galaxies were separated with a cut in the sSFR. The agreement between model and data is good. Passive galaxies are more clustered than active galaxies on all scales. Especially at small scales, the clustering of quenched galaxies is enhanced compared to star-forming galaxies, showing that the adopted satellite quenching model captures the environmental effects quite well. Also at large scale the clustering of passive galaxies is enhanced demonstrating that the mass quenching leads to active and passive galaxies being located in the correct haloes. Discussion {#sec:disc} ========== ![image](smf_serexp.pdf){width="99.00000%"} Our new empirical model [Emerge]{} links the formation of galaxies to the evolution of dark matter haloes. The resulting average build-up of the stellar component is in good agreement with previous empirical results, which have typically been used as a benchmark to test hydrodynamical simulations. For instance, the integrated conversion efficiency of simulated galaxies is often compared to the empirical findings to judge whether the simulation has formed the correct amount of stellar mass in a certain halo. The result is then often used to tune the unconstrained parameters of the relevant baryonic physics, most notably the efficiency of the supernova feedback. This test can also be applied to infer if a previously unconsidered process can release the tension. For example, @Stinson:2013aa have compared the SFHs of simulated galaxies to the empirical average SFHs by @Moster:2013aa, and found that supernova feedback alone can reduce star formation enough to match the integrated conversion efficiency at $z=0$, but the simulations still form too many stars before $z\sim2$. They then invoke ‘early feedback’, which mimics an ultraviolet ionisation source and provides pressure from the radiation of massive young stars, and show that the correct average SFHs can be reproduced. The limitation of previous empirical models was always the ability to only derive average galaxy properties for a given halo mass, while the scatter in the relations was typically introduced artificially, and did not reflect the formation history of the halo. It was therefore always unclear, if a particular simulated galaxy was simply an outlier to the relation, or was in disagreement with the empirical findings. In the first case the conclusion would be that the modelled baryonic physics can explain the observations, while in the second case one would modify the model or include new physics. Since [Emerge]{} computes the formation of galaxies in individual dark matter haloes in an empirical fashion that automatically reproduces a large number of observational constraints, it provides an ideal testbed for hydrodynamical simulations. It is possible to run the model on the halo merger tree provided by a dark matter simulation. If the hydrodynamical simulation with the same initial conditions is in good agreement with the empirical results for this specific system, the baryonic physics in the simulation is modelled such that statistical observations like SMFs can be reproduced for a larger simulation volume. In the case where the hydrodynamic simulation results in a large amount of stellar mass, while the empirical model predicts an average or lower amount of stars, then the efficiency of the feedback may not be modelled correctly and will lead to a disagreement with the SMF for a larger simulation volume. The conversion efficiency in massive haloes {#sec:smfz0} ------------------------------------------- The characteristic shape of the instantaneous and integrated baryon conversion efficiencies results from the interplay of the different physical processes that prevent the infalling gas in a dark matter halo from cooling and forming stars. The contribution of each process strongly depends on the mass of the halo. While in low-mass haloes, feedback from stars (e.g. supernova-driven winds or radiation pressure) can expel large amounts of gas from haloes with low escape velocities, in more massive haloes the gas cannot escape as easily and falls back to the centre. However, AGN feedback can heat the gas in the halo and prevent it from cooling and falling back to the centre. At the massive end, this feedback can thus dominate the conversion efficiency. Comparing the SHM ratio in simulations of massive galaxies to empirical constraints has therefore been a primary way to test models for AGN feedback and to determine its efficiency. The question whether AGN feedback is the dominant mechanism at high masses and how strong it has to can thus be inferred from observed galaxy properties. The main observational input for the empirical constraints has been the SMF. For low-redshift galaxies the massive end is well sampled by the SMF presented by @Li:2009aa, who use Petrosian magnitudes to compute stellar masses. In this work we used the updated SMF in @Guo:2010aa which are based on `cmodel` magnitudes. These have been derived by fitting an exponential and a de Vaucouleurs profile to the photometry and using the best fit to compute the total magnitude. In this way the light that falls outside the Petrosian aperture can be captured. Recently, @Bernardi:2013aa presented a SMF which is based on a fit of a Sersic profile or a two-component profile (Sersic and exponential) to the observed surface brightness. As these profiles are more realistic for objects that neither have a de Vaucouleurs nor a pure exponential profile, the authors claim their stellar masses provide a better estimate of the true mass. Since the profiles they adopt typically return more of the light in the outskirts, the derived stellar masses are generally larger then masses based on Petrosian or `cmodel` magnitudes. Using the @Bernardi:2013aa SMFs, @Kravtsov:2014aa apply the subhalo abundance matching method and find that the SHM ratio is significantly higher than previously found with SMFs based on Petrosian or `cmodel` magnitudes. While they acknowledge that empirical models predict the total stellar mass in the galaxy and the ICM to be of the same order as for their result, they conclude that the overall efficiency of star formation in massive halos is considerably less suppressed than previously thought, and that feedback in massive halos should be weaker than assumed in most of the current simulations. However, we stress that this conclusion heavily depends on what is defined as galaxy mass and what is defined as ICM, and can therefore be misleading. It is well established that the mass from accreted satellites is mainly located at the outskirts of massive galaxies [@Oser:2012aa; @Hilz:2013aa]. It is therefore not unambiguous if this material should be ranked among the central galaxy or the ICM, even if the light profile is continuous as there may be overlap. If extended profiles are used, more of this accreted mass is taken into account and assigned to the central galaxy. This can be illustrated with our model. Figure \[fig:serexp\] shows the SMF from $z\sim0.1$ to $z\sim4$ (left to right panels). The symbols are the same data as used to fit our model, with the exception of the SMF of @Bernardi:2013aa. Where before we have used their SMF based on `cmodel` magnitudes, the left panel shows the SMF based on their two-component (Sersic and exponential) fit. The red dashed lines correspond to our best-fit model with a fraction of stars in satellites that escape to the ICM during a merger of $f_\mathrm{esc}=0.39$. The black lines shows the same model but with an escape fraction of 10 per cent. While beyond $z>0.5$ the two models give almost identical results, the massive end of the $z\sim0.1$ SMF is very different. The best-fit model reproduces the @Li:2009aa SMF (as fitted), while the model with $f_\mathrm{esc}=0.1$ has a much shallower slope and reproduces the @Bernardi:2013aa SMF. This shows that while in both models the total amount of stars formed is equal, the SMF and thus the integrated baryon conversion efficiency can be boosted if more of the material is assigned to the galaxy instead of the ICM. In our model, the escape fraction $f_\mathrm{esc}$ regulates this assignment. The SFR is independent of this choice, and is additionally constrained by the sSFR measurements. In this sense, the instantaneous efficiency is given by the measured sSFR, while the escape fraction is fixed by the growth of the massive end of the SMF which is mainly though accretion. So while the escape fraction and the integrated efficiency depend on the assumed surface brightness profile and whether the mass is assigned to the galaxy or the ICM, the instantaneous efficiency provides a strong constraint for the feedback processes. As we have shown, the SFHs of massive galaxies are quenched strongly after $z\sim4$ falling from $\sim100{\ifmmode M_{\odot}{\rm~yr}^{-1} \else $M_{\odot}{\rm~yr}^{-1}$ \fi}$ to $<1{\ifmmode M_{\odot}{\rm~yr}^{-1} \else $M_{\odot}{\rm~yr}^{-1}$ \fi}$ at $z=0$. This implies that strong AGN feedback may still be necessary to prevent the gas from cooling and forming stars. If the integrated conversion efficiency in simulations is compared to empirical constraints, the definition of the surface brightness profiles does matter, i.e. what belongs to the galaxy and what belongs to the halo. In this case, it is important to ensure that the comparison between simulation and observation is as fair as possible. The ideal comparison would be to create mock images of the simulation with a radiative transfer code such as [Sunrise]{} [@Jonsson:2006aa] or [Grasil-3D]{} [@Dominguez-Tenreiro:2014aa], and to analyse these images in the exact same way as the observed data. If this is not possible, then the stellar mass in the simulation should be computed as close as possible to the observations. If the simulation is compared to SMFs based on Petrosian or `cmodel` magnitudes, then only the stellar mass within a few scale radii should be counted. For a comparison to SMFs based on profiles that capture the light out to large radii, then it is better to count all stellar mass within a large radius up to the virial radius, subtracting all stars bound to satellites. The conversion efficiency in low-mass haloes -------------------------------------------- The empirically determined SHM ratio has also often been used at the low mass end to provide a constraint for hydrodynamical simulations, and to test or tune the effects of stellar feedback, such as supernova-driven winds, radiation pressure, and cosmic rays. The integrated conversion efficiencies computed by different empirical models agree very well within the derived uncertainties up to the minimum mass that is used in the model. The minimum stellar mass for which observational constraints are available (typically from the SMF) is $m_\mathrm{min}\sim 10^{7.5}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$, with a corresponding halo mass of $M_\mathrm{min}\sim10^{10.5}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$. Below this minimum mass that has been used to fit the model, the empirical models do not provide any constraint. Still, it has become common to extrapolate the SHM mass relation down to very low masses. Since different empirical models can use different fitting functions this extrapolation can lead to orders of magnitude differences in the stellar mass at fixed halo mass, even if the models agree very well up to the minimum mass. In practice, the empirical model is used that fits the hydrodynamical results best when extrapolated, and success is then claimed for the simulations. We strongly caution against this practice, as extrapolating the conversion efficiency has several pitfalls. First, reionisation suppresses galaxy formation in haloes that have a low mass early-on [@Efstathiou:1992aa], which can lead to dark subhaloes that do not host a galaxy [@Sawala:2015aa]. This can lead to the mean SHM relation bending over with respect to the interpolation. Secondly, it is not clear that galaxies at these mass scales follow a tight SHM relation, as observations of their stellar populations show ‘bursty’ star formation histories. As our model shows, this can lead to a large scatter at low halo masses, and we have demonstrated that an increased scatter leads to a modified mean SHM relation. Thirdly, it is unlikely that the physical processes that govern galaxy formation at intermediate mass ($10^{11}<M/{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}<10^{12}$), are also the dominant processes at lower halo masses. As the processes can have different efficiencies, the low-mass slope of the SHM relation may change at the halo mass, where the efficiencies of two dominant processes cross, and currently it is difficult to find this transition mass. Finally, the main justification why simple empirical models such as abundance matching can be used to constrain the conversion efficiency is not because a monotonic SHM relation is an obvious choice, but because the models are able to correctly reproduce observations that have not been used to calibrate the model, such as galaxy clustering. This has been shown to be successful above the minimum stellar mass that has been used in the fit, but not yet below that mass. Until such an independent validation of the model can be performed, the results have to be regarded as tentative at best. Differences to a semi-analytic model ------------------------------------ The new empirical model presented here follows dark matter halo trees through cosmic time and populates them with galaxies according to a set of simple parameterised prescriptions where the parameters are adjusted to fit observed data. In this regard the question arises how this model differs from a SAM. To shed some light on this issue we first need to recall the purpose and history of SAMs which is excellently illustrated in the review by @Baugh:2006aa. When full hydrodynamical simulations were still unfeasible, the only possibility to model the formation of galaxies was to treat all gas physics semi-analytically [@White:1978aa; @White:1991aa]. However, as many of the physical processes are poorly understood, several of the adopted prescriptions need to be parameterised. As these parameters are directly connected to a physical process their values often cannot be chosen completely freely, but they need to comply with physical priors, typically from direct observations or more detailed simulations. The star-formation efficiency at low redshift is strongly constrained by the observed star-formation-relation [@Schmidt:1959aa; @Kennicutt:1998aa], for example. Within these limits the parameters are then typically chosen to reproduce a subset of statistical observations. Following the general philosophy of ab initio models, SAMs can then be employed to learn about galaxy formation. For a given model with certain values chosen for the parameters, the model is run and the results are compared to observations. Varying the parameter values within physical limits, this is repeated until the best agreement is found. If some subsets of the observations cannot be reproduced the model is changed, either by altering the parameterisations or by including new physical processes. A very insightful example is the inability of earlier SAMs to reproduce the bright end of the luminosity function. This eventually lead to the inclusion of feedback from AGN into the models showing that the data can be reproduced in this way. Compared to hydrodynamical simulations, SAMs have the great advantage that model prescriptions can easily be varied or disabled. Therefore, SAMs are very useful to to gain a better understanding of how the different physical processes impact a particular observation. Moreover, once all physical processes driving galaxy formation are fully understood, future SAMs that parameterise these processes will likely be the final way to describe galaxy formation. Empirical models on the other hand avoid directly modelling the physics of the baryon cycle, i.e. the different states of the baryons (hot gas, cold gas, and stars) and the physical processes that regulate the transition between these states, but rather employ empirical relations between galaxy and halo properties, marginalising over the baryon cycle. This can technically be done similarly to SAMs by populating halo merger trees with galaxies according to parameterised relations. However, the motivation determining the choice of the parameterised models is different. The parameters in an empirical model are not directly related to a physical process or quantity, so there are no physical priors on them. Therefore all model parameters can be fitted to reproduce the data with statistical methods such as an MCMC algorithm without concerns about the parameter values. Moreover, model selection criteria can easily be applied to find the simplest model that agrees with the data. In SAMs this is more complicated because of observed physical priors, e.g. outflow velocities. However, as empirical models do not explicitly implement physical processes it is difficult to use them directly to learn about the physics of galaxy formation, i.e. one cannot test the impact of certain physical processes on the galaxy population (empirical models can nevertheless provide strong constraints that can be used to study the physics). Still, this does not imply that empirical models are ‘unphysical’, unless they violate some law of physics – they simply do not model the physics of the baryon cycle explicitly, but rather provide a minimal set of assumptions to reproduce the observed data. Lately, several studies have used statistical methods such as MCMC methods to explore the parameter space of SAMs [e.g. @Henriques:2009aa; @Lu:2011aa; @Lu:2012aa; @Henriques:2013aa]. This is a very useful exercise as it provides many insights into the model, such as degeneracies between parameters or unconstrained parameters. This also provides uncertainties on model predictions and can thus help considerably with interpretation of the results. However, this does not imply that the models normalised in this way represent all the underlying physics, nor that they are unique. Rather they are a simple physically motivated representation that can reproduce the set of observations. For instance, if a SAM provides a very good fit to the low-mass end of the SMF, it is not guaranteed that the fitted supernova feedback parameters (e.g. the efficiency) represent the underlying physics. It cannot be ruled out that other processes that may not have been considered, such as cosmic ray feedback [e.g. @Pfrommer:2007aa], may be equally or even more relevant, and that their effects have just been mimicked by the supernova feedback model. This can make the interpretation of MCMC-fitted SAMs difficult. Unless there are strong priors on the physical parameters, a model fitting these parameters is no longer based on first principles and effectively becomes empirical. Given these differences, one could argue for employing a state-of-the-art SAM and fit all possible model parameters to reproduce observations. Even if the model effectively is empirical when fitted with a statistical method, it may still be able to reproduce the observations as well as a model that uses empirical relations between halo and galaxy properties. Galaxies with realistic properties can thus be followed through cosmic time if a limited physical interpretation of the model is accepted. However, there a some complications with this approach. First, as SAMs typically have a very large number of parameters it becomes difficult to sample the whole parameter space. Secondly, as SAMs directly model the physics of the baryon cycle, the parameters may degenerate and it becomes critical to find observational data that can break the degeneracies [e.g. @Henriques:2015aa]. Empirical models can be designed to minimise parameter degeneracies, as they do not aim to follow all components of the baryon cycle. Thirdly, because the relations in empirical models are not explicitly motivated by baryonic physics, it is straight-forward to use model selection statistics to optimise the model, which is much more difficult in a SAM. This means that if the goal is simply to track galaxies through cosmic time while having galaxy populations that agree as well as possible with observed galaxies, then Occam’s razor can favor empirical models. Summary {#sec:sum} ======= In this paper, we present the novel empirical galaxy formation model [Emerge]{}, which follows the evolution of galaxies in individual dark matter haloes through cosmic time. First, halo merger trees are extracted from cosmological $N$-body simulations and the growth rate is calculated for each halo at every redshift. Secondly, the SFR of the galaxy in each halo is determined as the product of the halo growth rate, which specifies how much material becomes available, and the instantaneous baryon conversion efficiency, which specifies how efficiently this material can be converted into stars. It captures the effects of all baryonic physics that governs galaxy formation, depends on halo mass and redshift, and has a peak around $10^{12}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$. The stellar mass of each galaxy is computed by integrating the SFHs taking into account mass loss from dying stars. Once a halo stops growing, i.e. when it starts to fall onto a larger halo and becomes a subhalo, its galaxy continues to form stars for a specified amount of time and is then rapidly quenched. If the halo has lost a significant fraction of its peak mass, the stars in the galaxy become unbound and are stripped to the ICM. When a satellite has lost its kinetic energy due to dynamic friction, it merges with the central galaxy and ejects a fraction of its stars to the ICM. We constrain our model with several sets of observed data. The conversion efficiency is constrained by SMFs, sSFRs, and the CSFRD up to $z\sim10$. The satellite quenching timescale is constrained by the fraction of quenched galaxies as function of stellar mass up to $z=4$, and the stellar stripping is constrained by small-scale galaxy clustering. The fraction of stars ejected to the ICM is determined with the low-redshift evolution of the massive end of the SMF and the sSFR of massive galaxies. We fit all model parameters with an MCMC ensemble sampler by requiring that all observed data be reproduced simultaneously. The adopted empirical relations are as flexible as possible. We increase the complexity of the model stepwise if the data require it, which is assessed by a number of different model selection statistics. The result is thus the simplest model that is in agreement with the data. For our best-fit model, we find that the characteristic halo mass where the instantaneous conversion efficiency peaks, decreases from $1.1\times10^{12}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ at high redshift to $2.2\times10^{11}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ at $z=0$. The peak efficiency at high redshift is 70 per cent and decreases to less than one per cent towards low redshift. We find a steep low-mass slope for the efficiency of 1.3 at high redshift that steepens to 3.3 towards low redshift, such that star formation is strongly suppressed in low-mass haloes. The high-mass slope is $\sim1$, independent of redshift, so as haloes become more massive the galaxy gets quenched. We find that almost 40 per cent of all stars in satellite galaxies get ejected to the ISM during a merger, and that satellite galaxies are tidally disrupted once the mass of their subhalo has dropped to 10 per cent of its peak value. Massive satellites with $m\ge10^{10}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ keep forming stars for about 4 dynamical halo times ($r_\mathrm{vir}/v_\mathrm{vir}$) after their halo has stopped growing. In lower-mass galaxies this quenching timescale is considerably longer, such as 10 dynamical halo times for satellites with $m=10^{9}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$. Using our best-fit model, we study the SFHs of individual galaxies, and find that even if the final halo masses are identical, galaxies can have very different SFHs. The average SFHs and accretion rates as a function of $z=0$ halo mass in our model agree very well with previous findings. The SFRs of central galaxies in massive haloes typically peak at high redshift, after which the galaxies become increasingly quenched, e.g. the SFR of a galaxy in a halo with a $z=0$ mass of $10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ peaks around $z=4$. In low-mass haloes the SFRs peak much later, e.g. in a halo with a $z=0$ mass of $10^{12}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ the SFR peaks around $z=1$. However, individual galaxies can deviate significantly from these overall trends. The average peak SFRs of central galaxies that are quenched at $z=0$ is slightly higher than that of their star-forming counterparts, as quenched galaxies generally form most of their mass at high redshift, where the peak is located. We find that the fraction of ex-situ formed (accreted) stars to be insignificant in low mass haloes, while in massive haloes these stars can dominate the total stellar mass of central galaxies, e.g. the accreted fraction of galaxies in haloes with $M=10^{14}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ is 55 per cent at $z=0$. We do not notice any difference between quenched and star-forming galaxies for the accreted fraction. Our model predicts the integrated baryon conversion efficiency $m/(f_\mathrm{b}M)$, i.e. the SHM ratio, for each individual system in contrast to previous models that only predicted the average SHM ratio at a given halo mass. The stellar mass of each galaxy depends on the formation history of the halo, such that scatter in the SHM relation is automatically produced. The average peak conversion efficiency at $z=0.1$ is 17 per cent at a halo mass of $M\sim10^{12}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$. As quenched galaxies formed most of their mass at high redshift where the conversion efficiency is higher, they have a larger SHM ratio than star-forming galaxies. We find a scatter of 0.16 dex at the massive end, which is in good agreement with observational estimates from satellite kinematics, while at the low-mass end the scatter is larger. This increased scatter causes the overall average SHM to be slightly lower than previously found. Still, the overall agreement with other empirical models is very good within the model uncertainties. We further find good agreement with direct measurements of the SHM ratio from weak lensing and satellite kinematics, while X-ray measurements indicate higher SHM ratios. Taking into account also the ICM, we find that the SHM ratio is strongly enhanced in massive systems at low redshift. For example, a halo with $M=10^{15}{\ifmmode {\rm M}_{\odot} \else ${\rm M}_{\odot}$ \fi}$ at $z=0.1$ has a conversion efficiency of 0.004 when only the mass in the central galaxy is taken into account, while if also the ICM is considered, which is larger by a factor of 8, the conversion efficiency increases to 0.037. Computing galaxy clustering for star-forming and quenched galaxies we find a very good agreement with observational constraints. Passive galaxies are more strongly clustered at all scales, while at small scales this effect is even enhanced. Quenched galaxies live in haloes that have low growth rates, and these haloes are more strongly clustered, so quenched galaxies are generally more clustered as well. Moreover, satellite galaxies get environmentally quenched, so at small scales, where the clustering is dominated by pairs in which at least one galaxy is a satellite and these are preferentially quenched, the clustering of passive galaxies is boosted. The good overall agreement indicates a realistic assignment of galaxies to haloes. Acknowledgements {#acknowledgements .unnumbered} ================ We thank all authors who provide their data in electronic form. We are also grateful to Peter Behroozi, Andreas Burkert, Darren Croton, Benedikt Diemer, George Efstathiou, Martin Haehnelt, Andrew Hearin, Bruno Henriques, Ben Hoyle, Houjun Mo, Jerry Ostriker, Debora Sijacki, Rachel Somerville, Jeremy Tinker, and Frank van den Bosch for enlightening discussions. The cosmological simulations used in this work were carried out at the Odin Cluster at the Max Planck Computing and Data Facility in Garching, and the Darwin Supercomputer of the University of Cambridge High Performance Computing Service. BPM acknowledges an Emmy Noether grant funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – MO 2979/1-1. TN acknowledges support from the DFG Cluster of Excellence “Origin and Structure of the Universe”. Model selection {#sec:bayes} =============== The philosophy of the empirical model [Emerge]{} is to construct the simplest self-consistent galaxy evolution model that is able to explain all available observed statistical data without being restricted by our limited understanding of baryonic physics. The model is thus designed to allow for a very broad range of possibilities such that all observational constraints can be fulfilled. We start with very simple models to describe the instantaneous efficiency and the processes for the satellite galaxies containing a minimal number of parameters, and increase the complexity and the number of parameters in a stepwise manner, as the data require it. We assess this with a number of different model selection criteria that are described below; for more details on the methods see @Liddle:2007aa. Model $M_\mathrm{z}$ $\epsilon_\mathrm{z}$ $\beta_\mathrm{z}$ $\gamma_\mathrm{z}$ $\tau_\mathrm{s}$ $f_\mathrm{s}$ $\tau_\mathrm{d}$ ----------- -------------------- --------------------------- ------------------------ ------------------------- ----------------------- -------------------- ----------------------- 1 0 0 0 0 0 $M_\mathrm{p}$ 0 2 F F 0 0 0 $M_\mathrm{p}$ 0 3 F F F 0 0 $M_\mathrm{p}$ 0 4 F F F F 0 $M_\mathrm{p}$ 0 5 F F F 0 F $M_\mathrm{p}$ 0 6 F F F 0 F $M_\mathrm{p}$ F 7 F F F 0 F $m_$ 0 : Summary of the tested models[]{data-label="tab:models"} \ Notes:* Columns are model number (1), $z$-evolution of $M_1$ (2), $N$ (3), $\beta$ (4), and $\gamma$ (5), slope of quenching time (6), threshold used for stripping (7), and star formation decay time-scale (8). A 0 indicates a parameter is set to $0$, F states that the parameter is free, and $M_\mathrm{p}$ signifies that the stripping threshold is taken with respect to the halo peak mass, while for $m_$ it is taken with respect to the present stellar mass. Model $\chi^2_\mathrm{min}$ $\chi^2_\mathrm{mean}$ $N_\mathrm{p}$ $p_\mathrm{D}$ AIC BIC DIC $-2\ln Z$ $-2\ln(Z/Z_5)$ ------------- ----------------------------- ------------------------------ ---------------------- ---------------------- ----------- ----------- ----------- ----------------- ---------------------- 1 2761.65 2768.94 7 7.29 2775.71 2814.09 2776.24 2859.32 990.67 2 1939.60 1948.75 9 9.16 1957.70 2007.04 1957.91 2023.47 154.82 3 1932.65 1943.08 10 10.42 1952.78 2007.01 1953.50 2020.29 151.64 4 1932.65 1943.77 11 11.12 1954.80 2015.07 1954.88 2045.40 176.75 5 1789.41 1800.42 11 11.01 1811.56 1871.83 1811.44 1868.65 0.0 6 1789.41 1801.55 12 12.14 1813.59 1879.33 1813.69 1908.77 40.12 7 1845.19 1856.16 11 10.97 1867.34 1927.62 1867.13 1917.29 48.64 \ Notes:* Columns are model number (1), minimum deviance $\chi^2_\mathrm{min}$ (2), mean deviance $\chi^2_\mathrm{mean}$ (3), number of free parameters $N_\mathrm{p}$ (4), effective number of parameters $p_\mathrm{D}$ (5), Akaike information criterion (6), Bayesian information criterion (7), deviance information criterion (8), twice the logarithmic marginal likelihood $Z$ (9), and twice the logarithmic Bayes factor w.r.t. the best model (10). Model selection criteria ------------------------ While models with a large number of free parameters can achieve a good fit more easily, the high complexity reduces the predictiveness of the model. Therefore, any model selection aims to balance the quality of the fit to observational data against the complexity of the model. This acts as Occam’s razor and prefers simpler models if the fits are similar. In practice, model selection statistics attach a number to each model, which is based on the quality of the fit and penalised for larger numbers of parameters. A simple statistic that can be applied easily to almost any model is the Akaike Information Criterion [AIC, @Akaike:1974aa], as it only requires the maximum likelihood a model can achieve $\mathcal{L}_\mathrm{max}$, rather than the full likelihood surface. It is defined as $$\label{eqn:AIC} \mathrm{AIC} = -2 \ln \mathcal{L}_\mathrm{max} + 2k \; ,$$ where $k$ is the number of model parameters. The best model is the one that minimises the AIC. For small sample sizes, a corrected AIC has been introduced by @Sugiura:1978aa: $$\label{eqn:AICc} \mathrm{AIC_c} = \mathrm{AIC} + \frac{2k(k+1)}{N-k-1} \; ,$$ where $N$ is the number of data points used in the fit. This strengthens the penalty for $N/k$ being only a few. The Bayesian Information Criterion [BIC, @Schwarz:1978aa] is an approximation of the Bayes factor, and is defined as $$\label{eqn:BIC} \mathrm{BIC} = -2 \ln \mathcal{L}_\mathrm{max} + k \ln N \; ,$$ assuming that the data points are independent and the parameters are not degenerate. The Deviance Information Criterion [DIC, @Spiegelhalter:2002aa] combines elements from Bayesian statistics and information theory. Unlike the AIC and BIC it accounts for situations where parameters are unconstrained by data. It requires knowledge of the full likelihood surface, but can be computed from a posterior sample generated by a MCMC method. First the Bayesian complexity is introduced: $$\label{eqn:pd} p_D = \langle\chi^2(\vec\theta)\rangle - \chi^2_\mathrm{min} \; ,$$ where the chevrons $\langle\rangle$ indicate the average over the posterior distribution. It corresponds to an effective number of model parameters indicating the number of parameters actually constrained by the data. The DIC is then defined as $$\label{eqn:DIC} \mathrm{DIC} = -2 \ln \mathcal{L}_\mathrm{max} + 2 p_D \; ,$$ so that it does not penalise the inclusion of parameters that are unconstrained by the data. In Bayesian statistics the Bayesian Evidence (also referred to as model likelihood or marginal likelihood) is the probability of the data $D$ given the model $\mathcal{M}$, and it can be computed as $$\label{eqn:evidence} Z = P(D\|\mathcal{M}) = \int \mathrm{d}\vec\theta \, P(\vec\theta\|\mathcal{M}) \, P(D\|\vec\theta,\mathcal{M}) \; ,$$ where $P(\vec\theta\|\mathcal{M})$ is the prior distribution and $P(D\|\vec\theta,\mathcal{M})$ is given by the likelihood function $\mathcal{L}(\vec\theta)$. With a prior probability $P(\mathcal{M}_i)$ of a model $\mathcal{M}_i$, the probability of this model given the data is specified by Bayes Theorem: $P(\mathcal{M}_i,D)=P(\mathcal{M}_i) P(D\|M)/P(D)$. Thus, the posterior odds of model $\mathcal{M}_i$ relative to model $\mathcal{M}_j$ become $$\label{eqn:bayes} \frac{P(\mathcal{M}_i\|D)}{P(\mathcal{M}_j\|D)} = \frac{P(\mathcal{M}_i)}{P(\mathcal{M}_j)} \, \frac{P(D\|\mathcal{M}_i)}{P(D\|\mathcal{M}_j)} = \frac{P(\mathcal{M}_i)}{P(\mathcal{M}_j)} \, \frac{Z_i}{Z_j}\; ,$$ and we can interpret the Evidence $Z$ as the support for a model given the data. In the absence of information about the model priors, a ratio of $P(\mathcal{M}_i)/P(\mathcal{M}_j)=1$ is adopted, and the relative probability of the models given the data is specified by the Bayes Factor $\mathcal{B}=Z_i/Z_j$. This property can be used for model selection and considers a combination of data fit and predictiveness. Moreover, the Evidence does not penalise parameters that are unconstrained by the data. Thus, models that fit the data by varying few parameters are favoured. Models with many unconstrained parameters therefore do not need to be discarded, only the unconstrained parameters. Calculating the Evidence by integrating over parameter space is difficult in practice, however. Here, we use the method presented by @Weinberg:2012aa and @Weinberg:2013aa to derive the Evidence from the posterior distributions computed with our adopted MCMC algorithm. We first define a define a region of high posterior probability employing a volume peeling strategy and then integrate the sample in this region numerically using a volume tessellation algorithm. This integration is performed for both the Riemann and the Lebesgue variants, and we have verified that both techniques result in the same Evidence within a few percent for all models, i.e. the values for $\ln Z$ differ by 0.2 at most. Tested models ------------- We start with a very simple model, test how well this model can fit the data, and then increase the complexity stepwise. For all models, we assume that at any redshift the instantaneous baryon conversion efficiency $\epsilon$ has a double power law dependence on halo mass, as specified by equation (\[eqn:epsilon\]). The four parameters may vary with redshift and depend linearly on the scale factor $a$: $$\begin{aligned} \log_{10} M_1(z)& = M_0 + M_\mathrm{z}(1-a) = M_0 + M_\mathrm{z}\frac{z}{z+1} \; ,\label{eqn:m1}\\ \epsilon_\mathrm{N}(z)& = \epsilon_0 + \epsilon_\mathrm{z}(1-a) = \epsilon_0 + \epsilon_\mathrm{z}\frac{z}{z+1} \; ,\label{eqn:epsN}\\ \beta(z)& = \beta_0 + \beta_\mathrm{z}(1-a) = \beta_0 + \beta_\mathrm{z}\frac{z}{z+1} \; ,\label{eqn:beta}\\ \gamma(z)& = \gamma_0 + \gamma_\mathrm{z}(1-a) = \gamma_0 + \gamma_\mathrm{z}\frac{z}{z+1}\; .\label{eqn:gamma}\end{aligned}$$ In model 1, we assume that the four parameters of $\epsilon(M)$ are constant through cosmic time, i.e. the efficiency does not evolve with redshift, and we set $M_\mathrm{z}=\epsilon_\mathrm{z}=\beta_\mathrm{z}=\gamma_\mathrm{z}=0$. Moreover we set the slope of the quenching timescale $\tau_\mathrm{s}$ to zero, so that the quenching time does not depend on stellar mass and is equal for all satellites. After the quenching time has elapsed, the SFR is set to zero instantly. The satellite is stripped to the halo once the mass of the associated subhalo has fallen below a fraction of its peak mass $M_\mathrm{p}$. In model 2 we allow for a redshift evolution of the characteristic halo mass $M_1$ and the normalisation $\epsilon_\mathrm{N}$ of the conversion efficiency as specified by equations, i.e. $M_\mathrm{z}$ and $\epsilon_\mathrm{z}$ are now free parameters, while everything else is identical to model 1. In model 3, we also let the low-mass slope $\beta$ evolve, i.e. $\beta_\mathrm{z}$ becomes a free parameter as well. All four parameters, including the high-mass slope $\gamma$ are allowed to evolve with redshift in model 4, i.e. $\gamma_\mathrm{z}$ is non-zero, too. The quenching and stripping in models 2 to 4 are identical to model 1. Model 5 is identical to model 3 ($M_\mathrm{z}$, $\epsilon_\mathrm{z}$, and $\beta_\mathrm{z}$ are free while $\gamma_\mathrm{z}=0$), but now we let the quenching timescale depend on stellar mass according to equation (\[eqn:satquenching\]), so that $\tau_\mathrm{s}$ is allowed to vary freely. In model 6, we also let the SFR decline exponentially on a timescale $\tau_\mathrm{d}$ after the quenching timescale has elapsed. Finally, model 7 is identical to model 5, but the galaxy is stripped to the halo once its subhalo has reached a mass that is a specific factor of the galaxy’s stellar mass (typically a factor of a few), instead of a fraction of its peak mass. Table \[tab:models\] summarises all models we have tested. For each model we first found the most likely parameters using the [Hybrid]{} method and then ran the ensemble MCMC as described in section \[sec:modelfitting\] using the same number of ensembles, walkers, and steps. ![image](modelcomp.pdf){width="95.00000%"} Model selection results ----------------------- We use the posterior distributions from the Markov chains to compute the information criteria and Bayesian Evidences. The results are summarised in Table \[tab:selectionresults\]. We present the best-fit results for a subset of the SMF, CSFRD, sSFR, quenched fraction and clustering data for all models in Figure \[fig:bayes\]. The simplest model, which has no redshift evolution of the efficiency (model 1), provides a rather poor fit to the data. The massive end of the SMF is underpredicted at intermediate redshift, the fraction of quenched galaxies is too high at the low-mass end and too low at the massive end, the CSFRD at low redshift is overpredicted as are the sSFRs of massive galaxies, and the small-scale clustering of all galaxies is underpredicted. Consequently, all information criteria have a very high value and the Evidence is very low. Adding two more free parameters, $M_\mathrm{z}$ and $\epsilon_\mathrm{z}$ (model 2), improves all model selection statistics significantly as the quality of the fit has increased. The SMFs can be reproduced much better, except for very high redshift, the fraction of quenched galaxies is only too high at very low masses, the CSFRD is slightly underpredicted at high redshift, and the small-scale clustering of low-mass galaxies is too low. The Evidence increases dramatically with a logarithmic Bayes factor of 418 with respect to model 1, so that model 2 cleary outperforms model 1. Letting also the low-mass slope $\beta$ evolve with redshift (model 3), improves the fit and thus the model selection statistics further. Now all SMFs, sSFRs, and the CSFRD can be reproduced very well. Still the small-scale clustering of low-mass galaxies is too low and the fraction of quenched low-mass galaxies is too high. With a logarithmic Bayes factor of 1.6, model 3 is preferred over model 2. Allowing the high-mass slope $\gamma$ to evolve with redshift as well (model 4), does not further improve the fit, and thus the model selection statistics become worse, as the model is penalised for the additional free parameter. The logarithmic Bayes factor of model 3 with respect to model 4 is 12.5, so that model 4 does not constitute an improvement, and model 3 is still favoured. Model 5 is identical to model 3, but adds one more free parameter, letting the quenching timescale depend on stellar mass, specifically low-mass galaxies can now have longer quenching timescales and form stars longer than in model 3. While the SMFs, sSFRs, and the CSFRD are still reproduced very well, this new addition significantly improves the fit to the quenched fractions of low-mass galaxies, as low-mass satellites keep forming stars longer. Moreover, the small-scale clustering of low-mass galaxies is improved. As the quality of the fit is much higher than for model 3, all model selection statistics are better for model 5, and the logarithmic Bayes factor with respect to model 3 is 75.8, so that model 5 is clearly favoured. Adding another free parameter by letting the SFR decline exponentially after the quenching timescale, does not improve the quality of the fit. In fact, the decline timescale $\tau_\mathrm{d}$ is constrained to zero, so that the data require the quenching to be rapid. Since the fit is not improved and constrained parameter is added, the Evidence drops such that the logarithmic Bayes factor of model 5 with respect to model 6 is 20. Finally, model 7 is identical to model 5, but the stripping method has been changed, and the subhalo mass is compared to the present stellar mass instead of the peak subhalo mass. While this retains the same number of parameters as model 5, the quality of the fit slightly decreases, as the small-scale clustering is not reproduced as well as for model 5. While this effect is relatively small, the best-fit $\chi^2$ is larger by 56 for model 7, and the logarithmic Bayes Factor of model 5 with respect to model 7 is 24. Therefore, the preferred model of our analysis is model 5, which has been presented in this paper. Correlations between parameters {#sec:cov} =============================== All model parameters have been fitted using an MCMC algorithm. This does not only provide the best-fit values and their uncertainties, but from the full likelihood surface we can also determine the correlations of the model parameters. The likelihood surface projected to 2-dimensional surfaces for each parameter pair and the probability density functions for each parameter are shown in Figure \[fig:cov\]. While most parameters are uncorrelated (especially the ones governing the behaviour of satellite galaxies), we can identify a number of interesting correlations. For the characteristic mass $M_1$, the normalisation $\epsilon_\mathrm{N}$, and the low-mass slope $\beta$ of the conversion efficiency, the $z=0$ values are anti-correlated with the respective evolution slopes. This means that they can be either (relatively) high/low at high/low redshift, or reverse, as long as their sum is roughly the same. For example, for a lower normalisation at low redshift, the normalisation at high redshift needs to be higher to compensate, so that the final amount of stellar mass is the same and the SMF is reproduced. Of course, other constraints specifically constrain the high redshift parameters, such as the sSFR for low-mass galaxies at high redshift specifically constrains $\beta_\mathrm{z}$. However, as these data have a non-negligible uncertainty, the high redshift parameters can vary, and as the final amount of stellar mass is rather well constrained, the low redshift parameters then adapt leading to correlations. We can also identify weaker correlations between other parameters, such as a slight anti-correlation between the low-mass slope at low redshift $\beta_0$ and the characteristic mass at low redshift $M_0$, and consequently a correlation between the low-mass slope at high redshift $\beta_\mathrm{z}$ and $M_0$. The first of these can be understood easily, as a lower value of $M_0$ leads to higher efficiencies for low-mass galaxies (the efficiency curve is simply shifted to lower halo masses), so that the low-mass slope $\beta$ increases and becomes steeper leading to again lowered efficiencies to compensate. The second correlation is just a consequence of the first correlation and the anti-correlation between $\beta_0$ and $\beta_\mathrm{z}$. Similarly, $M_\mathrm{z}$ is correlated with $\beta_0$, but anti-correlated with $\beta_\mathrm{z}$. ![image](cov.jpg){width="95.00000%"} \[lastpage\] [^1]: [email protected] [^2]: Empirical ModEl for the foRmation of GalaxiEs	{ "pile_set_name": "ArXiv" }
--- abstract: 'We investigate the existence of the electric field reversal in the negative glow of a dc discharge, its location, the width of the well trapping the electrons, the slow electrons scattering time, and as well the trapping time. Based on a stress-energy tensor analysis we show the inherent instability of the well. We suggest that the Fermi mechanism is a possible process for pumping out electrons from the through, and linking this phenomena with electrostatic plasma instabilities. A power law distribution function for trapped electrons is also obtained. Analytical expressions are derived which can be used to calculate these characteristics from geometrical dimensions and the operational parameters of the discharge.' address: 'Department of Physics and Centro de Fisica de Plasmas, Instituto Superior Técnico, Av. Rovisco Pais, 1049-001 Lisboa, Portugal' author: - 'Mário J. Pinheiro' title: Electron trapping by electric field reversal and Fermi mechanism --- Introduction ============ The phenomena of field reversal of the axial electric field in the negative glow of a dc discharge is of great importance, since the fraction of ions returning to the cathode depends on its existence and location. Technological application of gas discharges, particularly to plasma display panels, needs a better knowledge of the processes involved. The study of nonlocal phenomena in electron kinetics of collisional gas discharge plasma have shown that in the presence of field reversals the bulk electrons in the cathode plasma are clearly separated in two groups of slow electrons: trapped and free electrons[@Kolobov]. Trapped electrons give no contribution to the current but represent the majority of the electron population. The first field reversal it was shown qualitatively to be located near the end of the negative glow (NG) where the plasma density attains the greatest magnitude. If the discharge length is enough, it appears a second field reversal on the boundary between the Faraday dark space and the positive column. Also, it was shown in the previously referred theoretical work that ions produced to the left of this first reversal location move to the cathode by ambipolar diffusion and ions generated to the right of this location drift to the anode. For a review see also [@Godyak]. Those characteristic were experimentally observed by laser optogalvanic spectroscopy [@Gottscho]. Boeuf [et al]{} [@Boeuf] with a simple fluid model gave an analytical expression of the field reversal location which showed to depend solely on the cathode sheath length, the gap length, and the ionization relaxation length. They obtained as well a simple analytical expression giving the fraction of ions returning to the cathode and the magnitude of the plasma maximum density. In the present Letter we introduce a quite simple dielectric-like model of a plasma-sheath system. This approach have been addressed by other authors [@Taillet69; @Harmon76] to explain how the electrical field inversion occurs at the interface between the plasma sheath and the beginning of the negative glow. The aim of this Letter is to obtain more information about the fundamental properties related to field inversion phenomena in the frame of a dielectric model. It is obtained a simple analytical dependence of the axial location where field reversal occurs in terms of macroscopic parameters. In addition, it is obtained the magnitude of the minimum electric field inside the through, the trapped well length, and the trapping time of the slow electrons into the well. We emphasize in particular the description of the dielectric behavior and do not contemplate plasma chemistry and plasma-surface interactions. The analytical results hereby obtained could be useful for hybrid fluid-particle models (e.g., Fiala [et al.]{} [@Boeuf94]), since simple criteria can be applied to accurately remove electrons from the simulations. On the ground of the stress-energy tensor considerations it is shown the inherent instability of the field inversion sheath. The slow electrons distribution function is obtained assuming the Fermi [@Fermi] mechanism responsible for their acceleration from the trapping well. Theoretical model ================= Lets consider a plasma formed between two parallel-plate electrodes due to an applied dc electric field. We assume a planar geometry, but extension to cylindrical geometry is straightforward. The applied voltage is $V_a$ and we assume the cathode fall length is $l$ and the negative glow + eventually the positive column extends over the length $l_0$, such that the total length is $L=l + l_0$. We have $$\label{Eq1} -V_a = l E_s + l_0 E_p,$$ where $E_s$ and $E_p$ are, resp., the electric fields in the sheath and NG (possibly including the positive column). At the end of the cathode sheath it must be verified the following boundary condition by the displacement field $\mathbf{D}$ $$\label{Eq2} \mathbf{n}.(\mathbf{D}_p - \mathbf{D}_s) = \sigma.$$ Here, $\sigma$ is the surface charge density accumulated at the boundary surface and $\mathbf{n}$ is the normal to the surface. In more explicit form, $$\label{Eq3} \varepsilon_p E_p - \varepsilon_s E_s = \sigma.$$ Here, $\varepsilon_s$ and $\varepsilon_p$ are, resp., the electrical permittivity of the sheath and the positive column. We have to solve the following algebraic system of equations $$\label{Eq5} \begin{array}{cc} l_0 E_p + l E_s & = - V_a, \\ \varepsilon_p E_p - \varepsilon_s E_s & = \sigma. \\ \end{array}$$ They give the electric field strength in each region $$\label{Eq6} \begin{array}{cc} E_s = & -\frac{V_a}{L} \left(1-\alpha + \frac{l_o \sigma}{V_a \varepsilon_s}\right)\frac{1}{1-\frac{l\alpha}{L}}, \\ E_p = & -\frac{V_a}{L} \left(1-\frac{l \sigma}{V_a \varepsilon_s} \right) \frac{1}{1-\frac{l\alpha}{L}}.\\ \end{array}$$ Here, we define $\alpha=1-\frac{\varepsilon_p}{\varepsilon_s}=\frac{\omega_p^2}{\nu_{en}^2}$. Recall that in DC case, $\varepsilon_p=1-\frac{\omega_p^2}{\nu_{en}^2}$, and $\varepsilon_s=\varepsilon_0$, with $\omega_p$ denoting the plasma frequency and $\nu_{en}$ the electron-neutral collision frequency. In fact, our assumption $\varepsilon_s=\varepsilon_0$ is plainly justified, since experiments have shown the occurrence of a significant gas heating and a corresponding gas density reduction in the cathode fall region, mainly due to symmetric charge exchanges processes which lead to an efficient conversion of electrical energy to heavy-particle kinetic energy and thus to heating [@Hartog88]. Two extreme cases can be considered: [i]{}) $\omega_p > \nu_{en}$, implying $\varepsilon_p < 0$, meaning that $\tau_{coll} > \tau_{plasma}$, i.e, non-collisional regime prevails; [ii]{}) $\omega_p < \nu_{en}$, $\varepsilon_p > 0$, and then $\tau_{coll} > \tau_{plasma}$, i.e, collisional regime dominates. From the above Eqs. \[Eq6\] we estimate the field inversion should occurs for the condition $1-\frac{l_o \alpha}{L}=0$, which give the position on the axis where field inversion occurs: $$\label{Eq8} \frac{l_o}{L} = \frac{\nu_{en}^2}{\omega_p^2}.$$ From Eq. \[Eq8\] we can resume a criteria for field reversal: it only occurs in the non-collisional regime; by the contrary, in the collisional regime and to the extent of validity of this simple model, no field reversal will occur, since the slow electrons scattering time inside the well is higher than the the well lifetime, and collisions (in particular, coulombian collisions) and trapping become competitive processes. A similar condition was obtained in [@Nishikawa69] when studying the effect of electron trapping in ion-wave instability. Likewise, a self-consistent analytic model [@Kolobov] have shown that at at sufficiently high pressure, field reversal is absent. Due to the accumulation of slow electrons after a distance $\xi_c=L-l_0$, real charges accumulated on a surface separating the cathode fall region from the negative glow. Naturally, it appears polarization charges on each side of this surface and a double layer is created with a surface charge $-\sigma_1' < 0$ on the cathode side and $\sigma_2'$ on the anode side. But, $\sigma' = (\mathbf{P} \cdot \mathbf{n})$, $\mathbf{P}=\mathbf{\mathbf{\varepsilon}_0} \chi_e \mathbf{E}$ with $\varepsilon = \varepsilon_0 (1 + \chi_e)$, $\chi_e$ denoting the dimensionless quantity called electric susceptibility. As the electric displacement is the same everywhere, we have $\mathbf{D}_0 = \mathbf{D}_1 = \mathbf{D}_2$. Thus, the residual (true) surface charge in between is given by $$\label{Eq9} \sigma = - \sigma_1' + \sigma_2'.$$ After a straightforward but lengthy algebraic operation we obtain $$\label{Eq10} \sigma = \varepsilon_p V_a \frac{B}{A},$$ where $$\label{Eq11} A = L \left( -1+ \frac{\varepsilon_0 - \varepsilon_s}{\varepsilon_p} \right) + l\left( - \frac{\varepsilon_p}{\varepsilon_s} + \frac{\varepsilon_s}{\varepsilon_p} \right),$$ and $$\label{Eq12} B = \frac{\varepsilon_0 (\varepsilon_s - \varepsilon_p)}{ \varepsilon_s \varepsilon_p }.$$ We can verify that $\sigma$ must be equal to $$\label{Eqsig} \sigma = \alpha \frac{V_a \varepsilon_0}{2 l_0}.$$ Considering that $\sigma = \varepsilon_0 \chi_e E$, we determine the minimum value of the electric field at the reversal point: $$\label{Eq13} E_m = \frac{\omega_p^2}{\nu_{en}^2} \frac{V_a}{2 l_0 \chi_e}.$$ Here, $\chi_e=\varepsilon_{rw}-1$, with $\varepsilon_{rw}$ designating the relative permittivity of the plasma trapped in the well. From the above equation we can obtain a more practical expression for the electrical field at its minimum strength $$\label{Eq14} E_m = - \frac{n_{ep}}{n_{ew}}\frac{\nu_{enw}^2}{\nu_{en}^2}\frac{V_a}{e l_0} \approx - \frac{n_{ep}}{n_{ew}} \frac{T_{ew}}{T_{ep}} \frac{V_a}{2 l_0}.$$ The magnitude of the minimum electric field depends on the length of the negative glow $l_0$. This also means that without NG there is no place for field reversal, and also the bigger the length the minor the electric field. The length of the negative glow can be estimated by the free path length $l_0$ of the fastest electrons possessing an energy equal to the cathode potential fall value $eV_a$: $$\label{} l_0 = \int_{0}^{eV_a} \frac{d w}{(N F(w))}.$$ Here, $w$ is the electrons kinetic energy and $NF(w)$ is the stopping power. For example, for He, it is estimated $p l_0=0.02 eV_a$ [@Kolobov] (in cm.Torr units, with $V_a$ in Volt). We denote by $n_{ew}$ the density of trapped electrons and by $T_{ew}$ their respective temperature. Altogether, $n_{ep}$ and $T_{ep}$ are, resp., the electron density and electron temperature in the negative glow region. By other side, we can estimate the true surface charge density accumulated on the interface of the two regions by the expression $$\label{Eq15} \sigma = \frac{Q}{A} = - \frac{n_{ep} e A \Delta \xi}{A}.$$ Here, $Q$ is the total charge over the cross sectional area where the current flows and $\Delta \xi$ is the width of the potential well. Instability and width of the potential well ------------------------------------------- From Eqs. \[Eqsig\] and \[Eq15\] it is easily obtained the trapping well width $$\label{Eq16} \Delta \xi = - \frac{e V_a}{2 m l_0 \nu_{enw}^2}.$$ It is expected that the potential trough should have a characteristic width of the order in between the electron Debye length ($\lambda_{De}=\sqrt{\frac{\varepsilon_0 kT_e}{n_e e^2}}$) and the mean scattering length. Using Eq. \[Eq16\], in a He plasma and assuming $V_a=1$ kV, $l_0=1$ m and $\nu_{en}=1.85 \times 10^{9}$ s$^{-1}$ (with $T_e=0.03$ eV) at 1 Torr ($n=3.22 \times 10^{16}$ cm$^{-3}$) we estimate $\Delta \xi \approx 2.6 \times 10^{-3}$ cm, while the Debye length is $\lambda_{De}=2.4 \times 10^{-3}$ cm. So, our Eq. \[Eq16\] gives a good order of magnitude for the potential width, which is expected to be in fact of the same order of magnitude than the Debye length. Table I present the set of parameters used to obtain our estimations. We give in Table II the estimate of the minimum electric field attained inside the well. The first field reversal at $\xi_c \approx l_{NG}$ corresponds to the maximum density $n_{ew} \gg n_{ep}$ [@Boeuf; @Tsendin2001]. So, the assumed values for the ratio of electron temperatures and densities of the trapped electrons and electrons on the NG are typical estimates. Gas $T_e$ (eV) $\sigma$ ($10^{-16}$ cm$^{2}$) ------- ------------ -------------------------------- Ar 8 4.0 He 35 2.0 O$_2$ 6 4.5 N$_2$ 4 9.0 H$_2$ 8 6.0 : Data used for $E/p=100$ V$/$cm$/$Torr. Cross sections and electron temperatures are taken from Siglo Data base, CPAT and Kinema Software, http://www.Siglo-Kinema.com []{data-label="Table1"} It can be shown that there is no finite configuration of fields and plasma that can be in equilibrium without some external stress [@Longmire]. Consequently, this trough is dim to be unstable and burst electrons periodically (or in a chaotic process), releasing the trapped electrons to the main plasma. This phenomena produces local perturbation in the ionization rate and the electric field giving rise to ionization waves (striations). In the next section, we will calculate the time of trapping with a simple Brownian model. \[Table2\] $E_m$ (V.cm$^{-1}$) $\Delta \xi$ (cm) --------------------- ---------------------- $\lesssim -2.5$ $2.6 \times 10^{-3}$ : Minimum electric field at reversal point and well width. Conditions: He gas, $p=1$ Torr, $l_0=20$ cm, $V_a=1$ kV, $\frac{T_{ew}}{T_{ep}}=0.1$, $\frac{n_{ew}}{n_{ep}}=10$.[]{data-label="Table1"} From Eq. \[Eq8\] we calculate the cathode fall length for some gases. For this purpose we took He and H$_2$ data as reference for atomic and molecular gases, resp. The orders of magnitude are the same, with the exception of Ar. Due to Ramsauer effect direct comparison is difficult. In Table III it is shown a comparison of the experimental cathode fall distances to the theoretical prediction, as given by Eq. \[Eq16\]. Taking into account the limitations of this model these estimates are well consistent with experimental data [@Brown59]. Gas $\xi_c^{teo}$ (cm) $\xi_c^{exp}$ (cm) ------- -------------------- -------------------- Ar 7.40 0.29 (Al) He 1.32 1.32 (Al) $H_2$ 0.80 0.80 (Cu) $N_2$ 0.45 0.31 (Al) $Ne$ 0.80 0.64 (Al) $O_2$ 0.30 0.24 (Al) : Comparison between theoretical and experimental cathode fall distance at p=1 Torr, $E/p$=100 V$/$cm$/$Torr. Experimental data are collected from Ref. [@Brown59]. []{data-label="Table2"} Lifetime of a slow electron in the potential well ------------------------------------------------- The trapped electrons most probably diffuse inside the well with a characteristic time much shorter than the lifetime of the through. Trapping can be avoided by Coulomb collisions [@Nishikawa69] or by the ion-wave instability, both probably one outcome of the stress energy unbalance as previously mentioned. We consider a simple Brownian motion model for the slow electrons to obtain the scattering time $\tau$, and the lifetime T of the well. A Fermi-like model will allow us to obtain the slow electron energy distribution function. Considering the slow electron jiggling within the well, the estimated scattering time is $$\label{scat1} \tau = \frac{(\Delta \xi)^2}{ \mathcal{D}_e }.$$ Here, $\mathcal{D}_e$ is the electron diffusion coefficient at thermal velocities. ----------------------------------------------------------------------------------------------------------------------------------------------------- Gas $\mathcal{D}_e$ (cm$^2$.s$^{-1}$)[^1] $\nu_{enw} (s^{-1})$ [^2] $\Delta \xi (cm)$ $\tau$ (s) T (s) -------- --------------------------------------- --------------------------- ----------------------- ------------------------ ----------------------- Ar $2.52 \times 10^6$ $8.10 \times 10^9$ $1.34 \times 10^{-3}$ $7.10 \times 10^{-13}$ $3.97 \times 10^{-5}$ He $5.99 \times 10^6$ $2.39 \times 10^9$ $1.54 \times 10^{-2}$ $3.95 \times 10^{-11}$ $1.70 \times 10^{-5}$ N$_2$ $6.11 \times 10^5$ $6.15 \times 10^9$ $2.32 \times 10^{-3}$ $8.81 \times 10^{-12}$ $1.64 \times 10^{-4}$ CO$_2$ $1.70 \times 10^6$ $3.60 \times 10^9$ $6.78 \times $2.70 \times 10^{-11}$ $5.90 \times 10^{-5}$ 10^{-3}$ ----------------------------------------------------------------------------------------------------------------------------------------------------- The fluctuations arising in the plasma are due to the breaking of the well and we can estimate the amplitude of the fluctuating field by means of Eq. \[Eq14\]. We obtain $$\label{} \delta E_m = \frac{n_{ep}}{n_{ew}}\frac{\nu_{enw}^2}{\nu_{en}^2}\frac{V_a}{e l_0^2} \Delta \xi.$$ Then, we have $$\label{} \mathcal{E}_c = \frac{\delta E_m}{E_m} = \frac{\Delta \xi}{l_0}.$$ In Table IV we summarize scattering and trapping times for a few gases. Power-law slow electrons distribution function ---------------------------------------------- As slow electrons are trapped by the electric field inversion, some process must be at work to pull them out from the well. We suggest that fluctuations of the electric field in the plasma (with order of magnitude of $\mathcal{E}_c$)act over electrons giving energy to the slow ones, which collide with those irregularities as with heavy particles. From this mechanism it results a gain of energy as well a loss. This model was first advanced by E. Fermi [@Fermi] when developing a theory of the origin of cosmic radiation. We shall focus here on the rate at which energy is acquired. The average energy gain per collision by the trapped electrons (in order of magnitude) is given by $$\label{} \Delta w = \overline{U} w(t),$$ with $\overline{U}\cong \mathcal{E}_c^2$ and where $w$ is their kinetic energy. After $N$ collisions the electrons energy will be $$\label{} w(t) = \varepsilon_{t} \exp \left( \frac{\overline{U}t}{\tau} \right),$$ with $\varepsilon_t$ being their thermal energy, typical of slow electrons. ![Slow electrons distribution function vs. energy, for the same conditions as presented in Table IV. Solid curve: Ar, broken curve: N$_2$.[]{data-label=""}](Function1.EPS "fig:"){width="3" height="4.5"}\ The time between scattering collisions is $\tau$. Assuming a Poisson distribution $P(t)$ for electrons escaping from the trapping, then we state $$\label{5} P(t)=\exp(-t/\tau)dt/T.$$ The probability distribution of the energy gained is a function of one random variable (the energy), such as $$\label{6} f_w(w)d w = P\{ w<\bar{w}<w+dw \}.$$ This density $f_w(w)$ can be determined in terms of the density P(t). Denoting by $t_1=T$ the real root of the equation $w=w(t_1=T)$, then it can be readily shown that slow electrons obey in fact to the following power-law distribution function $$\label{7} f_w(w) d w = \frac{\tau}{\bar{U}T} \varepsilon_{t}^{\frac{\tau}{\bar{U}T}} \frac{d w}{w^{1+\tau/\bar{U}T}}.$$ Like many man made and naturally occurring phenomena (e.g., earthquakes magnitude, distribution of income), it is expected the trapped electron distribution function to be a power-law (see Eq. \[7\]), hence $1 + \frac{\tau}{\mathcal{E}_c^2 T} = n$, with $n=2 \div 4$ as a reasonable guess. Hence, we estimate the trapping time to be $$\label{traptime} T \approx \frac{\tau}{\mathcal{E}_c^2 n}.$$ Fig.1 shows the slow electrons distribution function pumped out from the well for two cases: Ar (solid curve), and N$_2$ (broken curve). It was chosen a power exponent $n=2$. Those distributions show that the higher confining time is associated with less slow electrons present in the well. When the width of the well increases (from solid to broken curve) the scattering time become longer, and as well the confining time, due to a decrease of the relative number of slow electrons per given energy. This mechanism of pumping out of slow (trapped) electrons from the well can possibly explains the generation of electrostatic plasma instabilities. Note that the trapping time is, in fact, proportional to the length of the NG and inversely proportional to the electrons diffusion coefficient at thermal energies: $$\label{} T \approx \frac{l_0^2}{\mathcal{D}_e}.$$ The survival frequency of trapped electrons is $\nu_t=1/T$. As the electrons diffusion coefficient are typically higher in atomic gases, it is natural to expect plasma instabilities and waves with higher frequencies in atomic gases. This result is in agreement with a kinetic analysis of instabilities in microwave discharges [@Tatarova1]. In addition, the length of the NG will influence the magnitude of the frequencies registered by the instabilities, since wavelengths have more or less space to build-up. Table \[Table 4\] summarizes the previous results for some atomic and molecular gases. The transport parameters used therefor where calculated by solving the electron Boltzmann equation, under the two-term approximation, in a steady-state Townsend discharge [@Pinheiro] Conclusion ========== We have shown in the framework of a simple dielectric model that the magnitude of the minimum electric field (on the edge of the negative glow) depends directly on the applied voltage and is inversely proportional to the NG length. The width of the well trapping the slow electrons is directly dependent on the applied electric field and is inversely proportional to the square of the electron-neutral collision frequency for slow electrons. It is, as well, inversely proportional to the NG length, and has typically the extension of a Debye length. We state that for typical conditions of a low-pressure glow-discharge, field reversal occurs whenever $\omega_p > \nu_{en}$, due to a lack of collisions necessary to pump out electrons from the well. Furthermore, the analytical expressions obtained for the scattering and trapping time of the slow electrons are potentially useful in hybrid fluid-particle plasma modelling. [1]{} V. I. Kolobov and L. D. Tsendin, Phys. Rev. A [46]{}(12), 7837 (1992) Vladimir I. Kolobov and Valery A. Godyak, IEEE Transactions on Plasma Science, [23]{}(4), 503 (1995) Richard A. Gottscho, Annette Mitchell, Geoffrey R. Scheller, Yin-Yee Chan, David B. Graves, Phys. Rev. A, [40]{}(1), 6407 (1989) J. P. Boeuf and L. C. Pitchford, J. Phys. D: Appl. Phys. [28]{}, 2083 (1995) J. Taillet, Am. J. Phys. [37]{}, 423 (1969) Gerald S. Harmon, Am. J. Phys. [44]{}(9), 869 (1976) A. Fiala, L. C. Pitchford, and J. P. Boeuf, Phys. Rev. E [49]{} (6), 5607 (1994) E. Fermi, Phys. Rev. [75]{} (8), 1169 (1949) E. A. Den Hartog, D. A. Doughty, and J. E. Lawler, Phys. Rev. A, [38]{} (5), 2471 (1988) K. Nishikawa and Ching-Sheng Wu, Phys. Rev. Lett. [23]{}(18), 1020 (1969) A. A. Kudryatsev and L.D. Tsendin, Technical Physics Letters, [27]{}(4), 284 (2001) Sanborn C. Brown, [Basic data of plasma physics]{} (The MIT Press, Cambridge, 1959) Conrad L. Longmire, [Elementary Plasma, Physics]{} (John Wiley $\&$ Sons, New York, 1963), Section 3.7 A. Shivarova, E. Tatarova, and V. Angelova, J. Phys. D: Appl. Phys. [21]{} 1605 (1988) M. J. Pinheiro and J. Loureiro, J. Phys. D: Appl. Phys. [35]{}, 3077 (2002) [^1]: Data obtained through resolution of the homogeneous electron Boltzmann equation with two term expansion of the distribution function in spherical harmonics, M. J. Pinheiro and J. Loureiro, J. Phys. D.: Appl. Phys. [35]{} 1 (2002) [^2]: Same remark as in ${}^a$	{ "pile_set_name": "ArXiv" }
\ $^{2}$$^{}$\ $^{1}$ Chemistry and Physics of Materials Unit,\ Jawaharlal Nehru Centre for Advanced Scientific Research,\ Jakkur, Bangalore 560064, India.\ $^{2}$Solid State and Structural Chemistry Unit, Indian Institute of Science,\ Bangalore 560012, India.\ [Abstract]{} [The dynamics of hydrogen bonds among water molecules themselves and with the polar head groups (PHG) at a micellar surface have been investigated by long molecular dynamics simulations. The lifetime of the hydrogen bond between a PHG and a water molecule is found to be much longer than that between any two water molecules, and is likely to be a general feature of hydrophilic surfaces of organized assemblies. Analyses of individual water trajectories suggest that water molecules can remain [bound to the micellar surface for more than a hundred picosecond]{}. The activation energy for such a transition from the bound to a free state for the water molecules is estimated to be about 3.5 kcal/mole.]{} The study of hydrogen bond dynamics has proven to be a very useful tool [@chandler; @chandra; @bakker] to understand the origin of many fascinating dynamical properties of water which are due to its extended hydrogen bond network [@franck]. However, the same properties exhibit quite different behavior for water molecules at the surfaces of self-organized assemblies and biological macromolecules [@fleming95; @faraday96; @lang99; @nandi00; @zewail02]. Recent time domain spectroscopic measurements have shown that their dynamics is considerably slower than their counterparts in bulk water, sometimes slower by [more than two orders of magnitude]{}! While the reorientation of water molecules and solvation of ions or dipoles in bulk water proceeds with an average time constant of less than a picosecond (ps), the same at protein surfaces gets extended to hundreds of picoseconds [@fukuzaki95; @lang99; @nandi00; @zewail02]. Such slow dynamics has been observed in proteins [@lang99; @zewail02], microemulsions [@nandi00; @sarkar96], micelles [@nandi00; @pal], lipid vesicles and bilayers [@poland-cpl]. The slow dynamics of water molecules at heterogeneous surfaces seems to be universal and could be a collective effect originating from the surface and/or from the nature of the hydrogen bond network at the surface. The work of Cheng and Rossky [@rossky98] have already shown that water molecules at the surface of a protein can be structurally different from those in the bulk. A phenomenological theory developed recently proposes to explain the emergence of slow dynamics in terms of a dynamical equilibrium between bound and free water molecules at the surface [@nandi97]. Recent atomistic molecular dynamics simulations of an aqueous micellar assembly, the cesium pentadecaflurooctanoate (CsPFO), have shown that the orientational correlation function of the water molecules near the micelle show a dramatic slowing down in its long time decay [@bbjpc01]. Note that micelles are not only of great interest themselves, but they also mimic surfaces of many biological molecules. The presence of a compact hydrophobic core and a hydrophilic surface resembles proteins rather closely. In this study, we explore the two most important microscopic aspects of water dynamics near the micellar surface – the hydrogen bond lifetime of the water-PHG hydrogen bond and also the [distance dependence]{} of the dynamics of the water-water hydrogen bond. The best way to study this dynamics is to investigate the hydrogen bond time correlation functions, $S_{\rm HB}(t)$ and $C_{\rm HB}(t)$ (defined below), introduced originally by Rapaport [@rapaport] and used by Chandler [et al]{} [@chandler] to study pure water and more recently by Chandra [@chandra] to explore effects of ions in water on the lifetime of the hydrogen bond. The time correlation function $S_{\rm HB}^{\rm wPHG}(t)$ for the hydrogen bond between the polar head group and the water molecules on the surface is defined by $$S_{\rm HB}^{\rm wPHG}(t) = \langle h_{\rm wPHG}(0) H_{\rm wPHG}(t) \rangle /\langle h_{\rm wPHG} \rangle$$ where the population variable $h_{\rm wPHG}(t)$ is unity when a particular tagged hydrogen bond between a water molecule and a polar head group is hydrogen bonded at time $t$ according to an adopted definition, and zero, otherwise. On the other hand, $H_{\rm wPHG}(t)=1$ if the tagged polar head group-water hydrogen bond remains [continuously]{} hydrogen bonded during the time duration $t$ and zero otherwise. Thus, $S_{\rm HB}^{\rm wPHG}(t)$ describes the lifetime of a tagged pair. The angular brackets denote averaging over initial time values and over all water-PHG pairs. We employed a geometric definition of the $\rm wPHG$ hydrogen bond such that a water molecule was assumed to be hydrogen bonded to a surfactant if the distance between the oxygen of the water molecule and the carbon of the headgroup was less than 4.35Å, and that the oxygen of the water molecule and the oxygen of the headgroup was less than 3.5Å. These distance criteria were obtained from the first minimum in the respective pair correlation functions. For water-water hydrogen bonds, we have used the same criteria as used by Chandra [@chandra], which were found to hold good for water-water hydrogen bonds near the interface also. In order to understand the structural relaxation of a tagged hydrogen bond, one also defines the following time correlation function, $C_{\rm HB}^{\rm wPHG}(t)$, $$C_{\rm HB}^{\rm wPHG}(t) = \langle h_{\rm wPHG}(0) h_{\rm wPHG}(t) \rangle /\langle h_{\rm wPHG}\rangle$$ Unlike $S_{\rm HB}^{\rm wPHG}(t)$, this correlation function is expected to exhibit a longtime tail [@chandler; @chandra]. The surfactant in this simulation is pentadecafluorooctanoate, with cesium being the counterion, commonly referred to as CsPFO. The CsPFO-H$_2$O system has been well studied experimentally [@boden7993] and is regarded as a typical binary to exhibit micellization. The amphiphiles form disc shaped micelles, stable over an extensive range of concentration and temperature. Details of our MD simulations have been discussed elsewhere [@bbjpc01; @sbbjcp02]. The molecular dynamics simulation was carried out in the NVT ensemble for an aggregate of 62 CsPFO molecules in 10,562 water molecules. The potential for water molecules is the extended simple point charge (SPC/E) model [@berendsen87], the counterions carry a unit positive charge, which is compensated by a $+0.4e$ charge on the carbon of the octanoate headgroup and a $-0.7e$ charge on each of the oxygens of the headgroup [@watanabe91]. The equations of motion were integrated with the RESPA scheme [@tuckerman92] using the PINY-MD program [@martyna_unpub] with an outer timestep of 4 fs. The analyses reported here were carried out from different sections of a subsequent 3.5 ns trajectory. The S(t) and C(t) functions were obtained with time resolutions of 12 fs and 1 ps respectively. The results reported here are at a temperature of 300K. Figure 1 shows the time dependence of the correlation function $S_{\rm HB}^{\rm wPHG}(t)$, with the inset showing the $S_{\rm HB}^{\rm ww}(t)$ (i.e., water-water) correlation function in bulk water, for comparison. Note the lengthening of the time scale in the H-bonding with the PHG of the micelle. $S_{\rm HB}^{\rm wPHG}(t)$ can be fitted to a sum of three exponentials, with the longest time constant being equal to 9 ps which is to be compared with 0.97 ps in pure water. The amplitudes and time constants are given in Table 1 [@note1]. From these, one can obtain an average time constant of around 6.8 ps, which is thirteen times larger than its value for the water-water hydrogen bond in pure water. It is difficult to pinpoint the reason for the slowdown of $S_{\rm HB}(t)$ at the surface. We found that the water molecules at the surface form bridge hydrogen bonds involving PHGs of the nearest neighbor surfactant molecules which could stabilize the $\rm wPHG$ hydrogen bond. In figure 2, we show the hydrogen bond time correlation function $C_{\rm HB}^{\rm wPHG}(t)$. In the inset we show $C_{\rm HB}^{\rm ww}(t)$ for bulk water. Note again the lengthening of the long time decay. The values of the time constants and the respective amplitudes are given in Table 1. In this case, the long time constant stretches to more than one hundred picoseconds. The reason for this unusual long decay time can be traced back to those trajectories which leave the micellar surface to go to the bulk but return after a long time to get bonded to the [same PHG]{} at the surface. This is in line with the explanation of the long time decay of the $C_{\rm HB}(t)$ function in bulk water [@chandler; @note2]. In the present problem, the existence of such trajectories indicate presence of correlations in the surface region. The $S(t)$ function is thus a more accurate representation of the lifetime dynamics of the hydrogen bond than the $C(t)$ function, although the latter contains rich information on the correlated pair diffusion of the water molecules. Analysis of the trajectories of the individual water molecules at the micellar surface reveals an amazing richness of events. In figure 3, we show representative trajectories over 800 ps, of four arbitrarily chosen water molecules that were hydrogen bonded to the surface at time zero. The plot shows the shortest distance, $D_{\rm W-PHG}$, of each of the water molecules to the micellar surface as a function of time [@note3]. For example, the water molecule shown in the top left figure, rattles near the surface for about 150 ps, then diffuses into the bulk region, stays there for around 500 ps, and then revisits the micelle surface, presumably to form a new hydrogen bond with a PHG. It should also be noted that the micelle itself is quite fluxional over these time scales. Note also that while the $S_{\rm HB}^{\rm wPHG}(t)$ and $C_{\rm HB}^{\rm wPHG}(t)$ functions study the dynamics of a particular wPHG hydrogen bond (with average lifetimes of 6.9 and 43.7ps respectively), the apparently longer lifetime of bound water molecules shown in the trajectories in figure 3 arises from the fact that a water molecule can remain bound to the micellar surface even after its hydrogen bond with a particular PHG is broken. Such dynamical processes involving water molecules affect the way they form hydrogen bond between themselves. We have explored this aspect by studying the water-water hydrogen bond time correlation functions, for water molecules that belong to different interfacial layers. In figure 4 we show the lifetime correlation functions, $C_{\rm HB}^{\rm ww}(t)$ and $S_{\rm HB}^{\rm ww}(t)$ between two water molecules that both exist at several regions away from the surface [@note3]. This function shows a sharp slow down as the surface is approached closely. The time constant obtained from the $S_{\rm HB}$ function for the water molecules [within]{} 6Å from the micellar surface is around 25-30% larger than that for water molecules in the bulk region. The time constant for the same water molecules obtained from the $C_{\rm HB}$ function is about 45% slower than that for bulk water. Again, the $S_{\rm HB}$ function decays much faster than the $C_{\rm HB}$ one. The simulation results can be combined with a theoretical model to obtain useful information on the micelle-water interaction. The model assumes a dynamical equilibrium between the “bound” and “free” states of water molecules at the micellar surface [@nandi97]. The dynamical variable $h(t)$ represents the instantaneous population of the bound state and the correlation function, $C^{\rm wPHG}_{\rm HB}$(t) gives, under the regression hypothesis (or linear response theory) [@chandler_statmech], the decay of an initial (t=0) bound state and its formation at a later time. Therefere, we can set, in the long time, $C^{\rm wPHG}_{\rm HB}$(t) $\sim$ e$^{-{t\over\tau_{\rm BF}}}$, where $\tau_{B\rm F}$ is the time constant for the bound to free transition. From Table 1, we find the average time constant at 300K to be 43.7ps. We can obtain an activation energy $E_{\rm A}$, for this transition using transition state theory to express the rate constant ($\frac{1}{\tau_{\rm BF}}$), as, $$\frac{1}{\tau_{BF}} = \frac{k_BT}{h}e^{-\frac{E_A}{k_BT}}$$ where, $k_{\rm B}$ is the Boltzmann constant, $T$ is temperature, and $h$ is the Planck constant. $E_{\rm A}$ is then found to be 3.34 kcal/mole. A similar analysis of $C^{\rm wPHG}_{\rm HB}$(t) at 350K (data not shown here) yields a time constant of 18.9ps. Using an Arrhenius dependence of the time constant with temperature, we find that the activation energy is 3.50 kcal/mole, in close agreement with the result of the transition state theory presented above, showing the robustness of the above analysis. To conclude, we have carried out large scale atomistic molecular dynamics simulations of an aqueous micellar system, with emphasis on understanding the hydrogen bond breaking dynamics at the micellar surface. We find the rather surprising result that the hydrogen bond between the micellar polar head group and a water molecule has a much longer lifetime – almost 13 times larger than that in the bulk between two tagged water molecules. This longer lifetime could originate from the bridge bonds that the surface water molecules form with the PHGs and also from the coupling to the micelle which acts as a bath for HB excitations. Another interesting result is that the lifetime of hydrogen bonds between two tagged water molecules increase only by about 25-30% as one approaches the micellar surface. Although this slowdown is not as dramatic as the slowdown in the orientational relaxation of the water molecules with large residence times near the interface, it is rather sharp. Analysis of the individual trajectories reveal that the quasi-bound water at the micellar surface (immobilized by double or more hydrogen bond bridges) may be partly responsible for the slow dynamics. The activation energy for the transition from a bound to a free state for the water molecules on the micellar surface, estimated from both the transition state theory and from the temperature dependence of the hydrogen bond lifetime, is about 3.5 kcal/mole. The agreement between the rather simple model and the temperature dependent simulation data is remarkable. We believe that the results discussed here are generic to organized assemblies and biological macromolecules that possess a hydrophilic surface or hydrophilic pockets that are accessible to water. [Acknowledgments]{}\ It is a pleasure to thank Prof. A. Chandra for helpful comments on an earlier version of this manuscript and for pointing out Ref. [@rapaport]. The research reported here was supported in part by grants from the Council of Scientific and Industrial Research (CSIR) and the Department of Science and Technology (DST), Government of India. \ [abcd-uf]{} (a) A. Luzar, and D. Chandler, Phys. Rev. Lett. [76]{}, 928 (1996); Nature (London) [379]{}, 53 (1996); (b) A. Luzar, J. Chem. Phys. [113]{}, 10663 (2000). A. Chandra, Phys. Rev. Lett. [85]{}, 768 (2000). M.F. Kropman, and H. Bakker, Science, [291]{}, 2118 (2001); H. Xu, and B. Berne, J. Phys. Chem. B, [105]{}, 11930 (2001); S. Raugei, and M.L. Klein, J. Am. Chem. Soc., [123]{}, 9484 (2001); F. Francks, Ed. [Water, A Comprehensive Treatise]{}, (Plenum Press, New York, 1972-1982); F. Sciortino [et al.]{} Phys. Rev. Lett. [64]{}, 1686 (1990). S. Vajda [et al.]{} J. Chem. Soc. Farad. Trans., [91]{}, 867 (1995). See for example,[Hydration Processes in Biological and Macromolecular systems, Faraday Disc]{}. 1996, [103]{}, 1-394. X.J. Jordanides [et al.]{} J. Phys. Chem. B. [103]{}, 7995 (1999). N. Nandi, K. Bhattacharyya, and B. Bagchi, Chem. Rev. [100]{}, 2013 (2000). S.K. Pal, J. Peon, A.H. Zewail, Proc. Nat. Acad. Sci. [99]{}, 1763 (2002). M. Fukuzaki [et al.]{} J. Phys. Chem. [99]{}, 431 (1995); G. Otting, in [Biological Magnetic Resonance]{}, (ed. N. Ramakrishna, L.J. Berliner), KluwerAcademic/Plenum, New York, 1999, vol. 17, pp. 485. N. Sarkar [et al.]{} J. Phys. Chem. [100]{}, 15483 (1996). S.K. Pal [et al.]{} Chem. Phys. Lett. [327]{}, 91 (2000); R.E. Riter, D.M. Willard, and N.E. Levinger, J. Phys. Chem. B, [102]{}, 2705 (1998). T. Rog, K. Murzyn, and M. Pasenkiewicz-Gierula, Chem. Phys. Lett., [352]{}, 323 (2002). Y.-K. Cheng and P.J. Rossky, Nature, [392]{}, 696 (1998). N. Nandi, and B. Bagchi, J. Phys. Chem. B. [101]{}, 10954 (1997). S. Balasubramanian, and B. Bagchi, J. Phys. Chem. B [105]{}, 12529, (2001); [ibid]{} [106]{}, 3668 (2002). D.C. Rapaport, Mol. Phys., [50]{}, 1151 (1983). N. Boden, K.W. Jolley, and M.H. Smith, J. Phys. Chem. [97]{}, 7678 (1993); H. Iijima [et al.]{} J. Phys. Chem. B [102]{}, 990 (1998). S. Balasubramanian, S. Pal, and B. Bagchi, Curr. Sci. [82]{}, 845 (2002); S. Pal, S. Balasubramanian, and B. Bagchi, J. Chem. Phys. [117]{}, 2852 (2002). H.J.C. Berendsen, J.R. Grigera, and T.P. Straatsma, J. Phys. Chem. [91]{}, 6269 (1987). K. Watanabe, and M.L. Klein, J. Phys. Chem. [95]{}, 4158 (1991). M.E. Tuckerman, B.J. Berne, and G.J. Martyna, J. Chem. Phys. [97]{}, 1990 (1992). M.E. Tuckerman [et al.]{} Comp. Phys. Comm. [128]{}, 333 (2000). Note that even in bulk water, the lifetime correlation function $S_{\rm HB}^{\rm ww}(t)$ is non-exponential, with three time constants, 0.07 ps (14%), 0.49 ps (70%) and 0.97 ps (16%). The average lifetime is 0.51 ps. $C_{\rm HB}^{\rm ww}(t)$ is also non-exponential in bulk water, with time constants 3.1 ps (61%) and 17.25 ps (39%). The average time constant is 7.2 ps. The instantaneous positions of the carbon atoms of the headgroups are taken to denote the micellar surface. D. Chandler, [Introduction to Modern Statistical Mechanics*]{}, (Oxford Unversity Press, New York, 1987, p. 242). The $C_{\rm HB}^{\rm ww}(t)$ function for pure water can show a spuriously long time tail which denotes the re-formation of the hydrogen bond of a given pair. This effect should vanish in the thermodynamic limit, and the time correlation function will not possess a long time tail. The function in the inset for water molecules in the far region matches the correlation function obtained for a collection of 256 water molecules in bulk, exactly, upto 80 ps. Function Time constant \[ps\] Amplitude (%) ----------------- ---------------------- --------------- 0.3 5.9 $S_{\rm HB}(t)$ 3.6 31.5 9.1 62.6 3.4 16 $C_{\rm HB}(t)$ 29.0 63 118.5 21 : Parameters of multi-exponential fits to the water-PHG hydrogen bond time correlation functions shown in Figures 1 and 2. ![S$_{\rm HB}^{\rm wPHG}$(t) function for the hydrogen bond between the polar head group and water molecules. Inset shows the same between pairs of water molecules in pure water.](fig1.eps) ![C$_{\rm HB}^{\rm wPHG}$(t) function for the hydrogen bond between the polar head group and water molecules. Inset shows the same between pairs of water molecules in the region far away from the micelle [@note4].](fig2.eps) ![Trajectory of four water molecules for a time period of 800 ps. The shortest distance of the water molecules to the micellar surface, $D_{\rm W-PHG}$, is plotted against time.](fig3.eps) ![The location dependence of the time correlation function, C$_{\rm HB}^{\rm ww}$(t), for the hydrogen bond between pairs of water molecules located at different regions from the micellar surface [@note3]. Solid curve: Within 6Å ; Long-Dashed curve: Between 6Å and 9Å ; Dashed curve: Beyond 25Å . Inset shows the similar location dependence of the S$_{\rm HB}^{\rm ww}$(t) function for the hydrogen bond between pairs of water molecules.](fig4.eps)	{ "pile_set_name": "ArXiv" }
--- abstract: 'We investigate resolved kpc-scale stellar and nebular dust distribution in eight star-forming galaxies at $z\sim0.4$ in the GOODS fields. This is to get a better understanding of the effect of dust attenuation on measurements of physical properties and its variation with redshift. Constructing the observed Spectral Energy Distributions (SEDs) per pixel, based on seven bands photometric data from HST/ACS and WFC3, we performed pixel-by-pixel SED fits to population synthesis models and estimated small-scale distribution of stellar dust extinction. We use $\rm H\alpha / H\beta$ nebular emission line ratios from Keck/DEIMOS high resolution spectra at each spatial resolution element to measure the amount of attenuation faced by ionized gas at different radii from the center of galaxies. We find a good agreement between the integrated and median of resolved color excess measurements in our galaxies. The ratio of integrated nebular to stellar dust extinction is always greater than unity, but does not show any trend with stellar mass or star formation rate. We find that inclination plays an important role in the variation of the nebular to stellar excess ratio. The stellar color excess profiles are found to have higher values at the center compared to outer parts of the disk. However, for lower mass galaxies, a similar trend is not found for the nebular color excess. We find that the nebular color excess increases with stellar mass surface density. This explains the absence of radial trend in the nebular color excess in lower mass galaxies which lack a large radial variation of stellar mass surface density. Using standard conversions of star formation rate surface density to gas mass surface density, and the relation between dust mass surface density and color excess, we find no significant variation in the dust to gas ratio in regions with high gas mass surface densities, over the scales probed in this study.' author: - 'Shoubaneh Hemmati, Bahram Mobasher , Behnam Darvish , Hooshang Nayyeri , David Sobral , Sarah Miller' bibliography: - 'shooby.bib' title: 'Nebular and Stellar Dust Extinction across the Disk of emission-line galaxies on small (kpc) scales' --- Introduction ============ The existence of interstellar dust was suggested about a century ago, even before the great debate about the nature of galaxies (@Curtis1918). The presence of dust in galaxies was later established firmly, not only from dimming and reddening of light but also from dust scattering and Far Infrared (FIR) continuum emission. Dust can absorb more than half of the Ultraviolet (UV) and optical radiation budget of the Universe (e.g. @Calzetti2001). Most of the radiation from star formation in galaxies is emitted in the UV and optical, the wavelength range most susceptible to dust extinction and where most of the observations are taken. Therefore, without a thorough understanding of the attenuation of light of galaxies (coming from both stars and nebulae) by dust, interpretation of their physical properties (e.g. star formation rate, stellar age and stellar mass-to-light ratio) will not be accurate. A lot of progress has been made in our understanding of the attenuation of light by dust. It is now well established that the amount of extinction in galaxies is wavelength dependent. Variation of dust extinction as a function of wavelength, or the so called extinction curve, is studied from different observations of nearby galaxies and the Milky Way (e.g. @Prevot1984, @Cardelli1989, @Calzetti1994, @Gordon2003). In the extinction curve, there is information about chemical composition and sizes of dust grains. The smoothness of the FIR part of the extinction curve suggests that a variety of grain sizes exist. While the overall shape of the extinction curves measured from these local galaxies agrees, especially towards the infrared, there are some differences in the slopes, the normalizations, and the presence of the 2175 Å bump (e.g. @Stecher1965, @Calzetti2001, @Reddy2015, @Scoville2015). These differences are attributed to different metallicities and dust-to-gas ratios as well as differences in the composition of the dust grains (e.g. @Calzetti1994, @Reddy2015). Local extinction and attenuation curves are often used to correct for dust attenuation at intermediate and high redshifts. Recently, the study of dust attenuation at higher redshifts has become accessible by infrared surveys. Studies at $z \sim 1-2$ (e.g. @Scoville2015, @Reddy2015) have found attenuation curves very similar to the commonly used [@Calzetti2000] attenuation curve derived from nearby galaxies. However, other studies reported poor fits from nearby curves to high redshift galaxies and found strong spectral dependence in the attenuation curve in their sample (e.g. @Kriek2013), suggesting different star-dust geometry, dust grain properties or both. Comparison of attenuation towards nebular star forming regions inside galaxies with that of the stellar continuum hints about the geometry of dust relative to stars. Studies of local galaxies have found higher attenuation towards the nebular regions compared to the integrated dust from stellar continuum (e.g. @Calzetti2000, @Moustakas2006, @Wild2011, @Kreckel2013). This is consistent with the picture from the radiative transfer model by [@Charlot2000] in which recombination lines generated in the HII regions, the birth clouds of the most massive O stars, face an extra amount of attenuation by dust. However, at higher redshifts the picture is not as clear. By comparing star formation rates from different diagnostics, it was found that (e.g. @Erb2006, @Reddy2010, @Garn2010, @Shivaei2015, @Oteo2015) the best agreement between the SFRs is met with no extra color excess towards the nebular regions. Other studies have found quite the opposite, with higher attenuation needed towards the nebular regions (e.g. @Forster2009, @wuyts2011, @Ly2012 @Price2014, Darvish et al. (in prep)). Using a large sample of emission line galaxies from the MOSFIRE Deep Evolution Field (MOSDEF) survey (@Kriek2015), [@Reddy2015] demonstrated that the ratio of nebular to stellar attenuation is a function of the star formation and specific star formation rates of galaxies. However, there is a huge scatter in the relation. A detailed analysis of spatially resolved colors is needed to understand the source of this scatter. The amount of nebular attenuation by dust is also shown to correlate with physical properties of galaxies, such as luminosity, Star Formation Rate (SFR), mass and metallicity (e.g. @Wang1996, @Sullivan2001, @Pannella2009, @Asari2007). [@Garn2010] investigated these different dependencies and found stellar mass to be the best parameter predicting the amount of dust extinction for $z\sim0.1$ galaxies from the Sloan Digital Sky Survey (SDSS DR7; @Abazajian2009). This result was later confirmed by studies at high redshift galaxies (e.g. @Sobral2012, @Dominguez2013, @Ibar2013). It is however not clear whether these relations hold at smaller scales inside galaxies. High spatial resolution observations of galaxies in the local Universe enabled the derivation of well calibrated relations among the physical properties of galaxies (such as stellar mass, dust mass, or SFR) at $z\sim 0$ (e.g. @Calzetti2000, @Kennicutt2007, @Kreckel2013, @Boquien2011). However until recently, at intermediate and high redshifts, studies of resolved (kpc-scale) properties of galaxies was not possible. High resolution multi-waveband surveys using the Hubble Space Telescope such as Cosmic Assembly Near-infrared Deep Legacy Survey (CANDELS; @Grogin2011, @Koekemoer2011) have enabled such studies through optical and near-infrared photometric observations to $z\sim 2$ (e.g. @Wuyts2012, @Hemmati2014, @Guo2015). In these studies, resolved physical properties of galaxies were measured by fitting the observed Spectral Energy Distribution (SED) at each pixel to the theoretical stellar synthesis models. Therefore, the uncertainties inherent to SED fitting will be inevitable using only photometric data. Spectroscopic data from grism observations by surveys such as 3D-HST have improved measurements of resolved physical properties (such as SFR surface densities). Using a sample of massive galaxies at z 1 in the CANDELS and 3D-HST, [@Wuyts2013] showed that an extra amount of attenuation is needed for nebular gas for the integrated H$\alpha$ SFR to agree with the SED inferred SFR. However, one caveat of such studies is the poor spectral resolution of grism observations, which can not resolve the NII and H$\alpha$ emission lines. The contribution of the NII to the H$\alpha$+NII emission depends on the ionization radiation and metallicity of the galaxy and hence a constant ratio assumption could affect the inferred SFRs measured. The advent of Adaptive Optics (AO) aided Integral Field Spectrographs (IFS), enabled spatially resolved spectroscopic observations of intermediate redshift galaxies (e.g. @Forster2009, @Genzel2010, @Swinbank2012). While obtaining statistical samples of galaxies with IFS has become possible over the last years (e.g. @Sobral2013, @Wisnioski2015), the same is not yet true for AO aided high resolution IFS observations. In this work, we combine high resolution photometric data from CANDELS with complementary high spatial and spectral resolution spectroscopic data from Keck/DEIMOS observations to study dust distribution in a sample of emission-line galaxies. We demonstrate the usefulness of the technique on a small sample. We measure the stellar continuum and ionized gas dust extinction along the major axis of disk galaxies. Stellar Continuum and the ionized gas extinctions are measured from resolved SED-fitting per pixel (@Hemmati2014) and the Balmer decrements from Keck/DEIMOS spectra, respectively. We investigate how integrated dust measurements in galaxies compare to the spatial variation of dust along the disks. The structure of this paper is as follows. §2 presents the sample selection and data. In §3 and §4 we describe measurements from photometric and spectroscopic data, respectively. We present our results in §5 and in §6 we finish with a summary and discussion. Throughout this paper all magnitudes are expressed in AB system [@Oke1983] and we use standard cosmology with $H_{0}=70 \rm \:kms^{-1} Mpc^{-1}$, $\Omega_{M}=0.3$ and $\Omega_{\Lambda}=0.7$. Sample and Data =============== The sample for this study consists of eight disky galaxies at $z\sim0.4$ with median stellar mass of $\rm log(M_{}/M{\odot})=9.6$ and median SFR of $\rm SFR(M_{\odot}/yr)= 10.0$. These galaxies are selected from a parent sample described in detail in [@Bundy2005] and [@Miller2011]. Here, we present the selection criteria for the sample used in this study. \[\] [\|c\|c\|]{}\ \ \ \ \ \ Galaxies in the parent sample were initially adapted from a $Z_{F850LP}<22.5$ sample in the Great Observatories Origins Deep Survey (GOODS) North and South fields (@Giavalisco2004) in the redshift (spectroscopic when available) range of $0.2<z<1.3$ and were visually inspected to have prominent disks. A further magnitude cut of $K_{s} < 22.2$ was applied to ensure reliable stellar mass measurements. We then observed these galaxies with the DEIMOS (@Faber2003) instrument on KECK with long ($\sim$ 6-8 hours) exposures. The $1200$ $\rm l mm^{-1}$ grating and $1\arcsec$ slits with a central wavelength of 7500 Å were used, achieving FWHM spectral resolution of $\sim 1.7$ Å. The PA of slits are almost aligned with the galaxies major axis. This was to measure rotation curves for the galaxies in the parent sample (@Miller2011). Out of the galaxies with extended line emission, only eight have both H$\alpha$ and H$\beta$ covered in the DEIMOS spectra which are not contaminated by OH sky emission lines. These galaxies are selected for the analysis of this pilot study. The photometric data in this study comes from high resolution Hubble Space Telescope (HST) optical and near infrared images taken by the Advanced Camera for Surveys (ACS) and Wide Field Camera3 (WFC3) as part of the CANDELS. We use HST/ACS observations in the F435W, F606W, F775W and F850LP (hereafter $B_{F435W}$, $V_{F606W}$, $I_{F775W}$ and $Z_{F850lp}$) and HST/WFC3 observations in the F105W, F125W and F160W (hereafter $Y_{F105W}$, $J_{F125W}$ and $H_{F160W}$) filters. The ACS images have been multi-drizzled to the WFC3 pixel scale of 006 [@Koekemoer2011]. Figure 1 shows the stacked HST ($B_{F435W}$, $Z_{F850lp}$ and $H_{F160W}$) images as well as cutouts of H$\alpha$ and H$\beta$ emission lines from the DEIMOS 2D spectra of all galaxies in the sample. Photometric measurements ======================== Using the photometric data, we have made 2D maps of physical parameters of galaxies (such as stellar mass and star formation rate surface density, age and extinction) in the sample by measuring the observed SED for individual pixels and fitting them to template SEDs. The method and the uncertainties in the estimated parameters are discussed in [@Hemmati2014]. Here, we briefly explain the method for producing 2D maps. We use the 2D maps to produce 1D profiles along the disk of the galaxy. These profiles will then be used to directly combine/compare with spectroscopic data from DEIMOS. ID RA DEC Spec z sin (i) $Log(M_{}/M_{\odot})$ $SFR (M_{\odot}yr^{-1})$ $E(B-V)_{star}$ $E(B-V)_{nebular}$ ---- ------------- ------------- -------- --------- ------------------------ --------------------------- ----------------- -------------------- 1 53.0728469 -27.7376055 0.40 0.92 $8.98 \pm 0.11$ $1.74_{-1.41}^{+4.60}$ 0.10 $0.52\pm 0.26$ 2 189.2838000 62.3300500 0.32 1.00 $9.19 \pm 0.12$ $1.68_{-1.43}^{+5.07}$ 0.15 $0.59\pm 0.32$ 3 189.2643000 62.2753800 0.36 0.67 $9.28 \pm 0.11$ $3.31_{-2.48}^{+7.69}$ 0.15 $ 0.41\pm 0.27$ 4 189.2070900 62.2203900 0.47 0.98 $9.58 \pm 0.30$ $17.53_{-13.89}^{+40.34}$ 0.4 $0.91\pm 0.18$ 5 189.2692900 62.2811400 0.38 0.71 $9.63 \pm 0.22$ $13.83_{-8.45}^{+16.74}$ 0.4 $ 0.99\pm 0.06$ 6 189.3240100 62.2443400 0.41 0.87 $9.65 \pm 0.13$ $6.36_{-4.58}^{+9.13}$ 0.4 $0.81\pm 0.13$ 7 189.1654100 62.2574100 0.38 0.97 $9.72 \pm 0.20$ $29.28_{-18.88}^{+40.50}$ 0.45 $0.86\pm 0.15$ 8 189.2497900 62.2877800 0.36 0.88 $10.29 \pm 0.10$ $21.17_{-8.47}^{+12.86}$ 0.5 $1.22\pm0.21$ \[\] ![image](profile_example_73.pdf){width="90.00000%"} We first make $80\times80$ pixel cutouts of galaxies in seven bands HST science and rms error images. We PSF-match them to the resolution of the $H_{F160W}$ band. By multiplying the segmentation maps from SExtractor (@Bertin1996) to the PSF-matched cutouts we define the boundary of galaxy and remove any surrounding objects. Using the LePhare code (@Arnouts1999; @Ilbert2006), we then fit the SED measured for each pixel with spectral synthesis models to obtain the physical properties at that pixel (redshift is fixed to the spectroscopic redshift of the galaxy). The model library is built using BC03 (@BC03) models, Chabrier (@Chabrier2003) Initial Mass Function (IMF), solar metallicity, declining star formation histories with a range of $\tau$ (including constant and bursts) and ages less than the age of the Universe at the redshift of the galaxy in question. We use Calzetti starburst (@Calzetti2000) attenuation curve with a range of E(B-V) from zero to one. We also include nebular emission lines in the fitting procedure. LePhare code accounts for the contribution of emission lines with a simple recipe based on the Kennicutt relations (@Kennicutt1998). The following lines are included in this treatment: Ly$\alpha$, H$\alpha$, H$\beta$, \[OII\], \[OIII\]4959 and \[OIII\]5007 with different ratio with respect to \[OII\] line as described in @Ilbert2008. The intensity of the lines are scaled according to the intrinsic UV luminosity of the galaxy. The 2D maps of physical parameters from the SED fitting output correspond to the median of the probability distribution function marginalized over all other parameters. We use $16\%$ lower and higher values from the Maximum Likelihood analysis to measure the $1\sigma$ error for each parameter. Using the same library and code, we also measure integrated properties of galaxies by fitting the integrated light of all the pixels in the defined boundary of the galaxy in the same seven ACS and WFC3 bands. Table 1 summarizes the properties of the host galaxies. To be able to compare these measurements with their spectroscopic counterparts from Keck/DEIMOS, we need to convert 2D maps to 1D profiles. We combine measurements of all pixels which correspond to one spatial resolution in the DEIMOS spectra. In making the profiles, we only include pixels that are covered by the DEIMOS slit. This is to avoid uncertain slit loss corrections in the spectral measurements. Therefore, in the direction perpendicular to the major axis of the galaxy, we bin all pixels in the slit and bin again in the direction of major axis of the galaxy to match the spatial pixel scale of DEIMOS (0.12per pixel). Figure 2 shows the 2D maps and 1D profile of stellar mass, SFR and E(B-V) for one of the galaxies in the sample. Spectral measurements ===================== \[\] ![image](ha_hb73.pdf){width="70.00000%"} We measure nebular dust attenuation along the disk of the galaxies using the ratio of the first two Balmer transitions ($H\alpha$ and $H\beta$) or the so-called Balmer decrement. Ratio of the luminosities of Balmer transitions arising from HII regions around very massive stars is known to be the most practical indicator measuring attenuation by dust in these regions. In the absence of dust, the intrinsic theoretical H$\alpha$ to H$\beta$ line ratio, assuming Case B recombination, a temperature of $T=10^4K$, and an electron density of $n_e=10^2$ $\rm cm^{-3}$ is known ($F(H\alpha)/F(H\beta)=2.86$; @Osterbrock1989). Any deviation from this amount is indicative of the amount of dust attenuation. We model the line emission at each spatial resolution element of continuum subtracted DEIMOS spectra. We convert the H$\alpha$ to H$\beta$ line ratio at each spatial resolution to an extinction measure using: $$E(B-V)_{nebular}=\rm \frac{-2.5}{k(H\beta)-k(H\alpha)}log_{10}(\frac{2.86}{H\alpha/H\beta})$$ We use the Cardelli Galactic extinction curve (@Cardelli1989) for measuring $\rm k(H\beta)$ and $\rm k(H\alpha)$. A line-of-sight attenuation curve such as Cardelli, is more appropriate for recombination emission of compact HII regions compared to more extended stellar continuum attenuation curves such as [@Calzetti2000]. The latter would cause smaller $E(B-V)_{nebular}$ values by a factor of 0.9 (see @Reddy2015 for a more detailed discussion). We model and subtract the continuum by first making cutouts of $\sim 100$ Å around the emission line (H$\alpha$ or H$\beta$ in this case). We mask the wavelength range ($\sim 15-20$ Å) where we have the emission in the cutout and bin the rest of the spectra in the wavelength direction with bins of $\sim 5$ Å. The median values in each wavelength bin are then fitted with a Gaussian function in the spatial direction. We measure the continuum under emission by interpolating these values over the masked wavelength range and subtract this continuum from the cutout spectra to have the continuum free emission line. To model the line emission at each spatial resolution in the galaxy, we make small cutouts covering only the emission from the continuum subtracted spectra. To be able to trace the emission to furthest points from the center and to avoid tracing and fitting the noise in the background, we use the Otsu method (@Otsu1979). Otsu method is an unsupervised automatic thresholding algorithm which reduces a grey level image to a binary image by finding an optimum threshold to maximize the separability of the two classes. Here, using this method we separate the emission from the noise in the background by multiplying the binary image by the continuum subtracted cutout. We fitted Gaussian functions to the emission at each spatial resolution (in the wavelength/velocity direction) to measure the total flux from that resolution element. In figure 3, we show H$\alpha$ and H$\beta$ cutouts as well as the thresholded cutout and traced emission for one example galaxy. We have centered emission lines by the spectroscopic redshift, which comes from visually aligning all the emission lines. However, in order to measure the ratio of two lines we need to be more precise in centering the emissions. We center each emission line by fitting the peak of modeled Gaussians with an arctangent function (in the spatial direction) this takes care of velocity offsets due to rotation in the disk. We measure a correction factor from standard stars to account for flux calibration. This is an essential step even though we are only using the ratio of emission lines, due to the large difference between wavelength of H$\alpha$ and H$\beta$ and the wavelength dependence of flux calibration. This is a factor of $\sim 1.2-1.5$ from $z=0.32-0.47$. The Balmer decrement is then measured along the major axis of galaxies by setting a minimum signal-to-noise ratio (S/N) of 5 and 3 at each spatial resolution for H$\alpha$ and H$\beta$ lines respectively and then dividing the two. Nebular color excess ($E(B-V)_{nebular}$) along the disk is then calculated by the conversion in equation 1. The uncertainty in this ratio is calculated by the standard error propagation method using the H$\alpha$ and H$\beta$ flux uncertainties at each spatial position. In the resolved line measurement, we did not correct for the Balmer absorption because the corrections are very small and well within the uncertainty of the line measurements. Extinction uncertainties range from 0.03-0.2 magnitude and its variation is larger between galaxies compared to that of one single galaxy. We also measure the integrated or “global” Balmer decrement and nebular color excess values for the whole galaxy by extracting the 1D spectra and measuring line fluxes. We correct our measurements for the underlying Balmer absorption using the stellar population model fits. The corrections to the H$\alpha$ to H$\beta$ ratio are $\lesssim 4\%$. For simplicity, we refer to our dust measurements at each spatial distance from the center as resolved dust measurement. Results ======= Color Excess, Stellar Continuum vs. ionized gas ----------------------------------------------- ![Integrated nebular color excess of galaxies in the sample measured through the Balmer decrement compared to their stellar color excess measured through SED fitting. Nebular to stellar color excess ratio of 1 and 2.27 (@Calzetti2000) are shown with dashed and dotted-dashed lines respectively. Galaxies are color coded based on their sSFR and symbol size increases with stellar mass.](nebular_stellar_total.pdf){width="51.00000%"} \[\] [cc]{}\ \ \ \ The ratio of the nebular to stellar dust extinction has been studied extensively over the past decade. Studies of local star forming galaxies (e.g. @Calzetti2000; @Wild2011) have found larger attenuation towards the nebular regions compared to the stellar continuum. However, in almost all of these works, there is a large scatter in the nebular vs. stellar color excess relation. Recent works (@Reddy2015) have attributed the scatter in this relation, seen in samples of higher redshift star-forming galaxies, to physical properties of galaxies, specifically their stellar mass or sSFR. In Figure 4, we compare the integrated nebular and stellar color excesses of galaxies in our sample. We have color-coded our galaxies based on their sSFR and the symbol sizes increase with stellar mass. All our galaxies sit above the $1:1$ (dashed line) nebular to stellar color excess ratio, which means there is larger amount of nebular extinction compared to the continuum. While we do not find any clear trend in this ratio with neither sSFR or stellar mass, the sample size is too small to draw any strong conclusions. We now compare resolved stellar and nebular attenuation measures along the disk of galaxies (as explained in the previous section). Plotted in Figure 5, are nebular (blue circles) and stellar (red triangles) color excess as a function of distance (in kpc) from the center of the galaxy, as well as the global measurement for each galaxy (solid lines). An agreement is seen among the nebular measurements at the resolution elements with the global measured value for each galaxy. The difference between the median of the measured points and the integrated value in each galaxy ranges from zero to maximum of 0.1 magnitude (well within the uncertainty of the integrated value of each galaxy). The small offset towards higher median values, might be due to the S/N criteria which affects the spectra of the outer parts of the disk more than the central parts. The uncertainties in the measured values become considerably smaller in galaxies with larger stellar masses (the right panels). Almost all the galaxies show higher stellar continuum extinction in the central regions compared to the outer parts in the disk. This differs from the nebular color excess of the lower mass galaxies ($\rm ID=1-4$) which does not significantly vary as a function of their distance from the center. More massive galaxies ($\rm ID=5-8$) however, show higher nebular extinction towards the central parts of galaxies similar to the stellar continuum color excesses. As shown in Figure 4, the nebular to stellar color excess ratio, varies from galaxy to galaxy. By examining $\rm ID=5$ and $\rm ID=7$, two galaxies with the most extended emissions, highest S/N ratio, exact same redshift and comparable stellar masses, we see very different ionized to continuum color excess ratio. Between the two galaxies the one with the higher sSFR ($\rm ID=7$) has smaller color excess ratio, consistent with the findings of [@Price2014] for higher redshift galaxies but inconsistent with [@Reddy2015], and the framework depicted by radiative transfer models (e.g. @Charlot2000). An important factor that might be playing a role is the difference in the inclination of these two galaxies. The inclination affects observed physical properties of galaxies and in particular their surface brightness (e.g. @Holmberg1958). Many studies used inclination and surface brightness to measure the amount of extinction in disky galaxies (e.g. @Giovanelli1995, @Peletier1992), knowing that with the same surface brightness, the more inclined galaxy suffers more from dust extinction. More recent study by [@Yip2010], confirmed the result of higher stellar extinction in more inclined galaxies using a large sample of local SDSS disk galaxies. More interestingly, this study showed that the amount of the Balmer decrement stays constant with inclination. If true, this can explain some of the dispersion in the ionized vs. stellar dust extinction studies. In the case of this study, if the more inclined galaxy ($\rm ID=7$) were face-on, it would have had less stellar continuum color excess and therefore larger ratio of ionized to stellar color excess similar to the other galaxy ($\rm ID=5$). We do not find any significant correlation between the nebular to stellar color excess ratio and distance from the center of galaxies (see bottom panels of figure 5) except for the most massive galaxy ($\rm ID=8$) in which this ratio decreases with distance from the center. This is because the resolution of boxes in which the color excess is measured is still significantly larger compared to the sizes of original clouds (sub-kpc) producing the recombination lines. However, future telescopes such as Thirty Meter Telescope (TMT) equipped with integral field unit (IFU) technology and assisted by adaptive optics (AO) systems will be able to provide us with a wealth of information on these clouds at these redshifts. Variation of Color Excess with Stellar Mass Surface Density ----------------------------------------------------------- The typical dust extinction of a galaxy has been shown to depend upon different properties of the galaxy, most fundamentally on its stellar mass (e.g. @Garn2010, @Sobral2012, @Ibar2013). Here, we investigate the variation of color excess in galaxies as a function of the stellar mass surface density at each resolution element along the major axis of the disks. --------------------------------------------------------------- -- ![image](BM5.pdf){width="9cm"} ![image](BM7.pdf){width="9cm"} --------------------------------------------------------------- -- Figure 6, shows the color excess of the ionized gas as a function of stellar mass surface density for two of the galaxies. The data points are color-coded based on their distance from the center and the typical error bars are shown in the top-left corner of each panel. The stellar mass surface density maps from the resolved SED fitting are shown in the lower right part of each plot and over plotted with dashed black lines are the DEIMOS slit position and coverage. It is important to note that while the size of the disk’s major axis is about 20 kpc in these galaxies, the extent of the emission lines is only about 12 kpc. This can be either due to lack of strong emission at large distances in these galaxies, or due partly to the over subtraction of background in larger radii in the DEIMOS reduction pipeline. It is clear from figure 6 that the color excess of ionized gas is increasing from the outskirts towards the bulge of the galaxies symmetrically, similar to the stellar mass. Putting all the resolution elements of all galaxies in the sample together on the color excess vs. stellar mass surface density plot in Figure 7, we see the overall increase of the median of color excess in bins of stellar mass surface density. ![Variation of color excess as a function of stellar mass surface density in all resolution elements of the galaxies in the sample. Different colors correspond to different galaxies. There is an overall increasing trend of color excess with increasing stellar mass surface density clear from the median represented with blue stars. ](balmer_mass.pdf){width="9cm"} Dust to Gas Ratio ----------------- A tight correlation exists between the mean surface density of cold gas and the average SFR per unit area (the so-called SFR law) on global scales (e.g. @Kennicutt1998) as well as on resolved kpc-scales (@Kennicutt2007). This relation (“KS relation”) is parameterized using a power-law introduced by [@Schmidt1963]. Here, we convert our SFR surface density ($\Sigma_{SFR}$; in units of $\rm M_{\odot}yr^{-1}kpc^{-2}$) measurements to total gas (molecular and atomic) surface density ($\Sigma_{H}$; in units of $\rm M_{\odot}pc^{-2}$), using: $$\rm log \Sigma_{SFR}=(1.56\pm0.04) log \Sigma_{H} - (4.32\pm0.09)$$ derived by [@Kennicutt2007] using observations of the nearby spiral galaxy M51a. We note that as the gas surface densities in higher redshift star forming galaxies is larger compared to M51, using the KS relation is an extrapolation when dealing with high gas surface densities. There are only limited observations of local galaxies with high gas surface densities, and they show evidence of shift in the KS relation at higher redshifts @Hodge2015. Figure 8 shows the variation of nebular color excess per resolution elements along the major axis of galaxies in our sample as a function of gas surface density. There is an increasing trend of color excess with gas surface density at ($log \Sigma_{H} \gtrsim 2.5 $). We formalize this relation by fitting a third order polynomial to $E(B-V)_{nebular}$ versus $x \equiv log \Sigma_{H}$ (shown in Figure 8 with magenta dashed line). $$E(B-V)_{nebular} = 0.51x^3-3.53 x^2+7.98 x -5.21$$ It is important to note again that, the sample size is small and we might be missing a population of galaxies that could alter this fit. In figure 8, we over plot the relation of nebular color excess as a function of dust mass surface density (multiplied by a factor of hundred) derived by [@Kreckel2013], from resolved FIR (Herschel and Spitzer) and optical integral field observations of a sample of eight nearby disk galaxies. The nebular color excess versus dust mass surface density relation introduced by [@Kreckel2013] (solid cyan line in Figure 8 ), is well bracketed by the two extremes of dust geometry discussed in [@Calzetti1994] (a uniform foreground dust screen model and a mixed media model), suggesting a combination of the two effects. The shape of the color excess versus dust mass surface density resembles that with gas surface density at higher gas surface densities. This implies that the change in the dust to gas ratio is insignificant in these regions with different amounts of extinction, while this is not the case at lower gas mass surface densities. Many previous studies have assumed a fixed dust to gas ratio in galaxies (e.g. @Leroy2011, @Sandstrom2013), this result certifies the fairness of those assumption at high gas mass surface densities. The difference in the lower gas mass surface densities, however is suggestive of a variable gas to dust ratio. ![Variation of nebular color excess as a function of gas surface density in all resolution elements of the galaxies in the sample. Magenta stars show the median in bins of gas surface density and the magenta dashed line is a third order polynomial fit to blue data points. Cyan solid line shows the relation between the nebular E(B-V) and dust mass surface density (multiplied by a factor of 1500), from [@Kreckel2013].](balmer_gasmass.pdf){width="10cm"} Summary and Discussion ====================== In this work, we have measured resolved kpc-scale dust reddening along the major axis of eight emission-line disky galaxies at $z\sim 0.4$. We have used pixel-by-pixel SED fitting and Keck/DEIMOS spectra to infer stellar and nebular dust extinction at kpc-scales, respectively. While the sample size redshift range probed are small to draw robust statistical inferences about galaxy populations in general, we developed the methodology that can be practically used on much larger samples to measure dust inside galaxies at different radii from the center, using optical spectra from local galaxies to $z\sim 0.5$ and infrared spectra at higher redshifts. These kind of studies could then be used to address the variation of results for integrated galaxies at high-z found in the literature. The integrated nebular to stellar color excess ratio for the galaxies in our sample are larger than unity with median $(E(B-V)_{nebular}/E(B-V)_{stellar}) = 2.5$ and standard deviation $1.0$. Due to the small sample size a clear trend of nebular to stellar color excess ratio was not seen with either stellar mass or the sSFR. However the nebular to stellar color excess ratio dependence can be investigated for individual galaxies. We specifically compared two of the galaxies in the sample ($\rm ID=5\&7$) with the same redshift ($z=0.38$) and comparable stellar mass ($Log(M_{*}/M_{\odot}\sim 9.65$). Between the two, the one with higher SFR has smaller nebular to stellar color excess ratio. This is in agreement with the work of [@Price2014] but contrary to the overall trend seen in @Reddy2015 with integrated measurements for larger samples of galaxies at higher redshifts. The difference here can be explained due to differences in the orientations (inclinations) between the two galaxies, a parameter that is often overlooked in many studies. We also compared the integrated and resolved nebular color excess in galaxies and found a good agreement between the two, with the integrated value, being equal or slightly less than the median of resolved measurements (median $\Delta E(B-V)_{median-integrated} \sim 0.05$ mag). This small offset is mostly due to the signal to noise criteria applied to H$\alpha$ and H$\beta$ emission lines at resolution elements, which exclude the outer parts of the emission line (which also appear to have lower extinction) from the resolved measurements. We found that the resolved stellar continuum color excess profiles show higher extinction towards central regions of galaxies compared to outer parts of the disk. This is contrary the nebular color excess profiles in the lower mass galaxies which show almost flat radial profiles. The relation between the stellar mass and color excess has been studied at various redshifts (e.g. @Garn2010, @Sobral2012, @Dominguez2013). Here, we extended this relation to substructures inside galaxies, by studying the variation of nebular color excess as a function of stellar mass surface density and found an increasing amount of nebular color excess in regions with higher stellar mass surface density. This also explains to some extent, the lack of a correlation in the nebular attenuation as a function of distance from the center in lower mass galaxies in which the stellar mass surface density range covered is lower, when compared to higher mass galaxies. We also examined the relation between the nebular color excess and the gas mass surface densities, converted from the SFR surface densities. The shape of the attenuation relation for high gas mass surface densities resembles the attenuation versus dust mass surface density relation found in [@Kreckel2013], implying that the dust to gas mass ratio is not changing significantly as a function of extinction at physical scale probed in this study. It is however important to note the assumptions made in deriving this result, such as the conversion from SFR surface density to gas surface density and the dust mass surface densities which are all measured from local observations and might not hold true at higher redshifts. Resolved FIR observations with the Atacama Large Millimeter/submillimeter Array (ALMA) will be essential to examine the validity of these assumptions. We thank the anonymous referee for insightful comments which greatly improved the quality of this manuscript. This work is based on observations at the W.M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California and NASA. S.H wishes to thank B. Siana for very constructive comments and I. Shivaei for useful discussions. DS acknowledges financial support from the Netherlands Organization for Scientific research (NWO) through a Veni fellowship, from FCT through a FCT Investigator Starting Grant and Start-up Grant (IF/01154/2012/CP0189/CT0010) and from FCT grant PEst-OE/FIS/UI2751/2014.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We develop a method for counting words subject to various restrictions by finding a combinatorial interpretation for a product of weighted sums of Laguerre polynomials with parameter $\alpha = -1$. We describe how such a series can be computed by finding an appropriate ordinary generating function and applying a certain transformation. We use this technique to find the generating function for the number of $k$-ary words avoiding any vincular pattern that has only ones, as well as words cyclically avoiding vincular patterns with only ones whose runs of ones between dashes are all of equal length.' author: - \| Jair Taylor\ Department of Mathematics\ University of Washington\ Seattle, WA, U.S.A.\ [email protected]\ title: 'Counting words with Laguerre series' --- Introduction ============ Consider the following simple problem. How many arrangements of the word “WALLAWALLA” are there with no LLL, AAA or WW as consecutive subwords? Perhaps surprisingly, the answer can be calculated by performing a certain integral. In fact, there are $$\int_0^\infty e^{-t}\left(\frac{1}{24}t^4 - t^2 + t\right)^2 \left(\frac{1}{2}t^2 - t\right) \, dt = 1584$$ such words, as we will see. More generally, we will develop a technique by which the generating function for the number of words subject to various conditions can be found by integrating a product against $e^{-t}$, $t{\geqslant}0$. Since $\int_0^\infty e^{-t}t^n \, dt = n!$, this amounts to applying the linear functional $\Phi: \R[t] \mapsto \R$ given by $t^n \mapsto n!$. Define a [word]{} $W$ on an alphabet $S$ to be an ordered list $s_1 \cdots s_n$ of letters $s_i \in S$. A subword of $W$ is a word $s_l s_{l+1} \cdots s_{l+m}$. Note that we require the indices in a subword to be consecutive, while some authors do not. Define a [factorization]{} $\phi$ on $S$ to be an ordered list $(\phi_1)(\phi_2) \ldots (\phi_k)$ of non-empty words $\phi_1, \ldots, \phi_k$ using letters from $S$. We call $\phi_1, \ldots, \phi_k$ the [factors]{} or [parts]{} of $\phi$. We abuse notation slightly by identifying a word $W$ with the factorization $(W)$ in one part, and the empty word with the factorization having no parts, writing $\emptyset$ for both. Given a factorization $\phi$ on $S$ and $T \subseteq S$, let $\phi\|_T$ be the factorization whose parts are the maximal subwords of the parts of $\phi$ that use only letters from $T$, the parts of $\phi\|_T$ being ordered by their occurrence in $\phi$. We call $\phi\|_T$ the restriction of $\phi$ to $T$. For example, if $S = \{a,b\}$ and $T = \{a\}$, then the restriction of the factorization $(aabba)(aab)(b)(aaa)$ to $T$ is $\phi\|_T = (aa)(a)(aa)(aaa)$, while $\phi\|_{\{b\}} = (bb)(b)(b)$. If $\phi$ contains no letters from $T$, we define $\phi\|_T$ to be the empty factorization. The methods given here will allow us to count words where only certain restrictions can occur. For example, consider the words $W$ so that for each $s \in S$, the factors of $W\|_{\{s\}}$ are all length one. This is equivalent to requiring that $W$ have no adjacent, identical letters. Such words are sometimes called [Carlitz]{}, after Leonard Carlitz. Thus “PEAR” is Carlitz, but “APPLE” is not. More generally, let $A_1$ and $A_2$ be two sets of factorization on disjoint alphabets $S_1$, $S_2$, respectively. Let $S = S_1 \cup S_2$, and denote by $A_1* A_2$ the set of factorizations $\phi$ of words on $S$ so that $\phi\|_{S_1} \in A_1$ and $\phi\|_{S_2} \in A_2$. For example, if $A_1 = \{\emptyset, (a),(a)(a),(a)(a)(a), \ldots\}$, $B_2 = \{\emptyset, (b),(b)(b),(b)(b)(b), \ldots\}$, then $A_1 * A_2$ is the set of factorizations on $\{a,b\}$ so that each factor is a Carlitz word. Thus $(ab)(a)(a)(baba) \in A_1 * A_2$, but $(b)(aab)(ba) \notin A_1 A_2$. Perhaps surprisingly, the problem of counting words in $A_1A_2$ can be reduced to multiplication of certain power series. If $A$ is a set of factorizations, we define a weight $w$ on $A$ to be a polynomial-valued function on $A$ that obeys a certain combinatorial restriction. Then we will define a power series $f_{A,w}(t)$, the associated [Laguerre series]{} for $A$, in terms of the generalized Laguerre polynomials with parameter $\alpha = -1$. The key fact we will use is the rule (Theorem \[key\]) $$f_{A_1A_2,w}(t) = f_{A_1,w}(t) \cdot f_{A_2,w}(t).$$ Furthermore, we will show in Proposition \[phifact\] that $\Phi(f_{A,w}(t))$ gives the weight of all factorizations in $A$ with one or no parts, which we identify with words. By applying $\Phi$ to a product of Laguerre series we may count a variety of sets of restricted words, especially when the restrictions are on the length of runs of particular letters. In Section 3, we describe the transformation $T$ that turns certain ordinary generating functions into Laguerre series. The transformation can be described in terms of the Laplace transform, and so can be easily implemented in mathematics software packages. We use $T$ to determine the Laguerre series for different sets of factorizations $A$, and use them to derive formulas and generating functions which count words that obey various restrictions. In particular, we can use this technique to analyze certain pattern avoidance problems. A vincular, or generalized, pattern is a pattern with dashes such as $13 \d 2$. This is a generalization of classical permutation patterns where the dashes are used to indicate that the numbers on either side are not required to be adjacent, but all others are. These patterns were first studied by Babson and Steingrímsson [@Steingrimsson], who showed that many statistics of interest can be classified in terms of vincular patterns. The term [vincular]{} itself was coined by Claesson in [@Claesson], from the Latin [vinculare]{}, to bind. Words avoiding vincular patterns are studied in [@Bernini; @Burstein; @Mansour2; @Mansour1; @Mansour5; @Mansour6]. In this paper we will study vincular patterns with all ones, such as $111\d 11$. Although such patterns are useless in the context of permutations, where only the pattern $1$ can be contained, they are meaningful in the context of general words on the alphabet $\N$ where letters may be repeated. Since all of our patterns will have only ones, we give the definition in this context. A word $W = s_1 \cdots s_l$, with each $s_i$ in some alphabet $S$, contains a vincular pattern $\tau = 1^{m_1}\d\cdots\d 1^{m_n}$ if there is a subsequence of $W$ consisting of $m = m_1 + \ldots + m_n$ identical letters of which the first $m_1$ are consecutive, the next $m_2$ are consecutive, and so on. For example, a word contains $111\d 11$ if it has five appearances of the same letter in the word so that the first three and the last two are consecutive. Formally, we say that $W$ contains $\tau$ if there are indices $1 {\leqslant}i_1 < i_2 < \ldots < i_m {\leqslant}l$ with $s_{i_1} = \ldots = s_{i_m}$ and $i_{j+1} - i_j = 1$ for $j \neq m_1, m_1 + m_2, \ldots, m_1 + \ldots + m_{n-1}$. Otherwise, we say that $W$ avoids $\tau$. In Section $4$, we give a formula to calculate the generating function for the number of words avoiding any such vincular pattern with only ones. This formula involves the use of the maps $T$ and $\Phi$, but these can be easily calculated. For example, we can use Sage to compute the the generating function $\sum_W x^{\len(W)}$ where the sum is taken over all ternary words $W$ avoiding the pattern $11 \d 11$, where $\len(W)$ is the ${\it length}$ of $W$, the number of letters counting multiplicity: $$\frac{6 \, x^{7} - 6 \, x^{6} + 6 \, x^{5} - 2 \, x^{4} - 5 \, x^{3} + 9 \, x^{2} - 5 \, x + 1}{16 \, x^{4} - 32 \, x^{3} + 24 \, x^{2} - 8 \, x + 1} = 1 + 3 x + 9 x^{2} + 27 x^{3} + 78 x^{4} + 222 x^{5}+ \ldots.$$ We give a cyclic version of this result in Section $5$ for the case of patterns $1^m \d 1^m \d \cdots \d 1^m$, where all runs of ones are the same length. This gives the generating functions for words so that any cyclic permutation of their letters avoids such a pattern. This generalizes a result of Burstein and Wilf [@Wilf] which gives the generating function for the number of words cyclically avoiding $1^m$. Acknowledgments {#acknowledgments .unnumbered} =============== The author would like to thank Sara Billey, Ira Gessel, Silvia Heubach, Yannick van Huele, Toufik Mansour, Brendan Pawloski, Austin Roberts, Byron Schmuland and William Stein for their extraordinary support. Without them this paper would still be “floating in platonic heaven”. Laguerre polynomials and Laguerre series ======================================== Define the polynomials $l_k(t)$ by their generating function $$\begin{aligned} \label{laguerregf} \sum_{k=0}^\infty l_k(t) x^k = e^{\frac{tx}{1+x}}.\end{aligned}$$ The first few such polynomials are $$\begin{aligned} l_0(t) &= 1\\ l_1(t) &= t\\ l_2(t) &= \frac{1}{2}t^2 - t\\ l_3(t) &= \frac{1}{6}t^3 - t^2 + t.\end{aligned}$$ These polynomials are a form of Laguerre polynomial. Specifically, $l_k(t) = (-1)^k L_k^{(-1)}(t)$ where $$L_k^{(\alpha)}(t) = \sum_{i=0}^k (-1)^i \binom{k + \alpha }{k - i} \frac{t^i}{i!}$$ defines the generalized Laguerre polynomials. They have been found to have a number of interesting combinatorial properties, beginning with their use by Even and Gillis to count generalized derangements when $\alpha$ is set to $0$ in [@Even]. This was later extended by Foata and Zeilberger who use $\alpha$ to keep track of the number of cycles [@Doron3]. For our purpose, we will take $\alpha = -1$. Recall that a word on $[k] = \{1,2, \ldots, k\}$ is called $k$-ary, and a word in which no two adjacent letters are the same is called a Carlitz word. In 1988, Ira M. Gessel [@Gessel Section 6] used his theory of generalized rook polynomials to find an explicit formula for the number of Carlitz arrangements of a given multiset of letters. We present here an unlabeled version. \[fundamental\] Let $\Phi$ be the linear functional on polynomials in $t$ given by $\Phi(t^n) = n!$. Given nonnegative integers $n_1, \ldots, n_k$, the number of $k$-ary Carlitz words with the letter $i$ used exactly $n_i$ times is $$\Phi\biggl(\prod_{i=1}^k l_{n_i}(t)\biggr).$$ For example, in [@Mississippi] the authors consider the “Mississippi Problem”. How many arrangements of the letters in the word “MISSISSIPPI” have no adjacent letters the same? We can use the preceding theorem to calculate this directly. There is one $M$, four $I$’s, four $S$’s, and two $P$’s. So the solution is $$\Phi(l_1(t)l_4(t)l_4(t)l_2(t)) = \int_0^\infty e^{-t} \, \Big( t \Big) \, \left( \frac{1}{24}t^4 - \frac{1}{2}t^3 + \frac{3}{2} t^2 - t\right)^2\left(\frac{1}{2}t^2 - t\right)\, dt = 2016.$$ Our proof of Theorem \[fundamental\] does not involve rook polynomials, but is based on the following surprisingly simple expression for the generating function for the number of Carlitz words with a given set of letters. \[carlitzgf\] Let $\mathcal{C}$ be the set of Carlitz words over the alphabet $\N$. For a word $W = s_1\cdots s_n$ define its weight to be $w(W) = x_{s_1}\cdots x_{s_n}$, so that the exponent of $x_i$ records the number of occurrences of the letter $i$. Then $$\begin{aligned} \label{smirnov} \sum_{W \in \mathcal{C}} w(W) = \frac{1}{1-\sum_i \frac{x_i}{1+x_i}}.\end{aligned}$$ Lemma \[carlitzgf\] can be found in the book by Goulden and Jackson [@Goulden-Jackson2 2.4.16]. They attribute it MacMahon’s book from 1915 [@MacMahon] and Smirnov, Sarmanov, and Zaharov in (1966) [@Smirnov], and so counting Carlitz words is sometimes known as the Smirnov problem. The case $x_i = x^i$ was found, apparently independently, by Carlitz in 1977 [@Carlitz]. Asymptotics of Carlitz compositions are investigated using this formula by Knopfmacher and Prodinger in [@Knopfmacher]. This result can be generalized to the Carlitz-Scoville-Vaughan Theorem [@CSV], a reciprocity result relating the words for which only a given set of pairs of letters are adjacent to the words where none of these pairs are adjacent. We have $$\begin{aligned} \frac{1}{1-\sum_i \frac{x_i}{1+x_i}} &= \sum_{n=0}^\infty \left(\sum_i \frac{x_i}{1+x_i}\right)^n\\ &= \sum_{n=0}^\infty \left(\sum_i \left( x_i -x_i^2 + x_i^3 - x_i^4 + \ldots\right)\right)^n .\\\end{aligned}$$ We see that this sum represents a method of choosing a word $W$. We choose a number of steps, $n$, and at each step we choose a letter $i$ and append any number of copies of it. Suppose $W$ is not Carlitz, and write $W = W_1iiW_2$ where $W_1$ is Carlitz so that the subword $ii$ is the first adjacent repetition. If the $i$’s were chosen in separate steps we may merge the steps together, and if the $i$’s were chosen in the same step we may split the steps apart. This gives a bijection between the ways $W$ may be formed in these two cases, and the monomials representing them in the power series above will appear with opposite sign. Thus any such $W$ will be canceled in the sum. The remaining Carlitz words can only be represented in one way, with positive sign. Now we can give a simple proof of Gessel’s formula. Note that $$\Phi(e^{tf}) = \frac{1}{1-f}$$ for any $f$ that is constant with respect to $t$ when both sides are defined, where $\Phi$ is extended in the natural way to some power series. Then we see $$\begin{aligned} \Phi\left(\prod_{i=1}^\infty \exp\left(\frac{tx_i}{1+x_i}\right)\right) &= \Phi\left( \exp\left(t\sum_{i=1}^\infty\frac{x_i}{1+x_i}\right)\right)\\ &= \frac{1}{1 - \sum_{i=1}^\infty \frac{x_i}{1+x_i}}\\\end{aligned}$$ and if we take the coefficient of $x_1^{k_1} x_2^{k_2} \cdots x_m^{k_m}$ on both sides we get the desired formula by Lemma \[carlitzgf\] and the generating function . Using Theorem \[fundamental\], it is easy to see combinatorially that $$\begin{aligned} \label{orthogonality} \Phi(l_i(t) l_j(t)) = \left\{ \begin{array}{ll} 2 & \text{if $i = j$} \\ 1 & \text{if $\|i-j\| = 1$}\\ 0 & \text{if $\|i-j\| > 1$}\\ \end{array} \right.\end{aligned}$$ and so the polynomials $l_k(t)$ are “almost” orthogonal with respect to $\Phi$. Note that $l_k(t)$ is a polynomial of degree $k$; so the matrix of $l_k$’s expanded into powers of $t$ is triangular with no zeroes on the diagonal, and so $\{l_k\}_k$ forms a basis of $\R[t]$. It is natural to ask, then, what is the expansion of $l_i(t) l_j(t)$ in this basis? These are known as [linearization coefficients]{}. The linearization coefficients of general Laguerre polynomials, with $\alpha$ indeterminate, is known [@Doron3; @zeng], but we will need a combinatorial interpretation of the case $\alpha = -1$. Denote by $n_{i,j,k}$ the number of factorizations over the alphabet $\{a,b\}$ with $k$ parts and exactly $i$ $a$’s and $j$ $b$’s so that each part is Carlitz. For example, $n_{2,5,3} = 6$: the possibilities are $(bab)(bab)(b)$, $(babab)(b)(b)$ and the different permutations of these sets of factors. \[linearization\] We have, for all $i,j \in \N$, $$l_i(t)l_j(t) = \sum_k n_{i,j,k}l_k(t).$$ Note that if $p(t) = a_0 + a_1t + \ldots +a_nt^n$ is a polynomial and $\Phi(t^m p(t)) = 0$ for all $m$, then $$a_0m! +a_1(m+1)! + \ldots + a_n(n+m)! = 0.$$ This is a homogenous linear recurrence relation with constant coefficients for the factorial sequence, which is impossible unless $a_0 = a_1 = \ldots = a_n = 0$ since it the factorial is superexponential. Since $\{l_k(t)\}_k$ forms a basis for $\R[t]$, if $\Phi(p(t)l_k(t)) = 0$ for all $k$ then we can conclude $p(t) = 0$. So it is enough to show that for any nonnegative integers $i,j,m$, $$\Phi\left( l_i(t)l_j(t) l_m(t) \right) = \phi\left(\sum_k n_{i,j,k} l_k(t) l_m(t) \right).$$ We know that the left hand side counts the number of Carlitz arrangements of $i$ $a$’s, $j$ $b$’s, and $m$ $c$’s, while the right hand side gives the total number of pairs $(\phi, W)$ where $\phi$ is a factorization in $k$ parts with $i$ $a$’s and $j$ $b$’s with each part Carlitz, and $W$ is a Carlitz word with $k$ $x$’s and $m$ $c$’s. There is a simple bijection between these sets. Given such a pair $(\phi, W)$, we can get a Carlitz arrangement of $i$ $a$’s, $j$ $b$’s and $m$ $c$’s by replacing the $i$th $x$ of $W$ with the $i$th part of $\phi$. For example, if $\phi = (ab)(bab)$ and $W = cxcx$, we get the Carlitz word $cabcbab$. This process is reversible: given a Carlitz word on $a,b,c$ we replace the $c$’s by parentheses to make a factorization $\phi$ with only the letters $a$ and $b$, and to get $W$ we replace each maximal subword that does not contain $c$ by a single $x$, getting a word with only $c$’s and $x$’s. For example, given the word $abcbcab$, we get the pair $W = xcxcx$, and $\phi = (ab)(b)(ab)$. The maximality condition guarantees that $W$ will be Carlitz. Given a set of factorizations $A$ on an alphabet $S$, a [weight]{} is a function $w$ from $A$ and all of the restrictions of factorizations in $A$ into a polynomial ring $\R[x_1, x_2, \ldots]$ that commutes with restriction in the sense that if $\phi \in A$ and $T \subseteq S$, then $w(\phi) = w(\phi\|_T) \, w(\phi\|_{S\backslash T})$. Note that in particular, if $A = A_1 A_2$ for some sets of factorizations $A_1,A_2$ then $w$ is also a weight on $A_1$ and $A_2$. Also note that taking $T$ to be empty forces $w(\emptyset) = 1$. Typically we will take the weight $w(\phi)$ to be a monomial $x_1^{n_1(\phi)} x_2^{n_2(\phi)} \cdots x_m^{n_m(\phi)}$ where each $n_i(\phi)$ is a statistic so that $x_i^{n_i(\phi)}$ is multiplicative in the above sense. In all our examples it will be obvious that $w$ is a weight. Examples include - $\len(\phi)$, the length of $\phi$ - $\operatorname{sum}(\phi)$, the sum of the letters in $\phi$ if the symbols in $\phi$ are nonnegative integers - the number of distinct symbols in $\phi$ - the number of appearances of a particular symbol or simply $w=1$ if we wish to enumerate a finite set. We write $\parts(\phi)$ for the number of parts of $\phi$; but $x^{\parts(\phi)}$ is in general [not]{} a weight. Let $A$ be a set of factorizations on an alphabet $S$ and $w$ be a weight on $A$. Define the [Laguerre series]{} of $A$ with respect to $w$ to be the formal power series $$f_{A,w}(t) = \sum_{\phi \in A} w(\phi) l_{\parts(\phi)}(t)$$ when this sum is well-defined as a formal power series. For convenience we will omit the $w$ in the subscript when $w=1$, writing $f_{A,1}(t)$ as $f_A(t)$. Convergence of Laguerre series is studied in, e.g., [@Pollard; @szasz; @weniger], where they are defined as series $$\sum_n \lambda_n^{(\alpha)}L_n^{(\alpha)}(t).$$ Our definition is slightly different, as we always take $\alpha = -1$ and our coefficients may be weighted. \[phifact\] Assume $A$ is a set of factorizations and $w$ is a weight on $A$. Let $\Phi$ be the linear operator so that $\Phi(t^n) = n!$ and $\Phi$ fixes any other variables. Then $$\Phi( f_{A,w}(t)) = \sum_W w(W)$$ when both sides are defined, where the sum is over allowed words $W \in A$ which are factorizations with one or no parts. We have $$\begin{aligned} \Phi(f_{A,w}(t)) &= \sum_{\phi \in A} w(\phi) \Phi( l_{\parts(\phi)}(t))\\\end{aligned}$$ and $\Phi( l_{\parts(\phi)}(t))$ is $1$ when $\phi$ has one or no parts and is $0$ otherwise by since $l_0(t) = 1$ and $l_1(t)= t$. It is easy to see that if $A_1$ and $A_2$ are disjoint sets of allowed factorizations on a common alphabet $S$, $w$ is a weight on $A_1 \cup A_2$, then $$f_{A_1 \cup A_2,w}(t) = f_{A_1,w}(t) + f_{A_2,w}(t).$$ More interesting, perhaps, is the combinatorial interpretation of a product of Laguerre series. This is our main theorem on the combinatorial properties of Laguerre series. Recall that if $A_1,A_2$ are sets of factorizations on disjoint alphabets $S_1,S_2$, then $A_1* A_2$ was defined to be the set of factorizations $\phi$ of words on $S = S_1 \cup S_2$ so that $\phi\|_{S_1} \in A_1$ and $\phi\|_{S_2} \in A_2$. \[key\] Let $S_1$ and $S_2$ be disjoint alphabets with sets of allowed factorizations $A_1$,$A_2$ respectively, and let $w$ be a weight on $A_1 * A_2$ (and hence on $A_1$ and $A_2$.) Then $$f_{A_1 * A_2,w}(t) = f_{A_1,w}(t)\cdot f_{A_2,w}(t).$$ By Lemma 2.3, $$\begin{aligned} f_{A_1,w}(t)\cdot f_{A_2,w}(t) &= \sum_{\phi_1 \in A_1} \sum_{\phi_2 \in A_2} w(\phi_1) w(\phi_2) l_{\parts(\phi_1)}(t) l_{\parts(\phi_1)}(t)\\ &= \sum_{\phi_1 \in A_1, \phi_2 \in A_2, k{\geqslant}0} n_{\parts(\phi_1), \parts(\phi_2),k} w(\phi_1) w(\phi_2) l_k(t).\end{aligned}$$ Fix $\phi_1 \in A_1, \phi_2 \in A_2$. It is enough to show that $n_{\parts(\phi_1), \parts(\phi_2),k}$ is the number of factorizations with $k$ parts on $S_1 \cup S_2$ whose restrictions to $S_1$ and $S_2$ are $\phi_1$ and $\phi_2$, respectively. Then each allowed word of $A_1 * A_2$ will then be represented exactly once in the series $f_{A_1,w}(t)\cdot f_{A_2,w}(t)$, giving $$f_{A_1 * A_2,w}(t) = \sum_{\phi \in A_1 * A_2} w(\phi) l_{\parts(\phi)} = f_{A_1,w}(t)\cdot f_{A_2,w}(t).$$ For fixed $k$, we will construct a simple bijection from the set of triples $(\phi,\phi_1, \phi_2)$ where $\phi_1$, $\phi_2$ are factorizations in $A_1,A_2$ respectively and $\phi$ is a factorization on the alphabet $\{a,b\}$ with $\parts(\phi_1)$ $a$’s and $\parts(\phi_1)$ $b$’s so that each part is Carlitz, and the set of factorizations $\phi_3$ of $A_1 * A_2$ with $k$ parts. Let $\phi_3$ be the factorization created by replacing the $n$th $a$ in $\phi$ with the $n$th part of $\phi_1$, and the $n$th $b$ with the $n$th part of $\phi_2$. Then by construction $\phi_3 \in A_1 * A_2$: its restrictions are $\phi_1$ and $\phi_2$. Furthermore, given an allowed factorization $\phi_3 \in A_1 * A_2$ with $k$ parts so that $\phi_{S_1} = \phi_1$, $\phi_{S_2} = \phi_2$, we can reconstruct the factorization $\phi$ of a word on $\{a,b\}$ by replacing each subword of a factor of $\phi_3$ that uses only the letters of $S_1$, and is maximal with respect to this condition, by an $a$ and each maximal subword using only letters of $S_2$ by a $b$. For example, if $S_1 = \{1,2\}$ and $S_2 = \{3,4\}$, with $\phi_3 = (123,2213,34413)$, we get the word $\phi = (ab,ab,bab)$. No part of $\phi$ can have $aa$ or $bb$ by the maximality condition. These two algorithms are inverse to each other, establishing the theorem. Inductively, we see that if $A_1, \ldots, A_n$ are sets of factorizations on disjoint alphabets and $w$ is a weight on $A_1 * \cdots * A_n$, then $f_{A_1 \cdots A_n,w}(t) = f_{A_1,w}(t)\cdots f_{A_n,w}(t)$. Thus we have the following extension of Lemma \[linearization\]: Let $S = \{a_1, \ldots, a_m\}$ be an alphabet of $m$ distinct letters, and $r_1, \ldots, r_m$ be fixed nonnegative integers. Let $n_k$ denote the number of factorizations on $n$ letters in $k$ parts so that each part is Carlitz, and the $i$th letter $a_i$ of $S$ is used exactly $r_i$ times for each $i$. Then $$\prod_{i=1}^m l_{r_i}(x) = \sum_{k{\geqslant}0} n_k l_k(x).$$ Let $A_i$ consist only of the factorization $(a_i)(a_i) \cdots (a_i)$, with exactly $r_i$ factors. Then $A_1 * \cdots * A_m$ consists of those factorizations for which each part is Carlitz and the letter $a_i$ is repeated $n_i$ times so we may apply Theorem \[key\]. Alternately, we can adapt the proof of Lemma \[linearization\]. Computing Laguerre series ========================= The Laguerre series for a set of factorizations would not be especially useful if it was difficult to compute. Fortunately, there is an efficient method to calculate them in some situations. It may be difficult to find a convenient formula for the coefficients of $l_k(t)$ in a given Laguerre series $f_{A,w}(t)$, but this is not needed to find an expression for $f_{A,w}(t)$. It is enough to find the ordinary generating function. Specifically, we define $$g_{A,w}(u) = \sum_{ \phi \in A} w(\phi) u^k.$$ If a nice form of $g_{A,w}(u)$ is known, we may obtain the Laguerre series $f_{A,w}(t)$ by applying the linear transformation $T$ that sends $u^k$ to $l_k(t)$. As it happens, $T$ can be computed in many situations using the inverse Laplace transform. We have $$\mathcal{L}\{l_k(t)\} = \frac{1}{s(1-s)} \left( \frac{1-s}{s}\right)^k$$ for $k{\geqslant}1$, where $\mathcal{L}$ is the Laplace transform; this is easily proved from the formula for $l_k(t)$ in terms of the generalized Laguerre polynomials, the fact that $\mathcal{L}\{ t^i\} = \frac{i!}{s^{i+1}}$, and the binomial theorem. Therefore, if $g_{A,w}(0) = 0$, we have by linearity $$T\{g_{A,w}(u)\} = f_{A,w}(t) = \mathcal{L}^{-1}\left\{ \frac{g_{A,w}(\frac{1-s}{s})}{s(1-s)}\right\}$$ when the right-hand side is well-defined. If $g_{A,w}(0) \neq 0$, we can calculate $$f_{A,w}(t) = T\{g_{A,w}(u) - g_{A,w}(0)\} + g_{A,w}(0) = \mathcal{L}^{-1}\left\{ \frac{g_{A,w}(\frac{1-s}{s}) - g_{A,w}(0)}{s(1-s)}\right\} + g_{A,w}(0)$$ since $l_0(t) = 1$. The use of the inverse Laplace transform here is not central to the theory, but it is convenient since many software packages provide symbolic calculation of the inverse Laplace transform, making it easy to implement the transformation $T$. In many cases, we can use standard generating function techniques to calculate $g_{A,w}(u)$. For example, let $A$ be the set of all factorizations on a single-letter alphabet, and let $w$ be the weight $w(\phi)= x^{\len(\phi)}$. Then we see that $$g_{A,w}(u) = \sum_{k=0}^\infty u^k(x+x^2 + \ldots)^k$$ since we choose a factorization $\phi$ by choosing the number of parts and then choosing the size of each part. We may simplify this expression to get $g_{A,w}(u) = \frac{1+x}{1 + x - ux},$ and calculate $f_{A,w}(t) = T\{g_{A,w}(u)\} = e^{tx}$. Taking the coefficient of $x^k$ shows that the Laguerre series for all factorizations on this single-letter alphabet with length $k$ is $\frac{t^k}{k!}$. Thus Proposition \[phifact\] and Theorem \[key\] tell us that $$\Phi\left(\prod_{k=1}^n \sum_{i=0}^{r_k} \frac{t^i}{i!}\right)$$ is the total number of words on an alphabet $S = \{c_1, \ldots, c_n\}$ so that $c_k$ is used at most $r_k$ times, where $r_1, \ldots, r_n$ is a set of nonnegative integers. Of course, this formula can also be seen directly. Laguerre series are sometimes useful for counting words with certain subwords prohibited. Consider the following problem: given a word $W$ and a subword $W'$ of $W$, how many words are there avoiding $W'$ as a subword that can be made from $W$ using each symbol at most as many times as it appears in $W$? To answer this, we form the set $A_1$ of factorizations that can be made from $W$ that use only letters from $W'$ and do not contain $W'$ as a subword of any factor. Letting $A_2$ be the set of all factorizations of words that can be made from $W$ in the above sense that do not use any letters from $W'$, we see that $A_1 * A_2$ is the set of all factorizations of words that can be made from $W$ so that no part contains $W'$ as a subword. Setting $w=1$, we may perform a brute force calculation to compute the Laguerre series $f_{A_1}(t)$. Since there are no subword restrictions on the remaining letters, we have $$f_{A_2}(t) = \prod_c \sum_{i=1}^{n_c} \frac{t^i}{i!}$$ where the product is over the letters $c$ used in $W$ that are not used in $W'$, and $n_c$ is the number of times $c$ is used in $W$. So the solution is given by $\Phi(f_{A_1}(t)f_{A_2}(t))$. This method is practical if there are not too many letters used in the $W'$, and the number of times these letters are repeated in $W$ is small. For example, how many words are there that can be made from the letters in “CONSTANTINOPLE” that avoids the subword “TNT”? We focus on the letters $T$ and $N$, forming the set $A_1$ consisting of those factorization of words using only these letters, with three or fewer $N$’s and two or fewer $T$’s, so that no factor contains “TNT”. We then perform a brute force count using a programming language such as Sage to explicitly calculate the Laguerre series for $A_1$ with respect to $w=1$, getting $$f_{A_1}(t) = \frac{1}{12}t^5 + \frac{5}{12}t^4 + \frac{2}{3} t^3 + t^2 + t + 1.$$ Then forming $A_2$ as above, the set of factorizations of any words made from the remaining letters “COSAIOPLE”, we calculate $f_{A_2}(t) = (t + 1)^7(\frac{1}{2}t^2 + t +1)$ since all of the letters appear only once, except O which appears twice. Therefore, the answer is $$\int_0^\infty e^{-t} \left(\frac{1}{12}t^5 + \frac{5}{12}t^4 + \frac{2}{3} t^3 + t^2 + t + 1\right)\left(t +1\right)^7\left(\frac{1}{2}t^2 +t +1\right)\, dt = 9{,}854{,}474{,}467.$$ A more general method for the subword avoidance problem is the cluster method developed by Goulden and Jackson [@Goulden-Jackson1; @Goulden-Jackson2]. This a powerful tool using linear algebra and inclusion-exclusion to find rational generating functions for words avoiding any given set of subwords. Noonan and Zeilberger [@Doron1] give generalizations and a good introduction to the idea; Edlin and Zeilberger also give an extension to cyclic words [@Doron2]. Now consider the problem of counting words that have no subword consisting of $m$ identical letters. These are words that avoid the subword pattern $1^m$, and are sometimes called $m$-Carlitz words; when $m=2$ we have the ordinary Carlitz words. To find the generating function, let $A$ be the set of factorizations on a one-letter alphabet with each part having length smaller than $m$, and again let $w(\phi)= x^{\len(\phi)}$. We see that $$g_{A,w}(u) = \sum_{n = 0}^\infty u^n(x + \ldots + x^{m-1})^n = \frac{1 - x}{1 -x - u(x - x^m)}$$ and so we compute $$\begin{aligned} \label{lacefunforsmallruns} f_{A,w}(t) = T\{g_{A,w}(u)\} = \exp\left(t \cdot\frac{x - x^m}{1 - x^m}\right).\end{aligned}$$ Taking the coefficient of $x^n$ in gives the Laguerre series for the set of factorizations with length $n$ so that each part is smaller than $m$. This gives a generalization of Theorem \[fundamental\]. If $m_1, \ldots, m_k, n_1, \ldots, n_k$ are nonnegative integers, and $p_{m,n}(t)$ are polynomials defined by $\sum_{n=0}^\infty p_{m,n}(t) \,x^n = \exp{\left( \frac{t(x - x^m)}{1- x^m}\right)}$, we see that $$\Phi\left( \prod_{i =1}^k p_{m_i,n_i}(t)\right)$$ is the total number of $k$-ary words that use the letter $i$ exactly $n_i$ times and do not contain the subwords $i^{m_i}$. Thus the number of arrangements of the word “WALLAWALLA” with no LLL, AAA or WW as consecutive subwords is $$\begin{aligned} \int_0^\infty e^{-t} p_{3,4}(t)\cdot p_{3,4}(t) \cdot p_{2,2}(t)\, dt &= \int_0^\infty e^{-t}\left(\frac{1}{24}t^4 - t^2 + t\right)^2 \cdot \left(\frac{1}{2}t^2 - t\right) \, dt\\ &= 1584.\end{aligned}$$ Recalling again the formula $$\Phi(e^{tf}) = \frac{1}{1-f},$$\[phirule\]we see that the generating function for the number of $k$-ary $m$-Carlitz words of length $n$ is given by $$\Phi\left(\exp\left( kt \cdot\frac{ x - x^m}{1-x^m} \right)\right) = \frac{1 - x^m}{1 - kx - (k-1)x^m}.$$ Another derivation of this formula is given by Burstein and Mansour [@Mansour2 Example 2.2]. We may set indeterminates to count the number of occurrences of each symbol. Let $w(\phi) = x_1^{i_1} x_2^{i_2} \cdots x_k^{i_k}$, where $i_j$ is the number of appearances of $j$ in a factorization $\phi$ on the alphabet $[k]$. By and , we see that $$\begin{aligned} \displaystyle\Phi\left( \prod_{i=1}^k \exp\left(t \cdot\frac{x_i - x_i^m}{1-x_i^m}\right) \right) = \frac{1}{1 - \sum_{i=1}^k \frac{x_i - x_i^m}{1 - x_i^m}}\end{aligned}$$ gives the sum of all weights in $m$-Carlitz $k$-ary words. This can also be found by substituting $x_i + x_i^2 + \ldots +x_i^{m-1}$ for $x_i$ in Lemma \[carlitzgf\]. Vincular patterns ================= We are now ready to state a general formula for $k$-ary words avoiding vincular patterns with ones. We say that a $k$-ary factorization $\phi$ contains a vincular pattern $\tau$ with only ones if the word made from $\phi$ by inserting a single $0$ between each pair of adjacent factors contains $\tau$, and this copy of $\tau$ does not use $0$. Using the transformations $T$ and $\Phi$, we can reduce the problem to finding ordinary generating functions for factorizations that only use one symbol and avoid the given vincular pattern. [\[vincgf\]]{} Let $m_1, \ldots, m_n$ be positive integers, and let $A$ be the set of $k$-ary words avoiding the pattern $\tau = 1^{m_1} \d 1^{m_2} \d \cdots \d 1^{m_n}$. Then $$\sum_{a_1 \cdots a_l \in A} x_{a_1} x_{a_2} \cdots x_{a_l} = \Phi\left( \prod_{i=1}^k\left[ e^{tx_i} - T\left\{G_\tau(x_i, u)\right\}\right]\right)$$ where $T$ is the operator defined in Section 3, and $$\begin{aligned} \label{G} G_\tau(x,u) =\frac{ ux^{m_1}(1-x) }{(1-x - u(x-x^{m_i}))(1-x- ux)} \prod_{i=2}^n \left[ x^{m_i} + \frac{ux^{m_i}(1 - x^{m_i})}{1-x - u(x-x^{m_i})} \right].\end{aligned}$$ Earlier we observed that $e^{tx}$ is the Laguerre series for all factorizations using only a single letter with respect to the weight $w(\phi) = x^{\len(\phi)}$. So it is enough to show that $G_\tau(x,u) = g_{A,w}(u)$ where $A$ is the set of factorizations on some single letter alphabet containing the given vincular pattern; the difference will then count the factorizations on a single letter which avoid the pattern. For the remainder of the proof, we will use only the alphabet $\{1\}$. We will use repeatedly the following basic facts: Given two sets of factorizations $A_1$ and $A_2$, let $$A_3 = \{ (\phi_1)\ldots(\phi_l)(\psi_1)\ldots(\psi_m): \phi = (\phi_1)\ldots(\phi_l) \in A_1, \psi = (\psi_1)\ldots(\psi_m) \in A_2\}.$$ Then $g_{A_3,w}(u) = g_{A_1,w}(u) \cdot g_{A_2,w}(u)$ when both sides are defined. Let $A$ be a set of factorizations on the alphabet $\{1\}$. Then for a fixed integer $k{\geqslant}0$, let $A_k$ be the set consisting of each factorization of $\phi$ with $k$ ones appended to the last factor: $A_k = \{(\phi_1)(\phi_2) \cdots (\phi_n 1^k): (\phi_1) \cdots (\phi_n) \in A\}$. Then $g_{A_k,w}(u) = x^kg_{A,w}(u)$. We say that a factorization [minimally]{} contains the pattern $1^{m_1} \d 1^{m_2} \d \cdots \d 1^{m_n}$ if it contains the pattern, but no longer contains the pattern if the last letter of the last factor is removed. For example, $(11)(1)(1)(111)$ contains $111$ minimally, contains $11\d11$ but not minimally, and does not contain $1111$ at all. Define a [truncation]{} of a factorization $\phi = (\phi_1) \cdots (\phi_n)$ to be a factorization $\hat{\phi} = (\phi_1) \cdots (\phi_{i-1}) (\hat{\phi_i})$, where $1 {\leqslant}i {\leqslant}n$ and $\hat{\phi_i}$ is a nonempty initial substring of $\phi_i$. Any factorization containing a pattern has a unique truncation that contains that pattern minimally, so we will count the factorizations containing it minimally and then multiply by the appropriate expression to count all factorizations that contain the pattern. First, suppose the pattern is just $1^{m_1}$. A factorization minimally containing this pattern is any factorization whose last part has size $m_1$, and all other parts have size strictly smaller. So the generating function for these factorizations is $$\sum_{n=0}^\infty (u(x + \ldots + x^{m_1-1}))^nux^{m_1} = \left(\frac{ 1 - x}{1 - x - u(x - x^{m_1})}\right)ux^{m_1}.$$ Now let the generating function for the number of words minimally containing the pattern $1^{m_1} \d \cdots \d 1^{m_{n}}$ be $g_n(x,u)$; we will find $g_n(x,u)$ in terms of $g_{n-1}(x,u)$. Let $\phi$ be a factorization minimally containing $1^{m_1} \d\cdots\d 1^{m_{n-1}} \d 1^{m_n}$; then $\phi$ has a unique truncation $\hat{\phi}$ that minimally contains $1^{m_1} \d\cdots \d 1^{m_{n-1}}$. Thus to find $g_n(x,u)$ we multiply $g_{n-1}(x,u)$ by the appropriate factor, mimicking the following process by which any such $\phi$ is attained uniquely from some $\hat{\phi}$. If $\phi$ and $\hat{\phi}$ have the same number of factors, $\phi$ must be $\hat{\phi}$ with exactly $m_n$ $1$’s appended to the last factor; so we multiply by $x^{m_n}$. Otherwise, we can assume a new factor is necessary. In this case, we may lengthen the final part of $\phi$ by adding $1$’s, but to avoid having the pattern $1^{m_1} \d \cdots \d 1^{m_{n}}$ without adding any factors at most $m_n-1$ should be added. This means we should first multiply by $$(1 + x + \ldots x^{m_n-1}) = \frac{1-x^{m_n}}{1-x}$$. Then, we may append some (possibly empty) sequence of factors to $\hat{\phi}$, each having length less than $m_n$. This amounts to multiplying by $$\sum_{i=0}^\infty (u(x + \ldots + x^{m_n - 1}))^i = \frac{ 1 - x}{1 - x - u(x - x^{m_{n})}}.$$ Finally, we append a new factor which is just $1^{m_n}$, multiplying by $ux^{m_n}$. Thus we multiply by $$\frac{1-x^{m_n}}{1-x} \cdot \frac{ 1 - x}{1 - x - u(x - x^{m_{n})}}\cdot ux^{m_n}.$$ So the ordinary generating function minimally avoiding $1^{m_1} -\ldots -1^{m_{n}}$ is exactly $$g_n(x,u) = g_{n-1}(x,u) \cdot \left(x^{m_n} + \frac{ ux^{m_n}(1 - x^{m_n})}{1 - x - u(x - x^{m_n})}\right).$$ Finally, once we have a factorization minimally containing the pattern, we may add any number of ones to the last factor and append any list of factors and the resulting factorization will still contain the pattern. This amounts to multiplying by $$\sum_{j=0}^\infty\sum_{i=0}^\infty x^j \left(u( x + x^2 + \ldots)\right)^i = \frac{1}{1-x} \frac{1 -x}{1 - x - ux}.$$ Putting this all together, we see inductively that the generating function for factorizations containing the pattern is $$\begin{aligned} \frac{ ux^{m_1}(1-x) }{(1-x - u(x-x^{m_i}))} \prod_{i=2}^n \left[ x^{m_i} + \frac{1 - x^{m_i}}{1-x - u(x-x^{m_i})}\right] \frac{1}{1 - x - ux}.\end{aligned}$$ Cyclically avoiding patterns ============================ We say that a word $W$ cyclically avoids a vincular pattern $\tau$ if $W$ avoids $\tau$ no matter how its letters are cycled. More formally, let $r$ be the function that cycles $W$, moving the last letter into the first position: $r(a_1 \cdots a_n) = a_n a_1 \cdots a_{n-1} $. Then $W$ cyclically avoids $\tau$ if $r^k(W)$ avoids $\tau$ for each $k$. In order to find the generating function for the number of words cyclically avoiding the pattern $\tau = 1^m \d 1^m \d \cdots \d 1^m$, we will need a little more information than provided by the generating function $G_\tau(x,u)$ defined by . Let $H(x,u,v) = g_{A,w}(u)$ where $A$ is the set of factorizations on the alphabet $\{1\}$ [avoiding]{} the pattern $\tau$ and $\hat{w}$ is the weight $\hat{w}(\phi) = x^{\len(\phi)} u^{\parts(\phi)} v^{ \fst(\phi)}$ where $\fst(\phi)$ is the size of the first factor of $\phi$. Note that $\hat{w}$ is trivially a weight by our definition since we are using a singleton alphabet, but generally is not. We will find a closed-form expression for $H(x,u,v)$, although it is rather unwieldy. \[hlemma\] The generating function $H(x,u,v)$ is given by $$\begin{aligned} H(x,u,v) = 1 &+ \left[ \frac{1-x}{(1-vx)(1-x-ux)}\right] \biggl[{u(vx - (vx)^{mn})} +\biggl.\notag\\ & \left.\frac{{u^{2}x^{m} \left({\left(1-vx\right)} {\left(z - (vx)^{m}\right)} z^{n - 1} - {\left(1-(vx)^{m}\right)} {\left(z^n -(vx)^{m n}\right)}\right)} }{{\left(z-(vx)^{m}\right)} {\left({1-x - u\left(x - x^{m}\right)}\right)}}\right]\label{H}\end{aligned}$$ where $$z = x^m + \frac{ux^m(1 - x^m)}{1 - x - u(x - x^m)}.$$ Fix $n>0$, and let $\tau_n$ be the pattern $1^m \d 1^m \d \cdots \d 1^m$ with $n-1$ dashes. Then by the proof of Theorem \[vincgf\], we know that if $G_n(x,u)$ is the generating function for the number of factorizations on the alphabet $\{1\}$ containing $\tau_n$, then $$G_n(x,u) = \frac{ ux^{m}(1-x) }{(1-x - u(x-x^{m}))(1-x- ux)} \left( x^{m} + \frac{ux^{m}(1 - x^{m})}{1-x - u(x-x^{m})}\right)^{n-1}$$ since in this case $m_1 = m_2 = \ldots = m.$ We have different cases depending on how much of the pattern $\tau_n$ is contained in the first part of a factorization. Let $A$ be the set of all factorizations on the alphabet $\{1\}$ that contain $\tau_n$, and define define $A_0, A_1, \ldots, A_n$ by $$\begin{aligned} A_0 &= \{\phi = (\phi_1)\cdots(\phi_l) \in A : \phi_1\text{ avoids }\tau_1 = 1^m\}\\ A_i &= \{\phi = (\phi_1)\cdots(\phi_l)\in A: \phi_1 \text{ contains }\tau_i \text{ but }\phi_1 \text{ avoids } \tau_{i+1}\} \text{ for }1 {\leqslant}i {\leqslant}n-1\}\\ A_n &= \{\phi = (\phi_1)\cdots(\phi_l)\in A : \phi_1 \text{ contains }\tau_n \}.\end{aligned}$$ First, suppose that $\phi \in A_0$. Then the first part of $\phi$ is one of $1, 11, \ldots, 1^{m-1}$, and the factorization $(\phi_2)\cdots(\phi_k)$ of the remaining parts can be any factorization containing the pattern. With the exponent of $v$ recording the size of the first factor, we see that the generating function corresponding to this case is $$\begin{aligned} g_{A_0,w}(u) &= u(vx +(vx)^2 + \ldots + (vx)^{m-1}) G_n(x,u) \\ &= u\cdot \frac{vx - (vx)^m}{1 - vx} G_n(x,u).\end{aligned}$$ Now suppose that $\phi \in A_i$, $1 {\leqslant}i {\leqslant}n-1$. Then the first part of $\phi$ is one of $1^{im}, 1^{im+1}, \ldots, 1^{im + m-1}$ and $(\phi_2)\cdots(\phi_l)$ can be any factorization containing $\tau_{n-i}$. So the generating function corresponding to $A_i$ is $$\begin{aligned} g_{A_i,w}(u) &= u((vx)^{im} + (vx)^{im+1} + \ldots + (vx)^{im + m -1}) G_{\tau_{n-i}}(x,u)\\ &= u(vx)^{im}\frac{1 - (vx)^{m} }{1 - vx}G_{n-i}(x,u).\end{aligned}$$ If $\phi \in A_n$, then the first factor of $\phi$ can have any length that is $nm$ or more, and the remaining parts of $\phi$ can be any factorization. Therefore $$\begin{aligned} g_{A_n,w}(u) &= u((vx)^{nm} + (vx)^{nm+1} + \ldots)\sum_j(u(x +x^2 + \ldots))^j\\ &= \frac{ u(vx)^{nm}}{(1 - vx)(1 - \frac{ux}{1-x})}.\end{aligned}$$ To find the generating function for factorizations avoiding the pattern, we must subtract $g_{A_i,w}(u)$, $0 {\leqslant}i {\leqslant}n$, from the generating function for all one-letter factorizations, which is $$\begin{aligned} 1 + u(vx + (vx)^2 + \ldots)\sum_{k=0}^\infty(u(x + x^2 + \ldots))^k = 1 + \frac{uvx}{(1-vx)(1 - \frac{ux}{1-x})}.\\\end{aligned}$$ We have $$\begin{aligned} H(x,u,v) =&\, 1 + \frac{uvx}{(1-vx)(1 - \frac{ux}{1-x})} - \sum_{i=0}^n g_{A_i,w}(u)\\ =& \,1 + \frac{uvx}{(1-vx)(1 - \frac{ux}{1-x})} - u\cdot \frac{vx - (vx)^m}{1 - vx} G_{n}(x,u)- \frac{ u(vx)^{nm}}{(1 - vx)(1 - \frac{ux}{1-x})} \\ & - \sum_{i=1}^{n-1} u(vx)^{im}\frac{1 - (vx)^{i} }{1 - vx}G_{n-i}(x,u).\end{aligned}$$ After simplifying, we have the desired formula. \[cyclicvinc\] Let $H(x,u,v)$ be the generating function . Let $A$ be the set of words cyclically avoiding the pattern $\tau = 1^m \d 1^m \d\cdots \d 1^m$, with $n-1$ dashes, and let $w$ be the weight on $k$-ary words with $w(s_1 \ldots s_l) = x_{s_1} x_{s_2} \cdots x_{s_l}$. Then the generating function $\sum_{W\in A} w(W)$ is given by $$1 + \sum_{i=1}^k \Phi\biggl( t^{-1}\cdot T\left\{ u \frac{d^2}{dv\, du}\bigg\|_{v=1} H(x_i,u,v)\right\} \biggl(- 1 +\prod^k_{\substack{j=1\\ j\neq i}} T\{H(x_j,u,1)\}\biggr)\biggr) + \sum_{i=1}^k \frac{x_i - x_i^{mn}}{1-x}.$$ In particular, letting $x_i = x$ for each $i$ gives: $$\sum_{W \in A} x^{\len(W)} = 1 + k\cdot \Phi\biggl( t^{-1}\cdot T\left\{ u \frac{d^2}{dv\, du}\bigg\|_{v=1} H(x,u,v)\} \big(T\{H(x,u,1)\}\right)^{k-1}-1\big)\biggr) + \frac{k(x - x^{mn})}{1-x}.$$ We will count all words which avoid $\tau$ and begin with $i$, where $1 {\leqslant}i {\leqslant}k$, and then sum over $i$. Let $\mathcal{W}$ be the set of all $k$-ary words $W$ that begin with $i$, do not end in $i$, and avoid the pattern $\tau$. Any $W \in \mathcal{W}$ will in fact [cyclically]{} avoid $\tau$. To see this, it is enough to note that the restriction $r^p(W)\|_{\{j\}}$, for any $j,p$, will avoid the pattern. Such a factorization will either be a permutation of the parts of $(W)\|_{\{j\}}$, or will have some part of $(W)\|_{\{j\}}$ divided into the first and last part of $r^p(W)\|_{\{j\}}$. Permuting the factors will not cause $(W)\|_{\{j\}}$ to contain the pattern by the symmetry of $\tau$, and splitting a part into two cannot create an instance of the pattern either. We can characterize $\mathcal{W}$ as follows. Suppose $\phi = (\phi_1)\ldots(\phi_p)$ is a factorization over the alphabet $[k]\backslash\{i\}$. Then we can insert $i$’s appropriately in between factors so that the result starts with $i$ and avoids the pattern. Given a factorization $\psi = (\psi_1)\ldots(\psi_p)$ which avoids the pattern and has the same number of parts as $\phi$ but only uses the symbol $i$, let $W_{\phi,\psi} = \psi_1 \phi_1 \psi_2 \phi_2\ldots \psi_p \phi_p$ be the concatenation of the alternating factors of $\phi$ and $\psi$. Then $W_{\phi,\psi}$ start with $i$, will not end in $i$, and avoids $\tau$, so $W_{\phi,\psi} \in \mathcal{W}$. Furthermore, if $W \in \mathcal{W}$, let $\phi$ be the restriction $W\|_{[k]\backslash \{i\}}$, and $\psi = W\|_{\{i\}}$. Then $\phi$ and $\psi$ will have the same number of factors, and $W = W_{\phi,\psi}$. So we see that $\mathcal{W}$ coincides with the set of $W_{\phi,\psi}$ where $\phi,\psi$ are nonempty factorizations over the alphabets $[k]\backslash \{i\}$ and $\{i\}$, respectively, which avoid $\tau$ and have the same number of parts. The words $W \in \mathcal{W}$ we have described always end in a letter besides $i$; to get all words avoiding the pattern starting with $i$, we associate $W\in \mathcal{W}$ with a class $$\{W, r^{-1}W, r^{-2}W, \ldots,r^{-B(W)+1}W\},$$ where $B(W)$ is the number of adjacent $i$’s at the beginning of $W$. These rotated words will also cyclically avoid the pattern, as we have explained, and begin with at least one $i$. Furthermore, any word $V$ beginning with $i$ and avoiding the pattern will be in the class of a unique word $W$: let $W = r^CV$ where $C$ is the number of $i$’s on the end of $V$, if any. Therefore we will count all words $W$ that avoid the pattern and begin with $i$, but do not end with $i$, with the multiplicity $B(W)$, the number of $i$’s at the beginning of the word $W$. This will account for all words beginning with $i$ that use at least one letter $j \neq i$. So the weight of all $k$-ary words that begin with $i$, cyclically avoid the pattern $\tau$, and contain at least one letter $j \neq i$ is $$\sum_{W \in \mathcal{W}} B(W)w(W) = \sum_{\psi,\phi} \fst(\phi) w(\phi)w(\psi)$$ where the sum on the right-hand side is taken over all factorizations $\phi,\psi$ over the alphabets $[k]\backslash \{i\}$ and $\{i\}$, respectively, which avoid $\tau$ and have the same number of parts. For any $1 {\leqslant}j {\leqslant}k$ let $A_j$ be the set of nonempty factorizations on the alphabet $\{j\}$ that avoid the pattern. Define $$b_p = \sum_{{\substack{\phi \in A_1 * \cdots \hat{A_i} \cdots * A_k\\\parts(\phi) = p}}} w(\phi),$$ where the $\hat{A_i}$ indicates that $A_i$ is removed, and let $$a_p = \sum_{\phi \in A_i, \parts(\phi) = p} \fst(\phi)x_i^{\len(\phi)}.$$ Then by the previous discussion, we see that $$\sum_{W \in \mathcal{W}} B(W) w(W) = \sum_{p= 1}^\infty a_p b_p$$ is the weight of all $k$-ary words avoiding the pattern $\tau$ which begin with $i$ and use at least one letter $j \neq i$. It remains to evaluate the right-hand side. We make use of the well-known fact that the generalized Laguerre polynomials obey the orthogonality relation $$\begin{aligned} \label{lagortho} \int_0^\infty x^\alpha e^{-x} L_i^{(\alpha)}(x) L_j^{(\alpha)}(x) \, dx = \frac{\Gamma(i + \alpha + 1)}{i!} \delta_{i,j}. \end{aligned}$$ which can be found in, e.g., [@ortho p. 241]. Although this only holds in general when $\alpha > - 1$, it is true for $\alpha = -1$ if $i {\geqslant}1$. We have $$\begin{aligned} \label{mylagortho} \Phi(it^{-1} l_i(t) l_j(t)) = \delta_{i,j}\end{aligned}$$ for $i{\geqslant}1$. Thus for a function $f(t)$ that is a formal sum of the Laguerre polynomials $l_p(t)$, we can extract the coefficient of $l_p(t)$ by evaluating $\Phi( pt^{-1} l_p(t) f(t))$, for $p{\geqslant}1$. Furthermore, we see that if $f(t) = \sum_{p{\geqslant}1} pa_p l_p(t)$ and $g(t) = \sum_{p{\geqslant}1} b_p l_p(t)$, we have $$\begin{aligned} \label{ab} \Phi(t^{-1} f(t)g(t)) = \sum_{p{\geqslant}1} a_p b_p\end{aligned}$$ when both sides are well-defined. We require $p{\geqslant}1$ since the orthogonality relation fails when $i = 0$. In this case, $\sum_{p {\geqslant}1} b_p l_p(t)$ is the Laguerre series for [nonempty]{} $k$-ary words avoiding the pattern that do not use $i$; it is given by $\left(\prod_{j \neq i} T\{H(x_j,u,1)\}\right) - 1$. To find $\sum_{p {\geqslant}1} pa_p l_p(t)$, recall from the Lemma \[hlemma\] that $H(x,u,v) = \sum_{\phi \in A_i} x^{\len(\phi)} u^{\parts(\phi)} v^ {\fst(\phi)}$. So $$u \frac{d^2}{dv\, du}\bigg\|_{v=1} g(x_i,u,v) = \sum_{\phi \in A_i, \phi \neq \emptyset} \fst(\phi)\cdot \parts(\phi) \cdot x_i^{\len(\phi)} u^{\parts(\phi)}.$$ Then applying $T$, we see that $$T\left\{u \frac{d^2}{dv\, du}\bigg\|_{v=1}H(x_i,u,v)\right\} = \sum_{p {\geqslant}1} p a_p l_p(t).$$ Thus by , $$\begin{aligned} \label{beginningwithi} \Phi\left(t^{-1}\cdot T\left\{ u\frac{d^2}{dv\, du}\bigg\|_{v=1} H(x_i,u,v)\right\}\cdot \prod_{j \neq i} \bigl(T\{G(x_j,u)\}\bigr)\right)\end{aligned}$$ is the generating function for the number of $k$-ary words that begin with $i$, use at least one $j\neq i$, and cyclically avoid the pattern. This does not account for the words that use only the letter $i$ and avoid the pattern, which are the words $i^k$, with $k< mn$. So we must add $x_i + x_i^2 + \ldots + x_i^{mn-1} = \frac{x_i - x_i^{mn}}{1 - x_i}$ to to get the full generating function for all words that begin with $i$ and cyclically avoid $\tau$. Summing over $i$ and adding $1$ to represent the empty factorization gives $$1 + \sum_{i=1}^k \Phi\biggl( t^{-1}\cdot T\left\{ u \frac{d^2}{dv\, du}\bigg\|_{v=1} H(x_i,u,v)\right\} \biggl(- 1 +\prod^k_{\substack{j=1\\ j\neq i}} T\{H(x_j,u,1)\}\biggr)\biggr) + \sum_{i=1}^k \frac{x_i - x_i^{mn}}{1-x}$$ as desired. If we set $n=1$, considering $k$-ary words that cyclically avoid $1^m$, the formula simplifies considerably. After some computation, which we omit here, we arrive at the following formula. Let $A$ be the set of nonempty $k$-ary words avoiding $1^m$. Let $w$ be the weight $w(a_1 \cdots a_l) = x_{a_1} x_{a_2} \cdots x_{a_l}$. Then $$\sum_{W \in A} w(W) = \sum_{i = 1}^k \frac{ x_i^{2m} - m x_i^{m + 1} + (m - 1)x_i^{m}}{ (x_i^{m} - 1)(x_i -1) } + \frac{\sum_{i=1}^k \frac{ (m - 1)x_i^{m+1} - m x_i^{m} + x_i}{(x_i^{m} - 1)^2} }{1 - \sum_{i =1}^k \frac{x_i^{m} - {x_i}}{ x_i^{m} - 1}}.$$ In the book by Heubach and Mansour [@Mansour1], the authors define a cyclic Carlitz composition as a Carlitz composition so that the first and last parts are not equal; they ask (Research Direction 3.3) for the generating function for the number of cyclic Carlitz compositions. If we let $k$ approach infinity, $m=2$, and $x_i = x^i$, we get the following. Let $A$ be the set of cyclic Carlitz compositions. Then $$\sum_{W \in A} x^{\operatorname{sum}(W)} = \frac{\sum_{i=1}^\infty \frac{x^i}{(1 + x^i)^2}}{1 - \sum_{i=1}^\infty \frac{x^i}{1 + x^i}} + \sum_{i=1}^\infty \frac{x^{2i}}{1 + x^i}.$$ Setting $x_i = x$ in Corollary 5.3 and simplifying gives the following formula. Let $A$ be the set of nonempty $k$-ary words that cyclically avoid $1^m$. Then $$\sum_{W \in A} x^{\len(W)} = \frac{1 - x^{m-1}}{ 1 - x} \left(kx + (k-1)x\left(\frac{m - (m-1)kx}{1 - kx + (k-1)x^m} - \frac{m}{1 - x^{m}}\right)\right).$$ This was found by Burstein and Wilf [@Wilf]. They go on to show that the number of $k$-ary words of length $n$ cyclically avoiding $1^m$ is asymptotically $\beta^n$, where $\beta$ is the positive root of $x^{m+1} = (k - 1)(1 + x + x^2 + \ldots + x^m)$. In fact, they extract an explicit formula when $n$ is sufficiently large. We can also give a cyclic version of Theorem \[fundamental\], which can be derived by extracting the coefficient of a monomial $x_1^{n_1}\cdots x_k^{n_k}$ in the generating function from Theorem 5.2 for words cyclically avoiding $1^m$. Let $n_1, \ldots, n_k$ be positive integers, and let $p_{m ,n}(t)$ be defined as before by $\sum_{n=0}^\infty p_{m,n}(t) \,x^n = \exp{\left( \frac{t(x - x^m)}{1- x^m}\right)}$. Then $$N\cdot \Phi\left(t^{-1}\cdot \prod_{i =1}^k p_{m,n_i}(t)\right)$$ is the total number of $k$-ary words that use the letter $i$ exactly $n_i$ times and cyclically avoid $1^m$, where $N = \sum_{i=1}^k n_i$ is the total number of letters counted with multiplicity. Questions and future directions =============================== There are a number of related questions that remain open. Is it possible to extend the work of Section 5, finding generating functions for words avoiding other cyclic patterns? Is there a formula generalizing Theorem \[vincgf\] or Theorem \[cyclicvinc\] to to count the number of occurrences of a given pattern of ones? One might also look for a combinatorial interpretation of some form of [composition]{} of Laguerre series; empirically, it seems that $l_i(l_j(t))$ has nonnegative integer coefficients in the $l_k$-basis for $j>0$. Finally, it would be useful to develop bijections from sets of words with restrictions to other combinatorial objects that are not obviously described in terms of words, using the methods outlined here to count sets that may be otherwise difficult to enumerate. [10]{} Eric Babson and Einar Steingr[í]{}msson. Generalized permutation patterns and a classification of the [M]{}ahonian statistics. , 44: Art. B44b, 2000. Antonio Bernini, Luca Ferrari, and Renzo Pinzani. Enumeration of some classes of words avoiding two generalized patterns of length three. , 14(2), 2009. Gunnar Blom, Jan-Eric Englund, and Dennis Sandell. The [M]{}ississippi problem. , 52(1), 1998. Mireille Bousquet-M[é]{}lou, Anders Claesson, Mark Dukes, and Sergey Kitaev. (2+2)-free posets, ascent sequences and pattern avoiding permutations. , 117(7), 2010. Alexander Burstein. . PhD thesis, University of Pennsylvania, 1998. Alexander Burstein and Toufik Mansour. Counting occurrences of some subword patterns. , 2003. Alexander Burstein and Toufik Mansour. Words restricted by 3-letter generalized multipermutation patterns. , 7(1), 2003. Alexander Burstein and Herbert S. Wilf. On cyclic strings without long constant blocks. , 35(3), 1997. L. Carlitz. Restricted compositions. , 14(3), 1976. L. Carlitz, Richard Scoville, and Theresa Vaughan. Enumeration of pairs of sequences by rises, falls and levels. , 19, 1976. Anne E. Edlin and Doron Zeilberger. The [G]{}oulden-[J]{}ackson cluster method for cyclic words. , 25(2), 2000. S. Even and J. Gillis. Derangements and [L]{}aguerre polynomials. , 79(01), 1976. Dominique Foata and Doron Zeilberger:i. Laguerre polynomials, weighted derangements, and positivity. , 1988. Ira M. Gessel. Generalized rook polynomials and orthogonal polynomials. In D. Stanton, editor, [q-Series and Partitions]{}, pages 159–176. Springer-Verlag, New York, 1989. I. Goulden and D.M. Jackson. An inversion theorem for cluster decompositions of sequences with distinguished subsequences. , 2(20), 1979. I. Goulden and D.M. Jackson. . John Wiley, New York, 1983. Silvia Heubach and Toufik Mansour. . CRC Press, 2009. Arnold Knopfmacher and Helmut Prodinger. On [C]{}arlitz compositions. , 19, 1998. Percy A. MacMahon. . Cambridge \[Eng.\]: The University Press, 1915-1916. W. Magnus and F. Oberhettinger. . Springer, New York, 1966. Toufik Mansour. Restricted 132-avoiding k-ary words, [C]{}hebyshev polynomials, and continued fractions. , 36(2), 2006. John Noonan and Doron Zeilberger. The [G]{}oulden-[J]{}ackson cluster method: extensions, applications and implementations. , 5(4-5), 1999. Harry Pollard. Representation of an analytic function by a [L]{}aguerre series. , 48, 1948. N. V. Smirnov, O. V. Sarmanov, and V.K. Zaharov. A local limit for transition numbers in a [M]{}arkov chain, and its applications. , 7, 1966. Otto Szász and Nelson Yeardley. The representation of an analytic function by general [L]{}aguerre series. , 8, 1948. Ernst Joachim Weniger. On the analyticity of [L]{}aguerre series. , 41(42), 2008. Jiang Zeng. Weighted derangements and the linearization coefficients of orthogonal [S]{}heffer polynomials. , 3, 1992.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We present the results of a Chandra 30 ks observation of the low mass X-ray binary and atoll source 705. Here we concentrate on the study of discrete features in the energy spectrum at energies below $\sim 3$ keV, as well as on the iron K$\alpha$ line, using the HETG spectrometer on board of the Chandra satellite. Below 3 keV, three narrow emission lines are found at $1.47$, $2.0$, and $2.6$ keV. The $1.47$ and $2.6$ keV are probably identified with Ly-$\alpha$ emission from Mg XII and S XVI, respectively. The identification of the feature at $\sim 2.0$ keV is uncertain due to the presence of an instrumental feature at the same energy. The iron K$\alpha$ line at $\sim 6.5$ keV is found to be intrinsically broad ($FWHM \sim 1.2$ keV); its width can be explained by reflection from a cold accretion disk extending down to $\sim 15$ km from the neutron star center or by Compton broadening in the external parts of a hot ($\sim 2$ keV) Comptonizing corona. We finally report here precise X-ray coordinates of the source.' author: - 'T. Di Salvo, R. Iaria, M. Méndez, L. Burderi, G. Lavagetto, N. R. Robba, L. Stella, M. van der Klis' title: 'A broad iron line in the Chandra/HETG spectrum of 705' --- Introduction ============ Broad emission lines (FWHM up to $\sim 1$ keV) at energies in the range 6.4 – 6.7 keV are often observed in the spectra of Low Mass X-ray Binaries (hereafter LMXBs), both in systems containing black hole candidates (see e.g. Miller et al. 2002) and in systems hosting an old accreting neutron star (see e.g. Barret 2001 for a review). These lines are identified with the K$\alpha$ radiative transitions of iron at different ionization stages. Sometimes an iron absorption edge at energies $\sim 7-8$ keV has been detected. These features are powerful tools to investigate the structure of the accretion flow close to the central source; in particular, important information can be obtained from detailed spectroscopy of the iron K$\alpha$ emission line and absorption edge, since these are determined by the ionization stage, geometry, and velocity field of the reprocessing plasma. To explain the large width of these lines it has been proposed that they originate from emission reprocessed by the accretion disc surface illuminated by the primary Comptonized spectrum (Fabian et al. 1989). In this model, the combination of relativistic Doppler effects arising from the high orbital velocities and gravitational effects due to the strong field in the vicinity of the neutron star smears the reflected features. Therefore the line will have a characteristically broad profile, the detailed shape of which depends on the inclination and on how deep the accretion disk extends into the neutron star potential (e.g. Fabian et al. 1989; Stella 1990). An alternative location of the line emitting region is the inner part of the so-called Accretion Disk Corona (ADC), probably formed by evaporation of the outer layers of the disk illuminated by the emission of the central object (e.g. White & Holt 1982). In this case the width of the line is explained by thermal Comptonization of the line photons in the ADC. This produces a genuinely broad Gaussian distribution of line photons, with $\sigma \geq E_{\it Fe} (k T_e/m_e c^2)^{1/2}$, where $E_{\it Fe}$ is the centroid energy of the iron line and $k T_e$ is the electron temperature in the ADC (see Kallman & White 1989; Brandt & Matt 1994 for more detailed calculations). This mechanism can explain the width of the iron line for temperatures of the emitting region of few keV. The presence of several unresolved components, which can eventually be resolved by the high resolution X-ray instruments on board Chandra and XMM-Newton, can also contribute to broaden the line. 705 is a very interesting source of the atoll class (see Hasinger & van der Klis 1989) which also shows type-I X-ray bursts. As in other similar sources, the energy spectrum of 705 can be described as the sum of a (dominating) Comptonized component, a blackbody, and an emission line at $\sim 6.4$ keV. We selected this source for a Chandra observation because a broad (1.1 keV FWHM) iron emission line at 6.5 keV has been previously reported (White et al. 1986; Barret & Olive 2002). One of the goals of our Chandra observation was to study the iron line profile to discriminate among the various models that have been proposed to explain the large line width. The Chandra/HETGS observation demonstrates that the iron line is intrinsically broad (1.2 keV FWHM); possible broadening mechanism are discussed. Observations ============ 705 was observed using the High-Energy Transmission Grating Spectrometer (HETGS) on board of Chandra starting on 2001 July 1. Part of the observation, for a total integration time of 24.4 ks, was performed in Timed Graded mode, while a short part of the observation, $\sim 5$ ks, was performed in Continuous Clocking mode. The HETGS consists of two types of transmission gratings, the Medium-Energy Grating (MEG) and the High-Energy Grating (HEG). The HETGS affords high-resolution spectroscopy from 1.2 to 31 Å (0.4-10 keV) with a peak spectral resolution of $\lambda / \Delta\lambda \sim 1000$ at 12 Å for HEG first order. The dispersed spectra were recorded with an array of six charged coupled devices (CCDs) that are part of the Advanced CCD Imaging Spectrometer (ACIS-S; Garmire et al. 2003; see http://asc.harvard.edu/cdo/about$\_$chandra for more details). The current relative accuracy of the overall wavelength calibration is of order of 0.05%, leading to a worst-case uncertainty of 0.004 Å in the first-order MEG, and 0.006 Å in the first-order HEG. We processed the event lists using available software (FTOOLS and the CIAO v.3.1 packages). We computed aspect-corrected exposure maps for each spectrum, allowing us to correct for effects from the effective area of the CCD spectrometer. For the part of the observation performed in Timed Graded mode, the brightness of the source required additional efforts to mitigate “photon pile-up" effects. We applied a 400-raw “subarray" (with the first raw = 1) during the observation that reduced the CCD frame time to 1.4 s. The zeroth-order image is affected by heavy pile-up; the event rate is so high that two or more events are detected in the CCD during the 1.4-s frame exposure. Pile-up distorts the count spectrum because detected events overlap and their deposited charges are collected into single, apparently more energetic, events. Moreover, many events ($\sim 90\%$) are lost as the grades of the piled-up events overlap those of highly energetic background particles, and are thus rejected by the on-board software. We therefore will ignore the zeroth-order events in subsequent analysis. On the other hand, the grating spectra are not, or only moderately (less than 10%), affected by pile-up. In our analysis, we utilize the HEG first-order spectrum, which is less affected by pile-up (less than 6% at maximum, less that 3% below 3 keV and above 6 keV) with respect to the MEG first-order spectrum. To determine the zero-th point position in the image as precisely as possible we calculated the mean crossing point of the zero-th order readout trace and the tracks of dispersed HEG and MEG arms. This results in the following source coordinates: RA = 17$^{\rm h}$ 08$^{\rm m}$ 54$^{\rm s}$.47, DEC = $-44^\circ$ 06$'$ 07$''$.35 (J2000, uncertainty 0.5$''$). Note that this position is significantly different ($\sim 0.15$ arcmin) from the coordinates previously reported for this source (see Liu, van Paradijs, & van den Heuvel 2001, and references therein). The data collected in Continuous Clocking mode do not suffer from photon pile-up in the first order dispersed spectra. However, due to the short exposure time of this observation, the statistics are quite low and we only used these data to check the results obtained with the Timed Graded observation. Indeed the spectra acquired in Continuous Clocking mode are in good agreement with the HEG spectra in Timed Graded (both in the shape of the continuum emission and in the parameters of the iron line at 6.5 keV, see below). We therefore conclude that the effects of pile-up in the HEG spectra of 705 acquired in Timed Graded mode are negligible. Spectral Analysis ================= We selected the first-order spectra from the HEG. Data were extracted from regions around the grating arms; to avoid overlapping between HEG and MEG data we used a region size of 26 pixels for the HEG along the cross-dispersion direction. The background spectra were computed, as usual, by extracting data above and below the dispersed flux. The contribution from the background is $\sim 0.3\%$ of the total count rate. We used the standard CIAO tools to create detector response files (see Davis 2001) for the HEG $+1$ and $-1$ order (background-subtracted) spectra, which we fit simultaneously using the XSPEC v.11.2 data analysis package (Arnaud 1996). The HEG spectra were rebinned to $0.005 \AA$. Two large energy gaps at those wavelength corresponding to the junction between two CCDs are present in the HEG first order spectra, at 3–3.4 keV and 5.5–6.5 keV for the +1 and –1 order, respectively. In these intervals the effective area is much smaller and is known with less accuracy (see Proposers’ Observatory Guide, POG, at http://cxc.harvard.edu/helpdesk). Unfortunately in the interval between 6 and 6.6 keV, the HEG +1 and $-1$ order spectra give slightly different residuals with respect to the same continuum model. In the present work we therefore exclude the interval mentioned above corresponding to the junction between two CCDs and in which the two orders give different results. The energy ranges used in the following spectral analysis are: 1.3–10 keV for the $-1$ order, and 1.3–6 keV and 6.6–10 keV for the +1 order, respectively. We fit the HEG first order spectra of 705 to a continuum model. The best fit model consists of the Comptonization model [comptt]{} (Titarchuk 1994), modified by absorption from neutral matter, parametrized by the equivalent hydrogen column $N_H$, which gives a $\chi^2_{\rm red} (d.o.f.)$ of $1.05 (3265)$. This model also includes an overabundance of Si by a factor $\sim 2$ with respect to Solar abundances to fit a highly significant absorption edge at $\sim 1.84$ keV (the addition of this parameter reduces the $\chi^2$ by $\Delta \chi^2 \simeq 61$ at the expense of 1 degree of freedom). Note that we cannot exclude that this feature may be due to the presence of Si in the CCDs, and therefore this overabundance is not discussed further. Finally, in all the fits we include an instrumental feature at $2.06$ keV (usually present in the HETG spectra of bright sources, see Miller et al. 2002) which is fitted by an inverse edge (with $\tau \sim -0.1$). A soft blackbody component has often been reported in the X-ray spectra of this kind of sources and has been detected in the spectra of 705 as well. In 705, in spectral states similar to the one found during our Chandra observation, this component has a temperature of $\sim 1.9$ keV and can contribute up to 20% of the total source flux (Barret & Olive 2002). We therefore added a blackbody component to the [comptt]{} continuum model; this component has a temperature of $\sim 1.95$ keV and contributes up to 50% of the total X-ray flux. However, the decrease of the $\chi^2$ for the addition of this component ($\Delta \chi^2 = 5$ for the addition of two parameters) is not significant, and, for sake of simplicity, we preferred not to include this component in our continuum model. Residuals in units of $\sigma$ with respect to the continuum model described above are shown in Figure \[res\]; several discrete features are still clearly visible in the residuals with respect to this continuum model, at $\sim 1.5$, $2.0$, $2.6$ keV, and, particularly, in the 6–7 keV range, where the K$\alpha$ iron emission line is expected. From these residuals it is apparent that the iron line is intrinsically broad and shows a complex profile. The addition of a broad ($\sigma \sim 0.5$ keV) Gaussian line centered at 6.5 keV proves necessary, giving $\Delta \chi^2 = 167$ for the addition of three parameters. We also added three narrow emission lines to fit the other low energy residuals mentioned above. The addition of Gaussian emission lines at $\sim 1.5$, $2.0$, $2.6$ keV gives a reduction of the $\chi^2$ by 27, 29, and 32 units, respectively, for the addition of three parameters. The errors in the normalizations of these features give a detection at about $3 \sigma$ confidence level. This is not a highly significant detection, and needs a confirmation with future observations. However, the fact that the energies of these features are close to the energies of Ly$\alpha$ transitions of H-like ionization stages of the most abundant ions emitting in the observed range (that are Mg XII, Si XIV, and S XVI, respectively) adds further confidence that these lines may be real. Data and residuals in units of $\sigma$ with respect to this best-fit model are shown in Figure \[spect\]. The best fit model is reported in Table 1 together with the discrete features mentioned above as well as the identification of the line, when possible. The spectral analysis of the first order HEG spectra of 705 from the part ($\sim 5$ ks) of the Chandra observation performed in Continuous Clocking mode confirms the best fit model found for the Timed Graded spectra. In particular, the best fit continuum model is again given by the [comptt]{} model, which gives a $\chi^2/dof = 867/1097$. The addition of a broad Gaussian emission line at $6.6 \pm 0.1$ keV ($\sigma = 0.4 \pm 0.1$ keV, EW = 138 eV) improves the fit giving a $\chi^2/dof = 831/1094$. The residuals in the K-shell iron line range are quite complex (see Fig. \[res\]), showing a broad feature centered at $\sim 6.5$ keV. We therefore try to fit this feature with the line profile expected from a thin Keplerian accretion disk. Substituting the Gaussian line with the [diskline]{} model (Fabian et al. 1989), we obtain an equivalently good fit, $\chi^2(d.o.f.) = 3164/3252$ using the [diskline]{} model. The line best fit parameters for the [diskline]{} model are given in Table 2. Discussion ========== We have analysed a Chandra 30 ks observation of the X-ray burster and atoll source 705. The position of the zero-th order image of the source provides improved X-ray coordinates for 705 (RA = 17$^{\rm h}$ 08$^{\rm m}$ 54$^{\rm s}$.47, DEC = -44$^\circ$ 06$'$ 07$''$.35), which significantly differ (by about 0.15 arcmin) from the coordinates previously reported for this source. We performed a spectral analysis of the HEG first order spectra of 705. The continuum emission is well fitted by the Comptonization model [comptt]{}, with an equivalent hydrogen column of $\sim 1.4 \times 10^{22}$ cm$^{-2}$, and an overabundance of Si by a factor $\sim 2$ with respect to Solar abundance (which might be of instrumental origin). The inferred unabsorbed flux of the source in the 0.1–10 keV range is $\sim 1.0 \times 10^{-8}$ ergs cm$^{-2}$ s$^{-1}$, corresponding to a luminosity of $3.3 \times 10^{37}$ ergs s$^{-1}$ assuming a distance to the source of 7.4 kpc (Haberl & Titarchuk 1995). The Comptonization continuum is quite soft, with an electron temperature of $k T_e \sim 2.3$ keV and an optical depth of $\tau \sim 18$ for a spherical geometry. We have detected a broad emission feature at $6.4-6.5$ keV, which we interpret as K-shell fluorescent emission of lowly ionized iron. The high energy resolution of the HEG shows that the line is intrinsically broad ($FWHM \sim 1.2$ keV), in agreement with previous measures (see e.g.Barret & Olive 2002). Therefore the most probable origin of this line is an accretion disk (in this case the large width of the line would be due to Doppler and relativistic smearing effects) or a hot corona (in this case the large width of the line would be due to Compton broadening). In the case the line is produced by reflection in an accretion disk we estimate that the required inner radius of the disk is $\sim 7\; R_g$ or $\sim 15$ km for a $1.4\; M_\odot$ neutron star. Note that the quite small inner radius of the disk inferred from this model is in agreement with the quite soft X-ray spectrum of 705 during the Chandra observation, which would probably place the source in the banana state of its X-ray color-color diagram. In this model, the inclination of the disk with respect to the line of sight is constrained in the range $55^\circ - 84^\circ$. Alternatively, Comptonization in the corona could explain the large width of the line. Detailed calculations give $\sigma_{\rm Fe} = 0.019 E_{\it Fe} \tau_T (1 + 0.78 kT_e)$, where $\tau_T$ is the Thompson optical depth and $kT_e$ is in keV (Kallman & White 1989, see also Brandt & Matt 1994). Assuming an average electron temperature of $k T_e = 2.3$ keV, as derived from the fit of our data to the Comptonization model, we can explain the width of the iron line for a Thomson optical depth of $\tau_T \sim 1.4$. Therefore, it is possible that the line is produced in the outer region of the Comptonizing corona, where the optical depth might be lower (assuming that the temperature remains constant). This is not unreasonable given that any contribution to the line produced inside the Comptonizing region, where the optical depth can be as high as 10 (see Tab. 1), would be completely smeared by Comptonization. Therefore, as expected, we only see that part of the line that is produced in the outer Comptonizing region. Unfortunately with these data we are not yet able to discriminate between the two possible origins of the iron line and its broadening proposed above. Higher statistics or detection of line variability would be needed for this. Naturally, what would definitively discriminate between the two proposed models is the detection of a double peak in the iron line profile; this would exclude the Comptonization of line photons model and would indicate the relativistic/Doppler effects as the origin of the line broadening. This detection would be possible with higher statistics (large effective areas or long exposures). Hence, more XMM-Newton and Chandra observations (better if simultaneous) would be ideal to this aim. Snapshots observations at different intensity levels and/or spectral states of the source would also be useful to study the variability of the line parameters with the position of the source in the X-ray color-color diagram and/or the frequency of the so-called kiloHertz quasi-periodic oscillations (kHz QPOs). If the source position in the color-color diagram track is indeed determined by the mass accretion rate (see e.g. Hasinger & van der Klis 1989) and the frequencies of the kHz QPOs are related to the Keplerian frequency at the inner edge of the accretion disk (as envisaged by most of the current models, see e.g. Miller, Lamb, & Psaltis 1998; Stella & Vietri 1999), then one would expect that the radius of the disk as determined from the line profile (fitted with a [diskline]{} model) should change accordingly. These kind of studies could give other pieces of evidence in favour of one or the other model. To fit discrete residuals at low energies with respect to the continuum model we added to the model several Gaussian lines. In Table 1 we report the most significant of these features, together with a possible identification of the line. The lines at $1.5$ and $2.6$ keV are identified with Ly$\alpha$ transitions from H-like Mg at 1.4726 keV (or, less probably, L-shell transitions from highly ionized iron, Fe XXII – Fe XXIV) and S at 2.6227 keV, respectively. The identification of the line at $\sim 2.0$ keV is more uncertain due to the presence of a systematic feature at $\sim 2.07$ keV. Note, however, that this line is quite close to the Ly$\alpha$ transition of Si XIV at 2.0061 keV. Finally, we note that there seems to be a correlation between the energy and the width of these lines, with the Gaussian $\sigma$ increasing with the centroid energy of the line. If the line width is due to velocity dispersions, we have calculated that $\Delta E/ E_0 = v/c \sim 2.7 - 3.6 \%$ (where we have used for $\Delta E$ the FWHM), slightly increasing with the energy of the line, from Mg XII to S XVI, as expected if more ionized elements are produced closer to the central X-ray source. This work was partially supported by the Ministero della Istruzione, della Universitá e della Ricerca (MIUR). Arnaud, K.A., 1996 in Astronomical Data Analysis Software and Systems V, A.S.P. Conference Series, Vol. 101, G.H. Jacoby and J. Barnes, eds., p. 17 Barret, D. 2001, Advances in Space Research, 28, 307 Barret, D., & Olive, J.F. 2002, ApJ, 576, 391 Brandt, W.M., & Matt, G. 1994, MNRAS, 268, 1051 Davis, J.E. 2001, ApJ, 562, 575 Fabian, A.C., et al. 1989, MNRAS, 238, 729 Garmire, G. P., Bautz, M. W., Ford, P. G., Nousek, J. A., Ricker, G. R. 2003, SPIE, 4851, p.28-44 Haberl, F., & Titarchuk, L. 1995, A&A, 299, 414 Hasinger, G., & van der Klis, M. 1989, A&A, 225, 79 Kallman, T., & White, N.E. 1989, ApJ, 341, 955 Liu, Q. Z., van Paradijs, J., van den Heuvel, E. P. J. 2001, A&A, 368, 1021 Miller, J. M., Fabian, A. C., Wijnands, R., et al. 2002, ApJ, 578, 348 Miller, J. M., et al. 2002, American Astronomical Society, 34, 1206 Miller, M.C., Lamb, F.K., & Psaltis, D., 1998, ApJ, 508, 791 Stella, L. 1990, Nature, 344, 747 Stella, L., & Vietri, M., 1999, Phys. Rev., 82, L17 Titarchuk, L. 1994, ApJ, 434, 570 White, N.E., & Holt, S.S. 1982, ApJ, 257, 318 White, N.E., et al. 1986, MNRAS, 218, 129 Parameter Value --------------------------------------------- ----------------------- $N_{\rm H}$ $\rm (\times 10^{22}\;cm^{-2})$ $1.42 \pm 0.06$ Si / Si$_\odot$ $2.0 \pm 0.2$ $k T_0$ (keV) $0.50 \pm 0.02$ $k T_{\rm e}$ (keV) $2.29 \pm 0.09$ $\tau$ $17.7 \pm 0.7$ Flux (1.3–10 keV, erg cm$^{-2}$ s$^{-1}$) $7.82 \times 10^{-9}$ Final $\chi^2(d.o.f.)$ $3168 / 3255$ $E_1$ (keV, ID: Mg XII Ly$\alpha$) $1.476 \pm 0.007$ $\sigma_1$ (eV) $17 \pm 6$ I$_1$ ($10^{-3}$ cm$^{-2}$ s$^{-1}$) $2.3 \pm 0.9$ EW$_1$ (eV) $ 4.28 $ $E_2$ (keV, ID: Si XIV Ly$\alpha$ ?) $2.03 \pm 0.01$ $\sigma_2$ (eV) $28 \pm 10$ I$_2$ ($10^{-3}$ cm$^{-2}$ s$^{-1}$) $1.9 \pm 0.7$ EW$_2$ (eV) $ 4.17 $ $E_3$ (keV, ID: S XVI Ly$\alpha$) $2.64 \pm 0.02$ $\sigma_3$ (eV) $40 \pm 14$ I$_3$ ($10^{-3}$ cm$^{-2}$ s$^{-1}$) $2.3 \pm 0.8$ EW$_3$ (eV) $ 6.32 $ $E_{\rm Fe}$ (keV) $6.54 \pm 0.07$ $\sigma_{\rm Fe}$ (keV) $0.51 \pm 0.08$ I$_{\rm Fe}$ ($10^{-2}$ cm$^{-2}$ s$^{-1}$) $1.5 \pm 0.3$ EW$_{\rm Fe}$ (eV) $ 170 $ : Results of the fitting of the 705 HEG first order spectra in the 1.3–10 keV energy band. []{data-label="table:1"} \ The model consists of a Comptonized spectrum modeled by [comptt]{}, and four Gaussian emission lines. $k T_0$ is the temperature of the seed photon (Wien) spectrum, $k T_e$ the electron temperature and $\tau$ the optical depth in a spherical geometry. For the discrete features, $I$ is the intensity of the line and $EW$ is the corresponding equivalent width. Uncertainties are given at 90% confidence level. Parameter Value ---------------------------------- ------------------------ Energy (keV) $6.40 \pm 0.04$ R$_{\rm in}$ (R$_g$) $7^{+4}_{-1}$ $(< 11)$ R$_{\rm out}$ (R$_g$) $410^{+230}_{-130}$ Inclination (deg) $59^{+25}_{-4}$ Index $2.1 \pm 0.2$ I ($10^{-2}$ cm$^{-2}$ s$^{-1}$) $1.8 \pm 0.3$ Final $\chi^2(d.o.f.)$ $3164/3252$ : Iron line parameters from the [diskline]{} model.[]{data-label="table:2"} \ The other best fit parameters are compatible with those reported in Table 1. Index refers to the power-law dependence of emissivity which scales as $r^{-\rm Index}$. Uncertainties are 90% confidence level for a single parameter of interest.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We predict the fraction of dark halo lenses, that is, the fraction of lens systems produced by the gravitational potential of dark halos, on the basis of a simple parametric model of baryonic compression. The fraction of dark halo lenses primarily contains information on the effect of baryonic compression and the density profile of dark halos, and is expected to be insensitive to cosmological parameters and source population. The model we adopt comprises the galaxy formation probability $p_{\rm g}(M)$ which describes the global efficiency of baryonic compression and the ratio of circular velocities of galaxies to virial velocities of dark halos $\gamma_v=v_{\rm c}/v_{\rm vir}$ which means how the inner structure of dark halos is modified due to baryonic compression. The model parameters are constrained from the velocity function of galaxies and the distribution of image separations in gravitational lensing, although the degeneracy between model parameters still remains. We show that the fraction of dark halo lenses depends strongly on $\gamma_v$ and the density profile of dark halos such as inner slope $\alpha$. This means that the observation of the fraction of dark halos can break the degeneracy between model parameters if the density profile of dark halo lenses is fully settled. On the other hand, by restricting $\gamma_v$ to physically plausible range we can predict the lower limit of the fraction of dark halo lenses on the basis of our model. Our result indicates that steeper inner cusps of dark halos ($\alpha\gtrsim 1.5$) or too centrally concentrated dark halos are inconsistent with the lack of dark halo lenses in observations.' author: - Masamune Oguri title: Constraints on the Baryonic Compression and Implications for the Fraction of Dark Halo Lenses --- Introduction ============ Strong gravitational lensing offers a powerful probe of the matter distribution in the universe. So far $\sim60$ lensed quasars are known, and their properties are summarized by CfA/Arizona Space Telescope Lens Survey (CASTLES[^1]). Most of these lens systems have image separations $\theta\lesssim3''$. The cold dark matter (CDM) scenario, however, predicts sufficiently cuspy dark halos [e.g., @navarro96; @navarro97], thus dark halos can produce the significant amount of multiple images even at $\theta\gtrsim 3''$ [@wyithe01; @keeton01b; @takahashi01; @li02; @oguri02a]. It has been unclear whether the distribution of image separations should be computed based on galaxies [using the luminosity function and the density profile of galaxies; @turner84; @turner90; @fukugita91; @fukugita92; @maoz93; @kochanek96; @chiba99] or dark halos [using the mass function and the density profile of dark halos; @narayan88; @cen94; @wambsganss95; @kochanek95; @maoz97; @wambsganss98; @mortlock00; @wyithe01; @keeton01b; @takahashi01; @oguri02a]. The distribution of image separations, however, clearly indicates that the use of only dark matter properties cannot match observations. That is, the observed distribution of image separations is never reproduced from the theoretical calculation using the mass function of dark halos and only one population for lensing objects [e.g., @li02]. The possible solution to explain the distribution of image separations is the modification of inner structure of dark halos by introducing baryonic cooling [@keeton98; @porciani00; @kochanek01b; @keeton01a; @sarbu01; @li02]. In this picture, inside low mass halos baryons are efficiently compressed and form sufficiently concentrated galaxies, while inside larger halos such as group- or cluster-mass halos global baryon cooling hardly occurs and thus the inner structure of dark halos remains unmodified. Then a question comes to our mind: [what is the fraction of dark halo lenses?]{} Here the term “dark halo lenses” is used to describe lenses which are produced by the gravitational potential of dark halos. In other words, the lens objects of dark halo lenses are dark halos which have no central galaxies or have central galaxies but they are too small to dominate in gravitational lensing. Dark halo lenses exhibit characteristic properties such as the small flux ratios and the detectable odd images [@rusin02], thus can be distinguished from usual galaxy lenses. Since dark halos are expected to have a steep central cusp, the significant amount of dark halo lenses should be observed. But so far no confirmed dark halo lens system is observed in strong gravitational lensing survey. The exception is arc statistics in rich clusters [@bartelmann98; @williams99; @meneghetti01; @molikawa01; @oguri01; @oguri02b], but the known cluster lenses were all found by searching for lenses in detail after identifying a rich cluster. In the surveys which first identify source objects and see whether they are lensed or not, it seems that dark halo lenses have not been observed yet: statistical argument [@kochanek99] and individual properties [@rusin02] imply that current ambiguous quasar pairs are likely to be binary quasars. Even known lensing systems in clusters, such as Q0957+561 [@walsh79], are produced mainly by a galaxy in the cluster. The cluster potential contributes to lensing only as a perturbation. Therefore, it should be checked whether the lack of dark halo lenses in observations really reconciles with the theoretical prediction. Description of baryonic effects needs detailed models for the star formation and feedback [e.g., @cole00]. Instead, in this paper we predict the expected fraction of dark halo lenses on the basis of a simple (minimal) parametric model [@kochanek01a]. This model comprises the formation probability of galaxies, $p_{\rm g}(M)$, and the ratio of circular velocities of galaxies to virial velocities of dark halos, $\gamma_v$. The former describes the global efficiency of baryonic compression, and the latter models the modification of inner structure of dark halos due to baryonic compression. The model parameters are chosen so as to reproduce the velocity function of galaxies and the distribution of image separations. Although there remains the strong degeneracy between model parameters, this degeneracy can be broken from the observation of the fraction of dark halos if we fix the density profile of dark halos. On the other hand, by restricting a range of $\gamma_v$ from various theories and observations, we can also derive the lower limit of the fraction of dark halo lenses. Our main finding is that steep inner slopes of dark halos ($\alpha\gtrsim1.5$) or too centrally concentrated dark halos are inconsistent with the lack of dark halo lenses in observations, even if various uncertainties are taken into account. Although this constraint on the density profile is not so severe, a large lens sample obtained by e.g., Sloan Digital Sky Survey (SDSS) can put tighter constraints on the density profile of dark halos as well as the model of baryonic compression. The plan of this paper is as follows. In §\[sec:th\], we describe the model of baryonic compression. Section \[sec:const\] is devoted to constrain the model parameters, and §\[sec:frac\] presents our predictions for the fraction of dark halo lenses. Finally, we summarize conclusions in §\[sec:conc\]. Throughout this paper, we assume the lambda-dominated cosmology $(\Omega_0, \lambda_0,h,\sigma_8)=(0.3,0.7,0.7,1.04)$, where the Hubble constant in units of $100{\rm km\,s^{-1}Mpc^{-1}}$ is denoted by $h$. As shown below, however, our results are quite insensitive to a particular choice of cosmological parameters. Theoretical Model\[sec:th\] =========================== Effects of Baryons on Dark Halos\[sec:th:baryon\] ------------------------------------------------- The obvious difference between velocity functions of galaxies and dark halos at high velocity [e.g., @gonzalez00] indicates that the efficiency of the baryon compression changes from galaxy-mass scale to group- and cluster-mass scales. If the mass of dark halos is sufficiently large, the baryon cooling time $\tau_{\rm cool}$ which increases with halo mass [e.g., @cole00] becomes larger than the age of dark halos which slightly deceases with halo mass [e.g., @lacey93; @kitayama96]. Thus baryons inside such a massive dark halo do not form so large galaxies as to collect most of baryons in the dark halo. This yields a steep cutoff at high velocity in the galaxy velocity function. To model this, following @kochanek01a, we introduce the probability $p_{\rm g}(M)$ that a sufficiently large galaxy which collects most of internal baryons is formed inside a halo of mass $M$. For example, galaxy-mass halos with mass $M$ usually have large galaxies with mass $M_{\rm gal}\sim(\Omega_{\rm b}/\Omega_0)M$. On the other hand, groups or clusters also have galaxies but their mass are small even for galaxies which lie at the center of halos, $M_{\rm gal}\ll(\Omega_{\rm b}/\Omega_0)M$, here in this case $M\gtrsim10^{13}h^{-1}M_\odot$ [e.g., @yoshikawa01]. Therefore we regard galaxies in groups or clusters as “substructures” and do not take into account. We adopt a parametric model [@kochanek01a] $$\begin{aligned} p_{\rm g}(M)&=& \left\{ \begin{array}{@{\hspace{0.6mm}}ll} \displaystyle{1} & \mbox{($M<M_{\rm h}$)},\\ \displaystyle{\exp\left[1-\left(\frac{M}{M_{\rm h}}\right)^{\delta_{\rm h}}\right] } & \mbox{($M>M_{\rm h}$)}. \end{array} \right.\label{gfp}\end{aligned}$$ Although the overall factor of $p_{\rm g}(M)$ should become another parameter, we neglect it. The effect of the overall factor of $p_{\rm g}(M)$ on the fraction of dark halo lenses will be discussed in §\[sec:frac\]. Also, there is a difference in the slopes of velocity functions of dark halos and galaxies at small velocity [e.g., @gonzalez00; @nagamine01], and this demands the modification of $p_{\rm g}(M)$ at low $M$. We, however, neglect this effect because we are not interested in low velocity galaxies which hardly contribute to lensing statistics. The baryon compression also changes inner structure of dark halos [e.g., @blumenthal86; @mo98]. Before the baryon cooling occurs, we assume that the density profile of dark halos is well described by the one-parameter family of the form [@zhao96; @jing00a] $$\rho(r)=\frac{\rho_{\rm crit}\delta_{\rm c}} {\left(r/r_{\rm s}\right)^\alpha\left(1+r/r_{\rm s}\right)^{3-\alpha}}, \label{nfw}$$ where $r_{\rm s}=r_{\rm vir}/c_{\rm vir}$ and $c_{\rm vir}$ is the concentration parameter. We adopt the mass and redshift dependence reported by @bullock01: $$c_{\rm vir}(M, z)=\frac{8}{1+z}\left(\frac{M}{10^{14}h^{-1}M_{\odot}}\right)^{-0.13},\label{bul}$$ for $\alpha=1$, and generalize it to $\alpha\neq1$ by multiplying the factor $(2-\alpha)$ [@keeton01b]. We also take account of scatter of the concentration parameter which has a log-normal distribution with the dispersion of $\sigma_{\rm c}=0.18$ [@jing00b; @bullock01]. The characteristic density $\delta_{\rm c}$ can be computed using the spherical collapse model [see @oguri01]. While the correct value of $\alpha$ is still unclear, the existence of a cups with $1\lesssim\alpha\lesssim1.5$ has been established in recent N-body simulations [@navarro96; @navarro97; @moore99; @ghigna00; @jing00a; @klypin01; @fukushige01; @fukushige02; @power02]. Next consider the modification of inner structure of dark halos induced by baryonic compression. In general, cooled baryons (galaxies) are more centrally concentrated than dark matters. Although complicated physical models are needed to know the modified mass distribution, we simply assume that the mass distribution of galaxies is well described by the Singular Isothermal Sphere (SIS) approximation: $$\rho(r)=\frac{\sigma^2}{2\pi Gr^2},$$ where $\sigma$ is a one-dimensional velocity dispersion and is related to the circular velocity $v_{\rm c}$ as $v_{\rm c}=\sqrt{2}\sigma$. We also assume that $v_{\rm c}$ as a function of halo mass $M$ can be described as $$\gamma_v=\frac{v_{\rm c}}{v_{\rm vir}(M)},\label{gammav}$$ where $v_{\rm vir}(M)$ is the halo virial velocity defined by $v_{\rm vir}(M)=\sqrt{GM/r_{\rm vir}}$ and $\gamma_v$ is an arbitrary constant. The SIS approximation for galaxies is consistent with several observations such as dynamics [e.g., @rix97] and gravitational lensing [e.g., @rusin01a; @cohn01]. Velocity Function of Galaxies\[sec:th:vf\] ------------------------------------------ Once the probability $p_g(M)$ and the galaxy circular velocity $v_{\rm c}(M)$ are given, we can calculate the velocity function of galaxies from the mass function of dark halos: $$\frac{dn}{dv_{\rm c}}(v_{\rm c}(M))=p_{\rm g}(M)\left\|\frac{dv_{\rm c}(M)}{dM}\right\|^{-1}\frac{dn}{dM}(M). \label{vf}$$ For the mass function $dn/dM$, we adopt a fitting form derived by @sheth99: $$\begin{aligned} \frac{dn_{\rm ST}}{dM}&=&0.322\left[1+\left(\frac{\sigma_M^2} {0.707\delta_0^2(z)}\right)^{0.3}\right]\sqrt{\frac{1.414}{\pi}}\nonumber\\ &&\times\frac{\rho_0}{M}\frac{\delta_0(z)}{\sigma_M^2}\left\| \frac{d\sigma_M}{dM}\right \|\exp\left[-\frac{0.707\delta_0^2(z)}{2\sigma_M^2}\right]. \label{st}\end{aligned}$$ This fitting form coincides more accurately with numerical simulations than the analytic mass function derived by @press74. The velocity function is sometimes expressed in terms of $\log v_{\rm c}$ as $$\Psi(v_{\rm c})=\frac{dn}{d\log v_{\rm c}}.$$ In this calculation we neglect the contribution from “substructures” (i.e., galaxies in groups and clusters), because this mainly changes the normalization of the velocity function which we do not use as a constraint on the model parameters (see §\[sec:const:vf\]). Substructures, however, may affect the fraction of dark halo lenses directly, because it changes only the number of galaxy lenses. Therefore we consider the effect of substructures in predicting the fraction of dark halo lenses (see §\[sec:frac\]). Lensing Probability Distribution -------------------------------- Bearing the picture described in §\[sec:th:baryon\] in mind, we calculate the probability of gravitational lensing caused by bright galaxies and dark halos separately: $$\begin{aligned} P_{\rm gal}(>\theta; z_{\rm S}, L)=&&\nonumber\\ &&\hspace{-15mm}\int_0^{z_{\rm S}}dz_{\rm L}\int_{M_{\rm min}}^{\infty} dM\,p_{\rm g}(M)\sigma_{\rm SIS}\,B\frac{c\,dt}{dz_{\rm L}}(1+z_{\rm L})^3\frac{dn}{dM},\label{pgal}\\ P_{\rm dark}(>\theta; z_{\rm S}, L)=&&\nonumber\\ &&\hspace{-25mm}\int_0^{z_{\rm S}}dz_{\rm L}\int_{M_{\rm min}}^{\infty} dM\,\left\{1-p_{\rm g}(M)\right\}\sigma_{\rm NFW}\,B\frac{c\,dt}{dz_{\rm L}}(1+z_{\rm L})^3\frac{dn}{dM},\label{pdark}\end{aligned}$$ where $z_{\rm S}$ is the source redshift, $z_{\rm L}$ is the lens redshift, $L$ is the luminosity of the source, and $dn/dM$ is the comoving number density of dark halos (eq. \[\[st\]\]). Lensing cross sections $\sigma_{\rm SIS}$ and $\sigma_{\rm NFW}$ are given by $$\begin{aligned} \sigma_{\rm SIS}&=&16\pi^3\left(\frac{\sigma}{c}\right)^4\left(\frac{D_{\rm OL}D_{\rm LS}}{D_{\rm OS}}\right)^2,\label{cs_sis}\\ \sigma_{\rm NFW}&=&\pi\left(\eta_{\rm rad}\frac{D_{\rm OL}}{D_{\rm OS}}\right)^2,\end{aligned}$$ where $\eta_{\rm rad}$ is the radius of the radial caustic at source plane [e.g., @schneider92] and $D_{\rm OL}$, $D_{\rm OS}$, and $D_{\rm LS}$ are the angular diameter distances to the lens, to the source, and between the lens and source, respectively. The one-dimensional velocity dispersion $\sigma$ in equation (\[cs\_sis\]) can be represented as a function of halo mass $M$: $$\sigma=\frac{v_{\rm c}}{\sqrt{2}}=\frac{\gamma_vv_{\rm vir}(M)}{\sqrt{2}},$$ where equation (\[gammav\]) is used. The lower limit of integral by mass, $M_{\rm min}$, is determined by solving the equation $\theta=\theta(M_{\rm min},z_{\rm S},z_{\rm L})$. The magnification bias [@turner80; @turner84] is included in $B$ as follows: $$B=\frac{1}{\sigma_{\rm lens}\Phi(z_{\rm S}, L)}\int_{\rm multi}d^2\eta\,\Phi(z_{\rm S}, L/\mu(\vec{\eta}))\frac{1}{\mu(\vec{\eta})}, \label{biasfactor}$$ where $\Phi(z_{\rm S}, L)$ is the luminosity function of sources and $\mu(\vec{\eta})$ is the magnification factor at $\vec{\eta}$. The total lensing probability with image separation larger than $\theta$ is given by $$P(>\theta; z_{\rm S}, L)=P_{\rm gal}(>\theta; z_{\rm S}, L)+P_{\rm dark}(>\theta; z_{\rm S}, L). \label{dist}$$ Constraints on the Model\[sec:const\] ===================================== Velocity Function of Galaxies\[sec:const:vf\] --------------------------------------------- The velocity function is valuable because it is easy to handle theoretically, and is useful to test the model of galaxy formation and cosmology [@cole89; @shimasaku93; @gonzalez00; @kochanek01b]. Using the model described in §\[sec:th:vf\], we can calculate the velocity function of galaxies from the mass function of dark halos. By comparing the theoretical velocity function with observed velocity functions, one can constrain the model parameters such as $\delta_{\rm h}$, $M_{\rm h}$, and $\gamma_v$. The definite theoretical prediction, however, needs the [correct]{} value of cosmological parameters such as $\sigma_8$. Therefore we use the normalized velocity function $$\psi(v_{\rm c})\equiv\frac{\Psi(v_{\rm c})}{\Psi(200{\rm km/s})}, \label{nvf}$$ instead of $\Psi(v_{\rm c})$. Moreover, by using $\psi(v_{\rm c})$ we can neglect the effect of the overall factor of $p_{\rm g}(M)$. For the observed velocity functions, we use five velocity functions derived by @gonzalez00: velocity functions derived from Southern Sky Redshift Survey (SSRS2), Automatic Plate Measuring facility survey (APM), United Kingdom Schmidt Telescope survey (UKST), Las Campanas Redshift Survey (LCRS), and K-band survey by @gardner97. We restrict the comparison with observations in the region $200{\rm km/s}<v_{\rm c}<500{\rm km/s}$ because of the following two reasons. One reason is that observed velocity functions are not reliable at the high velocity region ($v_{\rm c}>500{\rm km/s}$) [@gonzalez00]. The other reason is the obvious difference between observed and theoretical number density of galaxies at the low velocity region ($v_{\rm c}<200{\rm km/s}$). The reason of this difference is that in our model we neglect effects of e.g., supernovae feedback [@dekel86] which substantially suppress the number of galaxies in low-mass halos [e.g., @gonzalez00; @nagamine01]. However we are interested in the gravitational lensing with angular separation larger than current angular resolution ($\theta\sim0.3''$), thus we can safely neglect such low-velocity galaxies which have little impact on our lensing results. Figure \[fig:chis\] shows constraints on model parameters from observed velocity functions. Contours are calculated from ratios of combined likelihood ${\cal L}\propto\prod\exp(-\frac{1}{2}\chi_i^2)$. Errors of observed velocity functions are estimated from errors of fitting parameters of velocity functions [see @gonzalez00]. As easily seen from the upper panel of Figure \[fig:chis\], we can put useful constraints on $\delta_{\rm h}$ and $v_{\rm h}\equiv v_{\rm c}(M_{\rm h})$. The best fit parameter set is $(\delta_{\rm h},v_{\rm h})\simeq(0.78,238{\rm km/s})$. This means that there still remains strong degeneracy between $M_{\rm h}$ and $\gamma_v$: parameter combinations which yield the same $v_{\rm h}$ cannot be discriminated from the velocity functions. This fact is also shown in the lower panel of Figure \[fig:chis\]. One of our best fit parameter sets $(\delta_{\rm h},M_{\rm h},\gamma_v)=(0.78,10^{12}h^{-1}M_\odot,1.67)$ is compared with observed velocity functions in Figure \[fig:vf\]. It is clearly shown that the prediction of our model shows good coincidence with observed velocity functions. Image Separation Distribution\[sec:const:sep\] ---------------------------------------------- The distribution of image separations in strong gravitational lensing becomes another test of our model. The distribution is calculated from equation (\[dist\]). For the observed lens sample, we consider the Cosmic Lens All-Sky Survey [CLASS; @helbig00]. This sample is complete at image separations $0.3''<\theta<15''$ [@helbig00; @phillips01], and has 18 lenses among $\sim 12,000$ radio sources. Sources have the flux distribution $dn/dS\propto S^{-2.1}$ [@rusin01b]. The mean redshift is estimated to be $\langle z_{\rm S} \rangle=1.27$, although the redshift distribution of sources is still poorly understood [@marlow00]. As in the case of the velocity function, actually we consider the normalized image separation distribution: $$p(>\theta; z_{\rm S}, L)=\frac{P(>\theta; z_{\rm S}, L)}{P(>0.3''; z_{\rm S}, L)}, \label{ndist}$$ instead of usual probability distribution. This is because the absolute probability may suffer from uncertainties of source redshifts, magnification bias, and cosmological parameters. On the other hand the distribution $p(>\theta)$ mainly contains information on the shape of the mass function and effects of baryonic compression [@kochanek01a; @kochanek01b; @keeton01a]. And the distribution $p(>\theta)$ is quite insensitive to the source population and cosmological parameters, thus can be used for samples in which the source population is unknown such as CASTLES sample. The distribution of image separations is tested by using a Kolmogorov-Smirnov (KS) test. For the observed distribution, we use the distribution of CLASS survey. The result is shown in Figure \[fig:ks\]. We plot contours in $M_{\rm h}$-$\gamma_v$ plane by fixing $\delta_{\rm h}=0.78$ which is constrained from the velocity function (see Figure \[fig:chis\]). From this figure, we find that the constraint from the distribution of image separations is consistent with the constraint from the velocity function. The exception is the case of $\alpha=1.5$, where $\alpha$ is the inner slope of the density profile of dark halos (eq. \[\[nfw\]\]). In this case, the lower value of $\gamma_v$ becomes inconsistent with the observation, because the lensing cross section for SIS is $\sigma_{\rm SIS}\propto \gamma_v^4$ (see eq. \[\[cs\_sis\]\]) and the lower value of $\gamma_v$ means that the contribution of dark halo lenses becomes more significant. That is, when $\gamma_v$ is low, the number of dark halo lenses is comparable to that of galaxy lenses, thus a steep cutoff in the distribution of image separations never appears. Therefore, we conclude that the constraint from the distribution of image separations is consistent with the constraint from the velocity function, except for the cutoff of KS probability at low $\gamma_v$ which may appear when $\alpha$ is large. We also compare the distribution of our model with the observed distribution in Figure \[fig:sep\]. The model parameters are chosen so as to satisfy the constraint from the velocity function, and same as those used in Figure \[fig:vf\]. We also plot the distribution assuming that all lens objects are well approximated by SIS. This assumption corresponds to $p_{\rm g}(M)=1$ for all $M$. It is obvious that our model well reproduces the observed distribution while the distribution of all SIS assumption is far from the observed distribution, as pointed out by @keeton98. Various Constraints on the Circular Velocity of Galaxies\[sec:const:vc\] ------------------------------------------------------------------------ In §\[sec:const:vf\] and §\[sec:const:sep\], we constrained model parameters from the velocity function of galaxies and the distribution of image separations. The strong degeneracy between $M_{\rm h}$ and $\gamma_v$, however, still remains. Therefore in this subsection we try to put constraints on $\gamma_v$ from various theories and observations. We consider following three constraints. First, @seljak02a gave values of $\gamma_v$ at several halo mass derived from the observation and analysis of galaxy-galaxy lensing [@mckay02; @guzik02] and Tully-Fisher/fundamental plane relations. The observation of galaxy-galaxy lensing allows us to determine the mass of dark halos and its relation to the luminosity of galaxies. Tully-Fisher/fundamental plane relations are used to derive the galaxy circular velocity from its luminosity. He found that $\gamma_v$ is significantly larger than 1, $\gamma_v\sim1.8$ around $L_$. He also found the decrease of $\gamma_v$ from $L_$ to $7L_$, $\gamma_v\sim1.4$ at $7L_$. Second, @cole00 semi-analytically calculated average values of $\gamma_v$ for galaxies with $-20<M_I-5\log h<-18$. They calculated $\gamma_v$ for various parameter sets of semi-analytic model, and found that most of parameter sets predict $\gamma_v\sim1.3-1.4$. The most important parameter for $\gamma_v$ is the baryon density $\Omega_{\rm b}$, but even in the extreme cases they examined, $\Omega_{\rm b}=0.01$ and $0.04$, $\gamma_v$ becomes $1.13$ and $1.96$, respectively. Third, @mo98 considered the analytic disk formation model on the basis of adiabatic compression [e.g., @blumenthal86]. In their model, $\gamma_v$ depends on the mass of cooled baryons and the angular momentum of galactic disc. More specifically, their model is characterized by following three parameters; the concentration parameter $c$, the ratio of disk mass to halo mass $m_{\rm d}$, and the angular momentum $\lambda'$ [see @mo98]. We assume that the concentration parameter is related to halo mass as in equation (\[bul\]). Then $\gamma_v$ at fixed mass depends on $m_{\rm d}$ and $\lambda'$. We examine following two extreme cases: $(m_{\rm d},\lambda')=(0.02,0.02)$ and $(0.02,0.2)$. Most of parameter sets which are physically reasonable predict the value of $\gamma_v$ between these two extreme cases. The result is summarized in Figure \[fig:vcvvir\]. Particularly we are interested in the strong gravitational lensing, thus we focus on the range $10^{12}h^{-1}M_\odot\lesssim M \lesssim 10^{13}h^{-1}M_\odot$ where gravitational lensing with separations $\theta>0.3''$ is most efficient. In this range of halo mass, various constraints indicate that $\gamma_v$ is restricted to $1\lesssim \gamma_v\lesssim 2$. In particular, the upper bound of $\gamma_v$ is quite robust because $\gamma_v$ is closely related to the amount of cooled baryons; to produce $\gamma_v\gg 2$, we need the extraordinarily large amount of cooled baryon which exceeds the global baryon mass ratio [@seljak02a]. Figure \[fig:vcvvir\] also indicates that $\gamma_v$ should slightly depend on halo mass. The mass dependence of $\gamma_v$, however, is not so important because our interest is restricted in a narrow mass range. Therefore we can safely assume that $\gamma_v$ is constant. The Fraction of Dark Halo Lenses\[sec:frac\] ============================================ According to the model described in §\[sec:th\], we predict the fraction of dark halo lenses. In our model, there are three parameters which govern effects of baryonic compression; $\delta_{\rm h}$, $M_{\rm h}$, and $\gamma_v$. Since the constraint from the velocity function of galaxies (§\[sec:const:vf\]) indicates that $\delta_{\rm h}$ should be restricted around $\delta_{\rm h}\sim 0.78$, we fix $\delta_{\rm h}=0.78$ in the remainder of this paper. On the other hand, the other parameters, $M_{\rm h}$ and $\gamma_v$, were poorly constrained from the velocity function of galaxies and the distribution of image separations. Therefore we should see the dependence of the fraction of dark halo lenses on both $M_{\rm h}$ and $\gamma_v$. Another important parameter we should examine is the inner slope of density profile ($\alpha$ in eq. \[\[nfw\]\]), because it is also known that the number of dark halo lenses is extremely sensitive to $\alpha$ [@wyithe01; @keeton01b; @takahashi01; @li02; @oguri02a]. The fraction of dark halo lenses at image separations $\theta_1<\theta<\theta_2$ is given by $$f_{\rm dark}(\theta_1<\theta<\theta_2)=\frac{P_{\rm dark}(>\theta_1)-P_{\rm dark}(>\theta_2)}{P(>\theta_1)-P(>\theta_2)},\label{fdark}$$ where $P_{\rm dark}(>\theta)$ and $P(>\theta)$ are calculated from equations (\[pdark\]) and (\[dist\]), respectively. Again, $f_{\rm dark}$ is almost independent of the source population (source redshift and flux distribution) and cosmological parameters because these primarily change the normalization of the overall lensing rate which we never use. For the range of image separations, we consider following two cases: (1) $0.3''<\theta<15''$. In this range of image separations, CLASS survey is complete [@helbig00; @phillips01]. Among 18 lenses observed in CLASS survey, virtually all lenses are known to be galaxy lenses [@rusin02]. (2) $0.3''<\theta<3''$. Most of current ambiguous quasar pairs which may be dark halo lenses have separations $\theta>3''$ and almost all lenses with smaller separations are known to be produced by normal galaxies [@kochanek99; @rusin02]. Therefore, in this case we can use all lenses with separations $0.3''<\theta<3''$ in comparison with our result. Below we assume that there is no dark halo lens for both cases. Figure \[fig:fd\_cont\] plots contours of $f_{\rm dark}$ in $M_{\rm h}$-$\gamma_v$ plane. The constraint from the velocity function of galaxies is also shown. From this figure, it is found that the fraction of dark halo lenses is sensitive to $\gamma_v$: $f_{\rm dark}$ increases as $\gamma_v$ decreases. The reason is that the cross section of SIS lenses scales as $\sigma_{\rm SIS}\propto \gamma_v^4$ (see eq. \[\[cs\_sis\]\]) and the number of galaxy lenses decreases as $\gamma_v$ decreases. The fraction of dark halo lenses also increases as $M_{\rm h}$ decreases, because the number of dark halos which act as dark halo lenses becomes large. Figure \[fig:fd\_cont\] indicates that $f_{\rm dark}$ cannot be determined uniquely, even if we restrict $M_{\rm h}$ and $\gamma_v$ such that they satisfy the constraint from the velocity function of galaxies. This fact is clearly shown in Figure \[fig:fd\]. In Figure \[fig:fd\], we plot $f_{\rm dark}$ as a function of $\gamma_v$. The parameter $M_{\rm h}$ is chosen so that it satisfies the constraint from the velocity function of galaxies. This figure indicates that $f_{\rm dark}$ is sensitive to $\gamma_v$. Even if we restrict $1\leq \gamma_v\leq 2$ as discussed in §\[sec:const:vc\], $f_{\rm dark}$ has uncertainty of about one order of magnitude. Figure \[fig:fd\] also shows that $f_{\rm dark}$ is sensitive to $\alpha$. From $\alpha=1.0$ to $1.5$, $f_{\rm dark}$ also changes about one order of magnitude. Therefore, our conclusion is that robust predictions of $f_{\rm dark}$ need information on both $\gamma_v$ and $\alpha$. If the density profile of dark halos is determined from several observations, we can break the degeneracy between $M_{\rm h}$ and $\gamma_v$ from the observation of $f_{\rm dark}$. Although $f_{\rm dark}$ strongly depends on $\gamma_v$, we obtain the lower limit $f_{\rm dark,min}$ by restricting $\gamma_v$. Discussions in §\[sec:const:vc\] suggest that it is safe to adopt $\gamma_v\leq 2$. Thus by setting $\gamma_v=2$ we can derive $f_{\rm dark,min}$, which is shown in Figure \[fig:fd\_alpha\]. In this figure $f_{\rm dark, min}$ is plotted against the inner slope $\alpha$. The lack of dark halo lenses places upper limit on the value of $f_{\rm dark}$ which is also shown in Figure \[fig:fd\_alpha\]. From this figure, we find that too cuspy inner slope of dark halos ($\alpha\gtrsim 1.5$) is inconsistent with the lack of dark halo lenses: even if we adopt sufficiently large $\gamma_v$, $\gamma_v=2$, we predict too much dark halo lenses to reconcile with the observation. This constraint has somewhat different meaning from the one derived from the statistics of wide separation lensing [@keeton01b; @li02] because such statistics may suffer from the uncertainties of cosmological parameters and the source population. Are there any other effects which may change $f_{\rm dark}$? We consider following two effects which may change our results. One is the effect of “substructures” (i.e., galaxies in groups and clusters) as discussed in §\[sec:th:vf\]. Substructures may affect the fraction of dark halo lenses directly, because it changes only the number of galaxy lenses. Moreover the situation that lens galaxies which lie in groups or clusters is not rare (e.g., Q0957+561), thus the effect should be addressed quantitatively. Although the effect of substructures seems difficult to estimate, @keeton00 calculated the fraction of lensing galaxies which lie in groups and clusters. They predicted that $\sim 25\%$ of lens galaxies are likely to be in groups or clusters. Following this, the effect of substructures are simply estimated by replace the probability of galaxy lensing as $P_{\rm gal}\rightarrow (4/3)P_{\rm gal}$. This decreases $f_{\rm dark}$ by a factor $3/4$ at most. The fraction of dark halo lenses including this effect is also displayed in Figure \[fig:fd\_alpha\]. As seen in the figure, our result that $\alpha\gtrsim 1.5$ is inconsistent with the observation is not so affected by the effect of substructures. The other effect is the existence of “empty halos”, that is, dark halos which do not host a galaxy. We have adopted the galaxy formation probability $p_{\rm g}(M)$ of the form (\[gfp\]), and this means that we have assumed there is no empty halo at $M<M_{\rm h}$. Although this assumption does not conflict with the observation of galaxy-galaxy lensing [@seljak02b], it is still possible that the significant amount of empty halos exists at galaxy-mass scale. This effect, however, only [increases]{} $f_{\rm dark}$. Therefore [lower limit]{} of $f_{\rm dark}$ used in Figure \[fig:fd\_alpha\] is never changed by this effect. The main result of @wyithe01 and @keeton01b is that the lensing probability of dark halos depends strongly on the choice of concentration parameters as well as inner slopes $\alpha$. Therefore, in Figure \[fig:fdmin\_cont\] we examine $f_{\rm dark,min}$ for different concentration parameters. In calculating this, we adopt the concentration parameter of the following form: $$c_{\rm vir}(M, z)=\frac{c_{\rm norm}}{1+z}\left(\frac{M}{10^{14}h^{-1}M_{\odot}}\right)^{-0.13},\label{bul2}$$ instead of equation (\[bul\]), and regard $c_{\rm norm}$ as a free parameter. We plot $f_{\rm dark,min}$ in $\alpha$-$c_{\rm norm}$ plane. From this figure, it is clearly seen that there is a strong degeneracy between $\alpha$ and $c_{\rm norm}$ as reported by @wyithe01 and @keeton01b. Thus actually we can constrain the combination of $\alpha$ and $c_{\rm norm}$, or the core mass fraction proposed by @keeton01b, instead of $\alpha$ and $c_{\rm norm}$ separately. Conclusion\[sec:conc\] ====================== We have studied the effect of baryonic compression assuming the simple parametric model used by @kochanek01a. Our model has following two elements: the galaxy formation probability $p_{\rm g}(M)$ (eq. \[\[gfp\]\]) which describes the global efficiency of baryonic compression, and the ratio of circular velocities of galaxies to virial velocities of dark halos $\gamma_v=v_{\rm c}/v_{\rm vir}$ which means how the inner structure of dark halos is modified due to baryonic compression. The model parameters are constrained from the observed velocity function of galaxies and the distribution of image separations in strong gravitational lensing, although the strong degeneracy between model parameters still remains. By using this model, we predict the fraction of dark halo lenses $f_{\rm dark}$ (eq. \[\[fdark\]\]). Here dark halo lenses mean the lens systems which are produce by the gravitational potential of dark halos. The fraction of dark halo lenses is independent of the normalization of total lensing rate, thus is insensitive to cosmological parameters and information of sources such as redshift and flux distribution. Instead the fraction of dark halo lenses is expected to have information on both the effect of baryonic compression and the density profile of dark halos such as the inner density profile for which we modeled $\rho\propto r^{-\alpha}$. We found that $f_{\rm dark}$ is indeed sensitive to the inner slope $\alpha$, concentration parameter $c_{\rm norm}$, and model parameters such as $\gamma_v$. Therefore, definite predictions of $f_{\rm dark}$ need correct knowledge of baryonic compression as well as the density profile of dark halos. This also means that we can constrain the model of baryonic compression from the observation of $f_{\rm dark}$ if the density profile of dark halos is well known. Although the fraction of dark halo lenses is difficult to predict, we can still derive the lower limit of $f_{\rm dark}$ by restricting $\gamma_v\leq2$ which is inferred from various theories and observations (see §\[sec:const:vc\]). We found that the steep inner profiles ($\alpha\gtrsim1.5$) or too centrally concentrated dark halos are inconsistent with the lack of dark halo lenses in observations. As described above, our result is quite insensitive to cosmological parameters and source population. Therefore, our result is complementary to the result of @keeton01b who obtained similar constraint on the density profile of dark halos using the total lensing probability which also depends strongly on cosmological parameters. One of possible systematic effects which may change $f_{\rm dark}$ is the effect of galaxies in groups or clusters. By using the result of @keeton00, we found that this effect changes $f_{\rm dark}$ by a factor $3/4$ at most. Therefore this effect does not change our main result so much. Other important systematic effect is the existence of empty halos which are neglected in our model. This effect is, however, not important for the lower limit of $f_{\rm dark}$, because this effect only increases the fraction of dark halo lenses. One possible criticism of our result is that the baryon compression model we use in this paper is too simple. Many of the previous work, however, have not addressed this problem. For example, most work of gravitational lensing statistics which use the mass function of dark halos assumed that circular velocities of galaxies are the same as virial velocities of dark halos (this corresponds to $\gamma_v=1$ in our model). But this assumption seems to be invalid [e.g., @seljak02a], and since the number of galaxy lenses scales as $\propto \gamma_v^4$ the deviation from $\gamma_v=1$ should not be dismissed. This fact is seen even in our simple model where the connection between galaxy lenses and dark halo lenses sensitively depends on $\gamma_v$, and is indeed difficult to be determined. The importance of baryonic compression, however, means that we can constrain the model of baryonic compression from observations of the fraction of dark halo lenses. Although the current sample of gravitational lensing may be too small for this purpose, the larger lens sample obtained by e.g., SDSS can strongly constrain the model of baryonic compression as well as the density profile of dark halos. The author would like to thank Yasushi Suto, Atsushi Taruya, and Mamoru Shimizu for useful discussions and comments. Bartelmann, M., Huss, A., Colberg, J. M., Jenkins, A., & Pearce, F. R. 1998, , 330, 1 Blumenthal, G. R., Faber, S. M., Flores, R., & Primack, J. R. 1986, , 301, 27 Bullock, J. S., Kolatt, T. S., Sigad, Y., Somerville, R. S., Kravtsov, A. V., Klypin, A. A., Primack, J. R., & Dekel, A. 2001, , 321, 559 Cen, R., Gott, J. R., Ostriker, J. P., & Turner, E. L. 1994, , 423, 1 Chiba, M., & Yoshii, Y. 1999, , 510, 42 Cohn, J. D., Kochanek, C. S., McLeod, B. A., & Keeton, C. R. 2001, , 554, 1216 Cole, S., & Kaiser, N. 1989, , 237, 1127 Cole, S., Lacey, C. G., Baugh, C. M., & Frenk, C. S. 2000, , 319, 168 Dekel, A., & Silk, J. 1986, , 303, 39 Fukugita, M., Futamase, T., Kasai, M., & Turner, E. L. 1992, , 393, 3 Fukugita, M., & Turner, E. L. 1991, , 253, 99 Fukushige, T., & Makino, J. 2001, , 557, 533 Fukushige, T., & Makino, J. 2002, , submitted (astro-ph/0108014) Gardner, J. P., Sharples, R. M., Frenk, C. S., & Carrasco, B. E. 1997, , 480, L99 Ghigna, S., Moore, B., Governato, F., Lake, G., Quinn, T., & Stadel, J. 2000, , 544, 616 Gonzalez, A. H., Williams, K. A., Bullock, J. S., Kolatt, T. S., & Primack, J. R. 2000, , 528, 145 Guzik, J., & Seljak, U. 2002, , in press (astro-ph/0201448) Helbig, P. 2000, preprint (astro-ph/0008197) Jing, Y. P. 2000, , 535, 30 Jing, Y. P., & Suto, Y. 2000, , 529, L69 Keeton, C. R. 1998, Ph. D. thesis, Harvard Univ. Keeton, C. R. 2001, , 561, 46 Keeton, C. R., Christlein, D., & Zabludoff, A. I. 2000, , 545, 129 Keeton, C. R., & Madau, P. 2001, , 549, L25 Kitayama, T., & Suto, Y. 1996, , 469, 480 Klypin, A., Kravtsov, A. V., Bullock, J. S., & Primack, J. R. 2001, , 554, 903 Kochanek, C. S. 1995, , 453, 545 Kochanek, C. S. 1996, , 466, 638 Kochanek, C. S. 2001, in STScI spring symposium, The Dark Universe: Matter, Energy, and Gravity, ed. M. Livio (Cambridge University Press), in press (astro-ph/0108160) Kochanek, C. S., Falco, E. E., & Mu[\~ n]{}oz, J. A. 1999, , 510, 590 Kochanek, C. S., & White, M. 2001, , 559, 531 Lacey, C., & Cole, S. 1993, , 262, 627 Li, L. X., & Ostriker, J. P. 2002, , 566, 652 Maoz, D., & Rix, H.-W. 1993, , 416, 425 Maoz, D., Rix, H.-W., Gal-Yam, A., & Gould, A. 1997, , 486, 75 Marlow, D. R., Rusin, D., Jackson, N., Wilkinson, P. N., Browne, I. W. A., & Koopmans, L. 2000, , 119, 2629 McKay T. A., et al. 2002, , submitted (astro-ph/0108013) Meneghetti, M., Yoshida, N., Bartelmann, M., Moscardini, L., Springel, V., Tormen, G., & White S. D. M. 2001, , 325, 435 Mo, H. J., Mao, S., & White, S. D. M. 1998, , 295, 319 Molikawa, K., & Hattori, M. 2001, , 559, 544 Moore, B., Quinn, T., Governato, F., Stadel, J., & Lake, G. 1999, , 310, 1147 Mortlock, D. J., & Webster, R. L. 2000, , 319, 872 Nagamine, K., Fukugita, M., Cen, R., & Ostriker, J. P. 2001, , 327, L10 Nakamura, T. T., & Suto, Y. 1997, Prog. Theor. Phys., 97, 49 Narayan, R., & White, S. D. M. 1988, , 231, 97P Navarro, J. F., Frenk, C. S., & White, S. D. M. 1996, , 462, 563 Navarro, J. F., Frenk, C. S., & White, S. D. M. 1997, , 490, 493 Oguri, M. 2002, , 573, 51 Oguri, M., Taruya, A., & Suto, Y. 2001, , 559, 572 Oguri, M., Taruya, A., Suto, Y., & Turner, E. L. 2002, , 568, 488 Phillips, P. M., et al. 2001, , 328, 1001 Porciani, C., & Madau, P. 2000, , 532, 679 Power, C., Navarro, J. F., Jenkins, A., Frenk, C. S., White, S. D. M., Springel, V., Stadel, J., & Quinn, T. 2002, , submitted (astro-ph/0201544) Press, W. H., & Schechter, P. 1974, , 187, 425 Rix, H.-W., de Zeeuw, P. T., Cretton, N., van der Marel, R. P., & Carollo, C. M. 1997, , 488, 702 Rusin, D. 2002, , 572, 705 Rusin, D., & Ma, C.-P. 2001, , 549, L33 Rusin, D., & Tegmark, M. 2001, , 553, 709 Sarbu, N., Rusin, D., & Ma, C.-P. 2001, , 561, L147 Schneider, P., Ehlers, J., & Falco, E. E. 1992, Gravitational Lenses (New York: Springer) Seljak, U. 2002a, , in press (astro-ph/0201450) Seljak, U. 2002b, , submitted (astro-ph/0203117) Sheth, R. K., & Tormen, G. 1999, , 308, 119 Shimasaku, K. 1993, , 413, 59 Takahashi, R., & Chiba, T. 2001, , 563, 489 Turner, E. L. 1980, , 242, L135 Turner, E. L. 1990, , 365, L43 Turner, E. L., Ostriker, J. P., & Gott, J. R. 1984, , 284, 1 Walsh, D., Carswell, R. F., & Weymann, R. J. 1979, , 279, 381 Wambsganss, J., Cen, R., & Ostriker, J. P. 1998, , 494, 29 Wambsganss, J., Cen, R., Ostriker, J. P., & Turner, E. L. 1995, Science, 268, 274 Williams, L. L. R., Navarro, J. F., & Bartelmann, M. 1999, , 527, 535 Wyithe, J. S. B., Turner, E. L., & Spergel, D. N. 2001, , 555, 504 Yoshikawa, K., Taruya, A., Jing, Y. P., & Suto, Y. 2001, , 558, 520 Zhao, H. S. 1996, , 278, 488 [^1]: Kochanek, C. S., Falco, E. E., Impey, C., Lehar, J., McLeod, B., & Rix, H.-W., http://cfa-www.harvard.edu/castles/	{ "pile_set_name": "ArXiv" }
--- abstract: 'We examine the effects of significant electron anti-neutrino fluxes on hydrogen burning. Specifically, we find that the bottleneck weak nuclear reactions in the traditional pp-chain and the hot CNO cycle can be accelerated by anti-neutrino capture, increasing the energy generation rate. We also discuss how anti-neutrino capture reactions can alter the conditions for break out into the $rp$-process. We speculate on the impact of these considerations for the evolution and dynamics of collapsing very- and super- massive compact objects.' author: - 'Chad T. Kishimoto and George M. Fuller' title: 'Neutrino-Accelerated Hot Hydrogen Burning' --- Introduction ============ Hydrogen burning involves the conversion of four protons into an alpha particle, two positrons, neutrinos and photons. The principal bottleneck involved in this process is the weak interaction conversion of protons into neutrons. For decades the primary mechanisms of hydrogen burning have been an astronomical staple. @bet38 first elucidated the proton-proton chain (pp-chain) where the weak conversion is accomplished by two protons interacting to become a deuteron, $p ( p, \nu_e e^+ ) d$. @vwe37 [@vwe38] and @bet39 independently described the CNO cycle, where carbon is used as a catalyst in hydrogen burning, and the weak conversion of protons to neutrons occurs through the positron decay of isotopes of oxygen with half lives of about 100 seconds. A large flux of electron anti-neutrinos ($\bar\nu_e$) could alter the hydrogen burning paradigm. Anti-neutrino capture could perform the necessary conversion of protons to neutrons. The $\bar\nu_e$-capture cross sections of relevance are very small, but depend strongly on neutrino energy. The smallness of these cross sections allows energetic neutrinos to escape from deep within a compact object, where the temperature and other energy scales are high, and freely stream to where hydrogen burning is occurring. Nevertheless, if $\bar\nu_e$-capture is to have a significant effect on hot hydrogen burning, a truly prodigious flux ($\phi_{\bar\nu_e} \gtrsim 10^{40} ~\mbox{cm}^{-2} ~\mbox{s}^{-1}$) and large neutrino energy ($\langle E_{\bar\nu_e} \rangle \gtrsim \mbox{a few MeV}$) would be necessary. It should be kept in mind, however, that to affect hydrogen burning, the $\bar\nu_e$-capture rates need only be comparable to the corresponding positron decay rates. The difficulty would be to find an environment capable of producing these fluxes of neutrinos, yet quiescent enough that simple hydrogen burning could be relevant, and the products of such burning could be ejected into space. High entropy electron-positron plasmas are efficient engines for the production of neutrinos and anti-neutrinos of all flavors. Possible environments that may merit future investigations into the effects of anti-neutrino capture on hydrogen burning include high mass accretion disks and collapsing very- and super- massive objects. In this paper, we investigate the effects of a prodigious neutrino flux on hot hydrogen burning. In section \[burnmech\] we point out the effects of anti-neutrino capture on the rate limiting steps in both the pp-chain and the $\beta$-limited CNO cycle, and its implications for the energy generation rates. In section \[consequences\], we examine the consequences for the $rp$-process and energy generation mechanisms. In section \[smo\], we consider the case of a supermassive star collapsing on the general relativistic Feynman-Chandrasekhar instability, and the effects of its internal neutrino production on hydrogen burning in its envelope. We give conclusions in section \[conclusions\]. Neutrino-Induced Hydrogen Burning Mechanisms {#burnmech} ============================================ The rate limiting step in the pp-chain is the weak interaction conversion of two protons into a deuteron, a positron, and an electron neutrino. A significant flux of electron anti-neutrinos allows an alternate mechanism to be favored, where anti-neutrino capture on a proton creates a neutron and a positron [$\bar\nu_e + p \rightarrow n + e^+$ has been considered in supermassive objects by @woo77 and; @ful97]. This step would be followed by a fast radiative proton capture to form a deuteron. Comparing the two reaction rates: $p( p, \nu_e e^+) d$ vs. $p( \bar\nu_e, e^+ ) n( p, \gamma ) d$, we find that for the prodigious anti-neutrino fluxes discussed in the introduction ($\phi_{\bar\nu_e} \gtrsim 10^{40} ~\mbox{cm}^{-2} ~\mbox{s}^{-1}$, $\langle E_{\bar\nu_e} \rangle \gtrsim \mbox{a few MeV}$) the anti-neutrino capture path is significantly faster in relevant astrophysical environments. This provides not only a new reaction path for hydrogen burning, but increases the energy generation rate by several orders of magnitude. The $\beta$-limited CNO cycle, or hot CNO cycle, proceeds at a rate dictated by the positron decay of [$^{14}\mathrm{O}$]{} and [$^{15}\mathrm{O}$]{}, with half lives of $71\,{\rm s}$ and $122\,{\rm s}$, respectively. [See for example @hoy65 and; @aud73 .] These decays likewise could be augmented by electron anti-neutrino capture, ${\ensuremath{^{14}\mathrm{O}}}(\bar\nu_e, e^+){\ensuremath{^{14}\mathrm{N}}}$ and ${\ensuremath{^{15}\mathrm{O}}}(\bar\nu_e, e^+){\ensuremath{^{15}\mathrm{N}}}$. Figure \[figure1\] shows the acceleration of the relevant weak rates as a function of total electron anti-neutrino flux for an assumed Fermi-Dirac $\bar\nu_e$-energy spectrum with average $\bar\nu_e$-energy $\langle E_{\bar\nu_e} \rangle = 10 ~\mbox{MeV}$ and zero chemical potential. The flux at which anti-neutrino capture becomes important scales appoximately as $\langle E_{\bar\nu_e} \rangle^{-2}$. For a large enough flux ($\phi_{\bar\nu_e} \gtrsim 10^{39} ~\mbox{cm}^{-2} ~\mbox{s}^{-1}$ in the case of Figure \[figure1\]), the reaction rates are proportional to the incident flux of electron anti-neutrinos. Our weak rate calculations are described in Appendix \[ratecalc\]. Additionally, the CNO cycle is accelerated by the presence of free neutrons. The strong interaction reactions ${\ensuremath{^{15}\mathrm{O}}} (n, p) {\ensuremath{^{15}\mathrm{N}}}$ and ${\ensuremath{^{14}\mathrm{O}}} (n, p) {\ensuremath{^{14}\mathrm{N}}}$ have a significantly larger cross section than the electromagnetic reaction $n (p, \gamma) d$. As a result, neutrons are diverted from the modified pp-chain into the CNO cycle. Figure \[figure4\] shows how the neutrons created by $p (\bar\nu_e, e^+) n$ are distributed between the competing reactions ${\ensuremath{^{15}\mathrm{O}}} (n, p) {\ensuremath{^{15}\mathrm{N}}}$, ${\ensuremath{^{14}\mathrm{O}}} (n, p) {\ensuremath{^{14}\mathrm{N}}}$ and $n (p, \gamma) d$. Notice that for an assumed Fermi-Dirac $\bar\nu_e$-energy spectrum with $\langle E_{\bar\nu_e} \rangle = 10 ~\mbox{MeV}$ and zero chemical potential, the ratio of neutron captures on [$^{15}\mathrm{O}$]{} to [$^{14}\mathrm{O}$]{} to $p$ is approximately $4.5 : 2 : 1$ for a large range of $\bar\nu_e$-fluxes. Figure \[figure2\] illustrates the most significant reaction flow paths involved in hydrogen burning when a significant $\bar\nu_e$-flux is present. The pp-chain is modified as anti-neutrino capture allows the circumvention of the slow $p ( p, \nu_e e^+ ) d$ reaction. Also included are the triple-alpha process, which would provide a path between the pp-chain and CNO cycle, and the break out into the $rp$-process via ${\ensuremath{^{15}\mathrm{O}}}( \alpha, \gamma ){\ensuremath{^{19}\mathrm{Ne}}}( p, \gamma ){\ensuremath{^{20}\mathrm{Na}}}$. [@wal81] Side Effects {#consequences} ============ A large flux of electron anti-neutrinos certainly accelerates the weak rates that provide the bottleneck in hot hydrogen burning. However, since this flux also increases the rates of other positron decays, a number of side effects are possible. A principal mechanism for break out into the $rp$-process involves the reaction path ${\ensuremath{^{15}\mathrm{O}}} ( \alpha, \gamma ) {\ensuremath{^{19}\mathrm{Ne}}}( p, \gamma ) {\ensuremath{^{20}\mathrm{Na}}}$. The criteria for break out into the $rp$-process can be found in the competition between proton capture on [$^{19}\mathrm{Ne}$]{}, and the decay of [$^{19}\mathrm{Ne}$]{} through positron emission and now, anti-neutrino capture. Thus for densities and temperatures that satisfy the inequality $$\label{rpbreakout} \rho X \lambda_{p \gamma} ({\ensuremath{^{19}\mathrm{Ne}}}) > \lambda_{w} ({\ensuremath{^{19}\mathrm{Ne}}}),$$ break out into the $rp$-process will occur [@wal81]. Here the density $\rho$ is in $\mbox{g} \,\mbox{cm}^{-3}$, $X$ is the hydrogen mass fraction, $\lambda_{w} ({\ensuremath{^{19}\mathrm{Ne}}})$ is the total weak decay rate of [$^{19}\mathrm{Ne}$]{} (positron emission and $\bar\nu_e$-capture), and $\lambda_{p\gamma} = N_A \langle \sigma v \rangle_{p\gamma}$, where $N_A$ is Avagadro’s number and the thermally averaged product of cross section and speed is taken from @cf88. Including a large flux of electron anti-neutrinos would result in higher weak decay rates ($\lambda_{w}$). This increase in the right hand side of equation (\[rpbreakout\]) would require an increase in temperature (increasing $\lambda_{p\gamma}({\ensuremath{^{19}\mathrm{Ne}}})$) for a given density at which break out into the $rp$-process would occur. Figure \[figure3\] shows the effects of an electron anti-neutrino flux on the conditions necessary for break out into the $rp$-process. The pp-chain is the dominant process of energy generation in the sun, while the CNO cycle is dominant in stars that are more massive. However, with a large flux of electron anti-neutrinos, these processes become independent of temperature so long as the temperature is high enough to guarantee that proton capture remains comparatively fast. A high flux of electron anti-neutrinos allows the pp-chain to compete favorably with the CNO cycle. For example, @pru05 have studied nucleosynthesis in supernova winds where hydrogen “burning” is completely dominated by $\nu_e$ and $\bar\nu_e$ capture on free nucleons [see also @qian93]. The scarcity of [$^{15}\mathrm{O}$]{} in comparison to free protons means that for large anti-neutrino fluxes and average energies, the pp-chain is the dominant mechanism in hydrogen burning at temperatures that the CNO cycle would traditionally dominate. Figure \[figure5\] shows a comparison between the energy generation rates of the pp-chain and the CNO cycle for $X / Z' = 10$ and $100$, where $X$ is the hydrogen mass fraction and $Z'$ is the mass fraction in carbon, nitrogen, and oxygen isotopes. For large anti-neutrino fluxes and average energies, the pp-chain is the dominant energy generation mechanism, while for low fluxes and average energies the CNO cycle takes over. Example: Supermassive Stars {#smo} =========================== Now we consider the case of a supermassive star, a star so massive that it collapses on the general relativistic Feynman-Chandrasekhar instability ($M \gtrsim 5 \times 10^4 M_\odot$) [@fww86 hereafter FWW]. If such objects did exist, for example in the early universe, their homologous cores would emit copious fluxes of neutrinos and anti-neutrinos of all flavors during their collapse [@shi98]. Shi and Fuller examined the collapsing core of a supermassive star, calculating the luminosity and energy spectrum of neutrinos emitted. The total neutrino luminosity was found to be $$L_\nu \approx 2.8 \times 10^{57} \left( {\ensuremath{M_5^{\mathrm{HC}}}}\right)^{-1.5} ~\mbox{erg} ~\mbox{s}^{-1},$$ where [$M_5^{\mathrm{HC}}$]{} is the mass of the homologous core in units of $10^5 M_\odot$. Additionally they found the energy spectrum of neutrinos of all flavors to fit remarkably well to a Fermi-Dirac spectrum with a higher temperature than the central plasma temperature ($T_\nu \approx 1.6 T$) and a degeneracy parameter (chemical potential divided by temperature) $\eta_\nu \approx 2$. We can check the effect of this flux of neutrinos an antineutrinos on the nuclear physics in the gas in the envelope of the star. As a point of reference, we choose a radius of 100 Schwarzschild radii ($r = 3 \times 10^{12} {\ensuremath{M_5^{\mathrm{HC}}}}~\mbox{cm}$) where the gravitational binding energy of a nucleon is approximately equal to the nuclear energy liberated in these reactions, so there is a chance that any new nuclear physics that occurs as a result of $\bar\nu_e$-capture could be relevant. By “relevant” we mean that it is conceivable that material from this location could avoid being swallowed by the black hole forming in the core. Only a detailed simulation with general relativistic hydrodynamics could reveal whether or not material affected by $\bar\nu_e$-capture is ever ejected into space. We are now free to repeat the analyses done above, but with one free parameter, the mass of the collapsing homologous core. Figure \[figure6\] shows the acceleration of the relevant weak rates as a function of homologous core mass. The anti-neutrino capture rate is proportional to $({\ensuremath{M_5^{\mathrm{HC}}}})^{-4}$. We see that the effects of anti-neutrino capture on the decay rates of [$^{14}\mathrm{O}$]{} and [$^{15}\mathrm{O}$]{} become insignificant for large homologous core masses (${\ensuremath{M_5^{\mathrm{HC}}}}\gtrsim 0.4$). If supermassive stars ever formed, it is conceivable that they were in the first generation of stars with primordial initial abundances. In this case the CNO cycle would be negligible, at least initially . However, even in this case, the energy generation rate of the pp-chain would be boosted by several orders of magnitude. It would be interesting to see if this added energy source would have a discernible effect on the eventual fate of a collapsing supermassive star. Including a hydrogen burning phase in the final stages of the collapse of a supermassive star may affect the eventual fate of its baryons. Hydrodynamic, post-Newtonian calculations done in FWW show that initially metal-free supermassive stars will collapse to black holes. @shi02 use a fully relativistic simulation in axial symmetry to deduce that the supermassive star collapses to a black hole surrounded by some remaining gas in an ambient disk. The principal nucleosynthetic issue is whether any material that had experienced $\bar\nu_e$ capture-affected hydrogen burning escapes being incorporated into a black hole. Of course, there is the prior issue of whether material at the relatively low temperatures and densities which characterize hydrogen burning ever experiences high $\bar\nu_e$-fluxes. Both issues are related: to see nucleosynthesis products of $\bar\nu_e$ capture-affected hydrogen burning, the material must be ejected before the point at which nuclear burning proceeds past simple hydrogen burning and approaches or attains nuclear statistical equilibrium (NSE). We are skeptical that these conditions can be met. Fully relativistic numerical simulations could settle these issues. Obviously, the NSE nucleosynthetic yield is uninteresting in the context of this paper. However, a mass shedding scenario could be conducive to conditions that favor hydrogen burning and the $rp$-process. Speeding up weak decays could affect the relative abundances of the $rp$-process elements. A simulation that follows these species and their chemical reactions would be necessary to address this issue. Conclusions =========== In this paper we have examined the effects of a prodigious flux of electron anti-neutrinos on hydrogen burning. We have found that the traditional positron decay bottlenecks in hydrogen burning can be removed and replaced by much faster $\bar\nu_e$-capture reactions under some conditions. This would result in an increase of several orders of magnitude in the energy generation rate over what would be expected without such a flux. Additionally, the $\bar\nu_e$-flux would alter the conditions necessary for break-out into the $rp$-process, increasing the temperature necessary to do so at a given density. If conditions allow the break-out into the $rp$-process, we could expect an acceleration of the flow toward the iron-peak facilitated by and accelerated by $\bar\nu_e$-capture. When applied to the neutrino flux emitted in the final stages of the collapse of a supermassive star, interesting changes from current simulations may occur on the lower end of the supermassive star mass spectrum. Whether or not these effects are relevant, remains an open question that can only be answered by simulations that are able to include hydrogen burning during the final collapse of the star. Important issues that remain open include finding an astrophysical environment where the effects discussed here could take place. Accretion disks surrounding black holes may provide a combination of high accretion rates and hot, high entropy disks which could produce the necessary fluxes of electron anti-neutrinos. (See for example @sur05 for a discussion of neutrino emission in lower mass accretion disks.) Supermassive stars may exhibit this effect, though there is uncertainty related to whether or not these objects ever existed. Computer simulations would be useful to determine any changes in the expected nucleosynthetic yield, and the effects of their possible distribution into the surrounding IGM. We would like to thank S. E. Woosley, Y.-Z. Qian and A. Heger for useful discussions. This work was supported in part by NSF grant PHY-04-00359 and the TSI collaboration’s DOE SciDAC grant at UCSD. C.T.K. would like to acknowledge a fellowship from the ARCS Foundation, Inc. Calculation of Weak Rates {#ratecalc} ========================= In this work we calculate the $\bar\nu_e$-capture rates in the manner described in @ffn1 [@ffn2] and @fm85. We employ measured discrete states only. Our ${\ensuremath{^{14}\mathrm{O}}} (\bar\nu_e, e^+) {\ensuremath{^{14}\mathrm{N}}}$ rate calculation includes only the [$^{14}\mathrm{O}$]{} ground state (spin and parity $J^\pi = 0^+$) and the measured weak branches to the [$^{14}\mathrm{N}$]{} ground state ($J^\pi = 0^+$, $\log_{10} ft = 7.3$), first excited state ($J^\pi = 1^+$, $\log_{10} ft = 3.5$), and second excited state ($J^\pi = 1^+$, $\log_{10} ft = 3.1$). Contributions to the stellar rate from thermal excitation of parent states are small here as a result of the high first excited state excitation energy ($5.17 ~\mbox{MeV}$) and the temperatures of interest. Likewise, branches to higher excited states in [$^{14}\mathrm{N}$]{} are not significant. A possible exception is the first isobaric analog state in [$^{14}\mathrm{N}$]{} ($J^\pi = 0^+$) at excitation energy of $8.62 ~\mbox{MeV}$. This branch will have a large matrix element but will be $Q$-value-hindered relative to the $0^+ \rightarrow 0^+$ ground state to first excited state, pure Fermi branch. Our ${\ensuremath{^{15}\mathrm{O}}} (\bar\nu_e, e^+) {\ensuremath{^{15}\mathrm{N}}}$ rate calculation includes only the ground state ($J^\pi = 1/2^-$) branch. This channel has a large matrix element, corresponding to $\log_{10} ft = 3.6$. Branches to [$^{15}\mathrm{N}$]{} excited states will not be significant. [$^{15}\mathrm{N}$]{} states below $9.15 ~\mbox{MeV}$ excitation energy have positive parity and the branches from the [$^{15}\mathrm{O}$]{} ground state to them will be forbidden. We note, however, that [$^{15}\mathrm{O}$]{} and [$^{15}\mathrm{N}$]{} are isospin mirrors. This can be a significant fact for stellar weak interaction rates, as it implies large Fermi and Gamow-Teller matrix elements coupling each parent state with its daughter isobaric analog state . Thermal excitation of [$^{15}\mathrm{O}$]{} excited states would open weak branches to corresponding isobaric analog states in [$^{15}\mathrm{N}$]{}. This is not likely at the temperatures of interest because the first excited state of [$^{15}\mathrm{O}$]{} is at about $5.2 ~\mbox{MeV}$ excitation. Our calculation of the ${\ensuremath{^{19}\mathrm{Ne}}}( \bar\nu_e, e^+ ) {\ensuremath{^{20}\mathrm{Na}}}$ rate includes only the ground state of [$^{19}\mathrm{Ne}$]{} ($J^\pi = 1/2^+$) and branches to the ground ($J^\pi = 1/2^+$) and third excited state ($J^\pi = 3/2^+$) of [$^{19}\mathrm{F}$]{}. The first of these branches, with $\log_{10} ft = 3.2$, dominates the rate. We note, however, that [$^{19}\mathrm{Ne}$]{} and [$^{19}\mathrm{F}$]{} are isospin mirrors. Since temperatures are high near CNO cycle breakout, thermal excitation of the first ($J^\pi = 5/2^+$) and second ($J^\pi = 1/2^-$) excited states can be expected to carry a fraction of the total weak rate. However, on the assumption that the matrix elements for these branches are identical to that for the ground-to-ground transitions, inclusion of these branches makes little difference ($< 1\%$) for the rates and our conclusions. Audouze, J., Truran, J. W., and Zimmerman, B. A. 1973, , 184, 493 Bethe, H. A. & Critchfield, C. L. 1938, Phys. Rev., 54, 248 Bethe, H. A. 1939, Phys. Rev., 55, 434 Caughlan, G. R. & Fowler, W. A. 1988, ADNDT, 40, 283 Fuller, G. M., Fowler, W. A., & Newman, M. J. 1980, , 42, 447 (FFN I) Fuller, G. M., Fowler, W. A., & Newman, M. J. 1982, , 252, 715 (FFN II) Fuller, G. M. & Meyer, B. S 1995, , 453, 792 Fuller, G. M. & Shi, X. 1997, , 487, L25 Fuller, G. M., Woosley, S. E., & Weaver, T. A. 1986, , 307, 675 (FWW) Hoyle, F. & Fowler, W. A. 1965, in Quasi-Stellar Sources and Gravitational Collapse, ed. I. Robinson, A. Schild, & E. L. Schucking (Chicago: University of Chicago Press), 62 Pruet, J., Hoffman, R. D., Woosley, S. E., Janka, H.-T., & Buras, R. 2005, astro-ph/0511194 Qian, Y.-Z., Fuller, G. M., Matthews, G. J., Mayle, R. W., Wilson, J. R., & Woosley, S. E. 1993, PRL, 71, 1965 Shi, X. & Fuller, G. M. 1998, , 503, 307 Shibata, M. & Shapiro, S. 2002, , 572, L39 Surman, R., McLaughlin, G. C., & Hix, W. R. 2005, astro-ph/0509365 von Weizsäcker, C. F. 1937, Physik. Zeitschr., 38, 176 von Weizsäcker, C. F. 1938, Physik. Zeitschr., 39, 633 Wallace, R. K. & Woosley, S. E. 1981, , 45, 389 Woosley, S. E. 1977, Nature, 269,42	{ "pile_set_name": "ArXiv" }
--- abstract: 'The notion of almost periodicity nontrivially generalizes the notion of periodicity. Strongly almost periodic sequences (=uniformly recurrent infinite words) first appeared in the field of symbolic dynamics, but then turned out to be interesting in connection with computer science. The paper studies the class of eventually strongly almost periodic sequences (i. e., becoming strongly almost periodic after deleting some prefix). We prove that the property of eventual strong almost periodicity is preserved under the mappings done by finite automata and finite transducers. The class of almost periodic sequences includes the class of eventually strongly almost periodic sequences. We prove this inclusion to be strict.' author: - Yuri Pritykin title: \| Strongly Almost Periodic Sequences under\ Finite Automata Mappings --- Introduction ============ Strongly almost periodic sequences (=uniformly recurrent infinite words) were studied in the works of Morse and Hedlund [@Symb01; @Symb02] and of many others (for example see [@Cass]). This notion first appeared in the field of symbolic dynamics, but then turned out to be interesting in connection with computer science. Evidently, the class of finite automata mappings of strongly almost periodic sequences (for definitions see below) contains the class of eventually strongly almost periodic sequences, i. e., becoming strongly almost periodic after deleting some prefix. Indeed, we use finite automaton with delay to get the sequence $a\omega$ from the strongly almost periodic sequence $\omega$: this automaton keeps the string $a$ in memory, first outputs this string and then outputs the input sequence with delay $\|a\|$ (always remembering last $\|a\|$ symbols of the sequence). The main result of the article (Theorem \[finiteAutomataStrong\]) states the equality of the classes. In other words, Theorem \[finiteAutomataStrong\] says that finite automata preserve the property of eventual strong almost periodicity. In the last section we consider the generalized version of finite automaton — finite transducer — and prove the same statement for it. The notion of almost periodic sequence was studied in [@AlPer]. The authors prove that the class of almost periodic sequences is also closed under finite automata mappings. It can easily be checked that the class of almost periodic sequences contains the class of eventually strongly almost periodic sequences. We prove this inclusion to be strict (Theorem \[alPer-NotEssAlPer\]). Let $A$ be a finite alphabet with at least two symbols. Consider the sequences over this alphabet — mappings $\omega\colon{\mathbb{N}}\to A$ (where ${\mathbb{N}}=\{0, 1, 2,\dots\}$). Denote by $A^$ the set of all finite strings over $A$ including the empty string $\Lambda$. If $i\le j$ are natural, denote by $[i,j]$ the segment of natural numbers with ends in $i$ and $j$, i. e., the set $\{i, i+1, i+2,\dots,j\}$. Also denote by $\omega[i,j]$ the segment of the sequence $\omega$ — the string $\omega(i)\omega(i+1)\dots\omega(j)$. A segment $[i,j]$ is an occurrence of a string $x\in A^$ in a sequence $\omega$ if $\omega[i,j]=x$. Denote by $\|x\|$ the length of the string $x$. We imagine the sequences going horizontally from the left to the right, so we use terms “to the right” or “to the left” to talk about greater and smaller indices respectively. Almost periodicity ================== A sequence $\omega$ is called almost periodic if for any string $x$ occurring in the sequence infinitely many times there exists a number $l$ such that any segment of $\omega$ of length $l$ contains at least one occurrence of $x$. We denote the class of these sequences by ${\mathcal{AP}}$. A sequence $\omega$ is called strongly almost periodic if for any string $x$ occurring in the sequence at least once there exists a number $l$ such that any segment of $\omega$ of length $l$ contains at least one occurrence of $x$ (and therefore $x$ occurs in $\omega$ infinitely many times). Denote by ${\mathcal{SAP}}$ the class of these sequences. For convenience we introduce one additional definition: a sequence $\omega$ is eventually strongly almost periodic if $\omega=a\nu$ for some $\nu\in{\mathcal{SAP}}$ and $a\in A^$. The class of these sequences we denote by ${\mathcal{EAP}}$. Every eventually strongly almost periodic sequence is obviously almost periodic. Let us show that ${\mathcal{EAP}}$ is a proper subclass of ${\mathcal{AP}}$. \[alPer-NotEssAlPer\] There exists a binary sequence $\omega$ such that $\omega\in{\mathcal{AP}}$, but $\omega\notin{\mathcal{EAP}}$. Construct a sequence of binary strings $a_0=1$, $a_1=10011$,\ $a_2=1001101100011001001110011$, and so on, by this rule: $$a_{n+1}=a_n\bar a_n\bar a_na_na_n,$$ where $\bar x$ is a string obtained from $x$ by changing every 0 to 1 and vice versa. Put $$c_n=\underbrace{a_na_n\dots a_n}_{10}$$ and $$\omega=c_0c_1c_2c_3\dots$$ Prove that $\omega$ is a required one. The length of $a_n$ is $5^n$, so the length of $c_0c_1\dots c_{n-1}$ is $10(1+5+\dots+5^{n-1})=\frac52(5^n-1)$. By definition, put $$l_n=\frac52(5^n-1)=\|c_0c_1\dots c_{n-1}\|.$$ Show that $\omega$ is almost periodic. Suppose $x\ne\Lambda$ occurs in $\omega$ infinitely many times. Take $n$ such that $\|x\|<5^n$. Suppose $[i,j]$ is an occurrence of $x$ in $\omega$ such that $i\ge l_n$. By construction, for any $k$ we can consider the part of $\omega$ starting from the position $l_k$ as a concatenation of strings $a_k$ and $\bar a_k$. Thus by definition of $i$ the string $x$ is a substring of one of four strings $a_na_n$, $a_n\bar a_n$, $\bar a_na_n$, $\bar a_n\bar a_n$. Note that the string 10011 contains all strings of length two 00, 01, 10, 11, so the string $a_{n+1}$ contains each of $a_na_n$, $a_n\bar a_n$, $\bar a_na_n$, $\bar a_n\bar a_n$. So, $x$ is a substring of $a_{n+1}$. Similarly, $x$ is a substring of $\bar a_{n+1}$. If the segment of length $2\|a_{n+1}\|$ occurs in the sequence to the right of $l_{n+1}$, then $a_{n+1}$ or $\bar a_{n+1}$ occurs in this segment. Hence for $l=\frac52(5^{n+1}-1)+2\cdot5^{n+1}$ it is true, that on every segment of length $l$ there exists an occurrence of $x$. Prove now that for any $n>0$ the string $c_n$ does not occur in $\omega$ to the right of $l_{n+1}$. In this case, $c_n$ occurs finitely many times in the sequence obtained from $\omega$ by deleting some prefix of the length at most $l_n$, i. e., this sequence is not strongly almost periodic. Therefore $\omega$ is not eventually strongly almost periodic. Let $\nu$ be the sequence obtained from $\omega$ by deleting the prefix of the length $l_{n+1}$. As above, for each $k$, $1\le k\le n+1$, $\nu$ is a concatenation of strings $a_k$ and $\bar a_k$. Assume $c_n$ occurs in $\nu$ and let $[i,j]$ be one of this occurrences. For $n>0$ the string $c_n$ begins with $a_1$, hence $[i,i+4]$ is an occurrence of $a_1$ in $\nu$. We see that $a_1=10011$ occurs in $a_1a_1=1001110011$, $a_1\bar a_1=1001101100$, $\bar a_1a_1=0110010011$ or $\bar a_1\bar a_1=0110001100$ only in zeroth or fifth position. Thus $5\|i$, i. e., $\nu$ and $c_n$ can be considered as constructed of “letters” $a_1$ and $\bar a_1$, and we assume that $c_n$ occurs in $\nu$. Now it is easy to prove by induction on $m$ that $5^m\|i$ for $1\leqslant m\leqslant n$, i. e., we can consider $\nu$ and $c_n$ as constructed of “letters” $a_m$ and $\bar a_m$, and assume that $c_n$ occurs in $\nu$ (the base for $m=1$ is already proved, but we can repeat the same argument changing 1 and 0 to $a_m$ and $\bar a_m$ and taking into account that $c_n$ begins with $a_m$ for each $1\le m\le n$). Therefore we have shown that $5^n\|i$, i. e., if we consider $\nu$ and $c_n$ to be constructed of “letters” $a_n$ and $\bar a_n$, then $c_n=\underbrace{a_na_n\dots a_n}_{10}$ occurs in $\nu$. However there exists an occurrence of “five-letter” string $a_{n+1}$ or $\bar a_{n+1}$ on each segment of 10 consequent “letters” $a_n$ and $\bar a_n$ in $\nu$, and $\bar a_n$ occurs in this string. This is a contradiction. Moreover, it is quite easy to modify the proof and to construct continuum of sequences in ${\mathcal{AP}}\setminus{\mathcal{EAP}}$. For example, for each sequence $\tau\colon{\mathbb{N}}\to\{9,10\}$ we can construct $\omega_\tau$ in the same way as in the proof of Theorem \[alPer-NotEssAlPer\], but instead of $c_n$ we take $$c_n^{(\tau)}=\underbrace{a_na_n\dots a_n}_{\tau(n)}.$$ Obviously, all $\omega_\tau$ are different for different $\tau$ and hence there exists continuum of various $\tau$. In conclusion, we can remark that, as it is shown in [@AlPer], there exists continuum of different sequences in ${\mathcal{SAP}}$ too. Finite automata mappings ======================== Finite automaton* is a tuple $F = \langle A, B, Q, \tilde q, f\rangle$, where $A$ and $B$ are finite sets called input and output alphabets respectively, $Q$ is a finite set of states, $\tilde q\in Q$ is the initial state, and $f\colon Q\times A\to Q\times B$ is the translation function. We say that the sequence $\langle p_n, \beta(n)\rangle_{n=0}^\infty$, where $p_n\in Q$, $\beta(n)\in B$, is the automaton mapping of the sequence $\alpha$ over alphabet $A$, if $p_0=\tilde q$ and for each $n$ we have $\langle p_{n+1}, \beta(n)\rangle = f(p_n, \alpha(n))$. Thus we say that $F$ outputs the sequence $\beta$ and denote this output by $F(\alpha)$. If $[i,j]$ is an occurrence of the string $x$ in the sequence $\alpha$, and $p_i=q$, then we say that automaton $F$ comes to this occurrence of $x$ in the state $q$. In [@AlPer] the following statement was proved: if $F$ is a finite automaton, and $\omega\in{\mathcal{AP}}$, then $F(\omega)\in{\mathcal{AP}}$. This theorem can be expanded. \[finiteAutomataStrong\] If $F$ is a finite automaton, and $\omega\in{\mathcal{EAP}}$, then $F(\omega)\in{\mathcal{EAP}}$. Obviously, it is enough to prove the theorem for $\omega\in{\mathcal{SAP}}$, as every eventually strongly almost periodic sequence keeps this property after addition a prefix. Thus let $\omega\in{\mathcal{SAP}}$. By the previous statement, $F(\omega)\in{\mathcal{AP}}$. Suppose $F(\omega)$ is not eventually strongly almost periodic. It means that for any natural $N$ there exists a string that occurs in $F(\omega)$ after position $N$ and does not occur to the right of it. Indeed, if we remove the prefix $[0,N]$ from $F(\omega)$, we do not get strongly almost periodic sequence, hence there exists a string occurring in this sequence only finitely many times. Then take the rightmost occurrence. Let $[i_0,j_0]$ be the rightmost occurrence of a string $y_0$ in $F(\omega)$. For some $l_0$ the string $x_0=\omega[i_0,j_0]$ occurs in every segment of the length $l_0$ in $\omega$ (by the property of strong almost periodicity). If $F$ comes to $i_0$ in the state $q_0$, then $F$ never comes to righter occurrences of $x$ in the state $q_0$ because in this case automaton outputs $y_0$ completely. Now let $[r,s]$ be the rightmost occurrence of some string $a$ in $F(\omega)$, where $r>i_0+l_0$. On the segment $\omega[r-l_0,r]$ there exists an occurrence $[r',s']$ of the string $x_0$. By definition of $r$ we have $r'>i_0$. Thus assume $$i_1=r',\ j_1=s,\ x_1=\omega[i_1,j_1],\ y_1=F(\omega)[i_1,j_1].$$ Since $a$ does not occur in $F(\omega)$ to the right of $r$, then $y_1$ does not occur in $F(\omega)$ to the right of $i_1$, for it contains $a$ as a substring. Therefore if the automaton comes to the position $i_1$ in the state $q_1$, then it never comes to righter occurrences of $x_1$ in the state $q_1$. Since $x_1$ begins with $\omega[r',s']=x_0$, and $r'>i_0$, we get $q_1\ne q_0$. We have found the string $x_1$ such that automaton $F$ never comes to occurrences of $x_1$ to the right of $i_1$ in the state $q_0$ or $q_1$. Let $m=\|Q\|$. Arguing as above, for $k<m$ we construct the strings $x_k=\omega[i_k,j_k]$ and corresponding different states $q_k$, such that $F$ never comes to occurrences of $x_k$ in $\omega$ to the right of $i_k$ in the states $q_0,q_1,\dots,q_k$. For $k=m$ we have a contradiction. Finite transducers ================== Let $A$ and $B$ be finite alphabets. The mapping $h\colon A^\to B^$ is called homomorphism, if for any $u,v\in A^$ we have $h(uv)=h(u)h(v)$. Clearly, any homomorphism is fully determined by its values on one-character strings. Let $\omega$ be the sequence over the alphabet $A$. By definition, put $$h(\omega)=h(\omega(0))h(\omega(1))h(\omega(2))\dots$$ Suppose $h\colon A^\to B^$ is a homomorphism, $\omega$ is an almost periodic sequence over $A$. In [@AlPer] it was shown that if $h(\omega)$ is infinite, then it is almost periodic. Thus it is obvious, that if $\omega$ is strongly almost periodic, and $h(\omega)$ is infinite, then $h(\omega)$ is also strongly almost periodic. Indeed, it is enough to show that any $v$ occurring in $h(\omega)$ occurs infinitely many times. However there is some string $u$ occurring in $\omega$ such that $h(u)$ contains $v$, but by the definition of strong almost periodicity $u$ occurs in $\omega$ infinitely many times. Evidently, for $\omega\in{\mathcal{EAP}}$ we have $h(\omega)\in{\mathcal{EAP}}$, if $h(\omega)$ is infinite. Now we modify the definition of finite automaton, allowing it to output any string (including the empty one) over output alphabet reading only one character from input. This modification is called finite transducer* (see [@FinTrans]). Formally, we only change the definition of translation function. Now it has the form $$f\colon Q\times A\to Q\times B^.$$ If the sequence $\langle p_n, v_n\rangle_{n=0}^\infty$, where $p_n\in Q$, $v_n\in B^$, is the mapping of $\alpha$, then the output is the sequence $v_0v_1v_2\dots$ Actually, we can decompose the mapping done by finite transducer into two: the first one is a finite automaton mapping and another is a homomorphism. Each of these mappings preserves the class ${\mathcal{AP}}$, so we get the corollary: finite transducers map almost periodic sequences to almost periodic. Similarly, by Theorem \[finiteAutomataStrong\] and arguments above we also get the following statement: finite transducers map eventually strongly almost periodic sequences to eventually strongly almost periodic (if the output is infinite). The author is grateful to Alexei Semenov and Andrej Muchnik for the help in the work and also to the participants of Kolmogorov seminar for useful discussions. [99]{} M. Morse, G. A. Hedlund. Symbolic dynamics. American Journal of Mathematics, 60, pp. 815–866, 1938. M. Morse, G. A. Hedlund. Symbolic dynamics II: Sturmian trajectories. American Journal of Mathematics, 62, pp. 1–42, 1940. A. Weber. On the valuedness of finite transducers. Acta Informatica, 27, pp. 749–780, 1989. J. Cassaigne. Recurrence in infinite words. Proceedings of the 18th Symposium on Theoretical Aspects of Computer Science (STACS 2001), Springer Verlag, 1–11, 2001. An. Muchnik, A. Semenov, M. Ushakov. Almost periodic sequences. Theoretical Computer Science, vol. 304, pp. 1–33, 2003.	{ "pile_set_name": "ArXiv" }
--- author: - 'G. Bruni' - 'K.-H. Mack' - 'F.M. Montenegro-Montes' - 'M. Brienza' - 'J.I. González-Serrano' title: \| Restarting radio activity and dust emission in radio-loud\ broad absorption line quasars --- [Broad absorption line quasars (BAL QSOs) are objects showing absorption from relativistic outflows, with velocities up to 0.2c. These manifest, in about 15% of quasars, as absorption troughs on the blue side of UV emission lines, such as [C[iv]{} ]{}and [Mg[ii]{} ]{}. The launching mechanism and duration of these outflows is not clear yet.]{} [In this work, we complement the information collected in the cm band for our previously presented sample of radio loud BAL QSOs (25 objects with redshifts 1.7<z<3.6) with new observations at m and mm bands. Our aim is to verify the presence of old, extended radio components in the MHz range, and probe the emission of dust (linked to star formation) in the mm domain.]{} [We observed 5 sources from our sample, already presenting hints of low-frequency emission, with the GMRT at 235 and 610 MHz. Other 17 sources (more than half the sample) were observed with bolometer cameras at IRAM-30m (MAMBO2, 250 GHz) and APEX (LABOCA and SABOCA, 350 and 850 GHz, respectively).]{} [All sources observed with the GMRT present extended emission at a scale of tens of kpc. In some cases these measurements allow us to identify a second component in the SED, at frequencies below 1.4 GHz, beyond the one already studied in the GHz domain. In the mm-band, only one source shows emission clearly ascribable to dust, detached from the synchrotron tail. Upper limits were obtained for the remaining targets.]{} [These findings confirm that BAL QSOs can also be present in old radio sources, or even in restarting ones, where favourable conditions for the outflow launching/acceleration are present. A suggestion that these outflows could be precursors of the jet comes from the fact that $\sim$70% of our sample can be considered in a GigaHertz Peaked Spectrum (GPS) or Compact Steep Spectrum (CSS)+GPS phase. This would confirm the idea proposed by other authors that these outflows could be recollimated to form the jet. Comparing with previous works in the literature, dust emission seems to be weaker than the what expected in ‘normal’ QSOs (both radio loud and radio quiet ones) suggesting that a feedback mechanism could inhibit star formation in radio-loud BAL QSOs.]{} Introduction ============ In the context of Active Galactic Nuclei (AGN) outflows from the central region are commonly detected as absorption lines in more bands. In the UV/Optical range, they can be present in $\sim$70% of type 1 AGN, with an extension up to kpc scales and velocities up to $\sim$1000 km/s (@Harrison). A similar fraction ($\sim$60%) has been found in the quasar (QSO) population (@Gan). In the X-ray band, more highly ionized outflows are detected as Ultra Fast Outflows (UFO) both in radio quiet (@Pounds) and radio loud AGN (@Tombesi14) with much higher velocities, in the range 0.03-0.4c. A feedback effect from outflows on the host galaxy has been proven to exist by different authors in the past years (e.g. @Feruglio [@Wang; @Sturm]), and lately they have been proven to hamper star formation (@Tombesi). BAL QSOs are among the objects presenting the fastest outflows. These are detected in about 15% of QSOs as broad absorption troughs in the UV spectrum, on the blue side of emission lines from ionised species, mainly [C[iv]{} ]{}and [Mg[ii]{} ]{}. They can be both detached or superimposed to the emission peak, and reach relativistic velocities of up to 0.2c (@Hewett). [@Allen] have found a dependence with redshift of the BAL fraction, decreasing a factor of 3.5$\pm$0.4 from z$\sim$4 to $\sim$2. The mechanism at the origin of these violent outflows has not been unveiled yet. The two main scenarios discussed in the literature tend to ascribe the BAL phenomenon to: 1) young objects, in which the strong nuclear starburst activity is still expelling a dust cocoon (@Briggs [@Sanders; @Farrah]), or 2) normal QSOs, whose outflows intercept the line of sight of the observer (@Elvis). In this case, relativistic outflows are supposed to be commonly present in QSOs, but detected only when orientation is favourable. Variability of the BAL troughs has been explored by many authors in the past years, thanks to the increasing amount of available spectroscopic surveys data (@Gibson08 [@Gibson10; @Capellupo11; @Capellupo12; @Vivek12]), and a typical duty cycle of about a thousand years for the BAL-producing outflow has been found (@Filiz12 [@Filiz13]). Several works have been published in the past years, trying to collect information in the different electromagnetic bands. In particular, the emission in the radio band has been used to probe the orientation and age of these objects (@Montenegro [@DiPompeo1; @Bruni; @Bruni2]). No clear hints of a favoured scenario arose from these works, resulting in indications of different possible orientations and different ages for radio loud (RL) BAL QSOs. In this work, we present follow-up observations of sources from our previously studied sample (@Bruni). We explored the emission properties at m- and mm-wavelengths, to complement the multi-wavelength view of these objects, already studied at cm-band in our previous work. The detection of a strong MHz emission can be safely interpreted as the presence of old extended radio plasma connected to a former AGN radio-active phase, that can be as old as 10$^7$-10$^8$ years (@Konar06 [@Konar13]). It has been shown indeed that jets in radio loud AGN can have multiple phases of activity (@Lara [@Schoen; @Saikia; @Nandi]) with a duty cycle that depends on the source radio power (@Best2 [@Shabala]). Hints of these components in BAL QSOs were already found from literature data, for the same sample studied in this paper, and presented in [@Bruni]. This kind of emission adds a significant information in the framework of the presented models for BAL QSOs. In particular, if the young-scenario would be the most realistic one, no further components than the one peaking in the MHz-GHz range should be present, since GigaHertz-Peaked Spectrum sources (GPS) and Compact Steep Spectrum sources (CSS, @ODea), together with High Frequency Peakers (HFP, @Dallacasa) are among the youngest radio sources. To date systematic searches of diffuse emission around GPS and CSS sources find about 20% detections (@Stan1 [@Stan2]). In light of this we use low-frequency GMRT observations to probe possible extended emission around BAL QSO with the aim of possibly confirming or discarding the youth scenario described in the previous paragraph. The radio phase itself does not seem to introduce significant differences in RL BAL QSOs with respect to Radio Quiet (RQ) ones (@Bruni3 [@Rochais]), thus being a valid tool to study general phenomenology of these objects. The continuum emission of dust, in the rest-frame far-infrared domain (FIR), can be detected at mm-wavelengths (over 100 GHz) for objects with z$\sim$2. Objects enshrouded by gas and dust can host star-formation regions (@Zahid), and thus show high star-formation rates, that may indicate a young age of the galaxy. A flux density excess in the FIR could be an indicator of a different age for BAL QSOs with respect to the non-BAL QSO population, and thus help in discriminating between the orientation and the evolutionary models. There are two major works presenting (sub)mm observations on samples of BAL QSOs. [@Willott] showed SCUBA measurements on a sample of 30 radio-quiet BAL QSOs and conclude that there is no difference between BAL QSOs and a comparison sample of non-BAL QSOs. Nevertheless, [@Priddey] based on SCUBA observations of 15 BAL QSOs found tentative evidence for a dependence of submm flux densities on the equivalent width of the characteristic [C[iv]{} ]{}BAL which ‘suggests that the BAL phenomenon is not a simple geometric effect \[…\] but that other variables, such as evolutionary phase, \[…\] must be invoked’. [@Cao] discuss the far-infrared properties and star-formation rates (SFR) of BAL QSOs (without distinguishing between RL and RQ ones), using data from the [Herschel]{}-ATLAS project. They found no differences with respect to non-BAL QSOs, concluding that a scenario in which BAL QSOs are objects expelling a dust cocoon is improbable. The main difference of the above samples compared to our target sample is the radio-loudness of our sources. With the radio-data presented in [@Bruni] we were able to characterise the synchrotron spectra of our sources, and thus to study the peak frequency and spectral index distributions with respect to the ‘normal’ QSO population. Also an upper limit to the synchrotron emission at mm-wavelengths can be constrained. A search for HFP, i.e. the youngest known radio sources with the highest turnover frequencies, shows only a very small percentage of sources with peak frequencies close to 20 GHz (@Dallacasa) with the most extreme case at 25 GHz, leading to a formal age of some 50 years only (@Orienti). As any upturn towards an even higher peak frequency would be visible in our Spectral Energy Distributions (SEDs) we can safely assume that the extrapolated synchrotron emission reflects its true contribution at 250 GHz and that any observed excess emission can be attributed to the presence of cold dust. Moreover, the variability study presented in [@Bruni] excludes any possible significant variability even at high frequencies over a 3 years time scale, for this sample of objects. The outline of the paper is as follows: in Sect. \[sec:sample\] we describe the BAL QSO sample. The radio observations are reported in Sect. \[sec:observations\]. In Sect. \[sec:results\] we present the results concerning morphology at MHz frequencies and dust abundance. Sect. \[sec:discussion\] is a discussion of the results, in the context of recent works about BAL QSOs. The cosmology adopted throughout the work assumes a flat universe and the following parameters: $H_0=71$ km s$^{-1}$ Mpc$^{-1}$, $\Omega_{\Lambda}=0.73$, $\Omega_{M}=0.27$. The RL BAL QSO sample {#sec:sample} ===================== The radio-loud BAL QSO sample studied in this paper is presented in [@Bruni]. All sources were chosen among objects from the 4th edition of SDSS Quasar Catalogue (@Schneider07), drawn from the 5th data release of the Sloan Digital Sky Survey (SDSS-DR5; @Adelman). To select radio-loud objects, we cross-matched the SDSS with the FIRST (Faint Images of the Radio Sky at Twenty-cm; @Becker2) and only those with a counterpart lying $<$2 arcsec away and having $S_{1.4\rm{GHz}}$ $>$ 30 mJy were considered. All of these satisfy the radio-loudness definition by [@Stocke]. Moreover, the selection has been limited to those objects whose redshifts lie in the range 1.7 $<z<$ 4.7 allowing the identification of both [C[iv]{} ]{}and [Mg[ii]{} ]{}absorption features on SDSS spectra. In order to select genuine BAL QSOs, only objects with an Absorption Index (AI) $>$ 100 $\rm{km/s}$ were considered, and only troughs broader than 1000 $\rm{km/s}$ were used for this calculation[^1]. This resulted in 25 radio-loud BAL QSOs. For a complete description of the sample and the selection procedure refer to [@Bruni]. \[Summaryofobservations\] Radio observations and data reduction {#sec:observations} ===================================== In this paper, we present observations complementary to the ones performed by [@Bruni] at cm wavelengths. The GMRT, the APEX single-dish and the IRAM 30-m telescope were used to extend the available SEDs extension. Table \[Summaryofobservations\] summarises the different runs and observing setups. Giant Metrewave Radio Telescope ------------------------------- Observations at frequencies of 235 MHz and 610 MHz with the GMRT were performed for five sources, during January 2010. We used the double frequency mode to observe simultaneously at the two frequencies. The total bandwidth for each band was 33 MHz, divided into 256 channels of 0.13 MHz each. We observed in snapshot mode to improve the UV coverage for the sources. Standard phase and amplitude calibration were performed, using 3C 286 as primary calibrator every $\sim$4 hours and suitable phase calibrators near targets every $\sim$30 minutes. Correlation was performed using the GSB software correlator at NCRA. Data were reduced with the AIPS[^2] package, using the standard procedures. Flux densities were extracted from images via Gaussian fit of the components, using task JMFIT inside AIPS. IRAM-30m single dish -------------------- We could observe 11 sources of the BAL QSO sample at 250 GHz with the IRAM-30m telescope, during the 2010 summer pool session. We used the MAMBO2 117-pixel bolometer in ON/OFF mode, since all of our sources are point-like for this telescope (HPBW=11 arcsec). With average atmospheric conditions the detector could reach a noise of $\sim$1 mJy/beam in $\sim$40 min of observing time: we observed each source for this duration in order to obtain a detection or a 3-$\sigma$ upper limit. Skydip, calibration and pointing scans were regularly performed during the runs and every time the observing direction in the sky significantly changed in elevation. Focus was repeated at sunrise and sunset. A standard reduction was done using the MOPSIC[^3] script provided by IRAM. APEX ---- From 2010 to 2012 we observed with APEX a total of 9 southern sources from the BAL QSO sample. All of them were observed with the SABOCA bolometer array (@Siringo2) at 850 GHz in photometry mode (HPBW $\sim$8 arcsec) and two of the sources (0044+00 and 1404+07) also in mapping mode. In addition, one of the sources (0044+00) was observed with the LABOCA bolometer array (@Siringo1) at 345 GHz in photometry mode (HPBW $\sim$19 arcsec). APEX observations were carried out in service mode, with typical integration times of 1 h per source, in order to reach RMS values around 20 mJy/beam. Calibration was based on observations of primary calibrators (Mars and Uranus) as well as skydips measured at the same azimuth of the targets to derive atmospheric opacity. A standard reduction was done using the version 2.15-1 of the CRUSH[^4] software which offers an improved pipeline for photometric data as compared to earlier versions. Error determination ------------------- In the flux density error calculation, different contributions were considered for the GMRT interferometric observations: - The thermal noise, $\Delta S _{\rm{noise}}$, which is estimated from the map, in empty regions of sky surrounding the target;\ - The fractional calibration error, $\Delta S_{\rm{calib}}$, estimated as the visibilities dispersion of the flux density calibrators;\ In particular, we followed the approach proposed by Klein et al. (2003). The expression used is reported below: $$\Delta I=\sqrt{(\Delta S_{\rm{calib}}\cdot S)^2 +(\Delta S _{\rm{noise}})^2 \cdot \frac{A_{\rm{src}}}{A_{\rm{beam}}}},$$ where $A_{\rm{beam}}$ and $A_{\rm{src}}$ are respectively the area of the synthesised beam and the aperture used to extract the source flux density. From their ratio we determine the number of beams contained in the source. For APEX and IRAM-30m data, obtained with bolometer receivers, the error was calculated using the respective packages for data reduction, estimating the noise from off-source subscans. The polarimetry campaign ------------------------ During 2011, we conducted a polarimetry campaign using the EVLA and the Effelsberg-100m single dish on this same sample, to implement data later presented in [@Bruni], and probe with deeper observations the polarisation of the faintest sources. The results from this campaign will be presented in a future paper, but we decided to use part of the obtained total flux density measurements to improve the SED coverage of this work (see Tab. \[errata\]). Also, one of the sources observed with the EVLA turns out to have a resolved structure (0849+27), and we present here the map (see Sect. \[sec:morphology\]). This same source was also observed during our mm campaign (see Tab. \[dust\_flux\]). Observations and data reduction were conducted as in [@Bruni]. --------- ----------- -------------- ----------------- ID Frequency S Telescope (J2000) \[GHz\] \[mJy\] 0756+37 43 5.2$\pm$0.8 VLA 0816+48 1.4 70.9$\pm$0.7 VLA 1335+02 43 8.4$\pm$1.3 VLA 0842+06 8.35 19.5$\pm$0.8 Effelsberg-100m 0849+27 4.86 27.6$\pm$0.7 EVLA --------- ----------- -------------- ----------------- : Top lines: revised flux densities for sources presented in [@Bruni] and in this work. Bottom lines: flux densities from our polarimetry campaign, used for this work. \[errata\] ![image](./0849_multi.pdf){width="18cm"} Results {#sec:results} ======= In the following, we present the results from our observing campaign: the morphology could be studied with the GMRT interferometer, while images from the IRAM and APEX bolometers were used for photometric measurements. The collected information is presented in Tab. \[GMRT\_size\], while SEDs, including flux densities at cm wavelengths from [@Bruni] and our polarimetry campaign, are presented in Fig. \[dust\] and discussed in Sect. \[sec:SED\_section\]. Morphology {#sec:morphology} ---------- From the GMRT and EVLA maps we were able to investigate the morphologies of the sources at arcsec scales. The frequency range explored with the GMRT allowed us to put some constraints to the presence of extended, old, radio components. ### GMRT maps {#gmrt-maps .unnumbered} We observed with the GMRT the five sources from our BAL QSO sample showing the strongest low-frequency emission in the flux densities collected from archival surveys data. The goal was the detection of extended emission at 235 or 610 MHz, indicating a previous radio-activity period of the central AGN, and thus putting a constraint on the age. Maps show components with deconvolved dimensions greater than zero in most cases, corresponding to a fraction of the beam (see Fig. \[GMRT\]). One source (1159+01) presents an elongated structure at 235 MHz, confirmed by the detection of a second component (B) at higher resolution in the 610 MHz map. This structure is compatible with the one seen in the pc-scale maps obtained by [@Tak], where a jet extension at a comparable position angle is visible. This could confirm the presence of two different radio phases, at different scales, as also highlighted by the SED of this object (see Sec. \[sec:SED\_section\]). Quantities for all sources at the two frequencies are presented in Table \[GMRT\_size\], together with projected linear sizes. These measurements confirm the presence of a low-frequency, older radio component in some BAL QSOs, thus excluding that they are a subclass of young radio objects. The size of these components is significantly larger than the values of a few kpc measured for the high-frequency, unresolved, counterparts (see @Bruni), thus suggesting different emitting regions for the two. The flux densities found nicely fit with collected data from surveys (see Fig. \[dust\]). ![image](./GMRT.pdf){width="18cm"} --------- ------------ ----------- ----------- ---------- ----------- ----------- ---------------- ----------- ----------- ---------- ----------- ID $S$ Maj. axis Min. axis Maj axis Min. axis Component $S$ Maj. axis Min. axis Maj axis Min. axis (mJy) (arcsec) (arcsec) (kpc) (kpc) (mJy) (arcsec) (arcsec) (kpc) (kpc) 0756+37 $<$85.2 - - - - 112$\pm$4 3.9 2.2 32.4 18.3 1159+01 720$\pm$65 19.6 7.0 164 58.4 A 312$\pm$10 4.6 2.3 39.5 19.8 B 74.8$\pm$4.2 10.2 0.9 86.3 7.6 1159+06 872$\pm$81 6.9 $<$10.1 58.7 $<$85.9 163$\pm$5 6.7 2.4 57.0 20.4 1406+34 249$\pm$26 14.8 $<$11.0 122 $<$90.9 168$\pm$5 3.9 1.1 32.2 9.1 1624+37 104$\pm$16 11.1 $<$11.0 84.9 $<$84.1 98.6$\pm$4.9 4.3 3.7 32.9 28.3 --------- ------------ ----------- ----------- ---------- ----------- ----------- ---------------- ----------- ----------- ---------- ----------- ### EVLA map of 0849+27 {#evla-map-of-084927 .unnumbered} During our polarimetry campaign, we could observe the peculiar BAL QSO 0849+27 with the EVLA at 4.86 GHz. The map of the resolved structure of this source at 1.4 GHz, obtained from the FIRST survey, was already presented in a previous work from our group (@Bruni). This resulted to be the most extended BAL QSO of our sample (44 arcsec, 382 kpc, between components A and C). We could obtain a map at 4.8 GHz from our subsequent EVLA observation in 2011, taking advantage of the improved performance of this instrument, and also a high resolution (pc-scale) map fo the core component from our VLBA program (see @Bruni2). Results are presented in Fig. \[0849\]. Four out of five components detected in the FIRST data are visible in our EVLA map (A, B, C, E), while the flux density of component D seems to drop below the 3-$\sigma$ significance level. Total flux densities as measured at 4.86 GHz are 27.6$\pm$0.7 mJy, 2.0$\pm$0.2 mJy, 0.49$\pm$0.05 mJy, and 0.41$\pm$0.05 mJy for component A, B, C, E, respectively. Components E was not classified in our previous FIRST map, but given the clear detection we obtained in the EVLA map, we extracted the flux density at the corresponding position in the FIRST map, where a single contour was present. This results to be 2.36$\pm$0.15 mJy. The obtained spectral index for these components, are $-$0.54$\pm$0.04, $-$1.10$\pm$0.17, $-$1.78$\pm$0.18, and $-$1.41$\pm$0.22 for components A, B, C, and E, respectively. A flat spectral index ($>-$0.5) usually identifies the core for non-Doppler-boosted components: component A shows a spectral index compatible with that value within the error, while B, C, and E have a steep spectral index ($<-$0.5). From our VLBA observations we could confirm that core is component A, since at pc-scale it shows a core-jet structure, with one of the two components (A1) having a flat spectral index of $-$0.12$\pm$0.20 between 5 and 8.4 GHz. The peculiar morphology of this source, explored at both kpc and pc scales, suggests a jet precession, or radio-activity phases with different jet axes. In fact, trajectories connecting the core (A) with the other components detected at kpc scale are all different, not permitting to identify one as the counter-jet of the other. Moreover, the pc-scale structure shows a further launching direction for the most recent component (A2), not corresponding to any among the ones at kpc scale. In this scenario, the BAL-producing outflow would be present in a source presenting multiple ongoing radio phases: this would not be easily explicable with the young scenario. SEDs shape from 235 MHz to 850 GHz {#sec:SED_section} ---------------------------------- We present here a study of the SED shape for the objects in this work. We implemented the new flux densities in the ones from [@Bruni], spanning from 74 MHz up to 43 GHz, in order to improve the overall frequency coverage. For three sources (0756+37, 0816+48, 1335+02) we provide here a revised flux density for the VLA measurements presented in [@Bruni]: for source 0816+48, during previous data reduction the vicinity of a strong source led to an incorrect phase referencing, resulting in an overestimated flux density measurement. While for sources 0756+37 and 1335+02, our flux-extraction algorithm missed the source, displaced of a few arcsec from the map centre, due to atmospheric effects. The revised values are given in Tab. \[errata\]. ### Flux densities at mm wavelengths {#flux-densities-at-mm-wavelengths .unnumbered} Flux density from the dust grey-body thermal emission can be described by the following equation (@Hughes): $$S^{\rm{obs}}= \frac{(1+z)}{D_L^2} \times M_dk_d^{\rm{rest}}B(\nu^{\rm{rest}},T_d), \label{Sdust}$$ where $S^{\rm{obs}}$ is the observed flux density at a given frequency $\nu_{obs}$, and $\nu_{rest}$ is the rest-frame frequency, $D_L$ is the luminosity distance, $z$ is the target redshift, $k_d^{\rm{rest}}$ is the mass absorption coefficient at a given rest-frequency, $B$ is the black-body Planck function, $T_d$ is the dust temperature, and $M_d$ is the dust mass. Typical AGN values for $T_d$ and $M_d$ from the latest works in the literature are 10<$T_d$<60 K, and $M_d\sim10^8 M_\odot$ (@Kal), while $k_d^{\rm{rest}}$ is usually calculated scaling down to the desired wavelength values estimated in previous works (e.g. $k_d^{\rm{850\rm{\mu m}}}$=0.077 m$^2$kg$^{-1}$ in @Dunne00 [@Dunne11]), approximating the trend vs wavelength as a power law: $$k_d^{\rm{rest}}\propto\lambda^\beta,$$ where $\beta$ is the dust emissivity index (@daCunha). We could collect flux densities at mm-wavelengths for a total of 17 objects from the BAL QSO sample. In most cases only upper limits could be derived (see Tab. \[dust\_flux\]), nevertheless providing useful constraints for the dust abundance. Only one source (0756+37) shows a flux density at 250 GHz clearly distinguishable from the expected synchrotron-emission tail: this is the only genuine example of dusty BAL QSO found among these objects. Despite the fact that we could not collect enough detections in the mm-band to perform a full fit of the dust emission using Eq. \[Sdust\], we can still compare the obtained upper limits with the results from previous works in the literature (@Omont [@Kal], see Sec. \[sec:discussion:dust\] for a discussion). ### SED fitting {#sed-fitting .unnumbered} For the SED fitting in the m-/cm-wavelengths domain, we adopted the same basic approach as in our previous work, i.e. a linear and a parabolic fit in logarithmic scale, determining which of the two functions fits best the data. The former is a simple model of power-law synchrotron emission, commonly used to fit the optically-thin part of the emission above the peak, for old sources peaking at frequencies below the available ones. The latter is a first-order approximation of an optically-thick (on the left side of the peak) plus an optically-thin synchrotron emission (on the right side of the peak) capable of fitting the SED of a young radio source, peaking in the MHz-GHz range. Upper limits were excluded from the dataset used for fitting, as well as datapoints clearly belonging to a second component in the MHz range (if any). After introducing the new data, we obtained best fits with the same functions (line or parabola) than previously adopted, except for sources 0842+06, that resulted to have a parabolic shape, and 1335+02, that after the introduction of the revised flux density from Tab. \[errata\] show a HFP component peaking at $\sim$20 GHz (see Fig. \[dust\]). Another source (1159+06) showed a second synchrotron component in the MHz range, peaking at about 100 MHz, while 0816+48, that was not fitted in [@Bruni], now presents a parabolic component in the GHz range using again the revised flux density at 1.4 GHz. In two of the three sources detected at 250 GHz the emission is most probably due to synchrotron component (1237+47, 1406+34), since it perfectly fits in the shape of the emission tail. For source 0756+37 we found a significant excess of emission at 250 GHz (2.0$\pm$0.5 mJy), with respect to the expected contribution of the synchrotron emission at the same frequency. Thus we can consider this emission as most probably produced by dust. ----------- ------------- -------- -------- -- -- 0044+00 - $<$260 $<$30 0756+37 2.0$\pm$0.5 - - 0816+48 $<$1.8 - - 0842+06 - - $<$71 0849+27 $<$2.4 - - 1014+05 $<$3.0 - - 1102+11 $<$3.0 - - 1159+01 - - $<$150 1159+06 $<$6.0 - $<$161 1229+09 $<$3.3 - $<$66 1237+47 4.6$\pm$1.0 - - 1304+13 $<$3.0 - - 1327+03 $<$3.6 - $<$201 1335+02 - - $<$78 1337$-$02 - - $<$66 1404+07 - - $<$90 1406+34 9.3$\pm$0.8 - - ----------- ------------- -------- -------- -- -- : Results for the 17 BAL QSOs observed in the mm-band: 250 GHz flux densities from IRAM-30m, 345 and 850 GHz from APEX. We give 3-$\sigma$ upper limits in case of non-detection.[]{data-label="dust_flux"} Discussion {#sec:discussion} ========== In the following, we discuss our findings and put them in the context of the works present in the literature. Low-frequency components ------------------------ A consistent number of restarting radio sources have been found in the past 20 years, and hypotheses about their nature have been proposed (@Czerny [@Wu; @Marecki]). A fraction corresponds to GPS sources or CSS sources, known to be young radio sources. In some cases, up to three radio phases are detectable in the same object (@Brocksopp). [@Czerny] associate the intermittent activity of the central engine with the radiation pressure instability of the accretion disk. Concerning our results in the MHz range, the fact that hints of old, extended, radio components were found, despite the limited number of objects observed with the GMRT, could indicate that an age of $10^7$-$10^8$ years for the BAL-hosting QSOs is not rare. This finding is in line with what arose from our previous works (@Bruni [@Bruni2]), where we discussed the presence of old components in the SED of our sample. In some cases a radio-restarting scenario, and the complex dynamics involved by that, could be invoked to explain a two-component SED (1159+01, 1159+06, 1335+02, 1406+34), showing both a GHz-peaked and a MHz component. In other cases (1014+04, 1229+09, 1304+13, 1327+03) a peak in the MHz range could still be present, but not seen because of insufficient frequency coverage. Considering this, and the SEDs of objects previously studied in [@Bruni], a $\sim$70% of our sample could be in a GPS or GPS+CSS phase. Once again this shows how BAL-producing outflows can be present not only in young radio sources, but also in more complex scenarios. For example, sources found to have multiple radio phases ongoing (CSS+GPS), should have gone through an unstable radiation pressure phase, causing an intermittent BAL-producing outflow acceleration (favoured during the high-pressure phases). This could include young, just-started radio sources (GPS/HFP) as well as restarted radio sources (e.g. those with a CSS+GPS SED). If this link with ignition/recollimation of the radio jet would be confirmed, an outflow collimation to form a jet (as already proposed by @Elvis) could be invoked to explain the BAL variability. Dust emission {#sec:discussion:dust} ------------- Several works in the past years, especially with the advent of the Herschel space observatory, has constrained the dust emission properties in AGN, and verified the connection with star-formation rate. Here we provide a comparison between our results for BAL QSOs in the mm-band and what found by other authors. A very similar study to ours, in terms of instrumentation, sensitivity, and setup, is the one presented by [@Omont]. They performed 250-GHz observations of 35 optically luminous RQ QSOs ($M_{B}<-$27.0), with a similar redshift range to ours (1.8$<z<$2.8), performed with the MAMBO bolometer at the same frequency. They found that 26$\pm$9% of the sources present an emission at that frequency. Since they reached an RMS very similar to our observations ($\sim$1 mJy) we can compare our detection rate with this percentage: only 1 out of 17 ($\sim$6%) of our sources shows 250-GHz emission attributable to dust, a substantially smaller fraction than the one found by these authors. More recently, [@Kal] published results of the [Herschel]{}-ATLAS project regarding FIR properties of radio-loud and radio-quiet QSOs. Using five different bands (100, 160, 250, 350, 500 $\mu$m) they fitted the dust emission for both groups. At the mean redshift of our sample (z$\sim$2.3) they found a mean flux density for radio-loud QSOs of 23.5$\pm$2.1 mJy at 350 $\mu$m (corresponding to our 850 GHz observations), and 21.3$\pm$2.8 mJy at 500 $\mu$m ($\sim$600 GHz). These are values below the sensitivity we reached with APEX, but using a simple power law to extract the expected mean flux density at 250 GHz from the previous values, we obtain a value of $\sim$17 mJy, that would be well detectable with the RMS of $\sim$1 mJy we could reach at IRAM-30m. This could suggest that our sample of RL BAL QSOs is also poorer in dust content than normal RL QSOs, in addition to RQ ones (as seen comparing with @Omont). In the light of these works, our results support the idea of BAL QSOs not being specially dusty objects. In addition, although with modest statistical significance, our work suggests that dust emission in the RL BAL QSO sub-class could be weaker than expected. [@Tombesi] found that fast outflows ($\sim$0.25c) from the accretion disk in AGN can hamper star formation, impacting the interstellar medium. BAL-producing outflows, although at a lower ionization stage than the ones considered by those authors, can present velocities up to 0.2c, and, in the light of these results, could have a similar effect on the SFR of the host galaxy. This needs to be investigated with further observations, that could confirm this on larger samples of RL BAL QSOs. Conclusions =========== We performed observations in the m-band with the GMRT of 5 RL BAL QSOs, that already showed hints of emission in the MHz range, and in the mm-bands for 17 RL BAL QSOs, from our previously studied sample. We aimed at exploring the emission in the low-frequency regime, and the grey-body emission from dust, respectively. The conclusions are the following: - All 5 objects observed at low frequencies present emission from extended components, indicating the presence of an old radio emission. In some cases a restarting radio-activity can be invoked to explain the double component, MHz and GHz-peaked, present in their SEDs. This could suggest an intermittent BAL phase, associated with periods of radio-restarting activity. This is also supported by the morphology of some objects, both from this work (0849+27) and from [@Bruni2], and by the fact that $\sim$70% of our RL BAL QSOs sample can be considered in a GPS or CSS+GPS phase, thanks to the data presented here. - Only 1 out of 17 sources ($\sim$6%) presents a clear contribution at 250 GHz from the dust grey-body emission. In the other cases 3-$\sigma$ upper limits have been derived. Comparing our results with the fraction of dust-rich RQ QSOs found by [@Omont], resulting in a percentage of $\sim$26%, we found that RL BAL QSOs do not present a larger fraction of dust-rich objects with respect to the RQ QSO population. Also comparing with more recent works from the Herschel-ATLAS collaboration (@Kal), we found a lack of dust emission with respect to mean values for RL QSOs. Since the amount of dust can be connected to star-formation and thus to the age of the host galaxy, this results suggest that RL BAL QSOs are not a hosted by particularly young galaxies. The fact that, despite of radio loudness, these objects present even less dust emission (and consequently a lower star formation rate) than RQ QSOs, could suggest that BAL-producing outflows are able to hamper star formation in the host galaxy. - Both the obtained results, from observations performed at m- and mm-wavelengths, suggest that BAL QSOs are not commonly young radio objects, or objects still expelling their dust cocoon from the central region. They could be radio-restarting objects, which present relativistic outflows in conjunction with some periods of favourable emission/acceleration conditions. Part of this work was supported by a grant of the Italian Programme for Research of National Interest (PRIN No. 18/2007, PI: K.-H. Mack) The authors acknowledge financial support from the Spanish Ministerio de Ciencia e Innovación under project AYA2008-06311-C02-02. We thank the staff of the GMRT that made these observations possible. GMRT is run by the National Centre for Radio Astrophysics of the Tata Institute of Fundamental Research. This publication is based on data acquired with the Atacama Pathfinder Experiment (APEX). APEX is a collaboration between the Max-Planck-Institut für Radioastronomie, the European Southern Observatory, and the Onsala Space Observatory. This work is partly based on observations carried out with the IRAM-30m Telescope. IRAM is supported by INSU/CNRS (France), MPG (Germany) and IGN (Spain). ![image](./SEDs.pdf){width="18.5cm"} Adelman-McCarthy, J.K., Agüeros, M.A., Allam, S.S. et al. 2007, AJ Supplement Series, 172, 2, 634 Allen, J.T., Hewett, P.C., Maddox, N. et al. 2011, MNRAS, 410, 860 Becker, R.H., White, R.L., Gregg, M.D. et al. 2001, ApJS, 135, 227 Best, P.N., Rottgering, H.J.A., Bremer, M.N. et al. 1998, MNRAS, 301, L15 Best, P. N., Kaufmann, G., Heckman, T. M. et al. 2005, MNRAS, 362, 25 Briggs, F. H., Turnsheck, D. A. & Wolfe M. 1984, ApJ, 287, 549 Brocksopp, C., Kaiser, C. R., Schoenmakers, A. P. et al. 2007, MNRAS 382, 1019 Bruni G., Mack, K.-H., Salerno E. et al., 2012, A&A, 542, A13 Bruni, G., Dallacasa, D., Mack, K.-H. et al. 2013, A&A, 554, A94 Bruni G., González-Serrano, J.-I., Pedani M. et al., 2014, A&A, 569, A87 Cao Orjales, J. M., Stevens, J. A., Jarvis, M. J. et al. 2012, MNRAS, 427, 1209 Capellupo, D. M., Hamann, F., Shields, J. C. et al. 2011, MNRAS, 413, 908 Capellupo, D. M., Hamann, F., Shields, J. C.et al. 2012, MNRAS, 422, 3249 Czerny, B., Siemiginowska, A., Janiuk, A. et al. 2009, Vol. 698, Issue 1, 840 Dallacasa, D., Stanghellini, C., Centoza, M. et al. 2000, A&A, 363, 887 da Cunha, E., Charlot, S., and Elbaz D. 2008, MNRAS, 388, 1595 DiPompeo, M. A., Brotherton, M. S., De Breuck, C. et al. 2011, ApJ, 743, 71 Dunne, L., Eales, S., Edmunds, M. et al. 2000, MNRAS, 315, 115 Dunne, L., Gomez, H. L., da Cunha, E. et al. 2011, MNRA, 417, 1510 Elvis, M. 2000, ApJ, 545, 63 Farrah, D., Lacy, M., Priddey, R. et al. 2007, ApJ, 662, Issue 2, 59 Harrison, C. M., Alexander, D. M., Mullaney, G. R. et al. 2014, MNRAS, 441, 3306 Feruglio, C., Maiolino, R., Piconcelli, E., et al. 2010, A&A, Vol. 518, id. L155 Filiz Ak, N., Brandt, W. N., Hall, P. B., et al. 2012, ApJ, 757, 114 Filiz Ak, N., Brandt, W. N., Hall, P. B. et al. 2013, ApJ, 777, 168 Ganguly, R. & Brotherton, M. S. 2008, ApJ, 672, 102 Gibson, R. R., Brandt, W. N., Schneider, D. P. et al. 2008, ApJ, 675, 985 Gibson, R. R., Brandt, W. N., Gallagher, S. C. et al. 2010, ApJ, 713, 220 Hall, P. B., Anderson, S. F., Strauss, M. A. et al. 2002, ApJS, 141, 267 Hewett, P.C. & Foltz, C.B. 2003, AJ, 125, 1784 Hughes, D. H., Dunlop J. S., and Rawlings S. 1997, MNRAS, 289, 766 Kalfountzou, E., Stevens, J. A., Jarvis, M. J. et al. 2014, MNRAS, 442, 1181 Konar, C.; Hardcastle, M. J.; Jamrozy et al. 2013, MNRAS, 430, 2137 Konar, C., Saikia, D. J., Jamrozy, M. et al. 2006, MNRAS, 372, 693 Lara, L., Márquez, I., Cotton, W. D. et al. 1999, A&A, 348, 699 Marecki, A., Thomasson, P., Mack, K.-H. et al. 2006, A&A, 448, 479 Montenegro-Montes, F.M., Mack, K.-H., Vigotti, M. et al. 2008, MNRAS, 388, 1853 Nandi, S., Roy, R., Saikia, D. J. et al. 2014, ApJ, 789, Issue 1, article id. 16 O’Dea, C.P. 1998, PASP, 110, 493 Omont, A., Beelen, A., Bertoldi, F. et al. 2003, A&A, 398, 857 Orienti, M. & Dallacasa, D. 2008, A&A, 477, 807 Pounds, K. A., Reeves, J. N., King, A. R. et al. 2003, MNRAS, 345, 705 Priddey, R.S., Gallagher, S.C., Isaak, K.G. et al., 2007, MNRAS, 374, 867 Rochais, T. B., DiPompeo, M. A., Myers, A. D. et al. 2014, MNRAS, 444, Issue 3, 2498 Saikia, D. J. & Jamrozy, M. 2009, BASI, 37, 63 Sanders D. M., ASP conference series, Vol 284, 2002 Schneider, D. P., Hall, P. B., Richards, G. T. et al 2007, AJ, 134, 102 Schoenmakers, A. P., de Bruyn, A. G., Röttgering, H. J. A. et al. 2000, MNRAS, 315, 371 Shabala, S. S., Ash, S., Alexander, P. et al. 2008, MNRAS, 388, 625 Siringo, G., Kreysa, E., Kovács, A. et al. 2009, A&A, 497, Issue 3, 945 Siringo, G., Kreysa, E., De Breuck et al. 2010, The Messenger, 139, 20 Stanghellini, C., Baum, S. A., O’Dea, C. P. et al. 1990, A&A, 233, 379 Stanghellini, C., O’Dea, C. P., Dallacasa, D. et al. 2005, A&A, 443, 891 Sturm, E., González-Alfonso, E., Veilleux, S. 2011, ApJ Letters, Vol. 733, Issue 1, id. L16 Stocke, J.T., Morris, S. L., Weymann, R.J. et al. 1992, ApJ, 396, 487 Hayashi, T. J., Doi, A. & Nagai, H. 2013, ApJ, 772, 4 Tombesi, F., Tazaki, F., Mushotzky, R. F. et al. 2014, MNRAS, 443, 2154 Tombesi, F., Meléndez, M., Veilleux, S. et al. 2015, Nature, Vol. 519, Issue 7544, 436 Vivek, M., Srianand, R., Petitjean, P., et al. 2012, MNRAS, 423, 2879 Zahid, H. J., Kashino, D., Silverman, J. D. et al. 2014, ApJ, 792, Issue 1, 75 Willott, C.J., Rawlings, S., Grimes, J.A., 2003, ApJ, 598, 909 Wang, P., Li, Z.-Y., Abel, T., et al. 2010, ApJ, Vol. 709, Issue 1, 27 Wu, Q. 2009, ApJ, 701, L95 [^1]: i.e. we adopted an AI defined as ${\rm{AI}}=\int_{0}^{25000}(1-\frac{f(v)}{0.9})\cdot Cdv$, as in @Hall, but with $C$=1 only for contiguous troughs $\ge$ 1000 km/s, and null otherwise. [^2]: http://www.aips.nrao.edu/index.shtml [^3]: http://www.iram.es/IRAMES/mainWiki/CookbookMopsic [^4]: http://www.submm.caltech.edu/ sharc/crush/	{ "pile_set_name": "ArXiv" }
--- abstract: 'Conventional automatic speech recognition (ASR) systems trained from frame-level alignments can easily leverage posterior fusion to improve ASR accuracy and build a better single model with knowledge distillation. End-to-end ASR systems trained using the Connectionist Temporal Classification (CTC) loss do not require frame-level alignment and hence simplify model training. However, sparse and arbitrary posterior spike timings from CTC models pose a new set of challenges in posterior fusion from multiple models and knowledge distillation between CTC models. We propose a method to train a CTC model so that its spike timings are guided to align with those of a pre-trained [guiding]{} CTC model. As a result, all models that share the same guiding model have aligned spike timings. We show the advantage of our method in various scenarios including posterior fusion of CTC models and knowledge distillation between CTC models with different architectures. With the 300-hour Switchboard training data, the single word CTC model distilled from multiple models improved the word error rates to 13.7%/23.1% from 14.9%/24.1% on the Hub5 2000 Switchboard/CallHome test sets without using any data augmentation, language model, or complex decoder.' address: \| $^\star$IBM Research - Tokyo\ $^\dagger$IBM T. J. Watson Research Center title: \| Guiding CTC Posterior Spike Timings\ for Improved Posterior Fusion and Knowledge Distillation --- Index Terms: End-to-end speech recognition, Connectionist Temporal Classification (CTC), Knowledge distillation Introduction {#sec:introduction} ============ End-to-end (E2E) automatic speech recognition (ASR) using the Connectionist Temporal Classification (CTC) loss function has been gathering interest since it significantly simplifies model training pipelines[^1][@graves2006connectionist; @miao2015eesen; @sak2015learning; @pundak16:_lower; @miao2016empirical; @zweig2017advances; @Soltau2017]. Prior to E2E ASR systems, the conventional training pipeline for Gaussian Mixture Model-Hidden Markov Model (HMM) systems [@rabiner1989tutorial] and Deep Neural Network (DNN)-HMM hybrid systems [@hinton2012deep] required output symbols for every input acoustic frame, that is, [frame-level alignment]{}, which made the training process complex and time-consuming. Instead, E2E ASR only requires pairs of input feature sequences and output symbol sequences [@graves2006connectionist; @graves2013speech; @graves2013generating; @graves2014towards; @hannun2014deep; @hannun2017sequence], such as phones [@chorowski2015attention], characters [@chan2016listen], words [@Audhkhasi2017; @Soltau2017], or their combinations [@liu2017gram; @audhkhasi2017building]. One advantage of training ASR models with frame-level alignments is that frame-level posterior fusion of multiple systems is easy. Models trained from the same frame-level alignment data have the same target output symbol for each frame. Thus, we can compute the average or a weighted average of the posteriors from multiple systems to yield better posteriors, which results in improved ASR accuracy [@george17:_englis_conver_telep_speec_recog_human_machin] because such posterior averaging effectively creates a mixture model comprising of the different systems being combined. We can also use the (weighted) average of posteriors from multiple acoustic models as a teacher to train a single student model via knowledge distillation[^2] [@ba2014deep; @li2014learning; @hinton2015distilling; @chebotar2016distilling; @fukuda2017efficient]. In contrast, as noted in prior literature [@graves2006connectionist; @sak2015learning], CTC models emit very [spiky]{} posterior distributions[^3] where most frames emit the garbage [blank]{} symbol with high probability and only a few frames emit the target symbols of interest. As a result, CTC models trained with the same training data can have different spike timings (typical examples are shown later in [Figure \[fig:uni\_phone\]]{} and \[fig:bi\_word\]), which poses a new set of challenges. Due to non-aligned spike timings, we cannot make better posteriors by computing their (weighted) average across multiple systems [@sak2015acoustic]. Hence, it’s difficult to improve ASR accuracy by a naïve posterior fusion, and also to use the (weighted) average of posteriors as a teacher to train a single model via knowledge distillation. Disagreement of CTC spike timings has been discussed in the context of knowledge distillation from bidirectional LSTM (BiLSTM) CTC to unidirectional LSTM (UniLSTM) CTC models [@kurata18:_improv_knowl_distil_from_bi] and from multiple accent-specific models to a single multi-accent model [@ghorbani2018advancing]. [@kurata18:_improv_knowl_distil_from_bi] proposed searching for a similar posterior from a BiLSTM CTC model at a different time index to train a UniLSTM model. This approach only focused on knowledge distillation from BiLSTM to UniLSTM and cannot be simply extended to a posterior fusion of multiple systems. [@ghorbani2018advancing] attempted to train an accent-specific model for each accent by feeding training data of the target accent to a common teacher model trained from the data of all accents via knowledge distillation. As a result, each accent-specific model has aligned spike timings with the common teacher model. This method did not address knowledge distillation between different neural network (NN) architectures, such as between BiLSTM and UniLSTM CTC models. In this paper, we propose a method to explicitly guide the CTC spike timings to be aligned with those from a pre-trained CTC model with the NN architecture of interest. We call this pre-trained CTC model a [guiding model]{}. More specifically, when training a CTC model, in addition to a normal CTC loss, we add a new loss term that forces the spikes from the model being trained to occur at the same time as those from the guiding model. Hence, models guided by the same guiding model have aligned spike timings. The advantages of our proposed method are as follows: Posterior fusion of multiple CTC models : \ We can make better posteriors by averaging aligned posteriors from multiple models that share the same guiding model and improve ASR accuracy as a result. In addition, we can use these averaged posteriors as a teacher to train a single model with improved ASR accuracy through knowledge distillation. : The spike timings of CTC models with different NN architectures can be significantly different, which makes knowledge distillation between such models difficult. A typical case is knowledge distillation from a BiLSTM CTC model to a UniLSTM CTC model. By training a teacher model using a guiding model that has the same architecture as the final desired student model, the trained teacher model can have the spike timings appropriate for the student model and can be used in knowledge distillation. Improved speech recognition accuracy : \ The CTC model guided by the guiding model has improved ASR accuracy compared to the model without the guiding model. The guiding model promotes a good alignment path and hence does not allow the CTC training to assign sufficient probabilities to bad alignment paths, which results in an improved accuracy. We will show these advantages in various experiments with the standard English Switchboard conversational telephone speech data. Connectionist Temporal Classification ===================================== Let $\bm{y}$ denote a length-$L$ sequence of target output symbols. Let $\bm{X}$ denote acoustic feature vectors over $T$ time steps. The conventional alignment-based DNN/HMM hybrid system training requires $L$ to be equal to $T$ [@hinton2012deep]. The alignment-free CTC introduces an extra blank symbol $\phi$ that expands the length-${L}$ sequence $\bm{y}$ to a set of length-${T}$ sequences $\Phi(\bm{y})$. Each sequence $\bm{{\hat y}} \in \Phi(\bm{y})$ is one of the [CTC alignments]{} between $\bm{X}$ and $\bm{y}$. The CTC loss is defined as the summation of symbol posterior probabilities over all possible CTC alignments: ${\cal L}_{CTC} = - \sum_{\bm{{\hat y}} \in \Phi(\bm{y})} P(\bm{{\hat y}} \| \bm{X}).$ We use phones and words as target output symbols, namely [phone CTC]{} and [word CTC]{}, for evaluation in [Section \[sec:experiments\]]{}. Guided CTC training {#sec:guided-ctc-training} =================== In order to guide the spike timings of CTC models, we propose a two-step training process and name it [guided CTC training]{}. First, we train a guiding CTC model that has appropriate spike timings for the final target use case. Then, we train another CTC model (the [guided model]{}) so that similar spike timings as that of the guiding model can be obtained. Note that as long as the set of the output symbols is the same, different NN architectures can be used for the guiding and the guided models. [Figure \[fig:guided\_ctc\_training\]]{} shows a schematic diagram on how to guide the spike timings of the guided model by using the pre-trained guiding CTC model, assuming that we have completed the training of the guiding model in advance. When training the guided CTC model, for a training sample $\bm{X}$ of $T$ time steps, we feed it to the guiding CTC model, do the forward pass, and obtain posteriors for each time index. We convert these posteriors to a mask $M(\bm{X})$ by setting a 1 at the output symbol with the highest posterior and 0 at other symbols at each time index. In cases where the blank symbol $\phi$ has the highest posterior, we set 0 for all symbols at this time index. This mask passes the output symbol that the guiding CTC model emits at each time index. Then we feed the same training sample to the guided CTC model being trained and obtain posteriors $P(\bm{X})$. By the Hadamard (element-wise) product of the mask and the posteriors, we obtain the masked posteriors $\hat{P}(\bm{X}) = M(\bm{X}) \circ P(\bm{X})$. The summation of masked posteriors $\hat{P}(\bm{X})$ becomes greater if the guided CTC model being trained has spikes for the same output symbol at the same times as the guiding CTC model. Thus, by maximizing this summation, we can guide the spike timings of the guided CTC model to be the same as those of the guiding CTC model. We call the summation of the masked posteriors multiplied by $-1$ the [guide loss]{} defined as ${\cal L}_{G} = -1 \cdot \sum \hat{P}(\bm{X})$ [^4]. The overall loss for the guided CTC training becomes the summation of the standard CTC loss and the guide loss as ${\cal L} = {\cal L}_{CTC} + {\cal L}_{G}$. Experiments {#sec:experiments} =========== We conducted ASR experiments on phone and word CTC models to determine the advantages of our proposed guided CTC training. We used the standard 300-hour Switchboard English conversational telephone speech as training data [@george17:_englis_conver_telep_speec_recog_human_machin]. All models were trained for 20 epochs using stochastic gradient descent with the Nesterov momentum of 0.9 and a learning rate starting from 0.03 and annealing at $\sqrt{0.5}$ per-epoch after the 10th epoch. We set the batch size to 128 for phone CTC models and 96 for word CTC models. We trained multiple models for each experiment to investigate the effect of posterior fusion. By using different seeds for random parameter initialization, even models with exactly the same architecture trained with the same procedure can have sufficient diversity to benefit from posterior fusion [@hinton2015distilling]. We follow this procedure and also randomized the order of the training data after the first epoch (SortaGrad) [@amodei2016deep]. Posterior fusion of UniLSTM phone CTC models {#sec:post-fusi-unilstm} -------------------------------------------- First, we conducted experiments on posterior fusion with UniLSTM phone CTC models. We used 40-dimensional logMel filterbank energies, their delta, and double-delta coefficients with frame stacking and skipping rate of 2 [@sak2015fast], resulting in 240-dimensional features. We used 44 phones from the Switchboard pronunciation lexicon [@george17:_englis_conver_telep_speec_recog_human_machin] and the blank symbol. For decoding, we trained a 4-gram language model with 24M words from the Switchboard + Fisher transcripts with a vocabulary size of 30K. We constructed a CTC decoding graph similar to the one in [@miao2015eesen]. For NN architecture, we stacked 6 UniLSTM layers with 640 units and a fully-connected linear layer of 640$\times$45, followed by a softmax activation function. All NN parameters were initialized to samples of a uniform distribution over $(-\epsilon,\epsilon)$, where $\epsilon$ is the inverse square root of the input vector size. For evaluation, we used the Hub5-2000 Switchboard (SWB) and CallHome (CH) test sets. [Figure \[fig:experiments\]]{} shows the flow of experiments and [Table \[tab:uni\_phone\_ctc\]]{} lists the results. We trained 4 UniLSTM models with the standard training. In [1A]{}, we averaged the Word Error Rates (WERs) for the 4 decoding outputs from the 4 models. For posterior fusion in [1B]{}, we averaged the posteriors from the 4 models and used it for decoding with the graph. For comparison, we also combined the 4 decoding outputs from the 4 models by using the Recognizer Output Voting Error Reduction (ROVER) [@fiscus1997post; @wang2017residual; @fritz2017simplified], as shown in [1C]{}. Comparing [1A]{} and [1B]{}, we can see that since the posteriors from the 4 models were not aligned, we did not benefit from posterior fusion. ROVER gave us a solid improvement in [1C]{}. For the proposed guided CTC training, as shown in the right of [Figure \[fig:experiments\]]{}, we first trained a UniLSTM phone CTC model with the same architecture and training data and then used it as a guiding model to train 4 guided models. [1D]{} indicates the average WERs of the 4 decoding outputs by the 4 guided UniLSTM models trained by the proposed guided CTC training. Comparing [1A]{} and [1D]{}, we can see a WER reduction by the proposed guided CTC training. The guide loss promotes a good CTC alignment path and assigns lower probabilities to bad CTC alignment paths on the basis of the guiding model during the CTC training[^5]. [1E]{} and [1F]{} show WERs by the posterior fusion and ROVER of the 4 models trained with the proposed guided CTC training. Due to the aligned posteriors, we benefit from the posterior fusion that outperformed the ROVER. Note that posterior fusion requires decoding with the graph just once, while ROVER needs 4 separate decodes. This is another advantage of posterior fusion realized by the proposed guided training. [Figure \[fig:uni\_phone\]]{} shows posteriors of multiple UniLSTM phone CTC models for the same utterance, “this (DH IH S) is (IH S) true (T R UW)”, in the SWB test set, where symbols in parentheses represent phones. As expected, phone spike positions were not aligned between the UniLSTM CTC models. [Figure \[fig:uni\_phone\_guided\]]{} shows posteriors of the guiding UniLSTM CTC model at the top and the UniLSTM CTC models guided by the guiding model at the top with the proposed training. The spikes from the guided models were temporally smoothed and the spike positions of the guiding models were covered by the smoothed spikes of the guided models. Hence, the spike timings from multiple guided models overlapped, which underpins the improved WERs by posterior fusion. Knowledge distillation from BiLSTM phone CTC to UniLSTM phone CTC models ------------------------------------------------------------------------ Next, we explored knowledge distillation from BiLSTM to UniLSTM phone CTC models. UniLSTM CTC models are more suitable for deployment in actual streaming ASR services and products [@sak2015acoustic; @soltau17:_reduc] and closing the ASR accuracy gap between UniLSTM and BiLSTM CTC models is critical [@kim18:_improv; @takashima18:_ctc; @kurata18:_improv_knowl_distil_from_bi]. Here, we used the same speaker-independent (SI) 240-dimensional feature vectors as the previous experiments while keeping the actual use-case of streaming ASR in mind. For teacher BiLSTM phone CTC models, we stacked 6 BiLSTM layers with 320 units each in the forward and backward layers and a fully-connected linear layer of 640$\times$45, followed by a softmax activation function. For UniLSTM phone CTC models. we used the same architecture as in the previous experiments. All NN parameters were initialized in the same fashion as the previous experiments. For evaluation, we used the same SWB and CH test sets. [Table \[tab:bi\_to\_uni\_phone\]]{} shows the results. [2A]{} and [2B]{} indicate the WERs by UniLSTM and BiLSTM phone CTC models trained with the standard procedure. As a naïve approach, in [2C]{}, we trained a UniLSTM model through knowledge distillation from BiLSTM models with minimizing the frame-wise KL divergence, but saw degradation in WERs. The spike timings from BiLSTM were not appropriate for UniLSTM, as also demonstrated in [Figure \[fig:bi\_and\_uni\_phone\]]{}, where the spike timings for the same utterance were completely different between the UniLSTM and BiLSTM models. In addition, posteriors from BiLSTM models trained with the standard training were not aligned and thus knowledge distillation from their posterior fusion was not successful. [2D]{} is the key step where we first trained a guiding UniLSTM model and then trained a BiLSTM model guided by the guiding UniLSTM model. The trained BiLSTM model had spike timings appropriate for UniLSTM model as shown in [Figure \[fig:bi\_to\_uni\_phone\]]{} while posteriors were estimated with bidirectional context. Comparing [2B]{} and [2D]{}, we can see that [2D]{} had a worse WER because the spike positions were unnatural for a BiLSTM. However, this model can serve as an appropriate teacher to train a UniLSTM model. We trained multiple BiLSTM models guided by the same guiding UniLSTM model and used their posterior fusion as a teacher to train a single student UniLSTM model by knowledge distillation, where the frame-wise KL divergence from the student model to the fused teacher posterior was minimized. [2E]{} shows the WERs by the UniLSTM models distilled from the posterior fusion of 1, 4, and 8 BiLSTM models guided by the same guiding UniLSTM model[^6]. Even in the case of just 1 guided BiLSTM model, the WER for SWB was reduced to 13.4%, which was equivalent to reducing the gap between BiLSTM and UniLSTM by 54.3%[^7]. By increasing the number of BiLSTM models, the WER for SWB was reduced to 12.9%, which equals a 68.6% accuracy gap reduction. As in the bottom two posteriors in [Figure \[fig:bi\_to\_uni\_phone\]]{}, the temporally smoothed spikes of the guided BiLSTM models overlapped around the spikes of the guiding UniLSTM model, which underpins the improved WERs by increasing the number of guided BiLSTM models. Posterior fusion and knowledge distillation of BiLSTM word CTC models --------------------------------------------------------------------- We applied our proposed guided CTC training to BiLSTM word CTC models. For input acoustic features, we added 100-dimensional i-vectors for each speaker extracted in advance and appended them to the same SI feature used in the previous experiments, resulting in 340-dimensional feature vectors [@dehak2011front]. We selected words with at least 5 occurrences in the training data [@Audhkhasi2017; @audhkhasi2017building]. This resulted in an output layer with 10,175 words and the blank symbol. We stacked 6 BiLSTM layers with 512 units each in the forward and backward layers (BiLSTM encoder), added 1 fully-connected linear layer with 256 units to reduce computation [@sainath2013low], and put 1 fully-connected linear layer of 256$\times$10,176, followed by a softmax activation function. For better convergence, we initialized the BiLSTM encoder part with the trained BiLSTM phone CTC model[^8] [@Audhkhasi2017; @audhkhasi2017building]. Other parameters were initialized in similar fashion as the phone CTC models. For decoding, we performed a simple peak-picking over the output word posterior distribution, and removed repetitions and blank symbols. For evaluation, in addition to SWB and CH, we used RT02, RT03, and RT04 test sets [@george17:_englis_conver_telep_speec_recog_human_machin]. [Table \[tab:bi\_word\_ctc\]]{} shows the results. We trained 4 BiLSTM models with the standard training and conducted posterior fusion and ROVER. [3A]{} indicates the average WERs of 4 decoding outputs from the 4 models and [3B]{} and [3C]{} indicate the WERs by posterior fusion and ROVER, respectively. Comparing [3A]{} and [3B]{}, since the posteriors from the 4 models were not aligned and the spikes from word models were much more sparse than those from phone models, we saw a significant degradation in accuracy. Due to the sparse non-aligned word spikes, ROVER also did not improve the ASR accuracy, as in [3C]{}. [Figure \[fig:bi\_word\]]{} shows the word posteriors from BiLSTM word CTC models that were sparse and non-aligned, which caused degradation in [3B]{} and [3C]{}. For the proposed guided CTC training, we first trained a BiLSTM word CTC model with the same architecture and training data and then used it as a guiding model to train the 4 guided models. [3D]{} indicates the average WERs of 4 decoding outputs by the 4 BiLSTM word models trained by the proposed guided CTC training. Comparing [3A]{} and [3D]{}, we can see WER reduction by the proposed guided CTC training, the same as with the case of phone CTC models in [Section \[sec:post-fusi-unilstm\]]{}. [3E]{} and [3F]{} show WERs by the posterior fusion and ROVER of the 4 guided models. Due to the aligned posteriors obtained by the guided CTC training, both posterior fusion and ROVER improved ASR accuracy compared with [3D]{} while posterior fusion outperformed ROVER. Finally, we used the posterior fusion from the guided BiLSTM models as a teacher and trained a single student BiLSTM word CTC model by knowledge distillation with minimizing the frame-wise KL divergence. [3G]{} shows the results where the WERs were further decreased from [3D]{} consistently over all test sets. [Table \[tab:comparison\]]{} shows a comparison on WERs with the published CTC-based direct acoustic-to-word models trained from the standard 300-hour Switchboard data without using any data augmentation or language model. To the best of our knowledge, our best single model trained by knowledge distillation outperformed the published CTC models trained in a purely E2E fashion in [4A]{} and [4B]{} [@audhkhasi2017building; @sanabria2018hierarchical]. By combining CTC and cross-entropy (CE) training using the alignment between the acoustic frame and the labels, as in a conventional non-E2E system, 13.0% and 23.4%[^9] were achieved in [4C]{} [@yu2018multistage]. Our best single model achieved the comparable WERs of 13.7% and 23.1% with purely E2E training without any frame-level alignments. Analysis on spike timings ------------------------- Looking at the posteriors in [Figure \[fig:posteriors\]]{}, we can see that spikes from models that were trained with the standard training (we call them [non-guided]{} models) were not aligned, while the spikes from the guiding and the guided models and those of the guided models that share the same guiding model were aligned. We quantitatively analyzed the coverage ratio by investigating if the posterior spikes (except for the blank and silence) from one model were covered by the spikes from the other model at the same time index. We used the training data and the test data (SWB) for this analysis; the results are shown in [Table \[tab:coverage\]]{}. Note that in the case of “guiding and guided models”, we investigated if the spikes from the guiding model were covered by the corresponding guided model. For other cases, we randomly picked 2 models trained using the same training procedure with changing the parameter initialization and the training data order. For both UniLSTM phone CTC and BiLSTM word CTC models, the coverage ratio of 2 non-guided models was low. Especially, due to sparse spikes, the coverage ratio of BiLSTM word CTC was much lower, which underpins the poor WERs by the posterior fusion in [3B]{} of [Table \[tab:bi\_word\_ctc\]]{}. The coverage ratio between the guiding and guided model was improved and the ratio between 2 guided models was slightly worse, but much higher than the ratio between the 2 non-guided models. Comparing the training and test data, the training data had a slightly higher coverage ratio, but the same trends were seen for all combinations of models, which supports the experimental results. Conclusion {#sec:conclusion} ========== We proposed a method to guide spike timings of CTC models by using a pre-trained CTC model with the NN architecture of interest. We demonstrated its advantages in various scenarios including posterior fusion of multiple CTC models and knowledge distillation between CTC models. Through the experiments, we achieved state-of-the-art WERs in the CTC-based direct acoustic-to-word setting without using any data augmentation or language model. By qualitatively and quantitatively investigating the posterior spike timings from the CTC models trained with the proposed guided CTC training, we confirmed that the spike timings were aligned between the guiding and the guided models and between the multiple guided models. [10]{} \[1\][\#1]{} url@samestyle \[2\][\#2]{} \[2\][[l@\#1=l@\#1\#2]{}]{} D. Bahdanau, J. Chorowski, D. Serdyuk, P. Brakel, and Y. Bengio, “End-to-end attention-based large vocabulary speech recognition,” in Proc. ICASSP, 2016, pp. 4945–4949. T. Hori, S. Watanabe, and J. Hershey, “Joint [CTC/attention]{} decoding for end-to-end speech recognition,” in Proc. ACL, vol. 1, 2017, pp. 518–529. A. Graves, S. Fern[á]{}ndez, F. Gomez, and J. Schmidhuber, “Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks,” in Proc. ICML, 2006, pp. 369–376. Y. Miao, M. Gowayyed, and F. Metze, “[EESEN]{}: End-to-end speech recognition using deep [RNN]{} models and [WFST-based]{} decoding,” in Proc. ASRU, 2015, pp. 167–174. H. Sak, A. Senior, K. Rao, O. Irsoy, A. Graves, F. Beaufays, and J. Schalkwyk, “Learning acoustic frame labeling for speech recognition with recurrent neural networks,” in Proc. ICASSP, 2015, pp. 4280–4284. G. Pundak and T. Sainath, “Lower frame rate neural network acoustic models,” in Proc. INTERSPEECH, 2016, pp. 22–26. Y. Miao, M. Gowayyed, X. Na, T. Ko, F. Metze, and A. Waibel, “An empirical exploration of [CTC]{} acoustic models,” in Proc. ICASSP, 2016, pp. 2623–2627. G. Zweig, C. Yu, J. Droppo, and A. Stolcke, “Advances in all-neural speech recognition,” in Proc. ICASSP, 2017, pp. 4805–4809. H. Soltau, H. Liao, and H. Sak, “Neural speech recognizer: Acoustic-to-word [LSTM]{} model for large vocabulary speech recognition,” in Proc. INTERSPEECH, 2017, pp. 3707–3711. L. R. Rabiner, “A tutorial on hidden [Markov]{} models and selected applications in speech recognition,” Proceedings of the IEEE, vol. 77, no. 2, pp. 257–286, 1989. G. Hinton, L. Deng, D. Yu, G. E. Dahl, A. Mohamed, N. Jaitly, A. Senior, V. Vanhoucke, P. Nguyen, T. N. Sainath, and B. Kingsbury, “Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,” IEEE Signal Processing Magazine, vol. 29, no. 6, pp. 82–97, 2012. A. Graves, A. Mohamed, and G. Hinton, “Speech recognition with deep recurrent neural networks,” in Proc. ICASSP, 2013, pp. 6645–6649. A. Graves, “Generating sequences with recurrent neural networks,” arXiv preprint arXiv:1308.0850, 2013. A. Graves and N. Jaitly, “Towards end-to-end speech recognition with recurrent neural networks,” in Proc. ICML, 2014, pp. 1764–1772. A. Hannun, C. Case, J. Casper, B. Catanzaro, G. Diamos, E. Elsen, R. Prenger, S. Satheesh, S. Sengupta, A. Coates et al., “Deep speech: Scaling up end-to-end speech recognition,” arXiv preprint arXiv:1412.5567, 2014. A. Hannun, “Sequence modeling with [CTC]{},” Distill, 2017, [https://distill.pub/2017/ctc]{}. J. K. Chorowski, D. Bahdanau, D. Serdyuk, K. Cho, and Y. Bengio, “Attention-based models for speech recognition,” in Proc. NIPS, 2015, pp. 577–585. W. Chan, N. Jaitly, Q. Le, and O. Vinyals, “Listen, attend and spell: A neural network for large vocabulary conversational speech recognition,” in Proc. ICASSP, 2016, pp. 4960–4964. K. Audhkhasi, B. Ramabhadran, G. Saon, M. Picheny, and D. Nahamoo, “Direct acoustics-to-word models for [English]{} conversational speech recognition,” in Proc. INTERSPEECH, 2017, pp. 959–963. H. Liu, Z. Zhu, X. Li, and S. Satheesh, “[Gram-CTC]{}: Automatic unit selection and target decomposition for sequence labelling,” in Proc. ICML, 2017, pp. 2188–2197. K. Audhkhasi, B. Kingsbury, B. Ramabhadran, G. Saon, and M. Picheny, “Building competitive direct acoustics-to-word models for [English]{} conversational speech recognition,” in Proc. ICASSP, 2018, pp. 4759–4763. G. Saon, G. Kurata, T. Sercu, K. Audhkhasi, S. Thomas, D. Dimitriadis, X. Cui, B. Ramabhadran, M. Picheny, L.-L. Lim, B. Roomi, and P. Hall, “English conversational telephone speech recognition by humans and machines,” in Proc. INTERSPEECH, 2017, pp. 132–136. J. Ba and R. Caruana, “Do deep nets really need to be deep?” in Proc. NIPS, 2014, pp. 2654–2662. J. Li, R. Zhao, J.-T. Huang, and Y. Gong, “Learning small-size [DNN]{} with output-distribution-based criteria,” in Proc. INTERSPEECH, 2014, pp. 1910–1914. G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,” arXiv preprint arXiv:1503.02531, 2015. Y. Chebotar and A. Waters, “Distilling knowledge from ensembles of neural networks for speech recognition.” in Proc. INTERSPEECH, 2016, pp. 3439–3443. T. Fukuda, M. Suzuki, G. Kurata, S. Thomas, J. Cui, and B. Ramabhadran, “Efficient knowledge distillation from an ensemble of teachers,” in Proc. INTERSPEECH, 2017, pp. 3697–3701. H. Sak, F. de Chaumont Quitry, T. Sainath, K. Rao et al., “Acoustic modelling with [CD-CTC-sMBR LSTM RNNs]{},” in Proc. ASRU, 2015, pp. 604–609. G. Kurata and K. Audhkhasi, “Improved knowledge distillation from bi-directional to uni-directional [LSTM]{} [CTC]{} for end-to-end speech recognition,” in Proc. SLT, 2018, pp. 411–417. S. Ghorbani, A. E. Bulut, and J. H. Hansen, “Advancing multi-accented [LSTM-CTC]{} speech recognition using a domain specific student-teacher learning paradigm,” in Proc. SLT, 2018, pp. 29–35. D. Amodei, S. Ananthanarayanan, R. Anubhai, J. Bai, E. Battenberg, C. Case, J. Casper, B. Catanzaro, Q. Cheng, G. Chen et al., “Deep speech 2: End-to-end speech recognition in [English]{} and [Mandarin]{},” in Proc. ICML, 2016, pp. 173–182. H. Sak, A. Senior, K. Rao, and F. Beaufays, “Fast and accurate recurrent neural network acoustic models for speech recognition,” arXiv preprint arXiv:1507.06947, 2015. J. G. Fiscus, “A post-processing system to yield reduced word error rates: Recognizer output voting error reduction ([ROVER]{}),” in Proc. ASRU, 1997, pp. 347–354. Y. Wang, X. Deng, S. Pu, and Z. Huang, “Residual convolutional [CTC]{} networks for automatic speech recognition,” arXiv preprint arXiv:1702.07793, 2017. L. Fritz and D. Burshtein, “Simplified end-to-end [MMI]{} training and voting for [ASR]{},” arXiv preprint arXiv:1703.10356, 2017. H. Soltau, H. Liao, and H. Sak, “Reducing the computational complexity for whole word models,” in Proc. ASRU, 2017, pp. 63–68. S. Kim, M. L. Seltzer, J. Li, and R. Zhao, “Improved training for online end-to-end speech recognition systems,” in Proc. INTERSPEECH, 2018, pp. 2913–2917. R. Takashima, S. Li, and H. Kaswai, “An investigation of a knowledge distillation method for [CTC]{} acoustic models,” in Proc. ICASSP, 2018, pp. 5809–5813. R. Sanabria and F. Metze, “Hierarchical multi task learning with [CTC]{},” in Proc. SLT, 2018, pp. 485–490. C. Yu, C. Zhang, C. Weng, J. Cui, and D. Yu, “A multistage training framework for acoustic-to-word model,” in Proc. INTERSPEECH, 2018, pp. 786–790. N. Dehak, P. J. Kenny, R. Dehak, P. Dumouchel, and P. Ouellet, “Front-end factor analysis for speaker verification,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 19, no. 4, pp. 788–798, 2011. T. N. Sainath, B. Kingsbury, V. Sindhwani, E. Arisoy, and B. Ramabhadran, “Low-rank matrix factorization for deep neural network training with high-dimensional output targets,” in Proc. ICASSP, 2013, pp. 6655–6659. [^1]: We do not focus on other major approaches to E2E ASR in this paper, such as attention-based encoder-decoder systems [@bahdanau2016end] and joint CTC/sequence-to-sequence systems [@hori2017joint]. [^2]: Knowledge distillation is also called teacher-student training, where a [student]{} model is distilled from a [teacher]{} model. [^3]: We call the highest posterior except for blank and silence at each time index a “spike” hereafter. [^4]: The logarithmic guide loss is equivalent to a frame-level cross-entropy where the target is a sequence of the output symbols with the highest posterior from the guiding model. Note that the mask $M(\bm{X})$ is derived from this target sequence. [^5]: An naïve alternative is to conduct the standard UniLSTM CTC training starting from a pre-trained guiding CTC model. By this approach, we obtained the WERs of 15.0% and 27.3%. The larger improvements obtained in [1D]{} demonstrate the advantage of our proposed guided CTC training. [^6]: “Posterior fusion of 1 guided model” is used for better readability and indicates the use of posteriors from the 1 guided model directly. [^7]: We reported a smaller accuracy gap reduction (45.2%) with a similar setting in our former paper [@kurata18:_improv_knowl_distil_from_bi]. Besides, this method could not be extended to use a posterior fusion as a teacher. [^8]: To initialize multiple word CTC models for posterior fusion and ROVER, we trained multiple phone CTC models with different parameter initialization and training data order. [^9]: By using data augmentation with speed perturbation, [@yu2018multistage] further improved the WERs to 11.4% and 20.8%.	{ "pile_set_name": "ArXiv" }
--- author: - 'F. Ursini, L. Bassani, A. Malizia, A. Bazzano, A. J. Bird, J. B. Stephen,' - 'P. Ubertini' bibliography: - 'mybib.bib' date: 'Received ...; accepted ...' title: 'The coronal temperature of NGC 4388 and NGC 2110 measured with INTEGRAL[^1]' --- [ We aim to measure the physical properties of the hot X-ray corona of two active galactic nuclei, [NGC 4388]{} and [NGC 2110]{}. ]{} [We analysed the hard X-ray (20–300 keV) [[INTEGRAL]{}]{} spectrum in conjunction with archival [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} data.]{} [The X-ray spectrum of both sources is phenomenologically well described by an absorbed cut-off power law. In agreement with previous results, we find no evidence of a Compton reflection component in these sources. We obtain a high-energy cut-off of [$200^{+ 75}_{- 40}$]{} keV for [NGC 4388]{} and [$320^{+ 100}_{- 60}$]{} keV for [NGC 2110]{}. A fit with a thermal Comptonisation model yields a coronal temperature of [$80^{+ 40}_{- 20}$]{} keV and [$75^{+ 20}_{- 15}$]{} keV, respectively, and an optical depth of approximately two, assuming a spherical geometry. The coronal temperature and luminosity of both sources are consistent with pair production that acts as a thermostat for the thermal plasma. These results emphasise the importance of good signal-to-noise X-ray data above 100 keV to probe the high-energy emission of AGNs. ]{} Introduction ============ One of the main components of active galactic nuclei (AGNs) is a hot corona, which is thought to produce the X-ray emission via thermal Comptonisation of optical and UV photons from the accretion disc. This process naturally explains the observed power-law shape of the X-ray spectra of AGNs and their high-energy cut-off, which is related to the coronal temperature. The cut-off has often been observed at $\sim 100$ keV thanks to both past and current X-ray missions, such as the Compton Gamma-Ray Observatory’s Oriented Scintillation Spectrometer Experiment [OSSE; @zdziarski2000], [[BeppoSAX]{}]{} [@perola2002], [[Swift]{}]{}’s Burst Alert Telescope [BAT; @bat70; @ricci2017], [[INTEGRAL]{}]{} [@malizia2014], and [[NuSTAR]{}]{} [@fabian2015; @tortosa2018 and references therein]. Good constraints on the high-energy cut-off have been, and are still, obtained for a number of AGNs with [[NuSTAR]{}]{}. This leads to measurements of the coronal temperature to range from 15–20 keV in the ‘coolest’ sources [@kara2017; @tortosa2017; @buisson2018; @turner2018] to $\sim 400$ keV in NGC 5506 [@matt20155506]. However, in many cases only lower limits to the cut-off have been reported [e.g. @kamraj2018] that can still be useful to constrain the geometrical and physical parameters of the hot corona [e.g. @matt2014ark120; @marinucci2015; @7213]. Thanks to its high-energy coverage, the Imager on-board [[INTEGRAL]{}]{} [IBIS; @ibis] allows us to obtain good measurements of the cut-off energy [@panessa2008; @molina2009; @derosa2012; @malizia2014] and of the coronal temperature [@lubinski2016]. In this paper, we present the [[INTEGRAL]{}]{}’s IBIS spectrum together with archival [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} observations of two of the brightest Seyfert galaxies in the hard X-rays: [NGC 4388]{} and [NGC 2110]{}. In Sect. \[sec:data\] we report the data selection and reduction. In Sect. \[sec:analysis\] we report the spectral analysis. In Sect. \[sec:discussion\] we discuss the results and summarise the conclusions. The main properties of the two sources are summarised below. NGC 4388 -------- [NGC 4388]{} is a nearby [$z=0.00842$, @lu1993], X-ray bright Seyfert galaxy, hosting a supermassive black hole of [($8.5 \pm 0.2) \times 10^{6}$]{} [M$_{\odot}$]{} [measured from the water maser; see @kuo2011]. This source has been observed by all of the major X-ray satellites. The hard X-ray spectrum is moderately absorbed by a column density of a few $\times {10^{23}}$ [cm$^{-2}$]{}, which is found to be variable on short time-scales [@elvis2004]. The soft X-ray spectrum below $\sim 2-3$ keV is dominated by emission of a hot plasma in an extended X-ray nebula at kpc scales [@matt1994; @iwasawa2003; @beckmann2004; @bianchi2006]. No Compton reflection hump has been detected with [[NuSTAR]{}]{} [@kamraj2017] despite the presence of a strong [Fe K$\alpha$]{} emission line, that likely originates from Compton-thin material [@kamraj2017]. Past results indicate the presence of a high-energy cut-off at $\sim 200$ keV, albeit with some uncertainties (see Table \[tab:past\]). From [[INTEGRAL]{}]{} data taken in 2003 in combination with [[XMM–Newton]{}]{} spectra, [@beckmann2004] reported a lower limit of 180 keV to the high-energy cut-off. The same result is obtained by [@derosa2012], also from [[INTEGRAL]{}]{}+[[XMM–Newton]{}]{} data. Using high-energy [[INTEGRAL]{}]{} and BAT data, [@molina2013] report a high-energy cut-off of [$202^{+ 111}_{- 54}$]{} keV. From [[INTEGRAL]{}]{} data taken from 2003 to 2009, [@fedorova2011] report strong variations in both the flux and spectral slope in the 20–60 keV band on a few months time-scale. These authors also report a high-energy cut-off at $\sim 200$ keV with some indications of variability between 80–100 keV and [$> 320$]{} keV, but with limited statistics. From [[INTEGRAL]{}]{} data taken from 2003 to 2010, [@lubinski2016] report a temperature of [$53^{+ 17}_{- 9}$]{} keV and an optical depth of [$2.7^{+ 0.6}_{- 0.9}$]{}. However, no constraint on the cut-off has been found with [[NuSTAR]{}]{} [@kamraj2017] and only a lower limit of $\sim 100$ keV was reported from [[BeppoSAX]{}]{} [@risaliti2002] and [[Swift]{}]{}’s BAT [@ricci2017]. From the analysis of the 58-month [[Swift]{}]{}’s BAT light curves, [@caballero2012] report the detection of hard X-ray spectral variability in the 14–195 keV band. However, [@soldi2014] do not find strong evidence of long-term hard X-ray spectral variability from BAT data up to 66 months. NGC 2110 -------- [NGC 2110]{} is another nearby [$z=0.00779$, @gallimore1999], X-ray bright Seyfert galaxy. [@diniz2015] report a black hole mass of [$2.7^{+ 3.5}_{- 2.1} \times 10^{8}$]{} [M$_{\odot}$]{}, from the relation with the stellar velocity dispersion. From [[BeppoSAX]{}]{} data, [@malaguti1999] found the X-ray spectrum to be affected by complex absorption. This has been later confirmed by [@evans2007], who find the [[Chandra]{}]{}+[[XMM–Newton]{}]{} data to be well fitted with a neutral, three-zone, partial-covering absorber. [@rivers2014] find the [[Suzaku]{}]{} data to be well fitted with a stable full-covering absorber plus a variable partial-covering absorber. A soft excess below 1.5 keV is also present [@evans2007], and possibly due to extended circumnuclear emission seen with [[Chandra]{}]{} [@evans2006]. No Compton reflection hump has been detected with [[Suzaku]{}]{} [@rivers2014] or [[NuSTAR]{}]{} [@marinucci2015], despite the presence of a complex [Fe K$\alpha$]{} line. According to the multi-epoch analysis of [@marinucci2015], the [Fe K$\alpha$]{} line is likely the sum of a constant component (from distant, Compton-thick material) and a variable one (from Compton-thin material). Concerning the high-energy cut-off, ambiguous results have been reported in literature (see Table \[tab:past\]). [@ricci2017] report a value of [$448^{+ 63}_{- 55}$]{} keV, while [@lubinski2016] report a coronal temperature of [$230^{+ 51}_{- 57}$]{} keV and an optical depth of [$0.52^{+ 0.14}_{- 0.13}$]{}. From 2008-2009 [[INTEGRAL]{}]{} data, [@beckmann2010] report a cut-off of $\sim 80$ keV with a hard photon index, but these results are not confirmed by [[NuSTAR]{}]{} [@marinucci2015]. Indeed, only lower limits to the high-energy cut-off have been found with [[NuSTAR]{}]{} [210 keV: @marinucci2015], [[Suzaku]{}]{} [250 keV: @rivers2014] and [[BeppoSAX]{}]{} [143 keV: @risaliti2002]. No hard X-ray spectral variability has been detected by [@caballero2012] and [@soldi2014] from BAT data, despite the significant flux variability. [NGC 4388]{} [NGC 2110]{} ------------------------------- -------------------------- ------------------------- ${E_{\textrm{c}}}$ [$> 145$]{} [$> 143$]{} [$> 180$]{} [$82 \pm 9$]{} [$209^{+ 44}_{- 32}$]{} [$> 250$]{} \[0.5ex\] [$202^{+ 111}_{- 54}$]{} [$> 210$]{} [$> 104$]{} [$448^{+ 63}_{- 55}$]{} \[0.5ex\] ${kT_{\textrm{e}}}$ [$53^{+ 17}_{- 9}$]{} [$230^{+ 51}_{- 57}$]{} [$190 \pm 130$]{} : Previous constraints on high-energy cut-off and coronal temperature (both in keV) reported in literature.[]{data-label="tab:past"} Data selection and reduction {#sec:data} ============================ For both sources, we collected the archival [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} data to complement the IBIS spectrum (see Table \[tab:log\]). The [[XMM–Newton]{}]{} data were processed using the [[XMM–Newton]{}]{} Science Analysis System (<span style="font-variant:small-caps;">sas</span> v18). We focused on the EPIC-pn data because of the much larger effective area compared with the MOS detectors. The source extraction radii and screening for high-background intervals were determined through an iterative process that maximises the signal-to-noise ratio [see @pico2004]. The background were extracted from circular regions with a radius of 50 arcsec, while the source extraction radii were in the range 20–40 arcsec. The spectra were binned to have at least 30 counts per spectral bin and not oversampling the instrumental resolution by a factor larger than 3. The [[NuSTAR]{}]{} data were reduced using the standard pipeline (<span style="font-variant:small-caps;">nupipeline</span>) in the [[NuSTAR]{}]{} Data Analysis Software (<span style="font-variant:small-caps;">nustardas</span>, v1.8.0), using calibration files from [[NuSTAR]{}]{} [caldb]{} v20190410. We extracted the spectra using the standard tool [nuproducts]{} for each of the two hard X-ray detectors, which reside in the corresponding focal plane modules A and B (FPMA and FPMB). We extracted the source data from circular regions with a radius of 75 arcsec, and the background from a blank area close to the source. The spectra were binned to have a signal-to-noise ratio larger than 3 in each spectral channel and not to oversample the instrumental resolution by a factor greater than 2.5. The spectra from the two detectors were analysed jointly but not combined. Both [NGC 4388]{} and [NGC 2110]{} are detected with a good signal-to-noise in the hard X-ray band with IBIS ($> 6 \sigma$ in the 150–300 keV band, see Table \[tab:src\]). Here we use the data collected by the [[INTEGRAL]{}]{} Soft Gamma-Ray Imager [ISGRI: @isgri], that is the low-energy camera of the IBIS telescope. We used data up to [[INTEGRAL]{}]{} orbit 1500, that is from the launch in 2002 to the end of January 2015. ISGRI images for each available pointing were generated in 14 energy bands using the ISDC offline scientific analysis (OSA) software [@osa], version 10.2. Count rates at the position of the source were extracted from individual images to provide light curves in 14 energy bands between 20 and 300 keV. Weighted mean fluxes were then extracted in each band and combined to produce an average source spectrum [see @bird2007; @bird2010 for details]. Corresponding weighted response (arf) files for each source were created by weighting the published matrices for each response validity period according to the photon fluence from the source during that validity period. This method takes into account both the different exposure in each validity period, and any source variability present. A single rmf response is used, corresponding to the standard rmf, rebinned according to the 14 channels used for spectral analysis. For [NGC 4388]{}, the spectrum was derived from 4770 pointings with a total on-source time (not adjusted for off-axis response) of 7.5 Ms. The corresponding figures for [NGC 2110]{} are 2249 pointings for 3.8 Ms total on-source time. In both cases, the observations are spread reasonably uniformly across the time period analysed. $\sigma_{100-120}$ $\sigma_{120-150}$ $\sigma_{150-300}$ -------------- -------------------- -------------------- -------------------- [NGC 4388]{} 22.2 17.6 8.6 [NGC 2110]{} 17.6 12.9 6.6 : IBIS source significance of [NGC 4388]{} and [NGC 2110]{} in three energy bands: 100–120 keV, 120–150 keV, 150–300 keV.[]{data-label="tab:src"} [c c c c c]{} Satellite & Obs. Id. & Start time (<span style="font-variant:small-caps;">utc</span>) & Net exp. & label\ &&yyyy-mm-dd&(ks) &\ \ [[XMM–Newton]{}]{}& 0110930301 & 2002-07-07 & 3.8 & XMM1\ [[XMM–Newton]{}]{}& 0110930701 & 2002-12-12 & 7.3 & XMM2\ [[XMM–Newton]{}]{}& 0675140101 & 2011-06-18 & 37.7& XMM3\ [[NuSTAR]{}]{}& 60061228002 & 2013-12-27 & 21.4 & NUS\ \ [[XMM–Newton]{}]{}& 0145670101 & 2003-03-05 & 44.5 & XMM\ [[NuSTAR]{}]{}& 60061061002 & 2012-10-05 & 15.5& NUS1\ [[NuSTAR]{}]{}& 60061061004 & 2013-02-14 & 12.0& NUS2\ Spectral analysis {#sec:analysis} ================= Spectral analysis and model fitting was carried out with the [[xspec]{}]{} 12.10 package [@arnaud1996], using the ${\chi^{2}}$ minimisation technique. Errors are quoted at the 90 per cent confidence level for one interesting parameter. We assumed the element abundances of [@angr] and the photoelectric absorption cross-sections of [@vern]. We fitted the IBIS spectra over the full 20–300 keV energy band. NGC 4388 -------- The emission of this source below $\sim 3$ keV is known to be dominated by emission from a hot plasma in an extended X-ray nebula at kpc scales [@beckmann2004]. Therefore, since our main focus is the AGN-dominated hard X-ray emission, we restricted the broad-band spectral analysis to the data above 4 keV. We fitted the [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} spectra over the 4–10 keV and 4–79 keV energy bands, respectively. ### The IBIS spectrum As a first step, we focused on the IBIS spectrum, with a twofold purpose: first, testing for the presence of a curvature at high energies and, secondly, a consistency check with the low-energy spectra. We performed fits assuming absorption by a column density ${N_{\textrm{H}}}= {2.7 \times 10^{23}}$ [cm$^{-2}$]{} [@beckmann2004] since this value is not constrained by the data above 20 keV. Fitting the IBIS data with a power law, we obtain a poor fit (${\chi^{2}/\textrm{dof}}=17/10$) with negative residuals above 100 keV (Fig. \[fig:ratios\], first panel). To reproduce this curvature, we first tried to include Compton reflection, replacing the cut-off power law with [[pexrav]{}]{} [@pexrav]. This model includes Compton reflection off a neutral slab of infinite column density. We fixed the inclination angle at 60 deg since the fit was not sensitive to this parameter. We first fixed the reflection fraction ($\mathcal{R} = \Omega / 2 \pi$, where $\Omega$ is the solid angle subtended by the reflector) at 0.5. The model also includes an exponential high-energy cut-off, which we left free to vary. This yields a slightly better fit (${\chi^{2}/\textrm{dof}}=14/9$), but still with the indication of a curvature (Fig. \[fig:ratios\], second panel). Indeed, leaving $\mathcal{R}$ free to vary, the value is pegged at zero with an upper limit of 0.3. Then, we replaced [[pexrav]{}]{} with a cut-off power law (equivalent to [[pexrav]{}]{} with $\mathcal{R}=0$) finding a good fit (${\chi^{2}/\textrm{dof}}=8/9$) and a cut-off energy of [$300^{+ 400}_{- 100}$]{} keV. Finally, we replaced the phenomenological cut-off power law with the Comptonisation model [[compps]{}]{} [@compps]. We obtain a good fit (${\chi^{2}/\textrm{dof}}=7/9$) and a temperature of [$100^{+ 150}_{- 40}$]{} keV. We also repeated the same analysis assuming the partial-covering absorption with ${N_{\textrm{H}}}= {6.5 \times 10^{23}}$ [cm$^{-2}$]{} and a covering factor of 0.9, consistent with [[NuSTAR]{}]{} data [@kamraj2017], obtaining analogous results. Then we extrapolated the best-fitting cut-off power law down to 4 keV, without re-fitting, to compare it with the [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} spectra (Fig. \[fig:extrap\_4388\]). Despite the flux variability, the XMM1, XMM2, and NUS spectral shape appears in good agreement with the extrapolation, with minor discrepancies likely due to an imperfect modelling of absorption. XMM3 instead shows a significant difference in spectral shape. ![\[fig:ratios\] Residuals of fits of IBIS spectrum with different models. Upper panel: simple power law. Second panel: power law plus reflection ([[pexrav]{}]{}). Third panel: exponentially cut-off power law. Lower panel: thermal Comptonisation model ([[compps]{}]{}). ](ratios_4388){width="\columnwidth"} ![\[fig:extrap\_4388\] Upper panel: [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} spectra of NGC 4388 with the cut-off power law that best fits IBIS. Lower panel: data/model ratio. Only [[NuSTAR]{}]{}’s FPMA data are shown for clarity. The data were binned for plotting purposes.](extrap_4388){width="\columnwidth"} ### The broad-band fit As a second step, we performed a broad-band fit including the lower energy data from [[XMM–Newton]{}]{} and [[NuSTAR]{}]{}. We used the XMM1, XMM2, and NUS spectra, that is to say those in good agreement with IBIS. XMM3, despite having the deepest exposure, likely represents a different spectral state compared with the average and is discussed in Appendix \[appendix\]. First, we fitted the data with a phenomenological model. Following [@kamraj2017], we used a model consisting of a cut-off power law plus a Gaussian [Fe K$\alpha$]{} line, modified by partial covering absorption. The model reads <span style="font-variant:small-caps;">pcfabs(cutoffpl+zgauss)</span> in [[xspec]{}]{} terminology. We included a cross-normalisation constant free to vary among the XMM1, XMM2, and NUS observations, to account for the flux variations. We first assumed all the other parameters to be constant, meaning that they were tied among the different spectra. We obtained a good fit (${\chi^{2}/\textrm{dof}}=455/435$), but with some residuals near 6.4 keV indicating variability of the iron line flux not related to the primary continuum. We thus left the normalisation of the Gaussian line free to vary among the different spectra. We obtained an excellent fit (${\chi^{2}/\textrm{dof}}=433/432$), with $\Gamma \simeq 1.6$ and a well constrained cut-off energy of [$200^{+ 75}_{- 40}$]{} keV. We found no significant improvement by leaving the column density ${N_{\textrm{H}}}$, the covering factor ${\textrm{C}_{\textrm{F}}}$, or the photon index free to vary among the different observations. The best-fitting parameters are reported in Table \[tab:fits\_4388\], while the contour plots of the cut-off energy and photon index are shown in Fig. \[fig:cont\_cutoffpl\]. We also tried to include a reflection component, replacing the cut-off power law with [[pexrav]{}]{} and leaving the reflection fraction ${\mathcal{R}}$ free and tied among the observations. The fit is not improved and we obtain only an upper limit ${\mathcal{R}}< 0.12$. We note that the large equivalent width of the [Fe K$\alpha$]{} line is consistent with that reported by [@kamraj2017], who suggest the presence of a large amount of Compton-thin line-emitting material. Then, we fitted the data using a more physical Comptonisation model. We replaced the cut-off power law with [[compps]{}]{}, assuming a spherical geometry ($\textsc{geom}=0$). We assumed a seed photon temperature of 100 eV. Fixing this parameter at other values, like 10 eV, does not alter the results significantly. We fitted for the electron temperature ${kT_{\textrm{e}}}$ and the Compton parameter $y = 4\tau ({kT_{\textrm{e}}}/m_{\textrm{e}} c^2)$. We chose to use the Compton parameter instead of the optical depth to minimise the model degeneracy since the temperature and optical depth are generally correlated in the fitting procedure [e.g. @poptestingcompt2001; @pop2013mrk509]. We obtained a very good fit (${\chi^{2}/\textrm{dof}}=444/432$), with ${kT_{\textrm{e}}}= {80^{+ 40}_{- 20}}$ keV and $y = {1.08^{+ 0.05}_{- 0.08}}$, corresponding to an optical depth $\tau={1.7 \pm 0.7}$. The best-fitting parameters (${kT_{\textrm{e}}}, y$, and normalisation) are reported in Table \[tab:fits\_4388\], while the ${kT_{\textrm{e}}}- y$ contour plots are shown in Fig. \[fig:cont\_compps\]. The cross-normalisations, absorption and iron line parameters are consistent within the errors with the cut-off power law fit. ![\[fig:cont\_cutoffpl\] Contour plots of cut-off energy vs. photon index for NGC 4388 (black) and NGC 2110 (red). Solid, dashed, and dotted lines correspond to 68, 90, and 99 per cent confidence level, respectively.](cont_4388+2110_cutoffpl){width="\columnwidth"} ![\[fig:cont\_compps\] Contour plots of Compton parameter $y$ vs. electron temperature ${kT_{\textrm{e}}}$ for NGC 4388 (black) and NGC 2110 (red). Solid, dashed, and dotted lines correspond to 68, 90, and 99 per cent confidence level, respectively. Grey dotted lines correspond to constant values of optical depth.](cont_4388+2110_compps_new){width="\columnwidth"} ![\[fig:plot\_4388\] Upper panel: spectra of NGC 4388 with best-fitting [[compps]{}]{} model. Second panel: residuals, plotted as $\Delta \chi = $ (data-model)/error. Third panel: best-fitting model $E^2 f(E)$. The data were binned for plotting purposes.](plot_4388){width="\columnwidth"} [ l c c c c]{} & all obs. & XMM1 & XMM2 & NUS\ \ $\Gamma$ & [$1.60^{+ 0.07}_{- 0.06}$]{}&(t)&(t)&(t)\ ${E_{\textrm{c}}}$ (keV)& [$200^{+ 75}_{- 40}$]{}&(t)&(t)&(t)\ $N_{\textsc{pow}}$ ([$10^{-2}$]{}) & [$1.8^{+ 0.4}_{- 0.2}$]{}&(t)&(t)&(t)\ $K_{\textrm{IBIS--pn}}$ && [$0.29 \pm 0.02$]{} &[$0.86 \pm 0.04$]{}& -\ $K_{\textrm{IBIS--NusA}}$ &&- &-& [$0.29 \pm 0.01$]{}\ $K_{\textrm{IBIS--NusB}}$ &&- &-& [$0.30 \pm 0.01$]{}\ ${N_{\textrm{H}}}$ ([$10^{23}$]{} [cm$^{-2}$]{}) & [$3.8^{+ 0.5}_{- 0.2}$]{}&(t)&(t)&(t)\ ${\textrm{C}_{\textrm{F}}}$ & [$0.95^{+ 0.03}_{- 0.02}$]{} &(t)&(t)&(t)\ $E_{\textsc{ga}}$ (keV) & [$6.39^{+ 0.02}_{- 0.01}$]{}&(t)&(t)&(t)\ $N_{\textsc{ga}}$ ([$10^{-4}$]{}) && [$3.5 \pm 0.8$]{} & [$1.6^{+ 0.3}_{- 0.2}$]{} & [$2.7 \pm 0.6$]{}\ EW$_{\textsc{ga}}$ (eV) && [$390^{+ 100}_{- 90}$]{} & [$180 \pm 30$]{} & [$300 \pm 70$]{}\ ${\chi^{2}/\textrm{dof}}$ &433/432&&&\ \ ${kT_{\textrm{e}}}$ (keV)& [$80^{+ 40}_{- 20}$]{}&(t)&(t)&(t)\ $y$ & [$1.08^{+ 0.05}_{- 0.08}$]{}&(t)&(t)&(t)\ $N_{\textsc{compps}}$ ([$10^{4}$]{}) &[$6.5^{+ 1.5}_{- 1.0}$]{} &(t)&(t)&(t)\ ${\chi^{2}/\textrm{dof}}$ &444/432&&&\ NGC 2110 -------- The soft X-ray emission of [NGC 2110]{} below $\sim 2-3$ keV is affected by variable absorption from a complex medium [@evans2007; @rivers2014]. Therefore, we restricted the spectral analysis to the data above 3 keV. We fitted the [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} spectra over the 3–10 keV and 3–79 keV energy bands. ### The IBIS spectrum As we did for [NGC 4388]{}, we first focused on the IBIS spectrum. The spectrum above 20 keV is not strongly altered by absorption, however we assumed ${N_{\textrm{H}}}= {4 \times 10^{22}}$ [cm$^{-2}$]{} [@marinucci2015]. Fitting the data with a power law, we obtained a statistically good fit (${\chi^{2}/\textrm{dof}}=9/10$) with a hint of a curvature above 120 keV (Fig. \[fig:ratios2110\], first panel). We also tested three other models, as done for [NGC 4388]{}, namely: [[pexrav]{}]{} with $\mathcal{R} = 0.5$, a cut-off power law and [[compps]{}]{}. We obtained the same improvement in terms of ${\chi^{2}}$ for all the three models (${\chi^{2}/\textrm{dof}}=5/9$; see Fig. \[fig:ratios2110\]). Finally, we extrapolated the best-fitting cut-off power law down to 3 keV and compared it with the [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} spectra (Fig. \[fig:extrap\_2110\]). Despite the significant flux variability in the 3–79 keV band, there is no strong evidence of significant spectral variability among the different observations. ![\[fig:ratios2110\] Residuals of fits of IBIS spectrum with different models. Upper panel: simple power law. Second panel: power law plus reflection ([[pexrav]{}]{}). Third panel: exponentially cut-off power law. Lower panel: thermal Comptonisation model ([[compps]{}]{}). ](ratios_2110){width="\columnwidth"} ![\[fig:extrap\_2110\] Upper panel: [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} spectra of NGC 2110 with the cut-off power law that best fits IBIS. Lower panel: data/model ratio. Only [[NuSTAR]{}]{}’s FPMA data are shown for clarity. The data were binned for plotting purposes.](extrap_2110){width="\columnwidth"} ### The broad-band fit Next, we included the [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} data. We first fitted the data with a phenomenological model consisting of an absorbed cut-off power law plus a Gaussian [Fe K$\alpha$]{} line. In [[xspec]{}]{}, the model reads <span style="font-variant:small-caps;">phabs(cutoffpl+zgauss)</span>. According to the results of [@marinucci2015], the flux of the iron line varies among the [[XMM–Newton]{}]{} and the two [[NuSTAR]{}]{} observations, while the values of the line energy and intrinsic width are consistent with each other within the errors. We thus left the line flux free to vary among the observations, while the intrinsic width was tied. Keeping all the other parameters tied among the different spectra, we found a good fit (${\chi^{2}/\textrm{dof}}=975/1009$). However, we found an improvement by leaving the absorbing column density free to vary among the spectra, finally obtaining ${\chi^{2}/\textrm{dof}}=947/1007$ (${\Delta \chi^{2}}/{\Delta \textrm{dof}}= -28/-2$; the probability of chance improvement is [$ 4 \times 10^{-7}$]{} from an F-test), with $\Gamma \simeq 1.65$ and a ${E_{\textrm{c}}}= {320^{+ 100}_{- 60}}$ keV. No significant improvement is found by leaving the photon index free to vary, and the cut-off energy is consistent with the value reported above. The best-fitting parameters are reported in Table \[tab:fits\_2110\], while the contour plots of the cut-off energy and photon index are shown in Fig. \[fig:cont\_cutoffpl\]. To check for the presence of a reflection component, we replaced the cut-off power law with [[pexrav]{}]{}, leaving the reflection fraction ${\mathcal{R}}$ free and tied among the observations. The fit is not improved and we obtain a stringent upper limit ${\mathcal{R}}< 0.02$. Then, we replaced the cut-off power law with [[compps]{}]{}, assuming a spherical geometry and a seed photon temperature of 100 eV. We obtained a good fit (${\chi^{2}/\textrm{dof}}=971/1007$), with ${kT_{\textrm{e}}}= {75^{+ 20}_{- 15}}$ keV and $y = {1.20^{+ 0.01}_{- 0.02}}$, corresponding to an optical depth $\tau={2.1 \pm 0.5}$. The best-fitting parameters are reported in Table \[tab:fits\_2110\], while the ${kT_{\textrm{e}}}- y$ contour plots are shown in Fig. \[fig:cont\_compps\]. [ l c c c c]{} & all obs. & XMM & NUS1 & NUS2\ \ $\Gamma$ & [$1.65 \pm 0.02$]{}&(t)&(t)&(t)\ ${E_{\textrm{c}}}$ (keV)& [$320^{+ 100}_{- 60}$]{}&(t)&(t)&(t)\ $N_{\textsc{pow}}$ ([$10^{-2}$]{}) & [$2.27 \pm 0.09$]{}&(t)&(t)&(t)\ $K_{\textrm{IBIS--pn}}$ && [$0.34 \pm 0.01$]{} &-&-\ $K_{\textrm{IBIS--NusA}}$ &&- &[$2.63 \pm 0.06$]{}& [$1.79 \pm 0.05$]{}\ $K_{\textrm{IBIS--NusB}}$ && - &[$2.71 \pm 0.06$]{}& [$1.81 \pm 0.05$]{}\ ${N_{\textrm{H}}}$ ([$10^{22}$]{} [cm$^{-2}$]{}) & & [$4.4 \pm 0.2$]{}&[$3.5 \pm 0.3$]{}&[$3.6 \pm 0.3$]{}\ $E_{\textsc{ga}}$ (keV) & [$6.42 \pm 0.01$]{}&(t)&(t)&(t)\ $\sigma_{\textsc{ga}}$ (eV) & [$40 \pm 20$]{}&(t)&(t)&(t)\ $N_{\textsc{ga}}$ ([$10^{-4}$]{}) && [$1.5 \pm 0.2$]{} & [$0.23 \pm 0.11$]{} & [$0.72 \pm 0.16$]{}\ EW$_{\textsc{ga}}$ (eV) && [$140^{+ 10}_{- 20}$]{} & [$20 \pm 10$]{} & [$70^{+ 10}_{- 20}$]{}\ ${\chi^{2}/\textrm{dof}}$ &947/1007&&&\ \ ${kT_{\textrm{e}}}$ (keV) & [$75^{+ 20}_{- 15}$]{}&(t)&(t)&(t)\ $y$ & [$1.20^{+ 0.01}_{- 0.02}$]{}&(t)&(t)&(t)\ $N_{\textsc{compps}}$ ([$10^{4}$]{}) &[$5.6^{+ 0.4}_{- 0.3}$]{} &(t)&(t)&(t)\ ${\chi^{2}/\textrm{dof}}$ &971/1007&&&\ ![\[fig:plot\_2110\] Upper panel: spectra of NGC 2110 with best-fitting [[compps]{}]{} model. Second panel: residuals, plotted as $\Delta \chi = $ (data-model)/error. Third panel: best-fitting model $E^2 f(E)$. The data were binned for plotting purposes.](plot_2110){width="\columnwidth"} Discussion and conclusions {#sec:discussion} ========================== We presented the hard X-ray spectra of the two X-ray bright Seyferts [NGC 4388]{} and [NGC 2110]{}, from [[INTEGRAL]{}]{}’s IBIS data taken from 2003 to early 2015. From a joint fit with [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} data, we find that the spectrum up to 300 keV is in both cases well described by an absorbed power law with a well-constrained high-energy cut-off. In agreement with past observations, we find no evidence of a Compton hump. The primary continuum is generally consistent with being constant in spectral slope among the time-averaged IBIS spectrum and the [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} observations, with some absorption variability seen in [NGC 2110]{}. This is consistent with the lack of clear evidence of strong hard X-ray spectral variability in these objects [@soldi2014]. The exception is the 2011 [[XMM–Newton]{}]{} observation of [NGC 4388]{}, which has a flatter and less obscured spectrum (see Appendix \[appendix\]). On the other hand, the cross-normalisation constants indicate a significant flux variability in the [[XMM–Newton]{}]{} and [[NuSTAR]{}]{} energy bands, up to a factor of approximately eight in [NGC 2110]{}. Fitting the data with a phenomenological cut-off power law, we obtain photon indices of [$1.60^{+ 0.07}_{- 0.06}$]{} for [NGC 4388]{} and of [$1.65 \pm 0.02$]{} for [NGC 2110]{}, both perfectly consistent with the values previously reported from [[NuSTAR]{}]{} data alone [@kamraj2017; @marinucci2015]. As for the cut-off energies, we obtain [$200^{+ 75}_{- 40}$]{} keV for [NGC 4388]{} and [$320^{+ 100}_{- 60}$]{} keV for [NGC 2110]{}; the latter measurement is consistent with the lower limit of $\sim 200$ keV previously reported from [[Suzaku]{}]{} and [[NuSTAR]{}]{} data [@rivers2014; @marinucci2015]. Both values are also consistent within the errors with those reported by [@ricci2017]. In the case of [NGC 4388]{}, the measurement is consistent with previous [[INTEGRAL]{}]{} results [@fedorova2011; @molina2013]. However, the IBIS spectrum analysed here shows a clear and unambiguous high-energy turnover, not related to a putative reflection hump which could affect the results [@molina2013]. The cut-off energies of the two sources are relatively high compared with both the [[INTEGRAL]{}]{} high-energy cut-off distribution reported by [@malizia2014], which has a mean of 128 keV with a spread of 46 keV, and with the values recently measured with [[NuSTAR]{}]{} [@tortosa2018; @lanzuisi2019]. Among the 19 bright sources observed by [[NuSTAR]{}]{} and discussed by [@tortosa2018], only two have a cut-off $>200$ keV and a coronal temperature $>70$ keV with a percent error $\lesssim 50\%$. From a fit with a thermal Comptonisation model and assuming a spherical geometry, we obtain in both cases a coronal temperature of 70-80 keV and an optical depth close to two. In the case of [NGC 4388]{}, the temperature of [$80^{+ 40}_{- 20}$]{} keV is consistent within the errors with the upper limit of 70 keV reported by [@lubinski2016], who use the same model (i.e. [[compps]{}]{} with spherical geometry). In the case of [NGC 2110]{}, instead, the temperature of [$75^{+ 20}_{- 15}$]{} keV is much lower than the value reported by [@lubinski2016], who find a lower limit of 160 keV. However, [@lubinski2016] also report a relatively large reflection fraction of $\sim 0.6$, which is not consistent with the [[NuSTAR]{}]{} and [[Suzaku]{}]{} constraints [as already noted by @lubinski2016 Sect. 5.1.3]. We verified that imposing ${\mathcal{R}}= 0.6$ we obtain a temperature of $\sim 340$ keV, but with a worse fit ($\Delta {\chi^{2}}= +12$; F-test probability of [$ 4.4 \times 10^{-4}$]{}). [@lubinski2016] also note that the discrepancy could be due to the use of [[INTEGRAL]{}]{} data up to 2010 only, and that adding more recent [[INTEGRAL]{}]{} data result in a lower value of ${\mathcal{R}}$. Our results are roughly consistent with the temperature being a factor of two or three smaller than the cut-off energy, depending on the optical depth [@pop5548; @poptestingcompt2001]. The difference is actually by a factor of four in [NGC 2110]{}, but since the high-energy turnover of a Comptonisation spectrum is much sharper than an exponential cut-off [e.g. @stern1995; @zdziarski2003; @nied2019], this difference should not be overinterpreted [@poptestingcompt2001]. In both sources, the [Fe K$\alpha$]{} emission line is not accompanied by a significant Compton reflection hump, suggesting that the line originates from Compton-thin material, such as the broad-line region [@kamraj2017; @marinucci2015] or a Compton-thin torus [@borus]. The lack of a Compton hump due to reflection off the accretion disc or an optically thick torus could also indicate that the X-ray corona does not effectively illuminate this surrounding matter. This could be a geometrical effect, if the disc and the torus subtend a small solid angle from the corona. Alternatively, the X-ray coronal emission could be anisotropic [@beckmann2004; @kamraj2017]. For example, if the X-ray corona is outflowing at relativistic velocities, its emission would be beamed away from the disc, possibly producing a weaker reflection component [@malzac2001]. The 2–10 keV luminosity extrapolated from the IBIS time-averaged spectrum is [($1.3 \pm 0.1) \times 10^{43}$]{} [ergs s$^{-1}$]{} for [NGC 4388]{} and [($1.33 \pm 0.03) \times 10^{43}$]{} [ergs s$^{-1}$]{}for [NGC 2110]{}, respectively. Using the bolometric correction of [@marconi2004], we estimate a bolometric luminosity of $ \sim {2.5 \times 10^{44}}$ [ergs s$^{-1}$]{} for both sources. Given the different black hole mass, this yields a different accretion rate in Eddington units, of $\sim 0.23$ for [NGC 4388]{} and $\sim 0.01$ for [NGC 2110]{}. The latter is in agreement with the estimate reported by [@marinucci2015], with the caveat that the estimate of black hole mass in [NGC 2110]{} has a large uncertainty. The X-ray corona of AGNs can produce radio synchrotron emission , as recently observed in the two bright Seyferts IC 4329A and NGC 985 . This could explain the radio emission seen in radio-quiet Seyferts, whose origin is under debate [@panessa2019 and references therein]. In general, a coronal origin for the radio emission is suggested by the observed relation between the radio luminosity at 5 GHz $L_R$ and the X-ray (0.2–20 keV) luminosity $L_X$, with $L_R/L_X={10^{-5}}$ like in coronally active stars . Another hint is the presence of a mm-wave excess with respect to the spectral slope extrapolated from low frequencies [@behar2015; @behar2018]. [NGC 4388]{} has a complex radio structure [@falcke1998; @mundell2000], with a faint core having a flux density of 1.3 mJy at 1.6 GHz and $<0.55$ mJy at 5 GHz [@gp2009]. This source exhibits a mm excess [@behar2018], but it also has a $L_R/L_X$ ratio $< {10^{-7}}$ [from the upper limit to the 5-GHz luminosity reported by @gp2009]. [NGC 2110]{} shows an S-shaped jet-counterjet radio structure, with a relatively bright core having a flux density of 100 mJy at 5 GHz [@ulv1983; @mundell2000]. [@beckmann2010] argue that this object is borderline between a radio-quiet Seyfert and a radio-loud galaxy with a [@fr1974] type I morphology. No mm excess is found in [NGC 2110]{}, whose radio and mm emission is consistent with originating from the jet [@behar2018]. We conclude that the radio emission is unlikely to be of coronal origin in the two sources discussed here. According to our results, [NGC 4388]{} and [NGC 2110]{} have a corona with similar physical parameters and producing the same X-ray luminosity. However, they seem to have different accretion/ejection parameters: while [NGC 4388]{} has a high Eddington ratio and a weak radio core, [NGC 2110]{} has a relatively low Eddington ratio and a bright radio core. This is consistent with the inverse relation between radio loudness and Eddington ratio, found both in radio-quiet and radio-loud objects [@ho2002; @sikora2007; @panessa2007]. The physical origin of this anticorrelation is a matter of debate . Finally, we put [NGC 4388]{} and [NGC 2110]{} in the compactness-temperature ($\ell - {\Theta_{\textrm{e}}}$) diagram [@fabian2015] by calculating the dimensionless coronal temperature ${\Theta_{\textrm{e}}}= {kT_{\textrm{e}}}/{m_{\textrm{e}}}c^2$ and the compactness parameter $\ell = L \sigma_\textrm{T} / R {m_{\textrm{e}}}c^3$, where $L$ is the luminosity and $R$ is the radius of the corona. Following [@fabian2015], we used the luminosity of the primary continuum extrapolated to the 0.1–200 keV band. For both sources we assumed the flux level of the IBIS spectra. Denoting by $R_{10}$ the coronal size in units of 10 gravitational radii, we obtained $\ell \simeq 130 \, (R_{10})^{-1}$ and ${\Theta_{\textrm{e}}}= {0.16^{+ 0.08}_{- 0.04}}$ for [NGC 4388]{}, while for [NGC 2110]{} we obtained $\ell \simeq 5 \, (R_{10})^{-1}$ and ${\Theta_{\textrm{e}}}= {0.15^{+ 0.04}_{- 0.03}}$. Then, assuming a radius of 10 gravitational radii, both sources would lie below the pair runaway line, that is a maximum-luminosity curve delimiting a forbidden region in which pair production would exceed annihilation [@fabian2015]. Our results are thus in agreement with a scenario in which the so-called pair thermostat [e.g. @svensson1984; @zdziarski1985] controls the coronal temperature in AGNs [@fabian2015], possibly with a contribution from non-thermal particles [@fabian2017]. The compactness estimated for [NGC 2110]{} is not very high, but given the uncertainty on the black hole mass (and on the coronal size), it could easily be $>10$. Interestingly, for a pair-dominated corona, the spectral shape is almost constant unless the luminosity variations are strong [e.g. by a factor of 20: see @hmg1997]. We note that the detection of a high-energy roll-over in the two sources discussed above is enabled by a broad-band energy coverage, extending up to 300 keV thanks to the unique capabilities of IBIS. These results demonstrate that spectral information above 100 keV is crucial to constrain the properties of the hot corona in AGNs, especially in the high-temperature regime. Extending the census of AGN coronal parameters is in turn essential to test the pair thermostat scenario and to search for correlations between the different coronal parameters, that may carry significant information on the disc/corona system [@tortosa2018; @middei2019]. Future high-energy missions like HEX-P [@hexp] and ASTENA [@fuschino2018] will be key to provide single-epoch measurements of the high-energy cut-off. We thank the referee for comments that improved the manuscript. We acknowledge financial support from ASI and INAF under INTEGRAL ‘accordo ASI/INAF 2013-025-R1’. The 2011 [[XMM–Newton]{}]{} spectrum of [NGC 4388]{} {#appendix} ====================================================== To analyse the 4–10 keV spectrum of the XMM3 obs. (see Table \[tab:log\]), we first assumed the best-fitting model described in Sect. \[sec:analysis\], i.e. a cut-off power law plus a narrow Gaussian line modified by partial covering absorption. We fixed the cut-off energy at 200 keV, since we used only low-energy [[XMM–Newton]{}]{} data. We obtained a poor fit with ${\chi^{2}/\textrm{dof}}=153/85$ and strong, negative residuals in the 6.5-7.0 keV band (see Fig. \[fig:xmm3\]). We thus added one narrow Gaussian line in absorption, finding an improved but still unacceptable fit (${\chi^{2}/\textrm{dof}}=116/83$). Then, we added a further absorption line, obtaining a good fit with ${\chi^{2}/\textrm{dof}}=71/81$. The best-fitting parameters are reported in Table \[tab:fit\_xmm3\]. The photon index is [$1.50^{+ 0.02}_{- 0.08}$]{}, that is to say flatter than the value of [$1.60^{+ 0.07}_{- 0.06}$]{} reported in Sect. \[sec:analysis\]. The absorbing column density is also lower, amounting to $\sim {2.4 \times 10^{23}}$ [cm$^{-2}$]{} instead of $\sim {3.8 \times 10^{23}}$ [cm$^{-2}$]{}. The two absorption lines are found at rest-frame energies of [$6.72^{+ 0.02}_{- 0.03}$]{} keV and [$6.99^{+ 0.04}_{- 0.02}$]{} keV, and can be identified as the [K $\alpha$]{} lines of [Fe [xxv]{}]{} at 6.7 keV and of [Fe [xxvi]{}]{} at 6.966 keV, respectively. Ionized iron absorption lines have been detected in a number of objects. In many cases the line energies are blueshifted, implying an origin from a fast outflow [e.g. @tombesi2010]. However, absorption from ‘static’ or low-velocity ionised material has also been observed in some sources [@bianchi2005 and references therein]. For example, [@reeves2004] detected a variable absorption line from highly ionised iron in an [[XMM–Newton]{}]{} spectrum of the Seyfert 1 [NGC 3783]{}. This line was found to be stronger at higher continuum flux levels, and it was ascribed to a component of the warm absorber present in that source [@reeves2004]. A proper test of this hypothesis is difficult in an obscured object like [NGC 4388]{}, and is out of the scope of the present work. However, following [@reeves2004], we can tentatively ascribe the emergence of ionised iron absorption lines in the high-flux 2011 spectrum to a change in the ionisation level of the circumnuclear absorbing material due to an increase of the illuminating flux. ![\[fig:xmm3\] Upper panel: 2011 [[XMM–Newton]{}]{} spectrum of [NGC 4388]{} fitted with a an absorbed power law plus a narrow Gaussian line. Lower panel: residuals as $\Delta \chi$.](xmm3){width="\columnwidth"} -------------------------------------------------- ------------------------------ $\Gamma$ [$1.50^{+ 0.02}_{- 0.08}$]{} $N_{\textsc{pow}}$ ([$10^{-2}$]{}) [$1.8^{+ 0.1}_{- 0.3}$]{} ${N_{\textrm{H}}}$ ([$10^{23}$]{} [cm$^{-2}$]{}) [$2.4^{+ 0.1}_{- 0.4}$]{} ${\textrm{C}_{\textrm{F}}}$ [$> 0.95$]{} $E_1$ (keV) [$6.72^{+ 0.02}_{- 0.03}$]{} $N_1$ ([$10^{-4}$]{}) [$-3.2 \pm 0.7$]{} EW$_1$ (eV) [$-30 \pm 10$]{} $E_2$ (keV) [$6.99^{+ 0.04}_{- 0.02}$]{} $N_2$ ([$10^{-4}$]{}) [$-3.2 \pm 0.7$]{} EW$_2$ (eV) [$-40 \pm 10$]{} $E_3$ (keV) [$6.40^{+ 0.01}_{- 0.02}$]{} $N_3$ ([$10^{-4}$]{}) [$5.7 \pm 0.8$]{} EW$_3$ (eV) [$50 \pm 10$]{} -------------------------------------------------- ------------------------------ : \[tab:fit\_xmm3\] Best-fitting parameters for the XMM3 obs. of [NGC 4388]{}. The model includes an absorbed power law plus three narrow Gaussian lines (line 1 and 2 in absorption, line 3 in emission). [^1]: The reduced spectra (FITS files) are available at the CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or via http://cdsweb.u-strasbg.fr/cgi-bin/qcat?J/A+A/...	{ "pile_set_name": "ArXiv" }
--- abstract: 'It is investigated under which conditions an adiabatic adaption of the dynamic and spectral information of vector mesons to the changing medium in heavy ion collisions, as assumed in schematic model calculations and microscopic transport simulations, is a valid assumption. Therefore time dependent medium modifications of low mass vector mesons are studied within a nonequilibrium quantum field theoretical description. Timescales for the adaption of the spectral properties are given and non-equilibrium dilepton yields are calculated, leading to the result that memory effects are not negligible for most scenarios.' author: - 'B. Schenke$^{1}$ and C. Greiner$^{1}$' bibliography: - 'schenke.bib' title: 'Dilepton production from non-equilibrium hot hadronic matter' --- Introduction and Motivation {#Intro} ============================ High energy heavy ion reactions allow for studying strongly interacting matter under extreme conditions, i.e., high densities and temperatures. Photons and dileptons do not undergo strong interactions and thus may carry undistorted information on the early hot and dense phases of the fireball, because the production rates increase rapidly with temperature. Dilepton spectra are expected to play a central role in inferring the restoration of the spontaneous breaking of chiral symmetry from heavy ion reactions. In the low mass region they couple directly to the light vector mesons and reflect their mass distribution. They are thus considered the prime observable in studying mass (de-)generation related to the restoration of spontaneous chiral symmetry breaking. In this work we investigate dilepton production from the hot hadronic medium and we will concentrate on the low mass region, particularly the medium modifications of the $\rho$-meson. The medium and hence also possibly the properties of the regarded mesons undergo substantial changes over time. Such scenarios have been described within transport calculations using some quantum mechanically inspired off-shell propagation [@Effenberger:1999ay; @Cassing:1999wx]. It emerges the important question, whether a quasi instantaneous adaption of the dynamic and spectral information to the changing medium, as assumed in more schematic fireball model calculations [@Rapp:1999ej] and microscopic transport simulations, is a suitable assumption or whether the vector meson’s spectral information reacts to changes with a certain “quantum mechanical” retardation. The transport description of off-shell excitations is an open field of research and necessary in order to understand the transport dynamics of resonances. We employ a nonequilibrium quantum field theoretical description based on the formalism established by Schwinger and Keldysh [@Schw61; @Ke64]. We give a formula for the dynamic dilepton production rate and simulate modifications of the light vector mesons due to the dynamically changing medium in heavy ion collisions by parameterizing a certain time dependence of the $\rho$-meson self energies. We are able to analyze the mesons’ dynamic spectral properties as well as the resulting dilepton rate and the yield from an evolving fireball and compare to the quantities computed assuming adiabaticity. The nonequilibrium production rate {#dileptonproduction} ================================== We utilize the Schwinger-Keldysh formalism in order to derive the dynamic non-equilibrium rate of produced electron-positron pairs, coming from the decay of light vector mesons via virtual photons in a spatially homogeneous system (details in [@sg05]). Projecting on the particle number in the electron propagator $G^{<}$ and using the equations of motion for $G^<$, the so called Kadanoff-Baym equations, we find the production rate of electrons for a homogeneous, yet time dependent system, to read $$\begin{aligned} \partial_{\tau} N(\textbf{p},\tau) &=2 ~ \text{Im}\left[ \text{Tr}\left\{\frac{\slashed{p}+m}{2 E_{\textbf{p}}}\int_{t_0}^{\tau}d\bar{t} \left(\Sigma^<(\textbf{p},\tau,\bar{t})\right) e^{i E_{\textbf{p}}(\tau-\bar{t})}\right\}\right]\text{,}\label{pertfinal} \end{aligned}$$ with the electron self energy $\Sigma$ and $p_0 = E_{\textbf{p}}$. The free electron propagator can be used because due the electrons’ long mean free path the electrons are not expected to interact with the medium after being produced. The medium effects enter via the dressing of the virtual photon propagator in the electron self energy (see Fig. \[fig:selfenergy\]). ![$\Sigma(1,2)$ in coordinate space[]{data-label="fig:selfenergy"}](selfenergy){width="3cm"} \ $D_{\gamma}$ is the virtual photon propagator and $\Pi$ is its self energy. We have $$\begin{aligned} i \Sigma^{<}(\textbf{p},t_1,t_2)= - e^2 \gamma_{\mu}\left(\int \frac{d^3k}{(2\pi)^3}D_{\gamma}^{<,\mu\nu}(\textbf{k},t_1,t_2)G^{<}_{0}(\textbf{p}-\textbf{k},t_1,t_2)\right)\gamma_\nu\text{,} \end{aligned}$$ with $\textbf{k}$ the momentum of the virtual photon. On inserting this self energy and defining $p^{+}=k-p$ as the four-momentum of the positron and $p^{-}=p$ as that of the electron, equation (\[pertfinal\]) becomes $$\begin{aligned} E_+E_-\frac{dR}{d^3p^{+}d^3p^{-}}(\tau)=&\frac{2e^2}{(2\pi)^6} \left[p_{\mu}^+p_{\nu}^-+p_{\nu}^+p_{\mu}^--g_{\mu\nu}(p^+p^{-}+m^2)\right] \text{Re}\left[\int_{t_0}^{\tau}d\bar{t} i D_{\gamma}^{<,\mu\nu}(\textbf{k},\tau,\bar{t})e^{i(E_+ + E_-)(\tau-\bar{t})}\right]\text{,}\label{galekapustarate} \end{aligned}$$ with $E_+=E_{\textbf{p}}$ and $E_-=E_{\textbf{k}-\textbf{p}}$. Applying the equilibrium properties of $D^<$, it can be shown that equation (\[galekapustarate\]) is the generalization of the well known thermal production rate for lepton pairs in the stationary case [@gk91]: $$\begin{aligned} E_+E_-\frac{dR}{d^3p^{+}d^3p^{-}}(\tau)&=-\frac{2e^2}{(2\pi)^6} \left[p_{\mu}^+p_{\nu}^-+p_{\nu}^+p_{\mu}^--g_{\mu\nu}(p^+p^{-}+m^2)\right]\frac{1}{M^4} \frac{1}{e^{\beta E}-1}\text{Im}\Pi_{\gamma}^{\text{ret},\mu\nu}(k,\tau)\label{galekapustarate2} \end{aligned}$$ In the following, we will consider the mode $\textbf{k}=0$ exclusively, i.e., the virtual photon resting with respect to the medium. After projecting on the virtual photon momentum and taking the electron mass to zero, we get $$\begin{aligned} \frac{dN}{d^4xd^4k}(\tau,\textbf{k}=0,E)=&\frac{2e^2}{(2\pi)^6}\frac{2}{3}\pi(k_{\mu}k_{\nu}-k^2g_{\mu\nu}) \text{Re}\left[\int_{t_0}^{\tau}d\bar{t} i D_{\gamma}^{<,\mu\nu}(\textbf{k}=0,\tau,\bar{t})e^{i E (\tau-\bar{t})}\right]\text{.} \label{sigmalessrate} \end{aligned}$$ The dynamic information is inherent in the memory integral on the right that runs over all virtual photon occupation numbers $D_{\gamma}^{<}$ from the initial time to the present. This way the full nonequilibrium electron production rate at the present time $\tau$ is determined. We introduce the dynamic medium dependence by dressing the virtual photon propagator with the medium dependent $\rho$- or $\omega$-meson. This dressing enters with the self energy $\Pi^<$ via the relation $$\begin{aligned} D^{\gtrless}(1,1')=&\int_{t_0}^{\infty}d2\int_{t_0}^{\infty}d3D^{+}(1,2)\Pi^{\gtrless}(2,3)D^{-}(3,1')+\text{surface term,} \label{fdt} \end{aligned}$$ that follows from the Kadanoff-Baym equations. Using projectors on the different polarizations and the equality of each polarization ($\textbf{k}=0$), we find $$\begin{aligned} \frac{dN}{d^4xd^4k}(\tau,E,\textbf{k}=0)=&\frac{2}{3}\frac{e^2}{(2\pi)^5}(3E^2) \text{Re}\left[\int_{t_0}^{\tau}d\bar{t} i D_{\gamma,T}^{<}(\textbf{k}=0,\tau,\bar{t})e^{i E (\tau-\bar{t})}\right]\text{.}\label{photrate1} \end{aligned}$$ For dilepton production $\Pi^{\text{ret}}\propto e^2$ and $E$ is the invariant mass of the virtual photon. For the cases we are interested in, $\|\Pi^{\text{ret}}\|\ll E$ and we can approximate $ D_T^{<}=D_0^{\text{ret}}\otimes\Pi_T^{<}\otimes D_0^{\text{av}}\label{conv0}\text{.} $ Diverging contributions at early times (low frequencies), due to the undamped photon propagators, cause numerical problems. We introduce an additional cutoff $\Lambda$ for these propagators: $ D_0^{\text{ret}}(\tau-t_1) = (\tau-t_1) \rightarrow (\tau-t_1) e^{-\Lambda (\tau-t_1)} $ and analogously for $D_0^{\text{av}}(t_2-\bar{t})$. The exponential factors lead to a reduction of the rate, which we will overcome by renormalizing the final result. This does not affect the timescales we are interested in, and comparison of the dynamically computed rate for a stationary situation (constant self energy) with the analytic, thermal rate shows perfect agreement. Vector meson dominance (VMD) allows for the calculation of the photon polarization tensor $\Pi_T^<$, using the identity between the electromagnetic current and the canonical interpolating fields of the vector mesons, which leads to $ \Pi^<_{\alpha\beta}=\frac{e^2}{g_\rho^2}m_{\rho}^4 D_{\rho, \alpha\beta}^{<}\text{.} $ Once more, we apply the generalized fluctuation dissipation relation (\[fdt\]) to calculate $$\begin{aligned} D_{\rho,T}^{<}=D_{\rho,T}^{\text{ret}}\otimes\Sigma^<_{\rho,T}\otimes D_{\rho,T}^{\text{av}}\text{,}\label{photfdr} \end{aligned}$$ with the $\rho$-meson self energy $\Sigma^<_{\rho}$. The transverse parts of the retarded and advanced propagators $D_{\rho,T}^{\text{ret}}(\textbf{k},t_1,t_3)=D_{\rho,T}^{\text{av}}(\textbf{k},t_3,t_1)$ of the vector meson in a spatially homogeneous and isotropic medium follow the equation of motion $$\begin{aligned} \left(-\partial_{t_1}^2-m_{\rho}^2-\textbf{k}^2\right)D_{\rho,T}^{\text{ret}}(\textbf{k},t_1,t_3)-\int_{t_3}^{t_1}d\bar{t} \Sigma^{\text{ret}}_{\rho,T}(\textbf{k},t_1,\bar{t})D_{\rho,T}^{\text{ret}}(\textbf{k},\bar{t},t_3)=\delta(t_1-t_3)\text{.}\label{photdgl} \end{aligned}$$ The dynamic medium evolution is now introduced by hand via a specified time dependent retarded meson self energy $\Sigma^{\text{ret}(\tau, \omega)}$ [@sg05]. From that the self energy $\Sigma^<$, needed for solving equation (\[photfdr\]), follows by introducing a background temperature of the fireball. The fireball is assumed to generate the time dependent self energy $\Sigma^{\text{ret}}$ and, assuming a nearly quasi thermalized system, the $\rho$-meson current-current correlator $\Sigma^<$ is given via $\Sigma^<(\tau,\omega)=2i n_B(T(\tau)) Im \Sigma^{\text{ret}}(\tau,\omega)$ , which follows from the KMS relation [@Greiner:1998vd], being valid for thermal systems. The latter is a rather strong assumption, but necessary in order to proceed. The medium in nonequilibrium ============================ The medium effects are introduced via a specific evolving self energy of the vector meson. A simple self energy $$\begin{aligned} \Sigma^{\text{ret}}(\omega,\tau)=-i\omega\Gamma(\tau)\text{,} \end{aligned}$$ with a $\textbf{k}$- and $\omega$-independent width $\Gamma$, leads to a Breit-Wigner distribution for the spectral function. The time dependence is being accounted for by introduction of the parameter $\tau$. For the $\textbf{k}=0$ mode, the full self energy for coupling to $J^P=\frac{1}{2}^-$ -resonances is given by $$\begin{aligned} \Sigma_T(\omega,\textbf{k}=0)=\frac{\rho}{2} \left(\frac{f_{RN\rho}}{m_{\rho}}\right)^2 g_I \frac{\omega^2 \bar{E}}{(\omega+\frac{i}{2}\Gamma_R)^2-\bar{E}^2}-i\omega\Gamma \label{modelshenself}\text{,} \end{aligned}$$ with $\bar{E}=\sqrt{m_R^2+\textbf{k}^2}-m_N$ and $m_R$ and $m_N$ the masses of the resonance and the nucleon respectively. $\Gamma_R$ is the width of the resonance and $g_I$ the isospin factor [@po04]. For our purpose it suffices to retain the part of the self energy that creates the pole structure. We will neglect the $\omega^2$ in the numerator because it causes straightforward dispersion relations to become invalid (subtracted dispersion relations are needed in this case). We will shorten $\left(\frac{f_{RN\rho}}{m_{\rho}}\right)^2 g_I/2~\omega^2$, with $\omega^2=1 \text{GeV}^2$ by $C$, a dimensionless factor. ![Linear switching off of in-medium effects over a certain time $\Delta \tau$[]{data-label="fig:linearchange"}](linearchange){width="7cm"} Investigating which contributions in equation (\[sigmalessrate\]) come from which times in the past, shows that there are contributions from early times as well as alternating positive and negative contributions. An interpretation of this becomes difficult and it follows that only the time integrated yield is a physical quantity. In order to quantify the times that the mesons’ spectral properties need to adjust to the evolving medium we change the self energy linearly in time (see Fig. \[fig:linearchange\]). As a possible characteristic timescale we consider the difference of the final spectral function to the dynamically calculated one at the time where the medium effects are fully turned off, described by the difference in the moment $\int_0^{\infty} A(\omega)^2 \omega^2 d\omega$ of the two spectral functions or the difference in the peak position and height. All methods lead to similar results [@sg05]. We find an exponentially decreasing difference with increasing duration of the change $\Delta\tau$ (see Fig. \[fig:linearchange\]), from that we extract a time constant $\tilde{\tau}$. For the the $\rho$-meson we find a typical timescale of about 3 fm/c. That means that the spectral properties follow the changes in the medium nearly instantaneously only if the evolution is very slow as compared to 3 fm/c. This means that the vector mesons possess a certain memory of the past, and even if they decay outside the medium, they still carry information on the medium in that they were produced. This becomes important especially for the $\omega$-meson, having a width of only $8.49$ MeV ($\tilde{\tau}\approx 60$ fm/c). It turns out that $\tilde{\tau}$ is proportional to $c/\Gamma_2$ with $c$ lying between $2$ and $3.5$, depending on $m$ and $\Gamma_1$. This is significantly longer than the naively expected timescale of $1/\Gamma$. The time needed by the dilepton rate to follow changes is approximately equal to that of the spectral function. The quantum mechanical nature of the regarded systems leads to oscillations and negative values in the changing spectral functions, occupation numbers and production rates as well as interferences that one does not get in semi-classical, adiabatic calculations. The “rate” calculated here possesses the full quantum mechanical information incorporated and contains “memory” interferences that might cause cancellations - hence the rate has to be able to become negative while the time integrated yield always stays positive as the only observable physical quantity. An intriguing example for the occurring oscillations is shown in Fig. \[fig:oscillations\]. ![Production rate for the change of the mass from $m=400$ MeV to 770 MeV (constant $\Gamma$=150 MeV and constant $T=160$ MeV) directly after the self energy has reached its final form (after 7.8 fm/c) and 2 (and 4) fm/c later. Oscillations and negative values appear in the intermediate rates (left). The corresponding yield stays positive (right).[]{data-label="fig:oscillations"}](oscillationsyield){height="9cm"} To calculate the yield, we model the fireball evolution and fold it with the calculated time dependent rates, similar to [@Rapp:1999ej]. For the effective volume we choose a longitudinal Bjorken expansion together with an accelerating radial flow $$V_{eff}(t\geq t_0)=\pi c t (r_0+v_0(t-t_0)+0.5 a_0 (t-t_0)^2)^2\text{,} \label{veff}$$ with $r_0=6.5$ fm, $v_0=0.15~c$ and $a_0=0.05~c^2/\text{fm}$ (see also [@Greiner:2001uh]). From (\[veff\]) and the constraint of conserved entropy (given by a constant entropy per baryon $S/A=30$ for SPS energies [@Greiner:1991us]), temperature $T(\tau)$ and chemical potentials follow as functions of time. We start the calculation at the freezeout temperature of 175 MeV, whereas the final temperature, reached after a lifetime of about 7.8 fm/c is 120 MeV (thermal freezeout). At this point, we turn off further dilepton production by a rapid decrease of the temperature towards zero. With the time dependent temperatures we can integrate the rate and immediately find the yield per unit four momentum. The results for different scenarios are shown in Fig. \[fig:yieldmassres\]. We compare to Markov calculations that assume instantaneous adaption of the spectral function (and the rate) to the self energy, as employed in [@Rapp:1999ej] (dashed lines in Fig.\[fig:yieldmassres\]). The yield resulting from a mass shift following from assumed Brown-Rho scaling [@br91] shows an enhancement of about a factor of 2 in the dynamic calculation within the mass range where the CERES experiment [@ce95; @Wessels:2002ha] has measured a strongly enhanced dilepton yield compared to calculations with vacuum spectral functions. Also the coupling to the N(1520) resonance with no broadening shows an enhanced production around the resonance peak and the $\rho$ vacuum peak, but also a reduced yield in the region between the peaks. On the other hand, the difference to Markov calculations becomes smaller for large in-medium widths. ![Comparison of the dynamically computed dilepton yield from the fireball (solid line) to the one calculated assuming adiabaticty (as done in principle in [@Rapp:1999ej])(dashed line), to show the differences caused by memory effects. Mass shift to 400 MeV in-medium mass (left) and coupling to the N(1520) resonance (right). No broadening of either the resonance or the $\rho$-meson is included here. Also shown is the yield coming from a constant $\rho$ meson’s vacuum spectral function (dotted line) for comparison.[]{data-label="fig:yieldmassres"}](yieldmassres){height="9cm"} Summary and Conclusions {#conclusion} ======================= In the present work we introduced a method to calculate dilepton production rates within a non-equilibrium field theory formalism, based on the real time approach of Schwinger and Keldysh. We investigated possible medium modifications of the $\rho$ meson in a fireball created in a heavy ion collision. We considered mass shifts, broadening and coupling to resonances. Special attention was put to possible retardation effects concerning the off-shell evolution. The timescale on that the spectral function adjusts to changes in the self energy was found to be proportional to the inverse vacuum width of the meson $\Gamma_2$ like $c/\Gamma_2$, with $c$ approximately 3. Further dependence on the in-medium width as well as on the size of the medium modification is present. We find typical retardations for the $\rho$ of $3$ fm/c and about $60$ fm/c for the $\omega$, a very slow adjustment. The full quantum field theoretical treatment leads to oscillations in all mentioned quantities when changes in the self energy are performed. This oscillatory behavior reveals the quantum mechanical character of the many particle system, present in the investigated heavy ion reaction. The oscillations potentially cancel when the rate is integrated over time such that the measurable dilepton yield is always positive. Comparison of dynamically calculated yields with those calculated assuming adiabaticity reveals differences. About a factor of $2$ difference was found within the invariant mass range of $250$ to $500$ MeV for mass shifts predicted using Brown-Rho scaling for the $\rho$-meson in a fireball at SpS energies (158 AGeV). This is the range where CERES measures an increased dilepton yield as compared to calculations assuming the $\rho$’s vacuum shape. Similar results were found for the coupling of the $\rho$-meson to resonance-hole pairs. Our findings show that exact treatment of medium modifications in principle requires the consideration of memory effects. Further investigation of the $\omega$-meson is being done - due to its small width it causes numerical complications. It has to be seen how our results can be consistently (and probably in an approximative manner) incorporated in semi-classical nonequilibrium transport codes like UrQMD [@Bass:1998ca], BUU [@Effenberger:1999ay] or HSD [@Cassing:1999es].\	{ "pile_set_name": "ArXiv" }
--- abstract: \| We prove an improved version of the trace-Hardy inequality, so-called Kato’s inequality, on the half-space in Finsler context. The resulting inequality extends the former one obtained by [@AFV] in Euclidean context. Also we discuss the validity of the same type of inequalities on open cones. [Key words: trace-Hardy inequality, Finsler norm, hyper-geometric function.]{} [2010 Mathematics Subject Classification: 26D10, 35J20, 46E35.]{} author: - 'A. Alvino$^1$' - 'A. Ferone$^2$' - 'A. Mercaldo$^1$' - 'F. Takahashi$^3$' - 'R. Volpicelli$^1$' title: 'Finsler Hardy-Kato’s inequality ' --- Introduction {#section:Introduction} ============ In the last decades interests in Finsler geometry have increased due to its possible applications in different contexts of mathematics, such as anisotropic eigenvalue problems and anisotropic evolution problems. One of the basic idea is to endow the space $\R^N$ with the distance obtained by a Finsler metric and to extend classical results to such a new geometrical context. In this paper we are interested in the trace-Hardy inequality, so-called Hardy-Kato’s inequality, on the half-space $\R^N_+=\R^N\times[0,+\infty)$ endowed with a Finsler norm. That is Hardy inequality for Sobolev functions defined on $\R^N_+$ with non-zero trace on the boundary of $\R^N_+$ in Finsler context. More generally we also treat Hardy-Kato’s inequality on open cones endowed with Finsler norm. The interest in the theory of boundary trace for Sobolev functions and Hardy’s inequalities arises from the possible applications to boundary value problems for PDEs and nonlinear analysis. They have been developed by various authors via different methods in different settings: Here we just recall some recent papers and references therein [@FMT], [@AVV], [@AFV], [@VH]. Let us begin by discussing the case of the half-space $\R^N_+$. In [@AFV] a sharp trace-Hardy inequality has been proved: For any $2 \le \beta < N$ there exists a positive constant $K(N, \beta)$ such that $\displaystyle\lim_{\beta \to N} K(N, \beta) = 0$ and $$\label{HardyAFV} K(N,\beta) \int_{\pd {\re^N_{+}}}\frac {u^2}{\|x\|} \,dx + \frac {(\beta-2)^2}{4}\int_ {{\re^N_{+}}}\frac {u^2}{\|x\|^2+t^2} dx dt \le \int_ {{\re^N_{+}}} \|\nabla u\|^2 \, dx dt$$ holds for any function $u$ in the Sobolev space $W^{1,2}(\R^{N}_+)$. The constant $K(N, \beta)$ is computed explicitly as $$\label{constant} K(N,\beta) = 2\frac{\Gamma $ \frac{N+\beta}{4}-\frac{1}{2} $ \Gamma $ \frac{N-\beta}{4}+\frac{1}{2} $} {\Gamma $ \frac{N+\beta}{4}-1 $ \Gamma $ \frac{N-\beta}{4} $},$$ and both constants $K(N, \beta)$ and $\frac{(\beta-2)^2}{4}$ in are sharp. Inequality interpolates the classical Kato’s inequality, which corresponds to $\beta=2$ in , and the Hardy inequality on $\re^N_{+}$ obtained by letting $\beta$ go to $N$. As pointed out, our goal is to prove the trace-Hardy inequality in a more general geometrical framework. We consider $\R^{N}_{+}$ as the product space $\R^{N-1}\times \R_{+}$ and we endow it with a natural product metric generated by a Finsler norm $H^0$ on $\R^{N-1}$ and the usual Euclidian norm on $\R$. Denote each point $z \in \R^N_{+}$ as a couple $(x,t)$ where $x \in \R^{N-1}$, $t \in \R_{+}$, and consider the norm $\Phi^0$: $$\label{Phi0} \Phi^0(z) = \Phi^0(x,t)=\sqrt{[H^0(x)]^2+t^2}, \quad \>\> z=(x,t)\in \R^{N-1}\times \R_{+}.$$ The [dual norm]{} $\Phi$ of $\Phi^0$: $$\label{Phi} \Phi(\eta)=\Phi(\xi,t)=\sqrt{[H(\xi)]^2+t^2}, \quad \>\>\eta=(\xi,t) \in \R^{N-1} \times \R_{+}$$ is automatically introduced to evaluate the length of the gradient of a function, where $H = H(\xi)$ denotes the dual norm of $H^0 = H^0(x)$ defined on $\R^{N-1}$. We refer to §\[section:notations\] for the definitions, notations, and main properties of a Finsler norm. Our first main result is the following: \[Theorem:main\] Assume $N \ge 3$. Let $H$ be a Finsler norm on $\R^{N-1}$ and let $\Phi(\xi,t)$ be the Finsler norm in $\R_{+}^N$ defined by . Then for any $u \in W^{1,2}(\re^N_{+})$ and $2 \le \beta < N$, $$\label{Finsler-Kato} K(N,\beta) \int_{\pd {\re^N_{+}}} \frac{u^2(x,0)}{\Phi^0(x,0)} dx \le \int_{{\re^N_{+}}} \Phi^2(\nabla u) dz - \frac{(\beta-2)^2}{4} \int_{{\re^N_{+}}} \frac{u^2(z)}{[\Phi^0(z)]^2} dz$$ holds where $\Phi^0$ is defined by , $\nabla u(z) = ( \nabla_x u, \frac{\pd u}{\pd t})$, $dz = dx dt$ for $z = (x, t) \in \re^N_{+}$, and $K(N, \beta)$ is defined in . $K(N,\beta)$ is sharp in the sense that $$\label{optimal} K(N, \beta) = \inf_{u \in W^{1,2}(\R^N_+), u \ne 0} \frac{\displaystyle\int_{{\re^N_{+}}} \Phi^2(\nabla u) \, dz - \frac{(\beta-2)^2}{4} \displaystyle\int_{{\re^N_{+}}} \frac{u^2(z)}{[\Phi^0(z)]^2} dz} {\displaystyle\int_{\pd {\re^N_{+}}} \frac{u^2(x,0)}{\Phi^0(x,0)} dx}$$ holds true. The non-attainability of the optimal constant $K(N,\beta)$ can be seen as follows: If the infimum were attained by a function $u \in W^{1,2}(\R^N_{+})$, then it is proportional to the solution of the problem $$\label{Eq:P} \begin{cases} \Delta_{\Phi} \phi(x,t) + \disp{\frac{(\beta-2)^2}{4} \frac{\phi(x,t)}{[\Phi^0(x,t)] ^2}} = 0 & \textrm{in} \>\> \re^N_{+},\\ \phi(x,0) = [\Phi^0(x,0)] ^{-\frac{N}{2}+1} & \textrm {on} \>\> \pd \re^N_{+}. \end{cases}$$ Here $$\Delta_\Phi = \Delta_{H,x} + \frac{\pd^2}{\pd t^2},$$ and $$\Delta_{H,x} \phi = \sum_{j=1}^{N-1} \frac{\pd}{\pd \xi_j} $ H(\nabla_x \phi)(H_{\xi_j}(\nabla_x \phi) $$$ is the so-called Finsler-Laplace operator with respect to the Finsler norm $H$ on $\R^{N-1}$ (see §\[section:notations\] for the definition). However, we see that solution $\phi$ to satisfies $\phi \notin W^{1,2}(\R^N_{+})$. Actually in §\[section:extremal\] we prove that the solutions to are of the form $$\phi(x,y) = \Phi^0(x,t)^{-\frac N 2 +1}\, w(\sin^2\theta), \quad \theta = \arctan \frac t {H^0(x)}\,,$$ where the function $w$ is expressed in terms of the hyper-geometric series, i.e. $$\label{series} F(a,b,c;y)= 1 + \frac{\Gamma(c)}{\Gamma(a)\Gamma(b)}\sum_{k=1}^\infty \frac{\Gamma(a+k)\Gamma(b+k)}{\Gamma(c+k)}\frac{y^k}{k!}$$ (see in §\[section:extremal\]) and the optimal constant is given by $$\label{Kw} K(N,\beta)= -\lim_{\theta\to 0} (\sin 2\theta) w'(\sin^2\theta).$$ Theorem \[Theorem:main\] is obtained by using a very classical method of Calculus of Variations introduced by Weierstrass and developed by Schwartz, Lichtenstein and Morrey (we refer to [@GH] for the general theory and references therein). It has been adopted in [@AFV] and [@F] to prove inequality and previously, in [@Al2] to find an improvement of the classical Sobolev inequality. It consists of proving that a solution of the Euler-Lagrange equation of a suitable functional is, actually, a minimum. Such method is the crucial tool of our approach, since we deal with functions having non zero trace on the boundary. For more precise description of the method, we refer to [@AFV]. Finally in §\[section:cone\] we face the case of open cones and we show that the same method can be applied to prove the Hardy-Kato inequality in the cone $$C_\alpha = \left\{ (x,t)\in \R^N_+: t> (\tan\alpha)\, H^0(x) \right\}, \quad -\frac{\pi}{2} < \alpha < \frac{\pi}{2}.$$ (See [@Na1], [@Na2] and [@VH] for similar results). In the following, $$\label{area} d\sigma_{x,t} = \sqrt{1 + (\tan^2 \alpha) \|\nabla H^0(x)\|^2} dx, \quad x \in \re^{N-1}$$ denotes an $(N-1)$-dimensional surface measure on $\pd C_{\al}$. \[Theorem:cone\] Assume $N \ge 3$ and $2 \le \beta < N$. Let $\Phi^0(x,t)$ and $\Phi(\xi,t)$ be Finsler norms on $\re^N_{+}$ defined by and respectively. Then there exists a constant $K(N, \alpha, \beta) \in \re$ such that $$\begin{aligned} \label{Eq:Kato(cone)} K(N,\alpha, \beta) \sqrt{1 + \tan^2 \al} & \int_{\pd C_{\alpha}} \frac{u^2(x,t)}{\Phi^0(x,t)} \frac{d\sigma_{x,t}}{\sqrt{1 + (\tan^2 \al) \|\nabla H^0(x)\|^2}} \\ \le &\int_{C_{\alpha}} \Phi^2(\nabla u) \, dxdt - \frac {(\beta-2)^2}{4}\int_{C_{\alpha}} \frac{u^2(x,t)}{[\Phi^0(x,t)]^2} dxdt \notag\end{aligned}$$ holds true for any $u \in W^{1,2}(C_{\alpha})$. The constant $K(N, \alpha, \beta)$ is given by $$K(N,\alpha,\beta) = -(\sin 2\alpha) \frac{w^{\prime}(\sin^2 \alpha)}{A_{\al, \beta}},$$ where $w$ is defined in with $k = -K(N, \beta)$ for $K(N, \beta)$ in , and $A_{\al, \beta}$ is defined in . A proof of Theorem \[Theorem:cone\] is given in §\[section:cone\]. Note that the left-hand side of is written as $$\begin{aligned} K(N,\alpha, \beta) \int_{\re^{N-1}} \frac{u^2(x,(\tan \al) H^0(x))}{H^0(x)} dx.\end{aligned}$$ Note also that by and the fact $A_{\al, \beta} \Big \|_{\al = 0} = 1$, we clearly observe that $$\lim_{\alpha\to 0} K(N,\alpha,\beta) = K(N,\beta).$$ Notations and preliminary results {#section:notations} ================================= In this section, we introduce some notations. Let $n \in \N$ be an integer and let $H: \R^n \to [0,+\infty)$ be a continuous function satisfying the following properties $$\begin{aligned} \label{homogeneous} &H(\lambda \xi) = \|\lambda\| H(\xi), \quad \forall \xi \in \R^n, \forall \lambda \in \R, \\ \label{bound} &\gamma_1 \|\xi\| \le H(\xi) \le \gamma_2 \|\xi\|, \quad \forall \xi \in \R^{n}\end{aligned}$$ for two positive constants $0<\gamma_1 \le \gamma_2 < +\infty$. We denote the unit $H$-ball as $$B_H =\{ \xi \in \R^n \,: \, H(\xi) < 1 \}\,.$$ The [dual function]{}, or [polar function]{}, $H^0: \R^n \to [0,+\infty)$ of $H$ is defined by the formula $$H^0(x) = \sup_{\xi \in \R^n \setminus \{ 0 \}} \frac{\langle \xi, x \rangle_{n}}{H(\xi)} = \sup_{\xi \in B_H} \,\langle \xi, x \rangle_{n} \, ,\quad x \in \R^{n},$$ here and in the following, $\langle \xi ,x \rangle_{n} = \sum_{j=1}^n \xi_j x_j$ denotes the Euclidean inner product of $\R^n$. Note that by definition, it holds that $$\label{Schwarz} \|\langle \xi, x \rangle_n\| \le H(\xi) H^0(x), \quad \xi, x \in \R^n.$$ It is known that $H^0$ is a convex, continuous function on $\R^n$, which satisfies the following properties $$H^0(\lambda x) = \|\lambda\| H^0(x), \quad \forall x \in \R^n, \forall \lambda \in \R.$$ $$\label{bound0} \frac{1}{\gamma_2} \|x\| \le H^0(x) \le \frac{1}{\gamma_1} \|x\|, \quad \forall x \in \R^n.$$ A function $H \in C^2 $ \re^n \setminus \{0\} $$ is a [Finsler norm]{}, if it satisfies properties , , and it is strictly convex. For references about Finsler norms (or, more generally, for Finsler metrics) see [@BCS], [@BP]. Here we just recall the following properties: if $H$ is a Finsler norm, then $H$ is the polar function of $H^0$, that is the following equality holds true $$H(\xi) = (H^0)^0(\xi) = \sup_{x \in \re^n \setminus \{ 0 \}} \frac{\langle \xi ,x \rangle_{n}}{H^0(x)},$$ and $H^0$ is the [gauge function]{} of the closed convex set $\ol{B_H}$. Moreover we have the following basic identities whose proof can be found, for example, in [@BP] Lemma 2.1, 2.2, or [@VS] Proposition 6.2. $$\begin{aligned} \label{(1)} \nabla H (\la \xi) = \frac{\la}{\|\la\|} \nabla H(\xi), &\quad \forall \xi \in \R^{n} \setminus \{ 0\}, \forall \la \in \R \setminus \{ 0\}, \\ \label{(2)} \langle \nabla H(\xi), \xi \rangle_{n}= H(\xi), &\quad \forall \xi \in \R^{n} \setminus \{ 0\}, \\ \label{(3)} H $ \nabla H^0(x) $ = 1, &\quad \forall x \in \R^{n} \setminus \{ 0\}, \\ \label{(4)} \nabla H $\nabla H^0(x) $ =\frac{x}{H^0(x)}, &\quad \forall x \in \R^{n} \setminus \{ 0\}.\end{aligned}$$ Analogous properties hold true for $H^0$ by taking into account that $H(\xi) = (H^0)^0(\xi)$. Finally we recall that if $H: \R^n \to [0,+\infty)$ is a Finsler norm, the [Finsler-Laplace operator]{} $\Delta_H$ is defined as $$\begin{aligned} \Delta_H u (x) &= {\rm div} $ H(\nabla u) \nabla_{\xi} H(\nabla u) $(x) \\ &= \sum_{j=1}^{n} \frac{\pd}{\pd \xi_j} $ H(\xi) H_{\xi_j}(\xi) $ \Big \|_{\xi = \nabla u(x)} \end{aligned}$$ for any function $u\in C^2(\R^{n})$. Construction of exstremals {#section:extremal} ========================== This section is devoted to the construction of a smooth solution to . Let $N \ge 3$. We denote $\re_+ = [0,+\infty)$, $\re^N_{+} = \re^{N-1} \times \re_+$ and $z = (x,t) \in \re^N_{+}$. For a function $u=u(x,t)$ in the Sobolev space $W^{1,2}(\R^N_+)$, $\nabla u = (\nabla_x u,\frac{\pd u}{\pd t})$ denotes its full gradient where $\nabla_x u = $\frac{\partial u}{\partial x_1}, \cdots, \frac{\partial u}{\partial x_{N-1}}$$. \[prop:extremal\] Let $2\le \beta < N$ and let $K(N,\beta)$ be the constant defined in . Then the functions $$\begin{gathered} \label{extremal} \phi(x,t) = \frac{1}{\left[ \Phi^0(x,t) \right]^{\frac{N-2}{2}}} F$ \frac{N+\beta}{4}-1,\frac{N-\beta}{4},\frac{1}{2};\frac{t^2}{\left[ \Phi^0(x,t) \right]^2} $ \\ - t\frac{K(N,\beta)}{\left[ \Phi^0(x,t) \right]^{\frac{N}{2}}} F$\frac{N+\beta}{4}-\frac{1}{2},\frac{N-\beta}{4} + \frac{1}{2},\frac{3}{2};\frac{t^2}{\left[ \Phi^0(x,t) \right]^2}$\end{gathered}$$ are regular solutions to the problem . Moreover, $\phi$ in satisfies $$\label{normal_derivative} \frac{\pd \phi}{\pd t}(x, 0) = - \frac{K(N, \beta)}{[\Phi^0(x,0)]^{\frac{N}{2}}}.$$ \[Proof of Proposition \[prop:extremal\]\] Define new variables $$\label{rho_theta} \begin{cases} &\rho = \Phi^0(x,t)=\sqrt {[H^0(x)]^2+t^2},\\ &\theta = {\rm arctan}\disp\frac{t}{\Phi^0(x,0)} = {\rm arctan}\disp\frac{t}{H^0(x)}, \quad 0<\theta<\frac \pi 2. \end{cases}$$ Then we have $$\begin{aligned} \frac{\pd \rho}{\pd t} = \frac{t}{\rho}, &\quad \frac{\pd \theta}{\pd t} = \frac{H^0(x)}{\rho^2}, \\ \nabla_x \rho = \frac{H^0(x)}{\rho}\nabla H^0(x), &\quad \nabla_x \theta = -\frac{t}{\rho^2} \nabla_x H^0(x).\end{aligned}$$ Thus we see $$\begin{aligned} \label{phi_t} &\phi_t = \frac{\phi_\rho}{ \rho}t+\frac {\phi_\theta}{ \rho^2}H^0(x), \\ \label{phi_tt} &\phi_{tt} = \phi_{\rho\rho} \frac{t^2}{\rho^2} + 2\phi_{\rho\theta} \frac{t H^0(x)}{\rho^3} + \phi_{\theta\theta} \frac{(H^0(x))^2}{\rho^4} \\ &\hspace{4em} + \phi_{\rho} $ \frac{1}{\rho} - \frac{t^2}{\rho^3} $ - 2 \phi_{\theta} \frac{t H^0(x)}{\rho^4}, \notag \\ \label{gradx} &\nabla_x\phi(x,t) = $ \frac {\phi_\rho}{ \rho}-\tan \theta \frac {\phi_\theta}{ \rho^2} $ H^0(x) \nabla H^0(x).\end{aligned}$$ Moreover by , , , and , we have $$\label{Hgrad} H(\nabla_x\phi(x,t))\nabla H(\nabla_x\phi(x,t)) = $\frac{\phi_\rho}{\rho}-\tan \theta \frac{\phi_\theta}{\rho^2}$ x.$$ Thus by , we have $$\begin{aligned} \label{Delta_H} \Delta_{H, x} \phi &= {\rm div}_x $ H(\nabla_x\phi(x,t))\nabla H(\nabla_x\phi(x,t)) $ \\ &= \phi_{\rho\rho} \frac{(H^0(x))^2}{\rho^2} - 2\phi_{\rho\theta} \frac{t H^0(x)}{\rho^3} + \phi_{\theta\theta} \frac{t^2}{\rho^4} \notag \\ &+ \phi_{\rho} $ \frac{N-1}{\rho} - \frac{(H^0(x))^2}{\rho^3} $ + \phi_{\theta} $ \frac{2 t H^0(x)}{\rho^4} - \frac{N-2}{\rho^2} \tan \theta $. \notag\end{aligned}$$ Therefore by , , and the fact $\Delta_{\Phi} = \Delta_{H,x} + \frac{\pd^2}{\pd t^2}$, the equation in the new variables can be written as $$\label{Eq:polar} \phi_{\rho\rho} + (N-1)\frac{\phi_\rho}{\rho} - (N-2)\frac{\phi_\theta}{\rho^2} \tan \theta + \frac{\phi_{\theta\theta}}{\rho^2} = -$ \frac{\beta-2}{2} $^2 \frac{\phi}{\rho^2}.$$ Searching for solutions to of the form $$\label{phi_form} \phi(x,t)=\rho^{-\frac {N}{2}+1}f(\theta)\,,$$ we see that the problem is equivalent to the following limit problem: $$\label{Eq:f} \begin{cases} f''(\theta)-(N-2) (\tan \theta) f'(\theta)-$\frac {(N-2)^2}{4} - \frac {(\beta-2)^2}{4}$f(\theta) = 0 & \theta \in (0,\frac{\pi}{2}), \\ f(0) = 1, \quad \disp{\lim_{\theta \to \frac{\pi}{2}} f(\theta) \in \R}. \end{cases}$$ Problem is explicitly solved in [@PZ] (pp. 271, eq.131) (see also [@AFV]). Indeed, $f(\theta) = w(\sin^2\theta)$, and $w$ is given by $$\label{w_form} w(y) = F $\frac{N+\beta}{4}-1,\frac{N-\beta}{4},\frac{1}{2};y$ + k \sqrt y F $\frac{N+\beta}{4}-\frac{1}{2},\frac{N-\beta}{4}+\frac{1}{2},\frac{3}{2}; y $$$ for a suitable constant $k$. Here $F(a,b,c;y)$ is the hypergeometric series given in which is convergent for $0 \le y < 1$. Moreover in [@AFV], it is proved that $f$ is a bounded solution to $(\ref{Eq:f})$, i.e. $\lim_{y \to 1} w(y) \in \re$ holds true, if and only if $k = -K(N,\beta)$. Here we repeat those arguments for the sake of completeness, analyzing the behavior of a hypergeometric function near the point $y=1$. For this purpose, we recall that (see [@AS] pp. 559) $$\label{formula2} \lim_{y \to 1} \frac {F(a,b,c;y)}{\ln(1-y)}=-\frac {\Gamma(a+b)}{\Gamma (a)\Gamma(b)}, \quad if \quad c-a-b=0,$$ $$\label{formula3} \lim_{y \to 1} \frac {F(a,b,c;y)}{(1-y)^{c-a-b}}=\frac {\Gamma (c)\Gamma(a+b-c)}{\Gamma (a)\Gamma(b)}, \quad if \quad c-a-b<0.$$ An easy calculation shows that for both hypergeometric functions appearing in (\[w\_form\]), $c-a-b = \frac{3-N}{2} \le 0$ when $N \ge 3$. Let us first examine the case $N > 3$. We write $$\begin{aligned} &\lim_{y \to 1} w(y) \\ &=\lim_{y \to 1} (1-y)^{\frac {3-N}{2}} \left [ \frac{F$\frac{N+\beta}{4}-1,\frac{N-\beta}{4},\frac{1}{2};y$}{(1-y)^{\frac {3-N}{2}}} + k \sqrt y \frac{F$\frac{N+\beta}{4}-\frac {1}{2},\frac{N-\beta}{4}+\frac {1}{2},\frac{3}{2};y$}{(1-y)^{\frac {3-N}{2}}}\right ].\end{aligned}$$ By , the formula ([@AS], pp. 557) $$\label{formula_AS} \frac{d}{dy}F(a,b,c;y) = \,\frac{ab}{c} \,F(a+1,b+1,c+1;y),$$ and de l’Hopital Theorem, we have that the limit is finite if and only if $$\label{help} \frac{\Gamma $\frac{1}{2} $ \Gamma $\frac {N-3}{2}$}{\Gamma $\frac{N-\beta}{4}-1$ \Gamma $\frac{N+\beta}{4}$} + k \frac{\Gamma $\frac{3}{2}$ \Gamma $\frac {N-3}{2}$}{\Gamma $\frac{N+\beta}{4} -\frac{1}{2}$ \Gamma $\frac{N-\beta}{4} + \frac{1}{2} $} = 0.$$ Since $\Gamma(\frac{1}{2})= 2 \Gamma(\frac{3}{2})$, therefore $k = -K(N,\beta)$ for $K(N,\beta)$ in . The case $N=3$ follows in a similar way, by using instead of . Putting $k = -K(N,\beta)$ in and taking , into account, we deduce that $$\phi (x,t) = $ \Phi^0(x,t) $^{-\frac N 2 +1} w$\frac {t^2}{[{\Phi^0(x,t)}]^2} $$$ is a bounded solution to the equation in . Since $F(a,b,c;0)=1$, we have $w(0) = 1$ and that $\phi(x,0) = [\Phi^0(x,0)]^{-\frac{N}{2} +1}$. Also by and , we see $\phi_t(x,0) = f'(0)[\Phi^0(x,0)]^{-\frac{N}{2}}$. Let us evaluate $f'(0)$. To do this recall that $f(\theta)=w(\sin^2\theta)$ and $w$ is given by with $k=-K(N,\beta)$. Taking the formula into account and $F(a,b,c;0)=1$, it is easy to check that $f'(0)=-K(N,\beta)$. Thus we have and $\frac{\pd \phi}{\pd t}(x,0) = -K(N, \beta) [\Phi^0(x,0)]^{-\frac{N}{2}}$. This completes the proof of Proposition \[prop:extremal\]. Proof of Theorem \[Theorem:main\] {#section:proof} ================================= In this section, we prove Theorem \[Theorem:main\]. Let $\Phi$ be the Finsler norm in $\R^N_+$ defined in and let $\phi$ be the solution to the problem defined in (\[extremal\]). As stated in §\[section:Introduction\], we follow the arguments in [@AFV], [@F], while some modification is needed to apply them in the general Finsler context. Define a vector field ${\bf F}:\re^N_{+}\times \R \ni (z, h) \mapsto {\bf F}(z,h) \in \R^{N+1}$ where $z = (x,t) \in \re^N_{+}$ as $$\begin{gathered} \label{F} {\bf F}(z,h) \equiv \Bigg(\frac{2h}{\phi(z)}\Phi(\nabla\phi)\nabla\Phi(\nabla \phi)\,,\frac{h^2}{\phi^2(z)}\Phi^2(\nabla\phi)+\frac{(\beta-2)^2}4 \frac {h^2}{[\Phi^0(z)]^2}\Bigg )\,\\ = \Bigg(\frac{2h}{\phi(x,t)}H(\nabla_x\phi)\nabla H(\nabla_x\phi)\,,\frac{2h}{\phi(x,t)} \phi_t\,, \frac{h^2}{\phi^2(x,t)}[H^2(\nabla_x\phi)+(\phi_t)^2]+\frac{(\beta-2)^2}4 \frac {h^2}{[H^0(x)]^2+t^2}\Bigg ).\end{gathered}$$ Direct calculation shows that ${\bf F}$ is divergence free. Indeed, $$\langle \nabla\Phi(\nabla\phi),\nabla \phi \rangle_N = \Phi(\nabla \phi)$$ by and recalling that $\phi$ satisfies $$\Delta_{\Phi} \phi + \frac{(\beta-2)^2 }4 \frac {\phi}{[\Phi^0]^2} = 0, \quad z = (x, t) \in \R^N_{+},$$ we have $$\begin{gathered} {\rm div}_{z,h}{\bf F} = {\frac{2h}{\phi}}{\rm div}_z$\Phi(\nabla\phi)\nabla\Phi(\nabla\phi)$ -{\frac{2h}{\phi^2}}\Phi(\nabla\phi)\langle\nabla\Phi(\nabla\phi),\nabla \phi\rangle_N\\ +{\frac{2h}{\phi^2}}\Phi^2(\nabla\phi) +{\frac{2h}{\phi}} {\frac {(\beta-2)^2}{4}} {\frac {\phi}{[\Phi^0(z)]^2}} ={\frac{2h}{\phi}} \left[\Delta_{\Phi} \phi + \frac{(\beta-2)^2 }4 \frac {\phi}{[\Phi^0(z)]^2}\right] = 0.\end{gathered}$$ For every $r>0$, denote $$B_{\Phi^0}(r)=\{z \in \re^N_{+}:\, \Phi^0(z) < r \}.$$ Let $R>0$ and let $u\in C_0^\infty (\re^N)$ be a nonnegative function compactly supported on $B_{\Phi^0}(R)$. Denote by $\Omega$ the set of $\re^N_{+}\times \R_+$ given by the subgraph of $u$ which is projected into $B_{\Phi^0}(R)\setminus B_{\Phi^0}(r)$, for some $0<r<R$. We get that the flow of ${\bf F}$ across $\partial\Omega$ is zero, since ${\bf F}$ is divergence free. This means that, if $\nu$ is the unit outer normal to $\partial \Omega$, we have $$\label{div0} \int_{\partial \Omega}\langle {\bf F}(z,h)\,,\nu\rangle_{N+1}d\mathcal H^N=0\,.$$ Let us write explicitly the left hand side of . Note that $\partial\Omega$ consists of the union of the following $N$-dimensional surfaces $$\begin{aligned} \label{s1} &\Sigma_1=\left\{(z,0)\in \re^N_{+} \times \R : z \in B_{\Phi^0(R)} \setminus B_{\Phi^0(r)} \right\}, \\ \label{s2} &\Sigma_2=\left\{(z,h) \in \partial\re^N_{+} \times \R : r < \Phi^0(z)< R, \, 0\le h\le u(z) \right\}, \\ \label{s3} &\Sigma_3=\left\{(z,h) \in \re^N_{+} \times \R : \Phi^0(z)= r, \, 0 \le h \le u(z) \right\}, \\ \label{s4} &\Sigma_4=\left\{(z,h) \in \re^N_{+} \times \R : z \in B_{\Phi^0(R)}\setminus B_{\Phi^0(r)}, \, h= u(z) \right\}.\end{aligned}$$ If $\nu_i$ denotes the outer unit normal of $\Sigma_i$ $(i=1,2,3,4)$ with respect to the Euclidean norm, from we get $$\label{div_sum} \sum_{i=1}^4 \int_{\Sigma_i}\langle {\bf F}\,,\nu_i\rangle_{N+1} d\mathcal H^N = 0.$$ Since ${\bf F}(z,0)\equiv 0$, we have $$\label{div_s1} \int_{\Sigma_1}\langle {\bf F}\,,\nu_1\rangle_{N+1} d\mathcal H^N = 0.$$ As regards the flow across $\Sigma_2$, observe that $\nu_2\equiv -{\bf e}_N$, where ${\bf e}_N = (0, \cdots, 0, 1, 0)$ is the unit vector of the standard Euclidean basis of $\R^{N+1}$. Note that $d{\mathcal H}^N = dx dh$ on $\Sigma_2$. Since $z = (x, 0) \in \partial\re^N_{+}$ and by definition of $\bf F$, we get $$\begin{aligned} \int_{\Sigma_2}\langle \F, \nu_2 \rangle_{N+1} d{\mathcal H}^N &= -2\int_{r<\Phi^0(x,0)<R} dx \int_0^{u(x,0)} \frac{h}{\phi(x,0)} \phi_t(x,0) dh\notag\\ &=-\int_{r<\Phi^0(x,0)<R} \frac{u^2(x,0) }{\phi(x,0)}{\phi_t(x,0)} dx.\end{aligned}$$ Since $\phi$ satisfies and , we then obtain $$\label{div_s2} \int_{\Sigma_2}\langle \F, \nu_2 \rangle_{N+1} d{\mathcal H}^N = K(N,\beta) \int_{r<\Phi^0(x,0)<R} \frac{u^2(x,0)}{\Phi^0(x,0)} dx.$$ Let us now evaluate the flow across $\Sigma_3$. The unit normal to $\Sigma_3$ is given by $\nu_3 = $-\frac{\nabla \Phi^0}{\|\nabla \Phi^0\|},0 $$, so that by we deduce $$\label{div_s3} \int_{\Sigma_3}\langle \F, \nu_3 \rangle_{N+1} d{\mathcal H}^N = -\frac{1}{\|\nabla\Phi^0\|} \int_{\Sigma_3} \frac{2h}{\phi} \left\langle \Phi $ \nabla\phi $ \nabla\Phi $\nabla \phi $\,, \nabla \Phi^0 \right\rangle_N d{\mathcal H}^N.$$ By , , , and , it follows $$\begin{gathered} \label{Eq:Phi} \left\langle \Phi $\nabla\phi$ \nabla \Phi $\nabla \phi $, {\nabla\Phi^0} \right\rangle_N = \left\langle H(\nabla_x\phi)\,\nabla H(\nabla_x\phi)\,, \frac{H^0(x)\nabla H^0(x)}{\Phi^0(x,t)} \right \rangle_{N-1} + \frac{t \phi_t}{\Phi^0(x,t)} \\ = \rho^{-\frac{N}{2}-2} \Bigg [ $\( -\frac{N}{2} +1 $f(\theta) - \tan \theta f'(\theta )\) H^0(x)\, \langle x\,, \nabla H^0(x) \rangle_{N-1} \\ +t$\(-\frac{N}{2} +1 $f(\theta)t+f'(\theta )H^0(x)\) \Bigg ] =\rho^{-\frac{N}{2}} \left[$ -\frac{N}{2} +1$ f(\theta) \right] = -\frac {N-2}{2\rho} \phi\,.\end{gathered}$$ Note that by and we have $$\|\nabla\Phi^0\| \ge \frac{1}{\gamma_2} \Phi $\nabla\Phi^0 $ = \frac{1}{\gamma_2}\,.$$ Thus collecting and , we deduce $$\begin{gathered} \label{div_s3_estimate} \left \| \int_{\Sigma_3}\langle \F, \nu_3 \rangle_{N+1}\,d{\mathcal H}^N\right \| \le \gamma_2\frac{(N-2)}{r}\int_{\Sigma_3}h\,d{\mathcal H}^N=\gamma_2\frac {N-2}r \int_{\Phi^0(z)=r}d{\mathcal H}^{N-1}\int_0^{u(z)}hdh \\ \le \gamma_2\frac{(N-2)}{r}\,{\mathcal H}^{N-1}\left (\{z\in \R^{N-1}\times\R_+: \Phi^0(z)=r\}\right )\sup_{\Phi^0(z)=r}\frac {u^2(z)}2 \\ \le \gamma_2\frac{N-2}{r} \, \frac {N\,k_N \,r^{N-1}}2\sup_{\Phi^0(z)=r}\frac {u^2(z)}2=O(r^{N-2})\,,\end{gathered}$$ where $k_N$ is the measure of $B_{\Phi^0}$. It remains to estimates the flow of ${\bf F}$ across $\Sigma_4$. In such a case the normal $\nu_4$ is given by $$\nu_4 = \frac{1}{\sqrt{\|\nabla u\|^2+1}} $-\nabla u, 1 $ \in \re^{N+1},$$ and then by , it follows $$\begin{gathered} \label{div_s4} \int_{\Sigma_4}\langle {\bf F}\,, \nu_4 \rangle_{N+1}d{\mathcal H}^N= \int_{r<\Phi^0(z)<R} \langle {\bf F}(z,u(z)), (-{\bf\nabla} u(z), 1)\rangle_{N+1}dz \\ =\int_{r<\Phi^0(z)<R} \Bigg (-\frac{2u}{\phi} \Phi $\nabla \phi $\langle \nabla \Phi(\nabla \phi)\,, \nabla u \rangle_N + \frac {u^2}{\phi^2}\Phi(\nabla \phi)^2+\frac{(\beta-2)^2}{4} \frac{u^2(z)}{[\Phi^0(z)]^2}\Bigg )dz \,.\end{gathered}$$ Here note that $d{\mathcal H}^N = \sqrt{1 + \|\nabla u\|^2} dz$ on $\Sigma_4$. By convexity of $\Phi$, we get that $$\Phi(\nabla u) \ge \Phi(\nabla \phi) + \langle \nabla\Phi(\nabla\phi)\,,\nabla u - \nabla\phi \rangle_N \,,$$ and by , , and Young’s inequality, we obtain $$\begin{gathered} \label{convex} \frac{2u}{\phi}\Phi$\nabla \phi$\langle \nabla \Phi(\nabla \phi)\,, \nabla u\rangle_N\le \frac{2u}{\phi}\Phi$\nabla \phi$ \left [ \Phi(\nabla u)-\Phi(\nabla \phi)+ \langle \nabla\Phi(\nabla\phi)\,,\nabla\phi\rangle_N \right ]\\ =\frac{2u}{\phi}\Phi $\nabla \phi $\, \Phi(\nabla u)\le \frac {u^2}{\phi^2}\Phi$ \nabla \phi $^2+\Phi(\nabla u)^2\,.\end{gathered}$$ Finally, collecting and we deduce $$\label{div_s4_estimate} - \int_{\Sigma_4}\langle {\bf F}\,, \nu_4\rangle_{N+1}d{\mathcal H}^N \le \int_{r<\Phi^0(z)<R} $ \Phi(\nabla u)^2 - \frac {(\beta-2)^2}4 \frac {u^2(z)}{[\Phi^0(z)]^2} $dz\,.$$ Collecting , , , , and , we obtain $$\begin{gathered} \label{last} \int_{r<\Phi^0(z)<R} $ \Phi(\nabla u)^2 - \frac {(\beta-2)^2}4 \frac {u^2(z)}{[\Phi^0(z)]^2} $dz\\ \ge K(N,\beta) \int_{r<\Phi^0(x,0)<R} \frac{u^2(x,0) }{H^0(x)} dx+ O(r^{N-2})\,.\end{gathered}$$ Letting $r$ go to zero and $R$ go to infinity, we prove the inequality . To prove the optimality of the constant that appears in (\[constant\]), repeat all the previous arguments on replacing $u$ by $\phi$. In such a case both inequalities and hold as equality. Moreover, since $\phi $ is not compactly supported in $B_{\Phi^0}(R)$, the extra $N$-dimensional surface has to be considered $$\Sigma_5=\left \{(z,h)\in\re^N_{+}\times\R : \Phi^0(z)= R\,, \quad 0\le h\le \phi(z) \right \}\,.$$ The unit normal $\nu_5$ is given by $\nu_5=-\nu_3$, so that, by and , instead of , we obtain $$\begin{gathered} \label{last2} \int_{r<\Phi^0(z)<R} $ \Phi^2(\nabla \phi) -\frac{(\beta-2)^2}{4} \frac{\phi^2(z)}{[\Phi^0(x,t)]^2} $dz \\ = K(N,\beta) \int_{r<\Phi^0(x,0)<R} \frac{\phi^2(x,0)}{H^0(x)}dx - \frac{1}{\|\nabla \Phi^0\|} \frac {N-2}{2r}\int_{\Sigma_3} h d{\mathcal H}^N + \frac{1}{\|\nabla \Phi^0\|}\frac {N-2}{2R} \int_{\Sigma_5} h d{\mathcal H}^N \,.\end{gathered}$$ It is easy to check that the last two integrals in (\[last2\]) are equal. Indeed, by spherical coordinates, if $B_{\Phi^0}^+=B_{\Phi^0}\cap \R_+$, recalling , we have $$\begin{aligned} \label{uni} \frac 1 r \int_{\Sigma_3} h d{\mathcal H}^N &=\frac 1 r \int_{\partial B_{\Phi^0}^+}r^{N-1}$ \int_0^{\phi(rx', rt')}h dh $d{\mathcal H}^{N-1} \notag \\ &=\frac{r^{N-1}}{2r} \int_{\partial B_{\Phi^0}^+}\phi^2(rx',rt')d{\mathcal H}^{N-1} \notag \\ &= \frac{r^{N-2}}2 \int_{\partial B_{\Phi^0}^+}r^{-N+2}f^2(\theta)d{\mathcal H}^{N-1} =\frac 1 R \int_{\Sigma_5} h d{\mathcal H}^N.\end{aligned}$$ Collecting (\[last2\]) and (\[uni\]) we deduce $$\lim_{R \to \infty}\,\lim_{r \to 0} \frac{\int_{r<\Phi^0(z)<R} \Phi^2(\nabla \phi)(z) dz}{\int_{r<\Phi^0(x,0)<R} \frac{\phi^2(x,0)}{H^0(x)} dx} = K(N,\beta),$$ which shows the optimality of the constant. Finsler Hardy-Kato’s inequality in cones {#section:cone} ======================================== In this section, we give a proof of Theorem \[Theorem:cone\]. Let us consider the following open cone $$C_{\alpha} = \{(x,t) \in \re^{N-1} \times \re \,:\, t > (\tan \alpha) H^0(x) \}$$ in $\re^N$ for some $\alpha\in (-\frac{\pi}{2}, \frac{\pi}{2})$. Note that the unit outer normal vector on the $(N-1)$-dimensional surface $$\pd C_{\alpha} = \{(x,t) \in \re^{N-1} \times \re \,:\, t = (\tan \alpha) H^0(x) \}$$ is given by $$\label{nu} \nu^\alpha(x, t) = \frac{1}{\sqrt{1 + \tan^2 \alpha \|\nabla H^0(x)\|^2}} ( (\tan \alpha) \nabla H^0(x), -1) \in \re^{N-1} \times \re,$$ and the area element $d\sigma_{x,t}$ on $\pd C_{\al}$ is defined by . We repeat the same arguments used in §\[section:extremal\] and we look for solutions $\phi_{\al, \beta}$ to the problem $$\label{Eq:cone} \begin{cases} \Delta_{\Phi} \phi_{\alpha, \beta}(x,t) + \disp{\frac{(\beta-2)^2}{4} \frac{\phi_{\alpha, \beta}(x,t)}{[\Phi^0(x,t)] ^2}} = 0 & \textrm{in} \, C_{\alpha},\\ \phi_{\alpha, \beta} = [\Phi^0(x,t)] ^{-\frac{N}{2}+1} & \textrm {on} \, \pd C_{\alpha} \end{cases}$$ of the form $$\phi_{\alpha, \beta}(x,t) = \frac{1}{$ (H^0(x))^2 + t^2 $^{\frac{N-2}{4}}}w_{\alpha, \beta} $ \frac{t^2}{(H^0(x))^2 + t^2} $,$$ where $\Phi^0$, $\Phi$ are defined by , respectively. Then $w_{\alpha, \beta}(\sin^2 \theta) = g_{\al, \beta}(\theta)$ and $g_{\alpha, \beta}$ solves the problem $$\begin{cases} g''(\theta)-(N-2) (\tan \theta) g'(\theta)- $\frac {(N-2)^2}{4}- \frac{(\beta-2)^2}{4} $ g(\theta) = 0 & \theta \in (\alpha,\frac{\pi}{2}), \\ g(\alpha) = 1, \quad \lim_{\theta \to \frac{\pi}{2}} g(\theta) \in \re. \end{cases}$$ Thus $w_{\al, \beta}$ is described by using the hypergeometric function $$\label{w_ab} w_{\alpha, \beta}(y) = c_1 F $\frac{N+\beta}{4}-1,\frac{N-\beta}{4},\frac{1}{2};y $ + c_2 \sqrt{y} F $\frac{N+\beta}{4} - \frac {1}{2}, \frac{N-\beta}{4}+\frac{1}{2},\frac{3}{2};y $$$ for suitable choice of constants $c_1$ and $c_2$. The constants $c_1,c_2$ have to satisfy $$\label{limw1} w_{\al,\beta}(\sin^2\alpha)= 1, \quad \lim_{\theta \to \frac{\pi}{2}}w_{\al, \beta}(\sin^2\theta) \in \R.$$ By the first condition in , we get $$\begin{gathered} \label{limw2} c_1 F $ \frac{N+\beta}{4}-1,\frac{N-\beta}{4},\frac{1}{2};\sin^2\alpha $ \\ + c_2 \|\sin\alpha\| F $ \frac{N+\beta}{4}-\frac {1}{2},\frac{N-\beta}{4} + \frac{1}{2},\frac{3}{2};\sin^2\alpha $=1.\end{gathered}$$ For the second condition in , we have $$\label{c22} c_1 \frac{\Gamma $ \frac{1}{2} $ \Gamma $\frac{N-3}{2} $}{\Gamma $ \frac{N-\beta}{4} -1 $ \Gamma $\frac{N+\beta}{4} $} + c_2 \frac{\Gamma $ \frac{3}{2} $ \Gamma $\frac{N-3}{2} $}{\Gamma $ \frac{N+\beta}{4} -\frac{1}{2}$ \Gamma $\frac{N-\beta}{4} + \frac{1}{2} $}=0.$$ Collecting and , if we put $$\begin{gathered} \label{A_ab} A_{\alpha, \beta} = F$\frac{N+\beta}{4}-1,\frac{N-\beta}{4},\frac{1}{2};\sin^2\alpha $ \\ - K(N,\beta)\|\sin\alpha\| F$\frac{N+\beta}{4}-\frac {1}{2},\frac{N-\beta}{4}+\frac {1}{2},\frac{3}{2};\sin^2\alpha $,\end{gathered}$$ we get $$\label{c1c2} c_1 = \frac{1}{A_{\alpha, \beta}}, \quad c_2 = -\frac{K(N,\beta)}{A_{\alpha, \beta}}.$$ Therefore, if $w$ is as in with $k = -K(N, \beta)$ and $\phi$ is defined in , we obtain $w_{\alpha, \beta}(y) = \frac{1}{A_{\alpha, \beta}} w(y)$ and that $$\label{phi_ab} \phi_{\alpha, \beta}(x,t) = \frac 1{ A_{\alpha, \beta}} \phi(x,t)$$ is a solution to . Let us check what happens in the proof of Theorem \[Theorem:main\] when we work on $C_{\alpha}$. We start by defining the vector field $\F_{\al, \beta}$ by replacing $\phi$ with $\phi_{\al,\beta}$ in the definition . By , we have $\F_{\al, \beta} \equiv \F$ where $\F$ is defined by using $\phi$ in . Also, instead of the surface defined in - we deal with the following $N$-dimensional hypersurfaces in $\re^{N+1}$: $$\begin{aligned} &\Sigma_{C_\alpha,1}=\left \{(z,0)\in C_{\alpha} \times \R : z\in B_{\Phi^0}(R) \setminus B_{\Phi^0}(r) \right \}, \\ &\Sigma_{C_\alpha,2}=\left \{((x, t), h) \in \partial C_{\alpha} \times \R : r < \Phi^0(x,t)< R, \, 0 \le h \le u(x,t), \, t = (\tan \al) H^0(x) \right \}, \\ &\Sigma_{C_\alpha,3}=\left \{(z,h)\in C_{\alpha} \times \R : \Phi^0(z)= r, \, 0 \le h\le u(z) \right \}, \\ &\Sigma_{C_\alpha,4}=\left \{(z,h)\in C_{\alpha} \times \R : z\in B_{\Phi^0}(R) \setminus B_{\Phi^0}(r), \, h= u(z) \right \}.\end{aligned}$$ The unit outer normal vector on $\Sigma_{C_\alpha,2}$ is given by $\nu_2(x, t) =(\nu^\alpha (x,t),0)\in \R^{N+1}$ where $\nu^{\al}$ is defined in . Thus $$\begin{aligned} &\langle \F_{\al, \beta}, \nu_2 \rangle_{N+1} \Big\|_{(x, t, h) \in \Sigma_{C_\al,2}} = \langle $ \frac{2h}{\phi_{\al, \beta}} H(\nabla_x\phi_{\al, \beta})(\nabla H)(\nabla_x\phi_{\al, \beta}), \, \frac{2h}{\phi_{\al, \beta}} (\phi_{\al, \beta})_t $, \, \nu^{\al} \rangle_N \Big\|_{(x,t) \in \pd C_{\al}} \\ &= \frac{1}{\sqrt{1 + (\tan^2 \alpha) \|\nabla H^0(x)\|^2}} \frac{2h}{\phi_{\al, \beta}} \left\{ (\tan \alpha) H(\nabla_x\phi_{\al, \beta})(\nabla H)(\nabla_x\phi) \cdot \nabla H^0(x) - (\phi_{\al, \beta})_t \right\}.\end{aligned}$$ Note that $d{\mathcal H}^N = d\sigma_{x,t} dh = \sqrt{1 + (\tan^2 \alpha) \|\nabla H^0(x)\|^2} dx dh$ on $\Sigma_{C_\alpha,2}$. Thus noting the cancellation of the term $\sqrt{1 + (\tan^2 \alpha) \|\nabla H^0(x)\|^2}$, we see $$\begin{aligned} &\int_{\Sigma_{C_{\alpha},2}}\langle \F_{\al, \beta}, \nu_2 \rangle_{N+1} d{\mathcal H}^N \\ &= \int_0^{u(x, (\tan \al)H^0(x))} 2h \times \\ &\int_{\{ x \, : \, r<\Phi^0(x, (\tan \al)H^0(x))<R \}} \frac{1}{\phi_{\al, \beta}} \left\{ (\tan \alpha) H(\nabla_x\phi_{\al, \beta})(\nabla H)(\nabla_x\phi_{\al, \beta}) \cdot \nabla H^0(x) - (\phi_{\al, \beta})_t \right\} dxdh \\ &= \int_{\{ x \, : \, r<\Phi^0(x,(\tan \al) H^0(x))<R \}} \frac{u^2(x,(\tan \al) H^0(x))}{\phi_{\al, \beta}(x,(\tan \al) H^0(x))} \times \\ &\hspace{14em} \left\{ (\tan \alpha) H(\nabla_x\phi_{\al, \beta})(\nabla H)(\nabla_x\phi_{\al, \beta}) \cdot \nabla H^0(x) - (\phi_{\al, \beta})_t \right\} dx.\end{aligned}$$ Now, we compute $$\frac{1}{\phi_{\al,\beta}} \left\{ (\tan \alpha) H(\nabla_x\phi_{\al,\beta})(\nabla H)(\nabla_x\phi_{\al,\beta}) \cdot \nabla H^0(x) - (\phi_{\al,\beta})_t \right\}$$ on $\pd C_{\al}$. Since $\nabla_x \phi_{\al, \beta}(x,t) = A(x,t) \nabla H^0(x)$ where $$\begin{aligned} A(x,t) &= (\Phi^0(x,t))^{\frac{-N-2}{2}} $\frac{2-N}{2}$ w_{\al, \beta}$\frac{t^2}{(H^0(x))^2 + t^2} $ H^0(x) \\ &+ (\Phi^0(x,t))^{-\frac{N-6}{2}} w_{\al, \beta}^{\prime}$\frac{t^2}{(H^0(x))^2 + t^2}$ (-2t^2 H^0(x)),\end{aligned}$$ we check that $$H(\nabla_x\phi_{\al, \beta})(\nabla H)(\nabla_x\phi_{\al, \beta}) \cdot \nabla H^0(x) = A(x,t).$$ Also we have $\frac{t^2}{(H^0(x))^2 + t^2} = \sin^2 \alpha$ on the surface $\pd C_{\alpha}$. Thus since $t = (\tan \al) H^0(x)$ and $\Phi^0(x,t) = \sqrt{1 + \tan^2 \al} H^0(x)$ on $\pd C_{\al}$, we have $$\begin{aligned} &\frac{1}{\phi_{\al, \beta}} \left\{ (\tan \alpha) H(\nabla_x\phi_{\al, \beta})(\nabla H)(\nabla_x\phi_{\al, \beta}) \cdot \nabla H^0(x) - (\phi_{\al, \beta})_t \right\} = \frac{1}{\phi_{\al, \beta}} \left\{ (\tan \alpha) A(x,t) - (\phi_{\al, \beta})_t \right\} \\ &= \frac{1}{\phi_{\al, \beta}} \left\{ (\Phi^0(x,t))^{\frac{-N-2}{2}} $ \frac{2-N}{2} $ w_{\al, \beta}(\sin^2 \alpha) (\tan \al) H^0(x) \right. \\ &\left. \hspace{2em} + (\Phi^0(x,t))^{\frac{-N-6}{2}} w_{\al, \beta}^{\prime}(\sin^2 \alpha) (-2t^2 (\tan \al) H^0(x)) \right. \\ &\left. \hspace{2em} - $ \frac{2-N}{2} $ (\Phi^0(x,t))^{\frac{-N-2}{2}} w_{\al, \beta} (\sin^2 \alpha) t - (\Phi^0(x,t))^{\frac{-N-6}{2}} w_{\al, \beta}^{\prime}(\sin^2 \alpha) (2t(H^0(x))^2) \right\} \\ & = \frac{K(N, \al, \beta)}{\Phi^0(x,t)} \sqrt{1 + \tan^2 \al}\end{aligned}$$ on $\pd C_{\al}$, where we put $$K(N, \al, \beta) = -(\sin 2\al) w_{\al, \beta}^{\prime}(\sin^2 \alpha) = -(\sin 2\al) \frac{w^{\prime}(\sin^2 \alpha)}{A_{\al, \beta}}.$$ Summarizing, we have $$\begin{aligned} \label{div_S2} &\int_{\Sigma_{C_{\alpha},2}}\langle \F_{\al, \beta}, \nu_2 \rangle_{N+1} d{\mathcal H}^N \\ &= K(N, \al, \beta) \sqrt{1 + \tan^2 \al} \int_{\{ x \in \re^{N-1} \, : \, r<\Phi^0(x,(\tan \al) H^0(x))<R \}} \frac{u^2(x,(\tan \al) H^0(x))}{\Phi^0(x,(\tan \al) H^0(x))} dx \notag \\ &= K(N, \al, \beta) \sqrt{1 + \tan^2 \al} \int_{\{ (x, t) \in \pd C_{\al} \, : \, r < \Phi^0(x,t) < R \}} \frac{1}{\sqrt{1 + (\tan^2 \al)\|\nabla H^0(x)\|^2}} \frac{u^2(x,t)}{\Phi^0(x,t)} d\sigma_{x,t}. \notag\end{aligned}$$ On the other hand, since $\F_{\al, \beta} = \F$, we obtain the estimates $$\begin{aligned} \label{div_S1} &\int_{\Sigma_{C_{\alpha},1}}\langle \F_{\al, \beta}, \nu_1 \rangle_{N+1} d{\mathcal H}^N = 0, \\ \label{div_S3} &\int_{\Sigma_{C_{\alpha},3}}\langle \F_{\al, \beta}, \nu_3 \rangle_{N+1} d{\mathcal H}^N = O(r^{N-2}), \quad (r \to 0) \\ \label{div_S4} &-\int_{\Sigma_{C_{\alpha},4}}\langle \F_{\al, \beta}, \nu_4 \rangle_{N+1} d{\mathcal H}^N \le \int_{z \in C_{\al}, r<\Phi^0(z)<R} $ \Phi(\nabla u)^2 - \frac {(\beta-2)^2}4 \frac {u^2(z)}{[\Phi^0(z)]^2} $dz\end{aligned}$$ as in the case when $\al = 0$, where $\nu_i$ denotes the outer unit normal of $\Sigma_{C_{\alpha},i}$ $(i=1,3,4)$. Collecting , , , and $$\sum_{i=1}^4 \int_{\Sigma_{C_{\al},i}} \langle \F_{\al, \beta} \,, \nu_i \rangle_{N+1} d\mathcal H^N=0,$$ we obtain the conclusion as in §\[section:proof\]. [99]{} , “Handbook of mathematical functions with formulas, graphs, and mathematical tables", U.S. Government Printing Office, Washington, D.C., 1964. On a Sobolev-type inequality, Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl. [20]{} (2009), no. 4, 379-386. Sharp Hardy inequalities in the half space with trace remainder term., Nonlinear Anal. [75]{} (2012), no. 14, 5466-5472. On Hardy inequalities with a remainder term, Ric. Mat. [59]{} (2010), no. 2, 265-280. A remark on Hardy type inequalities with remainder terms, Discrete Contin. Dyn. Syst. Ser. S [4]{} (2011), no. 4, 801-807. , “An introduction to Riemann-Finsler geometry", Springer-Verlag, New York, 2000. Trace Hardy-Sobolev-Maz’ya inequalities for the half fractional Laplacian, Commun. Pure Appl. Anal. [14]{} (2015), no. 2, 373-382. Hokkaido Math. J., [25]{} (1996), 537–566. Best remainder norms in Sobolev-Hardy inequalities, Indiana Univ. Math. J. [58]{} (2009), no. 3, 1051-1096. Kato’s inequality in the half space: an alternative proof and relative improvements, Proceeding of INDAM Workshop [Geometric properties for parabolic and elliptic PDE’s]{}, 85–96, Springer INdAM Ser., 2, Springer, Milan, 2013. Critical Hardy-Sobolev inequalities, J. Math. Pures Appl. (9) [87]{} (2007), no. 1, 37-56. “Calculus of variations. I. The Lagrangian formalism", Grundlehren der Mathematischen Wissenschaften \[Fundamental Principles of Mathematical Sciences\], 310. Springer-Verlag, Berlin, 1996. Hardy-Sobolev inequalities in a cone. Problems in mathematical analysis, J. Math. Sci. (N.Y.) [132]{} (2006), no. 4, 419?427. The Hardy-Sobolev inequalities in a cone, St. Petersburg Math. J. [22]{} (2011), no. 6, 997?1006. “Handbook of exact solutions for ordinary differential equations", Second edition, Chapman and Hall/CRC, Boca Raton, FL, 2003. : [Some trace Hardy type inequalities and trace Hardy-Sobolev-Maz’ya type inequalities,]{} J. Funct. Anal. [270]{} (2016), no. 11, 4117–4151. : [Anisotropic symmetrization,]{} Ann. Inst. H. Poincaré Anal. Non Linéaire [23]{} (2006), no. 4, 539–565.	{ "pile_set_name": "ArXiv" }
--- abstract: \| We investigate asymptotic behavior of solutions for nonlocal elliptic boundary value problems in plane angles and in ${\mathbb R}^2\backslash\{0\}$. Such problems arise as model ones when studying asymptotics of solutions for nonlocal elliptic problems in bounded domains. We obtain explicit formulas for the asymptotic coefficients in terms of eigenvectors and associated vectors of both adjoint nonlocal operators (acting in spaces of distributions) and formally adjoint (with respect to the Green formula) nonlocal problems. --- [Asymptotics of Solutions for Nonlocal Elliptic Problems\ in Plane Angles]{} [^1] Pavel Gurevich sect1.tex sect2.tex sect3.tex sect4.tex sect5.tex appendA.tex bibl.tex [^1]: This work was partially supported by Russian Foundation for Basic Research (grant No 01-01-01030) and by Russian Ministry for Education (grant No E00-1.0-195).	{ "pile_set_name": "ArXiv" }
--- abstract: \| In the framework of the topcolor-assisted technicolor $(TC2)$ model, we study production of the neutral top-pion $\pi_{t}^{0}$ in association with a high-$p_{T}$ jet at the $LHC$, which proceeds via the partonic processes $gg\longrightarrow \pi_{t}^{0}g$, $gq\longrightarrow \pi_{t}^{0}q$, $q\bar{q}\longrightarrow \pi_{t}^{0}g$, $gb(\bar{b})\longrightarrow \pi_{t}^{0}b(\bar{b})$, and $b\bar{b}\longrightarrow \pi_{t}^{0}g$. We find that it is very challenging to detect the neutral top-pion $\pi_{t}^{0}$ via the process $pp\longrightarrow \pi_{t}^{0}+jet+X\rightarrow t\bar{t}+jet+X$, while the possible signatures of $\pi_{t}^{0}$ might be detected via the process $pp\longrightarrow \pi_{t}^{0}+jet+X\rightarrow(\bar{t}c+t\bar{c})+jet+X$ at the $LHC$. author: - \| Shi-Hai Zhu, Chong-Xing Yue, Wei Liu, and Li Ding\ [Department of Physics, Liaoning Normal University, Dalian 116029, P. R. China]{} [^1] title: 'Production of the neutral top-pion $\pi_{t}^{0}$ in association with a high-$p_{T}$ jet at the $LHC$' --- [1. Introduction]{} The Higgs mechanism for the electroweak symmetry breaking $(EWSB)$ is still the untested part of the standard model $(SM)$. Searching for the $SM$ Higgs boson is one of the main tasks of the forthcoming Large Hadron Collider $(LHC)$, which has considerable capability to discover and measure almost all of its quantum properties \[1\]. However, if the $LHC$ finds evidence for a new scalar state, it may not necessarily be the $SM$ Higgs boson. Most of new physics models beyond the $SM$ predict the existence of new scalar states. These new particles may has production cross sections and branching ratios which differ from those of the $SM$ Higgs boson. Distinguishing the various new physics scenarios is an important task for current and near future high energy collider experiments. Thus, studying the production and decay of the new scalar states at the $LHC$ is of special interest. Due to the large gluon luminosity, the main production mechanism for a scalar Higgs boson at the $LHC$ is the partonic gluon fusion process $gg\rightarrow H$ \[2\], which is the so-called inclusive single Higgs boson production channel. In order to fully explore the Higgs detection capabilities of the $LHC$, one should investigate more exclusive channels, like e.g. Higgs production in association with a high-$p_{T}$ hadronic jet \[3\]. The main advantage of this channel is the richer kinematical structure of the events which allows for refined cuts increasing the signal-to-background ratio. So far, this production channel has been extensively studied in the $SM$ \[4,5\]. In the minimal supersymmetric standard model $(MSSM)$, the analogous process, i.e. scalar Higgs production in association with a high-$p_{T}$ jet was also extensively studied in Refs.\[6,7\]. Among various kinds of dynamical $EWSB$ theories, the topcolor scenario is attractive because it can explain the large top quark mass and provides a possible $EWSB$ mechanism \[8\]. The topcolor-assisted technicolor $(TC2)$ model \[9\] is one of the phenomenologically viable models, which has all essential features of the topcolor scenario. This model predicts three $CP$ odd top-pions ($\pi_{t}^{0}$,$ \pi_{t}^{\pm}$) with large Yukawa couplings to the third family. The aim of this paper is to consider the production of the neutral top-pion $\pi_{t}^{0}$ associated with a high-$p_{T}$ jet and compare our results with those for the Higgs boson from the $SM$ or the $MSSM$. We hope that our work can help the upcoming $LHC$ to test topcolor scenario and to differentiate various kinds of new physics models. In the rest of this paper, we will give our results in detail. In section 2, we will calculate the production cross section of the hadronic process $pp\longrightarrow \pi_{t}^{0}+jet+X$ and give a simply phenomenological analysis at the $LHC$. Our conclusion is represented in section 3. [2. Production of the neutral top-pion $\pi_{t}^{0}$ associated with a high-$p_{T}$ jet]{} In the $TC2$ model \[9\], topcolor interactions, which are not flavor-universal and mainly couple to third generation fermions, generally generate small contributions to $EWSB$ and give rise to the main part of the top quark mass. Thus, the top-pions $ \pi_{t}^{0,\pm}$ have large Yukawa couplings to the third generation fermions. Such features can result in large tree-level flavor changing couplings of the top-pions to the fermions when one writes the interactions in the fermion mass eigen-basis. Just as for the $SM$ Higgs boson, the couplings of the top-pion to a pair of quarks are proportion to the quark masses. The explicit form for the couplings of the neutral top-pion $ \pi_{t}^{0}$ to quarks, which are related to our calculation, can be written as \[9,10\]: $$\begin{aligned} \frac{im_{t}}{\sqrt{2}F_{t}}\frac{\sqrt{\nu_{W}^{2}-F_{t}^{2}}} {\nu_{W}}[k_{UR}^{tt}k_{UL}^{tt^{}}\bar{t}\gamma^{5}t \pi_{t}^{0}+\frac{m_{b}-m_{b}'}{m_{t}}\bar{b}\gamma^{5}b \pi_{t}^{0}+k_{UR}^{tc^{}}k_{UL}^{tt}\bar{t}P_{R}c \pi_{t}^{0}],\end{aligned}$$ where $\nu_{W}=\nu/\sqrt{2}\approx174GeV$, $P_{R}=(1+\gamma^{5})/2$ is the right-handed projection operator, $F_{t}\approx50GeV$ is the top-pion decay constant, and $m_{b}'\approx0.1\varepsilon m_{t}$ is the part of the bottom quark mass generated by extended technicolor interactions. $k_{UL(R)}$ are rotation matrices that diagonalize the up-quark mass matrix $M_{U}$ for which the Cabibbo-Kobayashi-Maskawa ($CKM$) matrix is defined as $V_{CKM}=k_{UL}^{+}k_{DL}$. To yield a realistic form of $V_{CKM}$, it has been shown that the values of the matrix elements $k_{UL(R)}^{ij}$ can be taken as \[10\]: $$\begin{aligned} k_{UL}^{tt}\approx1,\hspace{2cm} k_{UR}^{tt}=1-\varepsilon, \hspace{2cm}k_{UR}^{tc}\leq\sqrt{2\varepsilon-\varepsilon^{2}}.\end{aligned}$$ In our numerical estimation, we will take $k_{UR}^{tc}=\sqrt{2\varepsilon-\varepsilon^{2}}$ and take $\varepsilon$ as a free parameter, which is assumed to be in the range of $0.01\sim0.1$. Similar to the Higgs boson predicted by the $SM$ or the $MSSM$, the neutral top-pion $\pi_{t}^{0}$ can be produced at the $LHC$ in association with a high-$p_{T}$ jet through three partonic processes: gluon fusion ($gg\longrightarrow g\pi_{t}^{0}$), quark-gluon scattering ($q(\bar{q})g\longrightarrow q(\bar{q})\pi_{t}^{0}$), and quark-antiquark annihilation ($q\bar{q}\longrightarrow g\pi_{t}^{0}$). Although the gluon fusion and quark-gluon scattering partonic processes give main contributions to the production cross section for the hadronic process $pp\longrightarrow \pi_{t}^{0}+jet+X$ at the $LHC$, our numerical analysis include all of above three processes, which proceed at one-loop, as shown in $Fig.1 (a)\sim(e)$. Considering the small value of the decay constant $F_{t}$ and the relatively large bottom-quark mass, we also consider the contributions of the tree-level partonic processes $gb\longrightarrow \pi_{t}^{0}b$ and $b\bar{b}\longrightarrow \pi_{t}^{0}g$, as shown in $Fig.1 (f)\sim(h)$. The variant amplitudes corresponding to the Feynman diagrams as shown in $Fig.1$ can be written as: $$\begin{aligned} M_{(a)}=&&\frac{i}{4\sqrt{2}\pi^{2}}T^{c_{3}}_{ij}T^{c_{4}}_{ji}f^{c_{1}c_{2}c_{3}}g_{s}^{3} \frac{m_{t}^{2}(1-\varepsilon)\sqrt{\nu_{W}^{2}-F_{t}^{2}}} {F_{t}\nu_{W}(p_{1}+p_{2})^{2}}C_{0(a)} [(p_{2}-p_{1})_{\mu}\epsilon(p_{1})\cdot{}\epsilon(p_{2})\nonumber\\ &&+(-2p_{2}-p_{1})\cdot{}\epsilon(p_{1}) \epsilon_{\mu}(p_{2})+(2p_{1}+p_{2})\cdot{}\epsilon(p_{2})\epsilon_{\mu}(p_{1})] \epsilon^{\mu\nu\rho\sigma}p_{3\rho}p_{4\sigma}\epsilon_{\nu}(p_{3}),\\ M_{(b)}=&&\frac{i}{4\sqrt{2}\pi^{2}}T^{c_{3}}_{ij}T^{c_{4}}_{ji}f^{c_{1}c_{2}c_{3}}g_{s}^{3} \frac{m_{t}^{2}(1-\varepsilon)\sqrt{\nu_{W}^{2}-F_{t}^{2}}} {F_{t}\nu_{W}(p_{1}-p_{3})^{2}}C_{0(b)} [(-p_{2}-p_{1})\cdot{}\epsilon(p_{3})\epsilon_{\mu}(p_{1})\nonumber\\ &&+(p_{2}-p_{3})\cdot{}\epsilon(p_{1}) \epsilon_{\mu}(p_{3})+(p_{3}+p_{1})_{\mu}\epsilon(p_{1})\cdot{}\epsilon(p_{3})] \epsilon^{\mu\nu\rho\sigma}p_{4\rho}(p_{1}-p_{3})_{\sigma}\epsilon_{\nu}(p_{2}),\\ M_{(c)}=&&\frac{i}{4\sqrt{2}\pi^{2}}T^{c_{1}}_{ij}T^{c_{2}}_{jk}T^{c_{3}}_{ki}g_{s}^{3} \frac{m_{t}(1-\varepsilon)\sqrt{\nu_{W}^{2}-F_{t}^{2}}} {F_{t}\nu_{W}}\nonumber\\&&(-D_{0(c)}m_{t}^{3}\epsilon^{\mu\nu\rho\sigma}p_{1\sigma} +D_{0(c)}m_{t}^{3}\epsilon^{\mu\nu\rho\sigma}p_{4\sigma} -D_{0(c)}m_{t}\epsilon^{\rho\sigma\alpha\beta}p_{1\sigma}p_{2\alpha}p_{4\beta}g^{\mu\nu}\nonumber\\&& +D_{0(c)}m_{t}\epsilon^{\nu\sigma\alpha\beta}p_{1\sigma}p_{2\alpha}p_{4\beta}g^{\mu\rho} -D_{0(c)}m_{t}\epsilon^{\nu\rho\sigma\alpha}p_{1}^\mu p_{2\sigma}p_{4\alpha} +D_{0(c)}m_{t}\epsilon^{\nu\rho\sigma\alpha}p_{1\sigma}p_{4\alpha}p_{2}^\mu\nonumber\\&& -D_{0(c)}m_{t}\epsilon^{\nu\rho\sigma\alpha}p_{1\sigma}p_{2\alpha}p_{4}^{\mu} -D_{0(c)}m_{t}\epsilon^{\mu\sigma\alpha\beta}p_{1\sigma}p_{2\alpha}p_{4\beta}g^{\nu\rho} +2D^\nu_{(c)}m_{t}\epsilon^{\mu\rho\sigma\alpha}p_{1\sigma}p_{4\alpha}\nonumber\\&& +2D^\nu_{(c)}m_{t}\epsilon^{\mu\rho\sigma\alpha}p_{2\sigma}p_{4\alpha} +D_{0(c)}m_{t}\epsilon^{\mu\rho\sigma\alpha}p_{2\sigma}p_{4\alpha}p_{1}^\nu -D_{0(c)}m_{t}\epsilon^{\mu\rho\sigma\alpha}p_{1\sigma}p_{4\alpha}p_{2}^\nu\nonumber\\&& +D_{0(c)}m_{t}\epsilon^{\mu\rho\sigma\alpha}p_{1\sigma}p_{2\alpha}p_{4}^\nu -2D^\nu_{(c)}m_{t}\epsilon^{\mu\rho\sigma\alpha}p_{2\sigma}p_{4\alpha} -D_{0(c)}m_{t}\epsilon^{\mu\nu\sigma\alpha}p_{2\sigma}p_{4\alpha}p_{1}^\rho\nonumber\\&& -D_{0(c)}m_{t}\epsilon^{\mu\nu\sigma\alpha}p_{1\sigma}p_{4\alpha}p_{2}^\rho +D_{0(c)}m_{t}\epsilon^{\mu\nu\sigma\alpha}p_{1\sigma}p_{2\alpha}p_{4}^\rho +D_{\nu(c)}{}D_{(c)}^\nu m_{t}\epsilon^{\mu\nu\rho\sigma}p_{1\sigma}\nonumber\\&& -D_{\nu(c)}{}D_{(c)}^\nu m_{t}\epsilon^{\mu\nu\rho\sigma}p_{4\sigma} -2D_{\nu(c)}\cdot{}p_{2}m_{t}\epsilon^{\mu\nu\rho\sigma}p_{1\sigma} +2D_{\nu(c)}\cdot{}p_{2}m_{t}\epsilon^{\mu\nu\rho\sigma}p_{4\sigma}\nonumber\\&& -2D_{\nu(c)}\cdot{}p_{4}m_{t}\epsilon^{\mu\nu\rho\sigma}p_{2\sigma} +D_{0(c)}p_{1}\cdot{}p_{2}m_{t}\epsilon^{\mu\nu\rho\sigma}p_{4\sigma} -D_{0(c)}p_{1}\cdot{}p_{4}m_{t}\epsilon^{\mu\nu\rho\sigma}p_{2\sigma}\nonumber\\&& +D_{0(c)}p_{2}\cdot{}p_{2}m_{t}\epsilon^{\mu\nu\rho\sigma}p_{1\sigma} -D_{0(c)}p_{2}\cdot{}p_{4}m_{t}\epsilon^{\mu\nu\rho\sigma}p_{1\sigma}) \epsilon_{\mu}(p_{1})\epsilon_{\nu}(p_{2})\epsilon_{\rho}(p_{3})\\ M_{(d)}=&&\frac{1}{4\sqrt{2}\pi^{2}}T^{c_{1}}_{ij}T^{c_{1}}_{kl}T^{c_{2}}_{lk}g_{s}^{3} \frac{m_{t}^{2}(1-\varepsilon)\sqrt{\nu_{W}^{2}-F_{t}^{2}}} {F_{t}\nu_{W}(p_{1}-p_{3})^{2}}C_{0(d)} \bar{u}(p_{3})\gamma^{\mu}u(p_{1})g_{\mu\nu}\nonumber\\&& \epsilon^{\nu\lambda\rho\sigma}p_{4\rho}(p_{1}-p_{3})_{\sigma} \epsilon_{\lambda}(p_{2}),\\ M_{(e)}=&&\frac{1}{4\sqrt{2}\pi^{2}}T^{c_{1}}_{ij}T^{c_{1}}_{kl}T^{c_{2}}_{lk}g_{s}^{3} \frac{m_{t}^{2}(1-\varepsilon)\sqrt{\nu_{W}^{2}-F_{t}^{2}}} {F_{t}\nu_{W}(p_{1}+p_{2})^{2}}C_{0(e)} \bar{u}(p_{2})\gamma^{\mu}v(p_{1})g_{\mu\nu}\nonumber\\&& \epsilon^{\nu\lambda\rho\sigma}p_{3\rho}p_{4\sigma} \epsilon_{\lambda}(p_{3}),\\ M_{(f)}=&&\frac{1}{\sqrt{2}}T^{c_{1}}_{ij}g_{s}\frac{1}{(p_{1}+p_{2})^{2}-m_{b}^{2}} \frac{m_{b}-m_{b}^{'}}{F_{t}} \frac{\sqrt{\nu_{W}^{2}-F_{t}^{2}}}{\nu_{W}}\nonumber\\&& \bar{u}(p_{3})\gamma^{5}(\not{p_{1}}+\not{p_{2}}-m_{b}) \gamma^{\mu}u(p_{1})\epsilon_{\mu}(p_{2}),\\ M_{(g)}=&&\frac{1}{\sqrt{2}}T^{c_{1}}_{ij}g_{s}\frac{1}{(p_{1}-p_{3})^{2}-m_{b}^{2}} \frac{m_{b}-m_{b}^{'}}{F_{t}} \frac{\sqrt{\nu_{W}^{2}-F_{t}^{2}}}{\nu_{W}}\nonumber\\&& \bar{u}(p_{4})\gamma^{\mu}(\not{p_{1}}-\not{p_{3}}-m_{b}) \gamma^{5}u(p_{1})\epsilon_{\mu}(p_{2}),\\ M_{(h)}=&&\frac{1}{\sqrt{2}}T^{c_{1}}_{ij}g_{s}\frac{1}{(p_{1}-p_{3})^{2}-m_{b}^{2}} \frac{m_{b}-m_{b}^{'}}{F_{t}} \frac{\sqrt{\nu_{W}^{2}-F_{t}^{2}}}{\nu_{W}}\nonumber\\&& \bar{v}(p_{2})\gamma^{\mu}(\not{p_{1}}-\not{p_{3}}-m_{b}) \gamma^{5}u(p_{1})\epsilon_{\mu}(p_{4}).\end{aligned}$$ Here $p_{1},p_{2}$ are the momenta of the incoming states, and $p_{3},p_{4}$ are the momenta of the outgoing final states. The $T^{c}_{ij}$ are the $SU(3)$ color matrices and the $f^{c_{1}c_{2}c_{3}}$ are the antisymmetric $SU(3)$ structure constants in which $i,j$ are the color indices and $c_{1},c_{2},c_{3}$ are the indices of gluon. The three-point and four-point standard functions $C_0,D_0,D_1,D_{\nu}$ \[11,12\] for different Feynman diagrams are defined as: $$\begin{aligned} &&C_{0(a)}=C_{0(a)}(p_{1}+p_{2},-p_{3},m_{t},m_{t},m_{t}), C_{0(b)}=C_{0(b)}(p_{1}-p_{3},p_{2},m_{t},m_{t},m_{t}),\\ &&C_{0(d)}=C_{0(d)}(p_{1}-p_{3},p_{2},m_{t},m_{t},m_{t}), C_{0(e)}=C_{0(e)}(p_{1}+p_{2},-p_{3},m_{t},m_{t},m_{t});\\ &&D_{0(c)}=D_{0(c)}(p_{1},p_{2},-p_{3},m_{t},m_{t},m_{t},m_{t}), D_{1(c)}=D_{1(c)}(p_{1},p_{2},-p_{3},m_{t},m_{t},m_{t},m_{t}),\\ &&D_{\nu(c)}=p_{1\nu}D_{1(c)}(1)+p_{2\nu}D_{1(c)}(2)+p_{3\nu}D_{1(c)}(3).\\\end{aligned}$$ Each loop diagram is composed of some scalar loop functions, which are calculated by using LoopTools \[12\]. The hadronic cross section at the $LHC$ is obtained by convoluting the partonic cross sections with the parton distribution functions $(PDFs)$. In our numerical calculation, we will use $CTEQ6L\hspace{0.2cm}PDFs$ \[13\] for the gluon and quark $PDFs$. The renormalization scale $\mu_{R}$ and the factorization scale $\mu_{F}$ are chosen to be $\mu_{R}=\mu_{F}=m_{\pi_{t}}$ for the gluons and the light quarks, and to be $\mu_{R}=\mu_{F}=m_{\pi_{t}}/4$ for the bottom-quark, in which $m_{\pi_{t}}$ is the mass of the neutral top-pion $\pi_{t}^{0}$. To make our predictions more realistic and high-$p_{T}$ jet not too close to the beam axis, we require that the transverse momentum $p_{T}$ and pseudorapidity $\eta$ of the hadronic jet satisfy: $p_{T}>30GeV$ and $\|\eta\|<4.5$, which have been used in previous $MSSM$ studies for the $LHC$ \[6,7\]. From the above discussions we can see that the production cross section $\sigma$ for the hadronic process $pp\rightarrow\pi_{t}^{0}+jet+X$ is dependent on the free parameters $\varepsilon$ and $m_{\pi_{t}}$. Similar with Ref.\[14\], we will assume that the free parameters $\varepsilon$ and $m_{\pi_{t}}$ are in the range of $0.01\sim0.1$ and $200GeV\sim500GeV$, respectively. Our numerical results are shown in $Fig.2$, in which we plot the cross section $\sigma$ as a function of the mass parameter $m_{\pi_{t}}$ for three values of the parameter $\varepsilon$. One can see from $Fig.2$ that $\sigma$ is insensitive to the free parameter $\varepsilon$. For $\varepsilon=0.05$ and $200GeV\leq m_{\pi_{t}}\leq 500GeV$, the value of the production cross section $\sigma$ is in the range of $18.3pb\sim2.1pb$. Observably, if we assume that the $\pi_{t}^{0}$ mass $m_{\pi_{t}}$ is equal to that of the $SM$ Higgs boson $H$ or the $MSSM$ Higgs boson $H^{0}$, the cross section for the production of the neutral top-pion $\pi_{t}^{0}$ associated with a high-$p_{T}$ is significantly larger than that of the $SM$ Higgs boson $H$ \[4,5\] or the $MSSM$ Higgs boson $H^{0}$ \[6,7\]. This is because the $\pi_{t}^{0} t \overline{t}$ coupling is larger than that for the $SM$ Higgs boson $H$ or the $MSSM$ Higgs boson $H^{0}$. To see contributions of the different partonic processes to the total hadronic cross section, we plot the hadronic cross sections of the partonic processes $gg\rightarrow\pi_{t}^{0}g, qg\rightarrow q\pi_{t}^{0} (q=u,c,d,s,\bar{u},\bar{c},\bar{d},\bar{s}), q\bar{q}\rightarrow\pi_{t}^{0}g (q=u,c,d,s), gb(\bar{b})\longrightarrow \pi_{t}^{0}b(\bar{b})$, and $b\bar{b}\longrightarrow \pi_{t}^{0}g$ for $\varepsilon$=0.05 in $Fig.3$. We see that the production of the neutral top-pion $\pi_{t}^{0}$ in association with a high-$p_{T}$ jet is dominated by the partonic process $gg\rightarrow\pi_{t}^{0}g$, which is similar with the Higgs boson production associated with a high-$p_{T}$ in the $SM$ and the $MSSM$. However, for the $MSSM$ model, the contributions of the $b\bar{b}$ channel can be significantly large, depending the free parameters. However, this is not the case for the $TC2$ model. For $0.02\leq\varepsilon\leq 0.08$ and $200GeV\leq m_{\pi_{t}}\leq 500GeV$, the hadronic cross section for the partonic process $b\bar{b}\longrightarrow \pi_{t}^{0}g$ is only in the range of $1.6fb\sim46fb$, which is several orders of magnitude smaller than that for the partonic process $gg\rightarrow\pi_{t}^{0}g$. It is well known that the mass of the $SM$ Higgs boson $H$ is generally smaller than $200GeV$, one can use the decay channels $H\rightarrow\gamma\gamma$, $H\rightarrow\tau^{+}\tau^{-}$ or $H\rightarrow W^{+}W^{-}$ to consider the $SM$ Higgs boson signatures generated by the hadronic process $pp\longrightarrow H+jet+X$ at the $LHC$ \[5\]. For the neutral top-pion $\pi_{t}^{0}$, its main decay modes are $t\bar{t}, \bar{t}c(t\bar{c}), b\bar{b}, gg$, and $ \gamma\gamma$. For $m_{t}\leq m_{\pi_{t}}\leq 2m_{t}$, $\pi_{t}^{0}$ mainly decays to $\bar{t}c$ and $t\bar{c}$. It has been shown that the value of the branching ratio $Br(\pi_{t}^{0}\rightarrow\bar{t}c+t\bar{c})$ is larger than 90% for $m_{\pi_{t}}=250GeV$ and $\varepsilon\geq0.02$ \[15\]. Thus, for $m_{t}<m_{\pi_{t}}\leq 2m_{t}$, the production of neutral top-pion $\pi_{t}^{0}$ associated with a high-$p_{T}$ hadronic jet can easily transfer to the $tc+jet$ event. This final state generates characteristic signatures at the $LHC$ experiments. So we further calculate its production rate. We find that, for $\varepsilon\leq0.08$ and $m_{\pi_{t}}\leq 350GeV$, the production cross section of the hadronic process $pp\rightarrow(\bar{t}c+ t\bar{c})+jet+X$ is larger than $19.4pb$. If we assume the yearly integrated luminosity $\pounds_{int}=100fb^{-1}$ for the $LHC$ with $\sqrt{s}=14TeV$, then there will be $1.94\times 10^{6}\sim 5.3\times 10^{5} \hspace{0.2cm} tc+jet$ events to be generated per year for $0.02\leq\varepsilon\leq0.08$ and $200GeV\leq m_{\pi_{t}}\leq 340GeV$, as shown in Fig.4. For the $tc+jet$ event, the peak of the invariant mass distribution of $tc$ is narrow. To identify $tc$, one needs reconstruct top quark from its mainly decay mode $Wb$ and the b-tagging and c-tagging are also needed. Furthermore, in the case of the $W$ hadronic decay, the $tc+jet$ event will generate the $bjjcj$ final state, while for the $W$ leptonic decay, it will generate the $bl\nu cj$ final state. For the former final state, the $SM$ background is $jjjjj$ and the $SM$ backgrounds of the later final state mainly come from the $t\bar{t}$, $tW$ and $Wjjj$ production process, which have been analyzed in Ref.\[16\]. They have shown that suitable kinematical cuts on the observed particles is more than enough to obtain a clear and statistically meaningful flavor-changing signal. Thus we expect that the possible signatures of the neutral top-pion $\pi_{t}^{0}$ might be detected via the decay channel $\pi_{t}^{0}\rightarrow\bar{t}c+t\bar{c}$ at the $LHC$ experiments. For $m_{\pi_{t}}>2m_{t}$, the neutral top-pion $\pi_{t}^{0}$ mainly decays to $t\bar{t}$ and the hadronic process $pp\rightarrow\pi_{t}^{0}+jet+X$ can give rise to the $t\bar{t}+jet$ event. Its production rate can reach 15pb for $m_{\pi_{t}}\geq 400GeV$ and $\varepsilon\leq 0.08$. This kind of events have been calculated at $NLO$ in the $SM$ \[17\]. It has shown that, for the renormalization and factorization scales having $\mu_{R}=\mu_{F}=\mu=m_{t}$, the $NLO$ cross section for $t\bar{t}+jet$ production at the $LHC$ is larger than $500pb$. Thus, the production cross section of the $t\bar{t}+jet$ final state coming from $TC2$ is smaller than that coming from the $SM$ by at least two orders of magnitude. It is very challenging to detect the possible signals of $\pi_{t}^{0}$ via the process $pp\longrightarrow \pi_{t}^{0}+jet+X\rightarrow t\bar{t}+jet+X$. [3. Conclusion]{} The production of a scalar state (the $SM$ Higgs boson, the $MSSM$ Higgs boson, etc) associated with a high-$p_{T}$ jet allows for refined cuts increasing the signal-to-background ratio, which is considered advantageous for scalar detection even though its production rate is lower than that for totally inclusive single scalar state production. In the context of the $TC2$ model, we consider the production of the neutral top-pion $\pi_{t}^{0}$ accompanied by a high-$p_{T}$ jet at the $LHC$. This production channel proceeds by the partonic processes $gg\longrightarrow \pi_{t}^{0}g$, $gq\longrightarrow \pi_{t}^{0}q$, $q\bar{q}\longrightarrow \pi_{t}^{0}g$, $gb(\bar{b})\longrightarrow \pi_{t}^{0}b(\bar{b})$, and $b\bar{b}\longrightarrow \pi_{t}^{0}g$. We find that, for $m_{\pi_{t}}$ equaling to the mass of the scalar sate predicted by the $MSSM$, the hadronic production cross section of the process $pp\longrightarrow \pi_{t}^{0}+jet+X$ is much larger than that for the $MSSM$ scalar state. For $m_{t}<m_{\pi_{t}}\leq2m_{t}$, the main decay channel is $\pi_{t}^{0}\rightarrow\bar{t}c+t\bar{c}$. There will be a large number of the $ tc+jet$ events to be generated which can generate characteristic signal at the $LHC$ experiment. So we might detect the possible signatures of the neutral top-pion $\pi_{t}^{0}$ via the process $pp\longrightarrow \pi_{t}^{0}+jet+X\rightarrow(\bar{t}c+t\bar{c})+jet+X$ at the $LHC$. [Acknowledgments]{} Shi-Hai Zhu would like to thank Lei Wang for useful discussions. This work was supported in part by the National Natural Science Foundation of China under Grants No.10675057 and Foundation of Liaoning Educational Committee(2007T086). [99]{} For a recent review, see: N. E. Adam et al, arXiv:* 0803.1154\[hep-ph\]. H. M. Georgi, S. L. Glashow, M. E. Machacek and D. V. Nanopoulos, [Phys. Rev. Lett.]{} [40]{}, 692(1978). M. Chaichian, I. Liede, J. Lindfors, D. P. Roy, [Phys. Lett. B]{}[198]{}, 416(1987)\[Erratum-ibid [B]{}[205]{}, 595(1988)\]; R. K. Ellis, I. Hinchliffe, M. Soldate, J. J. van der Bij, [Nucl. Phys. B]{}[297]{}, 221(1988); U. Baur, E. W. Nigel Glover, [Nucl. Phys. B]{}[339]{}, 38(1990). S. Abdellin et al., [Phys. Lett. B]{}[431]{}, 410(1998); B. Mellado, W. Quayle, S. L. Wu, [Phys. Lett. B]{}[611]{}, 60(2005); [Phys. Rev. D]{}[76]{}, 093007(2007). C. Anastasiou, K. Melnikov and F. Petriello, [Phys. Rev. Lett.]{} [93]{}, 262002(2004); [Nucl. Phys. B]{}[ 724]{}, 197(2005); G. Bozzi, S. Catani, D. de Florian and M. Grazzini, [Phys. Lett. B]{} [564]{}, 65(2003); S. Catani, D. de Florian, M. Grazzini and P. Nason, [JHEP]{} [0307]{}, 028(2003); D. de Florian, M. Grazzini and Z. Kunszt, [Phys. Rev. Lett.]{} [82]{}, 5209(1999); D. de Florian, Z. Kunszt and W. Vogelsang, [JHEP]{} [0602]{}, 047(2006). O. Brein and W. Hollik, [Phys. Rev. D]{}[68]{}, 095006(2003); B. Field, S. Dawson and J. Smith, [Phys. Rev. D]{}[69]{}, 074013(2004); U. Langenegger, M. Spira, A. Statradumov, P. Trucb, [JHEP]{} [0606]{}, 035(2006). O. Brein and W. Hollik, [Phys. Rev. D]{}[76]{}, 035002(2007); K. Bonciani, G. Degrassi, A. Vicini, [JHEP]{} [0711]{}, 095(2007); O. Brein and W. Hollik, arXiv: 0710.4781\[hep-ph\]. C. T. Hill and E. H. Simmons, [Phys. Rept.]{} [381]{}, 235(2003); \[[Erratum-ibid]{}, [390]{}, 553(2004)\]. C. T. Hill, [Phys. Lett. B]{} [345]{}, 483(1995); K. D. Lane and E. Eichten, [Phys. Lett. B]{} [352]{}, 382(1995); K. D. Lane, [Phys. Lett. B]{} [433]{}, 96(1998); G. Cvetic, [Rev. Mod. Phys.]{} [71]{}, 513(1999). G. Burdman, [Phys. Rev. Lett.]{} [83]{}, 2888(1999); H.-J. He, C.-P. Yuan, [Phys. Rev. Lett.]{} [ 83]{}, 28(1999); H.-J. He, S. Kanemura, C.-P. Yuan, [Phys. Rev. Lett.]{} [89]{}, 101803(2002). G. ’t Hooft and M. J. G. Veltman, [Nucl. Phys. B]{}[ 153]{}, 365(1979). T. Hahn, M. Perez-Victora, [Comput. Phys. Commun.]{} [ 118]{}, 153(1999); T. Hahn, [Nucl. Phys. Proc. Suppl.]{} [ 135]{}, 333(2004). J. Pumplin et al. (CTEQ Collaboration), [JHEP]{} [02]{}, 032(2006). C. -X. Yue, Z. -J. Zong, L. -L. Xu and J. -X. Chen, [Phys. Rev D]{}[73]{}, 015006(2006). C. -X. Yue, Y. -P. Kuang, and G. -G. Lu, [J. Phys. G]{}[23]{}, 163(1997); C. -X. Yue, Q. -J. Xu, G. -L. Liu, and J. -T. Li, [Phys. Rev D]{}[63]{}, 115002(2001). F. del Aguila and J. A. Aguilar-Saavedra, [Nucl. Phys. B]{}[ 576]{}, 56(2000). S. Diffmaier, P. Uwer and S. Weinzierl, [Phys. Rev. Lett.]{} [98]{}, 262002(2007);arXiv: 0804.4389\[hep-ph\]. [^1]: E-mail:[email protected]	{ "pile_set_name": "ArXiv" }
--- abstract: \| The C++ Standard Template Library is the flagship example for libraries based on the generic programming paradigm. The usage of this library is intended to minimize the number of classical C/C++ errors, but does not warrant bug-free programs. Furthermore, many new kinds of errors may arise from the inaccurate use of the generic programming paradigm, like dereferencing invalid iterators or misunderstanding remove-like algorithms. In this paper we present some typical scenarios that may cause runtime or portability problems. We emit warnings and errors while these risky constructs are used. We also present a general approach to emit “customized” warnings. We support the so-called “believe-me marks” to disable warnings. We present another typical usage of our technique, when classes become deprecated during the software lifecycle. title: C++ Standard Template Library by template specialized containers --- Introduction {#intro} ============ The C++ Standard Template Library (STL) was developed by generic programming approach [@austern:stl]. In this way containers are defined as class templates and many algorithms can be implemented as function templates. Furthermore, algorithms are implemented in a container-independent way, so one can use them with different containers [@stroustrup:cpp]. C++ STL is widely-used because it is a very handy, standard C++ library that contains beneficial containers (like list, vector, map, etc.) and a large number of algorithms (like sort, find, count, etc.) among other utilities [@czarnecki:generative]. The STL was designed to be extensible [@musser:generic]. We can add new containers that can work together with the existing algorithms. On the other hand, we can extend the set of algorithms with a new one that can work together with the existing containers. Iterators bridge the gap between containers and algorithms [@becker:iterators]. The expression problem [@torgersen:expr] is solved with this approach. STL also includes adaptor types which transform standard elements of the library for a different functionality [@alexandrescu:modern]. However, the usage of C++ STL does not guarantee bugless or error-free code [@devai:tool]. Contrarily, incorrect application of the library may introduce new kinds of problems [@stlmetric]. One of the problems is that the error diagnostics are usually complex, and very hard to figure out the root cause of a program error [@zolman:message; @zolyomi:introspection]. Violating requirement of special preconditions (e.g. sorted ranges) is not tested, but results in runtime bugs [@pataki:overhead]. A different kind of stickler is that if we have an iterator object that pointed to an element in a container, but the element is erased or the container’s memory allocation has been changed, then the iterator becomes invalid [@pataki:ranges]. Further reference of using invalid iterators causes undefined behaviour [@pataki:safestl]. Another common mistake is related to removing algorithms. The algorithms are container-independent, hence they do not know how to erase elements from a container, just relocate them to a specific part of the container, and we need to invoke a specific erase member function to remove the elements phisically. Since, for example the `remove` algorithm does not actually remove any element from a container [@meyers:STL]. Some of the properties are checked at compilation time [@gregor:concepts]. For example, the code does not compile if one uses sort algorithm with the standard list container, because the list’s iterators do not offer random accessibility [@jarvi:specialization]. Other properties are checked at runtime [@pirkelbauer:runtime], like the standard vector container offers an `at` method which tests if the index is valid and it raises an exception otherwise [@pataki:soundness]. Unfortunately, there are still a large number of properties that are tested neither at compilation-time nor at run-time. The observance of these properties is in the charge of the programmers [@devai:stl]. On the other hand, type systems can provide a high degree of safety at low operational costs. As part of the compiler, they discover many semantic errors very efficiently. Associative containers (e.g. `multiset`) use functors exclusively to keep their elements sorted. Algorithms for sorting (e.g. `stable_sort`) and searching in ordered ranges (e.g. `lower_bound`) are typically used with functors because of efficiency. These containers and algorithms need strict weak ordering. Containers become inconsistent if used functors do not meet the requirement of strict weak ordering [@pataki:functors]. Certain containers have member functions with the same names as STL algorithms. This phenomenon has many different reasons, for instance efficiency, safety or avoidance of compilation errors. For example, as mentioned before list’s iterators cannot be passed to `sort` algorithm, hence code with this mistake cannot be compiled [@lupin:modularization]. To overcome this problem list has a member function called `sort`. In these cases, although the code compiles, the member function calls are preferable to the usage of generic algorithms. Whereas C++ STL is pre-eminent in a sequential realm, it is not aware of multicore environment [@austern:range]. For example, the Cilk++ language aims at multicore programming. This language extends C++ with new keywords and one can write programs for multicore architectures easily. Although the language does not contain an efficient multicore library, just the C++ STL only which is an efficiency bottleneck in multicore environment. We develop a new STL implementation for Cilk++ to cope with the challenges of multicore architectures[@lupin:multicore]. This new implementation can be safer solution, too. Hence, our safety extensions will be included in the new implementation. However, the advised techniques presented in this paper concern to the original C++ STL, too. In this paper we argue for an approach that generates warnings or errors when a template container is instantiated with improper parameters. These instantiations mean erroneous, unportable code or other weird compilation effects. A general technique is presented to express custom warnings at compilation time. Our technique is able to indicate the usage of deprecated classes. This paper is organized as follows. In Section \[warnings\] we present an approach to generate “customized” warnings at compilation time. After, in Section \[vectorbool\] we describe the specialized vector container which contains boolean values. We show why this container is problematic, and argue for warnings when it is in use. We explain the forbidden containers of auto pointers and present an approach to disable their usage by template specializations. In Section \[marks\] the so-called believe-me marks are introduced. Finally, this paper concludes in Section \[conc\]. Generation of warnings {#warnings} ====================== Compilers cannot emit warnings based on the erroneous usage of the library. STLlint is the flagship example for external software that is able to emit warnings when the STL is used in an incorrect way [@gregor:stllint]. We do not want to modify the compilers, so we have to enforce the compiler to indicate these kinds of potential problems. However, `static_assert` as a new keyword is introduced in C++0x to emit compilation errors based on conditions, but no similar construct is designed for warnings. template <class T> inline void warning( T t ) { } struct VECTOR_BOOL_IS_IN_USE { }; // ... warning( VECTOR_BOOL_IS_IN_USE() ); When the `warning` function is called, a dummy object is passed. This dummy object is not used inside the function template, hence this is an unused parameter. Compilers emit warning to indicate unused parameters. Compilation of `warning` function template results in warning messages, when it is referred and instantiated [@pataki:funcframework]. No warning message is shown if it is not referred. In the warning message the template argument is referred. New dummy type has to be written for every new kind of warning. Different compilers emit this warning in different ways. For instance, Visual Studio emits the following message: warning C4100: 't' : unreferenced formal parameter ... see reference to function template instantiation 'void warning<VECTOR_BOOL_IS_IN_USE>(T)' being compiled with [ T=VECTOR_BOOL_IS_IN_USE ] And g++ emits the following message: In instantiation of 'void warning(T) [with T = VECTOR_BOOL_IS_IN_USE]': ... instantiated from here ... warning: unused parameter 't' Unfortunately, implementation details of warnings may differ, thus there is no universal solution to generate custom warnings. This approach of warning generation has no runtime overhead inasmuch as the compiler optimizes the empty function body. On the other hand—as the previous examples show—the message refers to the warning of unused parameter, incidentally the identifier of the template argument type is appeared in the message. The weirdest vector {#vectorbool} =================== In this section we present the basic idea behind the specialized `vector<bool>` container. We present the pros and cons of this weird type. We argue for generate warnings at compilation-time if a programmer uses `vector<bool>` because it is the embodiment of the weird container. Many programmers think that the `vector<bool>` is the instantiation of STL’s `vector` template, but it is not true. On many platforms `sizeof( int ) == sizeof( bool )` because of reverse compatibility. (In the C programming language `int` type has been used to represent Boolean values.) Hence, the `vector<bool>` is a template specialized container to develop a more advanced, denser implementation for boolean values. This representation is able to represent 32 boolean values on 4 bytes. The following code sketch represents the connection between `vector<bool>` and `vector` template: template <class T, class Alloc = std::alloc> class vector { T* p; size_t capacity; size_t size; public: vector() { // ... } void push_back( const T& t ) { // ... } // ... }; template <class Alloc> class vector<bool, class Alloc> { // dense representation of vector bool // No bool* member public: // public interface is similar to the previous one void push_back( const bool& t ) { // ... } vector() { //... } }; So, the `vector<bool>` has a special representation to handle dense boolean values. It is designed to be effective when someone stores boolean values. But it has weird behaviour compared to the `vector` template: std::vector<int> a; a.push_back( 3 ); int* p = &a[0]; std::vector<bool> b; b.push_back( true ); bool* q = &b[0]; The previous code does not compile because of the `bool* q = &b[0];` assignment. However, when the template `vector` is in use, its counterpart does compile. It is a contradiction in terms, because this way the `vector<bool>` cannot meet the requirements of C++ Standard. Hence, it is not advised to use. Let us see the background of this compilation issue: template <class T, class Alloc = std::alloc> class vector { T* p; //.. public: T& operator[]( int idx ) { return p[idx]; } const T& operator[]( int idx ) const { return p[idx]; } // ... }; template <class Alloc> class vector<bool, class Alloc> { // dense representation of vector bool // No bool* member public: class bool_reference { // ... }; bool_reference operator[]( int idx ) { // ... } }; Because the `vector<bool>` does not hold actual bool values it cannot return `bool&`. Hence, a proxy class is developed which actually simulates `bool&`. However, conversions cannot be defined between pointer to a `bool_reference` and a pointer to a `bool`. This behaviour can be much more appalling, when the programmer uses `vector` as a base class. Arcane error messages are emitted when the subtype is instantiated with `bool`. Unfortunately, most of STL references hardly mention that `vector<bool>` is not the instantiation of template, but a completely different class. It would be useful if the compiler indicated if the programmer used `vector<bool>` container, even intentionally or inadvertently. Now it is not difficult to emit warning with the presented function. Fortunately, `vector<bool>` is still a class template because the type of its allocator is a template parameter. So, the compilation warning is emitted only when this template class is instantiated, hence someone uses it: template<class Allocator> class vector<bool, Allocator> { // ... public: vector() { warning( VECTOR_BOOL_IS_IN_USE() ); // ... } template<class InputIterator> vector( InputIterator first, InputIterator last ) { warning( VECTOR_BOOL_IS_IN_USE() ); // ... } vector( size_t n, const bool& value = bool() ) { warning( VECTOR_BOOL_IS_IN_USE() ); // ... } vector( const vector& rhs) { warning( VECTOR_BOOL_IS_IN_USE() ); // ... } }; In Section \[warnings\] the emitted warning message can be seen. Containers of auto pointers {#coaps} =========================== In this section the containers of auto pointers are detailed. We present their motivation and reason why are they problematic. We present a solution to forbid the usage of these kinds of containers. Usually, auto pointers (`std::auto_ptr` objects) make easier to manage objects in the heap memory. This class assists in memory management. The auto pointers deallocate the pointed memory when they are gone out of scope [@stroustrup:cpp]. Hence, they prevent memory leaks: void f() { std::auto_ptr<int> p( new int( 5 ) ); // no memory leak } Containers of STL are template classes, so technically they should be instantiated with auto pointers and store auto pointers that point to the heap: std::vector<std::auto_ptr<int> > v; v.push_back( new int( 7 ) ); // ... The previous code snippet seems to be safe. On the other hand, the C++ Standardization Committee forbid the usage of containers of auto pointers (COAPs). The motivation behind this idea is that the copy of auto pointers is strange: std::auto_ptr<int> p( new int( 3 ) ); std::auto_ptr<int> q = p; // At this point p is null pointer The copied auto pointer becomes null pointer. Only one auto pointer is able to point to any object in the heap. This one is responsible for the deallocation. So, if COAPs are not be forbidden, the following code snippet results in a very strange behaviour: struct Auto_ptr_less { bool operator()( const std::auto_ptr<int>& a, const std::auto_ptr<int>& b ) { return a < b; } }; std::vector<std::auto_ptr<int> > v; v.push_back( new int( 7 ) ); // ... std::sort( v.begin(), v.end(), Auto_ptr_less() ); Some of the pointers may become null pointer because of the assignments during swapping vector’s elements when it is necessary. This is the reason why COAPs are forbidden. Unfortunately, some of the compilers and STL platforms are still permitting the usage of COAPs, some of them are not. This inhibits the writing of portable code [@meyers:STL]. We argue for an extension to emit compilation error if COAPs are in use. We have to create specializations for auto pointers. The trick that is we do not write the implementation for the auto pointer specializations. Thus, these specializations are declared, but are not defined types. For instance, the vector declaration can be the following: template <class T, class Alloc> class vector< std::auto_ptr<T>, Alloc>; The instantiation of a COAP results in the hereinafter error message: error: aggregate 'std::vector<std::auto_ptr<int>, std::allocator<std::auto_ptr<int> > > v' has incomplete type and cannot be defined We have to develop these declarations for all standard containers. These declarations mean bugless and more portable code. Believe-me marks {#marks} ================ Generally, warnings should be eliminated. On the other hand, the usage of `vector<bool>` does not necessarily mean a problem. It can be used safely. However, we cannot disable the generated warning if it is in use. Believe-me marks [@kto:subtype] are used to identify the points in the program text where the type system cannot obtain if the used construct is risky. For instance, in the hereinafter example, the user of the library asks the type system to “believe” that the programmer is conscious of the specialized vector container. This way we enforce the user to reason about the parameters of containers. First, we create a new type which stands for the believe-me mark: struct I_KNOW_VECTOR_BOOL { }; After, we extend the vector template container with one new template parameter. The new template parameter has default parameter value, so it is reverse compatible with the original container. This parameter has not been taken advantage of, and has no effect on the implementation: template <class T, class Alloc = std::alloc, class Info = int> class vector { }; Let us consider, that the original implementation of `vector<bool>` which does not generate warning has been removed to a new template class: template <class Alloc> class __VectorBool { // original implementation of vector<bool> }; The new template parameter has effect on the `vector<bool>` specialization: template <class Alloc> class vector<bool, I_KNOW_VECTOR_BOOL, Alloc>: public __VectorBool<Alloc> { }; template <class Alloc> class vector<bool, Alloc, I_KNOW_VECTOR_BOOL>: public __VectorBool<Alloc> { }; template <class Alloc, class Info> class vector<bool, Alloc, Info>: public __VectorBool<Alloc> { public: vector(): __VectorBool<Alloc>() { warning( VECTOR_BOOL_IS_IN_USE() ); } template<class InputIterator> vector( InputIterator first, InputIterator last ): __VectorBool<Alloc>( first, last ) { warning( VECTOR_BOOL_IS_IN_USE() ); } vector( size_t n, const bool& value = bool() ): __VectorBool<Alloc>( n, value ) { warning( VECTOR_BOOL_IS_IN_USE() ); } vector( const vector& rhs): __VectorBool<Alloc>( rhs ) { warning( VECTOR_BOOL_IS_IN_USE() ); } }; In this case no compilation warning is emitted if the last added template parameter is `I_KNOW_VECTOR_BOOL`, otherwise the mentioned warning can be seen during compilation. Deprecated classes {#deprecated} ================== In section \[vectorbool\] we generated warnings when a template-specialized class was used. A similar idea can be mentioned. It would be useful to generate warnings when the usage of classes becomes unsupported. A common idea during a software lifecycle is, that some of the classes are not deleted from the project, but their usage is not advised. These classes are called deprecated. Deprecated annotation can be added to classes in Java. Instantiation of deprecated classes results in compilation warnings [@juhasz:teachingjava]. However, no similar technique is used in C++. First, we create some utility classes for warning generation: struct DeprecatedClass { }; template <class DEPRECATED> struct Deprecated { Deprecated() { warning( DeprecatedClass() ); } }; The role of the template parameter in `Deprecated` struct is to pass the identifier of deprecated class to the emitted warning. Now, let us consider that the following class becomes deprecated during software lifecycle: class Foo { // ... public: Foo( int a, int b) { // ... } }; The user has to add one more base class to the deprecated class. This does not mean limitation because the C++ programming language supports multiple inheritance. For example: class Foo: public Deprecated<Foo> { // very same... }; The following warning is received from the compiler: In instantiation of 'void warning(T) [with T = DeprecatedClass]': ... instantiated from 'Deprecated<DEPRECATED>::Deprecated() [with DEPRECATED = Foo]' ... instantiated from here ... warning: unused parameter 't' However, this message is received irrespectively of its usage. If the usage is important, the deprecated class or a called method or constructor must be a template. This transformation cannot be executed automatically with the respect of client code. Our future work is to overcome this situation. We do not advise to make believe-me marks for the deprecated classes inasmuch as always exists a better approach to use. Conclusions {#conc} =========== C++ STL is the most widely-used library based on the generic programming paradigm. It is efficient and convenient, but the incorrect usage of the library results in weird or undefined behaviour. In this paper we argue for some extensions to make the STL itself safer. Not supported or not advised instantiations result in compilation warnings and errors to prevent unportable or defective code. We present an effective approach to generate custom warnings. Believe-me marks are also written to disable warning messages. With our technique classes can be marked deprecated, too. Acknowledgements {#acknowledgements .unnumbered} ================ The Project is supported by the European Union and co-financed by the European Social Fund (grant agreement no. TÁMOP 4.2.1/B-09/1/KMR-2010-0003). [99]{} [A. Alexandrescu](http://erdani.com/), Modern C++ Design: Generic Programming and Design Patterns Applied, [Addison-Wesley](http://www.pearsonhighered.com/), Reading, MA, 2001. M. H. Austern, Generic Programming and the STL: Using and Extending the C++ Standard Template Library, [Addison-Wesley](http://www.pearsonhighered.com/), Reading, MA, 1998. M. H. Austern, R. A. Towle, A. A. Stepanov, Range partition adaptors: a mechanism for parallelizing STL, ACM SIGAPP Applied Computing Review 4, 1 (1996) 5–6. T. Becker, STL & generic programming: writing your own iterators, C/C++ Users Journal 19, 8 (2001) 51–57. K. [Czarnecki](http://gsd.uwaterloo.ca/kczarnec), U. W. Eisenecker, Generative Programming: Methods, Tools and Applications, [Addison-Wesley](http://www.pearsonhighered.com/), Reading, MA, 2000. G. [Dévai](http://deva.web.elte.hu/index.html), N. Pataki, Towards verified usage of the C++ Standard Template Library, Proc. 10th Symposium on [Programming Languages](http://ikportal.inf.elte.hu:8080/rendezvenyek/splst.html) and Software Tools (SPLST) 2007, Dobogókő, Hungary, pp. 360–371. G. [Dévai](http://deva.web.elte.hu/index.html), N. Pataki, A tool for formally specifying the C++ Standard Template Library, Ann. Univ. Sci. Budapest. [Comput.](http://ac.inf.elte.hu/Vol_031_2009/147.pdf) 31 (2009) 147–166. D. Gregor, J. [Järvi](http://parasol.tamu.edu/~jarvi/), J. [Siek](http://ecee.colorado.edu/~siek/), B. [Stroustrup](http://www2.research.att.com/~bs/), G. Dos Reis, A. [Lumsdaine](http://osl.iu.edu/~lums/), Concepts: linguistic support for generic programming in C++, Proc. 21st Annual [ACM SIGPLAN 2006](http://www.oopsla.org/2006/) Conference on Object-oriented Programming Systems, Languages, and Applications (OOPSLA 2006), Portland, Oregon, USA, pp. 291–310. D. Gregor, S. Schupp, Stllint: lifting static checking from languages to libraries, Software – Practice & Experience, 36, 3 (2006) 225-–254. J. [Järvi](http://parasol.tamu.edu/~jarvi/), D. Gregor, J. Willcock, A. [Lumsdaine](http://osl.iu.edu/~lums/), J. [Siek](http://ecee.colorado.edu/~siek/), Algorithm specialization in generic programming: challenges of constrained generics in C++, Proc. [ACM SIGPLAN 2006](http://research.microsoft.com/en-us/um/redmond/events/pldi06/) Conference on Programming Language Design and Implementation (PLDI 2006), Ottawa, Canada, pp. 272–282. Z. Juhász, M. Juhás, L. [Samuelis](http://kpi1.fei.tuke.sk/person/samuelis/dcicard.php), Cs. [Szabó](http://kpi1.fei.tuke.sk/person/szabo/dcicard.php), Teaching Java programming using case studies, [Teaching](http://tmcs.math.klte.hu/) Mathematics and Computer Science 6, 2 (2008) 245–256. T. [Kozsik](http://kto.web.elte.hu/), Tutorial on subtype marks, in Proc. Central European [Functional Programming School](http://plc.inf.elte.hu/cefp/cefp2005/) (CEFP 2005), Lecture Notes in [Comput. Sci.](http://www.springer.com/series/558) 4164 (2006) 191–222. S. [Meyers](http://www.aristeia.com/), Effective STL – 50 Specific Ways to Improve Your Use of the Standard Template Library, [Addison-Wesley](http://www.pearsonhighered.com/), Reading, MA, 2001. D. R. [Musser](http://www.cs.rpi.edu/~musser/), A. A. Stepanov, Generic programming, Proc. International Symposium ISSAC’88 on Symbolic and Algebraic Computation, Lecture Notes in [Comput. Sci.](http://www.springer.com/series/558) 358 (1989) 13–25. N. Pataki: C++ Standard Template Library by safe functors, Abstracts 8th Joint Conference on [Mathematics and Computer Science](http://www.selyeuni.sk/macs/) (MaCS’10), Komárno, Slovakia, July 14-17, 2010, p. 44. N. Pataki, Advanced functor framework for C++ Standard Template Library, Stud. Univ. Babeş-Bolyai [Inform.](http://www.cs.ubbcluj.ro/~studia-i/2011-1/09-Pataki.pdf) 56, 1 (2011) 99–113. N. Pataki, C++ Standard Template Library by ranges, Proc. 8th International Conference on [ Applied Informatics](http://icai.ektf.hu/) (ICAI 2010) Vol. 2., pp. 367–374. N. Pataki, Z. [Porkoláb](http://gsd.web.elte.hu), Z. [Istenes](http://quasar.inf.elte.hu), Towards soundness examination of the C++ Standard Template Library, Proc. Electronic [ Computers and Informatics](http://hornad.fei.tuke.sk/kpi/eci/) (ECI 2006), pp. 186–191. N. Pataki, Z. Szűgyi, G. [Dévai](http://deva.web.elte.hu/index.html), C++ Standard Template Library in a safer way, Proc. Workshop on [Generative Technologies 2010](http://wgt2010.elte.hu) (WGT 2010), Paphos, Cyprus, pp. 46–55. N. Pataki, Z. Szűgyi, G. [Dévai](http://deva.web.elte.hu/index.html), Measuring the overhead of C++ Standard Template Library safe variants, Electronic Notes in [Theoret. Comput. Sci.](http://entcs.org) 264, 5 (2011) 71–83. P. Pirkelbauer, S. Parent, M. [Marcus](http://www.emarcus.org/), B. [Stroustrup](http://www2.research.att.com/~bs/), Runtime concepts for the C++ Standard Template Library, Proc. ACM Symposium on [Applied Computing 2008](http://www.acm.org/conferences/sac/sac2008/), Fortaleza, Ceará, Brazil, pp. 171–177. Z. [Porkoláb](http://gsd.web.elte.hu/), Á. Sipos, N. Pataki, Inconsistencies of metrics in C++ Standard Template Library, Proc. 11th ECOOP Workshop on [Quantitative Approaches](http://www.inf.usi.ch/faculty/lanza/QAOOSE2007/) in Object-Oriented Software Engineering (QAOOSE), Berlin, Germany, pp. 2–6. B. [Stroustrup](http://www2.research.att.com/~bs/), The C++ Programming Language (Special edition), [Addison-Wesley](http://www.pearsonhighered.com/), Reading, MA, 2000. Z. Szűgyi, Á. Sipos, Z. [Porkoláb](http://gsd.web.elte.hu/), Towards the modularization of C++ concept maps, Proc. Workshop on [Generative Programming](http://wgt2008.elte.hu/) (WGT 2008), Budapest, Hungary, pp. 33–43. Z. Szűgyi, M. Török, N. Pataki, Towards a multicore C++ Standard Template Libary, Proc. Workshop on [Generative Programming](http://wgt.elte.hu/2011/) (WGT 2011), Saarbrücken, Germany, pp. 38–48. M. [Torgersen](http://www.daimi.au.dk/~madst/), The expression problem revisited – Four new solutions using generics, Proc. European Conference on [Object-Oriented Programming](http://www.ifi.uio.no/ecoop2004/) (ECOOP) 2004, Lecture Notes in [Comput. Sci.](http://www.springer.com/series/558) 3086 (2004) 123–143. L. Zolman: An STL message decryptor for visual C++, C/C++ Users Journal, 19(7) (2001) 24–30. I. Zólyomi, Z. [Porkoláb](http://gsd.web.elte.hu/): Towards a general template introspection library, Proc. of Generative Programming and Component Engineering: Third International Conference (GPCE 2004), Lecture Notes in [Comput. Sci.](http://www.springer.com/series/558) 3286, 266–282.	{ "pile_set_name": "ArXiv" }
--- author: - Qiang Lu bibliography: - 'comprehensive.bib' title: Safe and Efficient Intersection Control of Connected and Autonomous Intersection Traffic --- Introduction {#chapter:intro} ============ Literature Review {#chapter:lr} ================= DTOT-based Intersection Traffic Management {#chapter:dica} ========================================== Computational Complexity Improvements of DICA {#chapter:e-dica} ============================================= Reactive DICA: an Approach for Expedited Crossing of Emergency Vehicles {#chapter:r-dica} ======================================================================= Conclusion and Future Work {#chapter:conclusion} ==========================	{ "pile_set_name": "ArXiv" }
--- abstract: 'We present a protocol for quantum state transfer and remote state preparation across spin chains which operate in their anti-ferromagnetic mode. The proposed mechanism harnesses the inherent entanglement of the ground state of the strongly correlated many-body systems which naturally exists for free. The uniform Hamiltonian of the system does not need any engineering and, during the whole process, remains intact while a single qubit measurement followed by a single-qubit rotation are employed for both encoding and inducing dynamics in the system. This, in fact, has been inspired by recent progress in observing spin waves in optical lattice experiments, in which manipulation of the Hamiltonian is hard and instead local rotations and measurements have become viable. The attainable average fidelity stays above the classical threshold for chains up to length $~ 50$ and the system shows very good robustness against various sources of imperfection.' author: - Sima Pouyandeh - Farhad Shahbazi - Abolfazl Bayat title: Measurement induced dynamics for spin chain quantum communication and its application for optical lattices --- Introduction ============ Strongly correlated many-body systems often have highly entangled nontrivial ground states. The dynamics of such systems can be used for propagating information [@bose03] across distant sites and has been studied intensively in the last decade [@bose-review; @bayat-review-book]. Very recently, experimental realization of quantum state transfer through the natural dynamics of many-body systems have been achieved in NMR [@state-transfer-NMR] and coupled optical fibers in linear optics [@kwek-perfect-transfer]. Most of the proposals so far (see [@bose-review; @bayat-review-book] and the references therein), with very few exceptions like [@bayat-densecoding], are based on attaching an extra qubit, which encodes an “unknown" quantum state, to a chain of strongly interacting particles which is usually initialized to its ground state unless for certain engineered XX chains in which local end-chain operations makes it to work for any initialization [@DiFranco-initialization]. This mode of transmission does not seek to harness the intrinsic entanglement of many-body systems and the symmetries of the Hamiltonian seems to be more important [@bayat-xxz]. Moreover, attaching and detaching a single qubit to a many-body system is practically hard and needs a very fine control over the interaction of particles which is missing in many physical systems such as cold atoms. Although at the receiver site taking the quantum state for further process ultimately may need a swap operator or equivalently controlling some local interactions, for encoding the quantum state at the sender site not demanding such fine control will simplify the fabrication significantly. One can also think of sending a “known" quantum state from the sender to receiver. This occurs in a few occasions such as the remote quantum state preparation [@Bennet-remote] in which preparing the quantum state at some place is impossible due to practical issues, like inaccessibility of certain sites. Thus, the quantum state has to be prepared at one location and then transferred to the less accessible ones. There might be also several users for whom the quantum states are prepared in a single location (which needs simpler fabrication) and then distributed between them (see Refs. [@bayat-router; @Kay-perfect] for more details on information routers). In all these cases known quantum states have to be transferred from one place to another. It may be argued that by knowing the quantum state, the sender can simply send the Bloch vector $(n_x,n_y,n_z)$ of the qubit to the receiver via classical communication to prepare the state at the receiver site and there is no need for quantum communication. This possibility is indeed correct, however, the parameters of the Bloch vector are real numbers and sending them may need very long string of classical bits which may not be desired and has to be compensated by loosing precision in using a shorter set of classical bits. A single quantum state, however, can take all that information in a single shot. Hence, sending known quantum states, either considered as state transfer or remote state preparation, has its own merit while has hardly been studied for spin spin chain communication [@bayat-densecoding]. Quantum measurement is one of the mysteries of physics which has been hardly understood since the birth of quantum mechanics. According to quantum theory, measuring any observable results in a random output which is one of the eigenvalues of a Hermitian operator that is associated to that particular observable. The probability of such an outcome is determined by the overlap of the initial wave function and the corresponding eigenvector of the observable operator. In fact, after the measurement the wave function of the system goes under an abrupt change and collapses to that particular eigenstate of the observable operator. So far, the quantum measurement has been exploited for quantum communication via teleportation [@Bennet-teleportation] and measurement-based quantum computation [@Briegel-MBQC]. In conventional spin chain quantum communications, however, the random nature of measurement has been an obstacle for incorporating it in quantum state transfer protocols. On the other hand, since in quantum measurement the state of the system collapses instantaneously it can be used to induce dynamics in the system by changing its state and thus may be used as an alternative approach to attaching scenarios for quantum communication in strongly correlated systems. Cold atoms in an optical lattice are excellent test bed for many-body experiments. Both bosons [@Mott-insulator-boson] and fermions [@Mott-insulator-fermion] have been realized in the Mott insulator phase, where there is exactly one atom per site, and by properly controlling the intensity of laser beams one can tune the interaction between neutral atoms to behave as an effective spin Hamiltonian [@Lukin]. Local addressability of atoms with the resolution of single sites [@Bloch-single-site1; @Bloch-single-site2] has opened a totally new window for exploring many-body systems. Single site unitary operations and measurements [@Bloch-single-site2; @Meschede; @Bloch-spin-wave; @Bloch-magnon] are in fact becoming viable and accessible with high fidelities. Thanks to these new advancements, the correlated particle-hole pairs and string orders [@Bloch-particle-hole] together with their time evolution [@Bloch-correlation-time] have been explored experimentally. Furthermore, in recent experiments the propagation of a single impurity spin [@Bloch-spin-wave] and magnon bound states [@Bloch-magnon] in a ferromagnetic spin chain have been investigated. New cooling techniques [@Medley2010] have enabled, reaching for the first time, the temperatures required for observing quantum magnetic phases emerged due to spin interactions. In view of these, it is very timely to put forward new proposals which are doable with current achievements in cold atom experiments. In particular, one may think of new ways for quantum communication across a strongly correlated many-body interacting systems. In this paper, we introduce a mechanism for exploiting the inherent entanglement of many-body systems for quantum communication across a spin chain. The encoding of information is done through a single qubit measurement followed by the operation of a unitary gate which is determined by the random outcome of the measurement. The following measurement induced dynamics propagates the quantum state through the chain till it reaches the other side in which the information is captured by switching off the interaction couplings. The proposed protocol, which has been inspired by recent achievements for observing spin waves in ferromagnetic chains in optical lattices [@Bloch-spin-wave; @Bloch-magnon], exploits quantum measurement in order to induce quench dynamics in the system and can be seen as the first step for observing spin dynamics in anti-ferromagnetic chains. The simplicity of the protocol, with all its ingredients available in optical lattice experiments, allows for the experimentation of the proof of principles for measurement induced dynamics along an anti-ferromagnetic chain. Our protocol can also be interpreted as remote state preparation [@Bennet-remote] since a known quantum state is prepared on one side of the chain and then is transferred to the other side which might be inaccessible for some practical issues. In addition, our measurement induced transport can serve as information router in which the quantum state is prepared at one site of a network and then distributed between multiple users to reduce the complexity of fabrication. The structure of the paper is as following. In section \[sec2\] the model is introduced, in section \[sec3\] the unrestricted measurement induced dynamics is introduced, in section \[sec4\] the proposal for restricted measurement is discussed and in section \[sec5\] entanglement distribution is analyzed. Then in section \[sec6\] odd chains which do not have SU(2) symmetry are investigated and imperfections are studied in section \[sec7\]. In section \[sec8\] the application of our mechanism in optical lattices is explored. Finally in section \[sec9\] we summarize our results. ![ (Color online) (a) An arrays of interacting qubits for which the interaction type is Heisenberg with the exchange coupling $J$. A local control is available for the first qubit to operate a quantum gate or perform a spin measurement. (b) A Bell measurement on the first qubits of two noninteracting chains (note that labeling of the atoms are reversed in each chain) is used for entanglement distribution along the two spin chains. []{data-label="fig1"}](Fig1.eps){width="8cm" height="6cm"} Introducing the model {#sec2} ===================== We assume a system of $N$ spin-1/2 particles interacting via an anti-ferromagnetic Heisenberg Hamiltonian $$\label{H} H=\sum_{k=1}^{N-1} J_k \overrightarrow{\sigma}_k.\overrightarrow{\sigma}_{k+1}$$ where $\overrightarrow{\sigma_k}=(\sigma_k^x,\sigma_k^y,\sigma_k^z)$ is the vector of Pauli operators acting on site $k$ and $J_k$ is the exchange coupling which is assumed to be uniform (i.e. $J_k=J>0$ for all $k$’s) unless it is stated. A schematic picture of this system is shown in Fig. \[fig1\](a). System is cooled down to its ground state $\|GS\ra$. For the moment we consider even chains (i.e. even $N$) in which due to the SU(2) symmetry of the Hamiltonian the ground state is unique and lies in the subspace that half of the spins are up. Moreover, in even chains the SU(2) symmetry implies that the reduced density matrix of each spin is maximally mixed. This allows us to write the the ground state $\|GS\ra$ in a very generic form of $$\label{gs} \|GS\ra=\frac{\|\uparrow_k,\Downarrow\ra-\|\downarrow_k,\Uparrow\ra}{\sqrt{2}},$$ where $\uparrow_k$ ($\downarrow_k$) means site $k$ is in spin up (spin down) and $\Uparrow$ represents a quantum state for the rest of the system in which there are $N/2$ spins up and $N/2-1$ spins down (similarly for $\Downarrow$ there are $N/2$ spins down and $N/2-1$ spins up). The detailed structure of $\|\Uparrow\ra$ and $\|\Downarrow\ra$ are very complex and due to their different parities these two states are orthogonal. In addition due to the the SU(2) symmetry of the system the generic form of the ground state in Eq. (\[gs\]) remains valid for any basis of spins. By measuring a single spin at site $k$, in an arbitrary basis, the quantum state of the whole system collapses according to the outcome of the measurement. For instance, if the measurement is in the $\sigma^z$ basis on site $k$ then with probability of 1/2 the outcome of measurement is spin up and the quantum state of the system collapses to $\|\uparrow_k,\Downarrow\ra$. This new state still remains in the subspace of the ground state but is no longer an eigenvector of the Hamiltonian and as the results system evolves under the action of the Hamiltonian $H$. However, the outcome of the measurement is a random process and cannot be used directly for quantum communication across the spin chain. In the rest of the paper we try to exploit the random measurement induced dynamics for the purpose of quantum communication. Quantum State Transfer: Unrestricted Basis {#sec3} ========================================== In this section we assume that a general projecting measurement, in any arbitrary basis, is possible at the sender site, which is taken to be site 1. This measurement followed by a conditional unitary operation, which depends on the outcome of the measurement, are used to initialize our desired quantum state in the sender spin. The following unitary time evolution of the system transfers this quantum state to the receiver site. The most general pure quantum state can be written as $$\label{psi_p1} \|\psi^{(+1)}\ra=\cos{(\frac{\theta}{2})}\|\uparrow\ra+e^{i\phi}\sin{(\frac{\theta}{2})}\|\downarrow\ra$$ where $0\leq \theta \leq \pi$ and $0\leq \phi \leq 2 \pi$ are the two angles in the spherical coordinates which determine a single point on the surface of the Bloch sphere. This state is the eigenvector of the Hermitian operator $\overrightarrow{\sigma}.\overrightarrow{{\bf n}}$ (with eigenvalue $+1$) where the unit vector $\overrightarrow{{\bf n}}$ is defined as $\overrightarrow{{\bf n}}=(\sin(\theta)\cos(\phi),\sin(\theta)\sin(\phi),cos(\theta))$. The other eigenvector corresponding to the negative eigenvalue (with eigenvalue $-1$) is $$\label{psi_m1} \|\psi^{(-1)}\ra=\cos{(\frac{\theta}{2})}\|\downarrow\ra- e^{-i\phi} \sin{(\frac{\theta}{2})}\|\uparrow\ra.$$ one can transfer one of these eigenvectors to another by using the following unitary operator $$\label{R_unitary} R_u=\|\psi^{(+1)}\ra \la \psi^{(-1)}\|+\|\psi^{(-1)}\ra \la \psi^{(+1)}\|.$$ To initialize the quantum state $\|\psi^{(+1)}\ra$ in the sender site we measure the Hermitian operator $\overrightarrow{\sigma}.\overrightarrow{{\bf n}}$ at site 1. With probability of 1/2 the outcome is $+1$ and the initialization is done otherwise with probability of 1/2 the output is $-1$ and thus the unitary operator $R$ should act on site 1 to convert its state into $\|\psi^{(+1)}\ra$. As the result of this measurement the quantum state of the whole system changes accordingly. Depending on the outcome of the measurement the quantum state of the system initialized to one of the following states $$\begin{aligned} \label{initi1_0} \|\Psi^+(0)\ra &=& \sqrt{2} P^{(+1)} \|GS\ra \cr \|\Psi^-(0)\ra &=& \sqrt{2} R_u P^{(-1)} \|GS\ra\end{aligned}$$ where $P^{(\pm 1)}=\|\psi^{(\pm 1)}\ra \la\psi^{(\pm 1)}\|$ are the projecting operators and $\sqrt{2}$ is the normalization factor. Each of these states are obtained by probability of 1/2 and as it is clear the unitary operation $R_u$ acts only when the outcome of the measurement is $\|\psi^{(-1)}\ra$. Since neither of these states are the eigenvector of the Hamiltonian they evolve as $$\label{psi_t} \|\Psi^\pm (t)\ra=e^{-iHt} \|\Psi^\pm(0)\ra.$$ By tracing out all spins except the receiver, which is taken to be the last spin $N$, one can get the density matrix of received state $$\label{rho_N_t} \rho_N^{\pm}(t)=Tr_{\widehat{N}} \|\Psi^\pm (t)\ra \la \Psi^\pm (t)\|.$$ To quantify the quality of state transfer one can compute the fidelity as $$\label{Fid1_t} F_u^{\pm}(t)=\la \psi^{(+ 1)}\| \rho_N^{\pm}(t) \|\psi^{(+ 1)}\ra.$$ Thanks to the SU(2) symmetry of the system $F_u^{\pm}(t)$ is independent of $\theta$ and $\phi$ which means that all quantum states are transferred by the same fidelity. A general proof for this statement is given in Appendix A. ![(Color online) (a) The two fidelities $F_u^{+}(t)$ and $F_u^{-}(t)$ in terms of $Jt$ in a chain of length $N=20$ for the unrestricted measurement protocol. (b) The maximal fidelity $F_{max}^+$ as a function of $N$. (c) The maximal fidelity $F_{max}^-$ as a function of $N$. (d) The optimal time $Jt_{opt}$ versus length $N$.[]{data-label="fig2"}](Fig2.eps){width="9cm" height="7cm"} In Fig. \[fig2\](a) the fidelity $F_u^{+}(t)$ and $F_u^{-}(t)$ are both plotted as functions of time. As it is clear from the figures the fidelity starts evolving after a certain time that information reaches the last site. Then due to constructive quantum interferences at a particular time $t=t_{opt}$ the information reaches the receiver site and fidelity peaks for the first time. Though the later peaks might be larger it is physically unwise to wait for such long times as in practical cases the interaction with environment and its induced decoherence deteriorates the quality of transmission. So that we focus on the first peak at which the fidelity takes its maximal value, i.e. $F_{max}^{\pm}=F_u^{\pm}(t_{opt})$. In Figs. \[fig2\](b) and (c) the maximal fidelities $F_{max}^{+}$ and $F_{max}^{-}$ are plotted versus length $N$. As it is clear from these figures the fidelities are both high and go down almost linearly with very small slopes. A linear fit to data shows that $F_{max}^{+}=-0.007N+1.024$ and $F_{max}^{-}=-0.005N+1.016$. One can use these linear fits to extrapolate the fidelities in longer chains which shows that for chains up to $N \sim 50$ the fidelities are still above the classical threshold 2/3. This indeed shows the very high potential of this strategy for quantum state transfer across a many-body system. In Fig. \[fig2\](d) the optimal time $t_{opt}$ is plotted versus $N$ which also shows a linear dependence on $N$. Quantum State Transfer: Restricted Basis {#sec4} ======================================== Very often due to practical issues it is not possible to accomplish quantum measurement in any arbitrary basis on a single spin as needed in the encoding of the previous section. Instead quantum projecting measurement may be possible only for a certain basis, let say $\sigma_z$. The outcome of the measurement is thus either $\|\uparrow\ra$ or $\|\downarrow\ra$ and the quantum state of the whole system collapses to $\|\uparrow\Downarrow\ra$ or $\|\downarrow\Uparrow\ra$ respectively. To initialize the spin into a general superposition like Eq. (\[psi\_p1\]) a further unitary operation on first site is needed. Depending on the outcome of the measurement we apply one of the following unitary operators to the first spin $$\begin{aligned} \label{R1_R2} R_\uparrow &=& \|\psi^{(+1)}\ra \la \uparrow \| + \|\psi^{(-1)}\ra \la \downarrow \| \cr R_\downarrow &=& \|\psi^{(-1)}\ra \la \uparrow \| + \|\psi^{(+1)}\ra \la \downarrow \|\end{aligned}$$ where $R_\uparrow$ ($R_\downarrow$) is applied if the outcome of the measurement in the $\sigma_z$ basis is $\|\uparrow\ra$ ($\|\downarrow\ra$) to rotate it to $\|\psi^{(+1)}\ra$. The resulted states are not eigenstates of the Hamiltonian $H$ and thus system evolves accordingly. At any time $t$ one can see that the quantum state of the system is one of the following states depending on the measurement result $$\begin{aligned} \label{initi2_0} \|\Psi^\uparrow(t)\ra &=& e^{-iHt}R_\uparrow \otimes I \|\uparrow\Downarrow\ra, \cr \|\Psi^\downarrow(t)\ra &=& e^{-iHt}R_\downarrow \otimes I \|\downarrow\Uparrow\ra,\end{aligned}$$ As before we compute the density matrix of the last spin by tracing out the rest $$\label{rho_n_ud_t} \rho_N^{\alpha}(t)=Tr_{\widehat{N}} \|\Psi^\alpha (t)\ra \la \Psi^\alpha (t)\| \ \ \ \text{for $\alpha=\uparrow,\downarrow$}.$$ To quantify the quality of the state transfer we compute the fidelity as $$\label{fid_r_ud_t} F_r^\alpha(t)= \la \psi^{(+1)}\| \rho_N^{\alpha}(t) \|\psi^{(+1)}\ra.$$ Unlike the fidelity $F_u^\pm (t)$ for unrestricted measurement basis the $F_r^\alpha(t)$ depends on input parameters $\theta$ and $\phi$. To have an input independent quantity one may compute the average fidelity for all possible pure input states on the surface of the Bloch sphere $$\label{Fav_ud_t} F_{av}(t)= \frac{1}{4\pi} \int{F_r^\alpha(t) \sin(\theta) d\theta d\phi}.$$ Using a little bit of maths one can show that $$\begin{aligned} \label{Fav_t} F_{av}(t)&=& \frac{1}{6} \left\{ \la\downarrow\Downarrow\|e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\downarrow\Downarrow\ra + \la\uparrow\Downarrow\|e^{+iHt} \|\downarrow_N\ra \la\downarrow_N\| e^{-iHt} \|\uparrow\Downarrow\ra \right\} \cr &+& \frac{1}{3} \left\{ \la\uparrow\Downarrow\|e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Downarrow\ra + \la\downarrow\Downarrow\|e^{+iHt} \|\downarrow_N\ra \la\downarrow_N\| e^{-iHt} \|\downarrow\Downarrow\ra \right\} \cr &+& \frac{1}{3} abs\left\{ \la\downarrow\Downarrow\|e^{+iHt} \|\downarrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Downarrow \ra \right\}.\end{aligned}$$ where in the above formula it is assumed that the outcome of the measurement is spin up and to have the formula for the outcome spin down one has to only replace $\Downarrow$ with $\Uparrow$ in Eq. (\[Fav\_t\]). In fact, due to the symmetries of the system $F_{av}(t)$ is identical for both $\alpha=\uparrow,\downarrow$ and thus we drop the index $\alpha$. ![(Color online) (a) The average fidelity $F_{av}(t)$ as a function of $Jt$ for a chain of length $N=20$ in a restricted basis protocol. (b) The maximal average fidelity $F_{av}(t_{opt})$ in terms of length $N$.[]{data-label="fig3"}](Fig3.eps){width="9cm" height="5cm"} In Fig. \[fig3\](a) we plot $F_{av}(t)$ as a function of time. At $t=t_{opt}$ the average fidelity peaks for the first time. In Fig. \[fig3\](b) the maximum of average fidelity is depicted in terms of $N$ which can be well fitted by a linear function as $F_{av}(t_{opt})= -0.006 N+1.020$. This shows that for chains up to length $N\approx 60$ the average fidelity is above the classical threshold 2/3. \[table\_I\] $ N $ 4 6 8 10 12 14 16 18 20 ---------------------- -------- -------- -------- -------- -------- -------- -------- -------- -------- $F_{av}(projection)$ 0.9991 0.9867 0.9735 0.9604 0.9482 0.9368 0.9264 0.9171 0.9082 $F_{av}(attaching)$ 0.9554 0.9212 0.8986 0.8826 0.8693 0.8584 0.8496 0.8425 0.8365 For the sake of completeness we compare the attainable average fidelity of our proposed mechanism for the restricted basis with the widely studied attaching procedures, in which one extra qubit that carries our desired quantum state is attached to a spin chain initialized in its ground state just as Ref. [@bayat-xxz]. The results have been given in TABLE I and as it is clear from the data the projective mechanism gives higher fidelity in comparison to the attaching scenarios. The same sort of improvement is observed for the unrestricted projective measurement (not shown in the TABLE I). Entanglement Distribution {#sec5} ========================= The proposed measurement induced dynamics for state transfer can also be used for entanglement distribution. To fulfill such task we consider two independent chains which do not interact with each other as shown in Fig. \[fig1\](b). Initially both chains are prepared in their ground states and hence the quantum state of the system is $\|GS\ra_L \otimes \|GS\ra_R$. A Bell measurement is performed on the first spins of both chains which projects them on one of the following four possible maximally entangled Bell states $$\begin{aligned} \label{Bell_states} \|B_0\ra &=& \frac{\|\uparrow\downarrow\ra-\|\downarrow\uparrow\ra}{\sqrt{2}} \cr \|B_1\ra &=& \frac{\|\uparrow\downarrow\ra+\|\downarrow\uparrow\ra}{\sqrt{2}} \cr \|B_2\ra &=& \frac{\|\uparrow\uparrow\ra-\|\downarrow\downarrow\ra}{\sqrt{2}} \cr \|B_3\ra &=& \frac{\|\uparrow\uparrow\ra+\|\downarrow\downarrow\ra}{\sqrt{2}}\end{aligned}$$ Since the two chains do not interact any of these four possible outcomes will occur with the probability of 1/4. The symmetry of the system implies that the final entanglement is the same for all of them and thus we assume that the outcome of the measurement is the singlet $\|B_0\ra$. After measurement the first sites of the two chains get entangled and hence at any time $t$ the quantum state of the system can be written as $$\label{psi_ent_t} \|\psi(t)\ra=2 e^{-iH_Tt} P^{B_0}_{1_L,1_R} \|GS\ra_L \otimes \|GS\ra_R,$$ where $P^{B_0}_{1_L,1_R}=\|B_0\ra \la B_0\|$ projects the first sites of the two chains (i.e. spins at sites $1_L$ and $1_R$ as depicted in Fig. \[fig2\](b)) into a singlet state $\|B_0\ra$, the factor 2 at the beginning of the formula is for normalization and $H_T=H\otimes I+I\otimes H$ is the total Hamiltonian of the system. One can compute the reduced density matrix of the last two sites by tracing out all the rest. The special symmetries of the system and conservation of parity during the evolution implies that $$\label{psi_ent_t} \rho_{N_L,N_R}(t)=\frac{1}{2}\left( \begin{array}{cccc} a(t) & 0 & 0 & 0 \\ 0 & 1-a(t) & b(t)& 0 \\ 0 & b(t) & 1-a(t) & 0 \\ 0 & 0 & 0 & a(t) \\ \end{array} \right)$$ where both $a$ and $b$ are real numbers and can be written as $$\begin{aligned} \label{a_b_t} a(t) = \frac{1}{2} &\{& \la\uparrow\Downarrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Downarrow\ra \times \la\downarrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\downarrow\Uparrow\ra \cr &+& \la\uparrow\Downarrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\downarrow\Uparrow\ra \times \la\downarrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Downarrow\ra \cr &+& \la\downarrow\Downarrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\downarrow\Downarrow\ra \times \la\uparrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Uparrow\ra \cr &+& \la\uparrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Uparrow\ra \times \la\downarrow\Downarrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\downarrow\Downarrow\ra \cr &+& \la\downarrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Downarrow\ra \times \la\uparrow\Downarrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\downarrow\Uparrow\ra \cr &+& \la\downarrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\downarrow\Uparrow\ra \times \la\uparrow\Downarrow\| e^{+iHt} \|\uparrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Downarrow\ra \}\end{aligned}$$ $$\begin{aligned} \label{a_b_t} b(t) = \frac{-1}{2} &\{& \la\uparrow\Downarrow\| e^{+iHt} \|\downarrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Uparrow\ra \times \la\downarrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\downarrow_N\| e^{-iHt} \|\downarrow\Downarrow\ra \cr &+& \la\downarrow\Downarrow\| e^{+iHt} \|\downarrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Downarrow\ra \times \la\uparrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\downarrow_N\| e^{-iHt} \|\downarrow\Uparrow\ra \cr &+& \la\downarrow\Downarrow\| e^{+iHt} \|\downarrow_N\ra \la\uparrow_N\| e^{-iHt} \|\downarrow\Uparrow\ra \times \la\uparrow\Uparrow\| e^{+iHt} \|\uparrow_N\ra \la\downarrow_N\| e^{-iHt} \|\uparrow\Downarrow\ra \cr &+& \la\downarrow\Uparrow\| e^{+iHt} \|\downarrow_N\ra \la\uparrow_N\| e^{-iHt} \|\uparrow\Uparrow\ra \times \la\uparrow\Downarrow\| e^{+iHt} \|\uparrow_N\ra \la\downarrow_N\| e^{-iHt} \|\downarrow\Downarrow\ra \}\end{aligned}$$ ![(Color online) (a) Entanglement $E(t)$ as a function of $Jt$ for a chain of length $N=40$ (i.e. $N_L=N_R=20$). (b) The maximal entanglement $E_{max}$ versus length $N$.[]{data-label="fig4"}](Fig4.eps){width="9cm" height="4.5cm"} One can compute the entanglement, quantified by concurrence [@concurrence], between the two qubits which becomes $$\label{concurrence} E(t)=max\{0,b(t)-a(t)\}.$$ In Fig. \[fig4\](a) the entanglement $E(t)$ is plotted as a function of time. It is worth mentioning that as entanglement propagates in two disconnected chains the distance over which the entanglement is generated at $t=t_{opt}$ is double the distance of state transfer. In Fig. \[fig4\](b) the maximum attainable entanglement $E_{max}=E(t_{opt})$ is plotted versus distance $N$. As it is clear from the figure entanglement decays almost linearly by increasing $N$ with a small slope such that it reaches $E_{max}=0.49$ for a large distance of $N=40$. \[table\_1\] $N (even)$ 4 6 8 10 12 14 16 18 20 ---------------- -------- -------- -------- -------- -------- -------- -------- -------- -------- $F_{av}(even)$ 0.9991 0.9867 0.9735 0.9604 0.9482 0.9368 0.9264 0.9171 0.9082 $N (odd)$ 3 5 7 9 11 13 15 17 19 $F_{av}(odd)$ 0.9715 0.9526 0.9367 0.9236 0.9118 0.9013 0.8915 0.8834 0.8761 Odd Chains {#sec6} ========== So far we have only considered even chains for which the ground state is unique and supports the SU(2) symmetry with total excitation of zero. In contrast, the odd chains have doubly degenerate ground states $\|GS_\uparrow\ra$ and $\|GS_\downarrow\ra$ that each can be converted to another by applying $\prod_k \sigma_k^x$. In a chain of length $N$, the ground state $\|GS_\uparrow\ra$ ($\|GS_\downarrow\ra$) lies in the manifold of parity +1 (-1) in which $(N+1)/2$ number of spins are up (down) and the rest are down (up). In such states there is no SU(2) symmetry and one can split their degeneracy by applying a small magnetic field in the $z$ direction to choose one the ground states. Due to the absence of the SU(2) symmetry the fidelity of state transfer in both restricted and unrestricted basis depends on input parameter $\theta$. So, to quantify the quality of state transfer we consider a system of length $N$ initially prepared in one of its ground states, let say $\|GS_\uparrow\ra$. Then a restricted measurement in $\sigma^z$ basis is performed on the first spin of the chain which projects the first qubit on either spin $\uparrow$ or spin $\downarrow$. Depending on the outcome of the measurement a further application of $R_\uparrow$ or $R_\downarrow$ rotates the first spin into $\|\psi^{(+1)}\ra$ and initialization process is accomplished. A further time evolution of the system transfers this quantum states through out the chain. Just as before one can trace out the state of all spins but the last one and get the reduced density matrix of the last site $\rho_N(t)$ from which the fidelity is computed just as in Eq. (\[fid\_r\_ud\_t\]). To have an input independent quantity one can also average over all possible input states on the surface of the Bloch sphere just as the one in Eq. (\[Fav\_ud\_t\]) to get the average fidelity $F_{av}^{odd}(t)$. Just as before we consider the first peak of the average fidelity at the optimal time $t_{opt}$. In TABLE II we give a comparison for the average fidelity of even and odd chains versus length $N$ when the outcome of the measurement is spin up. By comparing the values one can realize that the quality of transfer is slightly lower for odd chains. For instant the average fidelity in the odd chain of length $N=19$ is $0.88$ while for a longer even chain of $N=20$ is $0.91$. This means that the SU(2) symmetry of the ground state in the even chains makes the quality of transfer even higher than the slightly shorter chains but with an odd length. ![image](Fig5.eps){width="15cm" height="5cm"} Imperfections {#sec7} ============= Preparing the system in its anti-ferromagnetic ground state needs cooling to zero temperature which in reality cannot be achieved. Hence, the initial state of the system is inevitably a thermal state of the form $\rho_{th}=\frac{e^{-\beta H}}{Z}$, where $\beta=1/K_{B}T$ in which $T$ is temperature, $K_{B}$ is the Boltzmann constant and $Z$ is the partition function. The transport mechanism is just the same as before. The projective measurement on the first qubit and the following unitary dynamics transfers information across chain just as the case that system has been initialized in its ground state. In fact, the assumption of a unitary evolution is valid only when the thermalization time is much longer than our optimal time $t_{opt}$. In Fig. \[fig5\](a), the maximal attainable fidelity $F_{max}$ is plotted in terms of $K_BT/J$ for a chain of length $N=10$. As SU(2) symmetry remains valid in the thermal initial state, the fidelity is independent of the basis of measurement. As it is evident from the figure, there is a plateau for $F_{max}$ at low temperatures which its width is determined by the finite size energy gap of the system. It is worth mentioning that the optimal time at which the fidelity peaks does not change with temperature which is consistent with the results of [@bayat-thermal]. In practical situations, it is impossible to isolate the system from its environment. To study such effects, we assume that the system is initialized in its ground state and the projective measurement is performed on the first spin just as before. However, we replace the unitary time evolution of the system with a Lindblad type master equation as $$\label{Lindblad} \dot{\rho}(t)= -i[H,\rho(t)]+\gamma\sum_{k=1}^{N}\sum_{\mu=1}^2(L^\mu_{k}\rho(t)L^{\mu \dagger}_{k} - \frac{1}{2} \{ L^\mu_{k}L^{\mu \dagger}_{k},\rho(t) \} )$$ where $L^1_{k}=\sigma^+_k$ and $L^2_{k}=\sigma^-_k$ are the Lindblad operators which add and subtract spin excitations into the system respectively and the coefficient $\gamma$ represents the coupling with the environment. By tracing out all spins but the last one can compute the fidelity which peaks at $t=t_{opt}$ no matter how strong is the coupling $\gamma$. In Fig. \[fig5\](b) we plot $F_{max}$ as a function of $\gamma$ for chain of length $N=10$ when the first spin is projected to $\|+\ra=(\|\uparrow\ra+\|\downarrow\ra)/\sqrt{2}$. As it is clear, the fidelity goes down by increasing $\gamma$ and stays above $0.75$ even for $\gamma\simeq0.1J$. Another imperfection is randomness in the coupling of the Hamiltonian as making a uniform chain might be very challenging in some physical realizations. This means that in the Hamiltonian of Eq. (\[H\]) we have $J_{k}=J(1+\delta_{k})$, where, $\delta_{k}$ is a dimensionless random number with a uniform distribution in the interval $[-\epsilon,\epsilon]$. In fact, $\epsilon$ determines the strength of randomness in the couplings. We fix the optimal time to be $t_{opt}$, determined from the uniform chain (i.e. $\epsilon=0$), as the real time at which fidelity peaks depends on all couplings $J_k$’s. We then average the fidelity $F(t_{opt})$ over several different realizations (we did for 100) of the system for a fixed $\epsilon$. In Fig. \[fig5\](c) we depict the fidelity $\la F(t_{opt}) \ra$ averaged over 100 different realization as a function of $\epsilon$ when the first qubit is projected into the state $\|+\ra$. It is seen that although the average fidelity decreases by increasing the randomness the mechanism shows a relatively high resistance against this destructive effect as fidelity remains above $0.85$ even for twenty percent of randomness (i.e. $\epsilon=0.2$). ![(Color online) (a) Cold atoms in an optical lattice prepared in a Mott insulator phase with exactly one atom per site realizes the Heisenberg Hamiltonian of Eq. (\[H\]). A local focused laser beam is used to manipulate the first qubit for both the gate operation and spin measurement. (b) The single qubit measurement is accomplished by a perpendicular focused laser beam which applies a strong radiation pressure to the state $\|\downarrow\ra$ and leaves the atom unaffected if its quantum state is $\|\uparrow\ra$. (c) Two parallel arrays in a two dimensional optical lattice are used for entanglement distribution in which a Bell measurement is needed. (d) Bell measurement is fulfilled by tilting the optical lattice such that the singlets tend to occupy a single site and triple pairs remain separated.[]{data-label="fig6"}](Fig6.eps){width="8cm" height="6cm"} Application for Optical Lattices {#sec8} ================================ The proposed mechanism is most suitable for realization in optical lattices in which an array of cold atoms in their Mott insulator phase sit in the minimums of a periodic potential, formed by counter propagating laser beams, as shown in Fig. \[fig6\](a). In the limit of high on-site energy the double occupancy is prohibited and the interaction between atoms is effectively explained by a spin Hamiltonian [@Lukin]. Changing the intensity of the laser beams tunes the tunneling rate of the atoms and thus controls the exchange coupling of the spin chain globally. In two or three dimensional lattices by tuning the intensity of the corresponding laser beams one can independently control the coupling of the atoms in each dimension globally. Recently, local addressability of the atoms have also been possible in optical lattices [@Bloch-single-site1; @Bloch-single-site2], makes local measurements and spin rotations, the two essential ingredients of our proposal accessible. Using such local operations the propagation of a single [@Bloch-spin-wave] and double [@Bloch-magnon] spin flips in a ferromagnetic chain have been experimentally observed. To perform spin measurement on a single site one can use the techniques developed in Ref. [@Meschede]. In that methodology an intense perpendicular leaser beam is focused to the target atom and couples one of the atomic levels which represents $\|\downarrow\ra$ to one of the excited states. This generates a strong radiation pressure which pushes the atom out of the lattice only when atom is in state $\|\downarrow\ra$ and does not affect it otherwise. This leaves the site empty if its atom is in state $\|\downarrow\ra$ and full if the atom is in state $\|\uparrow\ra$ as it is shown schematically in Fig. \[fig6\](b). So, the result of the measurement is revealed through a following fluorescent picture to see whether the atom is still sitting in its initial position (projecting to $\|\uparrow\ra$) or has gone (projecting to $\|\downarrow\ra$). Notice that in this technique by probability of 1/2, for which the atom is in the state $\|\downarrow\ra$ and thus leaves the lattice, the protocol fails which reduces the rate of communication by half. This means that if a two dimensional optical lattice is used to provide several equivalent parallel noninteracting spin chains (just as the one for ferromagnetic case in Refs. [@Bloch-spin-wave; @Bloch-magnon]) and the measurement is performed instantaneously on all the first qubits of parallel chains only half of them can be used to extract final information as there will be no hole in those chains and the rest should be discarded. Apart from single qubit measurement we also need to perform unitary operations (such as $R_\uparrow$ and $R_\downarrow$ in Eq. (\[R1\_R2\])) to accomplish the initialization and encoding information. To apply such unitary operators on the target atom (i.e. site $1$) a focused laser beam is exploited to generate Rabi oscillation between the qubit levels as shown in Fig. \[fig6\](a). This local operation is much quicker ($\sim 10\mu s$) [@Meschede] than the time evolution of the system ($\sim 1-10$ ms) [@Bloch-spin-wave; @Bloch-magnon] and can be considered as a sudden action. To have a pure local gate operation and avoid affecting the neighboring qubits one may apply a weak magnetic field gradient [@Meschede], which splits the hyperfine levels of all qubits position dependently, or use a tightly focused laser beam [@Bloch-single-site2] to only split the hyperfine levels of the target atom. So then a microwave pulse, tuned only for the target qubit, operates the gate locally as it has been realized in Refs. [@Bloch-single-site2; @Meschede]. For instance, a weak magnetic field gradient of $27.4 Gcm^{-1}$ is enough for applying $\sigma^x$ on a target qubit with a pulse of duration $10\mu s$ without affecting the neighboring sites [@Meschede]. According to the proposed mechanism for entanglement distribution a Bell measurement on the first qubits of the two chains is essential for initializing the system. We consider a geometry, shown in Fig. \[fig6\](c), in which two arrays of atoms sit in two parallel rows with the first atoms recite in the neighboring sites. To perform the Bell measurement we first raise the barriers between the atoms to switch off the interactions along the chains (i.e. $J=0$ in both spin chains). We use the fact that the energy levels for the singlet and triple pairs are different in a single well such that the singlet state is lower in energy. To operate the Bell measurement one has to tilt the lattice adiabatically such that the atoms in the right chain tunnel into the next row and sit along the left chain. Though, the atom in the first site of the right chain has to compensate an extra on site energy $U$ for its tunneling as its target site is already occupied by the first atom of the left chain. If the amount of tilting is tuned to be resonant only with the singlet state of two atoms in the doubly occupied site then the double occupancy occurs only for the singlet state as shown in Fig. \[fig6\](d). As the other Bell states are off resonant and energetically inaccessible, the double occupancy never occurs for such states. A further florescent picture of the system, which can be done without disturbing the internal states [@Gibbons-Nondestructive], will determine the number of atoms in the first site and reveals if the two atoms are in a singlet state or not. A backward adiabatic evolution (i.e. returning the lattice back to its normal) restores all the atoms into their initial position while the first spins are either projected to singlet $\|B_0\ra$ or one of the three other Bell states. If the output of the projecting measurement is singlet $\|B_0\ra$ (its probability is 1/4) then initialization is complete and by decreasing the horizontal barriers along the chains the propagation begins. On the other hand if the result is not $\|B_0\ra$ then the density matrix of the two qubits is an equal mixture of all other Bell states (its probability is 3/4). One then can apply $\sigma_z$ to the atoms in site $1_L$ (or $1_R$) in order to convert the $\|B_1\ra$ part of the mixture into $\|B_0\ra$ and repeat the adiabatic tilting to see if the projection to singlet is accomplished or not. This time the probability of success increases to 1/3. In the case of failure the state of the two atoms become a mixture of $\|B_2\ra$ and $\|B_3\ra$ which a local unitary operation $\sigma_y$ transforms these two states into $\|B_0\ra$ and $\|B_1\ra$ respectively. An extra repeating of the adiabatic tilting either directly gives a singlet state $\|B_0\ra$ (with the probability of 1/2) for the pair or project them into $\|B_1\ra$ (again with probability of 1/2) which then can be transformed to $\|B_0\ra$ locally. Hence, at the worst case the adiabatic tilting of the lattice has to be done three times for the initialization. Then by letting the system to evolve one can generate entanglement between the distant atoms at both sides of the system. Conclusion {#sec9} ========== In this paper we put forward a timely proposal for quantum communication in anti-ferromagnetic Heisenberg Hamiltonian using only local operations for encoding the information. This harnesses the intrinsic entanglement of the system for inducing dynamics via a single site quantum measurement. As the outcome of measurement is ultimately random a following unitary operation which is determined by the outcome of the measurement is essential for encoding the information within the intrinsically entangled ground state of the system. By finishing the encoding procedure system is left to evolve freely and after a certain time (set by the length $N$ and the strength of the exchange coupling $J$) information reaches the receiver site which can be taken for further computational process. The quality of state transfer remains above the threshold limit for chains up to length $N\sim 50$ while system is not engineered and no extra modulation is needed. In comparison with the widely studied attaching scenarios, our proposed mechanism not only introduces a different encoding of quantum states into a many-body system which harnesses the intrinsic entanglement of the ground state but also provides a new way for inducing quantum quench in such systems. From the perspective of quantum communication our measurement induced transport gives higher average fidelity and does not need local control over interaction at least on the sender site. One application of our proposal can be information router in which the quantum state is prepared at a particular site to simplify the fabrication and then is distributed among multiple users. In fact, an immediate generalization of our idea is to design an information router based on the proposed measurement induced transport which has to be pursued in a separate project. Alternatively, one may see our protocol as remote quantum state preparation [@Bennet-remote] in which a known quantum state is generated remotely at the output via the free evolution of a many-body strongly correlated system. In addition, we considered several imperfections which may arise in different realizations including thermal fluctuations, interaction with environment and the effect of random couplings. Since the encoding of information and performing the quantum quench in the system is done by only local operations the proposed mechanism is most suitable to be realized in optical lattices. The recent experiments for spin wave propagation [@Bloch-spin-wave] and transferring magnon bound states [@Bloch-magnon] show that all the ingredients we need is already available in the laboratory. Based on these new achievements, our proposal is just timely for being pursued in experiments and indeed can be realized with current technology. [Acknowledgements:-]{} Discussions with Sougato Bose, Leonardo Banchi and Bedoor Alkurtas are warmly acknowledged. AB thanks EPSRC grant $EP/K004077/1$. SU(2) symmetry of the Heisenberg Hamiltonian ============================================ The SU(2) group and its corresponding SU(2) Lie algebra are fully determined by the Pauli operators as the generators of the algebra. Any element of the SU(2) group in its $2\times 2$ representation can be written as $$\label{SU2_U} \mathcal{U}(\alpha,\widehat{n})=e^{i \alpha \overrightarrow{\sigma}.\widehat{n}}$$ where $\alpha$ is a real number and $\widehat{n}$ is a unit vector in the three dimensional space. The SU(2) symmetry of the Heisenberg Hamiltonian of Eq. (\[H\]) means $$\label{SU2_symmetry} \mathcal{U}(\alpha,\widehat{n})^{\otimes N} H \mathcal{U}^\dagger(\alpha,\widehat{n})^{\otimes N}=H.$$ Being a spin singlet, the ground state of the Hamiltonian for even $N$ is also invariant, up to an irrelevant global phase, under the action of $\mathcal{U}(\alpha,\widehat{n})^{\otimes N}$ such that $$\label{SU2_GS} \mathcal{U}(\alpha,\widehat{n})^{\otimes N}\|GS\ra=e^{i\beta}\|GS\ra$$ where $\beta$ is a global phase. One can show that the $R_\uparrow$ operator, defined in Eq. (\[R1\_R2\]), is an element of SU(2) group as $R_\uparrow=\mathcal{U}(\alpha^,\widehat{n}^)$ for a particular choice of $$\begin{aligned} \label{SU2_theta-n} \alpha^= \theta/2 \h \widehat{n}^= \left (\sin(\phi), \cos(\phi), 0 \right )\end{aligned}$$ where $\theta$ and $\phi$ are the qubit parameters in Eq. (\[psi\_p1\]). Following the Eqs. (\[SU2\_symmetry\]) and (\[SU2\_GS\]), this implies that $$\label{SU2_R_GS_H} R_\uparrow^{\dagger \otimes N} \|GS\ra=e^{i\beta}\|GS\ra, \h R_\uparrow^{\dagger \otimes N} H R_\uparrow^{\otimes N}=H.$$ We now have all the ingredients to prove that the fidelity of the unrestricted basis strategy is independent of the qubit parameters $\theta$ and $\phi$. Let’s assume that the projection is made in the basis of $\{ \|\uparrow\ra,\|\downarrow\ra \}$ (which corresponds to $\theta=0$ while $\phi$ is arbitrary) and the outcome of the measurement is $\|\uparrow\ra$ (namely the $+1$ solution). We show that the fidelity is in fact the same for all other values of $\theta$ and $\phi$ provided that the measurement outcome is $+1$. The time evolution of the system can be written as $$\label{SU2_psi_t} \|\Psi(t)\ra=\sqrt{2} e^{-iHt} P_1^{\uparrow} \|GS\ra$$ where $P_1^{\uparrow}=\|\uparrow_1\ra \la \uparrow_1\|$ is the projection on the first qubit. The fidelity then can be written as $$\begin{aligned} \label{SU2_Fp} \nonumber F^+_u (\theta=0, \phi)&=& \la \Psi(t)\| P_N^\uparrow \|\Psi(t) \ra \cr &=& 2 \la GS\| P_1^\uparrow e^{+iHt} P_N^\uparrow e^{-iHt} P_1^{\uparrow} \|GS \ra .\end{aligned}$$ Using the equalities in Eq. (\[SU2\_R\_GS\_H\]) one can insert $R_\uparrow^{\otimes N}$ and its Hermitian conjugate on both sides of the time evolution operators and apply it to the ground state $\|GS\ra$ without changing the fidelity. With a straight forward calculation one gets $$\begin{aligned} \label{SU2_Fp2} F^+_u (\theta=0, \phi)&=& 2 \la GS\| P_1^{(+1)} e^{+iHt} P_N^{(+1)} e^{-iHt} P_1^{(+1)} \|GS \ra \cr &=& F^+_u (\theta, \phi).\end{aligned}$$ where we have used the fact that $$\label{SU2_Proj_p_up} P_k^{(+1)}=R_\uparrow P_k^\uparrow R_\uparrow^\dagger, \h k=1,N.$$ By arriving to the Eq. (\[SU2\_Fp2\]) the proof is complete. S. Bose, Phys. Rev. Lett. [91]{}, 207901 (2003). S. Bose, Contemporary Physics [48]{}, 13 (2007). G. M. Nikolopoulos, Igor Jex, [Quantum State Transfer and Network Engineering]{}, Springer (2013). K. R. Koteswara Rao, T. S. Mahesh, A. Kumar, arXiv:1307.5220. A. Perez-Leija, R. Keil, A. Kay, H. Moya-Cessa, S. Nolte, L. C. Kwek, B. M. Rodríguez-Lara, A. Szameit, and D. N. Christodoulides, Phys. Rev. A [87]{}, 012309 (2013). S. Yang, A. Bayat, S. Bose, Phys. Rev. A [84]{}, 020302(R) (2011). C. Di Franco, M. Paternostro, M. S. Kim, Phys. Rev. Lett. [101]{}, 230502 (2008). A. Bayat and S. Bose, Phys. Rev. A [81]{}, 012304 (2010). C. H. Bennett, et al., Phys. Rev. Lett. [87]{}, 077902 (2001). A. Bayat, S. Bose, P. Sodano, Phys. Rev. Lett. [105]{}, 187204 (2010) P. J. Pemberton-Ross and A. Kay, Phys. Rev. Lett. [106]{}, 020503 (2011). C. H. Bennett, et al., Phys. Rev. Lett. [70]{}, 1895 (1993). H. J. Briegel, D. E. Browne, W. Dür, R. Raussendorf and M. Van den Nest, Nature Phys. 5, 19 (2009). W. S. Bakr, [et al.]{}, Nature [462]{}, 74 (2009); M. Greiner, [et al.]{}, Nature [415]{}, 39 (2002); W. S. Bakr, [et al.]{}, Science [329]{}, 547 (2010). R. Jördens, [et al.]{}, Nature [455]{}, 204 (2008); U. Schneider, [et al.]{}, Science [322]{}, 1520 (2008). L. Duan, E. Demler and M. D. Lukin, Phys. Rev. Let. [91]{}, 090402 (2003). J. F. Sherson, [et al.]{}, Nature [467]{}, 68 (2010). C. Weitenberg, [et al.]{}, Nature [471]{}, 319 (2011). M. Karski, [et al.]{}, New J. Phys. [12]{}, 065027 (2010). T. Fukuhara, [et al.]{}, Nature Phys. [9]{}, 235 (2013). T. Fukuhara, [et al.]{}, Nature [502]{}, 76 (2013). M. Endres, [et al.]{}, Science [334]{}, 200 (2011); M. Endres, [et al.]{}, arXiv:1303.5652. M. Cheneau, [et al.]{}, Nature [481]{}, 484 (2012). P. Medley, D. M. Weld, H. Miyake, D. E. Pritchard, W. Ketterle, Phys. Rev. Lett. [106]{}, 195301 (2011). W. K. Wootters, Phys. Rev. Lett. [80]{}, 2245 (1998). A. Bayat, V. Karimipour, Phys. Rev. A, [71]{}, 042330 (2005). M. J. Gibbons, C. D. Hamley, C. Y. Shih and M. S. Chapman, Phys. Rev. Lett. [106]{}, 133002 (2011).	{ "pile_set_name": "ArXiv" }
--- address: 'Université Côte d’Azur, CNRS, LJAD, France' author: - 'Julie <span style="font-variant:small-caps;">Déserti</span>' bibliography: - 'biblio.bib' nocite: '\nocite{}' title: 'The <span style="font-variant:small-caps;">Cremona</span> group and its subgroups' --- [^1] À Benoît The main purpose of the present treatise is to draw a portrait of the $n$-dimensional <span style="font-variant:small-caps;">Cremona</span> group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. The study of this group started in the XIXth century; the subject has known a lot of developments since the beginning of the XXIth century. Old and new results are discussed; unfortunately we will not be exhaustive. The <span style="font-variant:small-caps;">Cremona</span> group is approached through the study of its subgroups: algebraic, finite, normal, nilpotent, simple, torsion subgroups are evoked but also centralizers of elements, representation of lattices, subgroups of automorphisms of positive entropy etc A chapter is devoted to generators and relations and an other one to some algebraic properties of the <span style="font-variant:small-caps;">Cremona</span>. We of course recall the isometric action of the <span style="font-variant:small-caps;">Cremona</span> group on an infinite dimensional hyperbolic space which which plays a considerable role in the study of subgroups of the <span style="font-variant:small-caps;">Cremona</span> group but we voluntarily keep proofs that do not use this action. In this short introduction we will just introduce birational self maps of the plane and the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group from a geometrical point of view; any chapter has its own introduction. A plane collineation is a transformation in which to any point $p$ of the plane there is made to correspond a point $q$ whose coordinates are proportional to assigned linear homogeneous functions of those of $p$. Such maps leave the projective properties of curves unaltered. In advancing beyond such properties let us introduce other maps of the plane into itself that establish relations between curves of differents orders and possessing different sets of singularities. The most general rational map of the plane is defined by equations of the form $$\phi\colon(z_0:z_1:z_2)\dashrightarrow\big(\phi_0(z_0,z_1,z_2):\phi_1(z_0,z_1,z_2):\phi_2(z_0,z_1,z_2)\big)$$ where $\phi_0$, $\phi_1$ and $\phi_2$ are homogeneous polynomials of degree $n$ without common factor of positive degree. Such a map makes correspond to a point $p$ with coordinates $(p_0:p_1:p_2)$ a point $\phi(p)=q$ with coordinates $(q_0:q_1:q_2)$ where $$\label{eq:crem} \delta q_0=\phi_0(p_0,p_1,p_2),\quad \delta q_1=\phi_1(p_0,p_1,p_2),\quad \delta q_2=\phi_2(p_0,p_1,p_2).$$ Consider the net of curves $\Lambda_\phi$ defined by the equation $$\alpha\phi_0+\beta\phi_1+\gamma\phi_2=0$$ where $\alpha$, $\beta$ and $\gamma$ are arbitrary parameters. As $p$ describes any given curve $\mathcal{C}$ of $\Lambda_\phi$ the corresponding point $q$ evidently describes a line. The curves of the net $\Lambda_\phi$ are thus correlated by $\phi$ with the lines of the plane. Conversely given any net $\Lambda$ of curves such as $\Lambda_\phi$ a linear representation of the curves of $\Lambda$ on the lines of the plane is equivalent to a rational map of the plane. The curves of $\Lambda_\phi$ may have base-points $p_i$ common to them all. Each such point is a common zero of $\phi_0$, $\phi_1$ and $\phi_2$ so the equations (\[eq:crem\]) to determine its corresponding point are illusory. Conversely each point, termed a base-point of $\phi$, which renders equation (\[eq:crem\]) illusory is a base-point of $\Lambda_\phi$. In other words The base-points of any rational map are the base-points of the associated net of curves. Any two general curves $\mathcal{C}$ and $\mathcal{C}'$ of $\Lambda_\phi$ defined a pencil of curves $\mathcal{C}+\alpha\mathcal{C}'$ of the net. Denote by $n$ the number of free intersections of $\mathcal{C}$ and $\mathcal{C}'$ not occuring at the base-points $p_i$; denote by $r_1$, $r_2$, $\ldots$, $r_n$ these points. The integer $n$ is called the grade of $\Lambda_\phi$. To curves of the arbitrary pencil $\mathcal{C}+\alpha\mathcal{C}'$ there correspond by the map $\phi$ lines of a pencil $L+\alpha L'$. Furthermore if the vertex of the latter pencil is $q$, then clearly every point $r_i$ corresponds to $q$. Conversely if any two points of the plane have the same transform $q$, then they belong to the same free intersection set of some pencil in $\Lambda_\phi$. Let $\phi$ be a rational self map of the plane. Let $\Lambda_\phi$ be its associated net and let $n$ be the grade of $\Lambda_\phi$. An arbitrary point $q$ is the transform of $n$ points $r_1$, $r_2$, $\ldots$, $r_n$ which together form the free intersection set of a pencil of curves of $\Lambda_\phi$. In other words the general rational map of the plane is a $(n,1)$ correspondence between the points $p$ and $q$. And this means that, when the ratios of $q_0$, $q_1$, $q_2$ are given the equations (\[eq:crem\]) have in general $n$ distinct solutions for the ratios of $p_0$, $p_1$ and $p_2$. If $n=1$, [i.e.]{} if these equations have only one solution, $(p_0:p_1:p_2)$ are rational functions of $(q_0:q_1:q_2)$. In this case the equations of the reverse map will be of the form $$\begin{aligned} & \alpha p_0=\psi_0(q_0,q_1,q_2)&& \alpha p_1=\psi_1(q_0,q_1,q_2)&& \alpha p_2=\psi_2(q_0,q_1,q_2)\end{aligned}$$ where $\psi_0$, $\psi_1$ and $\psi_2$ are homogeneous polynomials of degree $n'$. A <span style="font-variant:small-caps;">Cremona</span> map is a rational map whose reverse is also rational, we also speak about birational self map of the plane. The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is the group of birational self maps of the plane. A homaloidal net of curves in the plane is one whose grade is $1$. Equations (\[eq:crem\]) define a birational map $\phi$ if and only if the associated net $\Lambda_\phi$ is homaloidal. Conversely from any given homaloidal net we can derive many birational self maps of the plane; if $\overline{\phi_0}$, $\overline{\phi_1}$ and $\overline{\phi_2}$ are three independent linear combinations of $\phi_0$, $\phi_1$ and $\phi_2$, the net $$\alpha\,\phi_0+\beta\,\phi_1+\gamma\,\phi_2=0$$ can also be expressed in the form $$\alpha'\,\overline{\phi_0}+\beta'\,\overline{\phi_1}+\gamma'\,\overline{\phi_2}=0$$ and the map defined by $$(z_0:z_1:z_2)\dashrightarrow\big(\overline{\phi_0}(z_0,z_1,z_2):\overline{\phi_1}(z_0,z_1,z_2):\overline{\phi_2}(z_0,z_1,z_2)\big)$$ is based on the same net. Moreover To any birational self map of the plane there is associated a homaloidal net of curves. Conversely any homaloidal net of curves generates an infinity of birational self maps of the plane, any of which is the product of any other by a plane collineation. A collineation is the simplest kind of birational self map of the plane whose homaloidal net is composed of the lines of the plane. The degree of a birational self map of the plane is the degree of the curves of its generating homaloidal net. Let $\phi$ be a birational self map of the plane of degree $n$. Denote by $n'$ the degree of its inverse $\phi^{-1}$. If the number of intersections of two curves $\mathcal{C}$ and $\mathcal{C}'$ is denoted by $\mathcal{C}\cdot\mathcal{C}'$ and if $L$ and $L'$ are lines, then $$n=L\cdot\Lambda_\phi=\phi(L)\cdot\phi(\Lambda_\phi)=\Lambda_{\phi^{-1}}\cdot L'=n'.$$ Hence A birational self map of the plane and its inverse have the same degree. Let us finish this introduction by pointing out that this statement is not true in higher dimension: $$\begin{aligned} &\mathbb{P}^3_\mathbb{C}\dashrightarrow\mathbb{P}^3_\mathbb{C} && (z_0:z_1:z_2:z_3)\dashrightarrow(z_0^2:z_0z_1:z_1z_2:z_0z_3-z_1^2)\end{aligned}$$ is a birational self map of $\mathbb{P}^3_\mathbb{C}$ of degree $2$ whose inverse $$\begin{aligned} &\mathbb{P}^3_\mathbb{C}\dashrightarrow\mathbb{P}^3_\mathbb{C} && (z_0:z_1:z_2:z_3)\dashrightarrow \big(z_0^2z_1:z_0z_1^2:z_0^2z_2:z_1(z_0z_3+z_1^2)\big)\end{aligned}$$ has degree $3$. As we will see there are many other differences between the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group and the $n$-dimensional <span style="font-variant:small-caps;">Cremona</span> group, $n\geq 3$. Note that the study of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is central: if $S$ is a complex rational surface, then its group of birational self maps is isomorphic to $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Introduction {#chapter:intro} ============ This chapter is devoted to recalls and first definitions. In the first section morphisms between varieties, blow-ups, <span style="font-variant:small-caps;">Cremona</span> group and bubble space are for instance introduced. <span style="font-variant:small-caps;">Zariski</span> theorem, base-points, indeterminacy points are recalled. Examples of subgroups of the <span style="font-variant:small-caps;">Cremona</span> group are given: among them the group of automorphisms of $\mathbb{P}^n_\mathbb{C}$, the <span style="font-variant:small-caps;">Jonquières</span> group, the group of monomial maps. The second section is devoted to divisors (prime divisors, <span style="font-variant:small-caps;">Weil</span> divisors, principal divisors, <span style="font-variant:small-caps;">Picard</span> group) and intersection theory. The third section deals with a geometric definition of birational maps of the complex projective plane. First definitions and examples {#sec:firstdef} ------------------------------ Denote by $\mathbb{P}^n_\mathbb{C}$ the complex projective space of dimension $n$. A rational map $$\phi\colon V_1\subset\mathbb{P}^n_\mathbb{C}\dashrightarrow V_2\subset\mathbb{P}^k_\mathbb{C}$$ between two smooth projective complex varieties $V_1$ and $V_2$ is a regular map on a non-empty <span style="font-variant:small-caps;">Zariski</span> open subset of $V_1$ such that the image of the points where $\phi$ is well defined is contained in $V_2$. If $\phi$ is well defined on $V_1$ we say that $\phi$ is a morphism or a regular map, otherwise we denote by $\mathrm{Ind}(\phi)$ the set where $\phi$ is not defined and call it the indeterminacy set of $\phi$. A birational map between $V_1$ and $V_2$ is a rational map that admits an inverse which is rational. In other words it is an isomorphism between two non-empty <span style="font-variant:small-caps;">Zariski</span> open subsets of $V_1$ and $V_2$. Let us give an example of a birational morphism. Let $p$ be a point on a smooth algebraic surface $S$. We say that $\pi\colon Y\to S$ is a blow-up of $p$ if - $Y$ is a smooth surface, - $\pi_{\vert Y\smallsetminus\{\pi^{-1}(p)\}}\colon Y\smallsetminus\{\pi^{-1}(p)\}\to S\smallsetminus \{p\}$ is an isomorphism, - and $\pi^{-1}(p)\simeq\mathbb{P}^1_\mathbb{C}$. We call $\pi^{-1}(p)$ the exceptional divisor. If $\pi\colon Y\to S$ and $\pi'\colon Y'\to S$ are two blow-ups of the same point $p$, then there exists an isomorphism $\varphi\colon Y\to Y'$ such that $\pi=\pi'\circ\varphi$. We can thus speak about the blow-up of $p\in S$. Let us describe the blow-up of $(0:0:1)$ in $\mathbb{P}^2_\mathbb{C}$ endowed with the homogeneous coordinates $(z_0:z_1:z_2)$. Consider the affine chart $z_2=1$, [i.e.]{} let us work in $\mathbb{C}^2$ with coordinates $(z_0,z_1)$. Set $$V=\big\{\big((z_0,z_1),(u:v)\big)\in\mathbb{C}^2\times\mathbb{P}^1_\mathbb{C}\,\vert\, z_0v=z_1u\big\}.$$ Let $\pi\colon V\to\mathbb{C}^2$ be the morphism given by the first projection. Then - $\pi^{-1}(0,0)=\big\{\big((0,0),(u:v)\big)\,\vert\,(u:v)\in\mathbb{P}^1_\mathbb{C}\big\}$ so $\pi^{-1}(0,0)\simeq\mathbb{P}^1_\mathbb{C}$; - if $p=(z_0,z_1)$ is a point of $\mathbb{C}^2\smallsetminus\{(0,0)\}$, then $$\pi^{-1}(p)=\big\{((z_0,z_1),(z_0:z_1))\big\}\in V\smallsetminus\{\pi^{-1}(0,0)\}$$ and $\pi_{\vert V\smallsetminus\{\pi^{-1}(0,0)\}}$ is an isomorphism, the inverse being $$(z_0,z_1)\mapsto\big((z_0,z_1),(z_0:z_1)\big).$$ In other words $V=\mathrm{Bl}_{(0,0)}\mathbb{P}^2_\mathbb{C}$ is the surface obtained by blowing up the complex projective plane at $(0:0:1)$, $\pi$ is the blow up of $(0:0:1)$ and $\pi^{-1}(0,0)$ is the exceptional divisor. Let $V$ be a complex algebraic variety, and let $\mathrm{Bir}(V)$ be the group of birational maps of $V$. The group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is called the <span style="font-variant:small-caps;">Cremona</span> group. If we fix homogeneous coordinates $(z_0:z_1:\ldots:z_n)$ of $\mathbb{P}^n_\mathbb{C}$ every element $\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ can be described by homogeneous polynomials of the same degree $\phi_0$, $\phi_1$, $\ldots$, $\phi_n\in\mathbb{C}[z_0,z_1,\ldots,z_n]$ without common factor of positive degree: $$\phi\colon(z_0:z_1:\ldots:z_n)\dashrightarrow\big(\phi_0(z_0,z_1,z_2,\ldots,z_n):\phi_1(z_0,z_1,z_2,\ldots,z_n):\ldots:\phi_n(z_0,z_1,z_2,\ldots,z_n)\big).$$ The degree of $\phi$ is the degree of the $\phi_i$’s. In the affine chart $z_0=1$, the map $\phi$ is given by $(\varphi_1,\varphi_2,\ldots,\varphi_n)$ where for any $1\leq i\leq n$ $$\varphi_i=\frac{\phi_i(1,z_1,z_2,\ldots,z_n)}{\phi_0(1,z_1,z_2,\ldots,z_n)} \in\mathbb{C}(z_1,z_2,\ldots,z_n).$$ The subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ consisting of elements $\phi$ such that all the $\varphi_i$ are polynomials as well as the entries of $\phi^{-1}$ is exactly the group $\mathrm{Aut}(\mathbb{A}^n_\mathbb{C})$ of polynomial automorphisms of the affine space $\mathbb{A}^n_\mathbb{C}$ . Let $S$ be a smooth projective surface. The bubble space $\mathcal{B}(S)$ is, roughly speaking, the set of all points that belong to $S$, or are infinitely near to $S$. Let us be more precise: consider all surfaces $Y$ above $S$, [i.e.]{} all birational morphisms $\pi\colon Y\to S$ ; we identify $p_1\in Y_1$ and $p_2\in Y_2$ if $\pi_1^{-1}\circ\pi_2$ is a local isomorphism in a neighborhood of $p_2$ that maps $p_2$ onto $p_1$. The bubble space $\mathcal{B}(S)$ is the union of all points of all surfaces above $S$ modulo the equivalence relation generated by these identifications. A point $p\in\mathcal{B}(S)\cap S$ is a proper point. All points in $\mathcal{B}(S)$ that are not proper are called infinitely near. Let $S$ and $S'$ be two smooth projective surfaces. Let $\phi\colon S\dashrightarrow S'$ be a birational map. By <span style="font-variant:small-caps;">Zariski</span>’s theorem (see for instance [@Beauville:book]) we can write $\phi=\pi_2\circ\pi_1^{-1}$ where $\pi_1\colon Y\to S$ and $\pi_2\colon Y\to S'$ are finite sequences of blow-ups. We may assume that there is no $(-1)$-curve in $Y$, [i.e.]{} no $(-1)$-curve in $Y$ contracted by both $\pi_1$ and $\pi_2$. We then say that $\pi_2\circ\pi_1^{-1}$ is a minimal resolution of $\phi$. The base-points $\mathrm{Base}(\phi)$ of $\phi$ are the points blown up by $\pi_1$. The proper base-points of $\phi$ are called indeterminacy points of $\phi$. A birational morphism $\pi\colon S\to S'$ induces a bijection $\pi_\bullet\colon\mathcal{B}(S)\to\mathcal{B}(S')\smallsetminus\mathrm{Base}(\pi^{-1})$. A birational map of smooth projective surfaces $\phi\colon S\dashrightarrow S'$ induces a bijection $$\phi_\bullet\colon\mathcal{B}(S)\smallsetminus\mathrm{Base}(\phi)\to\mathcal{B}(S')\smallsetminus\mathrm{Base}(\phi^{-1})$$ by $\phi_\bullet=(\pi_2)_\bullet\circ(\pi_1)_\bullet^{-1}$ where $\pi_2\circ\pi_1^{-1}$ is a minimal resolution of $\phi$. Let us now give some subgroups of the <span style="font-variant:small-caps;">Cremona</span> group: - First consider the automorphism group of $\mathbb{P}^n_\mathbb{C}$. It is the subgroup formed by regular maps, [i.e.]{} maps well defined on $\mathbb{P}^n_\mathbb{C}$ and whose inverse is also well defined on $\mathbb{P}^n_\mathbb{C}$: $$\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})=\big\{\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\,\vert\,\mathrm{Base}(\phi)=\mathrm{Base}(\phi^{-1})=\emptyset\big\}.$$ To any $M=\big(a_{i,j}\big)_{0\leq i,\,j\leq n}\in\mathrm{PGL}(n+1,\mathbb{C})$ corresponds an element of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ of degree $1$: $$(z_0:z_1:\ldots:z_n)\mapsto\left(\displaystyle\sum_{j=0}^na_{0,j}z_j:\displaystyle\sum_{j=0}^na_{1,j}z_j:\ldots:\displaystyle\sum_{j=0}^na_{n,j}z_j\right)$$ and vice-versa. Such elements are biregular. Furthermore <span style="font-variant:small-caps;">Bezout</span> theorem implies that all biregular maps are linear. We thus have the following isomorphism $$\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})\simeq\mathrm{PGL}(n+1,\mathbb{C}).$$ - The $n$-dimensional subgroup of $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ consisting of diagonal automorphisms is denoted by $\mathrm{D}_n$. Note that $\mathrm{D}_n$ is the torus of highest rank of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$; indeed according to [@BialynickiBirula] if $\mathrm{G}$ is an algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ isomorphic to $(\mathbb{C}^)^k$, then $k\leq n$ and if $k=n$, then $\mathrm{G}$ is conjugate to $\mathrm{D}_n$. - Start with the surface $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ considered as a smooth quadric in $\mathbb{P}^3_\mathbb{C}$; its automorphism group contains $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$. By the stereographic projection the quadric is birationally equivalent to the plane, so that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ contains also a copy of $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$. If $\mathrm{G}$ is a semi-simple algebraic group, $\mathrm{H}$ is a parabolic subgroup of $\mathrm{G}$ and $V=\faktor{\mathrm{G}}{\mathrm{H}}$ is a homogeneous variety of dimension $n$, then $V$ is rational. Once a birational map $\pi\colon V\dashrightarrow\mathbb{P}^n_\mathbb{C}$ is given, $\pi\circ\mathrm{G}\circ\pi^{-1}$ determines an algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. - A fibration* of a surface $S$ is a rational map $\pi\colon S\dashrightarrow C$, where $C$ is a curve, such that the general fibers are one-dimensional. Two fibrations $\pi_1\colon S\dashrightarrow C$ and $\pi_2\colon S\dashrightarrow C'$ are identified if there exists an open dense subset $\mathcal{U}\subset S$ that is contained in the domains of $\pi_1$ and $\pi_2$ such that $\pi_{1\vert\mathcal{U}}$ and $\pi_{2\vert\mathcal{U}}$ define the same set of fibers. We say that a group $\mathrm{G}$ preserves a fibration $\pi$ if $\mathrm{G}$ permutes the fibers. A rational fibration of a rational surface $S$ is a rational map $\pi\colon S\dashrightarrow\mathbb{P}^1_\mathbb{C}$ such that the general fiber is rational. The following statement due to <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Enriques</span> says that, up to birational maps, there exists only one rational fibration of $\mathbb{P}^2_\mathbb{C}$: \[thm:fibration\] Let $S$ be a surface. Let $\pi\colon S\dashrightarrow C$ be a rational fibration. Then there exists a birational map $\phi\colon C\times\mathbb{P}^1_\mathbb{C}\dashrightarrow S$ such that $\pi\circ\phi$ is the projection onto the first factor. The <span style="font-variant:small-caps;">Jonquières</span> subgroup $\mathcal{J}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is the subgroup of elements that preserve the pencil of lines through the point $(0:0:1)\in\mathbb{P}^2_\mathbb{C}$. Any subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that preserves a rational fibration is conjugate to a subgroup of $\mathcal{J}$ (Theorem \[thm:fibration\]). With respect to affine coordinates $(z_0:z_1:1)$ an element of $\mathcal{J}$ is of the form $$(z_0,z_1)\dashrightarrow\left(\frac{\alpha z_0+\beta}{\gamma z_0+\delta},\frac{A(z_0)z_1+B(z_0)}{C(z_0)z_1+D(z_0)}\right)$$ where $\left(\begin{array}{cc}\alpha & \beta \\ \gamma & \delta\end{array}\right)$ belongs to $\mathrm{PGL}(2,\mathbb{C})$ and $\left(\begin{array}{cc} A & B\\ C & D\end{array}\right)$ to $\mathrm{PGL}(2,\mathbb{C}(z_0))$. This induces an isomorphism $$\mathcal{J}\simeq\mathrm{PGL}(2,\mathbb{C})\rtimes\mathrm{PGL}(2,\mathbb{C}(z_0)).$$ - Let $M=(a_{i,j})_{1\leq i,\,j\leq n}\in M(n,\mathbb{Z})$ be a $n\times n$ matrix of integers. The matrix $M$ determines a rational self map of $\mathbb{P}^n_\mathbb{C}$ given in the affine chart $z_0=1$ by $$\phi_M\colon (z_1,\ldots,z_n)\mapsto\Big(z_1^{a_{1,1}}z_2^{a_{1,2}}\ldots z_n^{a_{1,n}},z_1^{a_{2,1}}z_2^{a_{2,2}}\ldots z_n^{a_{2,n}},\ldots,z_1^{a_{n,1}}z_2^{a_{n,2}}\ldots z_n^{a_{n,n}}\Big).$$ The map $\phi_M$ is birational if and only if $M$ belongs to $\mathrm{GL}(n,\mathbb{Z})$. This yields an injective homomorphism $\mathrm{GL}(n,\mathbb{Z})\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ whose image is called <span style="font-variant:small-caps;">Weyl</span> group and denoted by $\mathrm{W}_n$. The group $\mathrm{W}_n$ is also called the group of monomial maps. - The well known result of <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> states that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is generated by the involution $$\sigma_2\colon(z_0:z_1:z_2)\dashrightarrow(z_1z_2:z_0z_2:z_0z_1)$$ and the group $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})=\mathrm{PGL}(3,\mathbb{C})$ (see for instance [@AlberichCarraminana]). For $n\geq 3$ the <span style="font-variant:small-caps;">Cremona</span> group is not generated by $\mathrm{PGL}(n+1,\mathbb{C})$ and $\mathrm{W}_n$ (see [@Hudson; @Pan:generation]). In other words the subgroup $\langle\mathrm{PGL}(n+1,\mathbb{C}),\,\mathrm{W}_n\rangle$ is a strict subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. The finite index subgroup of $\langle\mathrm{PGL}(n+1,\mathbb{C}),\,\mathrm{W}_n\rangle$ generated by $\mathrm{PGL}(n+1,\mathbb{C})$ and the involution $$\sigma_n\colon(z_0:z_1:\ldots:z_n)\dashrightarrow\left(\prod_{\stackrel{i=0}{i\not=0}}^nz_i:\prod_{\stackrel{i=0}{i\not=1}}^nz_i:\ldots:\prod_{\stackrel{i=0}{i\not=n}}^nz_i\right)$$ has been studied in [@BlancHeden; @Deserti:reg]. The group $\mathrm{G}(n,\mathbb{C})=\langle\sigma_n,\,\mathrm{PGL}(n+1,\mathbb{C})\rangle$ “looks like” $\mathrm{G}(2,\mathbb{C})=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ in the following sense ([@Deserti:reg]): - there is no non-trivial finite dimensional linear representation of $\mathrm{G}(n,\mathbb{C})$ over any field; - the group $\mathrm{G}(n,\mathbb{C})$ is perfect, [i.e.]{} $\big[\mathrm{G}(n,\mathbb{C}),\mathrm{G}(n,\mathbb{C})\big]=\mathrm{G}(n,\mathbb{C})$; - the group $\mathrm{G}(n,\mathbb{C})$ equipped with the <span style="font-variant:small-caps;">Zariski</span> topology is simple; - let $\varphi$ be an automorphism of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$; there exist an automorphism $\kappa$ of the field $\mathbb{C}$, and a birational self map $\psi$ of $\mathbb{P}^n_\mathbb{C}$ such that $$\varphi(\phi)={}^{\kappa}\!\,(\psi\circ\phi\circ\psi^{-1})\qquad\qquad \forall\,\phi\in\mathrm{G}(n,\mathbb{C}).$$ We will deal with - the <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> theorem in §\[subsection:nc1\] and §\[subsection:nc2\]; - the <span style="font-variant:small-caps;">Hudson</span> and <span style="font-variant:small-caps;">Pan</span> theorem in §\[subsection:hudsonandpan\]; - the fact that there is no non-trivial finite dimensional linear representation of $\mathrm{G}(2,\mathbb{C})$ over any field in §\[section:notlinear\]; - the fact that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})=\mathrm{G}(2,\mathbb{C})$ is perfect in §\[section:perfect\]; - the fact that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})=\mathrm{G}(2,\mathbb{C})$ equipped with the <span style="font-variant:small-caps;">Zariski</span> topology is simple in §\[section:closednormalsubgroups\]; - the description of $\mathrm{Aut}(\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}))=\mathrm{Aut}(\mathrm{G}(2,\mathbb{C}))$ in §\[sec:uncountableabelian\]. Divisors and intersection theory -------------------------------- Let $V$ be an algebraic variety. A prime divisor on $V$ is an irreducible closed subset of $V$ of codimension $1$. For instance if $V$ is a surface, then the prime divisors of $V$ are the irreducible curves that lie on it; if $V$ is the complex projective space, then the prime divisors are given by the zeros locus of irreducible homogeneous polynomials. A <span style="font-variant:small-caps;">Weil</span> divisor on $V$ is a formal finite sum of prime divisors with integer coefficients: $$\displaystyle\sum_{i=1}^ma_iD_i \qquad\qquad m\in\mathbb{N},\,a_i\in\mathbb{Z},\,D_i\text{ prime divisor of $V$}.$$ Let us denote by $\mathrm{Div}(V)$ the set of all <span style="font-variant:small-caps;">Weil</span> divisors of $V$. Let $f\in\mathbb{C}(V)^$ be a rational function, and let $D$ be a prime divisor. The multiplicity* $\nu_f(D)$ of $f$ at $D$ is defined by - $\nu_f(D)=k>0$ if $f$ vanishes on $D$ at the order $k$; - $\nu_f(D)=-k$ if $f$ has a pole of order $k$ on $D$; - $\nu_f(D)=0$ otherwise. To any rational function $f\in\mathbb{C}(V)^$ we associate a divisor $\mathrm{div}(f)$ defined by $$\mathrm{div}(f)=\displaystyle\sum_{\stackrel{\text{$D$ prime}}{\text{divisor}}}\nu_f(D)D.$$ Since $\nu_f(D)$ is zero for all but finitely many $D$ the divisor $\mathrm{div}(f)$ belongs to $\mathrm{Div}(V)$. Divisors obtained like that are called principal divisors. The set of principal divisors form a subgroup of $\mathrm{Div}(V)$; indeed $\mathrm{div}(fg)=\mathrm{div}(f)+\mathrm{div}(g)$ for any $f$, $g\in\mathbb{C}(V)^$. Let us introduce an equivalence relation on $\mathrm{Div}(V)$. Two divisors $D$, $D'$ are linearly equivalent if $D-D'$ is a principal divisor. The set of equivalence classes corresponds to the quotient of $\mathrm{Div}(V)$ by the subgroup of principal divisors. The <span style="font-variant:small-caps;">Picard</span> group of $V$ is the group of isomorphism classes of line bundles on $V$. When $V$ is smooth the quotient of $\mathrm{Div}(V)$ by the subgroup of principal divisors is isomorphic to the <span style="font-variant:small-caps;">Picard</span> group $\mathrm{Pic}(V)$. Let us determine $\mathrm{Pic}(\mathbb{P}^n_\mathbb{C})$. Consider the homomorphism of groups $$\theta\colon\mathrm{Div}(\mathbb{P}^n_\mathbb{C})\to\mathbb{Z}$$ which associates to any divisor $D$ of degree $d$ the integer $d$. Note that $\ker\theta$ is the subgroup of principal divisors of $\mathbb{P}^n_\mathbb{C}$: let $D=\sum a_iD_i$ be an element of $\ker\theta$ where each $D_i$ is a prime divisor given by an homogeneous polynomial $f_i\in\mathbb{C}[z_0,z_1,\ldots,z_n]$ of some degree $d_i$. Since $\sum a_id_i=0$, $f=\prod f_i^{a_i}$ belongs to $\mathbb{C}(\mathbb{P}^n_\mathbb{C})^$. By construction $D=\mathrm{div}(f)$ hence $D$ is a principal divisor. Conversely any principal divisor is equal to $\mathrm{div}(f)$ where $f=g/h$ for some homogeneous polynomials $g$, $h$ of the same degree. Thus any principal divisor belongs to $\ker\theta$. Since $\mathrm{Pic}(\mathbb{P}^n_\mathbb{C})$ is the quotient of $\mathrm{Div}(\mathbb{P}^n_\mathbb{C})$ by the subgroup of principal divisors, we get by restricting $\theta$ to the quotient an isomorphism between $\mathrm{Pic}(\mathbb{P}^n_\mathbb{C})$ and $\mathbb{Z}$. As an hyperplane is sent on $1$ we obtain that $\mathrm{Pic}(\mathbb{P}^n_\mathbb{C})=\mathbb{Z}H$ where $H$ is the divisor of an hyperplane. Let us now assume that $\dim V=2$ and so set $V=S$. We can define the notion of intersection: Let $S$ be a smooth projective surface. There exists a unique bilinear symmetric form $$\begin{aligned} &\mathrm{Div}(S)\times\mathrm{Div}(S)\to\mathbb{Z}&& (C,D)\mapsto C\cdot D\end{aligned}$$ such that - if $C$ and $D$ are smooth curves with transverse intersections, then $C\cdot D=\#(C\cap D)$; - if $C$ and $C'$ are linearly equivalent, then $C\cdot D=C'\cdot D$ for any $D$. In particular this yields an intersection form $$\begin{aligned} &\mathrm{Pic}(S)\times\mathrm{Pic}(S)\to\mathbb{Z}&& (C,D)\mapsto C\cdot D.\end{aligned}$$ Let $\pi\colon\mathrm{Bl}_pS\to S$ be the blow-up of the point $p\in S$. The morphism $\pi$ induces the map $$\begin{aligned} &\pi^\colon\mathrm{Pic}(S)\to\mathrm{Pic}(\mathrm{Bl}_pS), && C\mapsto \pi^{-1}(C).\end{aligned}$$ If $C$ is an irreducible curve on $S$, the strict transform $\widetilde{C}$ of $C$ is $\widetilde{C}=\overline{\pi^{-1}(C\smallsetminus\{p\})}$. If $C\subset S$ is a curve and if $p$ is a point of $S$, let us define the multiplicity $m_p(C)$ of $C$ at $p$. Recall that if $V$ is a quasi-projective variety and if $q$ is a point of $V$, then $\mathcal{O}_{q,V}$ denotes the set of equivalence classes of pairs $(\mathcal{U},\varphi)$ where $\varphi$ belongs to $\mathbb{C}[\mathcal{U}]$ and $\mathcal{U}\subset V$ is an open subset such that $q\in\mathcal{U}$. Let $\mathfrak{m}$ be the maximal ideal of $\mathcal{O}_{p,S}$. If $f$ is a local equation of $C$, then $m_p(C)$ is the integer $k$ such that $f$ belongs to $\mathfrak{m}^k\smallsetminus\mathfrak{m}^{k+1}$. Assume that $S$ is a rational surface. There exists a neighborhood $\mathcal{U}$ of $p$ in $S$ with $\mathcal{U}\subset\mathbb{C}^2$. We can assume that $p=(0,0)$ in this affine neighborhood and that $C$ is a curve described by the equation $\displaystyle\sum_{i=1}^nP_i(z_0,z_1)=0$ where $P_i$ is an homogeneous polynomial of degree $i$. The multiplicity $m_p(C)$ is the lowest $i$ such that $P_i$ is not equal to $0$. The following properties hold: - $m_p(C)\geq 0$, - $m_p(C)=0$ if and only if $p$ does not belong to $C$, - $m_p(C)=1$ if and only if $p$ is a smooth point of $C$. Assume that $C$ and $D$ are distinct curves with no common component ; we can define an integer $(C\cdot D)_p$ which counts the intersection of $C$ and $D$ at $p$: - if either $C$ or $D$ does not pass through $p$, it is equal to $0$; - otherwise let $f$, resp. $g$ be some local equation of $C$, resp. $D$ in a neighborhood of $p$ and define $(C\cdot D)_p$ to be the dimension of $\faktor{\mathcal{O}_{p,S}}{(f,g)}$. This number is related to $C\cdot D$ by the following statement: If $C$ and $D$ are distinct curves without any common irreducible component on a smooth surface, then $$C\cdot D=\displaystyle\sum_{p\in C\cap D}(C\cdot D)_p.$$ In particular $C\cdot D\geq 0$. Let $C$ be a curve on $S$, and let $p$ be a point of $S$. Take local coordinates $z_0$, $z_1$ at $p$ such that $p=(0,0)$. Set $k=m_p(C)$. The curve $C$ is thus given by $$P_k(z_0,z_1)+P_{k+1}(z_0,z_1)+\ldots+P_r(z_0,z_1)=0$$ where the $P_i$’s denote homogeneous polynomials of degree $i$. The blow up of $p$ can be viewed as $(u,v)\mapsto(uv,v)$, and the pull-back of $C$ is given by $$v^k\big(p_k(u,1)+vp_{k+1}(u,1)+\ldots+v^{r-k}p_r(u,1)\big)=0.$$ In other words the pull-back of $C$ decomposes into $k$ times the exceptional divisor $E=\pi^{-1}(0,0)=(v=0)$ and the strict transform. We can thus state: \[lem:tireenarriere\] Let $S$ be a smooth surface. Let $\pi\colon\mathrm{Bl}_pS\to S$ be the blow-up of a point $p\in S$. If $C$ is a curve on $S$, if $\widetilde{C}$ is its strict transform and if $E=\pi^{-1}(p)$ is the exceptional divisor, then $$\pi^(C)=\widetilde{C}+m_p(C)E.$$ We also have the following statement: \[pro:tireenarriere\] Let $S$ be a smooth surface, let $p$ be a point of $S$, and let $\pi\colon\mathrm{Bl}_pS\to S$ be the blow-up of $p$. Denote by $E\subset\mathrm{Bl}_pS$ the exceptional divisor $\pi^{-1}(p)\simeq\mathbb{P}^1_\mathbb{C}$. Then $$\mathrm{Pic}(\mathrm{Bl}_pS)=\pi^\mathrm{Pic}(S)+\mathbb{Z}E.$$ The intersection form on $\mathrm{Bl}_pS$ is induced by the intersection form on $S$ via the following formulas: - $\pi^C\cdot\pi^D=C\cdot D$ for any $C$, $D$ in $\mathrm{Pic}(S)$; - $\pi^C\cdot E=0$ for any $C$ in $\mathrm{Pic}(S)$; - $E^2=E\cdot E=-1$; - $\widetilde{C}^2=C^2-1$ for any smooth curve $C$ passing through $p$ and where $\widetilde{C}$ is the strict transform of $C$. If $V$ is an algebraic variety, then the nef cone* $\mathrm{Nef}(V)$ is the cone of divisors $D$ such that $D\cdot C\geq 0$ for any curve $C$ in $V$. A geometric definition of birational maps {#sec:geodef} ----------------------------------------- Let $\phi$ be the element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ given by $$\phi\colon(z_0:z_1:z_2)\dashrightarrow\big(\phi_0(z_0,z_1,z_2):\phi_1(z_0,z_1,z_2):\phi_2(z_0,z_1,z_2)\big)$$ where the $\phi_i$’s are homogeneous polynomials of the same degree $\nu$, and without common factor of positive degree. The linear system $\Lambda_\phi$ of $\phi$ is the preimage of the linear system of lines of $\mathbb{P}^2_\mathbb{C}$; in other words $$\Lambda_\phi=\displaystyle\bigcup_{(p_0:p_1:p_2)\in\mathbb{P}^2_\mathbb{C}}\big\{(z_0:z_1:z_2)\in\mathbb{P}^2_\mathbb{C}\,\big\vert\,\displaystyle\sum_{i=0}^2 p_i\phi_i(z_0,z_1,z_2)=0\big\}.$$ - If $A$ is an automorphism of $\mathbb{P}^2_\mathbb{C}$, then $\Lambda_\phi=\Lambda_{A\circ\phi}$. - The degree of the curves of $\Lambda_\phi$ is $\nu$. The linear system associated to $\sigma_2$ is the linear system of conics passing through $(1:0:0)$, $(0:1:0)$ and $(0:0:1)$. More generally let $D$ be a divisor on a surface $S$. The complete linear system associated to $D$ is the set of effective divisors of $S$ that are linearly equivalent to $D$: $$\vert D\vert=\big\{D'\in\mathrm{Div}(S)\,\vert\, D'\geq 0,\,D'\sim D\big\}.$$ It can be identified to a projective space of finite dimension. A linear system is a linear subspace of a complete linear system. If $p_1$ is a point of indeterminacy of $\phi$, then denote by $\pi_1\colon\mathrm{Bl}_{p_1}\mathbb{P}^2_\mathbb{C}\to\mathbb{P}^2_\mathbb{C}$ the blow-up of $p_1$ and by $\mathcal{E}_1$ the associated exceptional divisor. The map $\varphi_1=\phi\circ\pi_1$ is a birational map from $\mathrm{Bl}_{p_1}\mathbb{P}^2_\mathbb{C}$ to $\mathbb{P}^2_\mathbb{C}$. If $p_2$ is a point of indeterminacy of $\varphi_1$, we blow up $p_2$ via $\pi_2\colon\mathrm{Bl}_{p_1,p_2}\mathbb{P}^2_\mathbb{C}\to\mathbb{P}^2_\mathbb{C}$ and set $\mathcal{E}_2=\pi_2^{-1}(p_2)$. Again the map $\varphi_2=\varphi_1\circ\pi_1\colon\mathrm{Bl}_{p_1,p_2}\mathbb{P}^2_\mathbb{C}\dashrightarrow\mathbb{P}^2_\mathbb{C}$ is a birational map. We iterate this processus until $\varphi_r$ becomes a morphism. Set $E_i=(\pi_{i+1}\circ\ldots\circ\pi_r)^\mathcal{E}_i$ and $\ell=(\pi_1\circ\ldots\circ\pi_r)^L$ where $L$ is the divisor of a line. Applying $r$ times Proposition \[pro:tireenarriere\] we get $$\left\{ \begin{array}{lllll} \mathrm{Pic}(\mathrm{Bl}_{p_1,p_2,\ldots,p_r}\mathbb{P}^2_\mathbb{C})=\mathbb{Z}\ell\oplus\mathbb{Z}E_1\oplus\mathbb{Z}E_2\oplus\ldots\oplus\mathbb{Z}E_r,\\ \ell^2=\ell\cdot\ell,\\ E_i^2=E_i\cdot E_i=-1, \\ E_i\cdot E_j=0\quad\forall\,1\leq i\not=j\leq r,\\ E_i\cdot\ell=0\quad\forall\,1\leq i\leq r. \end{array} \right.$$ The curves of $\Lambda_\phi$ pass through the $p_i$’s with multiplicity $m_{p_i}(\phi)$. Applying $r$ times Lemma \[lem:tireenarriere\] the elements of $\Lambda_{\varphi_r}$ are equivalent to $$\nu L-\displaystyle\sum_{i=1}^rm_{p_i}(\phi)E_i$$ where $L$ is a generic line. As a result the curves of $\Lambda_{\varphi_r}$ have self intersection $\nu^2-\displaystyle\sum_{i=1}^rm_{p_i}(\phi)^2$. Note that all the members of a linear system are linearly equivalent and that the dimension of $\Lambda_{\varphi_r}$ is $2$; the self intersection has thus to be non-negative. As a consequence the number $r$ exists; in other words $\phi$ has a finite number of base-points. By construction $$\varphi_r\colon\mathrm{Bl}_{p_1,p_2,\ldots,p_r}\mathbb{P}^2_\mathbb{C}\to\mathbb{P}^2_\mathbb{C}$$ is a birational morphism which is the blow-up of the base-points of $\phi^{-1}$. Consider a general line $L$ of $\mathbb{P}^2_\mathbb{C}$ that does not pass through $p_1$, $p_2$, $\ldots$, $p_r$. Its pull-back $\varphi_r^{-1}(L)$ corresponds to a smooth curve on $\mathrm{Bl}_{p_1,p_2,\ldots,p_r}\mathbb{P}^2_\mathbb{C}$ which has self-intersection $-1$ and genus $0$. Hence $$\left\{ \begin{array}{ll} (\varphi_r^{-1}(L))^2=1, \\ \varphi_r^{-1}(L)\cdot K_{\mathrm{Bl}_{p_1,p_2,\ldots,p_r}\mathbb{P}^2_\mathbb{C}}=-3. \end{array} \right.$$ As the elements of $\Lambda_{\varphi_r}$ are equivalent to $\nu L-\displaystyle\sum_{i=1}^rm_{p_i}(\phi)E_i$ and since $K_{\mathrm{Bl}_{p_1,p_2,\ldots,p_r}\mathbb{P}^2_\mathbb{C}}=-3L+\displaystyle\sum_{i=1}^rE_i$ the following equalities hold: $$\left\{ \begin{array}{ll} \displaystyle\sum_{i=1}^rm_{p_i}(\phi)=3(\nu-1),\\ \displaystyle\sum_{i=1}^rm_{p_i}(\phi)^2=\nu^2-1. \end{array} \right.$$ - If $\nu=2$, then $r=3$ and $m_{p_1}(\phi)=m_{p_2}(\phi)=m_{p_3}(\phi)=1$. - If $\nu=3$, then $r=5$ and $m_{p_1}(\phi)=2$, $m_{p_2}(\phi)=m_{p_3}(\phi)=m_{p_4}(\phi)=m_{p_5}(\phi)=1$. An isometric action of the <span style="font-variant:small-caps;">Cremona</span> group on an infinite dimensional hyperbolic space {#chap:hyperbolicspace} ================================================================================================================================== If $S$ is a projective surface, the group $\mathrm{Bir}(S)$ of birational self maps of $S$ acts faithfully by isometries on a hyperbolic space $\mathbb{H}^\infty(S)$ of infinite dimension. After recalling some notions of hyperbolic geometry in the first section of this chapter we describe this construction in the second section. Let us now give an outline of it. Let $S$ be a projective surface. If $\pi\colon Y\to S$ is a birational morphism, then one obtains an embedding $\pi^\colon\mathrm{NS}(S)\to\mathrm{NS}(Y)$ of <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> groups. If $\pi_1\colon Y_1\to S$ and $\pi_2\colon Y_2\to S$ are two birational morphisms, then - $\pi_2$ is above $\pi_1$ if $\pi_1^{-1}\circ\pi_2$ is regular, - one can always find a third birational morphism $\pi_3\colon Y_3\to S$ that is above $\pi_1$ and $\pi_2$. Hence the inductive limit of all groups $\mathrm{NS}(Y_i)$ for all surfaces $Y_i$ above $S$ is well-defined; this limit $\mathcal{Z}(S)$ is the <span style="font-variant:small-caps;">Picard</span>-<span style="font-variant:small-caps;">Manin</span> space of $S$. The intersection forms on $Y_i$ yield to a scalar product $\langle\,,\,\rangle$ on $\mathcal{Z}(S)$. Consider all surfaces $Y$ above $S$, [i.e.]{} all birational morphisms $\pi\colon Y\to S$. We identify $p_1\in Y_1$ and $p_2\in Y_2$ if $\pi_1^{-1}\circ\pi_2$ is a local isomorphism in a neighborhood of $p_2$ that maps $p_2$ onto $p_1$. The bubble space $\mathcal{B}(S)$ of $S$ is the union of all points of all surfaces above $S$ modulo the equivalence relation generated by these identifications. If $p$ belongs to $\mathcal{B}(S)$, then $\mathbf{e}_p$ is the divisor class of the exceptional divisor of the blow up of $p$. The equalities $\mathbf{e}_p\cdot\mathbf{e}_p=-1$ and $\mathbf{e}_p\cdot\mathbf{e}_{p'}=0$ hold. The <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> group $\mathrm{NS}(S)$ is naturally embedded as a subgroup of the <span style="font-variant:small-caps;">Picard</span>-<span style="font-variant:small-caps;">Manin</span> space; this finite dimensional lattice is orthogonal to $\mathbf{e}_p$ for any $p\in\mathcal{B}(S)$. More precisely $$\mathcal{Z}(S)=\mathrm{NS}(S)\displaystyle\bigoplus_{p\in\mathcal{B}(S)}\mathbb{Z}\mathbf{e}_p.$$ As a result any element $v$ of $\mathcal{Z}(S)$ can be written as a finite sum $$v=w+\displaystyle\sum_{p\in\mathcal{B}(S)}m_p\mathbf{e}_p.$$ There is a completion process for which the completion $\mathrm{Z}(S)$ of $\mathcal{Z}(S)\otimes_\mathbb{Z}\mathbb{R}$ is $$\mathrm{Z}(S)=\Big\{w+\displaystyle\sum_{p\in\mathcal{B}(S)}m_p\mathbf{e}_p\,\vert\, w\in\mathrm{NS}(\mathbb{R},S),\,\displaystyle\sum_{p\in\mathcal{B}(S)}m_p^2<\infty\Big\}.$$ The intersection form extends as a scalar product with signature $(1,\infty)$ on this space. The hyperbolic space $\mathbb{H}^\infty(S)$ of $S$ is defined by $$\mathbb{H}^\infty(S)=\big\{w\in\mathrm{Z}(S),\,\vert\, \langle w,\,w\rangle=1,\,\langle w,\,a\rangle>0\text{ for all ample classes } a\in\mathrm{NS}(S)\big\}.$$ It is an infinite dimensional analogue of the classical hyperbolic space $\mathbb{H}^n$. One can define a complete distance $\mathrm{dist}$ on $\mathbb{H}^\infty(S)$ by $$\cosh(\mathrm{dist}(v,w))=\langle v,\,w\rangle.$$ Geodesics are intersection of $\mathbb{H}^\infty(S)$ with planes. The projection of $\mathbb{H}^\infty(S)$ in the projective space $\mathbb{P}(\mathrm{Z}(S))$ is one to one and the boundary of its image is the projection of the cone of isotropic vectors of $\mathrm{Z}(S)$: $$\partial\mathbb{H}^\infty(S)=\big\{\mathbb{R}_+v\subset\mathrm{Z}(S)\,\vert\,\langle v,\,v\rangle=0,\,\langle v,\,a\rangle>0\text{ for all ample classes }a\in\mathrm{NS}(S)\big\}.$$ The important fact is that $\mathrm{Bir}(S)$ acts faithfully on $\mathrm{Z}(S)$ by continuous linear endomorphisms preserving the intersection form, the effective cone, the nef cone, $\mathcal{Z}(S)$ and also $\mathbb{H}^\infty(S)$. If $\phi$ is an element of $\mathrm{Bir}(S)$, we denote by $\phi_$ its action on $\mathrm{Z}(S)$: it is a linear isometry with respect to the intersection form; we also denote by $\phi_$ the isometry of $\mathbb{H}^\infty(S)$ induced by this endomorphism of $\mathrm{Z}(S)$. Let $f$ be an isometry of $\mathbb{H}^\infty(S)$; the translation length of $f$ is $$L(f)=\inf\big\{\mathrm{dist}(v,f(v))\,\vert\, v\in\mathbb{H}^\infty(S)\big\}.$$ If this infimum is a minimum, then - either it is equal to $0$, $f$ has a fixed point in $\mathbb{H}^\infty(S)$ and $f$ is elliptic; - or it is positive and $f$ is loxodromic. When the infimum is not realized, $L(f)$ is equal to $0$ and $f$ is parabolic. This classification into three types holds for all isometries of $\mathbb{H}^\infty(S)$. For isometries $\phi_$ induced by birational maps $\phi$ of $S$ there is a dictionary between this classification and the geometric properties of $\phi$. We give this dictionary in the third section. Some hyperbolic geometry ------------------------ Consider a real <span style="font-variant:small-caps;">Hilbert</span> space $\mathcal{H}$ of dimension $n$, where $n$ can be infinite. Let $\mathbf{e}_0$ be a unit vector of $\mathcal{H}$, and let $\mathbf{e}_0^\perp$ be the orthogonal complement of the space $\mathbb{R}\mathbf{e}_0$. Denote by $(\mathbf{e}_i)_{i\in I}$ an orthonormal basis of $\mathbf{e}_0^\perp$. A scalar product with signature $(1,n-1)$ can be defined on $\mathcal{H}$ by setting $$\langle u,\,v\rangle=a_0b_0-\displaystyle\sum_{i\in I}a_ib_i$$ for any two elements $u=a_0\mathbf{e}_0+\displaystyle\sum_{i\in I}a_i\mathbf{e}_i$ and $v=b_0\mathbf{e}_0+\displaystyle\sum_{i\in I}b_i\mathbf{e}_i$ of $\mathcal{H}$. The set $$\big\{v\in\mathcal{H}\,\vert\,\langle v,v\rangle=1\big\}$$ defines a hyperboloid with two connected components. Let $\mathbb{H}^{n-1}$ be the connected component of this hyperboloid that contains $\mathbf{e}_0$. A metric can be defined on $\mathbb{H}^{n-1}$ by $$d(u,v):=\mathrm{arccosh}(\langle u,v\rangle).$$ A useful model for $\mathbb{H}^2$ is the <span style="font-variant:small-caps;">Poincaré</span> model: $\mathbb{H}^2$ is identified to the upper half-plane $\big\{z\in\mathbb{C}\,\vert\,\mathrm{Im}(z)>0\big\}$ with its Riemanniann metric given by $\mathrm{d}s^2=\frac{x^2+y^2}{y^2}$. Its group of orientation preserving isometries coincides with $\mathrm{PSL}(2,\mathbb{R})$, acting by linear fractional transformations. ### $\delta$-hyperbolicity and $\mathrm{CAT}$($-1$) spaces Let $(X,d)$ be a geodesic metric space. Let $x$, $y$, $z$ be three points of $X$. We denote by $[p,q]$ the segment with endpoints $p$ and $q$. A geodesic triangle with vertices $x$, $y$, $z$ is the union of three geodesic segments $[x,y]$, $[y,z]$ and $[z,x]$. Let $\delta\geq 0$. If for any point $m\in[x,y]$ there is a point in $[y,z]\cup[z,x]$ at distance less than $\delta$ of $m$, and similarly for points on the other edges, then the triangle is said do be $\delta$-slim. A $\delta$-hyperbolic space is a geodesic metric space whose all of geodesic triangles are $\delta$-slim. - Metric trees are $0$-hyperbolic: all triangles are tripods. - The hyperbolic plane is hyperbolic. In fact the incircle of a geodesic triangle is the circle of largest diameter contained in the triangle and any geodesic triangle lies in the interior of an ideal triangle, all of which are isometric with incircles of diameter $2\log 3$ (see [@CoornaertDelzantPapadopoulos]). - Nevertheless $\mathbb{R}^2$ endowed with the euclidian metric is not $\delta$-hyperbolic (for instance because of the existence of homotheties). Let us now introduce $\mathrm{CAT}$($-1)$ spaces[^2]. Let $(X,d_X)$ be a geodesic metric space. Consider a geodesic triangle $T$ in $X$ determined by the three points $x$, $y$, $z$ and the data of three geodesics between two of these three points. A comparison triangle of $T$ in the space $(X',d_{X'})$ is a triangle $T'$ such that $$\left\{ \begin{array}{lll} d_X(x,y)=d_{X'}(x',y')\\ d_X(x,z)=d_{X'}(x',z')\\ d_X(y,z)=d_{X'}(y',z')\\ \end{array} \right.$$ Let $p$ be a point of $[x,y]\subset T$. A point $p'\in[x',y']\subset T'$ is a comparison point of $p$ if $d_{X'}(x',p')=d_X(x,p)$. The triangle $T$ satisfies the $\mathrm{CAT}$$(-1)$ inequality if for any $(x,y)\in T^2$ $$d_X(x,y)\leq\vert\vert x'-y'\vert\vert_{\mathbb{H}^2}$$ where $T'$ is a comparison triangle of $T$ in $\mathbb{H}^2$ and $x'\in T'$ (resp. $y'\in T'$) is a comparison point of $x$ (resp. $y$). The space $X$ is $\mathrm{CAT}$$(-1)$ if all its triangles satisfy the $\mathrm{CAT}$($-1$) inequality. The $\mathrm{CAT}$($-1$) spaces are <span style="font-variant:small-caps;">Gromov</span> hyperbolic but the converse is false. Set $\mathcal{H}_{>0}=\big\{v\in\mathcal{H}\,\vert\, \langle v,v\rangle>0\big\}$. The image of $v$ by the map $$\begin{aligned} &\eta\colon\mathcal{H}_{>0}\to\mathbb{H}^\infty && v\mapsto \frac{v}{\sqrt{\langle v,v\rangle}}\end{aligned}$$ is called the normalization of $v$. Geometrically $\eta$ associates to a point $v\in\mathcal{H}_{>0}$ the intersection of $\mathbb{H}^\infty$ with the line through $v$. Note that if the intersection of $\mathcal{H}$ with a vectorial subspace of dimension $n+1$ of $\mathcal{H}$ is not empty, then it is a copy of $\mathbb{H}^n$. In particular there exists a unique geodesic segment between two points of $\mathbb{H}^\infty$ obtained as the intersection of $\mathbb{H}^\infty$ with the plane that contains these two points. Hence any triangle of $\mathbb{H}^\infty$ is isometric to a triangle of $\mathbb{H}^2$. As a result $\mathbb{H}^\infty$ is $\mathrm{CAT}$($-1$) and $\delta$-hyperbolic for the same constant $\delta $ as $\mathbb{H}^2$. ### Boundary of $\mathbb{H}^\infty$ Let $(X,d)$ be a geodesic metric space. Let $T$ be a geodesic triangle of $X$ given by $x$, $y$, $z\in X$ and geodesic segments between two of these three points. The triangle $T$ satisfies the $\mathrm{CAT}$$(0)$ inequality if for any $(x,y)\in T^2$ $$d_X(x,y)\leq\vert\vert x'-y'\vert\vert_{\mathbb{R}^2}$$ where $x'\in T'$ (resp. $y'\in T'$) is a comparison point of $x$ (resp. $y$) and $T'$ is a comparison triangle of $T$ in $\mathbb{R}^2$. The space $X$ is $\mathrm{CAT}$$(0)$ if all its triangles satisfy the $\mathrm{CAT}$($0$) inequality. A $\mathrm{CAT}$($-1$) space is a $\mathrm{CAT}$($0$) space. In particular $\mathbb{H}^\infty$ is a $\mathrm{CAT}$($0$) space. Since $\mathbb{H}^\infty$ is a $\mathrm{CAT}$($0$), complete metric space there exists a notion of boundary at infinity that generalizes the notion of boundary of finite dimensional <span style="font-variant:small-caps;">Riemann</span> varieties which are complete, simply connected and with negative curvature. The boundary of $\mathbb{H}^\infty$ is defined by $$\partial\mathbb{H}^\infty=\big\{v\in\mathcal{H}\,\vert\,\langle v,v\rangle=0,\,\langle v,\mathbf{e}_0\rangle>0\big\}.$$ A point of $\partial\mathbb{H}^\infty$ is called point at infinity. ### Isometries {#subsec:isometries} Denote by $\mathrm{O}_{1,n}(\mathbb{R})$ the group of linear transformations of $\mathcal{H}$ preserving the scalar product $\langle\,,\rangle$. The group of isometries $\mathrm{Isom}(\mathbb{H}^n)$ coincides with the index $2$ subgroup $\mathrm{O}_{1,n}^+(\mathbb{R})$ of $\mathrm{O}(\mathcal{H})$ that preserves the chosen sheet $\mathbb{H}^n$ of the hyperboloid $$\big\{u\in\mathcal{H}\,\vert\,\langle u,u\rangle=1\big\}.$$ This group acts transitively on $\mathbb{H}^n$, and on its unit tangent bundle. If $h$ is an isometry of $\mathbb{H}^n$ and $v\in\mathcal{H}$ is an eigenvector of $h$ with eigenvalue $\lambda$, then either $\vert\lambda\vert=1$ or $v$ is isotropic. Furthermore $\mathbb{H}^n$ is homeomorphic to a ball so $h$ has a least one eigenvector in $\mathbb{H}^n\cup\partial\mathbb{H}^n$. As a consequence according to [@BurgerIozziMonod] there are three types of isometries: - $h$ is elliptic if and only if $h$ fixes a point $p\in\mathbb{H}^n$. Since $\langle\,,\,\rangle$ is negative definite on $p^\perp$, $h$ fixes pointwise $\mathbb{R}p$ and acts by rotation on $p^\perp$ with respect to $\langle\,,\,\rangle$; - $h$ is parabolic if $h$ is not elliptic and fixes a vector $v$ in the isotropic cone. The line $\mathbb{R}v$ is uniquely determined by $h$. Let $p$ be a point of $\mathbb{H}^n$; there exists an increasing sequence $(n_i)\in\mathbb{N}^\mathbb{N}$ such that $(h^{n_i}(p))_{i\in\mathbb{N}}$ converges to the boundary point determined by $v$. - $h$ is loxodromic if and only if $h$ has an eigenvector $v_h^+$ with eigenvalue $\lambda>1$. Note that $v_h^+$ is unique up to scalar multiplication. There is another unique isotropic eigenline $\mathbb{R}v_h^-$ corresponding to the eigenvalue $\frac{1}{\lambda}$. On the orthogonal complement of $\mathbb{R}v_h^-\oplus\mathbb{R}v_h^+$ the isometry $h$ acts as a rotation with respect to $\langle\,,\,\rangle$. The boundary points determined by $v_h^-$ and $v_h^+$ are the two fixed points of $h$ in $\mathbb{H}^n\cup\partial\mathbb{H}^n$; the first one is an attracting fixed point $\alpha(h)$, the second one is a repelling fixed point $\omega(h)$. To an isometry $h$ of $\mathbb{H}^n$ one can associate the translation length of $h$: $$L(h)=\inf\big\{d(h(p),p)\,\vert p\in\mathbb{H}^n\big\}.$$ The isometry $h$ is elliptic if and only if $L(h)=0$ and the infimum is achieved, [i.e.]{} $h$ has a fixed point in $\mathbb{H}^n$. The isometry $h$ is parabolic if and only if $L(h)=0$ and the infinimum is not achieved. The isometry $h$ is loxodromic if and only if $L(h)>0$. In that case - $\exp(L(h))$ is the largest eigenvalue of $h$ - and $d(p,h^n(p))$ grows like $nL(h)$ as $n$ goes to infinity for any point $p\in\mathbb{H}^n$. The isometric action of $\mathrm{Bir}(S)$ on an infinite dimensional hyperbolic space ------------------------------------------------------------------------------------- ### The <span style="font-variant:small-caps;">Picard</span>-<span style="font-variant:small-caps;">Manin</span> space Let $S$ be a smooth, irreducible, projective, complex surface. As we see in Chapter \[chapter:intro\] the <span style="font-variant:small-caps;">Picard</span> group $\mathrm{Pic}(S)$ is the quotient of the abelian group of divisors by the subgroup of principal divisors ([@Hartshorne]). The intersection between curves extends to a quadratic form, the so-called intersection form: $$\begin{aligned} & \mathrm{Pic}(S)\times\mathrm{Pic}(S)\to\mathbb{Z}, && (C,D)\mapsto C\cdot D\end{aligned}$$ The quotient of $\mathrm{Pic}(S)$ by the subgroup of divisors $E$ such that $E\cdot D=0$ for all divisor classes $D$ is the <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> group $\mathrm{NS}(S)$. In case of rational surfaces we have $\mathrm{NS}(S)=\mathrm{Pic}(S)$. The <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> group is a free abelian group, and its rank, the <span style="font-variant:small-caps;">Picard</span> number is finite. The pull-back of a birational morphism $\pi\colon Y\to S$ yields an injection from $\mathrm{Pic}(S)$ into $\mathrm{Pic}(Y)$; we thus get an injection from $\mathrm{NS}(S)$ into $\mathrm{NS}(Y)$. The morphism $\pi\colon Y\to S$ can be written as a finite sequence of blow ups. Let $\mathbf{e}_1$, $\mathbf{e}_2$, $\ldots$, $\mathbf{e}_k\subset Y$ be the class of the irreducible components of the exceptional divisor of $\pi$, that is the classes contracted by $\pi$. We have the following decomposition $$\label{eq:decom} \mathrm{NS}(Y)=\mathrm{NS}(S)\oplus\mathbb{Z}\mathbf{e}_1\oplus\mathbb{Z}\mathbf{e}_2\oplus\ldots\oplus\mathbb{Z}\mathbf{e}_k$$ which is orthogonal with respect to the intersection form. Consider $\pi_1\colon Y\to S$ and $\pi_2\colon Y'\to S$ two birational morphisms of smooth projective surfaces. We say that $\pi_1$ is above $\pi_2$ if $\pi_2^{-1}\circ\pi_1$ is a morphism. For any two birational morphisms $\pi_1\colon Y\to S$ and $\pi_2\colon Y'\to S$ there exists a birational morphism $\pi_3\colon Y''\to S$ that lies above $\pi_1$ and $\pi_2$. Let us consider the set of all birational morphisms of smooth projective surfaces $\pi\colon Y\to S$. The corresponding embeddings of the <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> groups $\mathrm{NS}(S)\to\mathrm{NS}(Y)$ form a directed family; the direct limit $$\mathcal{Z}(S):=\displaystyle\lim_{\pi\colon Y\to S}\mathrm{NS}(Y)$$ thus exists. It is called the <span style="font-variant:small-caps;">Picard</span> <span style="font-variant:small-caps;">Manin</span> space of $S$. The intersection forms on the groups $\mathrm{NS}(Y)$ induce a quadratic form on $\mathcal{Z}(S)$ of signature $(1,\infty)$. Let $p$ be a point of the bubble space of $S$. Denote by $\mathbf{e}_p$ the divisor class of the exceptional divisor of the blow-up of $p$ in the corresponding <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> group. One deduces from (\[eq:decom\]) the following decomposition $$\displaystyle\mathcal{Z}(S)=\mathrm{NS}(S)\oplus\displaystyle\bigoplus_{p\in\mathcal{B}(S)}\mathbb{Z}\mathbf{e}_p.$$ Furthermore the following properties hold $$\left\{ \begin{array}{ll} \mathbf{e}_p\cdot \mathbf{e}_p=-1 \\ \mathbf{e}_p\cdot \mathbf{e}_q=0 \text{ for all $p\not=q$} \end{array} \right.$$ ### The hyperbolic space $\mathbb{H}^\infty(S)$ Let $S$ be a smooth projective surface, and let $\mathcal{Z}(S)$ be its <span style="font-variant:small-caps;">Picard</span>-<span style="font-variant:small-caps;">Manin</span> space. We define $\mathrm{Z}(S)$ to be the completion of the real vector space $\mathcal{Z}(S)\otimes\mathbb{R}$ $$\mathrm{Z}(S)=\Big\{v+\displaystyle\sum_{p\in\mathcal{B}(S)}m_p\mathbf{e}_p\,\vert\, v\in\mathrm{NS}(S)\otimes\mathbb{R},\,m_p\in\mathbb{R},\,\displaystyle\sum_{p\in\mathcal{B}(S)}m_p^2<\infty\Big\}.$$ The intersection form extends continuously to a quadratic form on $\mathrm{Z}(S)$ with signature $(1,\infty)$. Let $\mathrm{Isom}(\mathrm{Z}(S))$ be the group of isometries of $\mathrm{Z}(S)$ with respect to the intersection form. The set of vectors $v\in\mathrm{Z}(S)$ such that $\langle v,v\rangle=1$ is a hyperboloid. The subset $$\mathbb{H}^\infty(S)=\big\{v\in\mathrm{Z}(S)\,\vert\,\langle v,v\rangle=1,\, \langle v,\mathbf{e}_0\rangle> 0\big\}$$ is the sheet of that hyperboloid containing ample classes of $\mathrm{NS}(S,\mathbb{R})$. Let $\mathrm{Isom}(\mathbb{H}^\infty(S))$ be the subgroup of $\mathrm{Isom}(\mathcal{Z}(S))$ that preserves $\mathbb{H}^\infty(S)$. The space $\mathbb{H}^\infty(S)$ equipped with the distance defined by $$\cosh(d(v,v'))=\langle v,v'\rangle$$ is isometric to a hyperbolic space $\mathbb{H}^\infty$. Let $\partial\mathbb{H}^\infty(S)$ be the boundary of $\mathbb{H}^\infty(S)$. To simplify we will often write $\mathbb{H}^\infty$ (resp. $\partial\mathbb{H}^\infty$) instead of $\mathbb{H}^\infty(S)$ (resp. $\partial\mathbb{H}^\infty(S)$). ### An isometric action of $\mathrm{Bir}(S)$ Let us now describe the action of $\mathrm{Bir}(S)$ on $\mathbb{H}^\infty$ (see [@Manin; @Cantat:annals]). Let $\phi\colon Y\to S$ be a birational morphism of smooth projective surfaces. Denote by $p_1$, $p_2$, $\ldots$, $p_n\in\mathcal{B}(S)$ the points blown up by $\phi$. Denote by $\mathbf{e}_{p_i}$ the irreducible component of the exceptional divisor contracted to $p_i$. One has $$\mathrm{NS}(Y)=\mathrm{NS}(S)\oplus\mathbb{Z}\mathbf{e}_{p_1}\oplus\mathbb{Z}\mathbf{e}_{p_2}\oplus\ldots\oplus\mathbb{Z}\mathbf{e}_{p_n}.$$ The morphism $\phi$ induces the isomorphism $\phi_\colon\mathcal{Z}(Y)\to\mathcal{Z}(S)$ defined by $$\left\{ \begin{array}{lll} \phi_(\mathbf{e}_p)=\mathbf{e}_{\phi_\bullet(p)}\qquad\forall\,p\in\mathcal{B}(Y)\smallsetminus\mathrm{Base}(\phi)\\ \phi_(\mathbf{e}_{p_i})=\mathbf{e}_{p_i}\qquad\forall\,1\leq i\leq n\\ \phi_(D)=D\qquad\forall\,D\in\mathrm{NS}(S)\subset\mathrm{NS}(Y) \end{array} \right.$$ Let $\phi\colon Y\dashrightarrow S$ be a birational map of smooth projective surfaces. Let $\pi_2\circ\pi_1^{-1}$ be a minimal resolution of $\phi$. The map $\phi$ induces an isomorphism $\phi_\colon\mathcal{Z}(Y)\to\mathcal{Z}(S)$ defined by $$\phi_=(\pi_2)_\circ(\pi_1)_^{-1}.$$ Let $S$ be a smooth projective surface. Any element $\phi$ of $\mathrm{Bir}(S)$ induces an isomorphism $\phi_\colon\mathcal{Z}(S)\to\mathcal{Z}(S)$ and $\phi_$ yields an automorphism of $\mathcal{Z}(S)\otimes\mathbb{R}$ which extends to an automorphism of the completion $\mathrm{Z}(S)$ and preserves the intersection form. Let $\phi$ be a birational self map of $\mathbb{P}^2_\mathbb{C}$. Assume that $\phi$ has degree $d$. Then the base-point $\mathbf{e}_0$, [i.e.]{} the class of a line in $\mathbb{P}^2_\mathbb{C}$, is mapped by $\phi_$ to the finite sum $$d\mathbf{e}_0-\displaystyle\sum_im_i\mathbf{e}_{p_i}$$ where each $m_i$ is a positive integer and $\mathbf{e}_{p_i}$ are the classes of the exceptional divisors corresponding to the base-points of $\phi^{-1}$. For instance if $\phi=\sigma_2$ is the standard <span style="font-variant:small-caps;">Cremona</span> involution, then $$(\sigma_2)_\mathbf{e}_0=2\mathbf{e}_0-\mathbf{e}_{p_1}-\mathbf{e}_{p_2}-\mathbf{e}_{p_3}$$ where $p_1=(1:0:0)$, $p_2=(0:1:0)$ and $p_3=(0:0:1)$. An invariant structure is given by the canonical form. The canonical class of $\mathbb{P}^2_\mathbb{C}$ blown up in $n$ points $p_1$, $p_2$, $\ldots$, $p_n$ is equal to $-3\mathbf{e}_0-\displaystyle\sum_{j=1}^n\mathbf{e}_{p_j}$. By taking intersection products one obtains a linear form $\omega_\infty$ defined by $$\begin{aligned} & \omega_\infty\colon\mathcal{Z}(\mathbb{P}^2_\mathbb{C})\to\mathbb{Z}, && m_0\mathbf{e}_0-\displaystyle\sum_{j=1}^nm_j\mathbf{e}_{p_j}\mapsto -3m_0+\displaystyle\sum_{j=1}^nm_j\end{aligned}$$ Since the isometric action of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ on $\mathcal{Z}(\mathbb{P}^2_\mathbb{C})$ preserves the linear form $\omega_\infty$ we get the following equalities already obtained in §\[sec:geodef\]: if $\phi_\mathbf{e}_0=d\mathbf{e}_0-\displaystyle\sum_{j=1}^nm_j\mathbf{e}_{p_j}$, then $$\left\{ \begin{array}{ll} d^2=1+\displaystyle\sum_{j=1}^nm_j^2\\ 3d-3=\displaystyle\sum_{j=1}^nm_j \end{array} \right.$$ Let us understand the isometry $(\sigma_2)_$. Denote by $p_1$, $p_2$ and $p_3$ the base-points of $\sigma_2$, and set $S=\mathrm{Bl}_{p_1,p_2,p_3}\mathbb{P}^2_\mathbb{C}$. The involution $\sigma_2$ lifts to an automorphism $\widetilde{\sigma_2}$ on $S$. The <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> group $\mathrm{NS}(S)$ of $S$ is the lattice of rank $4$ generated by the class $\mathbf{e}_0$, coming from the class of a line in $\mathbb{P}^2_\mathbb{C}$, and the classes $\mathbf{e}_i=\mathbf{e}_{p_i}$ given by the three exceptional divisors. The action of $\widetilde{\sigma_2}$ on $\mathrm{NS}(S)$ is given by $$\left\{ \begin{array}{llll} (\widetilde{\sigma_2})_\mathbf{e}_0=2\mathbf{e}_0-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3\\ (\widetilde{\sigma_2})_\mathbf{e}_1=\mathbf{e}_0-\mathbf{e}_2-\mathbf{e}_3 \\ (\widetilde{\sigma_2})_\mathbf{e}_2=\mathbf{e}_0-\mathbf{e}_1-\mathbf{e}_3 \\ (\widetilde{\sigma_2})_\mathbf{e}_3=\mathbf{e}_0-\mathbf{e}_1-\mathbf{e}_2 \end{array} \right.$$ Then $(\widetilde{\sigma_2})_$ coincides on $\mathrm{NS}(S)$ with the reflection with respect to $\mathbf{e}_0-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3$: $$(\widetilde{\sigma_2})_(p)=p+\langle p,\mathbf{e}_0-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3\rangle\qquad\qquad\forall\,p\in\mathrm{NS}(S)$$ Let us blow up all points of $S$; we thus obtain a basis of $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$: $$\mathrm{Z}(\mathbb{P}^2_\mathbb{C})=\mathrm{NS}(S)\displaystyle\bigoplus_{p\in\mathcal{B}(S)}\mathbb{Z}\mathbf{e}_p.$$ The isometry $(\sigma_2)_$ of $\mathcal{Z}(\mathbb{P}^2_\mathbb{C})$ acts on $\mathrm{NS}(S)$ as the reflection $(\widetilde{\sigma_2})_$ and permutes each vector $\mathbf{e}_p$ with $\mathbf{e}_{\sigma_2(p)}$. Types and degree growth {#sec:degreegrowth} ----------------------- Consider an ample class $\mathbf{h}\in\mathrm{NS}(S,\mathbb{R})$ with self-intersection $1$. The degree of $\phi\in\mathrm{Bir}(S)$ with respect to the polarization $\mathbf{h}$ is defined by $$\deg_\mathbf{h}\phi=\langle\phi_(\mathbf{h}), \mathbf{h}\rangle=\cosh(d(\mathbf{h},\phi_\mathbf{h})).$$ Note that if $S=\mathbb{P}^2_\mathbb{C}$ and $\mathbf{h}=\mathbf{e}_0$ is the class of a line, then $\deg_\mathbf{h}\phi$ is the degree of $\phi$ as defined in Chapter \[chapter:intro\]. A birational map $\phi$ of a projective surface $S$ is - virtually isotopic to the identity if there is a positive iterate $\phi^n$ of $\phi$ and a birational map $\psi\colon Z\dashrightarrow S$ such that $\psi^{-1}\circ\phi^n\circ\psi$ is an element of $\mathrm{Aut}(Z)^0$; - a <span style="font-variant:small-caps;">Jonquières</span> twist if $\phi$ preserves a one parameter family of rational curves on $S$ but $\phi$ is not virtually isotopic to the identity; - a <span style="font-variant:small-caps;">Halphen</span> twist if $\phi$ preserves a one parameter family of genus one curves on $S$ but $\phi$ is not virtually isotopic to the identity. Note that the <span style="font-variant:small-caps;">Jonquières</span> twists (resp. <span style="font-variant:small-caps;">Halphen</span> twists) preserve a unique fibration ([@DillerFavre]). If $\phi$ is a <span style="font-variant:small-caps;">Jonquières</span> (resp. <span style="font-variant:small-caps;">Halphen</span>) twist, then, after conjugacy by a birational map $\psi\colon Z\dashrightarrow S$, $\phi$ permutes the fibers of a rational (resp. genus one) fibration $\pi\colon Z\to B$. If $z$ is the divisor class of the generic fiber of the fibration, then $z$ is an isotropic vector in $\mathcal{Z}(S)$ fixed by $\phi_$. In particular $\phi_$ can not be loxodromic. Let $\mathcal{C}$ and $\mathcal{C}'$ be two smooth cubic curves in the complex projective plane. By <span style="font-variant:small-caps;">Bezout</span> theorem $\mathcal{C}$ and $\mathcal{C}'$ intersect in nine points denoted $p_1$, $p_2$, $\ldots$, $p_9$. There is a pencil of cubic curves passing through these nine points. Let us blow up $p_1$, $p_2$, $\ldots$, $p_9$. We get a rational surface $S$ with a fibration $\pi\colon S\to\mathbb{P}^1_\mathbb{C}$ whose fibers are genus $1$ curves. More generally let us consider a pencil of curves of degree $3m$ for $m\in\mathbb{Z}_+$, blow up its base-points and denote by $S$ the surface we get. Such a pencil of genus $1$ curves is called a <span style="font-variant:small-caps;">Halphen</span> pencil and such a surface is called a <span style="font-variant:small-caps;">Halphen</span> surface of index $m$ . A surface $S$ is a <span style="font-variant:small-caps;">Halphen</span> one if $\vert -mK_S\vert$ satisfies the three following properties - it is one-dimensional, - it has no fixed component, - it is base-point free. According to [@CantatDolgachev] up to birational conjugacy - every pencil of genus $1$ curves of $\mathbb{P}^2_\mathbb{C}$ is a <span style="font-variant:small-caps;">Halphen</span> pencil, - <span style="font-variant:small-caps;">Halphen</span> surfaces are the only examples of rational elliptic surfaces. \[lem:UrechHalphen\] Let $S$ be a <span style="font-variant:small-caps;">Halphen</span> surface. Let $\phi$ be an element of $\mathrm{Bir}(S)$ that preserves the <span style="font-variant:small-caps;">Halphen</span> pencil. Then $\phi$ belongs to $\mathrm{Aut}(S)$. A subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that preserves a pencil of genus $1$ curves is conjugate to a subgroup of the automorphism group of some <span style="font-variant:small-caps;">Halphen</span> surface. The <span style="font-variant:small-caps;">Halphen</span> pencil is defined by a multiple of the class of the anticanonical divisor $-K_S$. As a result any birational map of a <span style="font-variant:small-caps;">Halphen</span> surface that preserves the <span style="font-variant:small-caps;">Halphen</span> fibration preserves the class of the canonical divisor $K_S$. Assume by contradiction that $\phi$ is not an automorphism. Take a minimal resolution of $\phi$ $$\xymatrix{ & Z\ar[rd]^\eta\ar[ld]_\pi & \\ S\ar@{-->}[rr]_\phi & & S }$$ Denote by $E_i$ and $F_i$ the total pull backs of the exceptional curves. On the one hand $$K_Z=\eta^(K_S)+\sum E_i$$ and on the other hand $$K_Z=\pi^(K_S)+\sum F_i.$$ The map $\phi$ preserves $K_S$ so $\eta^(K_S)=\pi^(K_S)$ and hence $\sum E_i=\sum F_i$. By assumption $\phi$ is not an automorphism, [i.e.]{} $\sum E_i$ contains at least one $(-1)$-curve $E_k$. Hence both $$E_k\cdot\Big(\sum E_i\Big)=-1$$ and $$E_k\cdot\Big(\sum F_i\Big)=-1$$ hold. This implies that $E_k$ is contained in the support of $\sum F_i$: contradiction with the minimality of the resolution. The automorphism groups of <span style="font-variant:small-caps;">Halphen</span> surfaces are studied in [@Gizatullin] and in [@CantatDolgachev]. On the contrary <span style="font-variant:small-caps;">Jonquières</span> twists are not conjugate to automorphisms of projective surfaces ([@DillerFavre; @BlancDeserti:degree]). Let $S$ be a projective complex surface with a polarization $H$. Let $\phi\colon S\dashrightarrow S$ be a birational map. The dynamical degree of $\phi$ is defined by $$\lambda(\phi)=\displaystyle\lim_{n\to +\infty}\deg_H(\phi^n)^{1/n}.$$ An element $\phi$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is called elliptic, (resp. parabolic, resp. loxodromic) if the corresponding isometry $\phi_$ is elliptic (resp. parabolic, resp. loxodromic). . The map $\phi$ is loxodromic if and only if $\lambda(\phi)>1$. As a consequence when $\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, $\lambda(\phi)>1$, the isometry $\phi_$ preserves a unique geodesic line $\mathrm{Ax}(\phi)\subset\mathbb{H}^\infty$ called the axis of $\phi$. This line is the intersection of $\mathbb{H}^\infty$ with a plane $P_\phi\subset\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ which intersects the isotropic cone of $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ in two lines $\mathbb{R}v^+_{\phi_}$ and $\mathbb{R}v^-_{\phi_}$ such that $$\phi_(p)=\lambda(\phi)^{\pm 1}p$$ for all $p\in\mathbb{R}v^\pm_{\phi_}$ (the lines $\mathbb{R}v^+_{\phi_}$ and $\mathbb{R}v^-_{\phi_}$ correspond to $\omega(\phi)$ and $\alpha(\phi)$ with the notations of §\[subsec:isometries\]). Take $\alpha\in\mathbb{R}v^-_{\phi_}$ and $\omega\in\mathbb{R}v^+_{\phi_}$ normalized so that $\langle\alpha,\,\omega\rangle=1$. The point $p=\frac{\alpha+\omega}{\sqrt{2}}$ lies on $\mathrm{Ax}(\phi)$. Since $\phi_(p)=\frac{\lambda(\phi)^{-1}\alpha+\lambda(\phi)\omega}{\sqrt{2}}$ one obtains $$\begin{aligned} \exp(L(\phi_))+\frac{1}{\exp(L(\phi_))}&=& 2\mathrm{cosh}(d(p,\phi_(p)))\\ &=& 2\langle p,\,\phi_(p)\rangle\\ &=&\lambda(\phi)+\frac{1}{\lambda(\phi)}.\end{aligned}$$ The translation length is thus equal to $\log\lambda(\phi)$. Consequently $\lambda(\phi)$ does not depend on the polarization and is invariant under conjugacy. There is a correspondence between the dynamical behavior of a birational map $\phi$ of $S$, in particular its degree, and the type of the induced isometry on $\mathbb{H}^\infty$: \[thm:dilfav\] Let $S$ be a smooth projective complex surface with a fixed polarization $H$. Let $\phi\colon S\dashrightarrow S$ be a birational map. Then one of the following holds: - $\phi$ is elliptic, $(\deg_H\phi^n)_n$ is bounded and $\phi$ is virtually isotopic to the identity; - $\phi$ is parabolic and either $\deg_H\phi^n\sim cn$ for some positive constant $c$ and $\phi$ is a <span style="font-variant:small-caps;">Jonquières</span> twist; or $\deg_H\phi^n\sim cn^2$ for some positive constant $c$ and $\phi$ is a <span style="font-variant:small-caps;">Halphen</span> twist; - $\phi$ is loxodromic and $\deg_H\phi^n=c\lambda(\phi)^n+O(1)$ for some positive constant $c$. <!-- --> - Any birational map of finite order is elliptic. Any automorphism of $\mathbb{P}^2_\mathbb{C}$ is elliptic. Any element of the group $$\big\{(z_0,z_1)\mapsto(\alpha z_0+P(z_1),\beta z_1+\gamma)\,\vert\,\alpha,\,\beta\in\mathbb{C}^,\,\gamma\in\mathbb{C},\,P\in\mathbb{C}[z_1]\big\}$$ is elliptic. - Any element of $\mathcal{J}$ of the form $$(z_0,z_1)\dashrightarrow\left(z_0,\frac{a(z_0)z_1+b(z_0)}{c(z_0)z_1+d(z_0)}\right)$$ with $\frac{(\mathrm{tr}\,M)^2}{\det M}\in\mathbb{C}(z_0)\smallsetminus\mathbb{C}$ where $$M=\left( \begin{array}{cc} a(z_0) & b(z_0)\\ c(z_0) & d(z_0) \end{array} \right)$$ is a <span style="font-variant:small-caps;">Jonquières</span> twist ([@CerveauDeserti:centralisateurs]). - Consider the family of birational self maps of $\mathbb{P}^2_\mathbb{C})$ given in the affine chart $z_2=1$ by ([@DillerFavre]) $$\phi_\varepsilon\colon(z_0,z_1)\dashrightarrow\left(z_1+1-\varepsilon,z_0\frac{z_1-\varepsilon}{z_1+1}\right).$$ If - $\varepsilon=-1$, then $\phi_\varepsilon$ is elliptic; - $\varepsilon\in\{0,\,1\}$, then $\phi_\varepsilon$ is a <span style="font-variant:small-caps;">Jonquières</span> twist; - $\varepsilon\in \left\{\frac{1}{2},\,\frac{1}{3}\right\}$, then $\phi_\varepsilon$ is a <span style="font-variant:small-caps;">Halphen</span> twist; - $\varepsilon\in\displaystyle\bigcup_{k\geq 4}\frac{1}{k}$, then $\phi_\varepsilon$ is loxodromic. - If $\phi\colon(z_0,z_1)\dashrightarrow(z_1,z_0+z_1^2)$, then $\deg(\phi^n)=(\deg\phi)^n=2^n$. If $\psi\colon(z_0,z_1)\dashrightarrow(z_0^2z_1,z_0z_1)$, then $\deg\psi^n\sim\left(\frac{3+\sqrt{5}}{2}\right)^n$; in particular $\deg(\psi^n)\not=(\deg\psi)^n$. - Let us finish with a more geometric example. Consider the elliptic curve $E=\faktor{\mathbb{C}}{\mathbb{Z}[\mathbf{i}]}$. The linear action of the group $\mathrm{GL}(2,\mathbb{Z}[\mathbf{i}])$ on the complex plane preserves the lattice $\mathbb{Z}[\mathbf{i}]\times\mathbb{Z}[\mathbf{i}]$. This yields to an action of $\mathrm{GL}(2,\mathbb{Z}[\mathbf{i}])$ by regular automorphisms on the abelian surface $S=E\times E$. Since this action commutes with $(z_0,z_1)\mapsto(\mathbf{i}z_0,\mathbf{i}z_1)$ one gets a morphism from $\mathrm{PGL}(2,\mathbb{Z}[\mathbf{i}])$ to $\mathrm{Aut}\left(\faktor{S}{(z_0,z_1)}\mapsto(\mathbf{i}z_0,\mathbf{i}z_1)\right)$. As $\faktor{S}{(z_0,z_1)}\mapsto(\mathbf{i}z_0,\mathbf{i}z_1)$ is rational one obtains an embedding of $\mathrm{PGL}(2,\mathbb{Z}[\mathbf{i}])$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. If we apply this construction to the map $(z_0,z_1)\mapsto(az_0+bz_1,cz_0+dz_1)$ we obtain a birational map $\psi$ of the complex projective plane such that $\lambda(\psi)$ is the largest eigenvalue of $\left( \begin{array}{cc} a & b \\ c & d \end{array} \right)$. More precisely if we start with $(z_0,z_1)\mapsto(z_0+z_1,z_1)$ one gets $\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ with $\deg\phi^n\sim n^2$. Any element virtually isotopic to the identity is regularizable, that is birationally conjugate to an automorphism. What can we say about two birational maps virtually isotopic to the identity ? We will see that if they commute they are simultaneously regularizable. Before proving it let us introduce a new notion. An element $\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is algebraically stable if $\deg\phi^n=(\deg\phi)^n$ for all $n\geq 0$. More generally if $S$ is a compact complex surface, then $\phi\in\mathrm{Bir}(S)$ is algebraically stable if $(\phi^)^n=(\phi^n)^$ for all $n\geq 0$. A geometric characterization of algebraically stable maps is the following: $\phi\in\mathrm{Bir}(S)$ is algebraically stable if and only if there is no curve $C\subset S$ such that $\phi^k(C)\in\mathrm{Ind}(\phi)$ for some integer $k$. <span style="font-variant:small-caps;">Diller</span> and <span style="font-variant:small-caps;">Favre</span> proved the following result: \[pro:DillerFavre\] Let $S$ be a compact complex surface, and let $\phi$ be a birational self map of $S$. There exists a composition of finitely many point blow-ups that lifts $\phi$ to an algebraically stable map. Before giving the proof, let us give its idea. Assume that $\phi$ is not algebraically stable. In other words there exist a curve $C\subset S$ and an integer $k$ such that $C$ is blown down onto $p_1$ and $p_k=\phi^{k-1}(p_1)$ belongs to $\mathrm{Ind}(\phi)$. The idea of <span style="font-variant:small-caps;">Diller</span> and <span style="font-variant:small-caps;">Favre</span> to get an algebraically stable map is the following: after blowing up the points $p_i=\phi^i(p_1)$, $i=1$, $\ldots$, $k$, the orbit of $C$ consists of curves. Doing this for any element of $\mathrm{Exc}(\phi)$ whose an iterate belongs to $\mathrm{Ind}(\phi)$ one gets the statement (note that the cardinal of $\mathrm{Exc}(\phi)$ is finite so the process ends). Let us write $\phi$ as follows $\phi=\phi_n\circ\phi_{n-1}\circ\ldots\circ\phi_1$ where - $\phi_i\colon S_{i-1}\to S_i$; - $S_0=S_n=S$; - and - either $\phi_i$ blows up a point $p_i=\mathrm{Ind}(\phi_i)\in S_i$, and we denote by $V_{i+1}=\mathrm{Exc}(\phi_i^{-1})\subset S_{i+1}$ the exceptional divisor of $\phi_i^{-1}$; - or $\phi_i$ blows down the exceptional divisor $E_i\subset S_i$; in this case we set $q_{i+1}:=\phi_i(E_i)\in S_{i+1}$. For any $j\in\mathbb{N}$ set $S_j:=S_{\text{$j$ mod $n$}}$ and $\phi_j:=\phi_{\text{$j$ mod $n$}}$. Assume that $\phi$ is not algebraically stable. Then there exist integers $1\leq i\leq N$ such that $\phi_i$ blows down $E_i$ and $$\phi_{N-1}\circ\phi_{N-2}\circ\ldots\circ\phi_i(E_i)=p_N\in\mathrm{Ind}(\phi_N).$$ Choosing a pair $(i,N)$ of minimal length we can assume that for all $i<j\leq N$ $$m_j:=\phi_j\circ\phi_{j-1}\circ\ldots\circ\phi_i(E_i)=\phi_j\circ\phi_{j-1}\circ\ldots\circ\phi_{i+1}(q_{i+1})$$ does not belong to $\mathrm{Ind}(\phi_i)\cup\mathrm{Exc}(\phi_i)$. First blow up $S_N$ at $m_N=p_N$. Then - $\phi_N$ lifts to a biholomorphism $\widehat{\phi}_N$; - $\widehat{\phi}_{N-1}$ either blows up the two distinct points $m_{N-1}$ and $p_{N-1}$ or blows up $m_{N-1}$ and blows down $E_{N-1}\notin m_{N-1}$; - $\sum\mathrm{Card}\big(\phi_j\big(\mathrm{Exc}(\phi_j)\big)\big)=\sum\mathrm{Card}\big(\widehat{\phi}_j\big(\mathrm{Exc}(\widehat{\phi}_j)\big)\big)$. Remark that modifying $S_N$ means modifying $S_{N+n}$, $S_{N-n}$, $\ldots$ nevertheless blowing up a point $m_j$ does not interfere with the behavior of the map $\phi_j$ around $m_{N+n}$, $m_{N-n}$, $\ldots$ (indeed if $j_1=j_2$ mod $n$ but $j_1\not=j_2$ then the points $m_{j_1}$, $m_{j_2}$ of $S_1=S_2$ are distinct), and these points can be blow up independently. Similarly blow up $m_{N-1}$, $m_{N-2}$, $\ldots$, $m_{i+2}$. At each step $\sum\mathrm{Card}(\phi_j(\mathrm{Exc}(\phi_j)))$ remains constant. Let us finish by blowing up $m_{i+1}=\phi_i(E_i)$; the situation is then different: $\phi_i$ becomes a biholomorphism $\widehat{\phi}_i$. The number of components of $\mathrm{Exc}(\phi_i)$ thus reduces from $1$ to $0$. As a consequence $$\label{eq:dim} \sum\mathrm{Card}\big(\widehat{\phi}_j\big(\mathrm{Exc}(\widehat{\phi}_j)\big)\big)= \sum\mathrm{Card}\big(\phi_j\big(\mathrm{Exc}(\phi_j)\big)\big)-1.$$ Repeating finitely many times the above argument either we produce an algebraically stable map $\widehat{\phi}=\widehat{\phi}_N\circ\widehat{\phi}_{N-1}\circ \ldots\circ\widehat{\phi}_1$, or thanks to (\[eq:dim\]) we eleminate all exceptional components of the factors of $\phi$. In both cases we get an algebraically stable map. \[lem:commas\] Let $\phi$, $\psi$ be two birational self maps of a compact complex surface $S$. Assume that $\phi$ and $\psi$ are both virtually isotopic to the identity. Assume that $\phi$ and $\psi$ commute. There exist a surface $Y$ and a birational map $\zeta\colon\ Y\dashrightarrow S$ such that - $\zeta^{-1}\circ\phi^\ell\circ\zeta\in\mathrm{Aut}(Y)^0$ for some integer $\ell$, - $\zeta^{-1}\circ\psi\circ\zeta$ is algebraically stable. Since $\phi$ is virtually isotopic to the identity we can assume that up to birational conjugacy and finite index $\phi$ is an automorphism of $S$. Let $N(\psi)$ be the minimal number of blow-ups needed to make $\psi$ algebraically stable (such a $N(\psi)$ exists according to Proposition \[pro:DillerFavre\]). If $N(\psi)=0$, then $\zeta=\mathrm{id}$ suits. Assume that Lemma \[lem:commas\] holds when $N(\psi)\leq j$. Consider a pair $(\phi,\psi)$ of birational self maps of $S$ such that - $\phi$ and $\psi$ are both virtually isotopic to the identity, - $\phi$ and $\psi$ commute, - $N(\psi)=j+1$. Since $\psi$ is not algebraically stable there exists a curve $C$ blown down by $\psi$ and such that $\psi^q(C)$ is a point of indeterminacy $p$ of $\psi$ for some integer $q$. The maps $\psi$ and $\phi$ commute so an iterate $\phi^k$ of $\phi$ fixes the irreducible components of $\mathrm{Ind}(\psi)$. Let us blow up $p$ via $\pi$. On the one hand $\pi^{-1}\circ\phi^k\circ\pi$ is an automorphism because $p$ is fixed by $\phi^k$ and on the other hand $N(\pi^{-1}\circ\psi\circ\pi)=j$. One can thus conclude by induction that there exist a surface $Y$ and a birational map $\zeta\colon Y\dashrightarrow S$ such that $\zeta^{-1}\circ\phi^\ell\circ\zeta\in\mathrm{Aut}(Y)^0$ for some integer $\ell$ and $\zeta^{-1}\circ\psi\circ\zeta$ is algebraically stable. \[pro:commas\] Let $\phi$, $\psi$ be two birational self maps of a surface $S$. Assume that $\phi$ and $\psi$ are both virtually isotopic to the identity. Assume that $\phi$ and $\psi$ commute. Then there exist a surface $Z$ and a birational map $\pi\colon Z\dashrightarrow S$ such that - $\pi^{-1}\circ\phi\circ\pi$ and $\pi^{-1}\circ\psi\circ\pi$ belong to $\mathrm{Aut}(Z)$; - $\pi^{-1}\circ\phi^k\circ\pi$ and $\pi^{-1}\circ\psi^k\circ\pi$ belong to $\mathrm{Aut}(Z)^0$ for some integer $k$. By assumption there exist a surface $\widetilde{S}$, a birational map $\eta\colon\widetilde{S}\dashrightarrow S$ and an integer $n$ such that $\eta^{-1}\circ\phi\circ\eta$ belongs to $\mathrm{Aut}(S)$ and $\eta^{-1}\circ\phi^n\circ\eta$ belongs to $\mathrm{Aut}(S)^0$. Let us now work on $\widetilde{S}$; to simplify denote by $\phi$ the automorphism $\eta^{-1}\circ\phi^n\circ\eta$ and by $\psi$ the birational map $\eta^{-1}\circ\psi\circ\eta$. According to Lemma \[lem:commas\] there exist a surface $Y$, a birational map $\upsilon\colon Y\dashrightarrow S$ and an integer $\ell$ such that $\zeta^{-1}\circ\widetilde{\phi}^\ell\circ\zeta$ belongs to $\mathrm{Aut}(Y)^0$ and $\zeta^{-1}\circ\widetilde{\psi}\circ\zeta$ is algebraically stable. Set $\overline{\phi}=\zeta^{-1}\circ\widetilde{\phi}^i\circ\zeta$ and $\overline{\psi}=\zeta^{-1}\circ\widetilde{\psi}\circ\zeta$. To get an automorphism from $\overline{\psi}$ let us blow down curves in $\mathrm{Exc}(\overline{\psi}^{-1})$. But curves blown down by $\overline{\psi}^{-1}$ are of self-intersection $<0$ and $\overline{\phi}$ fixes such curves since $\overline{\phi}$ is isotopic to the identity. We conclude by using the fact that $\mathrm{Card}\big(\mathrm{Exc}(\overline{\psi}^{-1})\big)$ is finite. Algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ {#Chapter:algebraicsubgroup} ============================================================== The first section of this chapter deals with the algebraic structure of the $n$-dimensional <span style="font-variant:small-caps;">Cremona</span> group, the fact that it is not an algebraic group of infinite dimension if $n\geq 2$, the obstruction to this, which is of a topological nature. By contrast, the existence of a Euclidean topology on the <span style="font-variant:small-caps;">Cremona</span> group which extends that of its classical subgroups and makes it a topological group is recalled. More precisely in [@Shafarevich] <span style="font-variant:small-caps;">Shafarevich</span> asked [10cm]{} “Can one introduce a universal structure of an infinite dimensional group in the group of all automorphisms (resp. all birational automorphisms) of arbitrary algebraic variety ?” We will see that the answer to this question is no ([@BlancFurter]). For any algebraic variety $V$ defined over $\mathbb{C}$ there is a natural notion of families of elements of $\mathrm{Bir}(V)$. These are maps $V(\mathbb{C})\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ compatible with the structures of algebraic varieties. Note that $\mathrm{Bir}(\mathbb{P}^1_\mathbb{C})\simeq\mathrm{PGL}(2,\mathbb{C})$ and families $V\dashrightarrow\mathrm{Bir}(\mathbb{P}^1_\mathbb{C})$ correspond to morphisms of algebraic varieties. If $n\geq 2$ the set $\mathrm{Bir}_d(\mathbb{P}^n_\mathbb{C})$ of all birational maps of $\mathbb{P}^n_\mathbb{C}$ of degree $d$ has the structure of an algebraic variety defined over $\mathbb{C}$ such that the families $V\to\mathrm{Bir}_d(\mathbb{P}^n_\mathbb{C})$ correspond to morphisms of algebraic varieties ([@BlancFurter]). The group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ thus decomposes into a disjoint infinite union of algebraic varieties, having unbounded dimension. <span style="font-variant:small-caps;">Blanc</span> and <span style="font-variant:small-caps;">Furter</span> established the following statement: \[thm:bf\] Let $n\geq 2$. There is no structure of algebraic variety of infinite dimension on $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ such that families $V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ would correspond to morphisms of algebraic varieties. The bad structures come from the degeneration of maps of degree $d$ into maps of smaller degree. A family of birational self maps of $\mathbb{P}^2_\mathbb{C}$ of degree $d$ which depends on a parameter $t$ may degenerate for certain values of $t$ onto a non-reduced expression of the type $P\,\mathrm{id}$ where $P$ denotes an homogeneous polynomial of degree $d-1$. Consider for instance the family $$\phi_{a,b,c}\colon(z_0:z_1:z_2)\dashrightarrow\Big(z_0\big(az_2^2+cz_0z_2+bz_0^2\big):z_1\big(az_2^2+(b+c)z_0z_2+(a+b)z_0^2\big):z_2\big(az_2^2+cz_0z_2+bz_0^2\big)\Big)$$ parameterized by the nodal plane cubic $a^3+b^3=abc$. The family $(\phi_{a,b,c})$ is globally defined by formulas of degree $3$ but each element $\phi_{a,b,c}$ has degree $\leq 2$ and there is no global parameterization by homogeneous formulas of degree $2$. In fact the obstruction to a positive answer to <span style="font-variant:small-caps;">Shafarevich</span> question comes only from the topology: \[thm:bf2\] There is no $\mathbb{C}$-algebraic variety of infinite dimension that is homeomorphic to $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. In $2010$ in the question session of the workshop “Subgroups of the <span style="font-variant:small-caps;">Cremona</span> group” in Edinburgh, <span style="font-variant:small-caps;">Serre</span> asked the following question [10cm]{} “Is it possible to introduce such topology on $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that is compatible with $\mathrm{PGL}(3,\mathbb{C})$ and $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$ ?” We will see that <span style="font-variant:small-caps;">Blanc</span> and <span style="font-variant:small-caps;">Furter</span> gave a positive answer to this question: \[thm:bf3\] Let $n\geq 1$ be an integer. There is a natural topology on $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$, called the Euclidean topology, such that: - $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ endowed with the Euclidean topology is a <span style="font-variant:small-caps;">Hausdorff</span> topological group, - the restriction of the Euclidean topology to algebraic subgroups in particular to $\mathrm{PGL}(n+1,\mathbb{C})$ and $\mathrm{PGL}(2,\mathbb{C})^n$ is the classical Euclidean topology. In the literature an algebraic subgroup $\mathrm{G}$ of $\mathrm{Bir}(V)$ corresponds to taking an algebraic group $\mathrm{G}$ and a morphism $\mathrm{G}\to\mathrm{Bir}(V)$ that is a group homomorphism and whose schematic kernel is trivial. We will see that in the case of $V=\mathbb{P}^n_\mathbb{C}$ one can give a more intrinsic definition (Corollary \[cor:agree\]) which corresponds to taking closed subgroups of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ of bounded degree and that these two definitions agree (Lemma \[lem:agree\]). An element $\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is algebraic if it is contained in an algebraic subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. It is equivalent to say that the sequence $(\deg\phi^n)_{n\in\mathbb{N}}$ is bounded. According to [@BlancFurter] the group $\mathrm{G}$ is thus an affine algebraic group. As a consequence $\phi$ decomposes as $\phi=\phi_s\circ\phi_u$ where $\phi_s$ is a semi-simple element of $\mathrm{G}$ and $\phi_u$ an unipotent element of $\mathrm{G}$. This decomposition does not depend on $\mathrm{G}$ (see [@Popov]). In particular there is a natural notion of semi-simple and unipotent elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. As we will see $\mathrm{G}$ could even by chosen to be the abelian algebraic subgroup $\overline{\big\{\phi^i\,\vert\,i\in\mathbb{Z}\big\}}$ of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. In all linear algebraic groups the set of unipotent elements is closed; <span style="font-variant:small-caps;">Popov</span> asked if it is the case in the context of the <span style="font-variant:small-caps;">Cremona</span> group. A natural and related question is the following one: is the set $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\text{alg}}$ of algebraic elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ closed ? The second section deals with the answers to these questions (Theorem \[thm:Blancelalg\]). In the third section the classification of maximal algebraic subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is given. In the fourth section we give a sketch of the proof of the fact that $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is topologically simple when endowed with the <span style="font-variant:small-caps;">Zariski</span> topology, [i.e.]{} it does not contain any non-trivial closed normal strict subgroup. The main ingredients of the proof are some clever deformation arguments. The fifth section is devoted to a modern proof of the regularization theorem of <span style="font-variant:small-caps;">Weil</span> which says that for every rational action $\rho$ of an algebraic group $\mathrm{G}$ on a variety $X$ there exist a variety $Y$ with a regular action $\mu$ of $\mathrm{G}$ and a dominant rational map $X\dashrightarrow Y$ with the following properties: for any $(g,p)\in\mathrm{G}\times X$ such that - $\rho$ is defined in $(g,p)$; - $\phi$ is defined in $p$ and $\rho(g,p)$; - $\mu$ is defined in $(g,\phi(p))$ we have $\phi(\rho(g,p))=\mu(g,\phi(p))$. Topologies and algebraic subgroups of the <span style="font-variant:small-caps;">Cremona</span> group {#sec:alg} ----------------------------------------------------------------------------------------------------- ### <span style="font-variant:small-caps;">Zariski</span> topology Take an irreducible algebraic variety $V$. A family of birational self maps of $\mathbb{P}^n_\mathbb{C}$ parameterized by $V$ is a birational self map $$\varphi\colon V\times\mathbb{P}^n_\mathbb{C}\to V\times\mathbb{P}^n_\mathbb{C}$$ such that - $\varphi$ determines an isomorphism between two open subsets $\mathcal{U}$ and $\mathcal{V}$ of $V\times\mathbb{P}^n_\mathbb{C}$ such that the first projection $\mathrm{pr}_1$ maps both $\mathcal{U}$ and $\mathcal{V}$ surjectively onto $V$, - $\varphi(v,x)=\big(v,\mathrm{pr}_2(\varphi(v,x))\big)$ where $\mathrm{pr}_2$ denotes the second projection; hence each $\varphi_v=\mathrm{pr}_2(\varphi(v,\cdot))$ is a birational self map of $\mathbb{P}^n_\mathbb{C}$. The map $v\mapsto \varphi_v$ is called a morphism from the parameter space $V$ to $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. A subset $S\subset\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is closed if for any algebraic variety $V$ and any morphism $V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ its preimage is closed. This yields to a topology on $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ called the <span style="font-variant:small-caps;">Zariski</span> topology. A subset $F\subset\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is closed if for any algebraic variety $V$ and any morphism $V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ the preimage of $F$ is closed. \[rem:homeo\] For any $\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ the maps from $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ into itself given by $$\begin{aligned} & \psi\mapsto\psi^k, &&\psi\mapsto \psi\circ\varphi, && \psi\mapsto \varphi\circ\psi, && \psi\mapsto\psi^{-1} \end{aligned}$$ are homeomorphisms of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ with respect to the <span style="font-variant:small-caps;">Zariski</span> topology. Indeed let $V$ be an irreducible algebraic variety. If $f$, $g\colon V\times\mathbb{P}^n_\mathbb{C}\to V\times\mathbb{P}^n_\mathbb{C}$ are two $V$-birational maps inducing morphisms $V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$, then $f\circ g$ and $f^{-1}$ are again $V$-birational maps that induce morphisms $V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. The map $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ given by $\psi\mapsto\psi^k$ is thus continuous. Let $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ (resp. $\mathrm{Bir}_d(\mathbb{P}^n_\mathbb{C})$) be the set of elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ of degree $\leq d$ (resp. of degree $d$); we have the following increasing sequence $$\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})=\mathrm{Bir}_{\leq 1}(\mathbb{P}^n_\mathbb{C})\subseteq\mathrm{Bir}_{\leq 2}(\mathbb{P}^n_\mathbb{C})\subseteq\mathrm{Bir}_{\leq 3}(\mathbb{P}^n_\mathbb{C})\subseteq\ldots$$ whose union gives the <span style="font-variant:small-caps;">Cremona</span> group. We will see that $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ is closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ and the topology of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is the inductive topology induced by the above sequence. As a result it suffices to describe the topology of $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ to understand the topology of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Take a positive integer $d$. Let $W_d$ be the set of equivalence classes of non-zero $(n+1)$-uples $(\phi_0,\phi_1,\ldots,\phi_n)$ of homogeneous polynomials $\phi_i\in\mathbb{C}[z_0,z_1,\ldots,z_n]$ of degree $d$ where $(\phi_0,\phi_1,\ldots,\phi_n)$ is equivalent to $(\lambda\phi_0,\lambda\phi_1,\ldots,\lambda\phi_n)$ for any $\lambda\in\mathbb{C}^$. We denote by $(\phi_0:\phi_1:\ldots:\phi_n)$ the equivalence class of $(\phi_0,\phi_1,\ldots,\phi_n)$. Let $H_d\subseteq W_d$ be the set of elements $\phi=(\phi_0:\phi_1:\ldots:\phi_n)\in W_d$ such that the rational map $\psi_\phi\colon\mathbb{P}^n_\mathbb{C}\dashrightarrow\mathbb{P}^n_\mathbb{C}$ given by $$(z_0:z_1:\ldots:z_n)\dashrightarrow \big(\phi_0(z_0,z_1,\ldots,z_n):\phi_1(z_0,z_1,\ldots,z_n):\ldots:\phi_n(z_0,z_1,\ldots,z_n)\big)$$ is birational. The map $$\begin{aligned} &H_d\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C}) && \phi\mapsto\psi_\phi\end{aligned}$$ is denoted by $\pi_d$. \[lem:propbf\] The following properties hold: - The set $W_d$ is isomorphic to $\mathbb{P}^k_\mathbb{C}$ where $k=(n+1)\binom{d+n}{d}-1$. - The set $H_d$ is locally closed in $W_d$ and thus inherits from $W_d$ the structure of an algebraic variety. - The map $\pi_d\colon H_d\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is a morphism, and $\pi_d(H_d)$ is the set $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$. - For all $\phi$ in $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ the set $\pi_d^{-1}(\phi)$ is closed in $W_d$ so in $H_d$. - If $S\subset H_\ell$ ($\ell\geq 1$) is closed, then $\pi_d^{-1}(\pi_\ell(S))$ is closed in $H_d$. Hence $W_d$ and $H_d$ are naturally algebraic varieties; we will see that $\mathrm{Bir}_d(\mathbb{P}^n_\mathbb{C})$ also but not $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$. - The set of homogeneous polynomials of degree $d$ in $(n+1)$ variables is a $\mathbb{C}$-vector space of dimension $\binom{d+n}{d}$; this implies the first assertion. - Denote by $Y\subseteq W_{d^{n-1}}\times W_d$ the set defined by $$\big\{(\varphi,\phi)\in W_{d^{n-1}}\times W_d\,\vert\, \varphi\circ \phi=P\,\mathrm{id}\text{ for some $P\in\mathbb{C}[z_0,z_1,\ldots,z_n]_d$}\big\}.$$ If $P$ is nonzero, then the rational maps $\psi_\phi$ and $\psi_\varphi$ are birational and inverses of each other. If $P$ is zero, then $\psi_\phi$ contracts the entire set $\mathbb{P}^n_\mathbb{C}$ onto a strict subvariety included in the set $\big\{\varphi_1=\varphi_2=\ldots=\varphi_n=0\big\}$. In particular for any pair $(\varphi,\phi)$ of $Y$ the rational map $\psi_\phi$ is birational if and only if its Jacobian is nonzero. As a consequence any element $\phi\in H_d$ corresponds to at least one pair $(\varphi,\phi)$ in $Y$ (indeed according to [@BassConnelWright] the inverse of a birational self map of $\mathbb{P}^n_\mathbb{C}$ of degree $d$ has degree $\leq d^{n-1}$). Hence $Y$ is closed in $W_{d^{n-1}}\times W_d$. The image $\mathrm{pr}_2(Y)$ of $Y$ by the second projection $\mathrm{pr}_2$ is closed in $W_d$ since $W_{d^{n-1}}$ is a complete variety and $\mathrm{pr}_2$ a <span style="font-variant:small-caps;">Zariski</span> closed morphism. One can write $H_d$ as $\mathcal{U}\cap\mathrm{pr}_2(Y)$ where $\mathcal{U}\subseteq W_d$ is the open set of elements having a nonzero Jacobian. As a result $H_d$ is locally closed in $W_d$ and closed in $\mathcal{U}$. - Consider the $H_d$-rational map $\phi$ defined by $$\begin{aligned} & \phi\colon H_d\times\mathbb{P}^n_\mathbb{C}\dashrightarrow H_d\times\mathbb{P}^n_\mathbb{C}&& (\varphi,z)\dashrightarrow(\varphi,\varphi(z)).\end{aligned}$$ Set $J=\det\left(\left(\frac{\partial \varphi_i}{\partial x_j}\right)_{0\leq i,\,j\leq n}\right)$. Let $\mathcal{V}\subset H_d\times\mathbb{P}^n_\mathbb{C}$ be the open set where $J$ is not zero. The restriction $f_{\vert\mathcal{V}}$ of $f$ to $\mathcal{V}$ is an open immersion. It then follows from the construction of $H_d$ that the image of $\pi_d$ is $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$. - Let $\phi$ be an element in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\leq d}$. It corresponds to a birational self map $\psi_\phi$ of $\mathbb{P}^n_\mathbb{C}$ given by $$\begin{aligned} \psi_\phi\colon (z_0:z_1:\ldots:z_n)\dashrightarrow(\phi_0(z_0,z_1,\ldots,z_n):\phi_1(z_0,z_1,\ldots,z_n):\ldots:\phi_n(z_0,z_1,\ldots,z_n))\end{aligned}$$ for some homogeneous polynomials of degree $k\leq d$ having no common divisor. Then $$(\pi_d)^{-1}(\psi_\phi)=\big\{(\varphi_0:\varphi_1:\ldots:\varphi_n)\in W_d\,\vert\, \varphi_i\phi_j=\varphi_j\phi_i\quad\forall\, 1\leq i<j\leq n\big\}\subset H_d.$$ This set is thus closed in $W_d$ and so in $H_d$. - If $\ell$ is a positive integer and $F$ a closed subset of $H_\ell$, then we denote by $Y_F$ the subset of $Y\times\overline{F}$ (where $Y\subset W_{d^{n-1}}\times W_d$ is as above and $\overline{F}$ is the closure of $F$ in $W_\ell$) given by $$Y_F=\big\{((\zeta,\phi),\varphi)\,\vert\, \text{$\phi$ and $\varphi$ yield the same map $\mathbb{P}^n_\mathbb{C}\dashrightarrow\mathbb{P}^n_\mathbb{C}$}\big\}.$$ In other words $$Y_F=\big\{((\zeta,\phi),\varphi)\,\vert\, \phi_i\varphi_j=\phi_j\varphi_i\quad \forall\, i,\,j\big\}.$$ Hence $Y_F$ is closed in $Y\times\overline{F}$ and also in $W_{d^{n-1}}\times W_d\times W_\ell$. The subset $\mathrm{pr}_2(Y_F)$ of $W_d$ is closed in $W_d$ and so in $\mathrm{pr}_2(Y)$; as a result $\mathrm{pr}_2(Y_F)\cap\mathcal{U}$ is closed in $\mathrm{pr}_2(Y)\cap\mathcal{U}$. We conclude using the fact that $\mathrm{pr}_2(Y_F)\cap\mathcal{U}=(\pi_d)^{-1}(\pi_\ell(F))$ and $\mathrm{pr}_2(Y)\cap\mathcal{U}=H_d$. \[lem:BlancFurter\] Let $V$ be an irreducible algebraic variety, and let $\upsilon\colon V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ be a morphism. There exists an open affine covering $(\mathcal{V}_i)_{i\in I}$ of $V$ such that for each $i$ there exist an integer $d_i$ and a morphism $\upsilon_i\colon \mathcal{V}_i\to H_{d_i}$ such that $\upsilon_{\vert \mathcal{V}_i}=\pi_{d_i}\circ\upsilon_i$. Consider a morphism $\tau\colon V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ given by a $V$-birational map $$\phi\colon V\times\mathbb{P}^n_\mathbb{C}\dashrightarrow V\times\mathbb{P}^n_\mathbb{C}$$ which restricts to an open immersion on an open set $\mathcal{U}$. Take a point $p_0$ in $V$. Let $\mathcal{V}_0\subset V$ be an open affine set containing $p_0$. Take an element $w_0=(p_0,y)$ of $\mathcal{U}$. Let us fix homogeneous coordinates $(z_0:z_1:\ldots:z_n)$ on $\mathbb{P}^n_\mathbb{C}$ such that - $y=(1:0:0:\ldots:0)$, - $\phi(w_0)$ does not belong to the plane $z_0=0$. Let us denote by $\mathbb{A}^n_\mathbb{C}\subset\mathbb{P}^n_\mathbb{C}$ the affine set where $z_0=1$; $$\begin{aligned} &x_1=\frac{z_1}{z_0}&&x_2=\frac{z_2}{z_0} &&\ldots && x_n=\frac{z_n}{z_0} \end{aligned}$$ are natural affine coordinates of $\mathbb{A}^n_\mathbb{C}$. The map $\phi$ restricts to a rational map of $\mathcal{V}_0\times\mathbb{P}^n_\mathbb{C}$ defined at $w_0$. Its composition with the projection on the $i$-th coordinate is a rational function on $\mathcal{V}_0\times\mathbb{A}^n_\mathbb{C}$ defined at $w_0$. Hence $\phi_{\vert \mathcal{V}_0\times\mathbb{A}^n_\mathbb{C}}$ can be written in a neighborhood of $w_0$ as $$(v,x_1,x_2,\ldots,x_n)\mapsto\left(\frac{P_1}{Q_1},\frac{P_2}{Q_2},\ldots,\frac{P_n}{Q_n}\right)$$ for some $P_i$, $Q_i$ in $\mathbb{C}[V][x_1,x_2,\ldots,x_n]$ such that none of the $Q_i$ vanish at $w_0$. As a result $\phi$ is given in a neighborhood of $w_0$ by $$\big(v,(z_0:z_1:\ldots:z_n)\big)\mapsto(P_0:P_1:\ldots:P_n)$$ where the $P_i\in\mathbb{C}[\mathcal{V}_0][z_0,z_1,\ldots,z_n]$ are homogeneous polynomials of the same degree $d_0$ such that not all vanish at $w_0$. Denote by $\mathcal{U}_0$ the set of points of $(V\times\mathbb{P}^n_\mathbb{C})\cap\mathcal{U}$ where at least one of the $P_i$ does not vanish; $\mathcal{U}_0$ is an open subset of $V\times\mathbb{P}^n_\mathbb{C}$. Its projection $\mathrm{pr}_1(\mathcal{U}_0)$ on $V$ is an open subset of $\mathcal{V}_0$ containing $p_0$. There thus exists an affine open subset $\widetilde{A_0}\subseteq\mathrm{pr}_1(\mathcal{U}_0)$ containing $p_0$. The $n$-uple $(P_0,P_1,\ldots,P_n)$ yields to a morphism $\upsilon_0\colon\widetilde{A_0}\to H_d$. By construction $\upsilon_{\vert \widetilde{A_0}}=\pi_d\circ\upsilon_0$. If we repeat this process for any point of $V$ we get an affine covering. This statement implies the following one: \[cor:BlancFurter\] - A set $S\subseteq\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is closed if and only if $\pi_d^{-1}(S)$ is closed in $H_d$ for any $d\geq 1$. - For any $d$,the set $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ is closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. - For any $d$, the map $\pi_d\colon H_d\to\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ is surjective, continuous and closed. In particular it is a topological quotient map. Let us prove the first assertion. Assume that $S$ is closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Recall that a subset of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ if and only if its preimage by any morphism is closed. Since any $\pi_d\colon H_d\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is a morphism $\pi_d^{-1}(S)$ is thus closed in $H_d$. Conversely suppose that $\pi_d^{-1}(S)$ is closed in $H_d$ for any $d$. Let $V$ be an irreducible algebraic variety, and let $\upsilon\colon V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ be a morphism. According to Lemma \[lem:BlancFurter\] there exists an open affine covering $(\mathcal{V}_i)_{i\in I}$ of $V$ such that for any $i$ there exist - an integer $d_i$, - a morphism $\upsilon_i\colon \mathcal{V}_i\to H_{d_i}$ with $\upsilon_{\vert \mathcal{V}_i}=\pi_{d_i}\circ\upsilon_i$. As $\pi_{d_i}^{-1}(S)$ is closed and $\upsilon^{-1}(S)\cap \mathcal{V}_i=\upsilon_i^{-1}(\pi_{d_i}^{-1}(F))$ one gets that $\upsilon^{-1}(S)\cap~\mathcal{V}_i$ is closed in $\mathcal{V}_i$ for any $i$. As a result $\upsilon^{-1}(S)$ is closed. We will now prove the second assertion. According to the first assertion it suffices to prove that $$\pi_\ell^{-1}\big(\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\big)=\pi_\ell^{-1}(\pi_d(H_d))$$ is closed in $H_\ell$ for any $\ell$. It follows from Lemma \[lem:propbf\]. Finally let us prove the third assertion. The surjectivity follows from the construction of $H_d$ and $\pi_d$ (see* [@BlancFurter]). Since $\pi_d$ is a morphism, $\pi_d$ is continuous. Let $S\subseteq H_d$ be a closed subset. According to Lemma \[lem:propbf\] the set $\pi_\ell^{-1}(\pi_d(S))$ is closed in $H_\ell$ for any $\ell$. The first assertion allows to conclude. The first and third assertions of Corollary \[cor:BlancFurter\] imply: The topology of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is the inductive limit topology given by the <span style="font-variant:small-caps;">Zariski</span> topologies of $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$, $d\in\mathbb{N}$, which are the quotient topology of $$\pi_d\colon H_d\to\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$$ where $H_d$ is endowed with its <span style="font-variant:small-caps;">Zariski</span> topology. Theorem \[thm:bf\] holds for any field, Theorem \[thm:bf2\] holds for any algebraically closed field, and Theorem \[thm:bf3\] holds for (locally compact) local field. ### Algebraic subgroups An algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is a subgroup $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ which is the image of an algebraic group $\mathrm{H}$ by a homomorphism $\upsilon$ such that $\upsilon\colon\mathrm{H}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is a morphism with respect to the <span style="font-variant:small-caps;">Zariski</span> topology; in particular any algebraic group is contained in some $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$, [i.e.]{} any algebraic group has bounded degree. Corollary \[cor:agree\] allows to give a more intrinsic definition which corresponds to taking closed subgroups of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ of bounded degree. Lemma \[lem:agree\] shows that these two definitions agree. \[pro:agree\] Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Assume that - $\mathrm{G}$ is closed for the <span style="font-variant:small-caps;">Zariski</span> topology; - $\mathrm{G}$ is connected for the <span style="font-variant:small-caps;">Zariski</span> topology; - $\mathrm{G}\subset\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ for some integer $d$. If $d$ is choosing minimal, then the set $(\pi_d)^{-1}(\mathrm{G}\cap\mathrm{Bir}_d(\mathbb{P}^n_\mathbb{C}))$ is non empty. Let us denote by $\mathrm{H}$ the closure of $(\pi_d)^{-1}(\mathrm{G}\cap\mathrm{Bir}_d(\mathbb{P}^n_\mathbb{C}))$ in $H_d$. Then - $\pi_d$ induces a homeomorphism $\mathrm{H}\to\mathrm{G}$ for the <span style="font-variant:small-caps;">Zariski</span> topology; - if $V$ is an irreducible algebraic variety, the morphisms $V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ having image in $\mathrm{G}$ correspond, via $\pi_d$, to the morphisms of algebraic varieties $V\to\mathrm{H}$; - the liftings to $\mathrm{K}$ of the map $$\begin{aligned} & \mathrm{G}\times\mathrm{G}\to\mathrm{G},\, (\varphi,\psi) && \mathrm{G}\to\mathrm{G},\,\varphi\mapsto\varphi^{-1}\end{aligned}$$ give rise to morphisms of algebraic varieties $\mathrm{K}\times\mathrm{K}\to\mathrm{K}$ and $\mathrm{K}\to\mathrm{K}$. This gives $\mathrm{G}$ a unique structure of algebraic group. \[cor:agree\] Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Assume that $\mathrm{G}$ is - closed for the <span style="font-variant:small-caps;">Zariski</span> topology, - of bounded degree. Then there exist an algebraic group $\mathrm{H}$ together with a morphism $\mathrm{H}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ inducing a homeomorphism $\pi\colon\mathrm{H}\to\mathrm{G}$ such that: - $\pi$ is a group homomorphism - and for any irreducible algebraic variety $V$ the morphisms $V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ having their image in $\mathrm{G}$ correspond, via $\pi$, to the morphisms of algebraic varieties $V\to\mathrm{H}$. Let us first prove that $\mathrm{G}$ has a finite number of irreducible components. The group $\mathrm{G}$ is closed in $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ hence its pullback $(\pi_d)^{-1}(\mathrm{G})$ is also closed in $H_d$. It thus had a finite number of irreducible components $C_1$, $C_2$, $\ldots$, $C_r$. The sets $\pi_d(C_1)$, $\pi_d(C_2)$, $\ldots$, $\pi_d(C_r)$ are closed and irreducible and cover $\mathrm{G}$ (third assertion of Corollary \[cor:BlancFurter\]). If we keep only the maximal ones we get the irreducible components of $\mathrm{G}$. As for algebraic groups ([@Humphreys §7.3]) one can show that: - exactly one irreducible component of $\mathrm{G}$ passes through $\mathrm{id}$; - this irreducible component is a closed normal subgroup of finite index in $\mathrm{G}$ whose cosets are the connected as well as irreducible components of $\mathrm{G}$. Then Proposition \[pro:agree\] allows to conclude. \[lem:agree\] Let $\mathrm{A}$ be an algebraic group. Let $\rho\colon \mathrm{A}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ be a morphism which is also a group homomorphism. Then the image $\mathrm{G}$ of $\mathrm{A}$ is a closed subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ which has bounded degree. If $\pi\colon\mathrm{H}\to\mathrm{G}$ is the homeomorphism constructed in Corollary \[cor:agree\], then there exists a unique morphism of algebraic groups $\widetilde{\rho}\colon\mathrm{A}\to\mathrm{H}$ such that $\rho=\pi\circ\widetilde{\rho}$. Lemma \[lem:BlancFurter\] asserts that $\mathrm{G}=\rho(\mathrm{A})$ has bounded degree. The closure $\overline{\mathrm{G}}$ of $\mathrm{G}$ is a subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$; indeed inversion being a homeomorphism $\overline{\mathrm{G}}^{-1}=\overline{\mathrm{G}^{-1}}=\overline{\mathrm{G}}$. Similarly translation by $g\in\mathrm{G}$ is a homeomorphism thus $g\overline{\mathrm{H}}=\overline{g\mathrm{H}}=\overline{\mathrm{H}}$, that is $\mathrm{G}\overline{\mathrm{G}}\subset\overline{\mathrm{G}}$. In turn, if $g\in\overline{\mathrm{G}}$, then $\mathrm{G}g\subset\overline{\mathrm{G}}$ so $\overline{\mathrm{G}}g=\overline{\mathrm{G}g}\subset\overline{\mathrm{G}}$. As a result $\overline{\mathrm{G}}$ is a subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. According to Corollary \[cor:agree\] there exist a canonical homeomorphism $\mathrm{H}\to\overline{\mathrm{G}}$ where $\mathrm{H}$ is an algebraic group and a lift $\widetilde{\rho}\colon\mathrm{A}\to\mathrm{H}$ of the morphism $\rho\colon\mathrm{A}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ whose image is contained in $\overline{\mathrm{G}}$. As $\rho$ is a group homomorphism $\widetilde{\rho}$ is a morphism of algebraic groups. As a result $\mathrm{im}\,\rho$ is closed, so $\mathrm{im}\,\rho=\mathrm{H}$. Therefore $\overline{\mathrm{G}}=\mathrm{G}$. Any algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is affine. Let $\mathrm{G}$ be an algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. One can show that $\mathrm{G}$ is linear, and this reduces to the connected case. By the regularization theorem of <span style="font-variant:small-caps;">Weil</span> (see §\[sec:WeilKraft\]) the group $\mathrm{G}$ acts by automorphisms on some (smooth) rational variety $V$. Assume that $\alpha_V\colon V\to A(V)$ is the <span style="font-variant:small-caps;">Albanese</span> morphism. According to the <span style="font-variant:small-caps;">Nishi</span>-<span style="font-variant:small-caps;">Matsumura</span> theorem the induced action of $\mathrm{G}$ on $A(V)$ factors through a morphism $A(\mathrm{G})\to A(V)$ with finite kernel (see for instance [@Brion]). But $V$ is rational hence $A(V)$ is trivial and so does $A(\mathrm{G})$. The structure theorem of <span style="font-variant:small-caps;">Chevalley</span> asserts that $\mathrm{G}$ is affine. Let us finish by some examples: - The <span style="font-variant:small-caps;">Cremona</span> group in one variable $\mathrm{Bir}(\mathbb{P}^1_\mathbb{C})$ coincides with the group of linear projective transformations $\mathrm{PGL}(2,\mathbb{C})$; it is an algebraic group of dimension $3$. - In dimension $2$ the <span style="font-variant:small-caps;">Cremona</span> group contains the two following algebraic subgroups: - the group $\mathrm{PGL}(3,\mathbb{C})$ of automorphisms of $\mathbb{P}^2_\mathbb{C}$; - the group $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$ obtained as follows: the surface $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ can be considered as a smooth quadric in $\mathbb{P}^3_\mathbb{C}$ whose automorphism group contains $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$; by stereographic projection the quadric is birationally equivalent to $\mathbb{P}^2_\mathbb{C}$. Hence $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ also contains a copy of $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$. - More generally $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})=\mathrm{PGL}(n+1,\mathbb{C})$ is an algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ and $$\underbrace{\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})\times\ldots\times\mathrm{PGL}(2,\mathbb{C})}_\text{$n$ times}$$ is an algebraic subgroup of $$\mathrm{Aut}(\underbrace{\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}\times\ldots\mathbb{P}^1_\mathbb{C}}_\text{$n$ times})\subset\mathrm{Bir}(\mathbb{P}^n_\mathbb{C}).$$ - If $\mathrm{G}$ is a semi-simple algebraic group, $\mathrm{H}$ is a parabolic subgroup of $G$ and $V=\faktor{\mathrm{G}}{\mathrm{H}}$, then the homogeneous variety $V$ of dimension $n$ is rational; $\pi\circ\mathrm{G}\circ\pi^{-1}$ determines an algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ for any birational map $\pi\colon V\dashrightarrow\mathbb{P}^n_\mathbb{C}$. ### Euclidean topology We can put the Euclidean topology on a complex algebraic variety; this gives any algebraic group the structure of a topological group. Note that the Euclidean topology is finer than the <span style="font-variant:small-caps;">Zariski</span> one. Let $n\geq 1$ be an integer. The group $\mathrm{Bir}(\mathbb{P}^1_\mathbb{C})=\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})=\mathrm{PGL}(2,\mathbb{C})$ is obviously a topological group. Assume now that $n\geq 2$; we will - first define the Euclidean topology on $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ and show that the natural inclusion $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\hookrightarrow\mathrm{Bir}_{\leq d+1}(\mathbb{P}^n_\mathbb{C}) $ is a closed embedding; - second define the Euclidean topology on $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ as the inductive limit topology induced by those of $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$, that is a subset $F\subset\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is closed if and only if $F\cap\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ is closed in $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ for each $d$. Finally we will prove that $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ endowed with the Euclidean topology is a topological group; - third give some remarks and properties. ### The Euclidean topology on $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ Let us recall that $W_d$ is a projective space and $H_d$ is locally closed in $W_d$ for the <span style="font-variant:small-caps;">Zariski</span> topology (Lemma \[lem:propbf\]). Let us put the Euclidean topology on $W_d$: the distance between $(p_0:p_1:\ldots:p_n)$ and $(m_0:m_1:\ldots:m_n)$ is (see [@Weyl]) $$\frac{\displaystyle\sum_{i<j}\vert p_im_j-p_jm_i\vert^2}{\left(\displaystyle\sum_{i}\vert p_i\vert^2\right)\left(\displaystyle\sum_{i}\vert m_i\vert^2\right)}$$ We then put the induced topology on $H_d$. The behavior of the <span style="font-variant:small-caps;">Zariski</span> topology on $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ yields to: The Euclidean topology on $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ is the quotient topology induced by the surjective map $\pi_d\colon H_d\to \mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ where we put the Euclidean topology on $H_d$. Recall that if $f\colon X\to Y$ is a quotient map between topological spaces, $A$ is a subspace of $X$, $A$ is open and $A=f^{-1}(f(A))$, then the induced map $A\to f(A)$ is a quotient map ([@Bourbaki Chapter I, §3.6]). Set $$H_{d,d}=(\pi_d)^{-1}(\mathrm{Bir}_d(\mathbb{P}^n_\mathbb{C})).$$ But $(\pi_d)^{-1}(\mathrm{Bir}_{\leq d-1}(\mathbb{P}^n_\mathbb{C}))$ is closed in $H_d$ so $H_{d,d}$ is open in $H_d$ for the <span style="font-variant:small-caps;">Zariski</span> topology and hence also for the Euclidean topology; $\pi_d$ restricts to a homeomorphism $H_{d,d}\to\mathrm{Bir}_d(\mathbb{P}^n_\mathbb{C})$ for any $d\geq 1$. Let $d\geq 1$ be an integer. The spaces $W_d$ and $H_d$ are locally compact metric spaces endowed with the Euclidean topology. In particular the sets $W_d$, $H_d$ and $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ are sequential spaces: a subset $F$ is closed if the limit of every convergent sequence with values in $F$ belongs to $F$. The construction of the topology implies that $W_d$ and $H_d$ are metric spaces. As $W_d$ is compact and $H_d$ is locally closed in $W_d$ (Lemma \[lem:propbf\]) the set $H_d$ is locally compact. But metric spaces are sequential spaces and quotients of sequential spaces are sequential ([@Franklin]). A map $f\colon X\to Y$ between two topological spaces is proper if it is continuous and universally closed: for each topological space $Z$ the map $f\times\mathrm{id}_Z\colon X\times Z\to Y\times Z$ is closed. A topological space is locally compact if it is <span style="font-variant:small-caps;">Hausdorff</span> and if each of its points has a compact neighborhood. Let $d\geq 1$ be an integer. Then - the topological map $\pi_d\colon H_d\to \mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ is proper and closed; - the topological space $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ is locally compact and <span style="font-variant:small-caps;">Hausdorff</span>. \[lem:top\] Let $d\geq 0$ be an integer. The natural injection $$\iota_d\colon\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\to\mathrm{Bir}_{\leq d+1}(\mathbb{P}^n_\mathbb{C})$$ is a closed embedding, that is a homeomorphism onto its image which is closed in $\mathrm{Bir}_{\leq d+1}(\mathbb{P}^n_\mathbb{C})$. Consider the map $$\begin{aligned} & \widehat{\iota_d}\colon H_d\to H_{d+1},&& (f_0:f_1:\ldots:f_n)\mapsto (z_0f_0:z_0f_1:\ldots:z_0f_n).\end{aligned}$$ It is a morphism of algebraic varieties that is a closed immersion. As a result it is continuous and closed with respect to the Euclidean topology. The diagram $$\xymatrix{ H_d \ar[r]^{\widehat{\iota_d}} \ar[d]_{\pi_d} & H_{d+1}\ar[d]^{\pi_{d+1}} \\ \mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C}) \ar[r]_{\iota_d} & \mathrm{Bir}_{\leq d+1}(\mathbb{P}^n_\mathbb{C}) }$$ commutes. The continuity of $\widehat{\iota_d}$ implies the continuity of $\iota_d$: let $\mathcal{U}$ be an open subset of $\mathrm{Bir}_{\leq d+1}(\mathbb{P}^n_\mathbb{C})$; the equality $(\pi_d)^{-1}((\iota_d)^{-1}(\mathcal{U}))=(\pi_{d+1}\circ\widehat{\iota_d})^{-1}(\mathcal{U})$ shows that $(\pi_d)^{-1}((\iota_d)^{-1}(\mathcal{U}))$ is open in $H_d$, that is $(\iota_d)^{-1}(\mathcal{U})$ is open in $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$. ### The Euclidean topology on $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ Thanks to Lemma \[lem:top\] one can put on $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ the inductive limit topology given by the $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$: a subset of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is closed (resp. open) if and only if its intersection with any $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ is closed (resp. open). In particular the injections $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\hookrightarrow\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ are closed embeddings. This topology is called the Euclidean topology of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Let us now prove that $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is a topological group endowed with the Euclidean topology. \[lem:bfcont\] Let $d\geq 1$ be an integer. The map $$\begin{aligned} &\mathcal{I}_d\colon\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\to\mathrm{Bir}_{\leq d^{n-1}}(\mathbb{P}^n_\mathbb{C}), && \phi\mapsto\phi^{-1}\end{aligned}$$ is continuous. As in Lemma \[lem:propbf\] we consider the set $Y\subset W_{d^{n-1}}\times W_d$ defined by $$Y=\big\{(\varphi,\phi)\in W_{d^{n-1}}\times W_d\,\vert\, \varphi\circ \phi=P\,\mathrm{id}\text{ for some $P\in\mathbb{C}[z_0,z_1,\ldots,z_n]_d$}\big\}.$$ Let $\mathcal{U}\subset W_d$ (resp. $\mathcal{U}'\subset W_{d^{n-1}}$) be the set of elements having a nonzero Jacobian. The set $Y$ is closed in $W_{d^{n-1}}\times W_d$ (see the proof of Lemma \[lem:propbf\]) and the set $\mathcal{U}$ is open in $W_d$. As a consequence $$L=Y\cap(W_{d^{n-1}}\times\mathcal{U})=Y\cap(\mathcal{U}'\times\mathcal{U})$$ is locally closed in the algebraic variety $W_{d^{n-1}}\times W_d$. The projection on the first factor is a morphism $\eta_1\colon L\to H_{d^{n-1}}$ which is not surjective in general. The projection on the second factor induces a surjective morphism $\eta_2\colon L\to H_d$. By construction the diagram $$\xymatrix{ H_d \ar[d]_{\pi_d} & L\ar[l]_{\eta_2}\ar[r]^{\eta_1} & H_{d^{n-1}}\ar[d]^{\pi_{d-1}} \\ \mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C}) \ar[rr]_{\mathcal{I}_d} & &\mathrm{Bir}_{\leq d^{n-1}}(\mathbb{P}^n_\mathbb{C}) }$$ commutes. Let us prove that $\eta_2$ is a closed map for the Euclidean topology. The set $W_{d^{n-1}}$ is compact so the second projection $W_{d^{n-1}}\times W_d\to W_d$ is a closed map. Its restriction $\eta'_2\colon Y\to W_d$ to the closed subset $Y$ of $W_{d^{n-1}}\times W_d$ is a closed map. But $L=(\eta'_2)^{-1}(H_d)$ is a closed map for the Euclidean topology (indeed if $\varphi\colon A\to B$ is a continuous closed map between topological spaces and $C$ is any subset of $B$, then $\varphi$ induces a continuous closed map $\varphi^{-1}(C)\to C$). As the diagram is commutative for any $F\subset\mathrm{Bir}_{\leq d^{n-1}}(\mathbb{P}^n_\mathbb{C})$ we have $$\eta_2\big((\pi_{d^{n-1}}\circ\eta_1)^{-1}(F)\big)=(\mathcal{I}_d\circ\pi_d)^{-1}(F);$$ furthermore this set corresponds to elements $(\phi_0:\phi_1:\ldots:\phi_n)\in W_d$ such that the rational map $\psi_\phi$ is the inverse of an element of $F$. Assume that $F$ is closed in $\mathrm{Bir}_{\leq d^{n-1}}(\mathbb{P}^n_\mathbb{C})$. The maps $\eta_1$ and $\pi_{d^{n-1}}$ are continuous for the Euclidean topology hence $(\pi_{d^{n-1}}\circ\eta_1)^{-1}(F)$ is closed in $L$. As a result $\pi_d^{-1}(\mathcal{I}_d^{-1}(F))=\eta_2\big((\pi_{d^{n-1}}\circ\eta_1)^{-1}(F)\big)$ is closed in $H_d$ and $\mathcal{I}_d^{-1}(F)$ is closed in $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$. Let us introduce the map $\mathcal{I}$ defined by $$\begin{aligned} &\mathcal{I}\colon\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\to \mathrm{Bir}(\mathbb{P}^n_\mathbb{C}),&&\phi\mapsto\phi^{-1}.\end{aligned}$$ The degree of the inverse of a birational self map of $\mathbb{P}^n_\mathbb{C}$ of degree $d$ has degree at most $d^{n-1}$. Consequently $\mathcal{I}$ restricts for any $d\geq 1$ to an injective map $$\mathcal{I}_d\colon\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\to \mathrm{Bir}_{\leq d^{n-1}}(\mathbb{P}^n_\mathbb{C}).$$ According to Lemma \[lem:bfcont\] the map $\mathcal{I}_d$ is continuous. The definition of the topology of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ implies that $\mathcal{I}$ is continuous. Since $\mathcal{I}=\mathcal{I}^{-1}$ one has: \[cor:bfcor1\] The map $$\begin{aligned} &\mathcal{I}\colon\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\to \mathrm{Bir}(\mathbb{P}^n_\mathbb{C}),&&\phi\mapsto\phi^{-1}\end{aligned}$$ is a homeomorphism. Let us now look at the composition of two birational maps. \[lem:bfcont2\] For any $d$, $k$ the map $$\begin{aligned} &\chi_{d,k}\colon\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\times \mathrm{Bir}_{\leq k}(\mathbb{P}^n_\mathbb{C})\to \mathrm{Bir}_{\leq dk}(\mathbb{P}^n_\mathbb{C}), && (\phi,\psi)\mapsto\phi\circ\psi\end{aligned}$$ is continuous. Let us consider the map $\widehat{\chi_{d,k}}\colon H_d\times H_k\to H_{dk}$ given by $$\big((\phi_0:\phi_1:\ldots:\phi_n),\,(\psi_0:\psi_1:\ldots:\psi_n)\big)\mapsto \big(\phi_n(\psi_0,\psi_1,\ldots,\psi_n)):\ldots:\phi_n(\psi_0,\psi_1,\ldots,\psi_n)).$$ The diagram $$\xymatrix{ H_d\times H_k \ar[d]_{\pi_d\times\pi_k}\ar[r]^{\widehat{\chi_{d,k}}} & H_{dk}\ar[d]^{\pi_{dk}} \\ \mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\times\mathrm{Bir}_{\leq k}(\mathbb{P}^n_\mathbb{C}) \ar[r]_{\mathcal{I}_d} & \mathrm{Bir}_{\leq dk}(\mathbb{P}^n_\mathbb{C}) }$$ commutes. The map $\widehat{\chi_{d,k}}$ is a morphism of algebraic varieties so is continuous for the Euclidean topology. Therefore if $F$ is a closed subset of $\mathrm{Bir}_{\leq dk}(\mathbb{P}^n_\mathbb{C})$, then $(\pi_{dk}\circ\widehat{\chi_{d,k}})^{-1}(F)$ is closed in $H_d\times H_k$. But the diagram is commutative so $$(\pi_d\circ\widehat{\chi_{d,k}})(F)=(\pi_d\times\pi_k)^{-1}\big((\chi_{d,k})^{-1}(F)\big).$$ The product of two proper maps is proper ([@Bourbaki Chapter 1,§10.1]); as a consequence $\pi_d\times\pi_k$ is proper and hence closed. This implies that $\pi_d\times\pi_k$ is a quotient map. Hence $(\chi_{d,k})^{-1}(F)$ is closed and $\chi_{d,k}$ is continuous. According to Lemma \[lem:bfcont2\] the map $$\chi_{d,k}\colon\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})\times \mathrm{Bir}_{\leq k}(\mathbb{P}^n_\mathbb{C})\to\mathrm{Bir}_{\leq dk}(\mathbb{P}^n_\mathbb{C})$$ is continuous for each $d$, $k$. As a consequence by definition of the topology of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ we get: \[cor:bfcor2\] The map $$\begin{aligned} & \mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\times \mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\to \mathrm{Bir}(\mathbb{P}^n_\mathbb{C}),&&(\phi,\psi)\mapsto\phi\circ\psi \end{aligned}$$ is continuous. Corollaries \[cor:bfcor1\] and \[cor:bfcor2\] complete the proof of: The $n$-dimensional <span style="font-variant:small-caps;">Cremona</span> group endowed with the Euclidean topology is a topological group. ### Remarks and properties Let us state some results proved in [@BlancFurter]: - let $\mathrm{G}$ be a <span style="font-variant:small-caps;">Zariski</span> closed subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ of bounded degree, let $\mathrm{K}$ be its associated algebraic group (Corollary \[cor:agree\]). We put on $\mathrm{G}$ the restriction of the Euclidean topology of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$, we get the Euclidean topology on $K$ via the bijection $\pi\colon\mathrm{K}\to\mathrm{G}$ which becomes a homeomorphism; - $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is <span style="font-variant:small-caps;">Hausdorff</span>; - any compact subset of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is contained in $\mathrm{Bir}_{\leq d}(\mathbb{P}^n_\mathbb{C})$ for some $d$; - any convergent sequence of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ has bounded degree; - for $n\geq 2$ the topological space $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is not locally compact and not metrisable; - the topological group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is compactly generated if and only if $n\leq 2$. Algebraic elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ ------------------------------------------------------------- Note that all the results of this section hold for $\mathrm{Bir}(\mathbb{P}^n_\Bbbk)$ where $\Bbbk$ is an algebraically closed field of characteristic $0$. An element $\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is algebraic if it is contained in an algebraic subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Let us denote by $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\text{alg}}$ the set of algebraic elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. \[thm:Blancelalg\] - For any $n\geq 2$ there exists a closed subset $U$ of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ canonically homeomorphic to $\mathbb{A}^1_\mathbb{C}$ such that the set of algebraic elements of $U$ is equal to the set of unipotent elements of $U$ and corresponds to the elements $t\in\mathbb{A}^1_\mathbb{C}$ belonging to the subgroup of $(\mathbb{C},+)$ generated by $1$. - For any $n\geq 2$ there exists a closed subset $S$ of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ canonically homeomorphic to $\mathbb{A}^1_\mathbb{C}\times(\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\})$ such that the set of algebraic elements of $S$ is equal to the set of semi-simple elements of $S$ and corresponds to the elements $(a,\xi)\in\mathbb{A}^1_\mathbb{C}\times(\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\})$ where $a=\xi^k$ for some $k\in\mathbb{Z}$. In particular $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\text{alg}}$ and the set of unipotent elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ are not closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Furthermore we will see that $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\text{alg}}$ is a countable union of closed sets of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. \[lem:plus\] Let $\phi$ be an element of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. The closure of $\big\{\phi^k\,\vert\,k\in\mathbb{Z}\big\}$ in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is a closed abelian subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Let us denote by $\Omega$ the closure of $\big\{\phi^k\,\vert\,k\in\mathbb{Z}\big\}$ in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. For any $j\in\mathbb{Z}$ the set $\phi^j(\Omega)$ is a closed subset of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. It contains $\big\{\phi^k\,\vert\,k\in\mathbb{Z}\big\}$ thus contains $\Omega$. As a result $\phi^k(\Omega)=\Omega$ for any $k\in\mathbb{Z}$. Set $$M=\big\{\psi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\,\vert\,\psi(\Omega)\subset\Omega\big\}=\displaystyle\bigcap_{\omega\in\Omega}\Omega\omega^{-1}.$$ As $M$ is closed and contains $\big\{\phi^k\,\vert\,k\in\mathbb{Z}\big\}$, the set $M$ contains $\Omega$. Therefore $M$ is closed under composition. Similarly the set $\big\{\psi^{-1}\,\vert\,\psi\in\Omega\big\}$ is closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ and contains $\big\{\phi^k\,\vert\,k\in\mathbb{Z}\big\}$. The set $\Omega$ is then a subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Let us now prove that $\Omega$ is abelian. The centralizer $$\mathrm{Cent}(\varphi)=\big\{\psi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\,\vert\,\psi\circ\varphi=\varphi\circ\psi\big\}$$ of an element $\varphi$ of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is the preimage of the identity by the continuous map $$\begin{aligned} & \mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C}) && \psi\mapsto \psi\circ\varphi\circ\psi^{-1}\circ\varphi^{-1}.\end{aligned}$$ Since a point of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is closed, $\mathrm{Cent}(\varphi)$ is closed. The closed subgroup $\mathrm{Cent}(\varphi)$ of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ contains $\big\{\varphi^j\,\vert\,j\in\mathbb{Z}\big\}$ hence it contains $\Omega$. Consequently each element of $\Omega$ commutes with $\varphi$. The set $$\big\{\psi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\,\vert\,\psi\circ\omega=\omega\circ\psi\,\forall\,\omega\in\Omega\big\}=\displaystyle\bigcap_{\omega\in\Omega}\mathrm{Cent}(\omega)$$ is closed and contains $\big\{\varphi^j\,\vert\,j\in\mathbb{Z}\big\}$ so contains $\Omega$. Therefore $\Omega$ is abelian. \[pro:2plus\] Let $\phi$ be an element of the $n$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. - If the sequence $(\deg \phi^k)_{k\in\mathbb{N}}$ is unbounded, then $\phi$ is not contained in any algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. - If the sequence $(\deg \phi^k)_{k\in\mathbb{N}}$ is bounded, then $\big\{\phi^j\,\vert\,j\in\mathbb{Z}\big\}$ is an abelian algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. A direct consequence is the following result: \[cor:3plus\] Let $\phi$ be a birational self map of $\mathbb{P}^n_\mathbb{C}$. The following assertions are equivalent: - the map $\phi$ is algebraic; - the sequence $(\deg\phi^k)_{k\in\mathbb{N}}$ is bounded, [i.e.]{} $\phi$ is elliptic. The first assertion follows from Lemma \[lem:agree\]. Let us now focus on the second assertion. Assume that the sequence $(\deg\phi^k)_{k\in\mathbb{N}}$ is bounded. According to [@BassConnelWright] one has for any $k$ $$\deg\phi^{-k}\leq(\deg\phi^k)^{n-1}.$$ As a consequence the set $\big\{\phi^j\,\vert\,j\in\mathbb{Z}\big\}$ is contained in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\leq d}$ for some $d$ and so does the closure $\Omega$ of $\big\{\phi^j\,\vert\,j\in\mathbb{Z}\}$. Lemma \[lem:plus\] allows to conclude. For any $k$, $d\in\mathbb{N}$ set $$\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{k,d}=\big\{\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\,\vert\,\deg\phi^k\leq d\big\}$$ and $$\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\infty,d}=\big\{\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\,\vert\,\deg\phi^k\leq d\,\forall\,k\in\mathbb{N}\big\}$$ Then - the set $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{k,d}$ is closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$; - the set $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\infty,d}=\displaystyle\bigcap_{i\in\mathbb{N}} \mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{i,d}$ is closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$; - the set $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\text{alg}}$ of all algebraic elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ coincides with the union of all $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\infty,d}$. The set $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{\leq d}$ is closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ for any $d$ (Corollary \[cor:BlancFurter\]), and the map $$\begin{aligned} & \mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C}), && \varphi\mapsto\varphi^k\end{aligned}$$ is continuous (Remark \[rem:homeo\]); the set $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})_{k,d}$ is thus closed in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. The first assertion clearly implies the second one. The third assertion follows from Corollary \[cor:3plus\]. Let us now deal with the first assertion of Theorem \[thm:Blancelalg\]. Assume $n\geq 2$. Consider the morphism $\rho\colon\mathbb{A}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ given by $$a\mapsto\big((z_0:z_1:\ldots:z_n)\dashrightarrow(z_0z_1:z_1(z_1+z_0):z_2(z_1+az_0):z_3z_1:z_4z_1:\ldots:z_nz_1\big).$$ It is clearly injective. Let $\widetilde{\rho}\colon\mathbb{P}^1_\mathbb{C}\to W_2$ be the closed embedding given by $$(\alpha:\beta)\to(\alpha z_0z_1:\alpha z_1(z_1+z_0):z_2(z_1+az_0):\alpha z_3z_1:\alpha z_4z_1:\ldots:\alpha z_nz_1).$$ Note that $\widehat{\rho}((0:1))$ does not belong to $H_2$. However for any $t\in\mathbb{A}^1_\mathbb{C}$ one has $\mathrm{pr}_2(\widehat{\rho}((1:t))=\rho(t)$. The restriction to $\mathbb{A}^1_\mathbb{C}$ thus yields a closed embedding $\mathbb{A}^1_\mathbb{C}\to H_2$. According to Proposition \[pro:2plus\] the restriction of $\mathrm{pr}_2$ to $\widehat{\rho}(\mathbb{P}^1_\mathbb{C}\smallsetminus\{(0:1)\})$ is an homeomorphism. - For $t\in\mathbb{C}$ the following conditions are equivalent: - $\rho(t)$ is algebraic, - $\rho(t)$ is unipotent, - $\rho(t)$ is conjugate to $\rho(0)\colon(z_1+1,z_2,\ldots,z_n)$, - $t$ belongs to the subgroup of $(\mathbb{C},+)$ generated by $1$. - The pull-back by $\rho$ of the set of algebraic elements is not closed. <!-- --> - A direct computation yields to $$\rho(a)^k\colon(z_1,z_2,\ldots,z_n)\mapsto\left(z_1+k,z_2\,\frac{(z_1+a)(z_1+a+1)\ldots(z_1+a+k-1)}{z_1(z_1+1)\ldots(z_1+m-1)},z_3,z_4,\ldots,z_n\right)$$ In particular the second coordinate of $\rho(a)^k(z_1,z_2,\ldots,z_n)$ is $$\frac{\displaystyle\prod_{i=0}^{k-1}(z_1+a+i)}{\displaystyle\prod_{i=0}^{k-1}(z_1+i)}$$ If $a$ does not belong to the subgroup of $(\mathbb{C},+)$ generated by $1$, then the degree growth of $\rho(a)^k$ is linear which implies that $\rho(a)$ is not algebraic. If $a$ belongs to the subgroup of $(\mathbb{C},+)$ generated by $1$, then $$\deg\rho(a)^k\leq \vert k\vert+1\qquad\forall\,k\in\mathbb{N}.$$ As a consequence $\rho(a)$ is algebraic. Furthermore $\rho(a)$ is conjugate to $$\rho(0)\colon(z_1,z_2,\ldots,z_n)\mapsto(z_1+1,z_2,\ldots,z_n)$$ via $$(z_1,z_2,\ldots,z_n)\dashrightarrow\left(z_1,\frac{z_2}{z_1(z_1+1)\ldots(z_1+a-1)},z_3,z_4,\ldots,z_n\right)$$ if $a>0$ or via $$(z_1,z_2,\ldots,z_n)\dashrightarrow\Big(z_1,z_2z_1(z_1-1)\ldots(z_1+a),z_3,z_4,\ldots,z_n\Big)$$ if $a<0$. In particular $\rho(a)$ is unipotent. - The second assertion follows from the first one and the fact that the subgroup of $(\mathbb{C},+)$ generated by $1$ is closed. Finally let us prove the second assertion of Theorem \[thm:Blancelalg\]. Assume $n\geq 2$. Consider the morphism $$\rho\colon\mathbb{A}^1_\mathbb{C}\times(\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\})\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$$ given by $$(a,\xi)\mapsto\Big((z_0:z_1:\ldots:z_n)\dashrightarrow(z_0(z_1+z_0):\xi z_1(z_1+z_0):z_2(z_1+az_0):z_3(z_1+z_0):\ldots:z_n(z_1+z_0)\Big).$$ It is injective. Let $\widehat{\rho}\colon\mathbb{P}^2_\mathbb{C}\to W_2$ be the closed embedding given by $$(\alpha:\beta:\gamma)\dashrightarrow\Big(\alpha z_0(z_1+z_0):\gamma z_1(z_1+z_0):z_2(\alpha z_1+\beta z_0):\alpha z_3(z_1+z_0):\ldots:\alpha z_n(z_1+z_0)\Big).$$ Note that $$(z_0:z_1:\ldots:z_n)\dashrightarrow\Big(\alpha z_0(z_1+z_0):\gamma z_1(z_1+z_0):z_2(\alpha z_1+\beta z_0):\alpha z_3(z_1+z_0):\ldots:\alpha z_n(z_1+z_0)\Big)$$ is a birational map if and only if $\alpha\gamma\not=0$. This yields a closed embedding $$\begin{aligned} & \mathbb{A}^1_\mathbb{C}\times(\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\})\to H_2, && (a,\xi)\mapsto\widehat{\rho}((1:a:\xi)).\end{aligned}$$ Furthermore $\mathrm{pr}_2(\widehat{\rho}(1:a:\xi))=\rho(a,\xi)$. Proposition \[pro:2plus\] says that the restriction of $\mathrm{pr}_2$ to the image is a homeomorphism. - For $(a,\xi)\in\mathbb{A}^1_\mathbb{C}\times(\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\})$ the following conditions are equivalent: - $\rho(a,\xi)$ is algebraic, - $\rho(a,\xi)$ is semi-simple, - $\rho(a,\xi)$ is conjugate to $\rho(1,\xi)\colon(z_1,z_2,\ldots,z_n)\mapsto(\xi z_1,z_2,z_3,\ldots,z_n)$, - there exists $k\in\mathbb{Z}$ such that $a=\xi^k$. - The pull-back by $\rho$ of the set of algebraic elements is not closed. <!-- --> - Note that $$\rho(a,\xi)^k\colon(z_1,z_2,\ldots,z_n)\dashrightarrow\left(\xi^kz_1,z_2\,\frac{(z_1+a)(\xi z_1+a)\ldots(\xi^k z_1+a)}{(z_1+1)(\xi z_1+1)\ldots(\xi^{k-1}z_1+1)},z_3,z_4,\ldots,z_n\right).$$ In particular the second coordinate of $\rho(a,\xi)^k$ is $$\frac{\displaystyle\prod_{i=0}^{k-1}(\xi^iz_1+a)}{\displaystyle\prod_{i=0}^{k-1}(\xi^iz_1+1)}.$$ If $a$ does not belong to $\langle\xi\rangle\subset(\mathbb {C},\cdot)$, then the degree growth of $\rho(a,\xi)^k$ is linear hence $\rho(a,\xi)$ is not algebraic. If $a$ belongs to $\langle\xi\rangle\subset(\mathbb {C},\cdot)$, then $a=\xi^k$ for some $k\in\mathbb{Z}$ and for any $j\in\mathbb{N}$ $$\deg\rho(a,\xi)^j\leq\vert k\vert+1$$ so $\rho(a,\xi)$ is algebraic. Remark that $\rho(a,\xi)$ is conjugate to $\rho(1,\xi)$ via $$(z_1,z_2,\ldots,z_n)\dashrightarrow\left(z_1,\frac{z_2}{z_1(z_1+1)\ldots(z_1+a-1)},z_3,z_4,\ldots,z_n\right)$$ if $k>0$ and via $$(z_1,z_2,\ldots,z_n)\dashrightarrow\Big(z_1,z_2z_1(z_1-1)\ldots(z_1+a),z_3,z_4,\ldots,z_n\Big)$$ if $k<0$. - The second assertion follows from the first one and the fact that $$\big\{(a,\xi)\in\mathbb{A}^1_\mathbb{C}\times(\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\})\,\vert\, a=\xi^k\text{ for some $k\in\mathbb{Z}$}\big\}$$ is not closed. Classification of maximal algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ ---------------------------------------------------------------------------------------- In [@Blanc:ssgpealg] the author gives a complete classification of maximal algebraic subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group and provides algebraic varieties that parametrize the conjugacy classes. The algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ have been studied for a long time. <span style="font-variant:small-caps;">Enriques</span> established in [@Enriques] the complete classification of maximal connected algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$: every such subgroup is the conjugate of the identity component of the automorphism group of a minimal rational surface. A modern proof was given in [@Umemura:maxconnalg1]. The case of $\mathrm{Bir}(\mathbb{P}^3_\mathbb{C})$ was treated by <span style="font-variant:small-caps;">Enriques</span> and <span style="font-variant:small-caps;">Fano</span> and more recently by <span style="font-variant:small-caps;">Umemura</span> ([@Umemura:dim3; @Umemura:maxconnalg1; @Umemura:imp]). <span style="font-variant:small-caps;">Demazure</span> has studied the smooth connected subgroups of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ that contain a split torus of dimension $n$ (see [@Demazure:sousgroupesalgebriques]). Only a few results are known for non-connected subgroups even in dimension $2$. Nevertheless there are a lot of statements in the case of finite subgroups which are algebraic ones ([@Wiman; @BayleBeauville; @deFernex; @BeauvilleBlanc; @Beauville; @Iskovskikh; @DolgachevIskovskikh; @Blanc:SMF; @Blanc:CRAS]) and we deal with in Chapter \[chapter:finite\]. But these results do not show which finite groups are maximal algebraic subgroups. As mentioned in [@DolgachevIskovskikh] there are some remaining open questions like the description of the algebraic varieties that parameterize conjugacy classes of finite subgroups $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. <span style="font-variant:small-caps;">Blanc</span> gives an answer to this question for - abelian finite subgroups $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ whose elements do not fix a curve of positive genus ([@Blanc:abelien]); - finite cyclic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ (see [@Blanc:cyclic]); - maximal algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ (see [@Blanc:ssgpealg]). Before specifying <span style="font-variant:small-caps;">Blanc</span> results let us recall some notions. If $S$ is a projective smooth rational surface and $\mathrm{G}$ a subgroup of $\mathrm{Aut}(S)$ we say that $(\mathrm{G},S)$ is a pair. A birational map $\varphi\colon X\dashrightarrow Y$ is $\mathrm{G}$-equivariant if the inclusion $\varphi\circ\mathrm{G}\circ\varphi^{-1}\subset\mathrm{Aut}(Y)$ holds. The pair $(\mathrm{G},S)$ is minimal if every birational $\mathrm{G}$-equivariant morphism $\varphi\colon S\dashrightarrow S'$ where $S'$ is a projective, smooth surface, is an isomorphism. A morphism $\pi\colon S\to \mathbb{P}^1_\mathbb{C}$ is a conic bundle if all generic fibers of $\pi$ are isomorphic to $\mathbb{P}^1_\mathbb{C}$ and if there exists a finite number of singular fibers which are the transverse union of two curves isomorphic to $\mathbb{P}^1_\mathbb{C}$. ### <span style="font-variant:small-caps;">del Pezzo</span> surfaces and their automorphism groups A <span style="font-variant:small-caps;">del Pezzo</span> surface is a smooth projective surface $S$ such that the anti-canonical divisor $-K_S$ is ample. Let us recall the classification of <span style="font-variant:small-caps;">del Pezzo</span> surfaces. The number $d=K_S^2$ is called the degree of $S$. By <span style="font-variant:small-caps;">Noether</span>’s formula $1\leq d\leq 9$. For $d\geq 3$, the anticanonical linear system $\vert -K_S\vert$ maps $S$ onto a non-singular surface of degree $d$ in $\mathbb{P}^d_\mathbb{C}$. If $d=9$, then $S\simeq\mathbb{P}^2_\mathbb{C}$. If $d=8$, then $S\simeq \mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ or $S\simeq\mathbb{F}_1$. For $d\leq 7$ a <span style="font-variant:small-caps;">del Pezzo</span> surface $S$ is isomorphic to the blow up of $n=9-d$ points in $\mathbb{P}^2_\mathbb{C}$ in general position, that is - no three of them are colinear, - no six are on the same conic, - if $n=8$, then the points are not on a plane cubic which has one of them as its singular point. There exist ([@Dolgachev Chapter 8]) - a unique isomorphism class of <span style="font-variant:small-caps;">del Pezzo</span> surfaces of degree $5$ (resp. $6$, resp. $7$, resp. $9$), - two isomorphism classes of <span style="font-variant:small-caps;">del Pezzo</span> surfaces of degree $8$, - and infinitely many isomorphism classes of <span style="font-variant:small-caps;">del Pezzo</span> surfaces of degree $1$, (resp. $2$, resp. $3$, resp. $4$). We will see that automorphism groups of <span style="font-variant:small-caps;">del Pezzo</span> surfaces are algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and that they are finite if and only if the degree of the corresponding surface is $\leq 5$. If $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $5$, then $\mathrm{Aut}(S)=\mathfrak{S}_5$. Automorphism groups of <span style="font-variant:small-caps;">del Pezzo</span> surfaces of degree $\leq 4$ are described in [@DolgachevIskovskikh §10]. In particular the authors got the following: \[thm:648\] If the automorphism group of a <span style="font-variant:small-caps;">del Pezzo</span> surface is finite, then it has order at most $648$. \[lem:te1\] If the automorphism group of a <span style="font-variant:small-caps;">del Pezzo</span> surface is finite, then it can be embedded into $\mathrm{GL}(8,\mathbb{C})$. Let $S$ be a <span style="font-variant:small-caps;">del Pezzo</span> surface such that $\mathrm{Aut}(S)$ is finite. Then $\deg S\leq 5$ and $S$ is isomorphic to $\mathrm{Bl}_{p_1,p_2,\ldots,p_r}\mathbb{P}^2_\mathbb{C}$ where $4\leq r=9-\deg S\leq 8$ and $p_1$, $p_2$, $\ldots$, $p_r$ are general points of $\mathbb{P}^2_\mathbb{C}$. Denote by $\mathbf{e}_0$ the pullback of the class of a line and by $\mathbf{e}_{p_i}$ the class of the exceptional line $E_{p_i}$ corresponding to the point $p_i$. The dimension of the <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> space $\mathrm{NS}(S)\otimes\mathbb{R}$ is $r+1$ and $\mathbf{e}_0$, $\mathbf{e}_{p_1}$, $\mathbf{e}_{p_2}$, $\ldots$, $\mathbf{e}_{p_r}$ is a basis of $\mathrm{NS}(S)\otimes\mathbb{R}$. Note that the equality $\mathbf{e}_{p_i}\cdot\mathbf{e}_{p_i}=-1$ implies that $E_{p_i}$ is the only representative of $\mathbf{e}_{p_i}$ on $S$. If $\varphi\in\mathrm{Aut}(S)$ acts as the identity on $\mathrm{NS}(S)\otimes\mathbb{R}$, then $\varphi$ preserves the exceptional lines $E_{p_i}$ for $1\leq i\leq r$. Hence $\varphi$ induces an automorphism of $\mathbb{P}^2_\mathbb{C}$ that fixes $p_1$, $p_2$, $\ldots$, $p_r$. As $r\geq 4$ and as the $p_i$ are in general position $\varphi$ is the identity. The action of $\mathrm{Aut}(S)$ on $\mathrm{NS}(S)\otimes\mathbb{R}$ is thus faithful and we get a faithful representation $$\mathrm{Aut}(S)\to\mathrm{GL}(r+1,\mathbb{C}).$$ Any element $\varphi$ of $\mathrm{Aut}(S)$ fixes $K_S$; as a result the one-dimensional subspace $\mathbb{R}\cdot K_S$ of $\mathrm{NS}(S)\otimes\mathbb{R}$ is fixed. By projecting the orthogonal complement of $K_S$ in $\mathrm{NS}(S)\otimes\mathbb{R}$ we obtain a faithful representation of $\mathrm{Aut}(S)$ into $\mathrm{GL}(r,\mathbb{C})$. A <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $6$ is isomorphic to the blow up of the complex projective plane in three general points, [i.e.]{} isomorphic to the surface $$S_6=\big\{\big((z_0:z_1:z_2),(a:b:c)\big)\in\mathbb{P}^2_\mathbb{C}\times\mathbb{P}^2_\mathbb{C}\,\vert\, az_0=bz_1=cz_2\big\}.$$ The automorphism group of $S_6$ is isomorphic to $(\mathbb{C}^)^2\rtimes\big(\mathfrak{S}_3\times\faktor{\mathbb{Z}}{2\mathbb{Z}}\big)$ where $\mathfrak{S}_3$ acts by permuting the coordinates of the two factors simultaneously, $\faktor{\mathbb{Z}}{2\mathbb{Z}}$ exchanges the two factors and $d\in(\mathbb{C}^)^2$ acts as follows $$d\cdot\big((z_0:z_1:z_2),(a:b:c)\big)=\big(d(z_0:z_1:z_2):d^{-1}(a:b:c)\big).$$ In other words $\mathrm{Aut}(S_6)$ is conjugate to $\big(\mathfrak{S}_3\times\faktor{\mathbb{Z}}{2\mathbb{Z}}\big)\ltimes \mathrm{D}_2 \subset\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. \[lem:te2\] The group $\mathrm{Aut}(S_6)$ can be embedded in $\mathrm{GL}(6,\mathbb{C})$. Consider the rational map $$\begin{aligned} & \phi\colon\mathbb{P}^2_\mathbb{C}\dashrightarrow\mathbb{P}^6_\mathbb{C}, && (z_0:z_1:z_2)\dashrightarrow(z_0^2z_1:z_0^2z_2:z_0z_1^2:z_1^2z_2:z_0z_2^2:z_1z_2^2:z_0z_1z_2).\end{aligned}$$ The rational action of $(\mathfrak{S}_3\times\faktor{\mathbb{Z}}{2\mathbb{Z}})\ltimes \mathrm{D}_2$ on $\phi(\mathbb{P}^2_\mathbb{C})$ extends to a regular action on $\mathbb{P}^6_\mathbb{C}$ that preserves the affine space given by $z_6\not=0$. This yields an embedding of $(\mathfrak{S}_3\times\faktor{\mathbb{Z}}{2\mathbb{Z}})\ltimes\mathrm{D}_2$ into $\mathrm{GL}(6,\mathbb{C})$. ### <span style="font-variant:small-caps;">Hirzebruch</span> surfaces and their automorphism groups {#subsec:hirz} Let us introduce the <span style="font-variant:small-caps;">Hirzebruch</span> surfaces. Consider the surface $\mathbb{F}_1$ obtained by blowing up $(1:0:0)\in\mathbb{P}^2_\mathbb{C}$; it is a compactification of $\mathbb{C}^2$ which has a natural fibration corresponding to the lines $z_1=$ constant. The divisor at infinity is the union of two rational curves which intersect in one point: - one of them is the strict transform of the line at infinity in $\mathbb{P}^2_\mathbb{C}$, it is a fiber denoted by $f_1$; - the other one, denoted by $s_1$, is the exceptional divisor which is a section for the fibration. Furthermore $f_1^2=0$ and $s_1^2=-1$. More generally for any $n$, $\mathbb{F}_n$ is a compactification of $\mathbb{C}^2$ with a rational fibration and such that the divisor at infinity is the union of two transversal rational curves: a fiber $f_n$ and a section $s_n$ of self-intersection $-n$. These surfaces are called <span style="font-variant:small-caps;">Hirzebruch</span> surfaces. One can go from $\mathbb{F}_n$ to $\mathbb{F}_{n+1}$ as follows. Consider the surface $\mathbb{F}_n$. Set $p=s_n\cap f_n$. Let $\mathfrak{p}_1$ be the blow up of $p\in\mathbb{F}_n$ and let $\mathfrak{p}_2$ be the contraction of the strict transform $\widetilde{f_n}$ of $f_n$. One goes from $\mathbb{F}_n$ to $\mathbb{F}_{n+1}$ via $\mathfrak{p}_2\circ\mathfrak{p}_1^{-1}$. We can also go from $\mathbb{F}_{n+1}$ to $\mathbb{F}_n$ via $\widetilde{\mathfrak{p}_2}\circ\widetilde{\mathfrak{p}_1}^{-1}$ where - $\widetilde{\mathfrak{p}_1}$ is the blow-up of a point $q$ such that $q\in f_{n+1}$, $q\not\in s_{n+1}$; - $\widetilde{\mathfrak{p}_2}$ is the contraction of the strict transform $\widetilde{f_{n+1}}$ of $f_{n+1}$. We will say that both $\mathfrak{p}_2\circ\mathfrak{p}_1^{-1}$ and $\widetilde{\mathfrak{p}_2}\circ\widetilde{\mathfrak{p}_1}^{-1}$ are elementary transformations. Their automorphism groups are $$\begin{aligned} &\mathrm{Aut}(\mathbb{P}^2_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})=(\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C}))\rtimes\langle(z_0,z_1)\mapsto(z_1,z_0)\rangle, && \mathrm{Aut}(\mathbb{P}^2_\mathbb{C})=\mathrm{PGL}(3,\mathbb{C})\end{aligned}$$ and $$\begin{aligned} \mathrm{Aut}(\mathbb{F}_n)&=&\left\{(z_0,z_1)\mapsto\left(\frac{az_0+P(z_1)}{(\gamma z_1+\delta)^n},\frac{\alpha z_1+\beta}{\gamma z_1+\delta}\right)\,\big\vert\,\right.\\ & & \hspace{0.5cm}\left.\left(\begin{array}{cc}\alpha & \beta \\ \gamma & \delta\end{array}\right)\in\mathrm{PGL}(2,\mathbb{C}),\,a\in\mathbb{C}^,\,P\in\mathbb{C}[z_1],\,\deg P\leq n\right\}.\end{aligned}$$ In other words as soon as $n\geq 2$ the group $\mathrm{Aut}(\mathbb{F}_n)$ is isomorphic to $\mathbb{C}[z_0,z_1]_n\rtimes\faktor{\mathrm{GL}(2,\mathbb{C})}{\mu_n}$ where $\mu_n\subset\mathrm{GL}(2,\mathbb{C})$ is the subgroup of $n$-torsion elements in the center of $\mathrm{GL}(2,\mathbb{C})$. \[lem:ru1\] If $n\geq 2$ is even, then $\faktor{\mathrm{GL}(2,\mathbb{C})}{\mu_n}$ is isomorphic as an algebraic group to $\mathrm{PGL}(2,\mathbb{C})\times\mathbb{C}^$. If $n$ is odd, then $\faktor{\mathrm{GL}(2,\mathbb{C})}{\mu_n}$ is isomorphic as an algebraic group to $\mathrm{PGL}(2,\mathbb{C})$. In particular all finite subgroups of $\mathrm{Aut}(\mathbb{F}_n)$ can be embedded into $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$ as soon as $n\geq 2$. ### Automorphism groups of exceptional fibrations An exceptional fibration $S$ is a conic bundle with singular fiber above $2n$ points in $\mathbb{P}^1_\mathbb{C}$ and with two sections $s_1$ and $s_2$ of self-intersection $-n$, where $n$ denotes an integer $\geq 2$ (see [@Blanc:ssgpealg]). \[lem:ru2\] Let $\pi\colon S\to\mathbb{P}^1_\mathbb{C}$ be an exceptional fibration. Then $\mathrm{Aut}(S,\pi)$ is isomorphic to a subgroup of $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$. ### $\Big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\Big)^2$-conic bundles A conic bundle $\pi\colon S\to\mathbb{P}^1_\mathbb{C}$ is a $\Big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\Big)^2$-conic bundle if - the group $\mathrm{Aut}(\faktor{S}{\mathbb{P}^1_\mathbb{C}})$ is isomorphic to $(\faktor{\mathbb{Z}}{2\mathbb{Z}})^2$, - each of the three involutions of $\mathrm{Aut}(\faktor{S}{\mathbb{P}^1_\mathbb{C}})$ fixes pointwise an irreducible curve $C$ such that $\pi\colon C\to\mathbb{P}^1_\mathbb{C}$ is a double covering that is ramified over a positive even number of points. The automorphism group $\mathrm{Aut}(S,\pi)$ of a $\Big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\Big)^2$-conic bundle is finite; its structure is given by the following exact sequence ([@Blanc:ssgpealg]) $$1 \longrightarrow V\longrightarrow\mathrm{Aut}(S,\pi)\longrightarrow H_V\longrightarrow 1$$ where $V\simeq\big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\big)^2$ and $H_V$ is a finite subgroup of $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C})$. Note that we also have the following property: \[lem:ru3\] Let $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ be an infinite torsion group. Assume that for any finitely generated subgroup $\Gamma\subset\mathrm{G}$ there exists a $(\faktor{\mathbb{Z}}{2\mathbb{Z}})^2$-conic bundle $S\to\mathbb{P}^1_\mathbb{C}$ such that $\Gamma$ is conjugate to a subgroup of $\mathrm{Aut}(S,\pi)$. Then any finitely generated subgroup of $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$. ### <span style="font-variant:small-caps;">Blanc</span> results First <span style="font-variant:small-caps;">Blanc</span> proved: Every algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is contained in a maximal algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. The maximal algebraic subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group are the conjugate of the groups $\mathrm{G}=\mathrm{Aut}(S,\pi)$ where $S$ is a rational surface and $\pi\colon S\to Y$ is a morphism such that 1. $Y$ is a point, $\mathrm{G}=\mathrm{Aut}(S)$ and $S$ is one of the following: - $\mathbb{P}^2_\mathbb{C}$, $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$; - a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $1$, $4$, $5$ or $6$; - a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $3$ (resp. $2$) such that the pair $(\mathrm{Aut}(S),S)$ is minimal and such that the fixed points of the action of $\mathrm{Aut}(S)$ on $S$ are lying on exceptional curves; 2. $Y\simeq\mathbb{P}^1_\mathbb{C}$ and $\pi$ is one of the following conic bundles: - the fibration by lines of the <span style="font-variant:small-caps;">Hirzebruch</span> surface $\mathbb{F}_n$ for $n\geq 2$; - an exceptional conic bundle with at least $4$ singular fibers; - a $\Big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\Big)^2$-conic bundle such that $S$ is not a <span style="font-variant:small-caps;">del Pezzo</span> surface. Moreover, in all these cases, the pair $(\mathrm{G},S)$ is minimal and the fibration $\pi\colon S\to Y$ is a $\mathrm{G}$-<span style="font-variant:small-caps;">Mori</span> fibration which is birationally superrigid. This means that two such groups $\mathrm{G}=\mathrm{Aut}(S,\pi)$ and $\mathrm{G}'=\mathrm{Aut}(S',\pi')$ are conjugate if and only if there exists an isomorphism $S\to S'$ which sends fibers of $\pi$ onto fibers of $\pi'$. Then <span style="font-variant:small-caps;">Blanc</span> described more precisely the structure of these minimal algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Furthermore he provides algebraic varieties that parameterize the conjugacy classes of these groups: \[thm:blanc11cases\] The maximal algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ belong up to conjugacy to one of the eleven following families: 1. $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})\simeq\mathrm{PGL}(3,\mathbb{C})$; 2. $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})\simeq\big(\mathrm{PGL}(2,\mathbb{C})\big)^2\rtimes\faktor{\mathbb{Z}}{2\mathbb{Z}}$; 3. $\mathrm{Aut}(S)\simeq(\mathbb{C}^)^2\rtimes\big(\mathfrak{S}_3\times\faktor{\mathbb{Z}}{2\mathbb{Z}}\big)$ where $S$ is the <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $6$; 4. $\mathrm{Aut}(\mathbb{F}_n)\simeq\mathbb{C}^{n+1}\rtimes\faktor{\mathrm{GL}(2,\mathbb{C})}{\mu_n}$ where $\mu_n$ is the $n$-th torsion of the center of $\mathrm{GL}(2,\mathbb{C})$ with $n\geq 2$; 5. $\mathrm{Aut}(S,\pi)$ where $(S,\pi)$ is an exceptional conic bundle with singular fibers over a set $\Delta\subset\mathbb{P}^1_\mathbb{C}$ of $2n$ distinct points, $n\geq 2$; the projection of $\mathrm{Aut}(S,\pi)$ onto $\mathrm{PGL}(2,\mathbb{C})$ gives an exact sequence $$1 \longrightarrow \mathbb{C}^\rtimes\faktor{\mathbb{Z}}{2\mathbb{Z}}\longrightarrow\mathrm{Aut}(S,\pi)\longrightarrow H_\Delta\longrightarrow 1$$ where $H_\Delta$ is the finite subgroup of $\mathrm{PGL}(2,\mathbb{C})$ formed by elements that preserve $\Delta$; 6. $\mathrm{Aut}(S)\simeq\mathfrak{S}_5$ where $S$ is the <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $5$; 7. $\mathrm{Aut}(S)\simeq\big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\big)^4\rtimes H_S$ where $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $4$ obtained by blowing up $5$ points in $\mathbb{P}^2_\mathbb{C}$ and $H_S$ is the group of automorphisms of $\mathbb{P}^2_\mathbb{C}$ that preserve this set of points; 8. $\mathrm{Aut}(S)$ where $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $3$ of the following form - the triple cover of $\mathbb{P}^2_\mathbb{C}$ ramified along a smooth cubic $\Gamma$. If $S$ is the Fermat cubic, then $\mathrm{Aut}(S)=\big(\faktor{\mathbb{Z}}{3\mathbb{Z}}\big)^3\rtimes\mathfrak{S}_4$, otherwise we have an exact sequence $$1 \longrightarrow \faktor{\mathbb{Z}}{3\mathbb{Z}}\longrightarrow\mathrm{Aut}(S)\longrightarrow H_\Gamma\longrightarrow 1$$ where $H_\Gamma$ is the group of automorphisms of $\mathbb{P}^2_\mathbb{C}$ that preserve $\Gamma$, $H_\Gamma$ contains a subgroup isomorphic to $\big(\faktor{\mathbb{Z}}{3\mathbb{Z}}\big)^2$; - the <span style="font-variant:small-caps;">Clebsch</span> cubic surface whose automorphism group is isomorphic to $\mathfrak{S}_5$; - a cubic surface given by $z_0^3+z_0(z_1^2+z_2^2+z_3^2)+\lambda z_1z_2z_3=0$ for some $\lambda\in\mathbb{C}$, $9\lambda^3\not=8\lambda$, $8\lambda^3\not=-1$ and whose automorphism group is isomorphic to $\mathfrak{S}_4$; 9. $\mathrm{Aut}(S)\simeq\faktor{\mathbb{Z}}{2\mathbb{Z}}\rtimes H_S$ where $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $2$ which is a double cover of a smooth quartic $Q_S\subset\mathbb{P}^2_\mathbb{C}$ such that $H_S=\mathrm{Aut}(Q_S)$ acts without fixed point on the quartic without its bitangent points; 10. $\mathrm{Aut}(S)$ where $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $1$, double cover of a quadratic cone $Q$, ramified along a curve $\Gamma_S$ of degree $6$, complete intersection of $Q$ with a cubic surface of $\mathbb{P}^3_\mathbb{C}$. We have the following exact sequence $$1 \longrightarrow \faktor{\mathbb{Z}}{2\mathbb{Z}}\longrightarrow\mathrm{Aut}(S)\longrightarrow H_S\longrightarrow 1$$ where $H_S$ denotes the automorphism group of $Q$ preserving the curve $\Gamma_S$; 11. $\mathrm{Aut}(S,\pi)$ where $(S,\pi)$ is a $\Big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\Big)^2$-conic bundle such that $S$ is not a <span style="font-variant:small-caps;">del Pezzo</span> surface. The projection of $\mathrm{Aut}(S,\pi)$ onto $\mathrm{PGL}(2,\mathbb{C})$ gives the following exact sequence $$1 \longrightarrow V\longrightarrow\mathrm{Aut}(S,\pi)\longrightarrow H_V\longrightarrow 1$$ where $V\simeq\big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\big)^2$ contains three involutions fixing an hyperelliptic curve ramified over points of $p_1$, $p_2$, $p_3\subset\mathbb{P}^1_\mathbb{C}$ and $H_V\subset\mathrm{Aut}(\mathbb{P}^1_\mathbb{C})$ is the finite subgroup preserving the set $\big\{p_1,\,p_2,\,p_3\big\}$. The eleven families are disjoint and the conjugacy classes in any family are parameterized respectively by - , $(2)$, $(3)$,eè $(6)$ the point; - there is only one conjugacy class for any integer $n\geq 2$; - for any integer $n\geq 2$ the set of $2n$ points of $\mathbb{P}^1_\mathbb{C}$ modulo the action of $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C})=\mathrm{PGL}(2,\mathbb{C})$; - the isomorphism classes of <span style="font-variant:small-caps;">del Pezzo</span> surfaces of degree $4$; - the isomorphism classes of cubic surfaces given respectively - by the isomorphism classes of elliptic curves; - for the <span style="font-variant:small-caps;">Clebsch</span> surface there is only one isomorphism class; - by the classes of $\big\{\lambda\in\mathbb{C}\,\vert\, 9\lambda^3\not=8\lambda,\,8\lambda^3\not=-1\big\}$ modulo the equivalence $\lambda\sim -\lambda$. - the isomorphism classes of smoooth quartics of $\mathbb{P}^2_\mathbb{C}$ having automorphism groups acting without fixed points on the quartic without its bitangent points; - the isomorphism classes of <span style="font-variant:small-caps;">del Pezzo</span> surfaces of degree $1$; - the triplets of ramification $\big\{p_1,\,p_2,\,p_3\big\}\subset\mathbb{P}^1_\mathbb{C}$ that determine $\big(\faktor{\mathbb{Z}}{2\mathbb{Z}}\big)^2$ conic bundles on surfaces that are not <span style="font-variant:small-caps;">del Pezzo</span> ones, modulo the action of $\mathbb{P}^1_\mathbb{C}$. The approach of <span style="font-variant:small-caps;">Blanc</span> used the modern viewpoint of <span style="font-variant:small-caps;">Mori</span>’s theory and <span style="font-variant:small-caps;">Sarkisov</span>’s program, aiming a generalization in higher dimension: - he described each maximal algebraic subgroup of the classification as a $\mathrm{G}$-<span style="font-variant:small-caps;">Mori</span> fibration; - he then proved that any algebraic subgroup is contained in one of the groups of the classification; - he also showed that any group of the classification is a minimal $\mathrm{G}$-fibration that is furthermore superrigid. Lemmas \[lem:te1\], \[lem:te2\] and Theorem \[thm:blanc11cases\] allow to prove the following statement: \[lem:ru4\] Let $\mathrm{G}$ be a subgroup of the two-dimensional <span style="font-variant:small-caps;">Cremona</span> group. Assume that $\mathrm{G}$ is conjugate to an automorphism group of a <span style="font-variant:small-caps;">del Pezzo</span> surface $S$. Then $\mathrm{G}$ can be embedded into $\mathrm{GL}(8,\mathbb{C})$. If $\deg S\leq 5$, then $\mathrm{Aut}(S)$ is finite and Lemma \[lem:te1\] allows to conclude. If $\deg S=6$, then $\mathrm{Aut}(S)$ can be embedded into $\mathrm{GL}(8,\mathbb{C})$ (Lemma \[lem:te2\]). If $\deg S=7$, then $\mathrm{Aut}(S)$ is conjugate to a subgroup of $$\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})\simeq\big(\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})\big)\rtimes\faktor{\mathbb{Z}}{2\mathbb{Z}}\subset \mathrm{GL}(6,\mathbb{C}).$$ If $\deg S=8$, then $S$ is isomorphic either to $\mathbb{F}_0=\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ or to $\mathbb{F}_1$. On the one hand $$\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})\simeq\big(\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})\big)\rtimes\faktor{\mathbb{Z}}{2\mathbb{Z}}\subset \mathrm{GL}(6,\mathbb{C}).$$ and on the other hand $\mathrm{Aut}(\mathbb{F}_1)$ is not a maximal algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ (Theorem \[thm:blanc11cases\]). If $\deg S=9$, then $S\simeq\mathbb{P}^2_\mathbb{C}$ and $\mathrm{Aut}(S)=\mathrm{PGL}(3,\mathbb{C})\subset\mathrm{GL}(8,\mathbb{C})$. Closed normal subgroups of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ {#section:closednormalsubgroups} ------------------------------------------------------------------ As we have seen we can endow the <span style="font-variant:small-caps;">Cremona</span> group with a natural <span style="font-variant:small-caps;">Zariski</span> topology induced by morphisms $V\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ where $V$ is an algebraic variety. In [@Shafarevich] <span style="font-variant:small-caps;">Mumford</span> discussed properties of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and in particular asked if it is a simple group with respect to the <span style="font-variant:small-caps;">Zariski</span> topology, [i.e.]{} if every closed normal subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is trivial. <span style="font-variant:small-caps;">Blanc</span> provided an affirmative answer to <span style="font-variant:small-caps;">Mumford</span> question: \[thm:simple2\] The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is simple with respect to the <span style="font-variant:small-caps;">Zariski</span> topology, [i.e.]{} it does not contain any non-trivial closed normal strict subgroup. An other result due to <span style="font-variant:small-caps;">Blanc</span> is \[pro:imp\] Let $n\geq 1$ be an integer. The group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is linearly connected. In other words for every $\psi$, $\phi$ in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ there exist - an open subset $\mathcal{U}\subset\mathbb{A}^1_\mathbb{C}$ containing $0$ and $1$, - and a morphism $\theta\colon\mathcal{U}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ such that $\theta(0)=\psi$ and $\theta(1)=\phi$. In particular $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is connected for every $n\geq 1$, yielding a positive answer to <span style="font-variant:small-caps;">Serre</span>’s question. The proof of Theorem \[thm:simple2\] is based on the classical <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> Theorem asserting that any element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is the composition of birational self maps of the complex projective plane of degree $2$ and an automorphism of $\mathbb{P}^2_\mathbb{C}$. <span style="font-variant:small-caps;">Blanc</span> proved that a non-trivial normal closed subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$, $n\geq 2$, contains $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ and $\mathrm{PSL}(2,\mathbb{C}(z_1,\ldots,z_{n-1}))$ but since the classical <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> factorization Theorem does not generalize to higher dimensions he can not state an analogue of Theorem \[thm:simple2\]. Nevertheless later <span style="font-variant:small-caps;">Blanc</span> and <span style="font-variant:small-caps;">Zimmermann</span> generalize Theorem \[thm:simple2\] as follows: \[thm:BlancZimmermann\] Let $n\geq 1$ be an integer. The group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is topologically simple when endowed with the <span style="font-variant:small-caps;">Zariski</span> topology. \[pro:chemin\] Let $\phi$ be an element of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. Let $p$ be a point of $\mathbb{P}^n_\mathbb{C}$ such that $\phi$ induces a local isomorphism at $p$, and fixes $p$. Then there exist morphisms $\nu\colon\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\}\to\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ and $\upsilon\colon\mathbb{A}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ such that: - $\upsilon(t)=\nu(t)^{-1}\circ\phi\circ \nu(t)$ for any $t\in\mathbb{C}$, moreover $\nu(1)=\mathrm{id}$ so $\upsilon(1)=\phi$; - $\upsilon(0)$ belongs to $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ and is the identity if and only if the action of $\phi$ on the tangent space is trivial. Up to conjugacy by an element of $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ we can assume that $p=(1:0:0:\ldots:0)$. In the affine chart $z_0=1$ one can write $\phi$ locally as $$\left(\frac{p_{1,1}(z_1,\ldots,z_n)+\ldots+p_{1,\ell}(z_1,\ldots,z_n)}{1+q_{1,1}(z_1,\ldots,z_n)+\ldots+q_{1,\ell}(z_1,\ldots,z_n)},\ldots,\frac{p_{n,1}(z_1,\ldots,z_n)+\ldots+p_{n,\ell}(z_1,\ldots,z_n)}{1+q_{n,1}(z_1,\ldots,z_n)+\ldots+q_{n,\ell}(z_1,\ldots,z_n)}\right)$$ where $p_{i,j}$, $q_{i,j}$ are homogeneous of degree $j$. For each $t\in\mathbb{C}\smallsetminus\{0\}$ the element $$\nu_t\colon(z_1,z_2,\ldots,z_n)\mapsto(tz_1,tz_2,\ldots,tz_n)$$ extends to a linear automorphism of $\mathbb{P}^n_\mathbb{C}$ that fixes $p$. Hence the map $t\mapsto\nu_t^{-1}\circ\phi\circ\nu_t$ gives rise to a morphism $\Theta\colon\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ and the image of $\Theta$ contains only conjugates of $\phi$ by linear automorphisms. Note that $$\begin{aligned} \Theta\colon t\mapsto & &\left(\frac{p_{1,1}(z_1,\ldots,z_n)+tp_{1,2}(z_1,\ldots,z_n)+\ldots+t^{\ell-1}p_{1,\ell}(z_1,\ldots,z_n)}{1+tq_{1,1}(z_1,\ldots,z_n)+t^2q_{1,2}(z_1,\ldots,z_n)+\ldots+t^\ell q_{1,\ell}(z_1,\ldots,z_n)},\right.\\ & &\qquad\left.\ldots,\frac{p_{n,1}(z_1,\ldots,z_n)+tp_{n,2}(z_1,\ldots,z_n)+\ldots+t^{\ell-1}p_{n,\ell}(z_1,\ldots,z_n)}{1+tq_{n,1}(z_1,\ldots,z_n)+t^2q_{n,2}(z_1,\ldots,z_n)+\ldots+t^\ell q_{n,\ell}(z_1,\ldots,z_n)}\right)\end{aligned}$$ and $\Theta(0)$ corresponds to the linear part of $\Theta$ at $p$ which is locally given by $$\big(p_{1,1}(z_1,\ldots,z_n),\ldots,p_{n,1}(z_1,\ldots,z_n)\big).$$ As $\phi$ is a local isomorphism at $p$, this linear part is an automorphism of $\mathbb{P}^n_\mathbb{C}$. Furthermore it is trivial if and only if the action of $\phi$ on the tangent space is trivial. Let $\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\smallsetminus\{\mathrm{id}\}$; it induces an isomorphism from $\mathcal{U}$ to $\mathcal{V}$ where $\mathcal{U}$, $\mathcal{V}\subset\mathbb{P}^n_\mathbb{C}$ are two non-empty open subsets. There exist a point $p$ in $\mathcal{U}$ and two automorphisms $\alpha_1$, $\alpha_2$ of $\mathbb{P}^n_\mathbb{C}$ such that - $\psi=\alpha_1\circ \phi\circ\alpha_2$ fixes $p$, - $\psi=\alpha_1\circ \phi\circ\alpha_2$ is a local isomorphism at $p$, - $D_p\psi$ is not trivial. According to Proposition \[pro:chemin\] there exist morphisms $\nu\colon\mathbb{A}^1_\mathbb{C}\smallsetminus\{0\}\to\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ and $\upsilon_1\colon\mathbb{A}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ such that - $\upsilon_1(t)=\nu(t)^{-1}\circ \psi^{-1}\circ\nu(t)$ for each $t\not=0$, - $\upsilon_1(0)$ is an automorphism of $\mathbb{P}^n_\mathbb{C}$. Consider the morphism $\upsilon_2\colon\mathbb{A}_\mathbb{C}^1\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ defined by $$\upsilon_2(t)=\alpha_1^{-1}\circ \psi\circ\upsilon_1(t)\circ\upsilon_1(0)^{-1}\circ\alpha_2^{-1}.$$ Since $\alpha_1$, $\alpha_2$, $\upsilon_1(0)$ and $\nu(t)$ are automorphisms of $\mathbb{P}^n_\mathbb{C}$ for all $t\not=0$ $$\upsilon_2(t)=\alpha_1^{-1}\circ\big(\psi\circ\nu(t)^{-1}\circ \psi^{-1}\big)\circ\nu(t)\circ\upsilon_1(0)^{-1}\circ\alpha_2^{-1}$$ belongs for any $t\not=0$ to the normal subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ generated by $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$. As a consequence $\phi=\upsilon_2(0)$ belongs to the closure of the normal subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ generated by $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$. But the normal subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ generated by $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ is dense in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ (see [@BlancFurter]). Let us mention that $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$, $n\geq 2$, is not simple as an abstract group (for $n=2$ see [@CantatLamy] or §\[CantatLamy:passimple\], for $n\geq 3$ see [@BlancLamyZimmermann]). Furthermore there is an analogue of Theorem \[thm:BlancZimmermann\] when $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is endowed with the Euclidean topology: \[thm:BZ2\] Let $n\geq 2$ be an integer. The topological group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is topologically simple when endowed with the Euclidean topology. This result of course does not hold for $n=1$: $\mathrm{PSL}(2,\mathbb{R})$ is a non-trivial normal strict subgroup of $\mathrm{PGL}(2,\mathbb{R})$ which is closed for the Euclidean topology. The proof of Theorem \[thm:BZ2\] is similar to the proof of Theorem \[thm:BlancZimmermann\]. Furthermore, one has: If $n\geq 1$, the group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is connected with respect to the <span style="font-variant:small-caps;">Zariski</span> topology. If $n\geq 2$, the group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is path-connected, and thus connected with respect to the Euclidean topology. Let us give an idea of the proof of this statement. We start with an example. \[eg:utile\] Let $n\geq 2$ and let $\alpha$ be an element of $\mathbb{C}^$. Consider the birational self map of $\mathbb{P}^n_\mathbb{C}$ given by $$\Phi\colon(z_0:z_1:\ldots:z_n) \dashrightarrow \left(\frac{z_0(z_1+\alpha z_2)+z_1z_2}{z_1+z_2}:z_1:z_2:\ldots:z_n\right).$$ The points $p=(0:1:0:0:\ldots:0)$ and $q=(0:0:1:0:0:\ldots:0)$ are fixed by $\Phi$. Applying Proposition \[pro:chemin\] to the points $p$ and $q$ we get two morphisms $\Theta_1$, $\Theta_2\colon\mathbb{A}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ such that - $\Theta_1(0)\colon(z_0:z_1:\ldots:z_n)\mapsto(z_0+z_2:z_1:z_2:\ldots:z_n)\in\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$, - $\Theta_2(0)\colon(z_0:z_1:\ldots:z_n)\mapsto(\alpha z_0+z_1:z_1:z_2:\ldots:z_n)\in\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$, - $\Theta_1(1)=\Theta_2(1)=\Phi$. Let us now establish a generalization of Proposition \[pro:imp\]: Let $n\geq 2$ be an integer. For any $\phi$, $\psi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ there is a morphism $\upsilon\colon\mathbb{P}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^1_\mathbb{C})$ such that $\upsilon(0)=\phi$ and $\upsilon(1)=\psi$. Up to composition with $\phi^{-1}$ one can assume that $\phi=\mathrm{id}$. Let us consider the subset $S$ of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ given by $$S=\left\{\varphi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\,\vert\,\exists\,\, \nu\colon\mathbb{A}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\text{ morphism such that }\nu(0)=\mathrm{id}\text{ and }\nu(1)=\varphi \right\}.$$ Let $\phi$ (resp. $\psi$) be an element of $S$; denote by $\nu_\phi$ (resp. $\nu_\psi$) the associated morphism. We can defined a morphism $\nu_{\phi\circ\psi}\colon\mathbb{A}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ by $\nu_{\phi\circ\psi}(t)=\nu_\phi(t)\circ\nu_\psi(t)$ which satisfies $\nu_{\phi\circ\psi}(0)=~\mathrm{id}$ and $\nu_{\phi\circ\psi}(1)=\phi\circ\psi$. For any $\varphi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ it is also possible to define a morphism $t\mapsto\varphi\circ\nu_\phi(t)\circ\varphi^{-1}$. Therefore $S$ is a normal subgroup of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$. The group $S$ contains $\mathrm{PSL}(n+1,\mathbb{C})$. Take $\alpha$, $\Phi$, $\Theta_1$ and $\Theta_2$ as in Example \[eg:utile\]; for $i\in\{1,\,2\}$ the morphisms $$t\mapsto\Theta_i(t)\circ(\Theta_i(0))^{-1}$$ show that $g\circ\big(\Theta_1(0)\big)^{-1}$ and $g\circ\big(\Theta_2(0)\big)^{-1}$ belong to $S$, hence $\Theta_1(0)\circ\big(\Theta_2(0)\big)^{-1}$ belong to $S$. But $\Theta_1(0)$ belongs to $\mathrm{PSL}(n+1,\mathbb{C})\subset S$ so $\Theta_2(0)$ belongs to $S$. Thus $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})=\mathrm{PGL}(n+1,\mathbb{C})$ is contained in $S$. Take $\phi\in\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ of degree $d\geq 2$. Let $p$ be a point of $\mathbb{P}^n_\mathbb{C}$ such that $\phi$ induces a local isomorphism at $p$. Consider an element $A$ of $\mathrm{PSL}(n+1,\mathbb{C})$ such that $A\circ\phi$ fixes $p$. There exists a morphism $\theta\colon\mathbb{A}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ such that $\theta(0)$ belongs to $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ and $\theta(1)=A\circ\phi$. Let us define $\theta'\colon\mathbb{A}^1_\mathbb{C}\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ by $\theta'(t)=\rho(t)\circ\theta(0)^{-1}$. Then $\theta'(1)=A\circ\phi\circ\theta(0)^{-1}$. But $\alpha$ and $\theta(0)$ belong to $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})\subset S$ so $\phi$ belongs to $S$. Regularization of rational group actions {#sec:WeilKraft} ---------------------------------------- The aim of [@Kraft:regularization] is to give a modern proof of the regularization theorem of <span style="font-variant:small-caps;">Weil</span> which says that for every rational action of an algebraic group $\mathrm{G}$ on a variety $X$ there exist a variety $Y$ with a regular action of $\mathrm{G}$ and a $\mathrm{G}$-equivariant birational map $X\dashrightarrow Y$ (see* [@Weil]). In this section the base field $\Bbbk$ is algebraically closed, a variety is an algebraic $\Bbbk$-variety, and an algebraic group is an algebraic $\Bbbk$-group. A rational map $\phi\colon X\dashrightarrow Y$ is called biregular in $p\in X$ if there is an open neighborhood $\mathcal{U}\subset (X\smallsetminus\mathrm{Base}(\phi))$ of $p$ such that $\phi_{\vert\mathcal{U}}\colon\mathcal{U} \hookrightarrow Y$ is an open immersion. As a result the subset $$X'=\big\{p\in X\,\vert\,\phi\text{ is biregular in }p\big\}$$ is open in $X$, and the induced morphism $\phi\colon X'\hookrightarrow Y$ is an open immersion. We can thus state: \[lem:Kraft1\] Let $X$ and $Y$ be two varieties. Let $\phi\colon X\dashrightarrow Y$ be a birational map. Then the set $$\mathrm{Breg}(\phi)=\big\{p\in X\,\vert\,\phi\text{ is biregular in }p\big\}$$ is open and dense in $X$. ### Rational group actions Let $X$ and $Z$ be two varieties. Let us recall that a map $\phi\colon Z\to\mathrm{Bir}(X)$ is a morphism if there exists an open dense set $\mathcal{U}\subset Z\times X$ such that - the induced map $\mathcal{U}\to X$, $(q,p)\mapsto\phi(q)(p)$ is a morphism of varieties; - for every $q\in Z$ the open set $\mathcal{U}_q=\big\{p\in X\,\vert\,(q,p)\in\mathcal{U}\big\}$ is dense in $X$; - for every $q\in Z$ the birational map $\phi(q)\colon X\dashrightarrow X$ is defined on $\mathcal{U}_q$. Equivalently there is a rational map $\phi\colon Z\times X\to X$ such that for every $q\in Z$ - the open subset $(Z\times X\smallsetminus\mathrm{Base}(\phi))\cap(\{q\}\times X)$ is dense in $\{q\}\times X$; - the induced birational map $\phi_q\colon X\dashrightarrow X$, $p\mapsto\phi(q,p)$ is birational. Recall that this definition allows to define the <span style="font-variant:small-caps;">Zariski</span> topology on $\mathrm{Bir}(X)$ (see §\[sec:alg\]). We can now define rational group actions on varieties. Let $\mathrm{G}$ be an algebraic group, and let $X$ be a variety. A rational action of $\mathrm{G}$ on $X$ is a morphism $\rho\colon\mathrm{G}\to\mathrm{Bir}(X)$ which is a morphism of groups. In other words there is a rational map still denoted $\rho$ $$\rho\colon\mathrm{G}\times X\dashrightarrow X$$ such that - the open set $((\mathrm{G}\times X)\smallsetminus\mathrm{Base}(\rho))\cap(\{g\}\times X)$ is dense in $\{g\}\times X$ for every $g\in\mathrm{G}$; - the induced map $\rho_g\colon X\dashrightarrow X$, $p\mapsto\rho(g,p)$ is birational for every $g\in\mathrm{G}$; - the map $g\mapsto\rho_g$ is a group morphism. \[thm:Kraft2\] Let $\rho\colon\mathrm{G}\to\mathrm{Bir}(X)$ be a rational action where $X$ is affine. Assume that there exists a dense subgroup $\Gamma\subset\mathrm{G}$ such that $\rho(\Gamma)\subset\mathrm{Aut}(X)$. Then the $\mathrm{G}$-action on $X$ is regular. Let $X$ and $Y$ be two varieties. Let $\rho$ be a rational $\mathrm{G}$-action on $X$. Let $\mu$ be a rational $\mathrm{G}$-action on $Y$. A dominant rational map $\phi\colon X\dashrightarrow Y$ is $\mathrm{G}$-equivariant if the following holds: for every $(g,p)\in\mathrm{G}\times X$ such that - $\rho$ is defined in $(g,p)$, - $\phi$ is defined in $p$ and in $\rho(g,p)$, - $\mu$ is defined in $(g,\phi(p))$, we have $\phi(\rho(g,p))=\mu(g,\phi(p))$. The set of $(g,p)\in\mathrm{G}\times X$ satisfying the previous assumptions is open and dense in $\mathrm{G}\times X$ and has the property that it meets all $\{g\}\times X$ in a dense open set. Let $X$ be a variety with a rational action $\rho\colon \mathrm{G}\times X\dashrightarrow X$ of an algebraic group $\mathrm{G}$. Consider $$\begin{aligned} &\widetilde{\rho}\colon\mathrm{G}\times X\dashrightarrow\mathrm{G}\times X,&& (g,p)\mapsto (g,\rho(g,p)).\end{aligned}$$ It is clear that $(\mathrm{G}\times X)\smallsetminus\mathrm{Base}(\widetilde{\rho})=(\mathrm{G}\times X)\smallsetminus\mathrm{Base}(\rho)$. Furthermore $\widetilde{\rho}$ is birational with inverse $\widetilde{\rho}^{-1}(g,p)=(g,\rho(g^{-1},p))$, that is $$\widetilde{\rho}^{-1}=\tau\circ\widetilde{\rho}\circ\tau$$ where $\tau$ is the isomorphism $$\begin{aligned} & \tau\colon\mathrm{G}\times X\to\mathrm{G}\times X, && (g,p)\mapsto(g^{-1},p).\end{aligned}$$ A point $x\in X$ is called $\mathrm{G}$-regular for the rational $\mathrm{G}$-action $\rho$ on $X$ if $\mathrm{Breg}(\widetilde{\rho})\cap(\mathrm{G}\times\{p\})$ is dense in $\mathrm{G}\times\{p\}$. In other words a point $p\in X$ is called $\mathrm{G}$-regular for the rational $\mathrm{G}$-action $\rho$ on $X$ if $\widetilde{\rho}$ is biregular in $(g,p)$ for all $g$ in a dense open set of $\mathrm{G}$. Denote by $X_{\text{reg}}\subset X$ the set of $\mathrm{G}$-regular points. Let $\lambda_g\colon \mathrm{G}\stackrel{\sim}{\to}\mathrm{G}$ be the left multiplication with $g\in\mathrm{G}$. For any $h\in\mathrm{G}$ the diagram $$\xymatrix{ \mathrm{G}\times X \ar@{-->}[r]^{\widetilde{\rho}} \ar[d]_{\lambda_h\times\mathrm{id}} & \mathrm{G}\times X \ar[d]^{\lambda_h\times\rho_h} \\ \mathrm{G}\times X \ar@{-->}[r]_{\widetilde{\rho}} & \mathrm{G}\times X }$$ commutes. This implies the following statement: \[lem:Kraft2\] If $\rho$ is defined in $(g,p)$ and if $\rho_h$ is defined in $g\cdot p$, then $\rho$ is defined in $(hg,p)$. If $\widetilde{\rho}$ is biregular in $(g,p)$ and if $\rho_h$ is biregular in $g\cdot p$, then $\widetilde{\rho}$ is biregular in $(hg,p)$. \[pro:Kraft3\] The set $X_{\text{reg}}$ of $\mathrm{G}$-regular points is open and dense in $X$. If $p$ belongs to $X_{\text{reg}}$ and if $\widetilde{\rho}$ is biregular in $(g,p)$, then $g\cdot p$ belongs to $X_{\text{reg}}$. Let $\mathrm{G}=\mathrm{G}_0\cup\mathrm{G}_1\cup\ldots\cup\mathrm{G}_n$ be the decomposition into connected components. Then $D_i=\mathrm{Breg}(\rho)\cap(\mathrm{G}_i\times X)$ is open and dense for all $i$ (Lemma \[lem:Kraft1\]); the same holds for the image $\overline{D_i}\subseteq X$ under the projection onto $X$. Since $X_{\text{reg}}=\displaystyle\bigcap_i\overline{D_i}$ the set $X_{\text{reg}}$ is open and dense in $X$. If $\widetilde{\rho}$ is biregular in $(g,p)$, then $\widetilde{\rho}^{-1}=\tau\circ\widetilde{\rho}\circ\tau$ is biregular in $(g,g\cdot p)$. As a consequence $\widetilde{\rho}$ is biregular in $\tau(g,g\cdot p)=(g^{-1},g\cdot p)$. If $p$ is $\mathrm{G}$-regular, then $\rho_h$ is biregular in $p$ for all $h$ in a dense open subset $\mathrm{G}'$ of $\mathrm{G}$. According to the second assertion of Lemma \[lem:Kraft2\] the birational map $\widetilde{\rho}$ is biregular in $(hg^{-1},g\cdot p)$ for all $h\in \mathrm{G}'$. Hence $g\cdot p$ belongs to $X_{\text{reg}}$. A consequence of Proposition \[pro:Kraft3\] allows us to only consider the case of a rational $\mathrm{G}$-action such every point is $\mathrm{G}$-regular. Every point is $\mathrm{G}$-regular for the rational action $\mathrm{G}$. \[lem:Kraft3\] Assume that $X=X_{\text{reg}}$. If $\rho_g$ is defined in $p$, [i.e.]{} if $p\in X\smallsetminus\mathrm{Base}(\rho_g)$, then $\rho_g$ is biregular in $p$. Suppose that $\rho_g$ is defined in $p\in X$. As $X=X_{\text{reg}}$ there exists a dense open subset $\mathrm{G}'$ of $\mathrm{G}$ such that for all $h\in\mathrm{G}'$ - $\rho_h$ is biregular in $g\cdot p$, - $\rho_{hg}$ is biregular in $p$. Since $\rho_{hg}=\rho_h\circ\rho_g$ the map $\rho_g$ is biregular in $p$. Let us recall that if $\phi\colon X\dashrightarrow Y$ is a rational map, its graph $\Gamma(\phi)$ is defined by $$\Gamma(\phi)=\big\{(x,y)\in X\times Y\,\vert\,x\in X\smallsetminus\mathrm{Base}(\phi)\text{ and } \phi(x)=y\big\}.$$ In particular $\mathrm{pr}_1(\Gamma(\phi))=X\smallsetminus\mathrm{Base}(\phi)$ and $\mathrm{pr}_2(\Gamma(\phi))=\phi(X\smallsetminus\mathrm{Base}(\phi))$. Let $\rho$ be a rational $\mathrm{G}$-action on a variety $X$. Suppose that every point of $X$ is $\mathrm{G}$-regular, that is $X=X_{\text{reg}}$. Then for every $g\in\mathrm{G}$ the graph $\Gamma(\rho_g)$ of $\rho_g$ is closed in $X\times X$. Denote by $\Gamma$ the closure $\overline{\Gamma(\rho_g)}$ of the graph of $\rho_g$ in $X\times X$. Let us prove that for any $x_0\in X\smallsetminus\mathrm{Base}(\rho_g)$ the rational map $\rho_g$ is defined in $x_0$. It is equivalent to prove that for any $(x_0,y_0)$ in $\Gamma$ the morphism $\mathrm{pr}_{1\vert\Gamma}\colon\Gamma\to X$ is biregular in $(x_0,y_0)$. Let $h$ be an element of $\mathrm{G}$ such that $\rho_{hg}$ is biregular in $x_0$ and $\rho_h$ is biregular in $y_0$. Consider the birational map $$\phi=(\rho_{hg},\rho_h)\colon X\times X\dashrightarrow X\times X.$$ If $\phi$ is defined in $(x,y)\in\Gamma(\rho_g)$, $y=g\cdot x$, then $\phi(x,y)=((hg)\cdot x,(hg)\cdot x)$ belongs to the diagonal $$\Delta(X)=\big\{(x,x)\in X\,\vert\,x\in X\big\}$$ of $X\times X$. It follows that $\overline{\phi(\Gamma)}\subseteq\Delta(X)$. Since $\phi$ is biregular in $(x_0,y_0)$, the restriction $\varphi=\phi_{\vert\Gamma}\colon\Gamma\dashrightarrow\Delta(X)$ of $\phi$ to $\Gamma$ is also biregular in $(x_0,y_0)$. By construction $$\rho_{hg}\circ\mathrm{pr}_{1\vert\Gamma}=\mathrm{pr}_{1\vert\Delta(X)}\circ\varphi;$$ indeed $$\xymatrix{ X\times X \ar@{-->}[r]^{\rho_{hg}\times\rho_h} & X\times X \\ \Gamma \ar@{-->}[r]^{\varphi} \ar[d]_{\mathrm{pr}_{1\vert\Gamma}}\ar@{^{(}->}[u] & \Delta(X) \ar[d]^{\mathrm{pr}_{1\vert\Delta(X)}}\ar@{^{(}->}[u] \\ X \ar@{-->}[r]^{\rho_{hg}} & X }$$ But $\rho_{hg}$ is biregular in $\mathrm{pr}_{1\vert\Gamma}(x_0,x_0)$, $\varphi$ is biregular in $(x_0,y_0)$ and $\mathrm{pr}_{1\vert\Delta(X)}$ is an isomorphism, so $\mathrm{pr}_{1\vert\Gamma}$ is biregular in $(x_0,y_0)$. \[lem:Kraft5\] Let $\rho$ be a rational action of $\mathrm{G}$ on a variety $X$. Suppose that there is a dense open subset $\mathcal{U}$ of $X$ such that $$\begin{aligned} & \widetilde{\rho}\colon\mathrm{G}\times\mathcal{U}\to\mathrm{G}\times X, && (g,p)\mapsto(g,\rho(g,p))\end{aligned}$$ defines an open immersion. Then the open dense subset $Y=\displaystyle\bigcup_{g\in\mathrm{G}}g\cdot \mathcal{U}\subseteq X$ carries a regular $\mathrm{G}$-action. Any $\rho_g$ induces an isomorphism $\mathcal{U}\stackrel{\sim}{\to} g\cdot\mathcal{U}$. Therefore $Y=\displaystyle\bigcup_{g\in\mathrm{G}} g\cdot X\subset X$ is stable under all $\rho_g$. By assumption the induced map on $\mathrm{G}\times\mathcal{U}$ is a morphism so the induced map on $\mathrm{G}\times g\cdot \mathcal{U}$ is a morphism for all $g\in\mathrm{G}$. As a result the induced map $\mathrm{G}\times Y\to Y$ is a morphism. ### Construction of a regular model \[thm:Kraft3\] Let $X$ be a variety with a rational action of $\mathrm{G}$. Suppose that every point of $X$ is $\mathrm{G}$-regular. Then there exists a variety $Y$ with a regular $\mathrm{G}$-action and a $\mathrm{G}$-equivariant open immersion. Assume now that $X$ is a variety with a rational $\mathrm{G}$-action $\rho$ such that $X_{\text{reg}}=X$. Consider a finite subset $S=\big\{g_0=e,\,g_1,\,g_2,\,\ldots,\,g_m\big\}$ of $\mathrm{G}$. Denote by $X^{(0)}$, $X^{(1)}$, $\ldots$, $X^{(m)}$ some copies of $X$. Consider the disjoint union $$X(S)=X^{(0)}\cup X^{(1)}\cup\ldots\cup X^{(m)}.$$ Let us define on $X^{(i)}$ the following relations $$\left\{ \begin{array}{ll} \forall\,i\quad p_i\sim p'_i \Longleftrightarrow p_i=p'_i\\ \forall\,i,\,j\quad i\not=j\quad p_i\sim p_j\Longleftrightarrow \rho_{g_j^{-1}g_i} \text{ is defined in $p_i$ and sends $p_i$ to $p_j$} \end{array} \right.$$ This defines an equivalence relation (Lemma \[lem:Kraft3\] is needed to prove the symmetry). Consider $\widetilde{X}(S)=\faktor{X(S)}{\sim}$ the set of equivalence classes endowed with the induced topology. The maps $\iota_i\colon X^{(i)}\to\widetilde{X}(S)$ are open immersions and endow $\widetilde{X}(S)$ with the structure of a variety. Let us fix the open immersion $\iota_0\colon X=X^{(0)}\hookrightarrow\widetilde{X}(S)$. Then $\mathrm{G}$ acts rationally on $\widetilde{X}(S)$ via $\overline{\rho}=\overline{\rho}_S$ such that $\iota_0$ is $\mathrm{G}$-equivariant. Consider any $X^{(i)}$ as the variety $X$ with the rational $\mathrm{G}$-action $$\rho^{(i)}(g,p)=\rho(g_igg_i^{-1},p);$$ by construction of $\widetilde{X}(S)$ the open immersions $$\iota_i\colon X^{(i)}\hookrightarrow\widetilde{X}(S)$$ are all $\mathrm{G}$-equivariant. \[lem:Kraft7\] Let $\widetilde{X}^{(i)}$ be the image of the open immersion $\iota_i\colon X^{(i)}\hookrightarrow\widetilde{X}(S)$. For all $i$ the rational map $\overline{\rho}_{g_i}$ is defined on $\widetilde{X}^{(0)}$. Furthermore $\overline{\rho}_{g_i}\colon\widetilde{X}^{(0)}\stackrel{\sim}{\to}\widetilde{X}^{(i)}$ defines an isomorphism. Consider the open immersion $$\tau_i=\iota_i\circ\iota_0^{-1}\colon\widetilde{X}^{(0)}\hookrightarrow\widetilde{X}(S).$$ Note that $\mathrm{im}\,\tau_i=\widetilde{X}^{(i)}$. Let us check that $\tau_i(\overline{p})=g_i\overline{p}$. It is sufficient to show that it holds on an open dense subset of $\widetilde{X}^{(0)}$. Let $\mathcal{U}\subseteq X$ be the open dense set where $g_i\cdot p$ is defined. Take $p$ in $\mathcal{U}$. On the one hand by definition $$\iota_0(g_i\cdot p)=\iota_i(p);$$ on the other hand $$\iota_0(g_i\cdot p)=g_i\cdot\iota_0(p).$$ As a result $g_i\cdot \iota_0(p)=\iota_i(p)$ and $$\tau_i(\overline{p})=\iota_i(\iota_0^{-1}(\overline{p}))=g_i\cdot \iota_0(\iota_0^{-1}(\overline{p}))=g_i\cdot \overline{p}$$ for any $\overline{p}\in\iota_0(\mathcal{U})$. Set $D=\mathrm{Breg}(\rho)\cap(\mathrm{G}\times X)$. Since $X_{\text{reg}}=X$ for any $p\in X$ there is an element $g$ in $\mathrm{G}$ such that $(g,p)\in D$. As a consequence $\displaystyle\bigcup_{g\in\mathrm{G}}g\cdot D=\mathrm{G}\times X$ where $\mathrm{G}$ acts on $\mathrm{G}\times X$ by left-multiplication on $\mathrm{G}$. Hence $\displaystyle\bigcup_ig_iD=\mathrm{G}\times X$ for a suitable finite subset $$S=\big\{g_0=e,\,g_1,\,g_2,\,\ldots,\,g_m\big\}.$$ Recall that $\widetilde{X}^{(0)}=\mathrm{im}(\iota_0)$. Let $D^{(0)}\subset\mathrm{G}\times\widetilde{X}^{(0)}$ be the image of $D$. Consider the rational map $$\begin{aligned} &\widetilde{\rho}_S\colon\mathrm{G}\times\widetilde{X}^{(0)}\dashrightarrow\mathrm{G}\times\widetilde{X}(S).\end{aligned}$$ The map $(g,p)\mapsto(g,g\cdot p)$ is the composition of $(g,p)\mapsto(g,(g_i^{-1}g)\cdot p)$ and $(g,y)\mapsto(g,g_i\cdot y)$. The first one is biregular on $g_i\cdot D^{(0)}$ and its image is contained in $\mathrm{G}\times\widetilde{X}^{(0)}$; the second is biregular on $\mathrm{G}\times\widetilde{X}^{(0)}$ (Lemma \[lem:Kraft7\]). As $\mathrm{G}\times\widetilde{X}^{(0)}=\displaystyle\bigcup_ig_i\cdot D^{(0)}$ the map $\widetilde{\rho}_S$ is biregular. As a consequence the rational action $\overline{\rho}$ of $\mathrm{G}$ on $\widetilde{X}(S)$ has the property that $$\widetilde{\rho}_S\colon\mathrm{G}\times\widetilde{X}^{(0)}\hookrightarrow\mathrm{G}\times\widetilde{X}(S)$$ defines an open immersion. Lemma \[lem:Kraft5\] allows to conclude. ### Proof of Theorem \[thm:Kraft2\] Let us start with the following statement: \[lem:Kraft8\] Let $X$, $Y$, $Z$ be varieties. Assume that $Z$ is affine. Let $\phi\colon X\times Y\dashrightarrow Z$ be a rational map. Suppose that - there exists an open dense subset $\mathcal{U}$ of $Y$ such that $\phi$ is defined on $X\times\mathcal{U}$; - there exists a dense subset $X'$ of $X$ such that the induced maps $\phi_p\colon\{p\}\times Y\to Z$ are morphisms for all $p\in X'$. Then $\phi$ is a regular morphism. Consider a rational action $\rho\colon\mathrm{G}\to\mathrm{Bir}(X)$ of an algebraic group on a variety $X$. Assume that there is a dense subgroup $\Gamma$ of $\mathrm{G}$ such that $\rho(\Gamma)\subset\mathrm{Aut}(X)$. - Let us first prove that the rational $\mathrm{G}$-action on the open dense set $X_{\text{reg}}\subseteq X$ is regular. For every $p\in X_{\text{reg}}$ there is $g\in\Gamma$ such that $\widetilde{\rho}$ is biregular in $(g,p)$. By assumption for any $h\in\Gamma$ the map $\rho_h$ is biregular on $X$, hence the map $\widetilde{\rho}$ is biregular in $(h,p)$ for any $h\in\Gamma$ (Lemma \[lem:Kraft2\]). Furthermore $h\cdot p$ belongs to $X_{\text{reg}}$ (Proposition \[pro:Kraft3\]), [i.e.]{} $X_{\text{reg}}$ is stable under $\Gamma$. According to Theorem \[thm:Kraft3\] there exists a $\mathrm{G}$-equivariant open immersion $$X_{\text{reg}}\hookrightarrow Y$$ where $Y$ is a variety with a regular $\mathrm{G}$-action. The complement $Y\smallsetminus X_{\text{reg}}$ is closed and $\Gamma$-stable so $Y\smallsetminus X_{\text{reg}}$ is stable under $\overline{\Gamma}=\mathrm{G}$. - From the previous point the rational map $$\rho\colon\mathrm{G}\times X\dashrightarrow X$$ has the following properties: - there is a dense open set $X_{\text{reg}}\subseteq X$ such that $\rho$ is regular on $\mathrm{G}\times X_{\text{reg}}$; - for every $g\in\Gamma$ the rational map $$\begin{aligned} &\rho_g\colon X\to X, && p\mapsto \rho(g,p)\end{aligned}$$ is a regular isomorphism. Lemma \[lem:Kraft8\] implies that $\rho$ is a regular action in case $X$ is affine. Generators and relations of the <span style="font-variant:small-caps;">Cremona</span> group {#chapter:gen} =========================================================================================== As we already say \[thm:noether\] The group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is generated by $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})=\mathrm{PGL}(3,\mathbb{C})$ and the standard quadratic involution $$\sigma_2\colon(z_0:z_1:z_2)\dashrightarrow(z_1z_2:z_0z_2:z_0z_1).$$ This result is well-known as the Theorem of <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span>. <span style="font-variant:small-caps;">Noether</span> was the first mathematician to state this result at the end of the XIXth century. Nevertheless the first exact proof is due to <span style="font-variant:small-caps;">Castelnuovo</span>. <span style="font-variant:small-caps;">Noether</span>’s idea was the following. Let us consider a birational self map $\phi$ of $\mathbb{P}^2_\mathbb{C}$. Take a quadratic birational self map $q$ of $\mathbb{P}^2_\mathbb{C}$ such that the three base-points of $q$ are three base-point of $\phi$ of highest multiplicity. Then $\deg(\phi\circ q)<\deg\phi$. By induction one gets a birational map of degree $1$. But such a quadratic birational map $q$ may not exist. This is for instance the case if one starts with the polynomial automorphism $$(z_0:z_1:z_2)\dashrightarrow(z_1^3-z_0z_2^2:z_1z_2^2:z_2^3).$$ In [@Alexander] <span style="font-variant:small-caps;">Alexander</span> fixes <span style="font-variant:small-caps;">Noether</span>’s proof by introducing the notion of complexity of a map: start with a birational self map $\phi$ of the complex projective plane; one can find a quadratic birational self map $q$ of the complex projective plane such that - either the complexity of $\phi\circ q$ is strictly less that the complexity of $\phi$; - or the complexities of $\phi\circ q$ and $\phi$ are equal but $\#\mathrm{Base}(\phi\circ q)<\#\mathrm{Base}(\phi)$. <span style="font-variant:small-caps;">Alexander</span>’s proof is a proof by induction on these two integers. One consequence of <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> theorem is: the <span style="font-variant:small-caps;">Jonquières</span> group and $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})=\mathrm{PGL}(3,\mathbb{C})$ generate $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. This result is “weaker” nevertheless it has the following nice property: Let $\phi$ be an element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. There exist $j_1$, $j_2$, $\ldots$, $j_k$ in $\mathcal{J}$ and $A$ in $\mathrm{PGL}(3,\mathbb{C})$ such that - $\phi=A\circ j_k\circ j_{k-1}\circ\ldots\circ j_2\circ j_1$; - for any $1\leq i\leq n-1$ $$\deg(A\circ j_k\circ j_{k-1}\circ\ldots\circ j_{i+1}\circ j_i)>\deg(A\circ j_k\circ j_{k-1}\circ\ldots\circ j_{i+2}\circ j_{i+1}).$$ The first presentation of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is given by <span style="font-variant:small-caps;">Gizatullin</span>: The <span style="font-variant:small-caps;">Cremona</span> group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is generated by the set $\mathcal{Q}$ of all quadratic maps. The relations in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ are consequences of relations of the form $q_1\circ q_2\circ q_3=\mathrm{id}$ where $q_1$, $q_2$, $q_3$ are quadratic birational self maps of $\mathbb{P}^2_\mathbb{C}$. In other words we have the following presentation $$\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})=\langle\mathcal{Q}\,\vert\,q_1\circ q_2\circ q_3=\mathrm{id}\text{ $\,\,\forall$ $q_1$, $q_2$, $q_3\in\mathcal{Q}$ such that $q_1\circ q_2\circ q_3=\mathrm{id}$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$} \rangle$$ Two years later <span style="font-variant:small-caps;">Iskovskikh</span> proved the following statement: \[thm:Iskovskikh\] The group $\mathrm{Bir}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ is generated by the group $B$ of birational maps preserving the fibration given by the first projection together with $\tau\colon(z_0,z_1)\mapsto(z_1,z_0)$. Moreover the following relations form a complete system of relations: - relations inside the groups $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ and $B$; - $\left(\tau\circ\left((z_0,z_1)\mapsto\left(z_0,\frac{z_0}{z_1}\right)\right)\right)^3=\mathrm{id}$; - $\left(\tau\circ\left((z_0,z_1)\mapsto(-z_0,z_1-z_0)\right)\right)^3=\mathrm{id}$. In $1994$ <span style="font-variant:small-caps;">Iskovskikh</span>, <span style="font-variant:small-caps;">Kabdykairov</span> and <span style="font-variant:small-caps;">Tregub</span> present a list of generators and relations of $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ over arbitrary perfect field $\Bbbk$ (see [@IskovskikhKabdykairovTregub]). The group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ hasn’t a structure of amalgamated product ([@Cornulier:amalgamatedproduct]). Nevertheless a presentation of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group in the form of a generalized amalgam was given by <span style="font-variant:small-caps;">Wright</span>: \[thm:Wright\] The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is the free product of $\mathrm{PGL}(3,\mathbb{C})$, $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ and $\mathcal{J}$ amalgamated along their pairwise intersections in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Twenty years later <span style="font-variant:small-caps;">Blanc</span> proved: \[thm:Blancrelations\] The group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is the amalgamated product of the <span style="font-variant:small-caps;">Jonquières</span> group with the group of automorphisms of the plane along their intersection, divided by the relation $\sigma_2\circ\tau=\tau\circ\sigma_2$ where $\sigma_2$ is the standard involution and $\tau$ is the involution $(z_0:z_1:z_2)\mapsto(z_1:z_0:z_2)$. As we have seen in Chapter \[Chapter:algebraicsubgroup\] there is an Euclidean topology on the <span style="font-variant:small-caps;">Cremona</span> group ([@BlancZimmermann]). With respect to this topology $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is a <span style="font-variant:small-caps;">Hausdorff</span> topological group. Furthermore the restriction of the Euclidean topology to any algebraic subgroup is the classical Euclidean topology. To show that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is compactly presentable with respect to the Euclidean topology <span style="font-variant:small-caps;">Zimmermann</span> established the following statement: The group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is isomorphic to the amalgamated product of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, $\mathrm{Aut}(\mathbb{F}_2)$, $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ along their pairwise intersection in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ modulo the relation $\tau\circ\sigma_2\circ\tau\circ\sigma_2=\mathrm{id}$ where $\sigma_2$ is the standard involution and $\tau$ the involution $\tau\colon(z_0:z_1:z_2)\mapsto(z_1:z_0:z_2)$. Recently <span style="font-variant:small-caps;">Urech</span> and <span style="font-variant:small-caps;">Zimmermann</span> got a presentation of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group with respect to the generators given by the Theorem of <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span>: \[thm:UrechZimmermann\] The <span style="font-variant:small-caps;">Cremona</span> group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is isomorphic to $$\langle\sigma_2,\,\mathrm{PGL}(3,\mathbb{C})\,\vert\,(\mathcal{R}_1)-(\mathcal{R}_5)\rangle$$ where - $(\mathcal{R}_1)$ $\quad$ $g_1\circ g_2\circ g_3^{-1}=\mathrm{id}$ for all $g_1$, $g_2$, $g_3\in\mathrm{PGL}(3,\mathbb{C})$ such that $g_1\circ g_2=g_3$; - $(\mathcal{R}_2)$ $\quad$ $\sigma_2^2=\mathrm{id}$; - $(\mathcal{R}_3)$ $\quad$ $\sigma_2\circ\eta\circ(\eta\circ\sigma_2)^{-1}=\mathrm{id}$ for all $\eta$ in the symmetric group $\mathfrak{S}_3\subset\mathrm{PGL}(3,\mathbb{C})$ of order $6$ acting on $\mathbb{P}^2_\mathbb{C}$ by coordinate permutations; - $(\mathcal{R}_4)$ $\quad$ $\sigma_2\circ d\circ\sigma_2\circ d=\mathrm{id}$ for all diagonal automorphisms $d$ in the subgroup $\mathrm{D}_2\subset\mathrm{PGL}(3,\mathbb{C})$ of diagonal automorphisms; - $(\mathcal{R}_5)$ $\quad$ $(\sigma_2\circ h)^3=\mathrm{id}$ where $h\colon(z_0:z_1:z_2)\mapsto(z_2-z_0:z_2-z_1:z_2)$. <!-- --> - The relations $(\mathcal{R}_2)$, $(\mathcal{R}_3)$ and $(\mathcal{R}_4)$ occur in the group $\mathrm{Aut}(\mathbb{C}^\times\mathbb{C}^)$ which is given by the group of monomial maps $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. - $(\mathcal{R}_5)$ is a relation from the group $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})^0\simeq\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$ which is considered as a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ by conjugation with the birational equivalence $$\begin{aligned} \mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}&\dashrightarrow&\mathbb{P}^2_\mathbb{C}\\ \big((u_0:u_1),(v_0:v_1)\big)&\dashrightarrow&(u_1v_0:u_0v_1:u_1v_1)\end{aligned}$$ All the results are stated on $\mathbb{C}$ but indeed - [@Cornulier:amalgamatedproduct; @UrechZimmermann; @Gizatullin:relations; @Iskovskikh:generatorsandrelations; @Iskovskikh:relations; @Wright; @Blanc:relations] work for the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group over an algebraically closed field, - [@Zimmermann:presentation] works for the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group over a locally compact local field. In the first section we recall the proof of <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> due to <span style="font-variant:small-caps;">Alexander</span>. In the second section we give an outline of the proof of the result of [@Cornulier:amalgamatedproduct] that says that the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group does not decompose as a non-trivial amalgam. We also recall the proof of Theorem \[thm:Wright\]. The third section is devoted to generators and relations in the <span style="font-variant:small-caps;">Cremona</span> group. We first give a sketch of the proof of Theorem \[thm:Blancrelations\]. We also give a sketch of the proof of Theorem \[thm:UrechZimmermann\]. We then explain why there is no <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> theorem in higher dimension. <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> theorem --------------------------------------------------------------------------------------------------------------------------- Let us now deal with the proof of Theorem \[thm:noether\] given by <span style="font-variant:small-caps;">Alexander</span> ([@Alexander]). Recall the two following formulas proved in §\[sec:geodef\]. Consider a birational self map $\phi$ of $\mathbb{P}^2_\mathbb{C}$ of degree $\nu$; denote by $p_1$, $p_2$, $\ldots$, $p_k$ the base-points of $\phi$ and by $m_{p_i}$ the multiplicity of $p_i$. Then $$\label{eq:cremona3} \displaystyle\sum_{i=0}^km_{p_i}=3(\nu-1)$$ $$\label{eq:cremona1} \displaystyle\sum_{i=0}^km_{p_i}^2=\nu^2-1.$$ From (\[eq:cremona1\]) and (\[eq:cremona3\]) one gets $$\label{eq:cremona2} \displaystyle\sum_{i=0}^km_{p_i}\big(m_{p_i}-1\big)=(\nu-1)(\nu-2).$$ Consider a birational self map of $\mathbb{P}^2_\mathbb{C}$ of degree $\nu$. If $\nu=1$, then according to (\[eq:cremona3\]) the map $\phi$ is an automorphism of $\mathbb{P}^2_\mathbb{C}$. So let us now assume that $\nu>1$. Let $\Lambda_\phi$ be the linear system associated to $\phi$. Denote by $p_1$, $p_2$, $\ldots$, $p_k$ the base-points (in $\mathbb{P}^2_\mathbb{C}$ or infinitely near) of $\phi$ and $m_{p_i}$ their multiplicity. Up to reindexation let us assume that $$m_{p_0}\geq m_{p_1}\geq \ldots \geq m_{p_k}\geq 1.$$ <span style="font-variant:small-caps;">Alexander</span> introduced the notion of complexity: the complexity of $\Lambda_\phi$ is the integer $2c=\nu-m_{p_0}$. It is the number of points except $p_0$ that belong to the intersection of a general line passing through $p_0$ and a curve of $\Lambda_\phi$. One has - $2c\geq 0$: the degree of the hypersurfaces of $\Lambda_\phi$ is $\nu$ so a point has multiplicity $\leq\nu$; - furthermore $2c\geq 1$; indeed if an homogeneous polynomial of degree $\nu$ has a point of multiplicity $\nu$, then the hypersurface given by this polynomial is the union of $\nu$ lines. Set $$C=\big\{p\in\mathrm{Base}(\phi)\smallsetminus\{p_0\}\,\vert\,m_p>c\big\}$$ and $$n=\#\,C.$$ <span style="font-variant:small-caps;">Bezout</span> theorem implies that the line through $p_0$ and $p_1$ intersects any curve of $\Lambda_\phi$ in $\nu$ points (counted with multiplicity). Furthermore it intersects any curve of $\Lambda_\phi$ at $p_0$ with multiplicity $m_{p_0}$. Consequently $m_{p_1}\leq \nu-m_{p_0}=2c$ and $$\label{eq:beurk} c<m_{p_k}\leq \ldots\leq m_{p_2}\leq m_{p_1}\leq 2c$$ \[lem:three\] There are at least three base-points of multiplicity $>c=\frac{\nu-m_{p_0}}{2}$, [i.e.]{} $n\geq 2$; hence $m_{p_0}>\frac{\nu}{3}$. Furthermore if $\nu\geq 3$, then $p_1$, $p_2$, $\ldots$, $p_n$ are not aligned. According to (\[eq:cremona1\]) and (\[eq:cremona2\]) one has on the one hand $$c\displaystyle\sum_{i=0}^km_{p_i}(m_{p_i}-1)-(c-1)\displaystyle\sum_{i=0}^km_{p_i}^2=\displaystyle\sum_{i=0}^k m_{p_i}(cm_{p_i}-c-cm_{p_i}+m_{p_i})=\displaystyle\sum_{i=0}^km_{p_i}(m_{p_i}-c)$$ and on the other hand $$(\nu-1)(\nu-2)c-(\nu^2-1)(c-1)=(\nu-1)(\nu c-2c-\nu c+\nu-c+1)=(\nu-1)(\nu-3c+1).$$ As a result $$\label{eq:bloub} \displaystyle\sum_{i=0}^km_{p_i}(m_{p_i}-c)=(\nu-1)(\nu-3c+1)$$ Since $m_{p_{n+i}}\leq c$ for any $i>0$ one gets $$\displaystyle\sum_{i=0}^nm_{p_i}(m_{p_i}-c)\geq\displaystyle\sum_{i=0}^km_{p_i}(m_{p_i}-c).$$ According to (\[eq:bloub\]) $$\displaystyle\sum_{i=0}^nm_{p_i}(m_{p_i}-c)\geq(\nu-1)(\nu-3c+1)=\nu(\nu-3c)+3c-1.$$ But $3c-1\geq\frac{1}{2}>0$ so $$\displaystyle\sum_{i=0}^nm_{p_i}(m_{p_i}-c)>\nu(\nu-3c)=\nu(m_{p_0}-c).$$ Consequently $$\displaystyle\sum_{i=0}^nm_{p_i}(m_{p_i}-c)>\nu(m_{p_0}-c)-m_{p_0}(m_{p_0}-c)=(\nu-m_{p_0})(m_{p_0}-c)=2c(m_{p_0}-c).$$ As $2c\geq m_{p_i}$ for any $i\geq 1$ (see (\[eq:beurk\])) one gets $2c\displaystyle\sum_{i=1}^n(m_{p_i}-c)>2c(m_{p_0}-c)$ and $$2c\displaystyle\sum_{i=1}^n(m_{p_i}-c)>2c(m_{p_0}-c)$$ that is $$\label{eq:bipbip} m_{p_i}-c>m_{p_0}-c$$ since $c>0$. But $m_{p_1}\leq m_{p_0}$ so $n\geq 2$. Therefore $m_{p_0}+m_{p_1}+m_{p_2}>3\left(\frac{\nu-m_{p_0}}{2}\right)$ and $m_{p_0}>\frac{\nu}{3}$. Let us assume that $n\geq 3$; then (\[eq:bipbip\]) can be rewritten $$\displaystyle\sum_{i=1}^n m_{p_i}-nc>m_{p_0}-c=\nu-3c$$ and $\displaystyle\sum_{i=1}^nm_{p_i}>\nu+(n-3)c\geq\nu$. A general quadratic birational self map of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ centered at $p$, $q$ $r$ is the map, up to linear automorphism, that blows up the three distinct points $p$, $q$, $r$ of $\mathbb{P}^2_\mathbb{C}$ and blows down the strict transform of the lines $(pq)$, $(qr)$ and $(pr)$. These lines are thus sent onto points denoted $p'$, $q'$ and $r'$. The line $(p'q')$ (resp. $(q'r')$, resp. $(p'r')$) corresponds to the exceptional line of the blow up of $r$ (resp. $p$, resp. $q$). \[lem:four\] Compose $\phi$ with a general quadratic birational self map of $\mathbb{P}^2_\mathbb{C}$ centered at $p_0$, $q$ and $r$ where $p_0$ is the base-point of $\phi$ of maximal multiplicity. The complexity of the new system is equal to the complexity of the old system if and only if $p_0'$ is of maximal multiplicity. If it is not the case, then the complexity of the new system is strictly less than the complexity of the old one. The complexity of the new system is $2c'=\nu'-m'_{\text{max}}$ where $m'_{\text{max}}$ denotes the highest multiplicity of the base-points of the new system. Then $$\begin{aligned} 2c'&=&\nu'-m'_{\text{max}}\\ &=&2\nu-m_{p_0}-m_q-m_r-m'_{\text{max}}\\ &=&\nu-m_{p_0}+(\nu-m_q-m_r)-m'_{\text{max}}\\ &=&\nu-m_{p_0}+m_{p'_0}-m'_{\text{max}}\\ &=&2c+m_{p'_0}-m'_{\text{max}}.\end{aligned}$$ Hence $c'\leq c$ and $c'=c$ if and only if $m_{p'_0}=m'_{\text{max}}$. \[lem:five\] If there exist two points $p_i$ and $p_j$ in $C=\big\{p_1,\,p_2,\,\ldots,\,p_n\big\}$ such that - $p_i$ and $p_j$ are not infinitely near; - $p_i$ and $p_0$ are not infinitely near; - $p_j$ and $p_0$ are not infinitely near. Then there exists a general quadratic birational self map of $\mathbb{P}^2_\mathbb{C}$ such that after composition with $\phi$ - either the complexity of the system decreases, - or $\#\,C=n$ decreases by $2$. Suppose that there exist two points $p_i$ and $p_j$ in $C=\big\{p_1,\,p_2,\,\ldots,\,p_n\big\}$ such that - $p_i$ and $p_j$ are not infinitely near; - $p_i$ and $p_0$ are not infinitely near; - $p_j$ and $p_0$ are not infinitely near. Let us now compose $\phi$ with a general quadratic birational self map of $\mathbb{P}^2_\mathbb{C}$ centered at $p_0$, $p_i$ and $p_j$. The degree of the new linear system $\Lambda'_\phi$ is $\nu'=2\nu-m_{p_0}-m_{p_i}-m_{p_j}$. Let us remark that $$\begin{aligned} \nu'&=&2\nu-m_{p_0}-m_{p_i}-m_{p_j}\\ &=&\nu+(\nu-m_{p_0}-m_{p_i}-m_{p_j})\\ &=&\nu+(2c-m_{p_i}-m_{p_j})\\ &<&\nu\end{aligned}$$ [i.e.]{} the degree has decreased. The new linear system $\Lambda'_\phi$ has complexity $c'$ and we denote by $C'$ the set of points of multiplicity $>c'$. The points $p_0$, $p_i$ and $p_j$ are no more points of indeterminacy; the other base-points and their multiplicity do not change. There are three new base-points which are $p'_0$, $p'_i$ and $p'_j$. By definition the multiplicity of $p'_0$ (resp. $p'_i$, resp. $p'_j$) is equal to the number of intersection points (counted with multiplicity) between the corresponding contracted line and the strict transform of a general curve of the linear system. From <span style="font-variant:small-caps;">Bezout</span> theorem we thus have $$\left\{ \begin{array}{lll} m_{p'_0}=\nu-m_{p_i}-m_{p_j}\\ m_{p'_i}=\nu-m_{p_0}-m_{p_j}\\ m_{p'_j}=\nu-m_{p_0}-m_{p_i} \end{array} \right.$$ - If $p'_0$ is not the point of highest multiplicity, the complexity of the system decreases (Lemma \[lem:four\]); - otherwise if $p'_0$ is the point of highest multiplicity, then the complexity remains constant (Lemma \[lem:four\]). Furthermore $p'_0$ belongs to $C'$ (Lemma \[lem:three\]). Since $m_{p_i}>c$, $m_{p_j}>c$ and $\nu-m_{p_0}=2c$, then $m_{p'_i}<c$ and $m_{p'_j}<c$, [i.e.]{} $p'_i\not\in C'$ and $p'_j\not\in C'$. As a consequence $n'=n-2$. \[lem:six\] Assume there exists a base-point $p_k$ in $C$ that is not infinitely near $p_0$. Then after composition by a general quadratic birational map, one can disperse the points above $p_0$ and $p_k$. The complexity of the system does not change, the cardinal of $C$ does not change. There is no point infinitely near $p'_0$. Consider a point $q$ of the complex projective plane such that - the lines $(p_0q)$ and $(p_kq)$ contain no base-point; - there is no point infinitely near $p_0$ in the direction of the line $(p_0q)$; - there is no point infinitely near $p_k$ in the direction of the line $(p_kq)$. Compose $\phi$ with a general quadratic birational map centered at $p_0$, $p_k$ and $q$. The degree of the new linear system is $$\nu'=2\nu-m_{p_0}-m_{p_k}=\nu+2c-m_{p_k}\geq \nu.$$ The point $p'_0$ is the point of highest multiplicity: $$\left\{ \begin{array}{lll} m_{p'_0}=\nu-m_{p_k}\geq\nu-m_{p_0}=2c\geq m_{p_1}\\ m_{p'_k}=\nu-m_{p_0}=2c>c\\ m_{q'}=\nu-m_{p_0}-m_{p_k}=2c-m_{p_k}<c \end{array} \right.$$ hence the complexity remains constant (Lemma \[lem:four\]). Note that $\#\,C'=\#\,C$. The assumptions on $q$ allow to say that a point infinitely near $p_k$ (resp. $p_0$) is not transformed in a point infinitely near $p'_0$. Similarly a point infinitely near $p_k$ (resp. $p_0$) is not transformed in a point infinitely near $q'$. \[lem:seven\] Assume that all the points of $C$ are above the point of highest multiplicity $p_0$. Then one can disperse them with a general quadratic birational self map; in other words there is no base-point of $C'$ infinitely near the point $p'_0$ of highest multiplicity of the new system. The cardinal $n$ increases by $2$ but the complexity of the system remains constant. Take two points $q$ and $r$ in $\mathbb{P}^2_\mathbb{C}$ such that the lines $(p_0r)$, $(p_0q)$ and $(rq)$ - do not contain base-points; - are not in the direction of the points infinitely near $p_0$. The degree of the new linear system is $\nu'=2\nu-m_{p_0}>\nu$. Since the curves of the system do not pass through $q$ and $r$ <span style="font-variant:small-caps;">Bezout</span> theorem implies that $m_{p'_0}=\nu$; it is thus the point of highest multiplicity. Furthermore $$2c'=2\nu-m_{p_0}-\nu=2c.$$ Any curve of the linear system intersects $(p_0r)$ and $(p_0q)$ in $\nu-m_{p_0}=2c$ points. As a result $m_{r'}=m_{q'}=2c>c=c'$. Moreover $r'$ and $q'$ belong to $C'$ and $n'=n+2$. The points infinitely near $p_0$ have been dispersed onto the line $(q'r')$. As there is no base-point on the line $(qr)$ there is no base-point infinitely near $p'_0$. Let us consider a birational self map $\phi$ of $\mathbb{P}^2_\mathbb{C}$ of degree $\nu$. Denote by $p_0$, $p_1$, $\ldots$, $p_k$ its base-points and by $\Lambda_\phi$ the linear system associated to $\phi$. Let $m_{p_i}$ be the multiplicity of $p_i$ and assume up to reindexation that $$m_{p_0}\geq m_{p_1}\geq \ldots\geq m_{p_k}.$$ Recall that the complexity of the system $\Lambda_\phi$ is $c$ where $2c=\nu-m_{p_0}$, that $$C=\big\{p\in\mathrm{Base}(\phi)\smallsetminus\{p_0\}\,\vert\, m_p>c\big\}$$ and that $n=\#\,C$. We will now compose $\phi$ with a sequence of general quadratic birational maps in order to decrease the complexity until the complexity equals to $1$. #### Step 1 {#step-1 .unnumbered} If all points of $C$ are above $p_0$, let us apply Lemma \[lem:seven\]. One gets that $p'_0$ has no more infinitely near base-points and that $n'=n+2$. Let us now apply Lemma \[lem:six\] until the points of $C'$ are all distinct; note that $C'$ and $n'$ do not change. According to Lemma \[lem:three\] the points of $C'$ are not aligned. Let us take two of these points, denoted by $p_i$ and $p_j$ such that there exist two base-points $p_k$ and $p_\ell$ with the following property: $p_k$ and $p_\ell$ do not belong to the lines $(p'_0p_i)$, $(p'_0p_j)$ and $(p_ip_j)$. Apply two times Lemma \[lem:five\] to the points $p_k$ and $p_\ell$. If the complexity decreases (the first or the second time anyway), then let us start this process again; otherwise the first application of Lemma \[lem:five\] yields to $n'=n$ and the second to $n'=n-2$. Furthermore there is no more base-point of $C'$ infinitely near $p'_0$ and we go to Step 2. #### Step 2 {#step-2 .unnumbered} We distinguish two possibilities: Step 2i. Either there are two base-points in $C$ that are not infinitely near and one applies Lemma \[lem:five\]. If the complexity decreases, come back to Step 1, otherwise come back to Step 2. Step 2ii. Or let us apply Lemma \[lem:six\], then there are two base-points that are not infinitely near and one can apply Step 2i. According to Lemma \[lem:three\] if $\nu>1$, then $\#\,C\geq 3$. As a result Step 1 and Step 2 allow to decrease the complexity. When the complexity is $1$, the point $p'_0$ has the highest multiplicity and from Lemmas \[lem:five\], \[lem:six\] and \[lem:seven\] one gets that $\#\,C$ decreases until $0$. In other words our system has at most one base-point. From (\[eq:cremona3\]) and (\[eq:cremona1\]) one gets that $\nu=1$ and that there is no base-point. Amalgamated product and <span style="font-variant:small-caps;">Cremona</span> group ----------------------------------------------------------------------------------- ### It is not an amalgamated product of two groups Let us recall that the group $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of polynomial automorphisms of the plane is the amalgamated product of the affine group $\mathrm{Aff}_2=\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})\cap\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ and the group $\mathcal{J}_{\mathbb{A}^2_\mathbb{C}}=\mathcal{J}\cap\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ along their intersection. On the contrary $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is not the amalgamated product of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ and $\mathcal{J}$. Indeed there exist elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of finite order which are neither conjugate to an element of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, nor to an element of $\mathcal{J}$ (see [@Blanc:cyclic]), contrary to the case of amalgamated products. More precisely <span style="font-variant:small-caps;">Cornulier</span> proved that the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group does not decompose as a non-trivial amalgam ([@Cornulier:amalgamatedproduct]); we will give a sketch of the proof in this section. A graph $\Gamma$ consists of two sets $X$ and $Y$, and two applications $$\begin{aligned} & Y\to X\times X,\quad y\mapsto (o(y),t(y)) && Y\to Y,\quad y\mapsto\overline{y}\end{aligned}$$ such that: $$\forall\, y\in Y \qquad \overline{\overline{y}}=y,\qquad \overline{y}\not=y,\qquad o(y)=t(y).$$ An element of $X$ is a vertex of $\Gamma$; an element $y\in Y$ is an oriented edge, and $\overline{y}$ is the reversed edge. The vertex $o(y)=t(\overline{y})$ is the origin of $y$, and the vertex $t(y)=o(\overline{y})$ is the terminal vertice. These two vertices are called the extremities of $y$. An orientation of a graph $\Gamma$ is a part $Y_+$ of $Y$ such that $Y$ is the disjoint union of $Y_+$ and $\overline{Y_+}$. An oriented graph is defined, up to isomorphism, by the data of two sets $X$ and $Y_+$, and an application $Y_+\to X\times X$. The set of edges of the corresponding graph is $Y=Y_+\bigsqcup\overline{Y_+}$. A graph is connected if two vertices are the extremities of at least one path. - Let $n$ be an integer. Let us consider the oriented graph ![image](ch.pdf) It has $n+1$ vertices $0$, $1$, $\ldots$, $n$ and the orientation is given by the $n$ egdes $[i,i+1]$, $0\leq i<n$ with $o([i,i+1])=i$ and $t([i,i+1])=i+1$. - Let $n\geq 1$ be an integer. Consider the oriented graph given by ![image](cir.pdf) The set of vertices is $\faktor{\mathbb{Z}}{n\mathbb{Z}}$, and the orientation is given by the $n$ edges $[i,i+1]$, $i\in\faktor{\mathbb{Z}}{n\mathbb{Z}}$, with $o([i,i+1])=i$ and $t([i,i+1])=i+1$. A path of length $n$ in a graph $\Gamma$ is a morphism from $\mathrm{Ch}_n$ to $\Gamma$. A cycle of length $n$ in a graph is a subgraph isomorphic to $\mathrm{Cir}_n$. A tree is a non-empty, connected graph without cycle. A group $\mathrm{G}$ is said to have property (FA) if every action of $\mathrm{G}$ on a tree has a global fixed point. A geodesic metric space is a metric space if given any two points there is a path between them whose length equals the distance between the points. A real tree can be defined in the following equivalent ways ([@Chiswell]): - a geodesic metric space which is $0$-hyperbolic in the sense of <span style="font-variant:small-caps;">Gromov</span>; - a uniquely geodesic metric space for which $[a,c]\subset[a,b]\cup[b,c]$ for all $a$, $b$ and $c$; - a geodesic metric space with no subspace homeomorphic to the circle. In a real tree a ray is a geodesic embedding of the half line. An end is an equivalence class of rays modulo being at bounded distance. For a group of isometries of a real tree, to stably fix an end means to pointwise stabilize a ray modulo eventual coincidence ([i.e.]{} it fixes the end as well as the corresponding <span style="font-variant:small-caps;">Busemann</span> function[^3]). A group has property (FR) if for every isometric action on a complete real tree every element has a fixed point. \[rem:FRFA\] Property (FR) implies property (FA). \[lem:propertyFR\] Let $\mathrm{G}$ be a group. Property (FR) has the following equivalent characterizations: - for every isometric action of $\mathrm{G}$ on a complete real tree every finitely generated subgroup has a fixed point; - every isometric action of $\mathrm{G}$ on a complete real tree either has a fixed point, or stably fixes a point at infinity. A group $\mathrm{G}$ decomposes as a non-trivial amalgam if $\mathrm{G}\simeq\mathrm{G}_1\ast_\mathrm{H}\mathrm{G}_2$ with $\mathrm{G}_1\not=\mathrm{H}\not=\mathrm{G}_2$. \[thm:serreFA\] A group $\mathrm{G}$ has property (FA) if and only if it does not decompose as a non-trivial amalgam. In the Appendix of [@Cornulier:amalgamatedproduct] the author has shown that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ satisfies the first assertion of Lemma \[lem:propertyFR\], hence: \[thm:notamalgamated\] The <span style="font-variant:small-caps;">Cremona</span> group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ has property (FR). According to Remark \[rem:FRFA\] the group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ thus has property (FA). From Theorem \[thm:serreFA\] one gets that: The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group does not decompose as a non-trivial amalgam. Let us give the main steps of the proof of Theorem \[thm:notamalgamated\]. From now on $\mathcal{T}$ is a complete real tree and all actions on $\mathcal{T}$ are isometric. #### Step 1 {#step-1-1 .unnumbered} Let $p_0$, $p_1$, $\ldots$, $p_k$ be points of $\mathcal{T}$ and $s\geq 0$. Suppose that the following equality holds $$d(p_i,p_j)=s\vert i-j\vert$$ for all $i$, $j$ such that $\vert i-j\vert\leq 2$. Then it holds for all $i$ and $j$. #### Step 2 {#step-2-1 .unnumbered} If $d\geq 3$, then $\mathrm{SL}(d,\mathbb{C})$ has property (FR). In particular if $d\geq 3$, then $\mathrm{PGL}(d,\mathbb{C})$ has property (FR). #### Step 3 {#step-3 .unnumbered} Let us recall that a torus $\mathrm{T}$ in a compact <span style="font-variant:small-caps;">Lie</span> group $\mathrm{G}$ is a compact, connected, abelian <span style="font-variant:small-caps;">Lie</span> subgroup of $\mathrm{G}$ (and therefore isomorphic to the standard torus $\mathbb{T}^n$ for some integer $n$). Given a torus $\mathrm{T}$, the <span style="font-variant:small-caps;">Weyl</span> group of $\mathrm{G}$ with respect to $\mathrm{T}$ can be defined as the normalizer of $\mathrm{T}$ modulo the centralizer of $\mathrm{T}$. A <span style="font-variant:small-caps;">Cartan</span> subgroup of an algebraic group is one of the subgroups whose <span style="font-variant:small-caps;">Lie</span> algebra is a <span style="font-variant:small-caps;">Cartan</span> subalgebra. For connected algebraic groups over $\mathbb{C}$ a <span style="font-variant:small-caps;">Cartan</span> subgroup is usually defined as the centralizer of a maximal torus. Let $\mathrm{C}$ be the normalizer of the standard <span style="font-variant:small-caps;">Cartan</span> subgroup of $\mathrm{PGL}(3,\mathbb{C})$, [i.e.]{} the semi-direct product of the diagonal matrices by the <span style="font-variant:small-caps;">Weyl</span> group (of order $6$). Set $\varsigma\colon (z_0,z_1)\mapsto(1-z_0,1-z_1)$. The group generated by $\varsigma$ and $\mathrm{C}$ coincides with $\mathrm{PGL}(3,\mathbb{C})$: $$\langle \mathrm{C},\varsigma\rangle=\mathrm{PGL}(3,\mathbb{C}).$$ #### Step 4 {#step-4 .unnumbered} Let $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ act on $\mathcal{T}$ so that $\mathrm{PGL}(3,\mathbb{C})$ has no fixed point and has a (unique) stably fixed end. Then $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ fixes this unique end. #### Step 5 {#step-5 .unnumbered} Note that $\sigma_2=\big(\varsigma\circ\sigma_2\big)\circ\varsigma\circ\big(\varsigma\circ\sigma_2\big)^{-1}$. Since $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})=\langle\sigma_2,\,\mathrm{PGL}(3,\mathbb{C})\rangle$ the groups $\mathrm{H_1}=\mathrm{PGL}(3,\mathbb{C})$ and $\mathrm{H}_2=\sigma_2\circ\mathrm{PGL}(3,\mathbb{C})\circ\sigma_2^{-1}$ generate $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let us consider an action of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ on $\mathcal{T}$. By Steps 2 and 4 it is sufficient to consider the case when $\mathrm{PGL}(3,\mathbb{C})$ has a fixed point. Let us prove that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ has a fixed point; suppose by contradiction that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ has no fixed point. Denote by $\mathcal{T}_i$ the set of fixed points of $\mathrm{H}_i$, $i=1$, $2$. These two trees are exchanged by $\sigma_2$, and as $\mathrm{H}_1$ and $\mathrm{H}_2$ generate $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ they are disjoint. Denote by $\mathcal{S}=[p_1,p_2]$, $p_i\in\mathcal{T}_i$, the minimal segment joining the two trees, and by $s>0$ its length. The segment $\mathcal{S}$ is thus fixed by $\mathrm{C}\subset\mathrm{H}_1\cap\mathrm{H}_2$, and reversed by $\sigma_2$. Step $1$ implies that for all $k\geq 1$, the distance between the points $p_1$ and $(\sigma_2\circ\varsigma)^kp_1$ is exactly $sk$. This contradicts the fact that $(\sigma_2\circ\varsigma)^3=\mathrm{id}$. ### It is an amalgamated product of three groups In [@Wright] the author shows that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})=\mathrm{Aut}_{\mathbb{C}}\mathbb{C}(z_0,z_1)$ acts on a $2$-dimensional simplicial complex $C$, which has as vertices certain models in the function field $\mathbb{C}(z_0,z_1)$ and whose fundamental domain consists of one face $F$. This yields a structure description of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ as an amalgamation of three subgroups along pairwise intersections. The subgroup $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ acts on $C$ by restriction; more precisely the face $F$ has an edge $E$ satisfying the following property: the $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$-translates of $E$ form a tree $T$, and the action of $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ on $T$ yields the well-known structure theory for $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ as an amalgamated product ([@Jung]). Let us give some details. Recall that $$\begin{aligned} & \mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})=\big(\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})\big)\rtimes\faktor{\mathbb{Z}}{2\mathbb{Z}}, && \mathcal{J}=\mathrm{PGL}(2,\mathbb{C})\ltimes\mathrm{PGL}(2,\mathbb{C}(z_0)).\end{aligned}$$ It is based on Theorem \[thm:Iskovskikh\]. Denote by $\mathrm{G}$ be the group obtained by amalgamating $\mathrm{PGL}(3,\mathbb{C})$, $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$, $\mathcal{J}$ along their pairwise intersections in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let $\tau$ be the involution $\tau\colon(z_0,z_1)\mapsto(z_1,z_0)$. Consider the group homomorphism $\alpha\colon\mathrm{G}\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ restricting to the identity on $\mathrm{PGL}(3,\mathbb{C})\cup\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})\cup\mathcal{J}$. As $\mathrm{im}\,\alpha$ contains $\mathcal{J}$ and $\tau\in\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ Theorem \[thm:Iskovskikh\] implies that $\alpha$ is surjective. Since $\big\{\mathrm{id},\,\tau\big\}\subset\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ Theorem \[thm:Iskovskikh\] gives a map $\widetilde{\beta}$ from the free product $\big\{\mathrm{id},\,\tau\big\}\ast\mathcal{J}$ to $\mathrm{G}$. Since $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})\subset\mathrm{G}$ the equality $$\tau\circ(\varphi_0,\varphi_1)\circ\tau=(\varphi_1,\varphi_0)\qquad\forall\, (\varphi_0,\varphi_1)$$ also holds in $\mathrm{G}$. Let us now prove that in $\mathrm{G}$ we have $(\tau\circ\varepsilon)^3=\sigma_2$ where $\varepsilon\colon(z_0,z_1)\mapsto\left(z_0,\frac{z_0}{z_1}\right)$. First note that the equality $\varepsilon=\rho\circ\sigma_2$, where $\rho\colon(z_0,z_1)\mapsto\left(\frac{1}{z_0},\frac{z_1}{z_0}\right)$, holds in $\mathcal{J}$ and so in $\mathrm{G}$. On the one hand $\sigma_2$ and $\rho$ commute in $\mathcal{J}$ thus in $\mathrm{G}$, on the other hand $\sigma_2$ and $\tau$ commute in $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ hence in $\mathrm{G}$. Therefore one has the following equality in $\mathrm{G}$ $$\label{eq:rel} (\tau\circ\varepsilon)^3=(\tau\circ\rho\circ\sigma_2)^3=(\tau\circ\rho)^3\circ\sigma_2^3$$ The maps $\tau$ and $\rho$ belong to $\mathrm{PGL}(3,\mathbb{C})$ and $(\tau\circ\rho)^3=\mathrm{id}$ in $\mathrm{PGL}(3,\mathbb{C})$; as a consequence $(\tau\circ\rho)^3=\mathrm{id}$ in $\mathrm{G}$. One has $\sigma_2^3=\sigma_2$ in $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ so in $\mathrm{G}$. From (\[eq:rel\]) one gets $(\tau\circ\varepsilon)^3=\sigma_2$ in $\mathrm{G}$. Consequently $\widetilde{\beta}$ induces a map $\beta\colon\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\to\mathrm{G}$ with the following property: $\beta$ restricts to the identity on $\mathcal{J}$ and $\big\{\mathrm{id},\,\tau\big\}$. According to Theorem \[thm:Iskovskikh\], $\alpha\circ\beta=\mathrm{id}$. The image of $\beta$ contains $\mathcal{J}\subset\mathrm{G}$ and $\tau\in(\mathrm{PGL}(3,\mathbb{C})\cap\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}))\subset\mathrm{G}$. But both $\mathrm{PGL}(3,\mathbb{C})$ and $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ are generated by their intersection with $\mathcal{J}$ (in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$) together with $\tau$; hence $\mathrm{PGL}(3,\mathbb{C})$ and $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$ belong to $\mathrm{im}\,\beta$. As $\mathrm{G}$ is generated by $\mathrm{PGL}(3,\mathbb{C})\cup\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})\cup\mathcal{J}$, $\beta$ is surjective. Therefore $\alpha$ is an isomorphism ($\alpha^{-1}=\beta$). The amalgamated product group structure of Theorem \[thm:Wright\] reflects the fact that it acts on a simply connected $2$-dimensional simplicial complex. This follows from a higher dimensional analogue of <span style="font-variant:small-caps;">Serre</span>’s tree theory (see for instance [@Soule; @Swan]). Let us detail it. A simplicial complex $\mathcal{K}$ is a finite collection of non-empty finite sets such that if $X\in\mathcal{K}$ and $\emptyset\not=Y\subseteq X$ then $Y\in\mathcal{K}$. The union of all members of $\mathcal{K}$ is denoted by $V(\mathcal{K})$. The elements of $V(\mathcal{K})$ are called the vertices of $\mathcal{K}$. The elements of $\mathcal{K}$ are called the simplices of $\mathcal{K}$. The dimension of a simplex $S\in\mathcal{K}$ is $\dim S=\vert S\vert-1$. The dimension of $\mathcal{K}$ is the maximum dimension of any simplex in $\mathcal{K}$. Admissible models A model is a reduced, irreducible, separated $\mathbb{C}$-scheme having function field $\mathbb{C}(z_0,z_1)$. Consider the set of models $S$ satisfying one of the three properties - $S\simeq \mathbb{P}^2_\mathbb{C}$, - $S\simeq \mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$, - $S\simeq\mathbb{P}^1_\Bbbk$ for some subfield $\Bbbk$ of $\mathbb{C}(z_0,z_1)$ necessarily of pure transcendance degree $1$ over $\mathbb{C}$. Such a $\mathbb{C}$-scheme $S$ will be called an admissible model. In the first (resp. second, resp. third) case, we say that $S$ is $\mathbb{P}^2_\mathbb{C}$ (resp. $S$ is a $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$, resp. $S$ is a $\mathbb{P}^1_\Bbbk$). The complex $C$ $C$ is constructed using as vertices the set of admissible models. The three models $S$, $V$ and $R$, where $S$ is a $\mathbb{P}^2_\mathbb{C}$, $V$ is a $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ and $R$ is a $\mathbb{P}^1_\Bbbk$, determine a face when there exist two distinct points $p$ and $q$ on $S$ such that - $V$ is the $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ ($\simeq\mathbb{F}_0$) obtained by blowing up $S$ at $p$ and $q$, then blowing down the proper transform of the line in $S$ containing $p$ and $q$; - $R$ is the generic $\mathbb{P}^1_\mathbb{C}$ obtained by blowing up $S$ at $p$. If $S$ is the standard $\mathbb{P}^2_\mathbb{C}$, $p=(0:1:0)$ and $q=(1:0:0)$, then $V$ is the standard $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$, and $R$ the standard $\mathbb{P}^1_{\mathbb{C}(z_0)}$. The standard models form a face called the standard face in $C$. Fundamental domain Note that from the construction of $C$ the group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ acts on $C$ without inverting any edge or rotating any face. A fondamental domain for the action is given by any one face ![image](domaine.pdf) If as before we choose $S$ to be the standard $\mathbb{P}^2_\mathbb{C}$, $p=(0:1:0)$ et $q=(1:0:0)$ one gets the standard face. For this choice the centralizer of $S$, $V$ and $R$ are respectively $\mathrm{PGL}(3,\mathbb{C})$, $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$, $\mathcal{J}$. Let us recall that two simplices $S_0$ and $S_n$ are $k$-connected if there is a sequence of simplices $S_0$, $S_1$, $S_2$, $\ldots$, $S_n$ such that any two consecutive ones share a $k$-face, [i.e.]{} they have at least $k+1$ vertices in common. The complex $\mathcal{K}$ is $k$-connected if any two simplices in $\mathcal{K}$ of dimension $\geq k$ are $k$-connected. <span style="font-variant:small-caps;">Wright</span> establishes the two following results ([@Wright]): - the simplicial complex $C$ is $1$-connected; - the complex $C$ contains the $\mathrm{Aut}(\mathbb{A}^2)$-tree. Two presentations of the <span style="font-variant:small-caps;">Cremona</span> group ------------------------------------------------------------------------------------ ### A simple set of generators and relations for $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ {#subsection:nc1} In [@Blanc:relations] <span style="font-variant:small-caps;">Blanc</span> gives a simple set of generators and relations for the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Namely he shows: The group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is the amalgamated product of the <span style="font-variant:small-caps;">Jonquières</span> group with the group $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ of automorphisms of the plane, divided by the relation $\sigma_2\circ\tau=\tau\circ\sigma_2$ where $\tau\colon(z_0:z_1:z_2)\mapsto(z_1:z_0:z_2)$. <span style="font-variant:small-caps;">Blanc</span>’s proof is inspired by <span style="font-variant:small-caps;">Iskovskikh</span>’s proof but <span style="font-variant:small-caps;">Blanc</span> stays on $\mathbb{P}^2_\mathbb{C}$. It is clear that $\sigma_2\circ\tau=\tau\circ\sigma_2$ so it suffices to prove that no other relation holds. <span style="font-variant:small-caps;">Blanc</span> first establishes the following statement: \[lem:tec1\] Let $\varphi$ be an element of $\mathcal{J}$ such that $\big\{p_1=(1:0:0),\,q\big\}\subset\mathrm{Base}(\varphi)$ where $q$ is a proper point of $\mathbb{P}^2_\mathbb{C}\smallsetminus\{p_1\}$. If $\nu\in\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ exchanges $p_1$ and $q$, then - $\psi=\nu\circ\varphi\circ\nu^{-1}$ belongs to $\mathcal{J}$, - the relation $\nu\circ\varphi^{-1}=\psi^{-1}\circ\nu$ is generated by the relation $\sigma_2\circ\tau=\tau\circ\sigma_2$ in the amalgamated product of $\mathcal{J}$ and $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$. Let $\phi$ be an element of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})\ast_{\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})\cap\mathcal{J}}\mathcal{J}$ modulo the relation $\sigma_2\circ\tau=\tau\circ\sigma_2$. Write $\phi$ as $$\begin{aligned} j_r\circ a_r\circ j_{r-1}\circ a_{r-1}\circ\ldots\circ j_1\circ a_1\end{aligned}$$ where $j_i\in\mathcal{J}$ and $a_i\in\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ for $i=1$, $\ldots$, $r$. Note that this decomposition is of course not unique. Let $\Lambda_0$ be the linear system of lines of $\mathbb{P}^2_\mathbb{C}$. For any $i=1$, $\ldots$, $r$ let us denote by $\Lambda_i$ the linear system $(j_i\circ a_i\circ\ldots\circ j_1\circ a_1)(\Lambda_0)$, and by $d_i$ the degree of $\Lambda_i$. Set $$\begin{aligned} & D=\max\big\{d_i\,\vert\,i=1,\,\ldots,\,r\big\}, && n=\max\big\{i\,\vert\,d_i=D\big\}, && k=\displaystyle\sum_{i=1}^n\big((\deg j_i)-1\big).\end{aligned}$$ Recall that $j_i$ belongs to $\mathcal{J}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and satisfies the following property: $$\deg j_i=\deg j_i(\Lambda_0)=\deg j_i^{-1}(\Lambda_0).$$ In particular $\deg j_i=1$ if and only if $j_i\in\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$. Let us give an interpretation of $k$: the number $k$ determines the complexity of the word $j_n\circ a_n\circ j_{n-1}\circ a_{n-1}\circ\ldots\circ j_1\circ a_1$ which corresponds to the birational self map $j_i\circ a_i\circ\ldots\circ j_1\circ a_1$ of the highest degree. Let us now give the strategy of the proof. If $D=1$, then each $j_i$ is an automorphism of $\mathbb{P}^2_\mathbb{C}$ and the word $\phi$ is equal to an element of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ in the amalgamated product. Since $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})\hookrightarrow\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ this eventuality is clear. Assume now that $D>1$, and prove the result by induction on the pairs $(D,k)$ (we consider the lexicographic order). #### Fact {#fact .unnumbered} We can suppose that $$\begin{aligned} &j_{n+1},\,j_n\in\mathcal{J}\smallsetminus\mathrm{Aut}(\mathbb{P}^2_\mathbb{C}), && a_{n+1}\in\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})\smallsetminus\mathcal{J}.\end{aligned}$$ #### Remark {#remark .unnumbered} The point $p=(1:0:0)$ is the base-point of the pencil associated to $\mathcal{J}$. As $a_{n+1}\not\in\mathcal{J}$, one has $a_{n+1}(p)\not=p$. #### Properties of the <span style="font-variant:small-caps;">Jonquières</span> maps {#properties-of-the-jonquières-maps .unnumbered} Since $j_n$, $j_{n+1}$ do not belong to $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, then $\deg j_n>1$, $\deg j_{n+1}>1$. Set $D_L=\deg j_{n+1}$, $D_R=\deg j_n$. The maps $j_{n+1}$ and $j_n$ preserve the pencil of lines through $p$. Furthermore $p$ is a base-point of $j_{n+1}$ (resp. $j_n$) of multiplicity $D_L-1$ (resp. $D_R-1$). Since $j_{n+1}^\pm(\Lambda_0)$ (resp. $j_n^\pm(\Lambda_0)$) is the image of the system $\Lambda_0$ it is a system of rational curves with exactly one free intersection point. The system $j_{n+1}^\pm(\Lambda_0)$ (resp. $j_n^\pm(\Lambda_0)$) has $2D_L-2$ (resp. $2D_R-2$) base-points distinct from $p$, which all have multiplicity $1$. Set $\Omega_L=(j_{n+1}\circ a_{n+1})^{-1}(\Lambda_0)$ and $\Omega_R=(j_n\circ a_n)(\Lambda_0)$. Since the automorphisms $a_{n+1}$, $a_n$ are changes of coordinates the following properties hold: - $\deg\Omega_L=D_L$ and $\ell_0=a_{n+1}^{-1}(p)\not=p$ is a base-point of $\Omega_L$ of multiplicity $D_L-1$; - $\deg\Omega_R=D_R$ and $r_0=p$ is a base-point of $\Omega_R$ of multiplicity $D_R-1$. The author uses these systems to compute the degrees $d_{n+1}$, resp. $d_{n-1}$ of the systems $\Lambda_{n+1}=(j_{n+1}\circ a_{n+1})(\Lambda_n)$, resp. $\Lambda_{n-1}=(a_n^{-1}\circ j_n^{-1})(\Lambda_n)$. Indeed for any $i$ the integer $d_i$ coincides with the degree of $\Lambda_i$ which is on the one hand the intersection of $\Lambda_i$ with a general line, on the other hand the free intersection of $\Lambda_i$ with $\Lambda_0$. So $d_{n+1}$ (resp. $d_{n-1}$) is the free intersection of $\Lambda_{n+1}=(j_{n+1}\circ a_{n+1})(\Lambda_n)$ (resp. $\Lambda_{n-1}=(a_n^{-1}\circ j_n^{-1})(\Lambda_n)$) with $\Lambda_0$ but also the free intersection of $\Lambda_n$ with $\Omega_L$ (resp. $\Omega_R$). Denote by $m(q)$ the multiplicity of a point $q$ as a base-point of $\Lambda_n$. Let $\ell_1$, $\ldots$, $\ell_{2D_L-2}$ (resp. $r_1$, $\ldots$, $r_{2D_R-2}$) be the base-points of $\Omega_L$ (resp. $\Omega_R$). Assume that up to reindexation $m(\ell_i)\geq m(\ell_{i+1})$ (resp. $m(r_i)\geq m(r_{i+1})$) and if $\ell_i$ (resp. $r_i$) is infinitely near to $\ell_j$ (resp. $r_j$), then $i>j$. The following equalities hold: $$\label{eq1} \left\{ \begin{array}{ll} d_{n+1}=D_Ld_n-(D_L-1)m(\ell_0)-\displaystyle\sum_{i=1}^{2D_L-2} m(\ell_i)<d_n\\ d_{n-1}=D_Rd_n-(D_R-1)m(r_0)-\displaystyle\sum_{i=1}^{2D_R-2} m(r_i)<d_n \end{array} \right.$$ Inequalities (\[eq1\]) imply $$\label{eq1b} \left\{ \begin{array}{ll} m(\ell_0)+m(\ell_1)+m(\ell_2)>d_n\\ m(r_0)+m(r_1)+m(r_2)\geq d_n \end{array} \right.$$ #### First case: $m(\ell_0)\geq m(\ell_1)$ and $m(r_0)\geq m(r_1)$ {#first-case-mell_0geq-mell_1-and-mr_0geq-mr_1 .unnumbered} Let $q$ be a point in $\big\{\ell_1,\,\ell_2,\,r_1,\,r_2\big\}\smallsetminus\{\ell_0,\,r_0\}$ with the maximal multiplicity $m(q)$ and so that $q$ is a proper point of $\mathbb{P}^2_\mathbb{C}$ or infinitely near to $\ell_0$ or $r_0$. - Either $\ell_1=r_0$, $m(q)\geq m(\ell_2)$ and $m(\ell_0)+m(r_0)+m(q)\geq m(\ell_0)+m(\ell_1)+m(\ell_2)>d_n$ by (\[eq1b\]). - Or $\ell_1\not=r_0$, $m(q)\geq m(\ell_1)\geq m(\ell_2)$ hence $m(\ell_0)+m(q)>\frac{2d_n}{3}$. The inequalities $m(r_0)\geq m(r_1)\geq m(r_2)$ imply $m(r_0)\geq \frac{d_n}{3}$ and then $m(\ell_0)+m(r_0)+m(q)>d_n$ holds. The inequality $m(\ell_0)+m(r_0)+m(q)>d_n$ implies that $\ell_0$, $r_0$ and $q$ are not aligned and there exists an element $\theta$ in $\mathcal{J}$ of degree $2$ with base points $\ell_0$, $r_0$, $q$. Note that $$\deg\theta(\Lambda_n)=2d_n-m(\ell_0)-m(r_0)-m(q)<d_n.$$ Let us recall that the automorphism $a_{n+1}$ of $\mathbb{P}^2_\mathbb{C}$ sends $\ell_0$ onto $r_0=p$. Take $\nu\in\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})\cap~\mathcal{J}$ such that $\nu$ fixes $r_0=p$ and sends $a_{n+1}(r_0)$ onto $\ell_0$. Replace $a_{n+1}$ (resp. $j_{n+1}$) by $\nu\circ a_{n+1}$ (resp. $j_{n+1}\circ\nu^{-1}$); we can thus assume that $a_{n+1}$ exchanges $\ell_0$ and $r_0$. As a consequence according to Lemma \[lem:tec1\] and modulo the relation $\sigma_2\circ\tau=\tau\circ\sigma_2$ $$j_{n+1}\circ a_{n+1}\circ j_n=j_{n+1}\circ a_{n+1}\circ\theta^{-1}\circ\theta\circ j_n=(j_{n+1}\circ\widetilde{\theta}^{-1})\circ a_{n+1}\circ(\theta\circ j_n)$$ where $\widetilde{\theta}=a_{n+1}\circ \theta\circ a_{n+1}^{-1}\in\mathcal{J}$. Both $j_{n+1}\circ \widetilde{\theta}^{-1}$ and $\theta\circ j_n$ belong to $\mathcal{J}$, but $a_{n+1}$ belongs to $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$. Since $\theta(\Lambda_n)=(\theta\circ j_n)(\Lambda_{n-1})$ has degree $<d_n$ this rewriting decreases the pair $(D,k)$. #### Second case: $m(\ell_0)<m(\ell_1)$ or $m(r_0)<m(r_1)$. {#second-case-mell_0mell_1-or-mr_0mr_1. .unnumbered} The author comes back to the first case by changing the writing of $\phi$ in the amalgamated product and modulo the relation $\sigma_2\circ\tau=\tau\circ\sigma_2$ without changing $(D,k)$ but reversing the inequalities. ### An other set of generators and relations for $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ {#subsection:nc2} The idea of the proof of Theorem \[thm:UrechZimmermann\] is the same as in [@Iskovskikh:generatorsandrelations; @Iskovskikh:relations; @Blanc:relations]. The authors study linear systems of compositions of birational maps of the complex projective plane and use the presentation of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ given in Theorem \[thm:Blancrelations\]. Before giving the proof of Theorem \[thm:UrechZimmermann\] let us state the following: \[pro:UrechZimmermann\] Let $\phi_1$, $\phi_2$, $\ldots$, $\phi_n$ be some elements of $\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}$. Suppose that $\phi_n\circ\sigma_2\circ\phi_{n-1}\circ\sigma_2\circ\ldots\circ\sigma_2\circ\phi_1=\mathrm{id}$ as maps. Then this expression is generated by relations $(\mathcal{R}_1)$-$(\mathcal{R}_5)$. Let $\mathrm{G}$ be the group generated by $\sigma_2$ and $\mathrm{PGL}(3,\mathbb{C})$ divided by the relations $(\mathcal{R}_1)$-$(\mathcal{R}_5)$ $$\mathrm{G}=\langle\sigma_2,\,\mathrm{PGL}(3,\mathbb{C})\,\vert\,(\mathcal{R}_1)-(\mathcal{R}_5)\rangle.$$ Denote by $\pi\colon\mathrm{G}\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ the canonical homomorphism that sends generators onto generators. Proposition \[pro:UrechZimmermann\] asserts that sending an element of $\mathcal{J}$ onto its corresponding word in $\mathrm{G}$ is well defined. Hence there exists a homomorphism $w\colon\mathcal{J}\to\mathrm{G}$ such that $$\pi\circ w=\mathrm{id}_{\mathcal{J}}.$$ In particular $w$ is injective. The universal property of the amalgamated product implies that there exists a unique homomorphism $$\varphi\colon\mathrm{PGL}(3,\mathbb{C})\ast_{\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}}\mathcal{J}\to\mathrm{G}$$ such that the following diagram commutes $$\xymatrix{ & & &\mathrm{G} \\ \mathcal{J}\ar@/^/[rrru] \ar[rr]^w & & \mathrm{PGL}(3,\mathbb{C})\ast_{\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}}\mathcal{J}\ar[ru]^{\varphi} & \\ \mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}\ar@{^{(}->}[u] \ar@{^{(}->}[rr] & & \mathrm{PGL}(3,\mathbb{C})\ar@{^{(}->}[u]\ar@/_2pc/[ruu] & }$$ According to Theorem \[thm:Blancrelations\] the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is isomorphic to $$\mathrm{PGL}(3,\mathbb{C})\ast_{\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}}\mathcal{J}$$ divided by the relation $\tau\circ\sigma_2\circ\tau\circ\sigma_2$ where $\tau\colon(z_0:z_1:z_2)\mapsto(z_1:z_0:z_2)$. Note that this relation holds as well in $\mathrm{G}$. As a consequence $\varphi$ factors through the quotient $$\mathrm{PGL}(3,\mathbb{C})\ast_{\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}}\mathcal{J}\Big/\langle\tau\circ\sigma_2\circ\tau\circ\sigma_2\rangle.$$ This yields a homomorphism $\overline{\varphi}\colon\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\to\mathrm{G}$. More precisely the homomorphisms $\pi\colon\mathrm{G}\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and $\overline{\varphi}\colon\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\to\mathrm{G}$ both send generators to generators $$\left\{ \begin{array}{ll} & \pi(\sigma_2)=\sigma_2\text{ and } \pi(A)=A\,\,\,\,\,\forall\,A\in\mathrm{PGL}(3,\mathbb{C})\\ &\overline{\varphi}(\sigma_2)=\sigma_2\text{ and } \overline{\varphi}(A)=A\,\,\,\,\,\forall\,A\in\mathrm{PGL}(3,\mathbb{C}) \end{array} \right.$$ The homomorphisms $\pi$ and $\overline{\varphi}$ are thus isomorphisms that are inverse to each other. Let us give some Lemmas and Remarks that allow to give a proof of Proposition \[pro:UrechZimmermann\]. In [@AlberichCarraminana] the author gave a general formula for the degree of a composition of two elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ but the multiplicities of the base-points of the composition is hard to compute in general. If we impose that one of the two maps has degree $2$ then it is a rather straight forward computation ([@AlberichCarraminana]). Denote by $m_p(\phi)$ the multiplicity of $\phi$ at the point $p$. For any $\phi\in\mathcal{J}$ of degree $d$ one has - $m_{(1:0:0)}(\phi)=d-1$, - $m_p(\phi)=1$ $\forall\,p\in\mathrm{Base}(\phi)\smallsetminus\{(1:0:0)\}$, so according to [@AlberichCarraminana] one has: \[lem:tecnik1\] Let $\phi$, resp. $\psi$ be a <span style="font-variant:small-caps;">Jonquières</span> map of degree $2$, resp. $d$. Let $p_1$, $p_2$ be the base-points of $\phi$ different from $(1:0:0)$ and $q_1$, $q_2$ be the base-points of $\phi^{-1}$ different from $(1:0:0)$ such that the pencil of lines through $p_i$ is sent by $\phi$ onto the pencil of lines through $q_i$. Then - $\deg(\psi\circ\phi)=d+1-m_{q_1}(\psi)-m_{q_2}(\psi)$, - $m_{(1:0:0)}(\psi\circ\phi)=d-m_{q_1}(\psi)-m_{q_2}(\psi)=\deg(\psi\circ\phi)-1$, - $m_{p_i}(\psi\circ\phi)=1-m_{q_j}(\psi)$ if $i\not=j$. \[rem:tecnik2\] These equalities can be translate as follows when $\Lambda_\psi$ denotes the linear system of $\psi$: - $\deg(\psi\circ\phi)=\deg(\phi^{-1}(\Lambda_\psi))=d+1-m_{q_1}(\Lambda_\psi)-m_{q_2}(\Lambda_\psi)$, - $m_{(1:0:0)}\big(\phi^{-1}(\Lambda_\psi)\big)=d-m_{q_1}(\Lambda_\psi)-m_{q_2}(\Lambda_\psi)=\deg\big(\phi^{-1}(\Lambda_\psi)\big)-1$, - $m_{p_i}\big(\phi^{-1}(\Lambda_\psi)\big)=1-m_{q_j}(\Lambda_\psi)\qquad i\not=j$. But the multiplicity of $\Lambda_\psi$ in a point different from $(1:0:0)$ is $0$ or $1$ so $$\left\{ \begin{array}{lll} \text{eiter }\deg\phi^{-1}(\Lambda_\psi)=\deg(\Lambda_\psi)+1\text{ and }m_{q_1}(\Lambda_\psi)=m_{q_2}(\Lambda_\psi)=0\\ \text{or }\deg\phi^{-1}(\Lambda_\psi)=\deg(\Lambda_\psi)\text{ and }m_{q_1}(\Lambda_\psi)+m_{q_2}(\Lambda_\psi)=1\\ \text{or }\deg\phi^{-1}(\Lambda_\psi)=\deg(\Lambda_\psi)-1\text{ and }m_{q_1}(\Lambda_\psi)=m_{q_2}(\Lambda_\psi)=1 \end{array} \right.$$ Furthermore <span style="font-variant:small-caps;">Bezout</span> theorem implies that $(1:0:0)$ and any other base-points of $\psi$ are not collinear; indeed $(1:0:0)$ is a base-point of multiplicity $d-1$, all other base-points of multiplicity $1$ (since $\psi$ belongs to $\mathcal{J}$) and a general member of $\Lambda_\psi$ intersects a line in $d$ points counted with multiplicity. \[lem:tecnik3\] Let $\phi$ be an element of $\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}$. Suppose that $\sigma_2\circ\phi\circ\sigma_2$ is linear. Then $\sigma_2\circ\phi\circ\sigma_2$ is generated by the relations $(\mathcal{R}_1)$, $(\mathcal{R}_3)$ and $(\mathcal{R}_4)$. By Lemma \[lem:tecnik1\] to say that $\sigma_2\circ\phi\circ\sigma_2$ is linear means that $$\mathrm{Base}(\sigma_2\circ\phi)=\mathrm{Base}(\sigma_2)=\big\{(1:0:0),\,(0:1:0),\,(0:0:1)\big\}.$$ Since $\phi$ belongs to $\mathcal{J}$ it fixes the point $(1:0:0)$ and so permutes $(0:1:0)$ and $(0:0:1)$. As a result there exist $\varphi$ in $\mathfrak{S}_3\cap\mathcal{J}$ and $d$ in $\mathrm{D}_2$ such that $\phi=d\circ\varphi$. Hence $$\sigma_2\circ\phi\circ\sigma_2\stackrel{(1)}{=}\sigma_2\circ d\circ\varphi\circ\sigma_2\stackrel{(3),\,(4)}{=}d^{-1}\circ\varphi.$$ \[lem:tecnik4\] Let $\phi$ be an element of $\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}$. Suppose that no three of the base-points of $\sigma_2$ and $\sigma_2\circ\phi$ are collinear. Then there exist $\varphi$, $\psi$ in $\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}$ such that $\sigma_2\circ\phi\circ\sigma_2=\varphi\circ\sigma_2\circ\psi$. Furthermore this expression is generated by relations $(\mathcal{R}_1)$, $(\mathcal{R}_3)$, $(\mathcal{R}_4)$ and $(\mathcal{R}_5)$. The assumption $\deg(\sigma_2\circ\phi\circ\sigma_2)=2$ implies that $\sigma_2$ and $\sigma_2\circ\phi$ have exactly two common base-points (Lemma \[lem:tecnik1\]), among them $(1:0:0)$ because $\sigma_2\circ\phi$ and $\sigma_2$ belong to $\mathcal{J}$. One can assume up to coordinate permutation that the second point is $(0:1:0)$. More precisely there exist $t_1$, $t_2$ in $\mathfrak{S}_3\cap\mathcal{J}$ such that $t_1\circ t_2$ fixes $(1:0:0)$ and $(0:1:0)$. As a result $$t_1\circ\phi\circ t_2\colon(z_0:z_1:z_2)\mapsto(a_1z_0+a_2z_2:b_1z_1+b_2z_2:cz_2)$$ for some complex numbers $a_1$, $a_2$, $b_1$, $b_2$, $c$. Since no three of the base-points of $\sigma_2$ and $\sigma_2\circ\phi$ are collinear, $a_2b_2$ is non-zero. There thus exist $d_1$, $d_2$ in $\mathrm{D}_2$ such that $$t_1\circ\phi\circ t_2=d_1\circ\zeta\circ d_2.$$ We get $$\begin{aligned} \sigma_2\circ\phi\circ\sigma_2&=&\sigma_2\circ t_1^{-1}\circ t_1\circ\phi\circ t_2\circ t_2^{-1}\circ\sigma_2 \\ &\stackrel{(1)}{=}& \sigma_2\circ t_1^{-1}\circ d_1\circ\zeta\circ d_2\circ t_2^{-1}\circ\sigma_2 \\ &\stackrel{(3),(4)}{=}& t_1^{-1}\circ d_1^{-1}\circ\sigma_2\circ\zeta\circ\sigma_2\circ d_2^{-1}\circ t_2^{-1}\\ &\stackrel{(5)}{=}& t_1^{-1}\circ d_1^{-1}\circ\zeta\circ\sigma_2\circ\zeta\circ d_2^{-1}\circ t_2^{-1} \\\end{aligned}$$ Finally $\varphi=t_1^{-1}\circ d_1^{-1}\circ\zeta$ and $\psi=\zeta\circ d_2^{-1}\circ t_2^{-1}$ suit. \[lem:tecnik5\] Let $\varphi_1$, $\varphi_2$, $\ldots$, $\varphi_n$ be elements of $\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}$. Then there exist $\psi_1$, $\psi_2$, $\ldots$, $\psi_n$ in $\mathrm{PGL}(3,\mathbb{C})\cap\mathcal{J}$ and $\phi$ in $\mathcal{J}$ such that $$\phi\circ\varphi_n\circ\sigma_2\circ\varphi_{n-1}\circ\sigma_2\circ\ldots\circ\sigma_2\circ\varphi_1\circ\phi^{-1}=\psi_n\circ\sigma_2\circ\psi_{n-1}\circ\ldots\circ\sigma_2\circ\psi_1,$$ and - the above relation is generated by relations $(\mathcal{R}_1)$-$(\mathcal{R}_5)$, - $\deg(\sigma_2\circ\psi_i\circ\sigma_2\circ\ldots\circ\sigma_2\circ\psi_1)=\deg(\sigma_2\circ\phi_i\circ\sigma_2\circ\ldots\circ\sigma_2\circ\phi_1)$ for all $1\leq i\leq n$, - $(\sigma_2\circ \psi_i\circ\sigma_2\circ \psi_{i-1}\circ\ldots\circ\sigma_2\circ \psi_1)^{-1}$ does not have any infinitely near base-points for all $1\leq i\leq n$. Let us introduce similar notations as in the proof of Theorem \[thm:Blancrelations\]. Let $\Lambda_0$ be the complete linear system of lines in $\mathbb{P}^2_\mathbb{C}$ and for $1\leq i\leq j$ let $\Lambda_i$ be the following linear system $$\Lambda_i:=\sigma_2\circ\varphi_{i-1}\circ\sigma_2\circ\ldots\circ\sigma_2\circ\varphi_1(\Lambda_0).$$ Set $\delta_i:=\deg\Lambda_i$, $D_i:=\max\big\{\delta_i\,\vert\,i=1,\,2,\,\ldots,\,j\big\}$, $n:=\max\big\{i\,\vert\,\delta_i=D\big\}$. Consider the lexicographic order. Let us prove the result by induction on pairs of positive integers $(D,n)$. If $D=1$, then $j=1$, and there is nothing to prove. Assume now that $D>1$. We can suppose that for $1\leq i\leq j$ the map $$\big(\phi_i\circ\sigma_2\circ\phi_{i-1}\circ\sigma_2\circ\ldots\circ\sigma_2\circ\phi_1\big)^{-1}$$ does not have any infinitely near base-points (Lemma \[lem:tecnik5\]). Furthermore we can do this without increasing the pair $(D,n)$. Hence any $\Lambda_i$, $1\leq i\leq j$, does not have any infinitely near base-points. The maps $\phi_i$ are <span style="font-variant:small-caps;">Jonquières</span> ones so fix $(1:0:0)$. The maps $\sigma_2\circ\phi_i$ and $\sigma_2$ always have $(1:0:0)$ as common base-points. In particular $\deg(\sigma_2\circ\phi_i\circ\sigma_2)\leq 3$ for any $1\leq i\leq j$ (Lemma \[lem:tecnik1\]). Let us now deal with the three distinct cases: $\deg(\sigma_2\circ\phi_n\circ\sigma_2)=1$, $\deg(\sigma_2\circ\phi_n\circ\sigma_2)=2$, $\deg(\sigma_2\circ\phi_n\circ\sigma_2)=3$. - First case: $\deg(\sigma_2\circ\phi_n\circ\sigma_2)=1$. According to Lemma \[lem:tecnik3\] the word $\sigma_2\circ\phi_n\circ\sigma_2$ can be replaced by the linear map $\phi'_n=\sigma_2\circ\phi_n\circ\sigma_2$ using relations $(\mathcal{R}_1)$, $(\mathcal{R}_3)$ and $(\mathcal{R}_4)$. We thus get a new pair $(D',n')$ with $D'\leq D$; moreover if $D=D'$, then $n'<n$. - Second case: $\deg(\sigma_2\circ\phi_n\circ\sigma_2)=2$. The maps $\sigma_2$ and $\phi_n\circ\sigma_2$ have exactly two common base-points, one of them being $(1:0:0)$. One can assume that the other one is $(0:1:0)$. More precisely there are two coordinate permutations $t_1$ and $t_2$ in $\mathfrak{S}_3\cap\mathcal{J}$ such that $t_1\circ\phi_n\circ t_2$ fixes $(1:0:0)$ and $(0:1:0)$, that is $$t_1\circ\phi_n\circ t_2\colon(z_0:z_1:z_2)\mapsto(a_1z_0+a_2z_2:b_1z_1+b_2z_2:cz_2)$$ for some $a_1$, $a_2$, $b_1$, $b_2$, $c$ in $\mathbb{C}$. Using $(\mathcal{R}_1)$ and $(\mathcal{R}_3)$ we get $$\begin{aligned} & &\phi_j\circ\sigma_2\circ\ldots\circ\phi_{n+1}\circ\sigma_2\circ t_1^{-1}\circ t_1\circ\phi_n\circ t_2\circ t_2^{-1}\circ\sigma_2\circ\ldots\circ\phi_1\\ & & \hspace{2mm}=\phi_j\circ\sigma_2\circ\ldots\circ\sigma_2\circ(\phi_{n+1}\circ t_1^{-1})\circ\sigma_2\circ(t_1\circ\sigma_2\circ t_2)\circ\sigma_2\circ(t_2^{-1}\circ\phi_{n-1})\circ\sigma_2\circ\ldots\circ\phi_1\end{aligned}$$ The pair $(D,n)$ is unchanged. Let us thus assume that $t_1=t_2=\mathrm{id}$ and $$\phi_n\colon(z_0:z_1:z_2)\mapsto(a_1z_0+a_2z_2:b_1z_1+b_2z_2:cz_2).$$ Recall that by assumption for any $1\leq i\leq n$ the maps $\phi_i\circ\sigma_2\circ\phi_{i-1}\circ\sigma_2\circ\ldots\circ\sigma_2\circ\phi_1$ have no infinitely other base-points. As a result $\Lambda_n$ has no infinitely near base-points. The product $a_2b_2$ is non-zero. Assume by contradiction that $a_2b_2=0$. Then $q:=\phi_n^{-1}(0:0:1)$ is a base-point of $\sigma_2\circ\phi_n$ that lies on a line contracted by $(\sigma_2\circ\phi_{n-1})^{-1}$. By Remark \[rem:tecnik2\] one has $$D-1=\delta_{n+1}=D+1-m_{(0:1:0)}(\Lambda_n)-m_q(\Lambda_n).$$ In particular $m_q(\Lambda_n)=2-m_{(0:1:0)}(\Lambda_n)=1$. As $q\not\in\mathrm{Base}(\sigma_2)$ one has: $q\not\in\mathrm{Base}\big((\sigma_2\circ\phi_{n-1})^{-1}\big)$. Its proper image by $(\sigma_2\circ\phi_{n-1})^{-1}$ is thus a base-point of $\Lambda_{n-1}$. But $a_2b_2=0$; as a result $q$ is an infinitely near point: contradiction. If $a_2b_2$ is non-zero, then no three of the base-points of $\sigma_2$ and $\sigma_2\circ\phi_n$ are collinear. According to Lemma \[lem:tecnik5\] there exist $\psi$ and $\varphi$ in $\mathrm{PGL}(3,\mathbb{C})$ such that the word $\sigma_2\circ\phi_n\circ\sigma_2$ can be replaced by the word $\psi\circ\sigma_2\circ\varphi$ using $(\mathcal{R}_1)$, $(\mathcal{R}_3)$, $(\mathcal{R}_4)$ and $(\mathcal{R}_5)$. We thus get a new pair $(D',n')$ where $D'\leq D$; moreover if $D=D'$, then $n'<n$. - Third case: $\deg(\sigma_2\circ\phi_n\circ\sigma_2)=3$. See [@UrechZimmermann]. Let us give an application of this new presentation ([@UrechZimmermann]). In [@Gizatullin:rep] <span style="font-variant:small-caps;">Gizatullin</span> has considered the following question: can a given group homomorphism $\varphi\colon\mathrm{PGL}(3,\mathbb{C})\to\mathrm{PGL}(n+1,\mathbb{C})$ be extended to a group homomorphism $\Phi\colon\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\to\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ ? He answers yes when $\varphi$ is the projective representation induced by the regular action of $\mathrm{PGL}(3,\mathbb{C})$ on the space of plane conics, plane cubics, or plane quartics. To construct these homomorphisms <span style="font-variant:small-caps;">Gizatullin</span> uses the following construction. Denote by $\mathrm{Sym}(n,\mathbb{C})$ the $\mathbb{C}$-algebra of symmetric $n\times n$ matrices. Define $\mathbb{S}(2,n)$ as the quotient $\big(\mathrm{Sym}(n,\mathbb{C})\big)^3\big/\mathrm{GL}(n,\mathbb{C})$ where the regular action of $\mathrm{GL}(n,\mathbb{C})$ is given by $$C\cdot(A_0,A_1,A_2)=\big(CA_0{}^{t}\!\,C,\,CA_1{}^{t}\!\,C,\,CA_2{}^{t}\!\,C\big).$$ The variety $\mathbb{S}(2,n)$ is a rational variety, and $\dim\mathbb{S}(2,n)=\frac{(n+1)(n+2)}{2}-1$. The variety $\mathbb{S}(2,n)$ has thus the same dimension as the space of plane curves of degree $n$. An element $A=(A_0,A_1,A_2)$ of $\mathrm{PGL}(3,\mathbb{C})$ induces an automorphism on $(\mathrm{Sym}(n,\mathbb{C}))^3$ by $$\phi(A_0,A_1,A_2):=\big(\phi_0(A_0,A_1,A_2),\phi_1(A_0,A_1,A_2),\phi_2(A_0,A_1,A_2)\big).$$ This automorphism commutes with the action of $\mathrm{GL}(n,\mathbb{C})$; we thus obtain a regular action of $\mathrm{PGL}(3,\mathbb{C})$ on $\mathbb{S}(2,n)$. Theorem \[thm:UrechZimmermann\] allows to give a short proof of the following statement: The regular action of $\mathrm{PGL}(3,\mathbb{C})$ extends to a rational action of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Define the birational action of $\sigma_2$ on $\mathbb{S}(2,n)$ by $$(A_0,A_1,A_2)\dashrightarrow(A_0^{-1},A_1^{-1},A_2^{-1}).$$ According to Theorem \[thm:UrechZimmermann\] to see that this indeed defines a rational action of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ on $\mathbb{S}(2,n)$ it is sufficient to see that $(\mathcal{R}_1)$-$(\mathcal{R}_5)$ are satisfied which is the case. ### Why no <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> theorem in higher dimension ? {#subsection:hudsonandpan} Let us give an idea of the proof of the fact that there is no <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> theorem in higher dimension: \[thm:hudsonpan\] Any set of group generators of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$, $n\geq 3$, contains uncountably many elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\smallsetminus\mathrm{PGL}(n+1,\mathbb{C})$. Let us first recall the following construction of <span style="font-variant:small-caps;">Pan</span> which given a birational self map of $\mathbb{P}^n_\mathbb{C}$ allows one to construct a birational self map of $\mathbb{P}^{n+1}_\mathbb{C}$. First introduce some notations: let $P\in\mathbb{C}[z_0,z_1,\ldots,z_n]_d$, $Q\in\mathbb{C}[z_0,z_1,\ldots,z_n]_\ell$ and $R_0$, $R_1$, $\ldots$, $R_{n-1}\in\mathbb{C}[z_0,z_1,\ldots,z_n]_{d-\ell}$ be some homogeneous polynomials of degree $d$, resp. $\ell$, resp. $d-\ell$. Consider $\widetilde{\psi}_{P,Q,R}$ and $\widetilde{\psi}_R$ the rational maps given by $$\begin{aligned} & \widetilde{\psi}_{P,Q,R}\colon(z_0:z_1:\ldots:z_n)\dashrightarrow(QR_0:QR_1:\ldots:QR_{n-1}:P),\\ & \widetilde{\psi}_R\colon(z_0:z_1:\ldots:z_n)\dashrightarrow(R_0:R_1:\ldots:R_{n-1}).\end{aligned}$$ \[lem:construction\] Let $d$ and $\ell$ be some integers such that $d\leq\ell+1\leq 2$. Take $Q\in\mathbb{C}[z_0,z_1,\ldots,z_n]_\ell$ and $P\in\mathbb{C}[z_0,z_1,\ldots,z_n]_d$ without common factors. Let $R_1$, $R_2$, $\ldots$, $R_n$ be some elements of $\mathbb{C}[z_0,z_1,\ldots,z_{n-1}]_{d-\ell}$. Assume that $$\begin{aligned} & P=z_nP_{d-1}+P_d&& Q=z_nQ_{\ell-1}+Q_\ell \end{aligned}$$ with $P_{d-1}$, $P_d$, $Q_{\ell-1}$, $Q_\ell\in\mathbb{C}[z_0,z_1,\ldots,z_{n-1}]$ of degree $d-1$, resp. $d$, resp. $\ell-1$, resp. $\ell$ and such that $(P_{d-1},Q_{\ell-1})\not=(0,0)$. The map $\widetilde{\psi}_{P,Q,R}$ is birational if and only if $\widetilde{\psi}_R$ is. This statement allows to prove that given a hypersurface of $\mathbb{P}^n_\mathbb{C}$ one can construct a birational self map of $\mathbb{P}^n_\mathbb{C}$ that blows down this hypersurface: \[lem:prescribed\] Let $n\geq 3$. Let $S$ be an hypersurface of $\mathbb{P}^n_\mathbb{C}$ of degree $\ell\geq 1$ having a point $p$ of multiplicity $\geq \ell-1$. Then there exists a birational self map of $\mathbb{P}^n_\mathbb{C}$ of degree $d\geq \ell+1$ that blows down $S$ onto a point. Let us assume without loss of generality that $p=(0:0:\ldots:0:1)$. Suppose that $S$ is given by $(Q=0)$. Take a generic plane passing through $p$ given by $(H=0)$. Choose $P=z_nP_{d-1}+P_d$ such that - $P_{d-1}\in\mathbb{C}[z_0,z_1,\ldots,z_{n-1}]$ of degree $d-1$ and $\not=0$; - $P_d\in\mathbb{C}[z_0,z_1,\ldots,z_{n-1}]$ of degree $d$; - $\mathrm{pgcd}(P,HQ)=1$. Set $\widetilde{Q}=H^{d-\ell-1}q$ and $R_i=z_i$. The statement then follows from Lemma \[lem:construction\]. Consider the family of hypersurfaces given by $Q(z_1,z_2,z_3)=0$ where $(Q=0)$ defines a smooth curve $\mathcal{C}_Q$ of degree $\ell$ on $\big\{z_0=z_4=z_5=\ldots=z_n=0\big\}$. Note that $(Q=0)$ is birationally equivalent to $\mathbb{P}^{n-2}_\mathbb{C}\times\mathcal{C}_Q$. Furthermore $(Q=0)$ and $(Q'=0)$ are birationally equivalent if and only if $\mathcal{C}_Q$ and $\mathcal{C}_{Q'}$ are isomorphic. Take $\ell=2$; the set of isomorphism classes of smooth cubics is a $1$-parameter family. For any $\mathcal{C}_Q$ there exists a birational self map of $\mathbb{P}^n_\mathbb{C}$ that blows down $\mathcal{C}_Q$ onto a point (Lemma \[lem:prescribed\]). As a result any set of group generators of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$, $n\geq 3$, has to contain uncountably many elements of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})\smallsetminus\mathrm{PGL}(n+1,\mathbb{C})$. As we have seen one consequence of <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> theorem is that the <span style="font-variant:small-caps;">Jonquières</span> group and $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})=\mathrm{PGL}(3,\mathbb{C})$ generate $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. This statement does also not hold in higher dimension ([@BlancLamyZimmermann]): let $n\geq 3$, the $n$ dimensional <span style="font-variant:small-caps;">Cremona</span> group is not generated by $\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$ and by <span style="font-variant:small-caps;">Jonquières</span> elements, [i.e.]{} elements that preserve a family of lines through a given point, which form a subgroup $$\mathrm{PGL}(2,\mathbb{C}(z_2,z_3,\ldots,z_n))\rtimes\mathrm{Bir}(\mathbb{P}^{n-1}_\mathbb{C})\subseteq \mathrm{Bir}(\mathbb{P}^n_\mathbb{C}).$$ A more precise statement has been established in dimension $3$ in [@BlancYasinsky]: the $3$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is not generated by birational maps preserving a linear fibration $\mathbb{P}^3_\mathbb{C}\dashrightarrow\mathbb{P}^2_\mathbb{C}$. Algebraic properties of the <span style="font-variant:small-caps;">Cremona</span> group ======================================================================================= The group $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ has many properties of linear groups so we wonder if $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ has a faithful linear representation; in the first section we show that the answer is no ([@CerveauDeserti:ptdegre; @Cornulier]). Still in the first section we give the proof of the following property: the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group contains non-linear finitely generated subgroups ([@Cornulier]). In the second section we give the proof of the facts that - the normal subgroup generated by $\sigma_2$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. - the normal subgroup, generated by a non-trivial element of $\mathrm{PGL}(3,\mathbb{C})=\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. As a consequence $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is perfect ([@CerveauDeserti:ptdegre]), that is $[\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}),\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})]=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. We finish this chapter by the description of the endomorphisms of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group; as a corollary we get the The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is hopfian, [i.e.]{} any surjective endomorphism of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is an automorphism. We use for that the classification of the representations of $\mathrm{SL}(3,\mathbb{Z})$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, we thus recall and establish it in the third section: \[thm:IMRN\] Let $\Gamma$ be a finite index subgroup of $\mathrm{SL}(3,\mathbb{Z})$. Let $\upsilon$ be an injective morphism from $\Gamma$ to $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Then, up to birational conjugacy, either $\upsilon$ is the canonical embedding, or $\upsilon$ is the involution $A\mapsto ({}^{t}\!\,A)^{-1}$. As a result we obtain the: \[cor:IMRN\] If a morphism from a subgroup of finite index of $\mathrm{SL}(n,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ has infinite image, then $n\leq 3$. The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is not linear {#section:notlinear} --------------------------------------------------------------------------------------------- <span style="font-variant:small-caps;">Cantat</span> and <span style="font-variant:small-caps;">Lamy</span> proved that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is not simple but the non-existence of a faithful representation does not imply the non-existence of a non-trivial representation. So let us deal with the following statement: \[thm:CerveauDesertinonlinear\] The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group has no faithful linear representation in characteristic zero. Before giving the proof let us mention that making an easy refinement of it provides the following stronger result: If $\Bbbk$ is an algebraically closed field, then there is no non-trivial finite dimensional linear representation for $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ over any field. Let us recall the following statement due to <span style="font-variant:small-caps;">Birkhoff</span>: Let $\Bbbk$ be a field of characteristic zero. Let $A$, $B$, and $C$ be three elements of $\mathrm{GL}(n,\Bbbk)$ such that - $[A,B]=C$, $[A,C]=[B,C]=\mathrm{id}$, - $C$ has prime order $p$. Then $p\leq n$. Assume that $\Bbbk$ is algebraically closed. Since $C$ is of order $p$ its eigenvalues are $p$-rooth of unity. If the eigenvalues of $C$ are all equal to $1$, then $C$ is unipotent and $p\leq n$. Otherwise $C$ admits an eigenvalue $\alpha\not=1$. Consider the eigenspace $E_\alpha=\big\{v\,\vert\, Cv=\alpha v\big\}$ of $C$ associated to the eigenvalue $\alpha$. By assumption $A$ and $B$ commute to $C$ so $E_\alpha$ is invariant by $A$ and $B$. From $[A,B]=C$ we get $[A_{\vert E_\alpha},B_{\vert E_\alpha}]=C_{\vert E_\alpha}$; but $C_{\vert E_\alpha}=\alpha\mathrm{id}_{\vert E_\alpha}$ hence $[A_{\vert E_\alpha},B_{\vert E_\alpha}]=\alpha\mathrm{id}_{\vert E_\alpha}$, that is $(B^{-1}AB)_{\vert E_\alpha}=\alpha A_{\vert E_\alpha}$. Note that $(B^{-1}AB)_{\vert E_\alpha}$ and $A_{\vert E_\alpha}$ are conjugate thus $(B^{-1}AB)_{\vert E_\alpha}$ and $A_{\vert E_\alpha}$ have the same eigenvalues. Furthermore these eigenvalues are non-zero. If $\lambda$ is an eigenvalue of $A_{\vert E_\alpha}$, then $\alpha\lambda$, $\alpha^2\lambda$, $\ldots$, $\alpha^{p-1}\lambda$ are also eigenvalues of $A_{\vert E_\alpha}$. As $p$ is prime and $\alpha$ distinct from $1$, the numbers $\alpha$, $\alpha^2$, $\ldots$, $\alpha^{p-1}$ are distinct, $\dim E_\alpha\geq p$, and $n\geq p$. Assume by contradiction that there exists an injective morphism $\zeta$ from $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ into $\mathrm{GL}(n,\Bbbk)$. For any prime $p$ let us consider in the affine chart $z_2=1$ the group generated by the maps $$\begin{aligned} & (z_0,z_1)\mapsto(\mathrm{e}^{-2\mathbf{i}\pi/p}z_0,z_1),&&(z_0,z_1)\dashrightarrow(z_0,z_0z_1), &&(z_0,z_1)\mapsto(z_0,\mathrm{e}^{-2\mathbf{i}\pi/p}z_1). \end{aligned}$$ The images of these three elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ satisfy the assumptions of <span style="font-variant:small-caps;">Birkhoff</span> Lemma; therefore $p\leq n$ for any prime $p$: contradiction. In [@Cornulier] <span style="font-variant:small-caps;">Cornulier</span> gives an example of a non-linear finitely generated subgroup of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. The existence of such subgroup is not new, for instance it follows from an unpublished construction of <span style="font-variant:small-caps;">Cantat</span>. The example in [@Cornulier] has the additional feature of being $3$-solvable. To prove its non-linearity <span style="font-variant:small-caps;">Cornulier</span> proves that it contains nilpotent subgroups of arbitrary large nilpotency length. Let $\mathrm{G}$ be a group. Recall that $[g,h]=g\circ h\circ g^{-1}\circ h^{-1}$ denotes the commutator of $g$ and $h$. If $\mathrm{H}_1$ and $\mathrm{H}_2$ are two subgroups of $\mathrm{G}$, then $[\mathrm{H}_1,\mathrm{H}_2]$ is the subgroup of $\mathrm{G}$ generated by the elements of the form $[g,h]$ with $g\in\mathrm{H}_1$ and $h\in\mathrm{H}_2$. We defined the derived series of $\mathrm{G}$ by setting $\mathrm{G}^{(0)}=\mathrm{G}$ and for all $n\geq 0$ $$\mathrm{G}^{(n+1)}=[\mathrm{G}^{(n)},\mathrm{G}^{(n)}].$$ The soluble length $\ell(\mathrm{G})$ of $\mathrm{G}$ is defined by $$\ell(\mathrm{G})=\min\big\{k\in\mathbb{N}\cup\{0\}\,\vert\,\mathrm{G}^{(k)}=\{\mathrm{id}\}\big\}$$ with the convention: $\min\emptyset=\infty$. We say that $\mathrm{G}$ is solvable if $\ell(\mathrm{G})<\infty$. The descending central series of a group $\mathrm{G}$ is defined by $C^0\mathrm{G}=\mathrm{G}$ and for all $n\geq 0$ $$C^{n+1}\mathrm{G}=[\mathrm{G},C^n\mathrm{G}].$$ The group $\mathrm{G}$ is nilpotent if there exists $j\geq 0$ such that $C^j\mathrm{G}=\{\mathrm{id}\}$. If $j$ is the minimum non-negative number with such a property, we say that $\mathrm{G}$ is of nilpotent class $j$. Take $f$ in $\mathbb{C}(z_0)$ and $g$ in $\mathbb{C}(z_0)^$; define $\alpha_f$ and $\mu_g$ by $$\begin{aligned} & \alpha_f\colon(z_0,z_1)\dashrightarrow\big(z_0,z_1+f(z_0)\big), && \mu_g\colon(z_0,z_1)\dashrightarrow\big(z_0,z_1g(z_0)\big).\end{aligned}$$ Note that $$\label{eq:uneetoile} \alpha_{f+f'}=\alpha_f\circ\alpha_{f'} \qquad \mu_{gg'}=\mu_g\circ\mu_{g'}\qquad \mu_g\circ\alpha_f\circ\mu_g^{-1}=\alpha_{fg}$$ Take $t\in\mathbb{C}$ and consider $s_t\colon(z_0,z_1)\mapsto(z_0+t,z_1)$. The following equalities hold $$\label{eq:deuxetoiles} s_t\circ\alpha_{f(z_0)}\circ s_t^{-1}=\alpha_{f(z_0-t)}, \qquad s_t\circ\mu_{g(z_0)}\circ s_t^{-1}=\mu_{g(z_0-t)}$$ Let $\Gamma_n$ be the subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ defined for any $n\geq 0$ by $$\Gamma_n=\langle s_1,\,\alpha_{z_0^n}\rangle.$$ Remark that $\Gamma_n$ is indeed a subgroup of the <span style="font-variant:small-caps;">Jonquières</span> group. It satisfies the following properties: The nilpotency length of $\Gamma_n$ is exactly $n+1$, and $\Gamma_n$ is torsion free. Let $\mathrm{A}_n$ be the abelian subgroup of the <span style="font-variant:small-caps;">Jonquières</span> group consisting of all $\alpha_P$ where $P$ ranges over polynomials of degree at most $n$. The group $\mathrm{A}_n$ is normalized by $s_1$, and $[s_1,\mathrm{A}_n]\subset \mathrm{A}_{n-1}$ for $n\geq 1$ while $\mathrm{A}_0=\{\mathrm{id}\}$. Therefore the largest group generated by $s_1$ and $\mathrm{A}_n$ is nilpotent of class at most $n+1$, and so is $\Gamma_n$. Consider now the $n$-iterated group commutator given by $$[s_1,[s_1,\ldots,[s_1,\alpha_{z_0^n}]\ldots]$$ It coincides with $\alpha_{\Delta^nz_0^n}$ where $\Delta$ is the discrete differential operator $\Delta P(z_0)=-P(z_0)+P(z_0-1)$. Remark that $\Delta^nz_0^n\not=0$ and $\Gamma_n$ is not $n$-nilpotent. Clearly $\Gamma_n$ is torsion-free. The group $$\mathrm{G}=\langle s_1,\,\alpha_1,\mu_{z_0}\rangle\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{Q})$$ satisfies the following properties: The group $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{Q})$ is solvable of length $3$, and is not linear over any field. A consequence of this statement is Proposition \[thm:CerveauDesertinonlinear\]. Relations (\[eq:uneetoile\]) and (\[eq:deuxetoiles\]) imply that $\langle s_1,\,\alpha_f,\,\mu_g\,\vert\, f\in\mathbb{C}(z_0),\,g\in\mathbb{C}(z_0)^\rangle$ is solvable of length at most three. The subgroup $$\langle s_1,\,\alpha_f,\,\mu_g\,\vert\,f\in\mathbb{C}(z_0),\,g=\displaystyle\prod_{n\in\mathbb{Z}}(z_0-n)^{k_n},\,k_n\text{ finitely supported }\rangle$$ contains $\Gamma_n$, and is torsion free. As $\mu_{z_0}^n\circ\alpha_1\circ\mu_{z_0}^{-n}=\alpha_{z_0^n}$, the group $\Gamma_n$ is contained in $\mathrm{G}$ for all $n$. But $\Gamma_n$ is nilpotent of length exactly $n+1$, hence $\mathrm{G}$ has no linear representation over any field. The <span style="font-variant:small-caps;">Cremona</span> group is perfect {#section:perfect} -------------------------------------------------------------------------- In this section let us prove the following statement \[thm:parfait\] The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is perfect, [i.e.]{} the commutator subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$: $$\big[\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}),\,\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\big]=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ Let $\mathrm{G}$ be a group, and let $g$ be an element of $\mathrm{G}$. We denote by $\ll g\gg_{\mathrm{G}}$ the normal subgroup of $\mathrm{G}$ generated by $g$: $$\ll g\gg_{\mathrm{G}}=\langle h\circ g\circ h^{-1},\,h\circ g^{-1}\circ h^{-1}\,\vert\,h\in \mathrm{G}\rangle.$$ Since $\mathrm{PGL}(3,\mathbb{C})$ is simple then $$\label{eq:simple} \ll A\gg_{\mathrm{PGL}(3,\mathbb{C})}=\mathrm{PGL}(3,\mathbb{C})$$ for any non-trivial element $A$ of $\mathrm{PGL}(3,\mathbb{C})$. Consider now a birational self map $\phi$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. The <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> Theorem implies that $$\label{eq:noether} \phi=(A_1)\circ\sigma_2\circ A_2\circ\sigma_2\circ A_3\circ\ldots\circ A_n\circ(\sigma_2)$$ with $A_i\in\mathrm{PGL}(3,\mathbb{C})$. The relation (\[eq:simple\]) implies that $$\ll(z_0,z_1)\mapsto(-z_0,-z_1)\gg_{\mathrm{PGL}(3,\mathbb{C})}=\mathrm{PGL}(3,\mathbb{C})$$ and thus any $A_i$ in (\[eq:noether\]) can be written $$B_1\circ\big((z_0,z_1)\mapsto(-z_0,-z_1)\big)\circ B_1^{-1}\circ B_2\circ\big((z_0,z_1)\mapsto(-z_0,-z_1)\big)\circ B_2^{-1}\circ\ldots\circ B_n\circ\big((z_0,z_1)\mapsto(-z_0,-z_1)\big)\circ B_n^{-1}$$ with $B_i\in\mathrm{PGL}(3,\mathbb{C})$. The involutions $(z_0,z_1)\mapsto(-z_0,-z_1)$ and $\sigma_2$ being conjugate via $(z_0,z_1)\mapsto\left(\frac{z_0+1}{z_0-1},\frac{z_1+1}{z_1-1}\right)\in\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$ any element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ can be written as a composition of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$-conjugates of $\sigma_2$. As a consequence one has \[pro:parfait1\] The normal subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ generated by $\sigma_2$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$: $$\ll \sigma_2\gg_{\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})}=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ Consider now a non-trivial automorphism $A$ of $\mathbb{P}^2_\mathbb{C}$. As $\ll A\gg_{\mathrm{PGL}(3,\mathbb{C})}=\mathrm{PGL}(3,\mathbb{C})$ (see (\[eq:simple\])) the involution $(z_0,z_1)\mapsto(-z_0,-z_1)$ can be written as a composition of $\mathrm{PGL}(3,\mathbb{C})$-conjugates of $A$. Since $(z_0,z_1)\mapsto(-z_0,-z_1)$ and $\sigma_2$ are conjugate via $(z_0,z_1)\mapsto\left(\frac{z_0+1}{z_0-1},\frac{z_1+1}{z_1-1}\right)\in\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$ one gets $$\sigma_2=\varphi_1\circ A\circ\varphi_1^{-1}\circ\varphi_2\circ A\circ\varphi_2^{-1}\circ\ldots\circ\varphi_n\circ A\circ\varphi_n^{-1}$$ with $\varphi_i\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. As a consequence the inclusion $\ll \sigma_2\gg_{\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})}\subset \ll A\gg_{\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})}$ holds. But $\ll \sigma_2\gg_{\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})}=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ (Proposition \[pro:parfait1\]) hence \[pro:parfait2\] Let $A$ be a non-trivial automorphism of the complex projective plane. Then $$\ll A\gg_{\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})}=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ According to (\[eq:noether\]) and Proposition \[pro:parfait2\] one has \[cor:parfait\] Any birational self map of $\mathbb{P}^2_\mathbb{C}$ can be written as the composition of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$-conjugates of the translation $(z_0,z_1)\mapsto(z_0,z_1+1)$. But the translation $(z_0,z_1)\mapsto(z_0,z_1+1)$ is a commutator $$\Big((z_0,z_1)\mapsto(z_0,z_1+1)\Big)=\left[(z_0,z_1)\mapsto(z_0,3z_1),\,(z_0,z_1)\mapsto\left(z_0,\frac{z_1+1}{2}\right)\right]$$ and Corollary \[cor:parfait\] thus implies Theorem \[thm:parfait\]. Representations of $\mathrm{SL}(n,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ for $n\geq 3$ --------------------------------------------------------------------------------------------------------- We will now give a sketch of the proofs of Theorem \[thm:IMRN\] and Corollary \[cor:IMRN\]. Let us introduce some notations. Given $A\in\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})=\mathrm{PGL}(3,\mathbb{C})$ we denote by ${}^{t}\!\,A$ the linear transpose of $A$. The involution $$A\mapsto A^\vee=({}^{t}\!\,A)^{-1}$$ determines an exterior and algebraic automorphism of the group $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ (see [@Dieudonne]). Let us recall some properties about the groups $\mathrm{SL}(n,\mathbb{Z})$ (see for instance [@Steinberg]). For any integer $q$ let us introduce the morphism $$\Theta_q\colon\mathrm{SL}(n,\mathbb{Z})\to\mathrm{SL}\left(n,\faktor{\mathbb{Z}}{q\mathbb{Z}}\right)$$ induced by the reduction modulo $q$ morphism $\mathbb{Z}\to\faktor{\mathbb{Z}}{q\mathbb{Z}}$. Denote by $\Gamma(n,q)$ the kernel of $\Theta_q$ and by $\widetilde{\Gamma}(n,q)$ the reciprocical image of the subgroup of diagonal matrices of $\mathrm{SL}\left(n,\faktor{\mathbb{Z}}{q\mathbb{Z}}\right)$ by $\Theta_q$. The $\Gamma(n,q)$ are normal subgroups called congruence subgroups. \[thm:structsl\] Let $n\geq 3$ be an integer. Let $\Gamma$ be a subgroup of $\mathrm{SL}(n,\mathbb{Z})$. If $\Gamma$ has finite index, there exists an integer $q$ such that the following inclusions hold $$\Gamma(n,q)\subset\Gamma\subset\widetilde{\Gamma}(n,q).$$ If $\Gamma$ has infinite index, then $\Gamma$ is finite. Take $1\leq i,\,j\leq n$, $i\not=j$. Let us denote by $\delta_{ij}$ the $n\times n$ <span style="font-variant:small-caps;">Kronecker</span> matrix and set $\mathrm{e}_{ij}=\mathrm{id}+\delta_{i,j}$. The group $\mathrm{SL}(3,\mathbb{Z})$ has the following presentation $$\langle \mathrm{e}_{ij}\,\vert\, [\mathrm{e}_{ij},\mathrm{e}_{k\ell}]=\left\{\begin{array}{lll} \mathrm{id}\text{ if $i\not=\ell$ and $j\not=k$}\\ \mathrm{e}_{i\ell}\text{ if $i\not=\ell$ and $j=k$}\\ \mathrm{e}_{kj}^{-1}\text{ if $i=\ell$ and $j\not=k$} \end{array}\right.,\,(\mathrm{e}_{12}\mathrm{e}_{21}^{-1}\mathrm{e}_{12})^4=\mathrm{id}\rangle.$$ The $\mathrm{e}_{ij}^q$’s generate $\Gamma(3,q)$ and satisfy relations similar to those verified by the $\mathrm{e}_{ij}$’s except $(\mathrm{e}_{12}\mathrm{e}_{21}^{-1}\mathrm{e}_{12})^4=\mathrm{id}$. The $\mathrm{e}_{ij}^q$’s are called the standard generators of $\Gamma(3,q)$. Let $k$ be an integer. A $k$-<span style="font-variant:small-caps;">Heisenberg</span> group is a group with the following presentation $$\mathcal{H}_k=\langle f,\,g,\,h\,\vert\,[f,g]=h^k,\,[f,h]=[g,h]=\mathrm{id}\rangle.$$ We will say that $f$, $g$ and $h$ are the standard generators of $\mathcal{H}_k$ . \[rem:heis\] - The subgroup of $\mathcal{H}_k$ generated by $f$, $g$ and $h^k$ is a subgroup of index $k$. - The groups $\Gamma(3,q)$ contain a lot of $k$-<span style="font-variant:small-caps;">Heisenberg</span> groups; for instance if $1\leq i\not=j\not=\ell\leq 3$, then $\langle \mathrm{e}_{ij}^q,\,\mathrm{e}_{i\ell}^q,\,\mathrm{e}_{j\ell}^q\rangle$ is a $q$-<span style="font-variant:small-caps;">Heisenberg</span> group of $\Gamma(3,q)$. Let $\mathrm{G}$ be a finitely generated group, let $\big\{a_1,\,a_2,\,\ldots,\,a_n\big\}$ be a generating set of $\mathrm{G}$, and let $g$ be an element of $\mathrm{G}$. The length $\vert\vert g\vert\vert$ of $g$ is the smallest integer $k$ for which there exists a sequence $(s_1,\,s_2,\ldots,\,s_k)$ with $s_i\in\big\{a_1,\,a_2,\,\ldots,\,a_n,\,a_1^{-1},\,a_2^{-1},\,\ldots,\,a_n^{-1}\big\}$ for any $1\leq i\leq k$, such that $$g=s_1s_2\ldots s_k.$$ We say that $$\displaystyle\lim_{k\to +\infty}\frac{\vert\vert g^k\vert\vert}{k}$$ is the stable length of $g$. A distorted element of $\mathrm{G}$ is an element of infinite order of $\mathrm{G}$ whose stable length is zero. \[lem:disto\] Let $\mathcal{H}_k=\langle f,\,g,\,h\,\vert\,[f,g]=h^k,\,[f,h]=[g,h]=\mathrm{id}\rangle$ be a $k$-<span style="font-variant:small-caps;">Heisenberg</span> group. The element $h^k$ is distorted. In particular the standard generators of $\Gamma(3,q)$ are distorted. Since $[f,g]=[g,h]=\mathrm{id}$ on has $[f^\ell,g^m]=h^{\ell m}$ for any integer $\ell$, $m$. In particular $h^{k\ell^2}=[f^\ell,g^\ell]$. As a result $\vert\vert h^{k\ell^2}\vert\vert\leq 4\ell$. Each standard generator of $\Gamma(3,q)$ satisfies $\mathrm{e}_{ij}^{q^2}=[\mathrm{e}_{i\ell}^q,\mathrm{e}_{\ell j}^q]$. \[lem:nilkeylemma\] Let $\mathrm{G}$ be a finitely generated group. Let $\upsilon$ be a morphism from $\mathrm{G}$ to $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Any distorted element $g$ of $\mathrm{G}$ satisfies $\lambda(\upsilon(g))=1$, [i.e.]{} $\upsilon(g)$ is an elliptic map or a parabolic one. Let $\big\{a_1,\,a_2,\,\ldots,\,a_n\big\}$ be a generating set of $\mathrm{G}$. The inequalities $$\lambda(\upsilon(g))^k\leq \deg(\upsilon(g)^k)\leq\displaystyle\max_i\big(\deg(\upsilon(a_i))\big)^{\vert\vert g^k\vert\vert}$$ imply the following ones $$0\leq\log\big(\lambda(\upsilon(g))\big)\leq\frac{\vert\vert g^k\vert\vert}{k}\log\big(\max_i\big(\deg(\upsilon(a_i))\big)\big).$$ But since $g$ is distorted $\displaystyle\lim_{k\to +\infty}\frac{\vert\vert g^k\vert\vert}{k}=0$ and $\log\big(\lambda(\upsilon(g))\big)=0$. We follow the proof of [@Deserti:IMRN]; nevertheless it is possible to “simplify it” by using the following result: any distorted element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is algebraic ([@BlancFurter:length; @CantatCornulier]). According to Corollary \[cor:3plus\] we thus have: any distorted element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is elliptic. Let $\phi_1$, $\phi_2$, $\ldots$, $\phi_k$ be some birational self maps of a rational surface $S$. Assume that $\phi_1$, $\phi_2$, $\ldots$, $\phi_k$ are virtually isotopic to the identity. We say that $\phi_1$, $\phi_2$, $\ldots$, $\phi_k$ are simultaneously virtually isotopic to the identity if there exists a surface $\widetilde{S}$, a birational map $\psi\colon\widetilde{S}\dashrightarrow S$ such that for any $1\leq i\leq k$ the map $\psi^{-1}\circ\phi_i\circ\psi$ belongs to $\mathrm{Aut}(\widetilde{S})$ and $\psi^{-1}\circ\phi_i^\ell\circ\psi$ belongs to $\mathrm{Aut}(\widetilde{S})^0$ for some integer $\ell$. \[pro:sim\] Let $\upsilon$ be a representation from $$\mathcal{H}_k=\langle f,\,g,\,h\,\vert\,[f,g]=h^k,\,[f,h]=[g,h]=\mathrm{id}\rangle$$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Assume that any standard generator $\upsilon(f)$, $\upsilon(g)$ and $\upsilon(h)$ of $\upsilon(\mathcal{H}_k)$ is virtually isotopic to the identity. Then $\upsilon(f)$, $\upsilon(g)$ and $\upsilon(h)$ are simultaneously virtually isotopic to the identity. According to Proposition \[pro:commas\] the maps $\upsilon(f)$ and $\upsilon(g)$ are simultaneously virtually isotopic to the identity. Since $g$ and $h$ commute, $\mathrm{Exc}(\upsilon(g))$ and $\mathrm{Ind}(\upsilon(g))$ are invariant by $\upsilon(h)$. The relation $[f,g]=h^k$ implies that both $\mathrm{Exc}(\upsilon(g))$ and $\mathrm{Ind}(\upsilon(g))$ are invariant by $\upsilon(f)$. A reasoning analogous to that of the proof of Proposition \[pro:commas\] and [@DillerFavre Lemma 4.1] allows us to establish the statement. The second assertion of Remarks \[rem:heis\] leads us to study the representations of <span style="font-variant:small-caps;">Heisenberg</span> $k$-groups into automorphisms groups of minimal rational surfaces. Let us deal with it. \[lem:notp1p1\] Let $\upsilon$ be a morphism from $\mathcal{H}_k$ into $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$. The morphism $\upsilon$ is not an embedding. We can assume that $\upsilon(f)$, $\upsilon(g)$ and $\upsilon(h)$ fixe the two standard fibrations (if it is not the case we can consider $\mathcal{H}_{2k}$ instead of $\mathcal{H}_k$); in other words we can assume that $\mathrm{im}\,\upsilon$ is contained in $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$. Denote by $\mathrm{pr}_i$, $i\in\{1,\,2\}$, the $i$-th projection. Note that $\mathrm{pr}_i(\upsilon(\mathcal{H}_{2k}))$ is a solvable subgroup of $\mathrm{PGL}(2,\mathbb{C})$. Furthermore $\mathrm{pr}_i(\upsilon(h^k))$ is a commutator. Hence $\mathrm{pr}_i(\upsilon(h^k))$ is conjugate to the translation $z\mapsto z+\beta_i$. Let us prove by contradiction that $\beta_i=0$; assume $\beta_i\not=0$. Then both $\mathrm{pr}_i(\upsilon(f))$ and $\mathrm{pr}_i(\upsilon(g))$ are also some translation since they commute with $\mathrm{pr}_i(\upsilon(h^k))$. But then $\mathrm{pr}_i(\upsilon(h^k))=[\mathrm{pr}_i(\upsilon(f)),\mathrm{pr}_i(\upsilon(g))]=\mathrm{id}$: contradiction with $\beta_i\not=0$. As a result $\beta_i=0$ and $\upsilon$ is not an embedding. \[lem:heishirz\] Let $\upsilon$ be a morphism from $\mathcal{H}_k$ into $\mathrm{Aut}(\mathbb{F}_n)$ with $n\geq 1$. Then up to birational conjugacy $\upsilon(\mathcal{H}_k)$ is a subgroup of $$\big\{(z_0,z_1)\mapsto(\alpha z_0+P(z_1),\beta z_1+\gamma)\,\vert\,\alpha,\,\beta\in\mathbb{C}^,\,\gamma\in\mathbb{C},\,P\in\mathbb{C}[z_1]\big\}.$$ Moreover up to birational conjugacy $$\upsilon(h^{2k})\colon(z_0,z_1)\mapsto(z_0+P(z_1),z_1)$$ for some $P\in\mathbb{C}[z_1]$. \[lem:heispgl3\] Let $\upsilon$ be an embedding of $\mathcal{H}_k$ into $\mathrm{PGL}(3,\mathbb{C})$. Up to linear conjugacy $$\begin{aligned} & \upsilon(f)\colon(z_0,z_1)\mapsto(z_0+\zeta z_1,z_1+\beta)&& \upsilon(g)\colon(z_0,z_1)\mapsto(z_0+\gamma z_1,z_1+\delta) \\ & \upsilon(h^k)\colon(z_0,z_1)\mapsto(z_0+k,z_1)&&\end{aligned}$$ where $\zeta$, $\delta$, $\beta$ $\gamma$ denote complex numbers such that $\zeta\delta-\beta\gamma=k$. The <span style="font-variant:small-caps;">Zariski</span> closure $\overline{\upsilon(\mathcal{H}_k)}$ of $\upsilon(\mathcal{H}_k)$ is an algebraic unipotent subgroup of $\mathrm{PGL}(3,\mathbb{C})$. By assumption $\upsilon$ is an embedding so the <span style="font-variant:small-caps;">Lie</span> algebra of $\overline{\upsilon(\mathcal{H}_k)}$ is isomorphic to $$\mathfrak{h}=\left\{\left(\begin{array}{ccc} 0 & \zeta & \beta\\ 0 & 0 & \gamma\\ 0 & 0 & 0 \end{array} \right)\,\vert\,\zeta,\,\beta,\,\gamma\in\mathbb{C}\right\}.$$ Let $\mathrm{pr}$ be the canonical projection from $\mathrm{SL}(3,\mathbb{C})$ into $\mathrm{PGL}(3,\mathbb{C})$. The <span style="font-variant:small-caps;">Lie</span> algebra of $\mathrm{pr}^{-1}(\upsilon(\mathcal{H}_k))$ coincides with $\mathfrak{h}$ up to conjugacy. Let us recall that the exponential map sends $\mathfrak{h}$ in the group $\mathrm{H}$ of upper triangular matrices and that $\mathrm{H}$ is a connected algebraic group. As a consequence $\big(\mathrm{pr}^{-1}(\overline{\upsilon(\mathcal{H}_k)})\big)^0=\mathrm{H}$. Any element of $\mathrm{pr}^{-1}(\overline{\upsilon(\mathcal{H}_k)})$ acts by conjugation on $\mathrm{H}$ so belongs to $\langle \mathrm{H},\,\mathbf{j}\cdot\mathrm{id}\,\vert\,\mathbf{j}^3=1\rangle$. As $\mathrm{pr}(\mathbf{j}\cdot\mathrm{id})=\mathrm{id}$, the restriction $\mathrm{pr}_{\vert\mathrm{H}}$ of $\mathrm{pr}$ to $\mathrm{H}$ is surjective on $\overline{\upsilon(\mathcal{H}_k)}$. It is also injective. Hence it is an isomorphism. Therefore $\upsilon$ can be lifted to a representation $\widetilde{\upsilon}$ from $\mathcal{H}_k$ into $\mathrm{H}$. The map $\widetilde{\upsilon}(h^k)$ can be written as a commutator; it is thus unipotent. The relations satisfied by the generators imply that up to conjugacy in $\mathrm{SL}(3,\mathbb{C})$ $$\begin{aligned} & \upsilon(f)\colon(z_0,z_1)\mapsto(z_0+\zeta z_1,z_1+\beta)&& \upsilon(g)\colon(z_0,z_1)\mapsto(z_0+\gamma z_1,z_1+\delta) \\ & \upsilon(h^k)\colon(z_0,z_1)\mapsto(z_0+k,z_1)&&\end{aligned}$$ with $\zeta\delta-\beta\gamma=k$. Let $\rho$ be an embedding of $\Gamma(3,q)$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. According to Lemma \[lem:disto\] and Lemma \[lem:nilkeylemma\] for any standard generator $\mathrm{e}_{ij}$ of $\mathrm{SL}(3,\mathbb{Z})$ one has $\lambda(\rho(e_{ij}))=1$. Theorem \[thm:dilfav\] implies that - either one of the $\rho(\mathrm{e}_{ij}^q)$ preserves a unique fibration that is rational or elliptic, - or any standard generator of $\Gamma(3,q)$ is virtually isotopic to the identity. Let us first assume that (i) holds. \[lem:fleur\] Let $\Gamma$ be a <span style="font-variant:small-caps;">Kazhdan</span> group that is finitely generated. Let $\rho$ be a morphism from $\Gamma$ into $\mathrm{PGL}(2,\mathbb{C}(z_1))$ $($resp. $\mathrm{PGL}(2,\mathbb{C}))$. Then $\rho$ has finite image. Denote by $\gamma_i$ the generators of $\Gamma$ and by $\left(\begin{array}{cc} a_i(z_1) & b_i(z_1)\\ c_i(z_1) & d_i(z_1) \end{array}\right)$ their image by $\rho$. A finitely generated $\mathbb{Q}$-group is isomorphic to a subfield of $\mathbb{C}$. Hence $\mathbb{Q}(a_i(z_0),b_i(z_0),c_i(z_0),d_i(z_0))$ is isomorphic to a subfield of $\mathbb{C}$ and one can assume that $\mathrm{im}\,\rho\subset\mathrm{PGL}(2,\mathbb{C})=\mathrm{Isom}(\mathbb{H}_3)$. As $\Gamma$ is <span style="font-variant:small-caps;">Kazhdan</span> any continuous action of $\Gamma$ by isometries of a real or complex hyperbolic space has a fixed point. The image of $\rho$ is thus up to conjugacy a subgroup of $\mathrm{SO}(3,\mathbb{R})$; according to [@delaHarpeValette] the image of $\rho$ is thus finite. \[pro:2fleurs\] Let $\rho$ be a morphism from $\Gamma(3,q)$ to $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. If one $\rho(\mathrm{e}_{ij}^q)$ preserves a unique fibration, then $\mathrm{im}\,\rho$ is finite. Let us assume without loss of generality that $\rho(\mathrm{e}_{12}^q)$ preserves a unique fibration $\mathcal{F}$. The relations satisfied by the $\mathrm{e}_{ij}^q$ imply that $\mathcal{F}$ is invariant by any $\rho(\mathrm{e}_{ij}^{q^2})$. Hence for any $\rho(\mathrm{e}_{ij}^{q^2})$ there exist - $F\colon\mathbb{P}^2_\mathbb{C}\to\mathrm{Aut}(\mathbb{P}^1_\mathbb{C})$ that defines $\mathcal{F}$, - and $h_{ij}\in\mathrm{PGL}(2,\mathbb{C})$ such that $F\circ\rho(\mathrm{e}_{ij}^{q^2})=h_{ij}\circ F$. Let $\upsilon$ be the morphism defined by $$\begin{aligned} &\upsilon\colon\Gamma(3,q^2)\to\mathrm{PGL}(2,\mathbb{C}), && \mathrm{e}_{ij}^{q^2}\mapsto h_{ij}.\end{aligned}$$ The group $\Gamma(3,q^2)$ is a <span style="font-variant:small-caps;">Kazhdan</span> group so $\Gamma=\ker\upsilon$ is of finite index (Lemma \[lem:fleur\]); as a consequence $\Gamma$ is a <span style="font-variant:small-caps;">Kazhdan</span> group. Remark that $\mathcal{F}$ can not be elliptic; indeed the group of birational maps that preserve fiberwise an elliptic fibration is metabelian and a subgroup of $\Gamma(3,q^2)$ of finite index can not be metabelian. Let us assume that $\mathcal{F}$ is a rational fibration. One can assume that $\mathcal{F}=(z_1=\text{ constant})$. The group of birational maps of the complex projective plane that preserves $\mathcal{F}$ is identified with $\mathrm{PGL}(2,\mathbb{C}(z_1))\rtimes\mathrm{PGL}(2,\mathbb{C})$ hence $\rho_{\vert\Gamma}\colon\Gamma\to\mathrm{PGL}(2,\mathbb{C}(z_1))$ has finite image (Lemma \[lem:fleur\]). Consider now the case (ii), [i.e.]{} assume that any standard generator of $\Gamma(3,q)$ is virtually isotopic to the identity. \[rem:hom\] Two irreducible homologous curves of negative self-intersection coincide. As a consequence an automorphism $\varphi$ of a surface $S$ isotopic to the identity fixes any curve of negative self-intersection. Furthermore for any sequence of blow-downs $\psi$ from $S$ to a minimal model $\widetilde{S}$ of $S$ the map $\psi\circ \varphi\circ\psi^{-1}$ is an automorphism of $\widetilde{S}$ isotopic to the identity. According to Remark \[rem:hom\], Proposition \[pro:sim\], Lemma \[lem:disto\] and Lemma \[lem:nilkeylemma\] the maps $\rho(\mathrm{e}_{12}^{qn})$, $\rho(\mathrm{e}_{13}^{qn})$, $\rho(\mathrm{e}_{23}^{qn})$ are, for some integer $n$, some automorphisms of a minimal rational surface, that is of $\mathbb{P}^2_\mathbb{C}$ or of $\mathbb{F}_n$, $n\geq 2$. Let us mention the case $\mathbb{F}_n$, $n\geq 2$ (see* [@Deserti:IMRN] for more details) and detail the case $\mathbb{P}^2_\mathbb{C}$. \[lem:hirzcong\] Let $\rho$ be a morphism from a congruence subgroup $\Gamma(3,q)$ of $\mathrm{SL}(3,\mathbb{Z})$ in the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. Assume that $\rho(\mathrm{e}_{12}^{q\ell})$, $\rho(\mathrm{e}_{13}^{q\ell})$ and $\rho(\mathrm{e}_{23}^{q\ell})$ belong to $\mathrm{Aut}(\mathbb{F}_n)$, $n\geq 2$, for some integer $\ell$. Then the image of $\rho$ is - either finite, - or contained in $\mathrm{PGL}(3,\mathbb{C})=\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ up to conjugacy. \[lem:pgl3cong\] Let $\rho$ be an embedding of a congruence subgroup $\Gamma(3,q)$ of $\mathrm{SL}(3,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. If $\rho(\mathrm{e}_{12}^{qn})$, $\rho(\mathrm{e}_{13}^{qn})$ and $\rho(\mathrm{e}_{23}^{qn})$ belong, for some integer $n$, to $\mathrm{PGL}(3,\mathbb{C})=\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, then $\rho(\Gamma(3,q^2n^2))$ is a subgroup of $\mathrm{PGL}(3,\mathbb{C})=\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$. To establish this statement we will need the two following results; the first one was obtained by <span style="font-variant:small-caps;">Cantat</span> and <span style="font-variant:small-caps;">Lamy</span> when they study the embeddings of lattices from simple <span style="font-variant:small-caps;">Lie</span> groups into the group of polynomial automorphisms $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ whereas the second one is a technical one. \[thm:CantatLamyaut\] Let $\mathrm{G}$ be a simple real <span style="font-variant:small-caps;">Lie</span> group. Let $\Gamma$ be a lattice of $\mathrm{G}$. If there exists an embedding of $\Gamma$ into $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$, then $\mathrm{G}$ is isomorphic to either $\mathrm{PSO}(1,n)$ or $\mathrm{PSU}(1,n)$ for some integer $n$. \[lem:tecn\] Let $\phi$ be an element of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. Assume that $\mathrm{Exc}(\phi)$ and $\mathrm{Exc}(\phi^2)$ are non-empty and contained in the line at infinity. If $\mathrm{Ind}(\phi)$ is also contained in the line at infinity, then $\phi$ is a polynomial automorphism of $\mathbb{A}^2_\mathbb{C}$. Lemma \[lem:heispgl3\] allows us to assume that $$\begin{aligned} & \rho(\mathrm{e}_{13}^{qn})\colon(z_0,z_1)\mapsto(z_0+qn,z_1), && \rho(\mathrm{e}_{12}^{qn})\colon(z_0,z_1)\mapsto (z_0+\zeta z_1,z_1+\beta),\\ & \rho(\mathrm{e}_{23}^{qn})\colon(z_0,z_1)\mapsto (z_0+\gamma z_1,z_1+\delta)&&\end{aligned}$$ where $\zeta\delta-\beta\gamma=q^2n^2$. - Let us first suppose that $\beta\delta\not=0$. Since $[\rho(\mathrm{e}_{13}^{qn}),\rho(\mathrm{e}_{21}^{qn})]=\rho(\mathrm{e}_{23}^{-q^2n^2})$ the curves blown down by $\rho(\mathrm{e}_{21}^{qn})$, if they exist, are of the type $z_1=$ constant. As $\rho(\mathrm{e}_{21}^{qn})$ and $\rho(\mathrm{e}_{23}^{qn})$ commute, the sets $\mathrm{Exc}(\rho(\mathrm{e}_{21}^{qn}))$ and $\mathrm{Ind}(\rho(\mathrm{e}_{21}^{qn}))$ are invariant by $\rho(\mathrm{e}_{23}^{qn})$. As a result $\mathrm{Exc}(\rho(\mathrm{e}_{21}^{qn}))$, $\mathrm{Ind}(\rho(\mathrm{e}_{21}^{qn}))$ and $\mathrm{Exc}((\rho(\mathrm{e}_{21}^{qn}))^2)$ are contained in the line at infinity. Hence $\rho(\mathrm{e}_{21}^{qn})$ belongs to either $\mathrm{PGL}(3,\mathbb{C})$ or $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ (Lemma \[lem:tecn\]). Note that if $\rho(\mathrm{e}_{21}^{qn})$ belongs to $\mathrm{PGL}(3,\mathbb{C})$, then $\rho(\mathrm{e}_{21}^{qn})$ preserves the line at infinity because $[\rho(\mathrm{e}_{21}^{qn}),\rho(\mathrm{e}_{23}^{qn})]=\mathrm{id}$. In other words $\rho(\mathrm{e}_{21}^{qn})$ also belongs to $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$. Using the relations $[\rho(\mathrm{e}_{13}^{qn}),\rho(\mathrm{e}_{32}^{qn})]=\rho(\mathrm{e}_{12}^{q^2n^2})$ and $[\rho(\mathrm{e}_{12}^{qn}),\rho(\mathrm{e}_{32}^{qn})]=\mathrm{id}$ we get that $\rho(\mathrm{e}_{23}^{qn})$ belongs to $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$. Finally any $\rho(\mathrm{e}_{ij}^{q^2n^2})$ is a polynomial automorphism of $\mathbb{A}^2_\mathbb{C}$ and $\rho$ is not an embedding (Theorem \[thm:CantatLamyaut\]). - Assume that $\beta\delta=0$. Since $\zeta\delta-\beta\gamma=q^2n^2$ one has $(\beta,\delta)\not=(0,0)$. Suppose that $\beta=0$. The conjugacy by $$(z_0,z_1)\mapsto\left(z_0+\frac{\gamma}{2}z_1-\frac{\gamma}{2\delta}z_1^2,z_1\right)$$ does change neither $\rho(\mathrm{e}_{13}^{qn})$, nor $\rho(\mathrm{e}_{12}^{qn})$, and sends $\rho(\mathrm{e}_{23}^{qn})$ onto $(z_0,z_1)\mapsto(z_0,z_1+\delta)$. One can thus assume that $$\begin{aligned} & \rho(\mathrm{e}_{13}^{qn})\colon(z_0,z_1)\mapsto(z_0+qn,z_1), && \rho(\mathrm{e}_{12}^{qn})\colon(z_0,z_1)\mapsto (z_0+\zeta z_1,z_1)\\ & \rho(\mathrm{e}_{23}^{qn})\colon(z_0,z_1)\mapsto (z_0,z_1+\delta).&&\end{aligned}$$ The map $\rho(\mathrm{e}_{21}^{qn})$ satisfies the relations $[\rho(\mathrm{e}_{13}^{qn}),\rho(\mathrm{e}_{21}^{qn})]=\rho(\mathrm{e}_{23}^{-q^2n^2})$ and $[\rho(\mathrm{e}_{21}^{qn}),\rho(\mathrm{e}_{23}^{qn})]=\mathrm{id}$ so does the element $\psi\colon(z_0,z_1)\mapsto(z_0,\delta nz_0+z_1)$ of $\mathrm{PGL}(3,\mathbb{C})$. Remark that the map $\phi=\rho(\mathrm{e}_{21}^{qn})\circ\psi^{-1}$ commute to both $\rho(\mathrm{e}_{13}^{qn})$ and $\rho(\mathrm{e}_{23}^{qn})$. As a consequence $$\phi\colon(z_0,z_1)\mapsto(z_0+a,z_1+b)$$ for some $a$, $b$ in $\mathbb{C}$. Finally up to conjugacy by $(z_0,z_1)\mapsto\left(z_0+\frac{b}{\delta},z_1\right)$ one has $$\rho(\mathrm{e}_{21}^{qn})\colon(z_0,z_1)\mapsto(z_0+a,\delta z_0+z_1);$$ in particular $\rho(\mathrm{e}_{21}^{qn})$ belongs to $\mathrm{PGL}(3,\mathbb{C})$. Similarly if $\varphi$ is the map given by $$(z_0,z_1)\mapsto\left(\frac{z_0}{1+\zeta z_1},\frac{z_1}{1+\zeta z_1}\right)$$ then the map $\rho(\mathrm{e}_{32}^{qn})\circ\varphi^{-1}$ commute to both $\rho(\mathrm{e}_{13}^{qn})$ and $\rho(\mathrm{e}_{12}^{qn})$. Therefore $$\rho(\mathrm{e}_{32}^{qn})\circ\varphi^{-1}\colon(z_0,z_1)\mapsto(z_0+b(z_1),z_1)$$ and $$\rho(\mathrm{e}_{32}^{qn})\colon(z_0,z_1)\mapsto\left(\frac{z_0}{1+\zeta z_1}+b\left(\frac{z_1}{1+\zeta z_1}\right),\frac{z_1}{1+\zeta z_1}\right).$$ Thanks to $[\rho(\mathrm{e}_{23}^{qn}),\rho(\mathrm{e}_{31}^{qn})]=\rho(\mathrm{e}_{21}^{q^2n^2})$, $[\rho(\mathrm{e}_{21}^{qn}),\rho(\mathrm{e}_{31}^{qn})]=\mathrm{id}$ and $[\rho(\mathrm{e}_{12}^{qn}),\rho(\mathrm{e}_{31}^{qn})]=\rho(\mathrm{e}_{32}^{-q^2n^2})$ we get $\rho(\mathrm{e}_{21}^{qn})\colon(z_0,z_1)\mapsto(z_0,\delta z_0+z_1)$. Finally since $\rho(\mathrm{e}_{31}^{qn})$ and $\rho(\mathrm{e}_{32}^{qn})$ commute, $b\equiv 0$ and $\mathrm{im}\,\rho\subset\mathrm{PGL}(3,\mathbb{C})$. Assume that $\delta=0$; using a similar reasoning we get a contradiction. Any $\rho(\mathrm{e}_{ij})$ is virtually isotopic to the identity (Lemma \[lem:nilkeylemma\] and Proposition \[pro:2fleurs\]). The maps $\rho(\mathrm{e}_{12}^n)$, $\rho(\mathrm{e}_{13}^n)$ and $\rho(\mathrm{e}_{23}^n)$ are, for some integer $n$, conjugate to automorphisms of a minimal rational surface (Proposition \[pro:sim\] and Remark \[rem:heis\]). Up to conjugacy one can assume that $\rho(\Gamma(3,n^2))\subset\mathrm{PGL}(3,\mathbb{C})$ (Lemmas \[lem:notp1p1\], \[lem:hirzcong\] and \[lem:pgl3cong\]). The restriction $\rho_{\vert\Gamma(3,n^2)}$ of $\rho$ to $\Gamma(3,n^2)$ can be extended to an endomorphism of $\mathrm{PGL}(3,\mathbb{C})$ (see [@Steinberg]). But $\mathrm{PGL}(3,\mathbb{C})$ is simple so this extension is both injective and surjective. The automorphisms of $\mathrm{PGL}(3,\mathbb{C})$ are obtained from inner automorphisms, automorphisms of the field $\mathbb{C}$ and the involution $u\mapsto u^\vee$ (see [@Dieudonne Chapter IV]). But automorphisms of the field $\mathbb{C}$ do not act on $\Gamma(3,n^2)$; hence up to linear conjugacy $\rho_{\vert\Gamma(3,n^2)}$ coincides with the identity or the involution $u\mapsto u^\vee$. Let $\phi$ be an element of $\rho(\mathrm{SL}(3,\mathbb{Z}))\smallsetminus\rho(\Gamma(3,n^2))$ that blows down at least one curve $\mathcal{C}$. The group $\Gamma(3,n^2)$ is a normal subgroup of $\Gamma$. As a consequence $\mathcal{C}$ is invariant by $\rho(\Gamma(3,n^2))$ and so by $\overline{\rho(\Gamma(3,n^2))}=\mathrm{PGL}(3,\mathbb{C})$ which is impossible. Finally $\phi$ does not blow down any curve, and $\rho(\mathrm{SL}(3,\mathbb{Z}))\subset\mathrm{PGL}(3,\mathbb{C})$. - Let $\Gamma$ be a subgroup of finite index of $\mathrm{SL}(4,\mathbb{Z})$, and let $\rho$ be a morphism from $\Gamma$ into the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. We will prove that $\mathrm{im}\,\rho$ is finite. To simplify let us suppose that $\Gamma=\mathrm{SL}(4,\mathbb{Z})$. Denote by $\mathrm{e}_{ij}$ the standard generators of $\mathrm{SL}(4,\mathbb{Z})$. The morphism $\rho$ induces a faithful representation $\widetilde{\rho}$ from $\mathrm{SL}(3,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$: $$\mathrm{SL}(4,\mathbb{Z})\supset\left( \begin{array}{cc} \mathrm{SL}(3,\mathbb{Z})& 0\\ 0 & 1 \end{array} \right)\stackrel{\widetilde{\rho}}{\to} \mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$$ According to Theorem \[thm:IMRN\] the map $\widetilde{\rho}$ is, up to birational conjugacy, the identity or the involution $u\mapsto u^\vee$. Let us first assume that up to birational conjugacy $\widetilde{\rho}=\mathrm{id}$. Assume that $\mathrm{Exc}(\rho(\mathrm{e}_{34}))\not=~\emptyset$. Since $[\mathrm{e}_{34},\mathrm{e}_{31}]=[\mathrm{e}_{34},\mathrm{e}_{32}]=\mathrm{id}$ the map $\rho(\mathrm{e}_{34})$ commutes with $$(z_0,z_1,z_2)\mapsto(z_0,z_1,az_0+bz_1+z_2)$$ where $a$, $b\in\mathbb{C}$ and $\mathrm{Exc}(\rho(\mathrm{e}_{34}))$ is invariant by $(z_0,z_1,z_2)\mapsto(z_0,z_1,az_0+bz_1+z_2)$. Moreover $\mathrm{e}_{34}$ commutes with $\mathrm{e}_{12}$ and $\mathrm{e}_{21}$, in other words $\mathrm{e}_{34}$ commutes with the following copy of $\mathrm{SL}(2,\mathbb{Z})$ $$\mathrm{SL}(4,\mathbb{Z})\supset\left( \begin{array}{ccc} \mathrm{SL}(2,\mathbb{Z})& 0 & 0\\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{array} \right)$$ The action of $\mathrm{SL}(2,\mathbb{Z})$ on $\mathbb{C}^2$ has no invariant curve so $\mathrm{Exc}(\rho(\mathrm{e}_{34}))$ is contained in the line at infinity. But the image of this line by $(z_0,z_1,z_2)\mapsto(z_0,z_1,az_0+bz_1+z_2)$ intersects $\mathbb{C}^2$: contradiction. Hence $\mathrm{Exc}(\rho(\mathrm{e}_{34}))=\emptyset$ and $\rho(\mathrm{e}_{34})$ belongs to $\mathrm{PGL}(3,\mathbb{C})$. Similarly we get that $\rho(\mathrm{e}_{43})$ belongs to $\mathrm{PGL}(3,\mathbb{C})$. The relations satisfied by the standard generators thus imply that $\rho(\mathrm{SL}(4,\mathbb{Z}))$ is contained in $\mathrm{PGL}(3,\mathbb{C})$. As a consequence $\mathrm{im}\,\rho$ is finite. A similar idea allows to conclude when $\widetilde{\rho}$ is, up to conjugacy, the involution $u\mapsto u^\vee$. - Let $n\geq 4$ be an integer. Consider a subgroup of finite index $\Gamma$ of $\mathrm{SL}(n,\mathbb{Z})$. Let $\rho$ be a morphism from $\Gamma$ to $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. According to Theorem \[thm:structsl\] the group $\Gamma$ contains a congruence subgroup $\Gamma(n,q)$. The morphism $\rho$ induces a representation $\widetilde{\rho}$ from $\Gamma(4,q)$ to $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. As we just see the kernel of this representation is infinite so does $\ker\rho$. The <span style="font-variant:small-caps;">Cremona</span> group is hopfian -------------------------------------------------------------------------- Let $V$ be a projective variety defined over a field $\Bbbk\subset\mathbb{C}$. The group $\mathrm{Aut}_\Bbbk(\mathbb{C})$ of automorphisms of the field extension $\faktor{\mathbb{C}}{\Bbbk}$ acts on $V(\mathbb{C})$, and on $\mathrm{Bir}(V)$ as follows $$\label{eq:aut} {}^{\kappa}\!\,\psi(p)=(\kappa\circ\psi\circ\kappa^{-1})(p)$$ for any $\kappa\in\mathrm{Aut}_\Bbbk(\mathbb{C})$, any $\psi\in\mathrm{Bir}(V)$, and any point $p\in V(\mathbb{C})$ for which both sides of (\[eq:aut\]) are well defined. As a consequence $\mathrm{Aut}_\Bbbk(\mathbb{C})$ acts by automorphisms on $\mathrm{Bir}(V)$. If $\kappa\colon\mathbb{C}\to\mathbb{C}$ is a field morphism, then this construction gives an injective morphism $$\begin{aligned} &\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})\to\mathrm{Aut}(\mathbb{P}^n_\mathbb{C}), && g\mapsto {}^{\kappa}\!\,g.\end{aligned}$$ Write $\mathbb{C}$ as the algebraic closure of a purely transcendental extension $\mathbb{Q}(x_i,\,i\in I)$ of $\mathbb{Q}$; if $f\colon I\to I$ is an injective map, then there exists a field morphism $$\begin{aligned} & \kappa\colon\mathbb{C}\to\mathbb{C}, && x_i\mapsto x_{f(i)}.\end{aligned}$$ Such a morphism is surjective if and only if $f$ is onto. The group $\mathrm{Aut}(\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}))$ has been described in [@Deserti:abelien] and [@Deserti:IMRN] via two different methods: Let $\varphi$ be an element of $\mathrm{Aut}(\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}))$. Then there exist a birational self map $\psi$ of $\mathbb{P}^2_\mathbb{C}$ and an automorphism $\kappa$ of the field $\mathbb{C}$ such that $$\varphi(\phi)={}^{\kappa}\!\,(\psi\circ\phi\circ\psi^{-1})\qquad\qquad \forall\,\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$$ The proof of [@Deserti:abelien] will be deal with in §\[sec:autbir\]. The proof of [@Deserti:IMRN] can in fact be used to describe the endomorphisms of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group: \[thm:hopfian\] Let $\varphi$ be a non-trivial endomorphism of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Then there exist $\psi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and an immersion $\kappa$ of the field $\mathbb{C}$ such that $$\varphi(\phi)={}^{\kappa}\!\,(\psi\circ\phi\circ\psi^{-1})\qquad\qquad \forall\,\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$$ Let us work in the affine chart $z_2=1$. The group of translations is $$\mathrm{T}=\big\{(z_0,z_1)\mapsto(z_0+\alpha,z_1+\beta)\,\vert\,\alpha,\,\beta\in\mathbb{C}\big\}.$$ \[lem:com\] Let $\varphi$ be a birational self map of $\mathbb{P}^2_\mathbb{C}$. Assume that $\varphi$ commutes with both $(z_0,z_1)\mapsto(z_0+1,z_1)$ and $(z_0,z_1)\mapsto(z_0,z_1+1)$. Then $\varphi$ belongs to $\mathrm{T}$. Let $\varphi=(\varphi_0,\varphi_1)$ be an element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that commutes with both $(z_0,z_1)\mapsto(z_0+1,z_1)$ and $(z_0,z_1)\mapsto(z_0,z_1+1)$. In particular $$\left\{ \begin{array}{ll} \varphi_0(z_0+1,z_1)=\varphi_0(z_0,z_1)+1\\ \varphi_1(z_0+1,z_1)=\varphi_1(z_0,z_1) \end{array} \right.$$ From $\varphi_1(z_0+1,z_1)=\varphi_1(z_0,z_1)$ we get that $\varphi_1=\varphi_1(z_1)$. The equality $\varphi_0(z_0+1,z_1)=\varphi_0(z_0,z_1)+1$ implies $$\frac{\partial\varphi_0}{\partial z_0}(z_0+1,z_1)=\frac{\partial\varphi_0}{\partial z_0}(z_0,z_1);$$ as a consequence $\frac{\partial\varphi_0}{\partial z_0}=a(z_1)$ and $\varphi_0=a(z_1)z_0+b(z_1)$ for some $a$, $b$ in $\mathbb{C}(z_1)$. Then $$\varphi_0(z_0+1,z_1)=\varphi_0(z_0,z_1)+1$$ yields $a(z_1)=1$. In other words $\varphi\colon(z_0,z_1)\dashrightarrow(z_0+b(z_1),\varphi_1(z_1))$. Let us now write that $\varphi\circ(z_0,z_1+1)\colon(z_0,z_1)\dashrightarrow(z_0,z_1+1)\circ\varphi$; we get that $\varphi\colon(z_0,z_1)\dashrightarrow(\varphi_0(z_0),z_1+c(z_0))$. Finally $\varphi\colon(z_0,z_1)\dashrightarrow(z_0+b(z_1),\varphi_1(z_1))$ and $\varphi\colon(z_0,z_1)\dashrightarrow(\varphi_0(z_0),z_1+c(z_0))$ imply that $\varphi$ belongs to $\mathrm{T}$. Since $\mathrm{PGL}(3,\mathbb{C})$ is simple the restriction $\varphi_{\vert\mathrm{PGL}(3,\mathbb{C})}$ is either trivial or injective. Let us first suppose that $\varphi_{\vert\mathrm{PGL}(3,\mathbb{C})}$ is trivial. Consider the element of $\mathrm{PGL}(3,\mathbb{C})$ given by $$\ell\colon(z_0,z_1)\mapsto\left(\frac{z_0}{z_0-1},\frac{z_0-z_1}{z_0-1}\right).$$ According to [@Gizatullin:relations] one has $(\ell\circ\sigma_2)^3=\mathrm{id}$. As a result $\varphi((\ell\circ\sigma_2)^3)=\mathrm{id}$. Since $\varphi(\ell)=\ell$ (recall that $\ell$ belongs to $\mathrm{PGL}(3,\mathbb{C})$) one gets that $\varphi(\sigma_2)=\mathrm{id}$. As the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is generated by $\mathrm{PGL}(3,\mathbb{C})$ and $\sigma_2$ one gets that $\varphi=\mathrm{id}$. Assume now that $\varphi_{\vert\mathrm{PGL}(3,\mathbb{C})}$ is injective. According to Theorem \[thm:IMRN\] the restriction $\varphi_{\vert\mathrm{SL}(3,\mathbb{Z})}$ of $\varphi$ to $\mathrm{SL}(3,\mathbb{Z})$ is, up to inner conjugacy, the canonical embedding or $A\mapsto A^\vee$. - Suppose first that $\varphi_{\vert\mathrm{SL}(3,\mathbb{Z})}$ is the canonical embedding. Denote by $\mathcal{U}$ the group of unipotent upper triangular matrices. Set $$\begin{aligned} & f_\beta=\varphi(z_0+\beta,z_1), && g_\alpha=\varphi(z_0+\alpha,z_1), && h_\gamma=\varphi(z_0,z_1+\gamma). \end{aligned}$$ Since $f_\beta$ and $h_\gamma$ commute to both $(z_0,z_1)\mapsto(z_0+1,z_1)$ and $(z_0,z_1)\mapsto(z_0,z_1+1)$ one gets from Lemma \[lem:com\] that $$\begin{aligned} & f_\beta\colon(z_0,z_1)\mapsto(z_0+\lambda(\beta),z_1+\zeta(\beta)) && h_\gamma\colon(z_0,z_1)\mapsto(z_0+\eta(\gamma),z_1+\mu(\gamma))\end{aligned}$$ where $\lambda$, $\zeta$, $\eta$ and $\mu$ are additive morphisms from $\mathbb{C}$ to $\mathbb{C}$. As $g_\gamma$ commutes with $(z_0,z_1)\mapsto(z_0+z_1,z_1)$ and $(z_0,z_1)\mapsto(z_0+1,z_1)$ there exists $a_\alpha$ in $\mathbb{C}(y)$ such that $$g_\gamma\colon(z_0,z_1)\mapsto(z_0+a_\alpha(z_1),z_1).$$ The equality $$(z_0+\alpha z_1,z_1)\circ(z_0,z_1+\gamma)\circ(z_0+\alpha z_1,z_1)^{-1}\circ(z_0,z_1+\gamma)^{-1}=(z_0+\alpha z_1,z_1)$$ implies that $g_\alpha\circ h_\alpha=f_{\alpha\gamma}\circ h_\gamma\circ g_\alpha$ for any $\alpha$, $\gamma$ in $\mathbb{C}$. As a consequence $$\begin{aligned} & f_\beta\colon(z_0,z_1)\mapsto(z_0+\lambda(\beta),z_1) && g_\alpha\colon(z_0,z_1)\mapsto(z_0+\theta(\alpha)z_1+\zeta(\alpha),z_1)\end{aligned}$$ and $\theta(\alpha)\mu(\alpha)=\lambda(\alpha\gamma)$. From $$\Big[\big((z_0,z_1)\mapsto(z_0+\alpha,z_1)\big),\big((z_0,z_1)\mapsto(z_0,z_1+\beta z_0)\big)\Big]=\big((z_0,z_1)\mapsto(z_0,z_1-\alpha)\big)$$ one gets $h_\gamma\colon(z_0,z_1)\mapsto(z_0,z_1+\mu(\gamma))$. In other words for any $\alpha$, $\beta\in\mathbb{C}$ one has $$\varphi(z_0+\alpha,z_1+\beta)=f_\alpha\circ h_\beta=(z_0,z_1)\mapsto(z_0+\lambda(\alpha),z_1+\mu(\beta)).$$ Therefore $\varphi(\mathrm{T})\subset\mathrm{T}$ and $\varphi(\mathcal{U})\subset\mathcal{U}$. Since $\mathrm{PGL}(3,\mathbb{C})=\langle\mathcal{U},\,\mathrm{SL}(3,\mathbb{Z})\rangle$ the inclusion $\varphi(\mathrm{PGL}(3,\mathbb{C}))\subset\mathrm{PGL}(3,\mathbb{C})$ holds. According to [@BorelTits] the action of $\varphi$ on $\mathrm{PGL}(3,\mathbb{C})$ comes, up to inner conjugacy, from an embedding of the field $\mathbb{C}$ into itself. - Assume now that $\varphi_{\vert\mathrm{SL}(3,\mathbb{Z})}$ is $A\mapsto A^\vee$. Similar computations and [@BorelTits] imply that $\varphi_{\vert\mathrm{PGL}(3,\mathbb{C})}$ comes, up to inner conjugacy, from the composition of $A\mapsto A^\vee$ and an embedding of the field $\mathbb{C}$ into itself. To finish let us assume for instance that $\varphi_{\vert\mathrm{PGL}(3,\mathbb{C})}$ comes, up to inner conjugacy, from the composition of $A\mapsto A^\vee$ and an embedding of the field $\mathbb{C}$ into itself. Set $(\eta_1,\eta_2)=\varphi\left((z_0,z_1)\mapsto\left(z_0,\frac{1}{z_1}\right)\right)$. From $$\left((z_0,z_1)\dashrightarrow\left(z_0,\frac{1}{z_1}\right)\right)\circ((z_0,z_1)\mapsto(\alpha z_0,\beta z_1))\circ\left((z_0,z_1)\dashrightarrow\left(z_0,\frac{1}{z_1}\right)\right)=\left((z_0,z_1)\mapsto\left(\alpha z_0,\frac{z_1}{\beta}\right)\right)$$ one gets $$\left\{ \begin{array}{ll} \eta_1\big(\lambda(\alpha^{-1})z_0,\lambda(\beta^{-1})z_1\big)=\lambda(\alpha^{-1})\eta_1(z_0,z_1)\\ \eta_2\big(\lambda(\alpha^{-1})z_0,\lambda(\beta^{-1})z_1\big)=\lambda(\beta)\eta_2(z_0,z_1) \end{array} \right.$$ Hence $\varphi\left((z_0,z_1)\mapsto\left(z_0,\frac{1}{z_1}\right)\right)=\left((z_0,z_1)\mapsto\left(\pm z_0,\pm\frac{1}{z_1}\right)\right)$. But $$\left(\big((z_0,z_1)\mapsto(z_1,z_0)\big)\circ\left((z_0,z_1)\mapsto\left(z_0,\frac{1}{z_1}\right)\right)\right)^2=\sigma_2$$ so $\varphi(\sigma_2)=\pm\sigma_2$. Furthermore $\varphi(\ell)=\big((z_0,z_1)\mapsto(-z_0-z_1-1,z_1)\big)$ as $\varphi_{\vert\mathrm{SL}(3,\mathbb{Z})}$ coincides with $A\mapsto A^\vee$. Then the second component of $\varphi\big(\ell\circ\sigma_2\big)^3$ is $\pm\frac{1}{z_1}$: contradiction with $\varphi\big(\ell\circ\sigma_2\big)^3=\mathrm{id}$. If $\varphi_{\vert\mathrm{PGL}(3,\mathbb{C})}$ comes, up to inner conjugacy, from an embedding of $\mathbb{C}$ similar computations imply that $\varphi(\sigma_2)=\sigma_2$ and one concludes with <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> theorem. Finite subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group {#chapter:finite} =================================================================================================== The classification of finite subgroups of $\mathrm{Bir}(\mathbb{P}^1_\mathbb{C})=\mathrm{PGL}(2,\mathbb{C})$ is well known and goes back to <span style="font-variant:small-caps;">Klein</span>. It consists of cyclic, dihedral, tetrahedral, octahedral and icosahedral groups. Groups of the same type and same order constitute a unique conjugacy class in $\mathrm{Bir}(\mathbb{P}^1_\mathbb{C})$. What about the $2$-dimensional case, [i.e.]{} what about the finite subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ ? The story starts a long time ago with <span style="font-variant:small-caps;">Bertini</span> ([@Bertini]) who classified conjugacy classes of subgroups of order $2$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Already the answer is drastically different from the $1$-dimensional case. The set of conjugacy classes is parameterized (see Theorem \[thm:BayleBeauville\]) by a disconnected algebraic variety whose connected components are respectively isomorphic to - either the moduli spaces of hyperelliptic curves of genus $g$, - or the moduli space of canonical curves of genus $3$, - or the moduli space of canonical curves of genus $4$ with vanishing theta characteristic. <span style="font-variant:small-caps;">Bertini</span>’s proof is considered to be incomplete; a complete and short proof was published only a few years ago by <span style="font-variant:small-caps;">Bayle</span> and <span style="font-variant:small-caps;">Beauville</span> ([@BayleBeauville]). In $1894$ <span style="font-variant:small-caps;">Castelnuovo</span> proved that any element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of finite order leaves invariant either a net of lines, or a pencil of lines, or a linear system of cubic curves with $n\leq 8$ base-points ([@Castelnuovo]). <span style="font-variant:small-caps;">Kantor</span> announced a similar result for arbitrary finite subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$; his proof relies on a classification of possible groups in each case ([@Kantor]). Unfortunately <span style="font-variant:small-caps;">Kantor</span>’s classification, even with some corrections made by <span style="font-variant:small-caps;">Wiman</span> ([@Wiman]), is incomplete in the following sense: - given some abstract finite group, it is not possible using their list to say whether this group is isomorphic to a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$; - the possible conjugation between the groups of the list is not considered. The Russian school has made great progress since the $1960$’s: <span style="font-variant:small-caps;">Manin</span> and <span style="font-variant:small-caps;">Iskovskikh</span> classified the minimal $G$-surfaces into automorphisms of <span style="font-variant:small-caps;">del Pezzo</span> surfaces and of conic bundles ([@Manin:rational2; @Iskovskih:minimal]). Many years after people come back to this problem. As we already mention <span style="font-variant:small-caps;">Bayle</span> and <span style="font-variant:small-caps;">Beauville</span> classified groups of order $2$. It is the first example of a precise description of conjugacy classes; it is shown that the non-rational curves fixed by the groups determine the conjugacy classes. Groups of prime order were also studied ([@BeauvilleBlanc; @deFernex; @Zhang]). <span style="font-variant:small-caps;">Zhang</span> applies <span style="font-variant:small-caps;">Bayle</span> and <span style="font-variant:small-caps;">Beauville</span> strategy to the case of birational automorphisms of prime order $p\geq 3$. It turns out that nonlinear automorphisms occur only for $p=3$ and $p=5$; the author describes them explicitly. The techniques of [@BayleBeauville] are also generalized by <span style="font-variant:small-caps;">de Fernex</span> to cyclic subgroups of prime order ([@deFernex]). The list is as precise as one can wish, except for two classes of groups of order $5$: the question of their conjugacy is not answered. <span style="font-variant:small-caps;">Beauville</span> and <span style="font-variant:small-caps;">Blanc</span> completed this classification ([@BeauvilleBlanc]); they prove in particular that a birational self map of the complex projective plane of prime order is not conjugate to a linear automorphism if and only if it fixes some non-rational curve. <span style="font-variant:small-caps;">Beauville</span> classified $p$-elementary groups ([@Beauville]). <span style="font-variant:small-caps;">Blanc</span> classified all finite cyclic groups ([@Blanc:CRAS]), and all finite abelian groups ([@Blanc:these]). The goal of [@DolgachevIskovskikh] is to update the list of <span style="font-variant:small-caps;">Kantor</span> and <span style="font-variant:small-caps;">Wiman</span>. The authors used the modern theory of $G$-surfaces, the theory of elementary links, and the conjugacy classes of <span style="font-variant:small-caps;">Weyl</span> groups. In the first section we recall the definitions of <span style="font-variant:small-caps;">Geiser</span> involutions, <span style="font-variant:small-caps;">Bertini</span> involutions and <span style="font-variant:small-caps;">Jonquières</span> involutions. We give a sketch of the proof of the classification of birational involutions of the complex projective plane due to <span style="font-variant:small-caps;">Bayle</span> and <span style="font-variant:small-caps;">Beauville</span>. In the second section we deal with finite abelian subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. Results due to <span style="font-variant:small-caps;">Dolgachev</span> and <span style="font-variant:small-caps;">Iskovskikh</span> are recalled. In the last section we state some results of <span style="font-variant:small-caps;">Blanc</span> about finite cyclic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, isomorphism classes of finite abelian subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ but also a generalization of a theorem of <span style="font-variant:small-caps;">Castelnuovo</span> which states that an element of finite order which fixes a curve of geometric genus $>1$ has order $2$, $3$ or $4$. Classification of subgroups of order $2$ {#sec:ordre2} ---------------------------------------- ### <span style="font-variant:small-caps;">Geiser</span> involutions {#geiser} Let $p_1$, $p_2$, $\ldots$, $p_7$ be seven points of the complex projective plane in general position. Denote by $L$ the linear system of cubics through the $p_i$’s. The linear system $L$ of cubic curves through the $p_i$’s is two-dimensional. Take a general point $p$, and consider the pencil of curves from $L$ passing through $p$. A general pencil of cubic curves has nine base-points; let us define $\mathcal{I}_G(p)$ as the ninth base-point of the pencil. The map $\mathcal{I}_G$ is a <span style="font-variant:small-caps;">Geiser</span> involution ([@Geiser]). The algebraic degree of a <span style="font-variant:small-caps;">Geiser</span> involution is equal to $8$. One can also see a <span style="font-variant:small-caps;">Geiser</span> involution as follows. The linear system $L$ defines a rational map of degree $2$, $$\psi\colon \mathbb{P}^2_\mathbb{C}\dashrightarrow\vert L\vert^\simeq \mathbb{P}^2_\mathbb{C}.$$ The points $p$ and $\mathcal{I}_G(p)$ lie in the same fibre. As a consequence $\mathcal{I}_G$ is a birational deck map of this cover. If we blow up $p_1$, $p_2$, $\ldots$, $p_7$ we get a <span style="font-variant:small-caps;">del Pezzo</span> surface $S$ of degree $2$ and a regular map of degree $2$ from $S$ to $\mathbb{P}^2_\mathbb{C}$. Furthermore the <span style="font-variant:small-caps;">Geiser</span> involution becomes an automorphism of $S$. Note that the fixed points of $\mathcal{I}_G$ lie on the ramification curve of $\psi$. It is a curve of degree $6$ with double points $p_1$, $p_2$, $\ldots$, $p_7$ and is birationally isomorphic to a canonical curve of genus $3$. A third way to see <span style="font-variant:small-caps;">Geiser</span> involutions is the following. Let $S$ be a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $2$. The linear system $\vert -K_S\vert$ defines a double covering $S\to\mathbb{P}^2_\mathbb{C}$, branched along a smooth quartic curve ([@Demazure:delPezzo]). The involution $\iota$ which exchanges the two sheets of this covering is called a <span style="font-variant:small-caps;">Geiser</span> involution; it satisfies $$\mathrm{Pic}(S)^\iota\otimes\mathbb{Q}\simeq\mathrm{Pic}(\mathbb{P}^2_\mathbb{C})\otimes\mathbb{Q}=\mathbb{Q}.$$ The exceptional locus of a <span style="font-variant:small-caps;">Geiser</span> involution is the union of seven cubics passing through the seven points of indeterminacy of $\mathcal{I}_G$ and singular at one of these seven points. ### <span style="font-variant:small-caps;">Bertini</span> involutions {#bertini} Let us fix in $\mathbb{P}^2_\mathbb{C}$ eight points $p_1$, $p_2$, $\ldots$, $p_8$ in general position. Consider the pencil of cubic curves through these points. It has a ninth base-point $p_9$. For any general point $p$ there is a unique cubic curve $\mathcal{C}(p)$ of the pencil passing through $p$. Take $p_9$ as the zero of the group law of the cubic $\mathcal{C}(p)$; define $\mathcal{I}_B(p)$ as the negative $-p$ with respect to the group law. The map $\mathcal{I}_B$ is a birational involution called <span style="font-variant:small-caps;">Bertini</span> involution* ([@Bertini]). The algebraic degree of a <span style="font-variant:small-caps;">Bertini</span> involution is equal to $17$. The fixed points of a <span style="font-variant:small-caps;">Bertini</span> involution lie on a canonical curve of genus $4$ with vanishing theta characteristic isomorphic to a nonsingular intersection of a cubic surface and a quadratic cone in $\mathbb{P}^3_\mathbb{C}$. Another way to see a <span style="font-variant:small-caps;">Bertini</span> involution is the following. Consider a <span style="font-variant:small-caps;">del Pezzo</span> surface $S$ of degree $1$. The map $S\to\mathbb{P}^3_\mathbb{C}$ defined by the linear system $\vert -2K_S\vert$ induces a degree $2$ morphism of $S$ onto a quadratic cone $Q\subset\mathbb{P}^3_\mathbb{C}$, branched along the vertex of $Q$ and a smooth genus $4$ curve ([@Demazure:delPezzo]). The corresponding involution, the <span style="font-variant:small-caps;">Bertini</span> involution, satisfies $\mathrm{rk}\,\mathrm{Pic}(S)^{\mathcal{I}_B}=1$. ### <span style="font-variant:small-caps;">Jonquières</span> involutions Let $\mathcal{C}$ be an irreducible curve of degree $\nu\geq 3$. Assume that $\mathcal{C}$ has a unique singular point $p$ and that $p$ is an ordinary multiple point with multiplicity $\nu-2$. To $(\mathcal{C},p)$ we associate a birational involution $\mathcal{I}_J$ that fixes pointwise $\mathcal{C}$ and preserves lines through $p$. Let $m$ be a generic point of $\mathbb{P}^2_\mathbb{C}\smallsetminus\mathcal{C}$. Let $r_m$, $q_m$ and $p$ be the intersections of the line $(mp)$ and $\mathcal{C}$. The point $\mathcal{I}_J(m)$ is the point such that the cross ratio of $m$, $\mathcal{I}_J(m)$, $q_m$ and $r_m$ is equal to $-1$. The map $\mathcal{I}_J$ is a <span style="font-variant:small-caps;">Jonquières</span> involution of degree $\nu$ centered at $p$; it preserves $\mathcal{C}$. More precisely its fixed points are the curve $\mathcal{C}$ of genus $\nu-2$ as soon as $\nu\geq 3$. If $\nu=2$, then $\mathcal{C}$ is a smooth conic ; the same construction can be done by choosing a point $p$ that does not lie on $\mathcal{C}$. \[lem:notconj\] Let $\mathrm{G}$ be a finite subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let $C_1$, $C_2$, $\ldots$, $C_k$ be non-rational irreducible curves on $\mathbb{P}^2_\mathbb{C}$ such that each of them contains an open subset $C_i^0$ whose points are fixed under all $g\in\mathrm{G}$. Then the set of birational isomorphism classes of the curves $C_i$ is an invariant of the conjugacy class of $\mathrm{G}$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Assume that $\mathrm{G}=\psi\circ\mathrm{H}\circ\psi^{-1}$ for some subgroup $\mathrm{H}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and some birational self map $\psi$ of the complex projective plane. Replacing $C_i^0$ by a smaller open subset if needed we assume that $\psi^{-1}(C_i^0)$ is defined and consists of fixed points of $\mathrm{H}$. As $C_i$ is not rational, $\psi^{-1}(C_i^0)$ is not a point. Its <span style="font-variant:small-caps;">Zariski</span> closure is thus a rational irreducible curve $C'_i$ birationally isomorphic to $C_i$ that contains an open subset of fixed points of $\mathrm{H}$. <span style="font-variant:small-caps;">Jonquières</span> involutions of degree $\geq 3$ are not conjugate to each other, not conjugate to projective involutions, not conjugate to <span style="font-variant:small-caps;">Bertini</span> involutions, not conjugate to <span style="font-variant:small-caps;">Geiser</span> involutions. <span style="font-variant:small-caps;">Bertini</span> involutions are not conjugate to <span style="font-variant:small-caps;">Geiser</span> involutions, not conjugate to projective involutions. <span style="font-variant:small-caps;">Geiser</span> involutions are not conjugate to projective involutions. The statement follows from Lemma \[lem:notconj\] and the above properties: - a connected component of the fixed locus of a projective map is a line or a point; - the fixed points of a <span style="font-variant:small-caps;">Geiser</span> involution lie on a curve birationally isomorphic to a canonical curve of genus $3$; - the fixed points of a <span style="font-variant:small-caps;">Bertini</span> involution lie on a canonical curve of genus $4$ with vanishing theta characteristic; - the set of fixed points of a <span style="font-variant:small-caps;">Jonquières</span> involution of degree $\nu\geq 3$ outside the base locus is an hyperelliptic curve of degree $\nu-2$. We can thus introduce the following definition. An involution is of <span style="font-variant:small-caps;">Jonquières</span> type if it is birationally conjugate to a <span style="font-variant:small-caps;">Jonquières</span> involution. An involution is of <span style="font-variant:small-caps;">Bertini</span> type if it is birationally conjugate to a <span style="font-variant:small-caps;">Bertini</span> involution. An involution is of <span style="font-variant:small-caps;">Geiser</span> type if it is birationally conjugate to a <span style="font-variant:small-caps;">Geiser</span> involution. The classification of subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of order $2$ is given by the following statement: \[thm:BayleBeauville\] A non-trivial birational involution of the complex projective plane is conjugate to one and only one of the following: - a <span style="font-variant:small-caps;">Jonquières</span> involution of a given degree $\nu\geq 2$; - a <span style="font-variant:small-caps;">Geiser</span> involution; - a <span style="font-variant:small-caps;">Bertini</span> involution. More precisely the parameterization of each conjugacy class is known. Before stating it let us give some definitions. \[rem:norm\] Let $S$, $S'$ be two rational surfaces and $\iota\in\mathrm{Bir}(S)$, $\iota'\in\mathrm{Bir}(S')$ be two involutions. They are birationally equivalent if there exists a birational map $\varphi\colon S\dashrightarrow S'$ such that $\varphi\circ\iota=\iota'\circ\varphi$. Note that in particular two involutions of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ are equivalent if and only if they are conjugate in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Assume that $\iota$ fixes a curve $C$. Then $\iota'=\varphi\circ\iota\circ\varphi^{-1}$ fixes the proper transform of $C$ under $\varphi$ which is a curve birational to $C$ except possibly if $C$ is rational; indeed, if $C$ is rational it may be contracted to a point. The normalized fixed curve of $\iota$ is the union of the normalizations of the non-rational curves fixed by $\iota$. This is an invariant of the birational equivalence class of $\iota$. The map which associates to a birational involution of $\mathbb{P}^2_\mathbb{C}$ its normalized fixed curve establishes a one-to-one correspondence between - conjugacy classes of <span style="font-variant:small-caps;">Jonquières</span> involutions of degree $\nu$ and isomorphism classes of hyperelliptic curves of genus $\nu-2$ $(\nu\geq 3)$; - conjugacy classes of <span style="font-variant:small-caps;">Geiser</span> involutions and isomorphism classes of non-hyperelliptic curves of genus $3$; - conjugacy classes of <span style="font-variant:small-caps;">Bertini</span> involutions and isomorphism classes of non-hyperelliptic curves of genus $4$ whose canonical model lies on a singular quadric. <span style="font-variant:small-caps;">Jonquières</span> involutions of degree $2$ form one conjugacy class. The approach of <span style="font-variant:small-caps;">Bayle</span> and <span style="font-variant:small-caps;">Beauville</span> is different from the approach of <span style="font-variant:small-caps;">Castelnuovo</span>. It is based on the following observation: any birational involution of $\mathbb{P}^2_\mathbb{C}$ is conjugate, via an appropriate birational isomorphism $S\stackrel{\sim}{\dashrightarrow}\mathbb{P}^2_\mathbb{C}$ to a biregular involution $\iota$ of a rational surface $S$. Therefore the authors are reduced to the birational classification of the pairs $(S,\iota)$. In [@Manin:rational2] <span style="font-variant:small-caps;">Manin</span> classified the pairs $(S,\mathrm{G})$ where $S$ is a surface and $\mathrm{G}$ a finite group. This question has been simplified by the introduction of <span style="font-variant:small-caps;">Mori</span> theory. This theory allows <span style="font-variant:small-caps;">Bayle</span> and <span style="font-variant:small-caps;">Beauville</span> to show that the minimal pairs $(S,\iota)$ fall into two categories, those which admit a $\iota$-invariant base-point free pencil of rational curves, and those with $\mathrm{rk}\,\mathrm{Pic}(S)^{\iota}=1$. The first case leads to the so-called <span style="font-variant:small-caps;">Jonquières</span> involutions whereas the second one leads to the <span style="font-variant:small-caps;">Geiser</span> and <span style="font-variant:small-caps;">Bertini</span> involutions. Let us now give some details. By a surface we mean a smooth, projective, connected surface over $\mathbb{C}$. We consider pairs $(S,\iota)$ where $S$ is a rational surface and $\iota$ a non-trivial biregular involution of $S$. Recall that the pair $(S,\iota)$ is minimal if any birational morphism $\psi\colon S\to S'$ such that there exists a biregular involution $\iota'$ of $S'$ with $\psi\circ\iota=\iota'\circ\psi$ is an isomorphism. The pair $(S,\iota)$ is minimal if and only if for any exceptional curve[^4] $E$ on $S$ the following hold: $$\begin{aligned} &\iota(E)\not=E &&E\cap\iota(E)\not=\emptyset.\end{aligned}$$ Suppose that $(S,\iota)$ is not minimal. Then there exist a pair $(S',\iota')$ and a birational morphism $\psi\colon S\to S'$ such that $\psi\circ\iota=\iota'\circ\psi$ and $\psi$ contracts some exceptional curve $E$. Then $\psi$ contracts the divisor $E+\iota(E)$. Therefore $(E+\iota(E))^2\leq 0$ and so $E\cdot\iota(E)\leq 0$, [i.e.]{} $\iota(E)=E$ or $E\cap\iota(E)=\emptyset$. Conversely assume that there exists an exceptional curve $E$ on $S$ such that $\iota(E)=E$ (resp. $E\cap\iota(E)=\emptyset$). Let $S'$ be the surface obtained by blowing down $E$ (resp. $E\cup\iota(E)$). Then $\iota$ induces an involution $\iota'$ of $S'$ so that $(S,\iota)$ is not minimal. The only piece of <span style="font-variant:small-caps;">Mori</span> theory used by <span style="font-variant:small-caps;">Bayle</span> and <span style="font-variant:small-caps;">Beauville</span> is the following one: \[lem:pencil\] Let $(S,\iota)$ be a minimal pair with $\mathrm{rk}\,\mathrm{Pic}(S)^\iota>1$. Then $S$ admits a base-point free pencil stable under $\iota$. It allows them to establish the: \[thm:minimalpairs\] Let $(S,\iota)$ be a minimal pair. One of the following holds: 1. there exists a smooth $\mathbb{P}^1_\mathbb{C}$-fibration $f\colon S\to\mathbb{P}^1_\mathbb{C}$ and a non-trivial involution $\mathcal{I}$ of $\mathbb{P}^1_\mathbb{C}$ such that $f\circ\iota=\mathcal{I}\circ f$; 2. there exists a fibration $f\colon S\to\mathbb{P}^1_\mathbb{C}$ such that $f\circ\iota=f$, the smooth fibres of $f$ are rational curves on which $\iota$ induces a non-trivial involution, any singular fibre is the union of two rational curves exchanged by $\iota$, meeting at one point; 3. $S$ is isomorphic to $\mathbb{P}^2_\mathbb{C}$; 4. $(S,\iota)$ is isomorphic to $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ with the involution $(z_0,z_1)\mapsto(z_1,z_0)$; 5. $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $2$ and $\iota$ is the <span style="font-variant:small-caps;">Geiser</span> involution; 6. $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $1$ and $\iota$ is the <span style="font-variant:small-caps;">Bertini</span> involution. - Assume $\mathrm{rk}\,\mathrm{Pic}(S)^\iota=1$. As $\mathrm{Pic}(S)^\iota$ contains an ample class, $-K_S$ is ample, [i.e.]{} $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface. If $\mathrm{rk}\,\mathrm{Pic}(S)=1$, then one obtains case $(3)$. If $\mathrm{rk}\,\mathrm{Pic}(S)>1$, then $-\iota$ is the orthogonal reflection with respect to $K_S^\perp$. Such a reflection is of the form $$x\mapsto x-2\,\frac{(\alpha\cdot x)}{(\alpha\cdot\alpha)}\,\alpha$$ with $(\alpha\cdot\alpha)\in\{1,\,2\}$ and $K_S$ proportional to $\alpha$. If $K_S$ is divisible, then $S$ is isomorphic to $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ and since $\iota$ must act non-trivially on $\mathrm{Pic}(S)$ we get case $(4)$. The only remaining eventualities are $K_S^2\in\{1,\,2\}$. The <span style="font-variant:small-caps;">Geiser</span> and <span style="font-variant:small-caps;">Bertini</span> involutions have the required properties (§\[geiser\], §\[bertini\]). An automorphism $\varphi$ of $S$ acting trivially on $\mathrm{Pic}(S)$ is the identity; indeed $S$ is the blow up of $\mathbb{P}^2_\mathbb{C}$ at $9-d$ points in general position, $\varphi$ induces an automorphism of $\mathbb{P}^2_\mathbb{C}$ which must fix these points. Hence <span style="font-variant:small-caps;">Geiser</span> and <span style="font-variant:small-caps;">Bertini</span> involutions are the only ones to have the required properties. - Suppose now that $\mathrm{rk}\,\mathrm{Pic}(S)^\iota>1$. According to Lemma \[lem:pencil\] the surface $S$ admits a $\iota$-invariant pencil $\vert F\vert$ of rational curves. This defines a fibration $f\colon S\to\mathbb{P}^1_\mathbb{C}$ with fibre $F$, and an involution $\mathcal{I}$ of $\mathbb{P}^1_\mathbb{C}$ such that $f\circ\iota=\mathcal{I}\circ f$. If $f$ is smooth, then this gives $(1)$ or a particular case of $(2)$. If $f$ is not smooth, let $F_0$ be a singular fibre of $f$. It contains an exceptional divisor $E$. Since $(S,\iota)$ is minimal, then $\iota(E)\not=E$ and $E\cdot\iota(E)\geq 1$. As a result $(E+\iota(E))^2\geq 0$ so $F_0=E+\iota(E)$ and $E\cdot\iota(E)=1$. Set $p=E\cap\iota(E)$. The involution induced by $\iota$ on $T_pS$ exchanges the directions of $E$ and $\iota(E)$; it thus has eigenvalues $1$ and $-1$. As a consequence $\iota$ fixes a curve passing through $p$; this curve must be horizontal and $\mathcal{I}$ trivial. Furthermore the fixed curve of $\iota$ being smooth, the involution induced by $\iota$ on a smooth fibre cannot be trivial. We get case $(2)$. <span style="font-variant:small-caps;">Bayle</span> and <span style="font-variant:small-caps;">Beauville</span> precised which pairs in the list of Theorem \[thm:minimalpairs\] are indeed minimal ([@BayleBeauville Proposition 1.7]). Let us now give the link between biregular involutions of rational surfaces and birational involutions of the complex projective plane: \[lem:linkbirinvbirinv\] Let $\iota$ be a birational involution of a surface $S_1$. There exists a birational morphism $\varphi\colon S\to S_1$ and a biregular involution $\mathcal{I}$ of $S$ such that $\varphi\circ\mathcal{I}=\iota\circ\varphi$. To prove it we need some results, let us state and prove them. \[thm:morph\] Let $S$ be a surface, and let $X$ be a projective variety. Let $\phi\colon S\dashrightarrow X$ be a rational map. Then there exist - a surface $S'$, - a morphism $\eta\colon S'\to S$ which is the composition of a finite number of blow-ups, - a morphism $\psi\colon S'\to X$ such that $$\xymatrix{ & S'\ar[rd]^\psi\ar[ld]_\eta & \\ S\ar@{-->}[rr]_\phi & & X }$$ commutes. As $X$ lies in some projective space we may assume that $X=\mathbb{P}^m_\mathbb{C}$. Furthermore we can suppose that $\phi(S)$ lies in no hypersurface of $\mathbb{P}^m_\mathbb{C}$. As a result $\phi$ corresponds to a linear system $P\subset\vert D\vert$ of dimension $m$ on $S$ without fixed component. If $P$ has no base-point, then $\phi$ is a morphism and there is nothing to do. Assume that $P$ has at least one base-point $p$. Consider the blow up $\varepsilon\colon \mathrm{Bl}_pS\to S$ at $p$. Set $S_1=\mathrm{Bl}_pS$. The exceptional curve $E$ occurs in the fixed part of the linear system $\varepsilon^P\subset\vert\varepsilon^D\vert$ with some multiplicity $k\geq 1$; that is, the system $P_1=\vert\varepsilon^P-kE\vert\subset\vert\varepsilon^D-kE\vert$ has no fixed component. It thus defines a rational map $\phi_1=\phi\circ\varepsilon\colon S_1\dashrightarrow\mathbb{P}^m_\mathbb{C}$. If $\phi_1$ is a morphism, then the result is proved. If not, we repeat the “same step”. We get by induction a sequence $\varepsilon_n\colon S_n\to S_{n-1}$ of blow ups and a linear system $P_n\subset\vert D_n\vert=\vert\varepsilon_n^D_{n-1}-k_nE_n\vert$ on $S_n$ with no fixed part. On the one hand $D_n^2=D_{n-1}^2-k_n^2<D_{n-1}^2$; on the other hand $P_n$ has no fixed part so $D_n^2\geq 0$ for any $n$. Consequently the process must end. More precisely after a finite number of blow ups we obtain a system $P_n$ with no base-points which defines a morphism $\psi\colon S_n\to\mathbb{P}^m_\mathbb{C}$ as required. \[lem:preim\] Let $S$ be an irreducible surface. Let $S'$ be a smooth surface. Let $\phi\colon S\to S'$ be a birational morphism. Assume that the rational map $\phi^{-1}$ is not defined at a point $p\in S'$. Then $\phi^{-1}(p)$ is a curve on $S$. We assume that $S$ is affine so that there is an embedding $j\colon S\hookrightarrow\mathbb{A}^n_\mathbb{C}$. The rational map $$j\circ \phi^{-1}\colon S'\dashrightarrow \mathbb{A}^n_\mathbb{C}$$ is defined by rational functions $g_1$, $g_2$, $\ldots$ $g_n$. One of them, say for instance $g_1$ is undefined at $p$, that is $g_1\not\in\mathcal{O}_{S',p}$. Set $g_1=\frac{u}{v}$ with $u$, $v\in\mathcal{O}_{S',p}$, $u$ and $v$ coprime and $v(p)=0$. Consider the curve $D$ on $S$ given by $\phi^v=0$. On $S\subset\mathbb{A}^n_\mathbb{C}$ denote by $z_0$ the first coordinate function. We have $\phi^u=z_0\phi^v$ on $S$. Hence $\phi^u=\phi^v=0$ on $D$. Consequently $D=\phi^{-1}(Z)$ where $$Z=\big\{u=v=0\big\}\subset S'.$$ By assumption $u$ and $v$ are coprime so $Z$ is finite. Shrinking $S'$ if necessary we can assume that $Z=\{p\}$. Finally $D=\phi^{-1}(p)$. \[lem:preim2\] Let $S$, $S'$ be two surfaces. Let $\phi\colon S\dashrightarrow S'$ be a birational map such that $\phi^{-1}$ is not defined at $p\in S'$. Then there exists a curve $C$ on $S$ such that $\phi(C)=\{p\}$. The map $\phi$ corresponds to a morphism $\psi\colon\mathcal{U}\to S'$ for some subset $\mathcal{U}$ of $S$. Denote by $$\Gamma=\big\{\big(u,\psi(u)\big)\,\vert\,u\in\mathcal{U}\big\}\subset\mathcal{U}\times S'$$ the graph of $\psi$. Let $\overline{\Gamma}$ be the closure of $\Gamma$ in $S\times S'$; it is an irreducible surface, possibly with singularities. The projections $$\begin{aligned} &\mathrm{pr}_1\colon\overline{\Gamma}\to S, && \mathrm{pr}_2\colon\overline{\Gamma}\to S'\end{aligned}$$ are birational morphisms and the diagram $$\xymatrix{ & \overline{\Gamma}\ar[ld]_{\mathrm{pr}_1}\ar[rd]^{\mathrm{pr}_2} & \\ S\ar@{-->}[rr]_\phi & & S' }$$ is commutative. By assumption $\phi^{-1}$ is not defined at $p\in S'$, so does $\mathrm{pr}_2^{-1}$. There is an irreducible curve $C'$ on $\overline{\Gamma}$ such that $\mathrm{pr}_2(C')=\{p\}$ (Lemma \[lem:preim\]). As $\overline{\Gamma}\subset S\times S'$ the image $\mathrm{pr}_1(C')$ of $C'$ by $\mathrm{pr}_1$ is a curve $C$ in $S$ such that $\phi(C)=\{p\}$. \[pro:universalpropertyofblowingup\] Let $X$ and $S$ be two surfaces. Let $\phi\colon X\to S$ be a birational morphism of surfaces. Suppose that the rational map $\phi^{-1}$ is not defined at a point $p$ of $S$. Then $$\xymatrix{ & \mathrm{Bl}_pS\ar[rd]^\varepsilon & \\ X\ar[ur]^\psi\ar[rr]_\phi & & S }$$ where $\psi\colon X\to\mathrm{Bl}_pS$ is a birational map and $\varepsilon\colon\mathrm{Bl}_pS\to S$ is the blow up at $p$. Set $\psi=\varepsilon^{-1}\circ\phi$. Suppose that $\psi$ is not a morphism, and let $m$ be a point of $X$ such that $\psi$ is not defined at $m$. On the one hand $\phi(m)=p$ and $\phi$ is not locally invertible at $m$; on the other hand there exists a curve in $\mathrm{Bl}_pS$ blown down onto $m$ by $\psi^{-1}$ (Lemma \[lem:preim2\]). This curve has to be the exceptional divisor $E$ associated to $\varepsilon$. Let $r$ and $q$ be two distinct points of $E$ at which $\psi^{-1}$ is well defined; consider $C$, $C'$ two germs of smooth curves transverse to $E$ at $r$ and $q$ respectively. Then $\varepsilon(C)$ and $\varepsilon(C')$ are two germs of smooth curves transverse at $p$, which are images by $\phi$ of two germs of curves at $m$. The differential of $\phi$ at $m$ has thus rank $2$: contradiction with the fact that $\phi$ is not invertible at $m$. There exists a birational morphism $\varphi\colon S\to S_1$ such that the rational map $\psi=\iota\circ\varphi$ is everywhere defined (Theorem \[thm:morph\]). Furthermore $\varphi$ can be written as $$\varphi=\varepsilon_{n-1}\circ\varepsilon_{n-2}\circ\ldots\circ\varepsilon_1$$ where $\varepsilon_i\colon S_{i+1}\to S_i$, $1\leq i\leq n-1$, is obtained by blowing up a point $p_i\in S_i$ and $S=S_n$. The map $\iota$ is not defined at $p_1$, so $\psi^{-1}=\varphi^{-1}\circ\iota$ is not defined at $p_1$. Proposition \[pro:universalpropertyofblowingup\] implies that $\psi$ factors as $$\xymatrix{ & S_2\ar[rd]^{\varepsilon_1} & \\ S\ar[rr]_\psi\ar[ur]^{g_1} & & S_1 }$$ Proceeding by induction we see that $\psi$ factors as $\varphi\circ\mathcal{I}$ where $\mathcal{I}$ is a birational morphism. Since $\varphi\circ\mathcal{I}^2=\varphi$, the map $\mathcal{I}$ is an involution. In other words Lemma \[lem:linkbirinvbirinv\] says that any birational involution of a surface is birationally equivalent to a biregular involution $\iota\colon S\to S$; furthermore $(S,\iota)$ can be assumed to be minimal. Therefore the classification of conjugacy classes of involutions in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is equivalent to the classification of minimal pairs $(S,\iota)$ up to birational equivalence. Recall that the $\mathbb{P}^1_\mathbb{C}$-bundles over $\mathbb{P}^1_\mathbb{C}$ are of the form $$\mathbb{F}_n=\mathbb{P}_{\mathbb{P}^1_\mathbb{C}}\big(\mathcal{O}_{\mathbb{P}^1_\mathbb{C}}\oplus\mathcal{O}_{\mathbb{P}^1_\mathbb{C}}(n)\big)$$ for some integers $n\geq 0$ (see §\[subsec:hirz\]). For $n\geq 1$ the fibration $$f\colon \mathbb{F}_n\to\mathbb{P}^1_\mathbb{C}$$ has a unique section of self-intersection $-n$. Consider a fibre $F$ of $f$, and a point $p$ of $F$. Assume that $\iota$ is a birational involution of $\mathbb{F}_n$ regular in a neighborhood of $F$ and fixing $p$. After the elementary transformation at $p$ we get a birational involution of $\mathbb{F}_{n+1}$ regular in a neighborhood of the new fibre. The unicity assertion follows from Remark \[rem:norm\]. Using Lemma \[lem:linkbirinvbirinv\] we will prove that the involutions of Theorem \[thm:minimalpairs\] are birationally equivalent to one of Theorem \[thm:BayleBeauville\]. Cases $(5)$ and $(6)$ give by definition the <span style="font-variant:small-caps;">Geiser</span> and <span style="font-variant:small-caps;">Bertini</span> involutions. An involution of type $(4)$ is birationally equivalent to a <span style="font-variant:small-caps;">Jonquières</span> involution of degree $2$. Indeed let $Q$ be a smooth conic in $\mathbb{P}^2_\mathbb{C}$, and let $p\in\mathbb{P}^2_\mathbb{C}\smallsetminus Q$ be a point. Consider the birational involution $\iota$ of $\mathbb{P}^2_\mathbb{C}$ that maps a point $x$ to its harmonic conjugate on the line $(px)$ through $p$ and $x$ with respect to the two points of $(px)\cap Q$. This involution is not defined at the following three points: $p$ and the two points $q$ and $r$ where the tangent line to $Q$ passes through $p$. Set $S=\mathrm{Bl}_{p,q,r}\mathbb{P}^2_\mathbb{C}$. The involution $\iota$ extends to a biregular involution $\mathcal{I}$ of $S$, the <span style="font-variant:small-caps;">Jonquières</span> involution of degree $2$. In case $(3)$ take a point $p\in\mathbb{P}^2_\mathbb{C}$ such that $\iota(p)\not=p$. Let us blow up $p$, $\iota(p)$ and then blow down the proper transform of the line $(p\iota(p))$ which is a $\iota$-invariant exceptional curve. We get a pair $(T,\iota')$ with $T\simeq\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ by stereographic projection and $\mathrm{rk}\,\mathrm{Pic}(T)^{\iota'}=1$: we are thus in case $(4)$, so in the case of a <span style="font-variant:small-caps;">Jonquières</span> involution of degree $2$. Let us now deal with case $(1)$. The surface $S$ is isomorphic to $\mathbb{F}_n$ for some $n\geq 0$. The involution $\iota$ has two invariant fibres, any of them containing at least two fixed points. One of these points does not belong to $s_n$ (section of self-intersection $-n$ on $\mathbb{F}_n$), hence after a (finite) sequence of elementary transformations we get $n=1$. Let us thus focus on the case $n=1$. Let $\mathbb{F}_1$ be the surface obtained by blowing up a point $p\in\mathbb{P}^2_\mathbb{C}$. Projecting from $p$ defines a $\mathbb{P}^1$-bundle $f\colon\mathbb{F}_1\to\mathbb{P}^1_\mathbb{C}$. Any biregular involution $\iota$ of $\mathbb{F}_1$ preserves this fibration hence defines a pair $(\mathbb{F}_1,\iota)$ of case $(1)$ or $(2)$. The involution $\iota$ preserves the unique exceptional curve $E_1$ of $\mathbb{F}_1$; the pair $(\mathbb{F}_1,\iota)$ is thus not minimal: $\iota$ induces a biregular involution of $\mathbb{P}^2_\mathbb{C}$. We finally get a <span style="font-variant:small-caps;">Jonquières</span> involution of degree $2$ as we just see. We now consider case $(2)$. Let us distinguish two possibilities: denote by $F_1$, $F_2$, $\ldots$, $F_s$ the singular fibres of $f$ and by $p_i$, $1\leq i\leq s$, the singular point of $F_i$. The fixed locus of $\iota$ is a smooth curve $C$ passing through $p_1$, $p_2$, $\ldots$, $p_s$. The degree $2$ covering $C\to\mathbb{P}^1_\mathbb{C}$ induced by $f$ is ramified at $p_1$, $p_2$, $\ldots$, $p_s$. - Either $f$ is smooth, $s=0$ and $C$ is the union of two sections of $f$ which do not intersect; - or $f$ is not smooth, $C$ is a hyperelliptic curve of genus $g\geq 0$ and $s=2g+2$. First assume that we are in case $(2a)$. After elementary transformations we can suppose that $S=\mathbb{F}_1$. The fixed locus of $\iota$ is the union of $E_1$ and a section which does not meet $E_1$. Blowing down $E_1$ one gets case $(4)$. Finally let us look at case $(2b)$ for $g\geq 0$. Let us blow down one of the components in each singular fibre. We thus have a birational involution on a surface $\mathbb{F}_n$, the fixed curve $C$ embedded into $\mathbb{F}_n$. After elementary transformations at general points of $C$ one gets a birational involution on a surface $\mathbb{F}_1$, the fixed curve $C$ embedded into $\mathbb{F}_1$. The genus formula implies that $E_1\cdot C=g$. Suppose that $C$ is tangent to $E_1$ at some point $q\in\mathbb{F}_1$. After an elementary transformation at $q$ then an elementary transformation at some general point of $C$ the order of contact of $C$ and $E_1$ at $q$ decreases by $1$. Proceeding in this way we arrive at the following situation: $E_1$ and $C$ meet transversally at $g$ distinct points. Let blow down $E_1$ to a point $p$ of $\mathbb{P}^2_\mathbb{C}$; the curve $C$ maps to a plane curve $\overline{C}$ of degree $g+2$ with an ordinary multiple point of multiplicity $g$ at $p$ and no other singularity. This yields to a birational involution of $\mathbb{P}^2_\mathbb{C}$ which preserves the lines through $p$ and admits $\overline{C}$ as fixed curve, [i.e.]{} a <span style="font-variant:small-caps;">Jonquières</span> involution with center $p$ and fixed curve $\overline{C}$. Finite abelian subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group ----------------------------------------------------------------------------------------------------------- <span style="font-variant:small-caps;">Dolgachev</span> and <span style="font-variant:small-caps;">Iskovskikh</span> used a modern approach to the problem initiated in the works of <span style="font-variant:small-caps;">Manin</span> and <span style="font-variant:small-caps;">Iskovskikh</span> who gave a clear understanding of the conjugacy problem via the concept of a $\mathrm{G}$-surface ([@Manin:rational2; @Iskovskih:minimal]). Let $\mathrm{G}$ be a finite group. A $\mathrm{G}$-surface is a pair $(S,\psi)$ where $S$ is a nonsingular projective surface and $\psi$ is an isomorphism from $\mathrm{G}$ to $\mathrm{Aut}(S)$. A morphism of the pairs $(S,\psi)\to(S',\psi')$ is defined to be a morphism of surfaces $\phi\colon S\to S'$ such that $$\psi'(\mathrm{G})=\phi\circ\psi(\mathrm{G})\circ\phi^{-1}.$$ In particular let us note that two subgroups of $\mathrm{Aut}(S)$ define isomorphic $\mathrm{G}$-surfaces if and only if they are conjugate inside $\mathrm{Aut}(S)$. Let $(S,\psi)$ be a rational $\mathrm{G}$-surface. Take a birational map $\phi\colon S\dashrightarrow\mathbb{P}^2_\mathbb{C}$. For any $g\in\mathrm{G}$ the map $\phi\circ g\circ\phi^{-1}$ belongs to $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. This yields to an injective homomorphism $$\iota_\phi\colon \mathrm{G}\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ Let $(S,\psi)$ and $(S',\psi')$ be two rational $\mathrm{G}$-surfaces. Let $\phi\colon S\dashrightarrow\mathbb{P}^2_\mathbb{C}$ and $\phi'\colon S\dashrightarrow\mathbb{P}^2_\mathbb{C}$ be two birational maps. The subgroups $\iota_\phi(\mathrm{G})$ and $\iota_{\phi'}(\mathrm{G})$ are conjugate if and only if there exists a birational map of $\mathrm{G}$-surfaces $S'\dashrightarrow S$. In other words a birational isomorphism class of $\mathrm{G}$-surfaces defines a conjugacy class of subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ isomorphic to $\mathrm{G}$. The following result shows that any conjugacy class is obtained in this way: Let $\mathrm{G}$ be a finite subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. There exist a rational $\mathrm{G}$-surface $(S,\psi)$ and a birational map $\phi\colon S\dashrightarrow\mathbb{P}^2_\mathbb{C}$ such that $$\mathrm{G}=\phi\circ\psi(\mathrm{G})\circ\phi^{-1}.$$ If $\phi$ belongs to $\mathrm{G}$, we denote by $\mathrm{dom}(\phi)$ an open subset on which $\phi$ is defined. Set $\mathcal{D}=\displaystyle\bigcap_{\phi\in\mathrm{G}}\mathrm{dom}(\phi)$. Then $\mathcal{U}=\displaystyle\bigcap_{\phi\in\mathrm{G}}g(\mathcal{D})$ is an open invariant subset of $\mathbb{P}^2_\mathbb{C}$ on which $\phi$ acts biregularly. Consider $\mathcal{U}'=\faktor{\mathcal{U}}{\mathrm{G}}$ the orbit space; it is a normal algebraic surface. Let us choose any normal projective completion $X'$ of $\mathcal{U}'$. Consider $S'$ the normalization of $X'$ in the field of rational functions of $\mathcal{U}$. It is a normal projective surface on which $\mathrm{G}$ acts by biregular transformations. A $\mathrm{G}$-invariant resolution of singularities $S$ of $S'$ suits ([@DeFernexEin]). Hence one has: There is a natural bijective correspondence between birational isomorphism classes of rational $\mathrm{G}$-surfaces and conjugate classes of subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ isomorphic to $\mathrm{G}$. Therefore the goal of <span style="font-variant:small-caps;">Dolgachev</span> and <span style="font-variant:small-caps;">Iskovskikh</span> is to classify $\mathrm{G}$-surfaces up to birational isomorphism of $\mathrm{G}$-surfaces. There is a $\mathrm{G}$-equivariant analogue of minimal surfaces: A minimal $\mathrm{G}$-surface is a $\mathrm{G}$-surface $(S,\psi)$ such that any birational morphism of $\mathrm{G}$-surfaces $(S,\psi)\to(S',\psi')$ is an isomorphism. Note that it is enough to classify minimal rational $\mathrm{G}$-surfaces up to birational isomorphism of $\mathrm{G}$-surfaces. The authors can rely on the following fundamental result: \[thm:classificationminimalGsurfaces\] Let $S$ be a minimal rational $\mathrm{G}$-surface. Then - either $S$ admits a structure of a conic bundle with $\mathrm{Pic}(S)^{\mathrm{G}}\simeq\mathbb{Z}^2$; - or $S$ is isomorphic to a <span style="font-variant:small-caps;">del Pezzo</span> surface with $\mathrm{Pic}(S)^{\mathrm{G}}\simeq\mathbb{Z}$. An analogous result from the classical literature is showed by using the method of the termination of adjoints, first introduced for linear system of plane curves in the work of <span style="font-variant:small-caps;">Castelnuovo</span>. This method is applied to find a $\mathrm{G}$-invariant linear system of curves in the plane in [@Kantor]; <span style="font-variant:small-caps;">Kantor</span> essentially stated the result above but without the concept of minimality. A first modern proof can be found in [@Manin:rational2] and [@Iskovskih:minimal]. Nowadays Theorem \[thm:classificationminimalGsurfaces\] follows from a $\mathrm{G}$-equivariant version of <span style="font-variant:small-caps;">Mori</span> theory ([@deFernex]). As a result to complete the classification <span style="font-variant:small-caps;">Dolgachev</span> and <span style="font-variant:small-caps;">Iskovskikh</span> need - to classify all finite groups $\mathrm{G}$ that may occur in a minimal $\mathrm{G}$-pair; - to determine when two minimal $\mathrm{G}$-surfaces are birationally isomorphic. To achieve $(i)$ the authors computed the full automorphisms group of a conic bundle surface on a <span style="font-variant:small-caps;">del Pezzo</span> surface and then made a list of all finite subgroups acting minimally on the surface. To achieve $(ii)$ the authors used the ideas of <span style="font-variant:small-caps;">Mori</span> theory to decompose a birational map of rational $\mathrm{G}$-surfaces into elementary links. Finite cyclic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ ------------------------------------------------------------------ In [@Blanc:CRAS] the author gave the list of finite cyclic subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group, up to conjugation. The curves fixed by one element of the group, and the action of the whole group on these curves, are often sufficient to distinguish the conjugacy classes. It was done in [@Blanc:these] in many cases, but some remain unsolved. In [@Blanc:these] the author completed this classification with the case of abelian non-cyclic groups. Its classification implies several results we will now mention. For any integer $n\geq 1$ there are infinitely many conjugacy classes of elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of order $2n$, that are non-conjugate to a linear automorphism. If $n>15$, a birational map of $\mathbb{P}^2_\mathbb{C}$ of order $2n$ is a $n$-th root of a <span style="font-variant:small-caps;">Jonquières</span> involution and preserves a pencil of rational curves. If an element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is of finite odd order and is not conjugate to a linear automorphism of $\mathbb{P}^2_\mathbb{C}$, then its order is $3$, $5$, $9$ or $15$. In particular any birational map of $\mathbb{P}^2_\mathbb{C}$ of odd order $>15$ is conjugate to a linear automorphism of the plane. Then <span style="font-variant:small-caps;">Blanc</span> generalized a theorem of <span style="font-variant:small-caps;">Castelnuovo</span> which states that an element of finite order which fixes a curve of geometric genus $>1$ has order $2$, $3$ or $4$ (see [@Castelnuovo]): Let $\mathrm{G}$ be a finite abelian group which fixes some curve of positive geometric genus. Then $\mathrm{G}$ is cyclic, of order $2$, $3$, $4$, $5$ or $6$, and all these cases occur. If the curve has geometric genus $>1$, then $\mathrm{G}$ is of order $2$ or $3$. Let $\mathrm{G}$ be a finite abelian subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. The following assertions are equivalent: - any $g\in\mathrm{G}\smallsetminus\big\{\mathrm{id}\big\}$ does not fix a curve of positive geometric genus; - the group $\mathrm{G}$ is birationally conjugate to a subgroup of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, or to a subgroup of $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$, or to the group isomorphic to $\faktor{\mathbb{Z}}{2\mathbb{Z}}\times\faktor{\mathbb{Z}}{4\mathbb{Z}}$ generated by the two following elements $$\begin{aligned} & (z_0:z_1:z_2)\mapsto\big(z_1z_2:z_0z_1:-z_0z_2\big), && (z_0:z_1:z_2)\mapsto\big(z_1z_2(z_1-z_2):z_0z_2(z_1+z_2):z_0z_1(z_1+z_2)\big).\end{aligned}$$ Furthermore this last group is conjugate neither to a subgroup of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, nor to a subgroup of $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$. In [@Beauville] <span style="font-variant:small-caps;">Beauville</span> gave the isomorphism classes of $p$-elementary subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. <span style="font-variant:small-caps;">Blanc</span> generalized it as follows: The isomorphism classes of finite abelian subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group are the following: - $\faktor{\mathbb{Z}}{m\mathbb{Z}}\times\faktor{\mathbb{Z}}{n\mathbb{Z}}$ for any integers $m$, $n\geq 1$, - $\faktor{\mathbb{Z}}{2n\mathbb{Z}}\times(\faktor{\mathbb{Z}}{2\mathbb{Z}})^2$ for any integer $n\geq 1$, - $\big(\faktor{\mathbb{Z}}{4\mathbb{Z}}\big)^2\times\faktor{\mathbb{Z}}{2\mathbb{Z}}$, - $(\faktor{\mathbb{Z}}{3\mathbb{Z}})^3$, - $(\faktor{\mathbb{Z}}{2\mathbb{Z}})^4$. In [@Blanc:cyclic] the author finished the classification of cyclic subgroups of finite order of the <span style="font-variant:small-caps;">Cremona</span> group, up to conjugation. He gave natural parameterizations of conjugacy classes, related to fixed curves of positive genus. The classification of finite cyclic subgroups that are not of <span style="font-variant:small-caps;">Jonquières</span> type was almost achieved in [@DolgachevIskovskikh]. Let us explain what we mean by “almost”: - a list of representative elements is available; - explicit forms are given; - the dimension of the varieties which parameterize the conjugacy classes are provided. What is missing ? A finer geometric description of the algebraic variety parameterizing conjugacy classes according to [@DolgachevIskovskikh]. The case of groups conjugate to subgroups of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ was studied in [@BeauvilleBlanc]: there is exactly one conjugacy class for each order $n$, representated by $$\langle(z_0:z_1:z_2)\mapsto(z_0:z_1:\mathrm{e}^{2\mathbf{i}\pi/n}z_2)\rangle.$$ <span style="font-variant:small-caps;">Blanc</span> completed the classification of cyclic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of finite order ([@Blanc:cyclic]). For groups of <span style="font-variant:small-caps;">Jonquières</span> type he applied cohomology group theory and algebraic tools to the group $\mathcal{J}$ and got: - For any positive integer $m$, there exists a unique conjugacy class of linearisable elements of order $n$, represented by the automorphism $$(z_0:z_1:z_2)\mapsto(z_0:z_1:\mathrm{e}^{2\mathbf{i}\pi/n}z_2).$$ - Any non-linearisable <span style="font-variant:small-caps;">Jonquières</span> element of finite order of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ has order $2n$, for some positive integer $n$, and is conjugate to an element $\phi$, such that $\phi$ and $\phi^n$ are of the following form $$\phi\colon(z_0,z_1)\dashrightarrow\left(\mathrm{e}^{2\mathbf{i}\pi/n}z_0,\frac{a(z_0)z_1+(-1)^\delta p(z_0^n)b(z_0)}{b(z_0)z_1+(-1)^\delta a(z_0)}\right)$$ $$\phi^n\colon(z_0,z_1)\dashrightarrow\left(z_0,\frac{p(z_0^n)}{z_1}\right)$$ where $a$, $b$ belongs to $\mathbb{C}(z_0)$, $\delta$ to $\big\{0,\pm 1\big\}$, and $p\in\mathbb{C}[z_0]$ is a polynomial with simple roots. The curve $\Gamma$ of equation $z_1^2=p(z_0^n)$, pointwise fixed by $\phi^n$, is hyperelliptic, of positive geometric genus, and admits a $(2:1)$-map $\phi_1^2\colon\Gamma\to\mathbb{P}^1_\mathbb{C}$. The action of $\phi$ on $\Gamma$ has order $n$, and is not a root of the involution associated to any $\phi_1^2$. Furthermore the above association yields a parameterization of the conjugacy classes of non-linearisable <span style="font-variant:small-caps;">Jonquières</span> elements of order $2n$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ by isomorphism classes of pairs $(\Gamma,\psi)$, where - $\Gamma$ is a smooth hyperelliptic curve of positive genus, - $\psi\in\mathrm{Aut}(\Gamma)$ is an automorphism of order $n$, which preserves the fibres of the $\phi_1^2$ and is not a root of the involution associated to the $\phi_1^2$. The analogous result for finite <span style="font-variant:small-caps;">Jonquières</span> cyclic groups holds, and follows directly from this statement. Note that if the curve $\Gamma$ has geometric genus $\geq 2$, the $\phi_1^2$ is unique, otherwise it is not. <span style="font-variant:small-caps;">Blanc</span> also dealt with cyclic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that are not of <span style="font-variant:small-caps;">Jonquières</span> type. Using the classification of [@DolgachevIskovskikh] and some classical tools on surfaces and curves he provided the parameterization of the $29$ families of such groups. The classification is divided in two parts: - find representative families and prove that each group is conjugate to one of these; - parameterize the conjugacy classes in each families by algebraic varieties. For cyclic groups of prime order the varieties parameterizing the conjugacy classes are the moduli spaces of the non-rational curves fixed by the groups. <span style="font-variant:small-caps;">Blanc</span> needs to generalize it, by looking for the non-rational curves fixed by the non-trivial elements of the group. Let us give the definition of this invariant which provides a simple way to decide whether two cyclic groups are conjugate. Recall that a birational map of the complex projective plane fixes a curve if it restricts to the identity on the curve. Let $\phi$ be a non-trivial element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of finite order. If no curve of positive geometric genus is (pointwise) fixed by $\phi$, then $\mathrm{NFC}(\phi)=\emptyset$; otherwise $\phi$ fixes exactly one curve of positive genus ([@BayleBeauville; @deFernex]), and $\mathrm{NFC}(\phi)$ is then the isomorphism class of the normalization of this curve. Two involutions $\phi$, $\psi$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ are conjugate if and only if $\mathrm{NFC}(\phi)=\mathrm{NFC}(\psi)$ (see §\[sec:ordre2\]). If $\phi$, $\psi$ are elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of the same prime order, then $\langle\phi\rangle$ and $\langle\psi\rangle$ are conjugate if and only if $\mathrm{NFC}(\phi)=\mathrm{NFC}(\psi)$ (see [@BeauvilleBlanc; @deFernex]). This is no longer the case for cyclic groups of composite order as observed in [@BeauvilleBlanc]: the automorphism $\phi$ of the cubic surface $z_0^3+z_1^3+z_2^3+z_3^3=0$ in $\mathbb{P}^3_\mathbb{C}$ given by $$\phi\colon(z_0:z_1:z_2:z_3)\mapsto\big(z_1:z_0:z_2:\zeta z_3\big)$$ where $\zeta^3=1$, $\zeta\not=1$ has only four fixed points while $\phi^2$ fixes the elliptic curve $z_3=0$. Let $\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ be a non-trivial element of finite order $n$. Then $\mathrm{NFCA}(\phi)$ is the sequence of isomorphism classes of pairs $$\Big(\mathrm{NFC}(\phi^k),\phi_{\vert\mathrm{NFC}(\phi^k)}\Big)_{k=1}^{n-1}$$ where $\phi_{\vert\mathrm{NFC}(\phi^k)}$ is the automorphism induced by $\phi$ on the curve $\mathrm{NFC}(\phi^k)$ (if $\mathrm{NFC}(\phi^k)=\emptyset$, then $\phi$ acts trivially on it). Let us now give a simple way to decide whether two cyclic subgroups of finite order of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ are conjugate: Let $\mathrm{G}$ and $\mathrm{H}$ be two cyclic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of the same finite order. Then $\mathrm{G}$ and $\mathrm{H}$ are conjugate in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ if and only if $\mathrm{NFCA}(\phi)=\mathrm{NFCA}(\psi)$ for some generators $\phi$ of $\mathrm{G}$ and $\psi$ of $\mathrm{H}$. Uncountable subgroups {#chapter:uncountable} ===================== All the results of this Chapter have been proved without the construction of the action of the isometric action of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ on the hyperbolic space $\mathbb{H}^\infty$ and we keep this point of view. Different ideas and tools are used in any section: foliations and group theory are the main ingredients. The study of the automorphis groups starts a long time ago. For instance for classical groups let us see [@Dieudonne]. Consider the automorphism group of the complex projective space $\mathbb{P}^n_\mathbb{C}$; it is $\mathrm{PGL}(n+1,\mathbb{C})$. The automorphism group of $\mathrm{PGL}(n+1,\mathbb{C})$ is generated by the inner automorphisms, the involution $M\mapsto M^\vee$ and the action of the field automorphisms of $\mathbb{C}$. In $1963$ <span style="font-variant:small-caps;">Whittaker</span> showed that any isomorphism between homeomorphism groups of connex topological varieties is induced by an homeomorphism between the varieties themselves ([@Whittaker]). In $1982$ <span style="font-variant:small-caps;">Filipkiewicz</span> proved a similar statement for differentiable varieties. \[thm:Filipkiewicz\] Let $V$, $W$ be two connected varieties of class $\mathcal{C}^k$, resp. $\mathcal{C}^j$. Let $\mathrm{Diff}^k(V)$ be the group of $\mathcal{C}^k$-diffeomorphisms of $V$. Let $\phi\colon\mathrm{Diff}^k(V)\to\mathrm{Diff}^j(V)$ be an isomorphism group. Then $k=j$ and there exists a $\mathcal{C}^k$-difffeomorphism $\psi\colon V\to W$ such that $$\phi(\varphi)=\psi\circ\varphi\circ\psi^{-1}\qquad\qquad\forall\,\varphi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ The description of uncountable maximal abelian subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group allows to characterize the automorphisms group of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$: \[thm:autofbir\] Let $\varphi$ be an automorphism of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. There exist a birational self map $\psi$ of the complex projective plane and an automorphism $\kappa$ of the field $\mathbb{C}$ such that $$\varphi(\phi)={}^{\kappa}\!\,(\psi\circ\phi\circ\psi^{-1})\qquad\qquad\forall\,\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ In other words the non-inner automorphism group of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ can be identified with the automorphisms of the field $\mathbb{C}$. In the first section we study uncountable maximal abelian subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$; let $\mathrm{G}$ be such a group. We give an outline of the proofs of the following results: - any element of $\mathrm{G}$ preserves at least one singular holomorphic foliation; - either no element of $\mathrm{G}$ is torsion-free, or $\mathrm{G}$ leaves invariant a holomorphic foliation; - if $\mathrm{G}$ is torsion-free, then $\mathrm{G}$ is conjugate to a subgroup of the <span style="font-variant:small-caps;">Jonquières</span> group. In the second section we describe the automorphism group of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. A study of the torsion-free maximal abelian subgroups of the <span style="font-variant:small-caps;">Jonquières</span> group shows that the group $$\mathcal{J}_a=\big\{(z_0,z_1)\dashrightarrow(z_0+a(z_1),z_1)\,\vert\,a\in\mathbb{C}(z_1)\big\}$$ is invariant by any automorphism of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Some work on special subgroups of $\mathcal{J}_a$ achieves the description of $\mathrm{Aut}(\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}))$. In a session problems during the International Congress of Mathematicians <span style="font-variant:small-caps;">Mumford</span> proposed the following ([@Mumford]): > “Let $\mathrm{G}=\mathrm{Aut}_{\mathbb{C}}\mathbb{C}(z_0,z_1)$ be the <span style="font-variant:small-caps;">Cremona</span> group (...) the problem is to topologize $\mathrm{G}$ and associate to it a <span style="font-variant:small-caps;">Lie</span> algebra consisting, roughly, of those meromorphic vector fields $D$ on $\mathbb{P}^2_\mathbb{C}$ which ”integrate“ into an analytic family of <span style="font-variant:small-caps;">Cremona</span> transformations.” In the third section we deal with a contribution in that direction: the description of $1$-parameter subgroups of quadratic birational self maps of $\mathbb{P}^2_\mathbb{C}$. In [@Ghys] <span style="font-variant:small-caps;">Ghys</span> showed that any nilpotent subgroup of $\mathrm{Diff}^\omega(\mathbb{S}^2)$ is metabelian; as a consequence he got that if $\Gamma$ is a subgroup of finite index of $\mathrm{SL}(n,\mathbb{Z})$, $n\geq 4$, then any morphism from $\Gamma$ into $\mathrm{Diff}^\omega(\mathbb{S}^2)$ has finite image. In the same spirit the nilpotent subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group are described in the fourth section: if $\Gamma$ is a strongly nilpotent group of length $>1$, then either $\mathrm{G}$ is metabelian up to finite index, or $\mathrm{G}$ is a torsion group. As a consequence as soon as $n\geq 5$ no subgroup of $\mathrm{SL}(n,\mathbb{Z})$ of finite index embeds into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. The description of centralizers of discrete dynamical systems is an important problem in real/complex dynamics. <span style="font-variant:small-caps;">Julia</span> ([@Julia1; @Julia2]) then <span style="font-variant:small-caps;">Ritt</span> ([@Ritt]) show that the set $$\mathrm{Cent}(\phi)=\big\{\psi\colon\mathbb{P}^1_\mathbb{C}\to\mathbb{P}^1_\mathbb{C}\,\vert\,\psi\circ \phi=\phi\circ \psi\big\}$$ of rational functions that commute to a rational function $\phi$ coincide in general [^5] with $\big\{\phi_0^n\,\vert\,n\in\mathbb{N}\big\}$ where $\phi_0$ is an element of $\mathrm{Cent}(\phi)$. In the $60$’s <span style="font-variant:small-caps;">Smale</span> considered generic diffeomorphisms $\phi$ of compact manifolds and asked if its centralizer coincides with $\big\{\phi^n\,\vert\,n\in\mathbb{Z}\big\}$. Many mathematicians have considered this question (for instance [@BonattiCrovisierWilkinson; @Palis; @PalisYoccoz; @PalisYoccoz2]). The fifth section deals with centralizers of elliptic birational maps, <span style="font-variant:small-caps;">Jonquières</span> twists and <span style="font-variant:small-caps;">Halphen</span> twists. Uncountable maximal abelian subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ {#sec:uncountableabelian} -------------------------------------------------------------------------------- Let $S$ be a complex compact surface. A foliation $\mathcal{F}$ on $S$ is given by a family $(\chi_i)_i$ of holomorphic vector fields with isolated zero defined on some open cover $(\mathcal{U}_i)_i$ of $S$. The vector fields $\chi_i$ have to satisfy the following conditions: there exist $g_{ij}\in\mathcal{O}^(\mathcal{U}_i\cap\mathcal{U}_j)$ such that $\chi_i=g_{ij}\chi_j$ on $\mathcal{U}_i\cap\mathcal{U}_j$. Let us remark that a non-trivial meromorphic vector field on $S$ defines such a foliation. Let $\mathrm{G}$ be an uncountable abelian subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. There exists a rational vector field $\chi$ such that $$\varphi_\chi=\chi\qquad\qquad\forall\,\varphi\in\mathrm{G}.$$ In particular $\mathrm{G}$ preserves a foliation. Since $\mathrm{G}$ is uncountable, there exists an integer $d$ such that $$\mathrm{G}_d=\mathrm{G}\cap\mathrm{Bir}_d(\mathbb{P}^2_\mathbb{C})$$ is uncountable. Hence the <span style="font-variant:small-caps;">Zariski</span> closure $\overline{\mathrm{G}_d}$ of $\mathrm{G}_d$ in $\mathrm{Bir}_{\leq d}(\mathbb{P}^2_\mathbb{C})$ is an algebraic set of dimension $\geq 1$. Consider a curve in $\overline{\mathrm{G}_d}$, [i.e.]{} a map $$\eta\colon\mathbb{D}\to\overline{\mathrm{G}_d},\qquad\qquad t\mapsto\eta(t).$$ Remark that elements of $\overline{\mathrm{G}_d}$ are rational maps that commute. Let us define the rational vector field $\chi$ at any $m\in\mathbb{P}^2_\mathbb{C}\smallsetminus\mathrm{Ind}(\eta(0)^{-1})$ by $$\chi(m)=\frac{\partial\eta(s)}{\partial s}\Big\vert_{s=0}\big(\eta(0)^{-1}(m)\big).$$ Let $\varphi$ be an element of $\overline{\mathrm{G}_d}$. If we differentiate the equality $$\varphi\eta(s)\varphi^{-1}(m)=\eta(s)(m)$$ with respect to $s$, $m$ being fixed, one gets: $\varphi_\chi=\chi$. In other words $\chi$ is invariant by the elements of $\overline{\mathrm{G}_d}$, and so by any element of $\mathrm{G}$. As a result for any uncountable abelian subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, there exists a foliation on $\mathbb{P}^2_\mathbb{C}$ invariant by $\mathrm{G}$. <span style="font-variant:small-caps;">Brunella</span>, <span style="font-variant:small-caps;">McQuillan</span> and <span style="font-variant:small-caps;">Mendes</span> have classified, up to birational equivalence, singular holomorphic foliations on projective, compact, complex surfaces ([@Brunella; @McQuillan; @Mendes]). If $S$ is a projective surface endowed with a foliation $\mathcal{F}$, we denote by $\mathrm{Bir}(S,\mathcal{F})$ (resp. $\mathrm{Aut}(S,\mathcal{F})$) the group of birational maps (resp. holomorphic maps) of $S$ preserving the foliation $\mathcal{F}$. In general $\mathrm{Bir}(S,\mathcal{F})$ coincides with $\mathrm{Aut}(S,\mathcal{F})$ and is finite. In [@CantatFavre] the authors dealt with the opposite case and got a classification. \[thm:CantatFavre1\] Let $\mathcal{F}$ be a foliation on $S$ such that $\mathrm{Aut}(X,\varphi^\mathcal{F})\subsetneq\mathrm{Bir}(X,\varphi^\mathcal{F})$ for any birational map $\varphi\colon X\dashrightarrow S$. Then, up to conjugacy, there exists an element of infinite order in $\mathrm{Bir}(S,\mathcal{F})$ and - either $\mathcal{F}$ is a rational fibration, - or up to a finite cover there exist some integers $p$, $q$, $r$, $s$ such that $$\mathrm{Bir}(\mathbb{P}^2_\mathbb{C},\mathcal{F})=\big\{(z_0,z_1)\dashrightarrow(z_0^pz_1^q,z_0^rz_1^s),\,(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)\,\vert\,\alpha,\,\beta\in\mathbb{C}^\big\}.$$ Before stating the opposite case $\mathrm{Aut}(S,\mathcal{F})$ infinite, let us give some definitions. Let $\Lambda$ be a lattice in $\mathbb{C}^2$; it induces a complex torus $\mathbb{T}=\faktor{\mathbb{C}^2}{\Lambda}$ of dimension $2$. For instance the product of an elliptic curve by itself is a complex torus. An affine map $\psi$ that preserves $\Lambda$ induces an automorphism of the torus $\mathbb{T}$. If the linear part of $\psi$ is of infinite order, then - either the linear part of $\psi$ is hyperbolic and $\psi$ induces an <span style="font-variant:small-caps;">Anosov</span> automorphism that preserves two linear foliations; - or the linear part of $\psi$ is unipotent and $\psi$ preserves an elliptic fibration. Sometimes there is a finite automorphism group of $\mathbb{T}$ normalized by $\psi$. Denote by $\widetilde{\faktor{\mathbb{T}}{\mathrm{G}}}$ the desingularization of $\faktor{\mathbb{T}}{\mathrm{G}}$. The automorphism induced by $\psi$ on $\widetilde{\faktor{\mathbb{T}}{\mathrm{G}}}$ preserves - the foliations induced the stable and unstable foliations preserved by $\psi$ when $\psi$ is hyperbolic; - an elliptic fibration when the linear part of $\psi$ is unipotent. If $\mathrm{G}=\big\{\mathrm{id},\,(z_0,z_1)\mapsto(-z_0,-z_1)\big\}$ we say that $\widetilde{\faktor{\mathbb{T}}{\mathrm{G}}}$ is a <span style="font-variant:small-caps;">Kummer</span> surface; otherwise $\widetilde{\faktor{\mathbb{T}}{\mathrm{G}}}$ is a generalized <span style="font-variant:small-caps;">Kummer</span> surface. \[thm:CantatFavre2\] Let $\mathcal{F}$ be a singular holomorphic foliation on a projective surface $S$. Assume that $\mathrm{Aut}(S,\mathcal{F})$ is infinite. Then $\mathrm{Aut}(S,\mathcal{F})$ contains at least one element $\varphi$ of infinite order and one of the following holds: - $\mathcal{F}$ is invariant by an holomorphic vector field; - $\mathcal{F}$ is an elliptic fibration; - the surface $S$ is a generalized <span style="font-variant:small-caps;">Kummer</span> surface, $\varphi$ can be lifted to an <span style="font-variant:small-caps;">Anosov</span> automorphism $\widetilde{\varphi}$ of the torus and $\mathcal{F}$ is the projection on $S$ of the unstable or stable foliation of $\widetilde{\varphi}$. <!-- --> - The foliations invariant by an holomorphic vector field are described in [@CantatFavre Proposition 3.8]. - The last two cases are mutually exclusive. Using these two statements one can prove the following one: \[thm:abmax\] Let $\mathrm{G}$ be an uncountable maximal abelian subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Then: - either $\mathrm{G}$ has an element of finite order; - or $\mathrm{G}$ is up to conjugacy a subgroup of the <span style="font-variant:small-caps;">Jonquières</span> group. Assume first that $\mathrm{Aut}(X,\phi^\mathcal{F})\subsetneq\mathrm{Bir}(X,\phi^\mathcal{F})$ for any birational map $\phi\colon X\dashrightarrow S$. Then according to Theorem \[thm:CantatFavre1\] either $\mathrm{G}$ preserves a rational fibration, and then $\mathrm{G}$ is up to birational conjugacy contained in the <span style="font-variant:small-caps;">Jonquières</span> group; or $\mathrm{G}$ is up to conjugacy and finite cover a subgroup of $$\big\{(z_0,z_1)\dashrightarrow(z_0^pz_1^q,z_0^rz_1^s),\,(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)\,\vert\,\alpha,\,\beta\in\mathbb{C}^,\,\alpha=\alpha^p\beta^q,\,\beta=\alpha^r\beta^s\big\}.$$ If $\mathrm{G}$ is conjugate to the diagonal group $\mathrm{D}=\big\{(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)\,\vert\,\alpha,\,\beta\in\mathbb{C}^\big\}$, then $\mathrm{G}$ contains elements of finite order. Otherwise since $\mathrm{G}$ is uncountable it can not be reduced to $$\langle(z_0,z_1)\dashrightarrow(z_0^pz_1^q,z_0^rz_1^s)\rangle.$$ Therefore there exists a non-trivial element $(z_0,z_1)\mapsto(\lambda z_0,\mu z_1)$ in $\mathrm{G}$ such that $\lambda=\lambda^p\mu^q$ and $\mu=\lambda^r\mu^s$. For any $\ell$ the map $(z_0,z_1)\mapsto(\lambda^\ell z_0,\mu^\ell z_1)$ satisfies these equalities so belongs to $\mathrm{G}$. Consider $\ell$ such that $\lambda^\ell=\mathbf{i}$; then $\mu^\ell=\mathrm{e}^{\mathbf{i}\pi \frac{1-p}{2q}}$ is also a root of unity and $(z_0,z_1)\mapsto(\lambda^\ell z_0,\mu^\ell z_1)$ is thus an element of finite order of $\mathrm{G}$. More precisely $\mathrm{G}$ contains periodic elements of any order. Suppose now that there exist a surface $S$ and a birational map $\psi\colon S\dashrightarrow\mathbb{P}^2_\mathbb{C}$ such that $\mathrm{Aut}(S,\psi^\mathcal{F})=\mathrm{Bir}(S,\psi^\mathcal{F})$. According to Theorem \[thm:CantatFavre2\] - either $\psi^\mathcal{F}$ is invariant by an holomorphic vector field on $S$; - or $\psi^\mathcal{F}$ is an elliptic fibration. Since $\mathrm{G}$ is uncountable the last eventuality can not occur ([@BHPV]). Let us thus assume that $\psi^\mathcal{F}$ is invariant by an holomorphic vector field on $S$. According to [@CantatFavre] one can assume up to conjugacy that $\mathrm{G}$ is a subgroup of $\mathrm{Aut}(\widetilde{S})$ where $\widetilde{S}$ is a minimal model of $S$. But minimal rational surfaces are $\mathbb{P}^2_\mathbb{C}$, $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ and the <span style="font-variant:small-caps;">Hirzebruch</span> surfaces $\mathbb{F}_n$, $n\geq 2$, and their automorphisms groups are known (see* Chapter \[Chapter:algebraicsubgroup\]). The description of the uncountable maximal abelian subgroups of minimal rational surfaces gives: Let $S$ be a minimal rational surface. Let $\mathrm{G}$ be an uncountable abelian subgroup of $\mathrm{Aut}(S)$ maximal in $\mathrm{Bir}(S)$. Then: - either $\mathrm{G}$ contains an element of finite order, - or $\mathrm{G}$ coincides with $\big\{(z_0,z_1)\mapsto(z_0+P(z_1),z_1)\,\vert\, P\in\mathbb{C}[z_1],\,\deg P\leq n\big\}$, - or $\mathrm{G}=\big\{(z_0,z_1)\mapsto(z_0+\alpha,z_1+\beta)\,\vert\,\alpha,\,\beta\in\mathbb{C}\big\}$. A study of the uncountable maximal abelian subgroups of the <span style="font-variant:small-caps;">Jonquières</span> group allows to refine Theorem \[thm:abmax\] as follows: Let $\mathrm{G}$ be an uncountable maximal abelian subgroup of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. Then up to conjugacy: - either $\mathrm{G}$ contains an element of finite order, - or $\mathrm{G}=\big\{(z_0,z_1)\dashrightarrow(z_0+a(z_1),z_1)\,\vert\,a\in\mathbb{C}(z_1)\big\}$, - or $\mathrm{G}=\big\{(z_0,z_1)\mapsto(z_0+\alpha,z_1+\beta)\,\vert\,\alpha,\,\beta\in\mathbb{C}\big\}$, - or any subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ acting by conjugacy on $\mathrm{G}$ is, up to finite index, solvable. Description of the automorphisms group of the <span style="font-variant:small-caps;">Cremona</span> group {#sec:autbir} --------------------------------------------------------------------------------------------------------- Let us give an idea of the proof of Theorem \[thm:autofbir\]. The description of uncountable maximal abelian subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ yields to Let $\varphi$ be an automorphism of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Set $$\mathrm{J}_a=\big\{(z_0,z_1)\dashrightarrow(z_0+a(z_1),z_1)\,\vert\,a\in\mathbb{C}(z_1)\big\}.$$ Up to birational conjugacy $\varphi(\mathrm{J}_a)=\mathrm{J}_a$ and $(z_0,z_1)\mapsto(z_0+1,z_1)$ is invariant by $\varphi$. Let us consider $$\begin{aligned} &\mathrm{T}_1=\big\{(z_0,z_1)\mapsto(z_0+\alpha,z_1)\,\vert\,\alpha\in\mathbb{C}\big\}, && \mathrm{T}_2=\big\{(z_0,z_1)\mapsto(z_0,z_1+\beta)\,\vert\,\beta\in\mathbb{C}\big\},\end{aligned}$$ and $$\begin{aligned} & \mathrm{D}_1=\big\{(z_0,z_1)\mapsto(\alpha z_0,z_1)\,\vert\,\alpha\in\mathbb{C}^\big\}, &&\mathrm{D}_2=\big\{(z_0,z_1)\mapsto(z_0,\beta z_1)\,\vert\,\alpha\in\mathbb{C}\big\}.\end{aligned}$$ \[pro:autbir\] Let $\varphi$ be an automorphism of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Assume that $\varphi(\mathrm{J}_a)=\mathrm{J}_a$ and $(z_0,z_1)\mapsto(z_0+1,z_1)$ is invariant by $\varphi$. Then up to birational conjugacy: - $\varphi(\mathrm{J}_a)=\mathrm{J}_a$, - $(z_0,z_1)\mapsto(z_0+1,z_1)$ is invariant by $\varphi$, - $\varphi(\mathrm{T}_1)=\mathrm{T}_1$ and $\varphi(\mathrm{T}_2)=\mathrm{T}_2$, - $\varphi(\mathrm{D}_1)=\mathrm{D}_1$ and $\varphi(\mathrm{D}_2)=\mathrm{D}_2$. As a consequence an automorphism of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ induces two automorphisms of the group $\mathrm{Aff}(\mathbb{C})$ of affine maps of the complex line. \[lem:autbir\] Let $\varphi$ be an automorphism of $\mathrm{Aff}(\mathbb{C})$. Then $\varphi$ is the composition of an inner automorphism and an automorphism of the field $\mathbb{C}$. The maximal abelian subgroups of $\mathrm{Aff}(\mathbb{C})$ are the group of translations $$\mathrm{T}=\big\{z\mapsto z+\beta\,\vert\,\beta\in\mathbb{C}\big\}$$ and the groups of affine maps that preserve a point $$\mathrm{D}_{z_0}=\big\{z\mapsto \alpha(z-z_0)+z_0\,\vert\alpha\in\mathbb{C}^\big\}.$$ Since $\mathrm{T}$ does not contain element of finite order, $\varphi$ sends $\mathrm{T}$ onto $\mathrm{T}$. In other words there exists an additive bijection $\kappa_2\colon\mathbb{C}\to\mathbb{C}$ such that $\varphi(z+\beta)=z+\kappa_2(\beta)$. Up to conjugacy by an element of $\mathrm{T}$ one can assume that $\varphi(\mathrm{D}_0)=\mathrm{D}_0$. In other words there exists a multiplicative bijection $\kappa_1\colon\mathbb{C}^\to\mathbb{C}^$ such that $\varphi(\alpha z)=\kappa_1(\alpha)z$. On the one hand $$\varphi\big(z\mapsto\alpha z+\alpha\big)=\varphi\big(z\mapsto z+\alpha\big)\circ\varphi\big(z\mapsto \alpha z\big)=\big(z\mapsto\kappa_1(\alpha)z+\kappa_2(\alpha)\big)$$ and on the other hand $$\varphi\big(z\mapsto \alpha z+\alpha\big)=\varphi\big(z\mapsto \alpha z\big)\circ\varphi\big(z\mapsto z+1\big)=\big(z\mapsto\kappa_1(\alpha)z+\kappa_1(\alpha)\kappa_2(1)\big).$$ Hence for any $\alpha$ the equality $z\mapsto\kappa_1(\alpha)z+\kappa_2(\alpha)=z\mapsto\kappa_1(\alpha)z+\kappa_1(\alpha)\kappa_2(1)$ holds. Since $\mu=\kappa_2(1)$ is non-zero, $\kappa_2$ is additive and multiplicative. As a result $\kappa_2$ is an isomorphism of the field $\mathbb{C}$ and $$\begin{aligned} \varphi\big(z\mapsto \alpha z+\beta)&=& \big(z\mapsto{}^{\kappa_1}\!\,\alpha z+{}^{\kappa_2}\!\,\beta\big)\\ &=& \big(z\mapsto{}^{\kappa_1}\!\,\alpha z+\mu{}^{\kappa_1}\!\,\beta\big)\\ &=& \big(z\mapsto{}^{\kappa_1}\!\,\big(\alpha z+{}^{\kappa_1^{-1}}\!\,\mu\beta\big)\big)\\ &=& \big(z\mapsto{}^{\kappa_1}\!\,\big(({}^{\kappa_1^{-1}}\!\,\mu z)\circ(\alpha z+\beta)\circ({}^{\kappa_1}\!\,\mu z)\big)\big)\\ &=& \big(z\mapsto{}^{\kappa_1}\!\,\big(({}^{\kappa_1}\!\,\mu z)^{-1}\circ(\alpha z+\beta)\circ({}^{\kappa_1}\!\,\mu z)\big)\big). \end{aligned}$$ Proposition \[pro:autbir\] and Lemma \[lem:autbir\] imply that for any $\alpha$, $\beta$ in $\mathbb{C}^$, for any $\gamma$, $\delta$ in $\mathbb{C}$ one has $$\varphi\big((z_0,z_1)\mapsto(\alpha z_0+\gamma,\beta z_1+\delta)\big)=\big((z_0,z_1)\mapsto({}^{\kappa_1}\!\,\alpha z_0+\mu{}^{\kappa_1}\!\,\gamma,{}^{\kappa_2}\!\,\beta z_1+\eta{}^{\kappa_2}\!\,\delta)\big)$$ where $\eta$, $\mu$ are two non-zero complex numbers and $\kappa_1$, $\kappa_2$ two automorphisms of the field $\mathbb{C}$. Since $(z_0,z_1)\mapsto(z_0+z_1,z_1)$ and $(z_0,z_1)\mapsto(\alpha z_0,\alpha z_1)$ commute their image by $\varphi$ also, and so $\kappa_1=\kappa_2$. As a consequence up to conjugacy by an inner automorphism and an automorphism of the field $\mathbb{C}$, the groups $$\mathrm{T}=\big\{(z_0,z_1)\mapsto(z_0+\alpha,z_1+\beta)\,\vert\,\alpha,\,\beta\in\mathbb{C}\big\}$$ and $$\mathrm{D}=\big\{(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)\,\vert\,\alpha,\,\beta\in\mathbb{C}^\big\}$$ are pointwise invariant. Then one can check that the involutions $(z_0,z_1)\mapsto\left(z_0,\frac{1}{z_1}\right)$ and $(z_0,z_1)\mapsto(z_1,z_0)$ are invariant by $\varphi$. But the group generated by $\mathrm{T}$, $\mathrm{D}$, $(z_0,z_1)\mapsto\left(z_0,\frac{1}{z_1}\right)$ and $(z_0,z_1)\mapsto(z_1,z_0)$ contains $\mathrm{PGL}(3,\mathbb{C})$. Furthermore $$\sigma_2=\left(\left((z_0,z_1)\mapsto\left(z_0,\frac{1}{z_1}\right)\right)\circ\left((z_0,z_1)\mapsto(z_1,z_0)\right)\right)^2$$ hence $\varphi(\sigma_2)=\sigma_2$. We conclude thanks to the <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> Theorem. An isomorphism of the semi-group of rational self maps of $\mathbb{P}^2_\mathbb{C}$ is inner up to the action of an automorphism of the field $\mathbb{C}$. In the spirit of the result of <span style="font-variant:small-caps;">Filipkiewicz</span> (Theorem \[thm:Filipkiewicz\]) one has: Let $S$ be a complex projective surface. Let $\varphi$ be an isomorphism between $\mathrm{Bir}(S)$ and $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. There exist a birational map $\psi\colon S\dashrightarrow\mathbb{P}^2_\mathbb{C}$ and an automorphism of the field $\mathbb{C}$ such that $$\varphi(\phi)={}^{\kappa}\!\,(\psi\circ\phi\circ\psi^{-1})\qquad\qquad\forall\,\phi\in\mathrm{Bir}(S).$$ The automorphism group of $\mathbb{C}(z_0,z_1)$ is isomorphic to the automorphisms group of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. According to [@Beauville] the groups $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ and $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ are isomorphic if and only if $n=2$. Note that there is no description of $\mathrm{Aut}(\mathrm{Bir}(\mathbb{P}^n_\mathbb{C}))$ for $n\geq 3$. Nevertheless there are two results in that direction: - the first one is Let $\varphi$ be an automorphism of $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$; there exist an automorphism $\kappa$ of the field $\mathbb{C}$, and a birational self map $\psi$ of $\mathbb{P}^n_\mathbb{C}$ such that $$\varphi(\phi)={}^{\kappa}\!\,(\psi\circ\phi\circ\psi^{-1})\qquad\qquad \forall\,\phi\in\mathrm{G}(n,\mathbb{C})=\langle\sigma_n,\,\mathrm{PGL}(n+1,\mathbb{C})\rangle.$$ - the second one is \[thm:Cantatcompos\] Let $V$ be a smooth connected complex projective variety of dimension $n$. Let $r$ be a positive integer and let $\rho\colon\mathrm{Aut}(\mathbb{P}^r_\mathbb{C})\to\mathrm{Bir}(V)$ be an injective morphism of groups. Then $n\leq r$. Furthermore if $n=r$, there exist a field morphism $\kappa\colon\mathbb{C}\to\mathbb{C}$ and a birational map $\psi\colon V\dashrightarrow\mathbb{P}^n_\mathbb{C}$ such that - either $\psi\circ\rho(A)\circ\psi^{-1}={}^{\kappa}\!\, A$ for all $A\in\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$, - or $\psi\circ\rho(A)\circ\psi^{-1}=({}^{\kappa}\!\, A)^\vee$ for all $A\in\mathrm{Aut}(\mathbb{P}^n_\mathbb{C})$. In particular $V$ is rational. Moreover $\kappa$ is an automorphism of $\mathbb{C}$ if $\rho$ is an isomorphism. Before giving an idea of the proof of this last result let us state two corollaries of it. The first shows that the <span style="font-variant:small-caps;">Cremona</span> groups $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ are pairwise non-isomorphic, thereby solving an open problem for $n\geq 4$. Let $n$ and $k$ be natural integers. The group $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ embeds into $\mathrm{Bir}(\mathbb{P}^k_\mathbb{C})$ if and only if $n\leq k$. In particular $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ is isomorphic to $\mathrm{Bir}(\mathbb{P}^k_\mathbb{C})$ if and only if $n=k$. The second characterizes rational varieties $V$ by the structure of $\mathrm{Bir}(V)$, as an abstract group: Let $V$ be an irreducible complex projective variety of dimension $n$. The following properties are equivalent: - $V$ is rational, - $\mathrm{Bir}(V)$ is isomorphic to $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ as an abstract group, - there is a non-trivial morphism from $\mathrm{PGL}(n+1,\mathbb{C})$ to $\mathrm{Bir}(V)$. The strategy that leads to the proof of Theorem \[thm:Cantatcompos\] is similar to the proof of Theorem \[thm:autofbir\] but requires several new ideas: - <span style="font-variant:small-caps;">Weil</span>’s regularization Theorem, that transforms a group of birational maps of $V$ with uniformly bounded degrees into a group of automorphisms of a new variety by a birational change of variables; - <span style="font-variant:small-caps;">Epstein</span> and <span style="font-variant:small-caps;">Thurston</span> work on nilpotent <span style="font-variant:small-caps;">Lie</span> subalgebras in the <span style="font-variant:small-caps;">Lie</span> algebra of smooth vector fields of a compact manifold ([@EpsteinThurston]). One-parameter subgroups of the <span style="font-variant:small-caps;">Cremona</span> group ------------------------------------------------------------------------------------------ ### Description of $1$-parameter subgroups of quadratic birational maps of $\mathbb{P}^2_\mathbb{C}$ A germ of flow in $\mathrm{Bir}_{\leq 2}(\mathbb{P}^2_\mathbb{C})$ is a germ of holomorphic application $t\mapsto\phi_t\in\mathrm{Bir}_{\leq 2}(\mathbb{P}^2_\mathbb{C})$ such that $$\left\{ \begin{array}{ll} \phi_{t+s}=\phi_t\circ\phi_s\\ \phi_0=\mathrm{id} \end{array} \right.$$ Since a germ of flow can be generalized we speak about flow. The set of lines blown down by the flow $\phi_t$ is a germ of analytic sets in the Grassmaniann of lines in $\mathbb{P}^2_\mathbb{C}$, [i.e.]{} in the dual space $(\mathbb{P}^2_\mathbb{C})^\vee$. Similarly the set of indeterminacy points of the $\phi_t$ is a germ of analytic sets of $\mathbb{P}^2_\mathbb{C}$. We call family of contracted curves a continuous map (indeed an analytic one) defined over a germ of closed sector $\Delta$ of vertex $0$ in $\mathbb{C}$ $$\mathcal{D}\colon\Delta\to(\mathbb{P}^2_\mathbb{C})^{\vee}$$ such that for any $t\in\Delta$ the lines $\mathcal{D}_t$ coincide with a line $\mathcal{D}(t)$ blown down by $\phi_t$. Similarly a family of indeterminacy points is a continuous map $t\mapsto p_t$ defined on a sector $\Delta$ such that any $p_t$ is an indeterminacy point of $\phi_t$. Let $\phi_t$ be a flow. Let $\mathcal{D}_t$ (resp. $p_t$) be a family of curves blown down by $\phi_t$ (resp. a family of indeterminacy points of $\phi_t$). If $\mathcal{D}_t$ (resp. $p_t$) is independent of $t$, the family is unmobile, otherwise it is mobile. A rational vector field $\chi$ on $\mathbb{P}^2_\mathbb{C}$ is rationally integrable if its flow is a flow of birational maps. A germ of flow in $\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$ is the flow of a rationally integrable vector field $\chi=\frac{\partial\phi_t}{\partial t}\Big\vert_{t=0}$ called infinitesimal generator of $\phi_t$. To this vector field is associated a foliation whose leaves are “grosso modo” the trajectories of $\chi$. Recall that a fibration by lines $\mathcal{L}$ of $\mathbb{P}^2_\mathbb{C}$ is given by $$\lambda\ell_1+\mu\ell_2=0$$ where $\ell_1$, $\ell_2$ are linear forms that are not proportional. The base-point is the intersection point $p$ of all these lines. We also say that $\mathcal{L}$ is a pencil of lines through $p$, or $\mathcal{L}$ is a foliation by lines singular at $p$. Recall that a birational self map of $\mathbb{P}^2_\mathbb{C}$ that preserves a rational fibration belongs up to birational conjugacy to $\mathcal{J}$. Let $\phi_t$ be a germ of flow in $\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$. Then the following properties hold: - assume that $\phi_t$ blows down a mobile line, then $\phi_t$ preserves a fibration by lines, more precisely the family of contracted lines belongs to a fibration invariant by any element of the flow ([@CerveauDeserti:ptdegre Proposition 2.5, Remark 2.6]); - there is at most one unmobile line blown down by $\phi_t$ (see [@CerveauDeserti:ptdegre Lemma 2.10]); - if $\phi_t$ blows down a unique line that is moreover unmobile, then there exists an invariant affine chart $\mathbb{C}^2$ such that $\phi_{t\vert\mathbb{C}^2}\colon\mathbb{C}^2\to\mathbb{C}^2$ is polynomial for any $t$ (see [@CerveauDeserti:ptdegre Proposition 2.12]); - assume that there exists an invariant affine chart $\mathbb{C}^2$ such that $\phi_{t\vert\mathbb{C}^2}\colon\mathbb{C}^2\to\mathbb{C}^2$ is polynomial for any $t$. Then $\phi_t$ preserves a pencil of lines. Furthermore either $\phi_t$ is affine, or there exists a normal form for $\phi_t$ up to linear conjugacy ([@CerveauDeserti:ptdegre Proposition 2.15]). Combining all these properties one can state the following result: \[thm:fibrationbis\] A germ of flow in $\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$ preserves a fibration by lines. Let $\phi_t$ be a quadratic birational flow, and let $\chi$ be its infinitesimal generator. A strong symmetry $Y$ of $\chi$ is a rationally integrable vector field of flow $\psi_s$ such that - $\phi_t$ and $\psi_s$ commute, [i.e.]{} $[\chi,Y]=0$, - $\psi_s\in\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$ for all $s$, - $\chi$ and $Y$ are not $\mathbb{C}$-colinear. Let $\phi_t$ be a flow in $\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$, and let $\chi$ (resp. $\mathcal{F}_\chi$) be the associated vector field (resp. foliation). We denote by $\overline{\langle\phi_t\rangle}^Z\subset\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$ the <span style="font-variant:small-caps;">Zariski</span> closure of $\langle\phi_t\rangle$ in $\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$. Let $\mathrm{G}(\chi)$ be the maximal algebraic abelian subgroup of $\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$ that contains $\overline{\langle\phi_t\rangle}^Z$. \[thm:flotalt\] Let $\phi_t$ be a germ of flow in $\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$, and let $\chi$ be its infinitesimal generator. - If $\dim\mathrm{G}(\chi)=1$, then $\mathcal{F}_\chi$ is a rational fibration. - If $\dim\mathrm{G}(\chi)\geq 2$, then $\mathcal{F}_\chi$ has a strong symmetry. In both cases $\mathcal{F}_\chi$ is defined by a rational closed $1$-form. Let us prove the first assertion. If $\dim\mathrm{G}(\chi)=1$, then $\overline{\langle\phi_t\rangle}^Z$ is the component of $\mathrm{G}(\chi)$ that contains the identity. This group viewed as a <span style="font-variant:small-caps;">Lie</span> group is isomorphic to $\mathbb{C}$, or $\mathbb{C}^$, or $\faktor{\mathbb{C}}{\Lambda}$. According to Theorem \[thm:fibrationbis\] the group $\overline{\langle\phi_t\rangle}^Z$ preserves a fibration by lines; let us assume that this fibration is given by $z_1=$ constant. One yields a morphism $$\pi\colon\overline{\langle\phi_t\rangle}^Z\to\mathrm{PGL}(2,\mathbb{C})$$ that describes the action of $\phi_t$ on the fibers. If $\pi$ is trivial ([i.e.]{} if the fibration is preserved fiberwise), then $\mathcal{F}_\chi=\big\{z_1=\text{ constant }\big\}$ and the result holds. Otherwise $\overline{\langle\phi_t\rangle}^Z$ is not isomorphic to $\faktor{\mathbb{C}}{\Lambda}$ because there is no $\faktor{\mathbb{C}}{\Lambda}$ among the subgroup of $\mathrm{PGL}(2,\mathbb{C})$. Hence the topological closure of $\overline{\langle\phi_t\rangle}^Z$ in $\mathbb{P}^{17}_\mathbb{C}\simeq\mathrm{Rat}_2$ is a rational curve. But according to <span style="font-variant:small-caps;">Darboux</span> a foliation of $\mathbb{P}^2_\mathbb{C}$ whose the closure of all leaves are algebraic curves has a non-constant rational first integral ([@Jouanolou]). In our case the curves are rational so $\mathcal{F}_\chi$ is a rational fibration. Let us now prove the second assertion. Assume $\dim\mathrm{G}(\chi)\geq 2$. One can find a germ of $1$-parameter group $\psi_s$ in $\mathrm{G}(\chi)$ not contained in $\langle\phi_t\rangle$. Let $Y$ be the infinitesimal generator of $\psi_s$. The vector fields $\chi$ and $Y$ commute and are not $\mathbb{C}$-colinear. Let us consider $\omega$ a rational $1$-form that define $\mathcal{F}_\chi$, [i.e.]{} $i_\chi\omega=0$. If $\chi$ and $Y$ are generically independent, then $\Omega=\frac{\omega}{i_Y\omega}$ is closed and define $\mathcal{F}_\chi$. If $\chi$ and $Y$ are not generically independent, then $Y=f\chi$ with $f$ rational and non-constant. Since $[\chi,Y]=0$ one has $\chi(f)=0$. As a result $\mathrm{d}f$ defines $\mathcal{F}_\chi$ and is closed. The last two statements can be generalized as follows: Let $\phi_t$ be a germ of flow in $\mathrm{Bir}_n(\mathbb{P}^2_\mathbb{C})$, and let $\chi$ be its infinitesimal generator. Denote by $\mathrm{G}(\chi)$ the abelian maximal algebraic group contained in $\mathrm{Bir}_n(\mathbb{P}^2_\mathbb{C})$ and that contains $\overline{\langle\phi_t\rangle}^Z$. Then - if $\dim\mathrm{G}(\chi)=1$, then $\mathcal{F}_\chi$ is either a rational fibration or an elliptic fibration; - if $\dim\mathrm{G}(\chi)\geq 2$, then $\chi$ has a strong symmetry. In both cases $\mathcal{F}_\chi$ is defined by a closed rational $1$-form. Any germ of birational flow in $\mathrm{Bir}_n(\mathbb{P}^2_\mathbb{C})$ preserves a rational fibration. ### A few words about the classification of germs of quadratic birational flows Let $\phi_t$ be a germ of flow in $\mathrm{Bir}_2(\mathbb{P}^2_\mathbb{C})$; then $\phi_t$ preserves a fibration by lines ([@CerveauDeserti:ptdegre Theorem 2.16]). In other words up to linear conjugacy $$\phi_t\colon(z_0,z_1)\dashrightarrow\left(\frac{A(z_1,t)z_0+B(z_1,t)}{C(z_1,t)z_0+D(z_1,t)},\nu(z_1,t)\right)$$ with - $\nu(z_1,t)=z_1$, or $z_1+t$, or $\mathrm{e}^{\beta t}z_1$; - $A$, $B$, $C$, $D$ are polynomials in $z_1$ and $\deg_{z_1}A\leq 1$, $\deg_{z_1}B\leq 2$, $\deg_{z_1}C=0$, $\deg_{z_1}D\leq 1$, - $B(z_1,0)=C(z_1,0)=0$ and $A(z_1,0)=D(z_1,0)$. The infinitesimal generator $\chi=\frac{\partial\phi_t}{\partial t}\Big\vert_{t=0}$ of $\phi_t$ can be written $$\frac{\alpha z_0^2+\ell(z_1)z_0+P(z_1)}{az_1+b}\frac{\partial}{\partial z_0}+\varepsilon (z_1)\frac{\partial}{\partial z_1}$$ with $\alpha$, $a$, $b\in\mathbb{C}$, $\ell$, $P\in\mathbb{C}[z_1]$, $\deg\ell=1$, $\deg P=2$ and up to linear conjugacy and scalar multiplication $\varepsilon\in\{0,\,1,\,z_1\}$. The above vector fields are classified up to automorphisms of $\mathbb{P}^2_\mathbb{C}$ and renormalization in [@CerveauDeserti:ptdegre Chapter 2, §2]; such vector fields are detected via the following methods: - compute explicitely the flow by integration; - or degenerate $\chi$ on another vector field $\chi_0$ that is not rationally integrable; - or show that a birational model of $\mathcal{F}_\chi$ has an isolated degenerate resonnant singular point (one and only one non-zero eigenvalue) and so $\mathcal{F}_\chi$ has no rational first integral. Then prove that there is no strong symmetry hence $\chi$ is not rationally integrable (Theorem \[thm:flotalt\]). Nilpotent subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ -------------------------------------------------------------- In [@Deserti:nilpotent] are described the nilpotent subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group: \[thm:nilpotent\] Let $\mathrm{N}$ be a nilpotent subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Assume that, up to finite index, $\mathrm{N}$ is not abelian. Then - either $\mathrm{N}$ is a torsion group; - or $\mathrm{N}$ is metabelian up to finite index, [i.e.]{} $[\mathrm{N},\mathrm{N}]$ is abelian up to finite index. Let $\alpha$ and $\beta$ be two non zero complex numbers; the group $$\langle(z_0,z_1)\mapsto(z_0+\alpha\beta,z_1),\,(z_0,z_1)\mapsto(z_0+\alpha z_1,z_1),\,(z_0,z_1)\mapsto(z_0,z_1+\beta)\rangle$$ is a non-abelian, non-finite and nilpotent subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. If $a$ belongs to $\mathbb{C}(z_1)$, then $$\langle(z_0,z_1)\mapsto(z_0+1,z_1),\,(z_0,z_1)\mapsto(z_0+z_1,z_1),\,(z_0,z_1)\mapsto(z_0+a(z_1),z_1-1)\rangle,$$ is a non-abelian, non-finite and nilpotent subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let $\mathrm{G}$ be a group. Assume that $\mathrm{G}$ contains a subgroup $\mathrm{N}$ such that - $\mathrm{N}$ is of nilpotent class $>1$, - $\mathrm{N}$ has no torsion, - $\mathrm{N}$ is not metabelian up to finite index. Then there is no faithfull representation of $\mathrm{G}$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. \[rem:distorted\] Let $\mathrm{G}$ be a nilpotent group of nilpotent class $n$. Take $f$ in $\mathrm{G}$, $g$ in $C^{(n-2)}\mathrm{G}$ and consider $h=[f,g]\in C^{(n-1)}\mathrm{G}$. Since $\mathrm{G}$ is of nilpotent class $n$, then $[f,h]=[g,h]=\mathrm{id}$. In other words any nilpotent group contains a distorted element. According to Remark \[rem:distorted\] and Lemma \[lem:nilkeylemma\] one has: Let $\mathrm{N}$ be a nilpotent subgroup of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. It contains a distorted element which is elliptic or parabolic. Take $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ a nilpotent subgroup of class $k$ which is not up to finite index of nilpotent class $k-1$. Denote by $\Sigma_\mathrm{G}$ the set of finitely generated nilpotent subgroups of $\mathrm{G}$ that are, up to finite index, not abelian. Then - either any element of $\Sigma_\mathrm{G}$ is finite and $\mathrm{G}$ is a torsion group; - or $\Sigma_\mathrm{G}$ contains a non-finite element $\mathrm{H}$. \[claim:nil\] The group $\mathrm{H}$ preserves a fibration $\mathcal{F}$ that is rational or elliptic. Any element of $C^{(k-1)}\mathrm{H}$ preserves fiberwise $\mathcal{F}$. Let $\phi$ be in $C^{(k-1)}\mathrm{H}$. As $[\phi,\mathrm{G}]=\mathrm{id}$, then - either $\phi$ preserves fiberwise two distinct fibrations; - of $\mathrm{G}$ preserves fiberwise $\mathcal{F}$. If a) holds, then $\phi$ is of finite order; if it is the case for any $\phi\in C^{(k-1)}\mathrm{H}$, then $\mathrm{H}$ is, up to finite index, of nilpotent class $k-1$: contradiction. If b) holds, then $\mathrm{G}$ is, up to finite index, metabelian. Let us detail why when $\mathcal{F}$ is rational. In that case $\mathrm{G}$ is, up to conjugacy, a subgroup of the <span style="font-variant:small-caps;">Jonquières</span> group $\mathcal{J}$. Let $\mathrm{pr}_2$ be the projection $\mathcal{J}\to\mathrm{PGL}(2,\mathbb{C})$. A non-finite nilpotent subgroup of $\mathrm{PGL}(2,\Bbbk)$, where $\Bbbk=\mathbb{C}$ or $\mathbb{C}(z_1)$, is up to finite index abelian. The group $\mathrm{pr}_2(\mathrm{G})$ is thus, up to finite index, abelian. We can thus assume that $\mathrm{pr}_2(C^{(i)}\mathrm{G})=\{\mathrm{id}\}$ for $1\leq i\leq k$. In particular $C^{(1)}\mathrm{G}$ is a nilpotent subgroup of $\mathrm{PGL}(2,\mathbb{C}(z_1))$ and thus is, up to finite index, abelian. Let us recall that $\mathrm{H}$ is a non-finite nilpotent subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ with the following properties: - $\mathrm{H}$ is finitely generated, - $\mathrm{H}$ is nilpotent of class $k>0$, - $\mathrm{H}$ is not, up to finite index, of nilpotent class $k-1$. Assume $C^{(k-1)}\mathrm{H}$ is not a torsion group. Then $\mathrm{H}$ preserves a fibration that is rational or elliptic. According to Lemma \[lem:nilkeylemma\] a non-trivial element of $C^{(k-1)}\mathrm{G}$ either preserves a unique fibration $\mathcal{F}$ that is rational or elliptic, or is an elliptic birational map. We have the following alternative: - either $C^{(k-1)}\mathrm{G}$ contains an element $h$ that preserves a unique fibration $\mathcal{F}$, - or any element of $C^{(k-1)}\mathrm{G}\smallsetminus\{\mathrm{id}\}$ is elliptic. Let us look at these eventualities: - Since $[h,\mathrm{G}]=\{\mathrm{id}\}$ any element of $\mathrm{G}$ preserves $\mathcal{F}$. - The group $C^{(k-1)}\mathrm{G}$ is finitely generated and abelian. Let $\big\{a_1,\,a_2,\,\ldots,\,a_n\big\}$ be a generating set of $C^{(k-1)}\mathrm{G}$. The $a_i$’s are elliptic maps so there exist a surface $S_i$, a birational map $\eta_i\colon S_i\dashrightarrow \mathbb{P}^2_\mathbb{C}$ and an integer $k_i>0$ such that $\eta_i^{-1}\circ a_i^{k_i}\circ\eta_i$ belongs to the neutral component $\mathrm{Aut}(S_i)^0$ of $\mathrm{Aut}(S_i)$. In particular the $a_i$’s fix any curve of negative self-intersection, we can thus assume that $S_i$ is a minimal rational surface. A priori all the $S_i$ are distinct. Nevertheless according to Proposition \[pro:commas\] there exist a minimal rational surface $S$, a birational map $\eta\colon S\dashrightarrow\mathbb{P}^2_\mathbb{C}$ and an integer $k>0$ such that for any $1\leq i\leq n$ the map $\eta^{-1}\circ a_i^{k}\circ\eta$ belongs to the neutral component $\mathrm{Aut}(S)^0$ of $\mathrm{Aut}(S)$. Minimal rational surfaces are $\mathbb{P}^2_\mathbb{C}$, $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ and <span style="font-variant:small-caps;">Hirzebruch</span> surfaces $\mathbb{F}_n$, $n\geq 2$. Using - on the one hand the description of the automorphisms groups of minimal rational surfaces (see* Chapter \[Chapter:algebraicsubgroup\]), - and on the other hand the fact that if $\mathrm{K}$ is an algebraic <span style="font-variant:small-caps;">Lie</span> subgroup of $\mathrm{GL}(n,\mathbb{C})$, then the semi-simple and nilpotent parts of any element of $\mathrm{K}$ belong to $\mathrm{K}$, we prove that $\mathrm{G}$ is, up to finite index and up to conjugacy, contained in the <span style="font-variant:small-caps;">Jonquières</span> group $\mathcal{J}$ (see [@Deserti:nilpotent]). It remains to consider the case “$C^{(k-1)}\mathrm{G}$ is a torsion group”; the ideas are similar (see [@Deserti:nilpotent Proposition 4.5]). Centralizers ------------ ### Centralizers of elliptic birational maps {#subsec:centrell} We will focus on the case of birational self maps of $\mathbb{P}^2_\mathbb{C}$ of infinite order. Note for instance that for birational self map of $\mathbb{P}^2_\mathbb{C}$ of finite order the situation is wild: consider for instance a birational involution $\phi$ of $\mathbb{P}^2_\mathbb{C}$. If $\phi$ is conjugate to an automorphism of $\mathbb{P}^2_\mathbb{C}$, then the centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is uncountable but if $\phi$ is conjugate to a <span style="font-variant:small-caps;">Bertini</span> (or a <span style="font-variant:small-caps;">Geiser</span>) involution, then the centralizer is finite ([@BlancPanVust]). According to [@BlancDeserti:degree] an elliptic birational self map of $\mathbb{P}^2_\mathbb{C}$ of infinite order is conjugate to an automorphism of $\mathbb{P}^2_\mathbb{C}$ which restricts to one of the following automorphisms on some open subset isomorphic to $\mathbb{C}^2$: - $(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)$ where $\alpha$, $\beta$ belong to $\mathbb{C}^$ and where the kernel of the group homomorphism $$\begin{aligned} & \mathbb{Z}^2\to\mathbb{C}^2 && (i,j)\mapsto \alpha^i\beta^j\end{aligned}$$ is generated by $(k,0)$ for some $k\in\mathbb{Z}$; - $(z_0,z_1)\mapsto(\alpha z_0,z_1+1)$ where $\alpha\in\mathbb{C}^$. We can describe the centralizers of such maps; let us start with the centralizer of $(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)$ where $\alpha$, $\beta$ belong to $\mathbb{C}^$ and where the kernel of the group homomorphism $$\begin{aligned} & \mathbb{Z}^2\to\mathbb{C}^2 && (i,j)\mapsto \alpha^i\beta^j\end{aligned}$$ is generated by $(k,0)$ for some $k\in\mathbb{Z}$. Recall that $\mathrm{PGL}(2,\mathbb{C})$ is the group of automorphisms of $\mathbb{P}^1_\mathbb{C}$ or equivalently the group of <span style="font-variant:small-caps;">Möbius</span> transformations $$z_0\dashrightarrow\frac{az_0+b}{cz_0+d}$$ A direct computation implies the following: for any $\alpha\in\mathbb{C}^$ $$\big\{\eta\in\mathrm{PGL}(2,\mathbb{C})\,\vert\,\eta(\alpha z_0)=\alpha\eta(z_0)\big\}=\left\{ \begin{array}{lll} \mathrm{PGL}(2,\mathbb{C}) \text{ if $\alpha=1$}\\ \big\{z_0\dashrightarrow \gamma z_0^{\pm 1}\,\vert\,\gamma\in\mathbb{C}^\big\} \text{ if $\alpha=-1$}\\ \big\{z_0\mapsto\gamma z_0\,\vert\,\gamma\in\mathbb{C}^\big\} \text{ if $\alpha^2\not=1$} \end{array} \right.$$ Let us consider $\phi\colon(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)$ where $\alpha$, $\beta$ belongs to $\mathbb{C}^$ and where the kernel of the group homomorphism $$\begin{aligned} & \mathbb{Z}^2\to\mathbb{C}^2 && (i,j)\mapsto \alpha^i\beta^j\end{aligned}$$ is generated by $(k,0)$ for some $k\in\mathbb{Z}$. The centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $$\big\{(z_0,z_1)\dashrightarrow(\eta(z_0),z_1a(z_0^k))\,\vert\, a\in\mathbb{C}(z_0),\,\eta\in\mathrm{PGL}(2,\mathbb{C}),\,\eta(\alpha z_0)=\alpha\eta(z_0)\big\}.$$ Let $\psi\colon(z_0,z_1)\dashrightarrow(\psi_0(z_0,z_1),\psi_1(z_0,z_1))$ be a birational self map of $\mathbb{P}^2_\mathbb{C}$ that commutes with $\phi$. Then $$\label{eqet} \psi_0(\alpha z_0,\beta z_1)=\alpha\psi_0(z_0,z_1)$$ and $$\label{eqtri} \psi_1(\alpha z_0,\beta z_1)=\beta\psi_1(z_0,z_1)$$ hold. Denote by $\phi^$ the linear automorphism of the $\mathbb{C}$-vector space $\mathbb{C}[z_0,z_1]$ given by $$\phi^\colon \varphi(z_0,z_1)\mapsto\varphi(\alpha z_0,\beta z_1).$$ Let us write $\psi_i$ as $\frac{P_i}{Q_i}$ for $i=0$, $1$ where $P_i$, $Q_i$ are polynomials without common factor. Note that $P_0$, $P_1$, $Q_0$, $Q_1$ are eigenvectors of $\phi^$, [i.e.]{} any of the $P_i$, $Q_i$ is a product of a monomial in $z_0$, $z_1$ with an element of $\mathbb{C}[z_0^k]$. Using (\[eqet\]) and (\[eqtri\]) we get that $$\left\{ \begin{array}{ll} \psi_0(z_0,z_1)=z_0a_0(z_0^k)\\ \psi_1(z_0,z_1)=z_1a_1(z_0^k) \end{array} \right.$$ But $\psi$ is birational so $\psi_0$ belongs to $\mathrm{PGL}(2,\mathbb{C})$. Furthemore $\psi_0$ satisfies $\psi_0(\alpha z_0)=\alpha\psi_0(z_0)$. Let us now deal with the other possibility: Let $\phi$ be the automorphism of $\mathbb{P}^2_\mathbb{C}$ given by $$\phi\colon(z_0,z_1)\mapsto(\alpha z_0,z_1+\beta)$$ where $\alpha\in\mathbb{C}^$, $\beta\in\mathbb{C}$. The centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $$\big\{(z_0,z_1)\dashrightarrow(\eta(z_0),z_1+a(z_0))\,\vert\,\eta\in\mathrm{PGL}(2,\mathbb{C}),\,\eta(\alpha z_0)=\alpha\eta(z_0),\,a\in\mathbb{C}(z_0),\,a(\alpha z_0)=a(z_0)\big\}$$ After conjugacy by $(z_0,z_1)\mapsto(z_0,\beta z_1)$ we can assume that $\beta=1$. If $\psi\colon(z_0,z_1)\dashrightarrow(\psi_0(z_0,z_1),\psi_1(z_0,z_1))$ is a birational map that commutes with $\phi$, then $$\label{eq:equ1} \psi_0(\alpha z_0,z_1+1)=\alpha\psi_0(z_0,z_1)$$ and $$\label{eq:equ2} \psi_1(\alpha z_0,z_1+1)=\psi_1(z_0,z_1)+1$$ From (\[eq:equ1\]) and [@Blanc:manuscripta] we get that $\psi_0$ only depends on $z_0$. Hence $\psi_0$ belongs to $\mathrm{PGL}(2,\mathbb{C})$ and commutes with $z_0\mapsto\alpha z_0$. From (\[eq:equ2\]) we get $$\left\{ \begin{array}{ll} \frac{\partial\psi_1}{\partial z_1}(\alpha z_0,z_1+1)=\frac{\partial\psi_1}{\partial z_1}(z_0,z_1)\\ \frac{\partial\psi_1}{\partial z_0}(\alpha z_0,z_1+1)=\frac{1}{\alpha}\frac{\partial\psi_1}{\partial z_0}(z_0,z_1) \end{array} \right.$$ which again means that both $\frac{\partial\psi_1}{\partial z_0}$ and $\frac{\partial\psi_1}{\partial z_1}$ only depend on $z_0$. Therefore $\psi_1\colon(z_0,z_1)\mapsto \gamma z_1+b(z_0)$ with $\gamma\in\mathbb{C}^$ and $b\in\mathbb{C}(z_0)$. Then (\[eq:equ2\]) can be rewritten $$b(\alpha z_0)=b(z_0)+1-\gamma$$ which implies that $$\frac{\partial b}{\partial z_0}(\alpha z_0)=\frac{1}{\alpha}\frac{\partial b}{\partial z_0}(z_0)$$ and thus that $z_0\frac{\partial b}{\partial z_0}(z_0)$ is invariant under $z_0\mapsto\alpha z_0$. If $\alpha$ is not a root of unity, then $\frac{\partial b}{\partial z_0}=\frac{\delta}{z_0}$ for some $\delta\in\mathbb{C}$. As $b$ is rational, $\delta$ is zero and $b$ is constant. As a consequence $b(\alpha z_0)=b(z_0)+1-\gamma$ implies $\gamma=1$, that is $\psi_1\colon(z_0,z_1)\dashrightarrow z_0+\beta$. Assume that $\alpha$ is a primitive $k$-th root of unity. The map $\psi\colon(z_0,z_1)\dashrightarrow(\eta(z_0),\gamma z_1+b(z_0))$ commutes with $$\phi^k\colon(z_0,z_1)\dashrightarrow(z_0,z_1+k)$$ if and only if $\gamma(z_1+k)+b(z_0)=\gamma z_1+b(z_0)+k$, [i.e.]{} if and only if $\gamma=1$. Then $b(\alpha z_0)=b(z_0)+1-1$ can be rewritten $b(z_0)=b(\alpha z_0)$. ### Centralizers of <span style="font-variant:small-caps;">Jonqui\` eres</span> twists Recall that the subgroup $\mathcal{J}$ of <span style="font-variant:small-caps;">Jonquières</span> maps is isomorphic to $\mathrm{PGL}(2,\mathbb{C}(z_1))\rtimes\mathrm{PGL}(2,\mathbb{C})$. Let us denote by $\mathrm{pr}_2$ the morphism $$\mathrm{pr}_2\colon\mathcal{J}\to\mathrm{PGL}(2,\mathbb{C}).$$ Geometrically it corresponds to look at the action of $\phi\in\mathcal{J}$ on the basis of the invariant fibration $z_1=$ cst. The kernel of $\mathrm{pr}_2$, [i.e.]{} the elements of $\mathcal{J}$ which preserve the fibration $z_1=$ cst fiberwise, is a normal subgroup $\mathcal{J}_0\simeq\mathrm{PGL}(2,\mathbb{C}(z_1))$ of $\mathcal{J}$. Up to a birational conjugacy an element $\phi$ of $\mathcal{J}_0$ is of one of the following form ([@Deserti:abelien]) $$\begin{aligned} & (z_0,z_1)\dashrightarrow(z_0+a(z_1),z_1), && (z_0,z_1)\dashrightarrow(b(z_1)z_0,z_1), && (z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+F(z_1)}{z_0+c(z_1)},z_1\right)\end{aligned}$$ with $a\in\mathbb{C}(z_1)$, $b\in\mathbb{C}(z_1)^$, $c\in\mathbb{C}(z_1)$, $F\in\mathbb{C}[z_1]$ and $F$ not a square. Still according to [@Deserti:abelien] the non-finite maximal abelian subgroups of $\mathcal{J}_0$ are $$\begin{aligned} \mathcal{J}_a&=&\big\{(z_0,z_1)\dashrightarrow(z_0+a(z_1))\,\vert\,a\in\mathbb{C}(z_1)\big\}\\ \mathcal{J}_m&=&\big\{(z_0,z_1)\dashrightarrow(b(z_1)z_0,z_1)\,\vert\,a\in\mathbb{C}(z_1)\big\}\\ \mathcal{J}_F&=&\big\{(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+F(z_1)}{z_0+c(z_1)},z_1\right)\,\vert\,a\in\mathbb{C}(z_1)\big\}\end{aligned}$$ where $F$ denotes an element of $\mathbb{C}[z_1]$ which is not a square. Note that we can assume up to conjugacy that $F$ is a polynomial with roots of multiplicity $1$. If $\phi$ belongs to $\mathcal{J}_0$, let us denote by $\mathrm{Ab}(\phi)$ the non-finite maximal abelian subgroup of $\mathcal{J}_0$ that contains $\phi$. Up to conjugacy - either $(z_0,z_1)\dashrightarrow(z_0+a(z_1),z_1)$ and $\mathrm{Ab}(\phi)=\mathcal{J}_a$; - or $(z_0,z_1)\dashrightarrow(b(z_1)z_0,z_1)$ and $\mathrm{Ab}(\phi)=\mathcal{J}_m$; - or $(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+F(z_1)}{z_0+c(z_1)},z_1\right)$ and $\mathrm{Ab}(\phi)=\mathcal{J}_F$. \[pro:centrfib\] Let $\phi$ be an element of $\mathcal{J}_0$ that is a <span style="font-variant:small-caps;">Jonquières</span> twist. Then the centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is contained in $\mathcal{J}$. Consider a birational self map $\varphi\colon(z_0,z_1)\dashrightarrow(\varphi_0(z_0,z_1),\varphi_1(z_0,z_1))$ of $\mathbb{P}^2_\mathbb{C}$ that commutes with $\phi$. If $\varphi$ does not belong to $\mathcal{J}$, then $\varphi_1=$ cst is a fibration invariant by $\phi$ distinct from $z_1=$ cst. Then $\phi$ is of finite order (Lemma \[lem:2fibr\]): contradiction with the fact that $\phi$ is a <span style="font-variant:small-caps;">Jonquières</span> twist. #### Centralizers of elements of $\mathcal{J}_a$ Note that elements of $\mathcal{J}_a$ are not <span style="font-variant:small-caps;">Jonquières</span> twists but elliptic maps. Hence their centralizers are described in §\[subsec:centrell\]. Let us give some details. Let $\phi\colon(z_0,z_1)\dashrightarrow(z_0+a(z_1),z_1)$ be a non-trivial element of $\mathcal{J}_a$ ([i.e.]{} $a\not\equiv 0$). Up to conjugacy by $(z_0,z_1)\dashrightarrow(a(z_1)z_0,z_1)$ one can assume that $a\equiv 1$. The centralizer of $(z_0,z_1)\dashrightarrow(z_0+1,z_1)$ is isomorphic to $\mathcal{J}_a\rtimes\mathrm{PGL}(2,\mathbb{C})$ (see* §\[subsec:centrell\]). Hence The centralizer of a non-trivial element of $\mathcal{J}_a$ is isomorphic to $\mathcal{J}_a\rtimes\mathrm{PGL}(2,\mathbb{C})$. #### centralizers of twists of $\mathcal{J}_m$ An element $\phi$ of $\mathcal{J}_m$ is a <span style="font-variant:small-caps;">Jonquières</span> twist if and only if up to birational conjugacy $$(z_0,z_1)\dashrightarrow(a(z_1)z_0,z_1)$$ with $a\in\mathbb{C}(z_1)\smallsetminus\mathbb{C}^$. Remark that if $a$ belongs to $\mathbb{C}^$, then $(z_0,z_1)\dashrightarrow(az_0,z_1)$ is an elliptic map whose centralizer is described in § \[subsec:centrell\]. Assume now that $\phi\in\mathcal{J}_m$ is a <span style="font-variant:small-caps;">Jonquières</span> twist. Let $a\in\mathbb{C}(z_1)\smallsetminus\mathbb{C}^$. Denote by $$\mathrm{Stab}(a)=\big\{\nu\in\mathrm{PGL}(2,\mathbb{C})\,\vert\, a(\nu(z_1))=a(z_1)^{\pm 1}\big\}$$ the subgroup of $\mathrm{PGL}(2,\mathbb{C})$ and by $$\mathrm{stab}(a)=\big\{\nu\in\mathrm{PGL}(2,\mathbb{C})\,\vert\, a(\nu(z_1))=a(z_1)\big\}$$ the normal subgroup of $\mathrm{Stab}(a)$. Consider also $$\overline{\mathrm{stab}(a)}=\big\{(z_0,\nu(z_1))\,\vert\, \nu\in\mathrm{stab}(a)\big\}$$ and $\overline{\mathrm{Stab}(a)}$ the group generated by $\overline{\mathrm{stab}(a)}$ and the elements $$(z_0,z_1)\dashrightarrow\left(\frac{1}{z_0},\nu(z_1)\right)$$ with $\nu\in\mathrm{Stab}(a)\smallsetminus\mathrm{stab}(a)$. Let $\phi$ be a <span style="font-variant:small-caps;">Jonquières</span> twist in $\mathcal{J}_m$. The centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $\mathcal{J}_m\rtimes\overline{\mathrm{Stab}(a)}$; in particular it is a finite extension of $\mathrm{Ab}(\phi)=\mathcal{J}_m$. One can write $\phi$ as $(z_0,z_1)\dashrightarrow(a(z_1)z_0,z_1)$ with $a\in\mathbb{C}(z_1)\smallsetminus\mathbb{C}^$. For generic $a$ the group $\overline{\mathrm{Stab}(a)}$ is trivial, so for generic $\phi\in\mathcal{J}_m$ the centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ coincides with $\mathcal{J}_m=\mathrm{Ab}(\phi)$. Write $\phi$ as $(z_0,z_1)\dashrightarrow(a(z_1)z_0,z_1)$ with $a\in\mathbb{C}(z_1)\smallsetminus\mathbb{C}^$. If $\psi$ commutes with $\phi$, then $\psi$ preserves the fibration $z_1=$ cst (Proposition \[pro:centrfib\]), [i.e.]{} $$\psi\colon(z_0,z_1)\dashrightarrow\left(\frac{A(z_1)z_0+B(z_1)}{C(z_1)z_0+D(z_1)},\nu(z_1)\right)$$ with $\left(\begin{array}{cc} A & B\\ C& D \end{array} \right)\in\mathrm{PGL}(2,\mathbb{C}(z_1))$ and $\nu\in\mathrm{PGL}(2,\mathbb{C})$. Since $\psi$ and $\phi$ commute, the following hold $$\left\{ \begin{array}{ll} A(z_1)C(z_1)\big(1-a(\nu(z_1))\big)=0\\ B(z_1)D(z_1)\big(1-a(\nu(z_1))\big)=0 \end{array} \right.$$ Therefore $AC\equiv0$ and $BD\equiv 0$, [i.e.]{} $B=C=0$ or $A=D=0$. Assume first that $B=C=0$, [i.e.]{} that $$\psi\colon(z_0,z_1)\dashrightarrow(A(z_1)z_0,\nu(z_1)).$$ The condition $\phi\circ\psi=\psi\circ\phi$ implies $a(\nu(z_1))=a(z_1)$. As $\overline{\mathrm{stab}(a)}$ is contained in the centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ the map $\phi$ belongs to $\mathcal{J}_m\rtimes\overline{\mathrm{stab}(a)}$. Suppose now that $A=D=0$, [i.e.]{} that $\psi\colon(z_0,z_1)\dashrightarrow\left(\frac{B(z_1)}{z_0},\nu(z_1)\right)$. The equality $\psi\circ\varphi=\varphi\circ\psi$ implies that $a(\nu(z_1))=a(z_1)^{-1}$. But $\overline{\mathrm{Stab}(a)}$ is contained in the centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ so $\psi$ belongs to $\mathcal{J}_m\rtimes\overline{\mathrm{Stab}(a)}$. #### Centralizers of elements of $\mathcal{J}_F$ Let $\phi$ be a twist in $\mathcal{J}_F$. Let us write $\phi$ as $$(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+F(z_1)}{z_0+c(z_1)},z_1\right)$$ with $c\in\mathbb{C}(z_1)^$ and $F\in\mathbb{C}[z_1]$ whose roots have multiplicity $1$. The curve $\mathcal{C}$ of fixed points of $\phi$ is given by $z_0^2=F(z_1)$. Since $F$ has simple roots one has $$\left\{\begin{array}{lll} \text{$\mathcal{C}$ is rational when $1\leq\deg F\leq 2$;}\\ \text{the genus of $\mathcal{C}$ is $1$ when $3\leq\deg F\leq 4$;}\\ \text{the genus of $\mathcal{C}$ is $\geq 2$ when $\deg F\geq 5$.} \end{array} \right.$$ - Assume first that the genus of $\mathcal{C}$ is positive. Let $$\begin{aligned} &\phi\colon(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+F(z_1)}{z_0+c(z_1)},z_1\right) && c\in\mathbb{C}(z_1)^, \,F\in\mathbb{C}[z_1]\end{aligned}$$ be a twist in $\mathcal{J}_F$. The curve $z_0^2=F(z_1)$ and the fibers $z_1=$cst are invariant and there is no other invariant curves. The map $\phi$ has two fixed points on a generic fiber which correspond to the intersection of the fiber with the curve $z_0^2=F(z_1)$. Assume by contradiction that there is an other invariant curve $\mathcal{C}$. The curve $\mathcal{C}$ intersects a generic fiber in a finite number of points that are invariant by $\phi$. But a <span style="font-variant:small-caps;">Möbius</span> transformation that preserves more than three points is periodic: contradiction with the fact that $\phi$ is a <span style="font-variant:small-caps;">Jonquières</span> twist so of infinite order. Let $$\begin{aligned} &\phi\colon(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+F(z_1)}{z_0+c(z_1)},z_1\right) && c\in\mathbb{C}(z_1)^, \,F\in\mathbb{C}[z_1]\end{aligned}$$ be a twist in $\mathcal{J}_F$. Assume that $F$ has only simple roots and $\deg F\geq 3$, [i.e.]{} the curve $z_0^2=F(z_1)$ has genus $\geq 1$. Then the centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is a finite extension of $\mathrm{Ab}(\phi)=\mathcal{J}_F$. Take $\alpha\in\mathbb{C}$ such that $F(\alpha)\not=0$. The restriction $\phi_{\vert z_1=\alpha}$ of $\phi$ on the fiber $z_1=\alpha$ has two fixed points: $(\pm\sqrt{F(\alpha)},\alpha)$. Note that the centralizer $\mathrm{Cent}(\phi)$ of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is contained in $\mathcal{J}$ (Proposition \[pro:centrfib\]). We will focus on elements $\psi$ of $\mathrm{Cent}(\phi)$ that preserve the fibration $z_1=$cst fiberwise, [i.e.]{} on the kernel of $$\mathrm{pr}_{2\vert\mathrm{Cent}(\phi)}\colon\mathrm{Cent}(\phi)\to\mathrm{PGL}(2,\mathbb{C}).$$ Remark that any $\psi\in\mathrm{Cent}(\phi)$ preserves $\mathcal{C}$ and that the automorphism $\psi_{\vert\mathcal{C}}$ of $\mathcal{C}$ preserves $\big\{(\pm\sqrt{F(\alpha)},\alpha)\big\}$. Hence either $\psi_{\vert\mathcal{C}}=\mathrm{id}$, that is $\psi\in\mathcal{J}_F$, or $\psi_{\vert\mathcal{C}}$ is the involution $(z_0,z_1)\mapsto(-z_0,z_1)$ of $\mathcal{C}$. Note that the restriction of $\tau\colon(z_0,z_1)\dashrightarrow\left(-\frac{F(z_1)}{z_0},z_1\right)$ to $\mathcal{C}$ is $\tau_{\vert\mathcal{C}}\colon(z_0,z_1)\dashrightarrow(-z_0,z_1)$. Therefore any birational self map of $\mathbb{P}^2_\mathbb{C}$ that preserves both $\mathcal{C}$ and the fibration $z_1=$cst fiberwise belongs either to $\mathcal{J}_F$ or to $\tau\circ\mathcal{J}_F$. But $\tau\circ\phi\circ\tau^{-1}=\tau\circ\phi\circ\tau=\phi^{-1}$ so $\tau$ does not belong to $\mathrm{Cent}(\phi)$. As a result $\ker\mathrm{pr}_{2\vert\mathrm{Cent}(\phi)}=\mathcal{J}_F$. Any $\varphi\in\mathrm{Cent}(\phi)$ has to preserve $\mathcal{C}$ and the fibration $z_1=$cst; the restriction $\varphi_{\vert\mathcal{C}}$ of $\varphi$ to $\mathcal{C}$ is an automorphism of $\mathcal{C}$ that commutes with the involution $\tau_{\vert\mathcal{C}}$. The group $\mathrm{Aut}_\tau(\mathcal{C})$ of such automorphisms is a finite group (more precisely if $F$ is generic, then $\mathrm{Aut}_\tau(\mathcal{C})=\{\mathrm{id},\,\tau_{\vert\mathcal{C}}\}$). - Assume that $\mathcal{C}$ is rational. Let $\phi\in\mathcal{J}_F$ be a <span style="font-variant:small-caps;">Jonquières</span> twist such that the curve $\mathcal{C}$ of fixed points of $\phi$ is rational. Any element that commutes with $\phi$ belongs to $\mathcal{J}$ and preserves $\mathcal{C}$. The curve of fixed points of $\phi$ is given by $z_0^2=F(z_1)$. Let $\psi$ be a birational self map of $\mathbb{P}^2_\mathbb{C}$ such that $\phi\circ\psi=\psi\circ\phi$. According to Proposition \[pro:centrfib\] the map $\psi$ preserves the fibration $z_1=$cst. Either $\psi$ contracts $\mathcal{C}$ or $\psi$ preserves $\mathcal{C}$. But $\mathcal{C}$ is transverse to the fibration $z_1=$cst so $\psi$ can not contract $\mathcal{C}$. As a result $\varphi$ is an element of $\mathcal{J}$ that preserves $\mathcal{C}$. Note that the case $\deg F\geq 3$ has already been studied so let us assume that $\deg F\leq 2$. Remark that if $$\phi\colon(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+z_1}{z_0+c(z_1)},z_1\right)$$ and if $$\varphi\colon(z_0,z_1)\dashrightarrow\left(\frac{z_0}{\gamma z_1+\delta},\frac{\alpha z_1+\beta}{\gamma z_1+\delta}\right)$$ then $\varphi^{-1}\circ\phi\circ\varphi$ is of the following type $$(z_0,z_1)\dashrightarrow\left(\frac{\widetilde{c}(z_1)z_0+(\alpha z_1+\beta)(\gamma z_1+\delta)}{z_0+\widetilde{c}(z_1)},z_1\right).$$ In other words thanks to $$(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+z_1}{z_0+c(z_1)},z_1\right)$$ we obtain all polynomials $(\alpha z_1+\beta)(\gamma z_1+\delta)$ of degree $2$ with simple roots. So one can suppose that $\deg F=1$. Note that if $\deg F=1$, [i.e.]{} $F(z_1)=\alpha z_1+\beta$, then up to conjugacy by $(z_0,z_1)\mapsto\left(z_0,\frac{z_1-\beta}{\alpha}\right)$ one can assume that $F\colon z_1\mapsto z_1$. \[lem:stab\] Consider the birational self map of $\mathbb{P}^2_\mathbb{C}$ given by $$\phi\colon(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+z_1}{z_0+c(z_1)},z_1\right)$$ with $c\in\mathbb{C}(z_1)^$. If $\psi$ is a birational self map of $\mathbb{P}^2_\mathbb{C}$ that commutes with $\phi$, then - either $\mathrm{pr}_2(\psi)=\frac{\alpha}{z_1}$ with $\alpha\in\mathbb{C}^$; - or $\mathrm{pr}_2(\psi)=\zeta z_1$ with $\zeta$ root of unity. Furthermore $\mathrm{pr}_2(\psi)$ belongs to the finite group $\mathrm{stab}\left(\frac{4c^2(z_1)}{c^2(z_1)-z_1}\right)$. For any $\alpha$ non-zero consider the dihedral group $$\mathrm{D}_\infty(\alpha)=\langle z_1\mapsto \frac{\alpha}{z_1},\,z_1\mapsto\zeta z_1\,\vert\,\zeta\text{ root of unity}\rangle$$ Note that all the $\mathrm{D}_\infty(\alpha)$ are conjugate to $\mathrm{D}_\infty(1)$. Let $\phi\in\mathcal{J}_F$ be a <span style="font-variant:small-caps;">Jonquières</span> twist such that the fixed curve of $\phi$ is rational. Up to conjugacy we can assume that $$\phi\colon(z_0,z_1)\dashrightarrow\left(\frac{c(z_1)z_0+z_1}{z_0+c(z_1)},z_1\right)$$ with $c\in\mathbb{C}(z_1)\smallsetminus\mathbb{C}$. The centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $$\mathcal{J}_{z_1}\rtimes\left(\mathrm{stab}\left(\frac{4c^2(z_1)}{c^2(z_1)-z_1}\right)\cap\mathrm{D}_\infty(\alpha)\right)$$ for some $\alpha\in\mathbb{C}^$. Denote by $\mathrm{Cent}(\phi)$ the centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, and by $\mathcal{C}$ the fixed curve of $\phi$. - Let us first assume that any element of $\mathrm{Cent}(\phi)$ preserves the fibration $z_1=$cst fiberwise. Then $\mathrm{Cent}(\phi)=\mathcal{J}_{z_1}$. - Assume now that there exists an element $\psi$ in $\mathrm{Cent}(\phi)$ that does not preserve the fibration $z_1=$cst fiberwise. According to Lemma \[lem:stab\] either $\mathrm{pr}_2(\psi)=\zeta z_1$ with $\zeta$ root of unity, or $\mathrm{pr}_2(\psi)=\frac{\alpha}{z_1}$ with $\alpha$ in $\mathbb{C}^$. If $\mathrm{pr}_2(\psi)=\zeta z_1$ with $\zeta$ root of unity, then $$\frac{4c^2(\zeta z_1)}{c^2(\zeta z_1)-\zeta z_1}=\frac{4c^2(z_1)}{c^2(z_1)-z_1}$$ [i.e.]{} $c^2(\zeta z_1)=\zeta c^2(z_1)$. There exists $\upsilon$ such that $\upsilon^2=\zeta$ and $c(\upsilon^2 z_1)=\upsilon c(z_1)$. Note that $\varphi\colon(z_0,z_1)\mapsto(\upsilon z_0,\upsilon^2z_1)$ belongs to $\mathrm{Cent}(\phi)$. Remark that $\mathrm{pr}_2(\psi\circ\varphi^{-1})=\mathrm{id}$ so $\psi\circ\varphi^{-1}$ belongs to $\mathcal{J}_{z_1}$. If $\mathrm{pr}_2(\psi)=\frac{\alpha}{z_1}$, then $$\frac{4c^2\left(\frac{\alpha}{z_1}\right)}{c^2\left(\frac{\alpha}{z_1}\right)-z_1}=\frac{4c^2(z_1)}{c^2(z_1)-z_1}$$ [i.e.]{} $c^2\left(\frac{\alpha}{z_1}\right)=\frac{\alpha}{z_1^2}c^2(z_1)$. There exists $\beta$ in $\mathbb{C}$ such that $\beta^2=\alpha$ and $c\left(\frac{\beta^2}{z_1}\right)=\frac{\beta}{z_1}c(z_1)$. Remark that the map $(z_0,z_1)\mapsto\left(\frac{\beta z_0}{z_1},\frac{\beta^2}{z_1}\right)$ commutes with $\phi$. The map $\psi\circ\varphi^{-1}$ belongs to $\mathrm{Cent}(\phi)$ and preserves the fibration $z_1=$cst fiberwise; hence $\psi\circ\varphi^{-1}$ belongs to $\mathcal{J}_{z_1}$. We thus have established: \[pro:centrfib2\] The centralizer of a <span style="font-variant:small-caps;">Jonquières</span> twist $\phi$ that preserves fiberwise the fibration in the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is a finite extension of $\mathrm{Ab}(\phi)$. #### Centralizers of elements of $\mathcal{J}\smallsetminus\mathcal{J}_0$ {#subsubsec:centrjonq} The description of the centralizers of elements of $\mathcal{J}_0$ (Proposition \[pro:centrfib2\]) allows to describe, up to finite index, the centralizer of elements of $\mathcal{J}$. Generically these maps have a trivial centralizer ([@CerveauDeserti:centralisateurs]). A consequence of the study of the centralizers of elements of $\mathcal{J}$ is: The centralizer of a <span style="font-variant:small-caps;">Jonquières</span> twist is virtually solvable. ### What about the others ? #### Let $\phi$ be an <span style="font-variant:small-caps;">Halphen</span> twist. Up to birational conjugacy one can assume that $\phi$ is an element of a rational surface $S$ with an elliptic fibration and that this fibration is $\phi$-invariant (§\[sec:degreegrowth\]). Furthermore we can assume that there is no smooth curve of self-intersection $-1$ in the fibers, [i.e.]{} that the fibration is minimal, and so that $\phi$ is an automorphism. The elliptic fibration is the unique $\phi$-invariant fibration ([@DillerFavre]). As a result the fibration is invariant by all elements that commute with $\phi$, and the centralizer of $\phi$ is contained in $\mathrm{Aut}(S)$. Since the fibration is minimal, the surface $S$ is obtained by blowing up the complex projective plane in the nine base-points of an <span style="font-variant:small-caps;">Halphen</span> pencil and the rank of its <span style="font-variant:small-caps;">Néron</span>-<span style="font-variant:small-caps;">Severi</span> group is equal to $10$. The group $\mathrm{Aut}(S)$ can be embedded in the endomorphisms of $H^2(S,\mathbb{Z})$ for the intersection form and preserves the class $[K_S]$ of the canonical divisor, that is the class of the fibration. The dimension of the orthogonal hyperplane to $[K_S]$ is $9$, and the restriction of the intersection form on its hyperplane is semi-negative: its kernel coincides with $\mathbb{Z}[K_S]$. As a consequence $\mathrm{Aut}(S)$ contains an abelian group of finite index with rank $\leq 8$. We can thus state: Let $\phi$ be an <span style="font-variant:small-caps;">Halphen</span> twist. The centralizer of $\phi$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ contains a subgroup of finite index which is abelian, free and of rank $\leq 8$. #### We finish the description of the centralizers of birational maps with the case of loxodromic maps in §\[subsec:centralisateursloxodromic\]. Consequences of the action of the <span style="font-variant:small-caps;">Cremona</span> group on an infinite dimensional hyperbolic space {#chap:hyper} ========================================================================================================================================= As we will see in this chapter one of the main techniques to better understand infinite subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is the construction of the action by isometries of the plane <span style="font-variant:small-caps;">Cremona</span> group on an infinite dimensional hyperbolic space detailed in Chapter \[chap:hyperbolicspace\] and the use of results from hyperbolic geometry and group theory. In the first section we recall results of <span style="font-variant:small-caps;">Demazure</span> and <span style="font-variant:small-caps;">Beauville</span> that suggest that the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group behaves like a rank $2$ group. We give an outline of the proof of the description of the centralizer of a loxodromic element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. On the one hand it finishes the description of the centralizer of the elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, on the other hand it suggests that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ behaves as a group of rank $1$. We end this section by recalling the description of the morphisms from a countable group with <span style="font-variant:small-caps;">Kazhdan</span> property $(T)$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ which also insinuates that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ behaves as a group of rank $1$. In the second section we give an outline of the proofs of the description of elliptic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, [i.e.]{} the subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ whose all elements are elliptic: if $\mathrm{G}$ is such a group, either $\mathrm{G}$ is a bounded subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, or $\mathrm{G}$ is a torsion subgroup ([@Urech:ellipticsubgroups]). It is thus natural to describe torsion subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. In the third section we give an outline of the proof of the fact that if $\mathrm{G}$ is a torsion subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, then $\mathrm{G}$ is isomorphic to a bounded subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$; furthermore it is isomorphic to a subgroup of $\mathrm{GL}(48,\mathbb{C})$. Let us mention the surprising fact that the proof uses model theory as <span style="font-variant:small-caps;">Malcev</span> already did in [@Malcev]. The fourth section deals with <span style="font-variant:small-caps;">Tits</span> alternative and <span style="font-variant:small-caps;">Burnside</span> problem. We recall the Ping Pong Lemma and give a sketch of the proof of the <span style="font-variant:small-caps;">Tits</span> alternative for the <span style="font-variant:small-caps;">Cremona</span> group, [i.e.]{} the proof of Every subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ either is virtually solvable, or contains a non-abelian free group. One consequence is a positive answer to the <span style="font-variant:small-caps;">Burnside</span> problem for the <span style="font-variant:small-caps;">Cremona</span> group: every finitely generated torsion subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is finite. The study of solvable groups is a very old problem. For instance let us recall the <span style="font-variant:small-caps;">Lie</span>-<span style="font-variant:small-caps;">Kolchin</span> theorem: any linear solvable subgroup is up to finite index triangularizable ([@KargapolovMerzljakov]). Note that the assumption “up to finite index” is essential: for instance the subgroup $$\langle\left(\begin{array}{cc} 1 & 0 \\ 1 & -1 \end{array} \right),\,\left(\begin{array}{cc} -1 & 1 \\ 0 & 1 \end{array} \right)\rangle$$ of $\mathrm{PGL}(2,\mathbb{C})$ is isomorphic to $\mathfrak{S}_3$ so is solvable but is not triangularizable. The fifth section dedicated to a sketch of the proof of the characterization of the solvable subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group ([@Deserti:resoluble; @Urech:ellipticsubgroups]). Let us recall a very old question, already asked in $1895$ in [@Enriques2]: [10cm]{} “Tuttavia altre questioni d’indole gruppale relative al gruppo <span style="font-variant:small-caps;">Cremona</span> nel piano (ed a più forte ragione in $S_n$, $n>2$) rimangono ancora insolute; ad esempio l’importante questione se il gruppo <span style="font-variant:small-caps;">Cremona</span> contenga alcun sottogruppo invariante (questione alla quale sembra probabile si debba rispondere negativamente)”. In $2013$ <span style="font-variant:small-caps;">Cantat</span> and <span style="font-variant:small-caps;">Lamy</span> established that $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ is not simple as soon as $\Bbbk$ is algebraically closed ([@CantatLamy]). Then in $2016$ <span style="font-variant:small-caps;">Lonjou</span> proved that $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ is not simple over any field ([@Lonjou]). The sixth section is devoted to normal subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and the non-simplicity of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Strategies of [@CantatLamy] and [@Lonjou] are evoked. A consequence of one result of [@Lonjou] is the following property: the <span style="font-variant:small-caps;">Cremona</span> group contains infinitely many characteristic subgroups ([@Cantat:survey]). Taking the results of the sixth section as a starting point <span style="font-variant:small-caps;">Urech</span> gives a classification of all simple groups that act non-trivially by birational maps on complex compact Kähler surfaces. In particular he gets the two following statements: \[thm:urechsimple1\] A simple group $\mathrm{G}$ acts non-trivially by birational maps on a rational complex projective surface if and only if $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{PGL}(3,\mathbb{C})$. \[thm:urechsimple2\] Let $\mathrm{G}$ be a simple subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Then - $\mathrm{G}$ does not contain loxodromic elements; - if $\mathrm{G}$ contains a parabolic element, then $\mathrm{G}$ is conjugate to a subgroup of $\mathcal{J}$; - if $\mathrm{G}$ is an elliptic subgroup, then $\mathrm{G}$ is either a simple subgroup of an algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, or conjugate to a subgroup of $\mathrm{G}$. In the last section we give a sketch of the proof of these results. A group of rank $1.5$ --------------------- ### Rank $2$ phenomenon Let $\Bbbk$ be a field. Consider a connected semi-simple algebraic group $\mathrm{G}$ defined over $\Bbbk$. Let $\Psi\colon\mathrm{G}\to\mathrm{Aut}(\mathrm{G})$ be the mapping $g\mapsto\Psi_g$ where $\Psi_g$ denotes the inner automorphism given by $$\begin{aligned} & \Psi_g\colon\mathrm{G}\to\mathrm{G}, && h\mapsto ghg^{-1}.\end{aligned}$$ For each $g$ in $\mathrm{G}$ one can define $\mathrm{Ad}_g$ to be the derivative of $\Psi_g$ at the origin $$\mathrm{Ad}_g=(D\Psi_g)_{\mathrm{id}}\colon \mathfrak{g}\to\mathfrak{g}$$ where $D$ is the differential and $\mathfrak{g}=T_{\mathrm{id}}\mathrm{G}$ is the tangent space of $\mathrm{G}$ at the identity element of $\mathrm{G}$. The map $$\begin{aligned} & \mathrm{Ad}\colon\mathrm{G}\to\mathrm{Aut}(\mathfrak{g}), && g\mapsto\mathrm{Ad}_g\end{aligned}$$ is a group representation called the adjoint representation of $\mathrm{G}$. The $\Bbbk$-rank of $\mathrm{G}$ is the maximal dimension of a connected algebraic subgroup of $\mathrm{G}$ which is diagonalizable over $\Bbbk$ in $\mathrm{GL}(\mathfrak{g})$. Such a maximal diagonalizable subgroup is a maximal torus. \[thm:demazure\] Let $\mathbb{G}_m$ be the multiplicative group over $\mathbb{C}$. Let $r$ be an integer. If $\mathbb{G}_m^r$ embeds as an algebraic subgroup in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$, then $r\leq n$. If $r=n$, then the embedding is conjugate to an embedding into the group of diagonal matrices in $\mathrm{PGL}(n+1,\mathbb{C})$. Theorem \[thm:demazure\] not only holds for $\mathbb{C}$ but also for any algebraically closed field $\Bbbk$. In other words the group of diagonal automorphisms $\mathrm{D}_n$ plays the role of a maximal torus in $\mathrm{Bir}(\mathbb{P}^n_\mathbb{C})$ and the <span style="font-variant:small-caps;">Cremona</span> group “looks like” a group of rank $n$. Furthermore <span style="font-variant:small-caps;">Beauville</span> has shown a finite version of Theorem \[thm:demazure\] in dimension $2$: Let $p\geq 5$ be a prime. If the abelian group $\Big(\faktor{\mathbb{Z}}{p\mathbb{Z}}\Big)^r$ embeds into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, then $r\leq 2$. Moreover if $r=2$, then the image of $\Big(\faktor{\mathbb{Z}}{p\mathbb{Z}}\Big)^r$ is conjugate to a subgroup of the group $\mathrm{D}_2$ of diagonal automorphisms of $\mathbb{P}^2_\mathbb{C}$. This statement not only holds for $\mathbb{C}$ but also for any algebraically closed field $\Bbbk$. Let us give an idea of the proof. Consider a finite group $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. It can be realized as a group of automorphisms of a rational surface $S$ (see for instance [@DeFernexEin]). Moreover one can assume that every birational $\mathrm{G}$-equivariant morphism of $S$ onto a smooth surface with a $\mathrm{G}$-action is an isomorphism. Then according to [@Manin:rational] - either $\mathrm{G}$ preserves a fibration $\pi\colon S\to\mathbb{P}^1$ with rational fibers, - or $\mathrm{Pic}(S)^{\mathrm{G}}$ has rank $1$. In the first case $\mathrm{G}$ embeds in the group of automorphisms of the generic fibre $\mathbb{P}^1_{\mathbb{C}(t)}$ of $\pi$ and <span style="font-variant:small-caps;">Beauville</span> classified the $p$-elementary subgroups of $\mathrm{Aut}(\mathbb{P}^1_{\mathbb{C}(t)})$. In the last case $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface and the group $\mathrm{Aut}(S)$ is well known. <span style="font-variant:small-caps;">Beauville</span> also classified the $p$-elementary subgroups of such groups. ### Rank $1$ phenomenon {#subsec:centralisateursloxodromic} Generic elements of degree $\geq 2$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ are loxodromic and hence can not be conjugate to elements of the maximal torus $\mathrm{D}_2$. The description of their centralizer is given by: \[thm:cent\] Let $\phi$ be a loxodromic element of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. The infinite cyclic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ generated by $\phi$ has finite index in the centralizer $$\mathrm{Cent}(\phi)=\big\{\psi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\,\vert\,\psi\circ\phi=\phi\circ\psi\big\}$$ of $\phi$. Theorem \[thm:cent\] holds for any field $\Bbbk$. The centralizer of a generic element of $\mathrm{SL}(n+1,\mathbb{C})$ is isomorphic to $(\mathbb{C}^)^n$; Theorem \[thm:cent\] suggests that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ behaves as a group of rank $1$. If $\psi$ commutes to $\phi$, then the isometry $\psi_$ of $\mathbb{H}^\infty$ preserves the axis $\mathrm{Ax}(\phi)$ of $\phi_$ and its two endpoints. Consider the morphism $\Theta$ which maps $\mathrm{Cent}(\phi)$ to the group of isometries of $\mathrm{Ax}(\phi)$. View it as a morphism into the group of translations $\mathbb{R}$ of the line. On the one hand the translation lengths are bounded from below by $\log(\lambda_L)$ where $\lambda_L$ is the <span style="font-variant:small-caps;">Lehmer</span> number, [i.e.]{} the unique root $>1$ of the irreducible polynomial $x^{10}+x^9-x^7-x^6-x^5-x^4-x^3+x+1$ (see* [@BlancCantat]). On the other hand every discrete subgroup of $\mathbb{R}$ is trivial or cyclic. As a result the image of $\Theta$ is a cyclic group. Its kernel is made of elliptic elements of $\mathrm{Cent}(\phi)$ fixing all points of $\mathrm{Ax}(\phi)$. Denote by $\mathbf{e}_\phi$ the projection of $\mathbf{e}_0$ on $\mathrm{Ax}(\phi)$. Since $\ker\Theta$ fixes $\mathbf{e}_\phi$, the inequality $$\mathrm{dist}(\psi_\mathbf{e}_0,\mathbf{e}_0)\leq 2\mathrm{dist}(\mathbf{e}_0,\mathbf{e}_\phi)$$ holds. As a consequence $\ker\Theta$ is a group of birational maps of bounded degree. From [@BlancFurter] the <span style="font-variant:small-caps;">Zariski</span> closure of $\ker\Theta$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is an algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let us denote by $\mathrm{G}$ the connected component of the identity in this group. Assume that $\ker\Theta$ is infinite. Then $\dim\mathrm{G}$ is positive and $\mathrm{G}$ is contained, after conjugacy, in the group of automorphisms of a minimal, rational surface ([@Blanc:ssgpealg; @Enriques]). Therefore $\mathrm{G}$ contains a <span style="font-variant:small-caps;">Zariski</span> closed abelian subgroup whose orbits have dimension $1$. Those orbits are organised in a pencil of curves that is invariant under the action of $\phi$: contradiction with the fact that $\phi_$ is loxodromic. As a result $\ker\Theta$ is finite. ### Rank $1$ phenomenon {#rank-1-phenomenon} To generalize <span style="font-variant:small-caps;">Margulis</span> work on linear representations of lattices of simple real <span style="font-variant:small-caps;">Lie</span> groups to non-linear representations <span style="font-variant:small-caps;">Zimmer</span> proposed to study the actions of lattices on compact varieties ([@Zimmer1; @Zimmer2; @Zimmer3; @Zimmer4]). One of the main conjectures of the program drawn by <span style="font-variant:small-caps;">Zimmer</span> is: let $\mathrm{G}$ be a connex, simple, real <span style="font-variant:small-caps;">Lie</span> group and let $\Gamma$ be a lattice of $\mathrm{G}$. If there exists a morphism from $\Gamma$ into the diffeomorphisms group of a compact variety $V$ with infinite image, then the real rank of $\mathrm{G}$ is less or equal to the dimension of $V$. In the context of birational self maps one has the following statement that can be see as another rank one phenomenum: \[thm:zimmer\] Let $S$ be a complex projective surface. Let $\Gamma$ be a countable group with <span style="font-variant:small-caps;">Kazhdan</span> property $(T)$. If $\upsilon\colon\Gamma\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is a morphism with infinite image, then $\upsilon$ is conjugate to a morphism into $\mathrm{PGL}(3,\mathbb{C})$. Theorem \[thm:zimmer\] indeed holds for any algebraically closed field $\Bbbk$. The first step is based on a fixed point property: since $\Gamma$ has <span style="font-variant:small-caps;">Kazhdan</span> property (T), then $\upsilon(\Gamma)$ acts by isometries on $\mathbb{H}^\infty$ and $(\upsilon(\Gamma))_$ has a fixed point. Then according to [@delaHarpeValette] all its orbits have bounded diameter. Hence $\rho(\Gamma)$ has bounded degree. There thus exists a birational map $\pi\colon X\dashrightarrow\mathbb{P}^2_\mathbb{C}$ such that - $\Gamma_S=\pi^{-1}\circ\Gamma\circ\pi$ is a subgroup of $\mathrm{Aut}(S)$; - $\mathrm{Aut}(S)^0\cap\Gamma_S$ has finite index in $\Gamma_S$. The classification of algebraic groups of maps of surfaces and the fact that every subgroup of $\mathrm{SL}(2,\mathbb{C})$ having <span style="font-variant:small-caps;">Kazhdan</span> property (T) is finite allow to prove that: since $\mathrm{Aut}(S)^0$ contains an infinite group with <span style="font-variant:small-caps;">Kazhdan</span> property (T) the surface $S$ is isomorphic to the projective plane $\mathbb{P}^2_\mathbb{C}$. Subgroups of elliptic elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ {#sec:ellipticgroup} ------------------------------------------------------------------------- A subgroup $\mathrm{G}$ of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is elliptic* if any element of $\mathrm{G}$ is an elliptic birational map. Let us give an example: a bounded subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is elliptic. But not all elliptic subgroups are bounded; indeed for instance - all elements of $\big\{(z_0,z_1+a(z_0))\,\vert\,a\in\mathbb{C}(z_0)\big\}$ are elliptic but $\big\{(z_0,z_1+a(z_0))\,\vert\,a\in\mathbb{C}(z_0)\big\}$ contains elements of arbitrarily high degrees; - <span style="font-variant:small-caps;">Wright</span> gives examples of subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ isomorphic to a subgroup of roots of unity of $\mathbb{C}^$ that are not bounded ([@Wright:abelian]). Let us be more precise. Set $\psi_0\colon(z_0,z_1)\mapsto(-z_0,-z_1)$ and for any $k\geq 1$ $$\begin{aligned} & \alpha_k=\exp\left(\frac{\mathbf{i}\pi}{2^k}\right), && \phi_k\colon(z_0,z_1)\mapsto(z_1,c_kz_1^{2^k+1}+z_0), && \varphi_k=\phi_k^2\circ \phi_{k-1}^2\circ\ldots\circ \phi_1^2\end{aligned}$$ where $c_k$ denotes an element of $\mathbb{C}^$. Consider $$\psi_k=\varphi_k^{-1}\circ\big((z_0,z_1)\mapsto(\alpha_kz_0,\alpha_k^pz_1)\big)\circ\varphi_k$$ where $p$ is an odd integer. The group $$\mathrm{G}=\displaystyle\bigcup_{k\geq 0}\langle \psi_k\rangle$$ is an abelian group obtained as a growing union of finite cyclic groups that does not preserve any fibration ([@Lamy:dynamique]). This gives all the possibilities for elliptic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$: \[thm:urechell1\] Let $\mathrm{G}$ be an elliptic subgroup of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. Then one of the following holds: - $\mathrm{G}$ is a bounded subgroup; - $\mathrm{G}$ preserves a rational fibration; - $\mathrm{G}$ is a torsion group. Furthermore he characterizes torsion subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$: \[thm:urechell2\] Let $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ be a torsion group. Then $\mathrm{G}$ is isomorphic to a bounded subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Furthermore $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{GL}(48,\mathbb{C})$. As a consequence he gets an analogue of the Theorem of <span style="font-variant:small-caps;">Jordan</span> and <span style="font-variant:small-caps;">Schur</span>: There exists a constant $\gamma$ such that every torsion subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ contains a commutative normal subgroup of index $\leq \gamma$. Theorems \[thm:urechell1\] and \[thm:urechell2\] allow to refine - the result of <span style="font-variant:small-caps;">Cantat</span> about <span style="font-variant:small-caps;">Tits</span> alternative (§\[sec:tits\]); - the description of solvable subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ (see §\[sec:solvable\]). The aim of the section is to prove Theorem \[thm:urechell1\]. We need the following technical lemmas. \[lem:2fibr\] Let $\phi$ be a birational self map of the complex projective plane that fixes pointwise two different rational fibrations. Then $\phi$ is of finite order. The intersections of the generic fibres of these two fibrations are finite, uniformly bounded. But these intersections are invariant by $\phi$ so $\phi$ is of finite order. \[lem:Urechmoinsde9\] An algebraic subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of dimension $\leq 9$ preserves a unique rational fibration. According to Theorem \[thm:blanc11cases\] the group $\mathrm{G}$ is conjugate to a subgroup of $\mathrm{Aut}(\mathbb{F}_n)$ for some <span style="font-variant:small-caps;">Hirzebruch</span> surface $\mathbb{F}_n$, $n\geq 2$. As a consequence $\mathrm{G}$ preserves a rational fibration $\pi\colon\mathbb{F}_n\to\mathbb{P}^1_\mathbb{C}$. The fibres of $\pi$ are permuted by $\mathrm{G}$, this yields to a homomorphism $$f\colon\mathrm{G}\to\mathrm{PGL}(2,\mathbb{C})$$ such that $\dim\ker f\geq 6$. Assume by contradiction that there exists a second rational fibration $\pi'\colon\mathbb{F}_n\to\mathbb{P}^1_\mathbb{C}$ preserved by $\mathrm{G}$; this yields to a second homomorphism $$g\colon\mathrm{G}\to\mathrm{PGL}(2,\mathbb{C}).$$ One has $\dim\ker\pi'_{\vert\ker\pi}>0$; therefore $\dim(\ker f\cap\ker g)>0$. In particular $\ker f\cap\ker g$ contains an element of infinite order: contradiction with Lemma \[lem:2fibr\]. \[lem:algcst\] Let $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ be an algebraic subgroup isomorphic as an algebraic group to $\mathbb{C}^$. There exists a constant $K(\mathrm{G})$ such that any elliptic element of $$\mathrm{Cent}(\mathrm{G})=\big\{\varphi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\,\vert\,\varphi\circ\psi=\psi\circ\varphi\quad\forall\,\psi\in\mathrm{G}\big\}$$ has degree $\leq K(\mathrm{G})$. Up to conjugacy by an element $\psi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ one can assume that $$\mathrm{G}=\big\{(z_0,z_1)\mapsto(\alpha z_0,z_1)\,\vert\,\alpha\in\mathbb{C}^\big\}.$$ An elliptic element of $\mathrm{Cent}(\mathrm{G})$ is of the following form $$\varphi\colon(z_0,z_1)\dashrightarrow(z_0\varphi_1(z_1),\varphi_2(z_1))$$ where $\varphi_1$, $\varphi_2$ are rational functions. Since $(\deg\varphi^n)_n$ is bounded, $\varphi_1$ is constant and so $\varphi_2\colon z_1\mapsto\frac{az_1+b}{cz_1+d}$ for some matrix $\left( \begin{array}{cc} a & b \\ c & d \end{array} \right)$ of $\mathrm{PGL}(2,\mathbb{C})$. In particular $\deg\varphi\leq 2$. The constant $K(\mathrm{G})$ thus only depends on the degree of $\psi$. \[lem:Urechmonomial\] A group $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of monomial elliptic elements is bounded. The group $\mathrm{G}$ is contained in $\mathrm{GL}(2,\mathbb{Z})\ltimes(\mathbb{C}^)^2$. Consider the projection $\pi\colon\mathrm{G}\to\mathrm{GL}(2,\mathbb{Z})$. On the one hand $\ker\pi$ is bounded, on the other hand all elements of $\pi(\mathrm{G})$ are bounded. All elliptic elements in $\mathrm{GL}(2,\mathbb{Z})\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ are of finite order so $\pi(\mathrm{G})$ is a torsion subgroup of $\mathrm{GL}(2,\mathbb{Z})$. Since there are only finitely many conjugacy classes of finite subgroups in $\mathrm{GL}(2,\mathbb{Z})$ the group $\pi(\mathrm{G})$ is finite. Therefore $\mathrm{G}$ is a finite extension of a bounded subgroup hence $\mathrm{G}$ is bounded. \[lem:Urechsemisimple\] Let $\mathrm{H}$ be a semi-simple algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ be a group of elliptic elements that normalizes $\mathrm{H}$. Then $\mathrm{G}$ is bounded. The group $\mathrm{H}$ is semi-simple; in particular its group of inner automorphisms has finite index in its group of algebraic automorphisms. As a result there exists $N\in\mathbb{Z}$ such that for any $\phi$ in $\mathrm{G}$ conjugation by $\phi^N$ induces an inner automorphism of $\mathrm{H}$. There exists thus an element $\psi$ in $\mathrm{H}$ such that $\phi^N\circ\psi$ centralizes $\mathrm{H}$. By assumption $\mathrm{H}$ is semi-simple so $\mathrm{H}$ contains a closed subgroup $\mathrm{D}$ isomorphic as an algebraic group to $\mathbb{C}^$ and this group is centralized by $\phi^N\circ\psi$. From Lemma \[lem:algcst\] we get that $\deg(\phi^N\circ\psi)$ is bounded by a constant that depends neither on $\phi$, nor on $N$. As $\mathrm{H}$ is an algebraic group both $\deg \psi$ and $\deg \phi$ are also bounded independently of $\phi$ and $N$. Finally $\mathrm{G}$ is bounded. \[lem:Urechtecn1\] Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that fixes a point of $\mathbb{H}^\infty$. Then - the degree of all elements in $\mathrm{G}$ is uniformly bounded; - there exist a smooth projective surface $S$ and a birational map $\varphi\colon\mathbb{P}^2_\mathbb{C}\dashrightarrow S$ such that $\varphi\circ\mathrm{G}\circ\varphi^{-1}\subset\mathrm{Aut}(S)$. Denote by $p\in\mathbb{H}^\infty$ the fixed point of $\mathrm{G}$, and by $\mathbf{e}_0\in\mathbb{H}^\infty$ the class of a line in $\mathbb{P}^2_\mathbb{C}$. Take an element $\psi$ of $\mathrm{G}$. The action of $\mathrm{G}$ on $\mathbb{H}^\infty$ is isometric hence $d(\psi(\mathbf{e}_0),p)=d(\mathbf{e}_0,p)$ and so $d(\psi(\mathbf{e}_0),p)\leq 2d(\mathbf{e}_0,p)$. This implies $$\langle\psi(\mathbf{e}_0),\mathbf{e}_0\rangle\leq \cosh(2d(\mathbf{e}_0,p))\qquad\forall\,\psi\in\mathrm{G}.$$ Since $\langle\psi(\mathbf{e}_0),\mathbf{e}_0\rangle=\deg\psi$ the previous inequality can be rewritten as follows $$\deg\psi\leq \cosh(2d(\mathbf{e}_0,p))\qquad\forall\,\psi\in\mathrm{G},$$ [i.e.]{} the degrees of all elements in $\mathrm{G}$ are uniformly bounded. According to <span style="font-variant:small-caps;">Weil</span> $\mathrm{G}$ can be regularized (§\[sec:WeilKraft\]). Let us recall the following statement due to <span style="font-variant:small-caps;">Cantat</span>: \[pro:etaumilieuCantat\] Let $\Gamma$ be a finitely generated subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of elliptic elements. Then - either $\Gamma$ is bounded, - or $\Gamma$ preserves a rational fibration, [i.e.]{} $\Gamma\subset\mathcal{J}$ up to birational conjugacy. \[lem:Urechtecalt\] Let $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ be a group of elliptic elements. Then one of the following holds: - $\mathrm{G}$ preserves a fibration and so up to birational conjugacy either $\mathrm{G}\subset\mathcal{J}$, or $\mathrm{G}\subset\mathrm{Aut}(S)$ where $S$ is a <span style="font-variant:small-caps;">Halphen</span> surface. - every finitely generated subgroup of $\mathrm{G}$ is bounded. Furthermore if $\mathrm{G}$ fixes a point $p\in\partial\mathbb{H}^\infty$ that does not correspond to the class of a rational fibration, then the second assertion holds. The group $\mathrm{G}$ fixes a point $p\in\mathbb{H}^\infty\cup\partial\mathbb{H}^\infty$ (Theorem \[thm:weak\]). If $p$ belongs to $\mathbb{H}^\infty$, then $\mathrm{G}$ is bounded. Let us now assume that $p$ belongs to $\partial\mathbb{H}^\infty$. Then either $p$ corresponds to the class of a general fibre of some fibration, or not. - If $p$ corresponds to the class of a general fibre of some fibration $\pi\colon Y\to\mathbb{P}^1_\mathbb{C}$ where $Y$ is a rational surface, then $\mathrm{G}$ preserves this fibration and is thus conjugate to a subgroup of $\mathcal{J}$ (if the fibration is rational), or to a subgroup of $\mathrm{Aut}(S)$ where $S$ is a <span style="font-variant:small-caps;">Halphen</span> surface (if the fibration consists of curves of genus $1$). - Suppose now that $p$ does not correspond to the class of a fibration. Let $\Gamma$ be a finitely generated subgroup of $\mathrm{G}$. Then either $\Gamma$ is bounded, or $\Gamma$ preserves a rational fibration (Proposition \[pro:etaumilieuCantat\]). If $\Gamma$ preserves a rational fibration $\mathcal{F}$, then $\Gamma$ fixes a point $q\in\partial\mathbb{H}^\infty$ that corresponds to the class of $\mathcal{F}$. Hence $p$ and $q$ are two distinct points preserved by $\mathrm{G}$ and $\mathrm{G}$ fixes the geodesic line through $p$ and $q$. In particular $\mathrm{G}$ fixes a point in $\mathbb{H}^\infty$ and according to Lemma \[lem:Urechtecn1\] the degrees of all elements in $\mathrm{G}$ are uniformly bounded. Consider a subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of elliptic elements. According to Lemma \[lem:Urechtecn1\] either $\mathrm{G}$ preserves a rational fibration, or any finitely generated subgroup of $\mathrm{G}$ is bounded. Assume that any finitely generated subgroup of $\mathrm{G}$ is bounded. Set $$n:=\sup\big\{\dim\overline{\Gamma}\,\vert\,\Gamma\subset\mathrm{G}\text{ finitely generated}\big\}.$$ - If $n=0$, then $\mathrm{G}$ is a torsion group. - If $n=+\infty$, then take $\Gamma$ a finitely generated subgroup of $\mathrm{G}$ such that $\dim\overline{\Gamma}\geq 9$. By Lemma \[lem:Urechmoinsde9\] the group $\Gamma$ preserves a unique fibration and this fibration is, again by Lemma \[lem:Urechmoinsde9\], preserved as well by $\langle\Gamma,\,\phi\rangle$ for any $\phi$ in $\mathrm{G}$. - Assume now $n\in\mathbb{N}^$. Let $\Gamma$ be a finitely generated subgroup of $\mathrm{G}$ such that $\dim\overline{\Gamma}~=~n$. Let $\overline{\Gamma}^0$ be the neutral component of $\Gamma$. For any $\varphi\in\mathrm{G}$ the group $\langle\overline{\Gamma}^0,\varphi\circ\overline{\Gamma}^0\circ\varphi^{-1}\rangle$ is connected and contained in $\langle\overline{\Gamma,\varphi\circ\Gamma\circ\varphi^{-1}}\rangle$ which is finitely generated and thus of dimension less or equal to $n$. As a consequence $\langle\overline{\Gamma}^0,\varphi\circ\overline{\Gamma}^0\circ\varphi^{-1}\rangle=\overline{\Gamma}^0$ for any $\varphi\in\mathrm{G}$ and $\overline{\Gamma}^0$ is normalized by $\mathrm{G}$. If $\overline{\Gamma}^0$ is semi-simple, Lemma \[lem:Urechsemisimple\] allows to conclude. Assume that $\overline{\Gamma}^0$ is not semi-simple. Denote by $R$ the radical of $\overline{\Gamma}^0$, [i.e.]{} $R$ is the maximal connected normal solvable subgroup of $\overline{\Gamma}^0$. Since $\overline{\Gamma}^0$ is semi-simple the inequality $\dim R>0$ holds. The radical is unique hence preserved by $\mathrm{Aut}(\overline{\Gamma}^0)$ and in particular normalized by $\mathrm{G}$. Denote by $$R^{(\ell+1)}=\big\{\mathrm{id}\big\}\subsetneq R^{(\ell)}\subset\ldots\subset R^{(2)}\subset R^{(1)}\subset R^{(0)}=R$$ the derived series of $R$ ([i.e.]{} $R^{(k+1)}=[R^{(k)},R^{(k)}]$). Note that $\dim R^{(\ell)}>0$ and $R^{(\ell)}$ is abelian. This series is invariant under $\mathrm{Aut}(\overline{\Gamma}^0)$ and so invariant under conjugation by elements of $\mathrm{G}$. In particular $\mathrm{G}$ normalizes $R^{(\ell)}$. Since $R^{(\ell)}$ is bounded, $R^{(\ell)}$ is conjugate to one of the groups of Theorem \[thm:blanc11cases\]; in particular $R^{(\ell)}$ can be regularized. In other words, up to birational conjugacy, $\mathrm{G}$ is a subgroup of $\mathrm{Bir}(S)$ for some smooth projective surface $S$ on which $R^{(\ell)}$ acts regularly. If all the orbits of $R^{(\ell)}$ have dimension $\leq 1$, then $\mathrm{G}$ preserves a rational fibration. Assume that $R^{(\ell)}$ has an open orbit $\mathcal{O}$. The group $\mathrm{G}$ normalizes $R^{(\ell)}$ so $\mathrm{G}$ acts regularly on $\mathcal{O}$. The action of $R^{(\ell)}$ is faithful; as a result $\dim R^{(\ell)}=1$ and $R^{(\ell)}\simeq \mathbb{C}^2$, or $R^{(\ell)}\simeq\mathbb{C}^\times\mathbb{C}$, or $R^{(\ell)}\simeq\mathbb{C}^\times\mathbb{C}^$. If $R^{(\ell)}\simeq\mathbb{C}^2$, then $\mathcal{O}$ is isomorphic to the affine plane, and the action of $R^{(\ell)}$ on $\mathcal{O}$ is given by translations. But the normalizer of $\mathbb{C}^2$ in $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ is the group of affine maps $\mathrm{GL}(2,\mathbb{C})\ltimes\mathbb{C}^2$ hence $\mathrm{G}$ is bounded. If $R^{(\ell)}\simeq\mathbb{C}^\times\mathbb{C}$, then we similarly get the inclusion $\mathrm{G}\subset\mathrm{Aut}(\mathbb{C}^\times\mathbb{C})$. The $\mathbb{C}$-fibration of $\mathbb{C}^\times\mathbb{C}$ is given by the invertible functions; it is thus preserved by $\mathrm{Aut}(\mathbb{C}^\times\mathbb{C})$. In particular $\mathrm{G}$ preserves a rational fibration. If $R^{(\ell)}\simeq\mathbb{C}^\times\mathbb{C}^$, then elements of $\mathrm{G}$ are monomial maps, and Lemma \[lem:Urechmonomial\] allows to conclude. Torsion subgroups ----------------- As we have seen at the beginning of §\[sec:ellipticgroup\] some torsion groups can be embedded into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ in such a way that they neither are bounded, nor preserve any fibration. However the group structure of torsion subgroups can be specified: \[thm:urechtorsion\] A torsion subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is isomorphic to a bounded subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Furthermore $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{GL}(48,\mathbb{C})$. <span style="font-variant:small-caps;">Malcev</span> used model theory to prove that if for a given group $\mathrm{G}$ every finitely generated subgroup can be embedded into $\mathrm{GL}(n,\Bbbk)$ for some field $\Bbbk$, then there exists a field $\Bbbk'$ such that $\mathrm{G}$ can be embedded into $\mathrm{GL}(n,\Bbbk')$. Let us briefly introduce the compactness theory from model theory; it states that a set of first order sentences has a model if and only if any of its finite subsets has a model. Let $\{x_i\}_{i\in I}$ be a set of variables. A condition is an expression of the form $$P(x_{i_1},x_{i_2},\ldots,x_{i_k})=0$$ or an expression of the form $$\big(P_1(x_{i_1},x_{i_2},\ldots,x_{i_k})\not=0\big) \vee \big(P_2(x_{i_1},x_{i_2},\ldots,x_{i_k})\not=0\big)\vee \ldots\vee \big(P_\ell(x_{i_1},x_{i_2},\ldots,x_{i_k})\not=0\big)$$ where $P$ and the $P_i$’s are polynomials with integer coefficients. A mixed system is a set of conditions. A mixed system $S$ is compatible if there exists a field $\Bbbk$ which contains values $\{y_i\}_{i\in I}$ that satisfy $S$. \[thm:Malcev\] If every finite subset of a mixed system $S$ is compatible, then $S$ is compatible. Let us now explain the proof of Theorem \[thm:urechtorsion\]. Let $\mathrm{G}$ be a torsion subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. If $\mathrm{G}$ is finite, then $\mathrm{G}$ is bounded; we can thus assume that $\mathrm{G}$ is infinite. Following Theorem \[thm:blanc11cases\] we will deal with different cases. - First assume that every finitely generated subgroup of $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{PGL}(3,\mathbb{C})$. Consider the closed embedding $\rho$ of $\mathrm{PGL}(3,\mathbb{C})$ into $\mathrm{GL}(8,\mathbb{C})$ given by the adjoint representation. Let $P_1$, $P_2$, $\ldots$, $P_n$ be polynomials in the set of variables $\{x_{ij}\}_{1\leq i,\,j\leq 8}$ such that $\rho(\mathrm{PGL}(3,\mathbb{C}))\subset\mathrm{GL}(8,\mathbb{C})$ is the zero set of $P_1$, $P_2$, $\ldots$, $P_n$. To any element $g\in\mathrm{G}$ we associate a $8\times 8$ matrix of variables $(x_{ij}^g)$. Consider the following mixed system $S$ defined by 1. the equations $(x_{ij}^f)(x_{ij}^g)=(x_{ij}^h)$ for all $f$, $g$, $h\in\mathrm{G}$ such that $f\circ g=h$; 2. the conditions $\big(\displaystyle\bigvee_i x_{ii}^g-1\not=0\big)\vee\big(\displaystyle\bigvee_{i\not=j} x_{ij}^g-1\not=0\big)$; 3. $x_{ii}^{\mathrm{id}}-1=0$ and $x_{ij}^{\mathrm{id}}=0$ for all $1\leq i\not=j\leq N$; 4. $P_k(\{x_{ij}\})=0$ for all $1\leq k\leq n$, for all $g\in\mathrm{G}$; 5. $p\not=0$ for all $p\in\mathbb{Z}^+$ primes. The system $S$ is compatible. According to Theorem \[thm:Malcev\] it suffices to show that every finite subset of $S$ is compatible. Let $c_1$, $c_2$, $\ldots$, $c_n\in S$ be finitely many conditions. Only finitely many variables $x_{ij}^g$ appear in $c_1$, $c_2$, $\ldots$, $c_n$. Let $\big\{g_1,\,g_2,\,\ldots,\,g_\ell\big\}\subset\mathrm{G}$ be the finite set of all elements $g\in\mathrm{G}$ such that for some $1\leq i,\,j\leq 8$ the variable $x_{ij}^g$ appears in one of the conditions $c_1$, $c_2$, $\ldots$, $c_n$. Consider the finitely generated subgroup $\Gamma=\langle g_1,\,g_2,\,\ldots,\,g_\ell\rangle$ of $\mathrm{G}$. By Theorem \[thm:burnside\] the group $\Gamma$ is finite. Therefore by assumption $\Gamma$ has a faithful representation to $\mathrm{PGL}(3,\mathbb{C})$. This representation implies that $\mathbb{C}$ contains values that satisfy the conditions $c_1$, $c_2$, $\ldots$, $c_n$. In other words $S$ is compatible. As a result there exists a field $\Bbbk$ such that $\Bbbk$ contains values $y_{ij}^g$ for all $1\leq i,\,j\leq 8$ and all $g\in\mathrm{G}$ satisfying conditions $(1)$ to $(5)$. Condition $(5)$ asserts that the characteristic of $\Bbbk$ is $0$. The group $\mathrm{G}$ has at most the cardinality of the continuum since $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$; the values $\{y_{ij}^g\}$ are thus contained in a subfield $\Bbbk'$ of $\Bbbk$ that has the same cardinality as $\mathbb{C}$. Hence $\Bbbk'$ can be embedded into $\mathbb{C}$ as a subfield. Hence we may suppose that $\Bbbk=\mathbb{C}$. Consider the map $$\begin{aligned} &\varphi\colon\mathrm{G}\to\mathrm{PGL}(3,\mathbb{C}), && g\mapsto (y_{ij}^g)_{i,j}.\end{aligned}$$ Note that - conditions $(1)$ imply that the image of any element of $\mathrm{G}$ is an invertible matrix and that $\varphi$ is a group automorphism; - conditions $(2)$ lead that this automorphism is injective; - conditions $(3)$ imply $\varphi(\mathrm{id})=\mathrm{id}$; - conditions $(4)$ lead that $\varphi(\mathrm{G})\subset\mathrm{PGL}(3,\mathbb{C})\subset\mathrm{GL}(8,\mathbb{C})$. - Denote by $S_6$ the <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $6$. If any finitely generated subgroup of $\mathrm{G}$ can be embedded into $\mathrm{Aut}(S_6)\simeq\mathrm{D}_2\rtimes(\faktor{\mathbb{Z}}{2\mathbb{Z}}\times\mathfrak{S}_3)$ a similar reasoning leads to: $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{Aut}(S_6)$. - If any finitely generated subgroup of $\mathrm{G}$ can be embedded into $$\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})\simeq\big(\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})\big)\rtimes\faktor{\mathbb{Z}}{2\mathbb{Z}},$$ then $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$. - If any finitely generated subgroup of $\mathrm{G}$ can be embedded into $$\mathrm{Aut}(\mathbb{F}_{2n})\simeq\mathbb{C}[z_0,z_1]_{2n}\rtimes\faktor{\mathrm{GL}(2,\mathbb{C})}{\mu_{2n}}$$ for some $n>0$ (and not necessarily the same for all finitely generated subgroups of $\mathrm{G}$), then $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{GL}(2,\mathbb{C})$ and thus can be embedded in $\mathrm{PGL}(3,\mathbb{C})$. - It remains to consider the case where $\mathrm{G}$ contains - a finitely generated subgroup $\Gamma_1$ that can not be embedded into $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, - a finitely generated subgroup $\Gamma_2$ that can not be embedded into $\mathrm{Aut}(S_6)$, - a finitely generated subgroup $\Gamma_3$ that can not be embedded into $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$, - a finitely generated subgroup $\Gamma_4$ that can not be embedded into $\mathrm{Aut}(\mathbb{F}_{2n})$ for all $n>0$. The finitely generated subgroup $\Gamma=\langle\Gamma_1,\,\Gamma_2,\,\Gamma_3,\,\Gamma_4\rangle$ is not isomorphic to any subgroup of infinite automorphisms group of a <span style="font-variant:small-caps;">del Pezzo</span> surface. Adding finitely many elements if needed we may assume that $\Gamma$ has order $>648$; as a consequence $\Gamma$ is isomorphic neither to any subgroup of an automorphisms group of a <span style="font-variant:small-caps;">del Pezzo</span> surface (Theorem \[thm:648\]), nor to a subgroup of $\mathrm{Aut}(\mathbb{F}_{2n})$ for all $n>0$. Consider a finitely generated subgroup $\mathrm{H}$ of $\mathrm{G}$. The finitely generated subgroup $\langle\Gamma,\,\mathrm{H}\rangle$, and in particular $\mathrm{H}$, is isomorphic to a subgroup of (Theorem \[thm:blanc11cases\]) - either $\mathrm{Aut}(S,\pi)$ where $\pi\colon S\to\mathbb{P}^1_\mathbb{C}$ is an exceptional conic bundle, - or $\mathrm{Aut}(S,\pi)$ where $(S,\pi)$ is a $(\faktor{\mathbb{Z}}{2\mathbb{Z}})^2$-conic bundle and $S$ is not a <span style="font-variant:small-caps;">del Pezzo</span> surface, - or $\mathrm{Aut}(\mathbb{F}_{2n+1})$ for some $n>0$. According to Lemmas \[lem:ru1\], \[lem:ru2\] and \[lem:ru3\] the group $\mathrm{H}$ is isomorphic to a subgroup of $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$. Therefore every finitely generated subgroup of $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$. The group $\mathrm{G}$ is thus isomorphic to a subgroup of $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$ (Theorem \[thm:Malcev\]) and hence to a subgroup of $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$. Every torsion subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is isomorphic to a subgroup of $\mathrm{GL}(48,\mathbb{C})$. Let $\mathrm{G}$ be a torsion group of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. - Assume that $\mathrm{G}$ is infinite. As we just see $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, $\mathrm{Aut}(\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C})$, $\mathrm{Aut}(S_6)$ or $\mathrm{Aut}(\mathbb{F}_n)$ for some $n\geq 2$. According to the structure of $\mathrm{Aut}(\mathbb{F}_n)$ and Lemma \[lem:ru1\] all torsion subgroups of $\mathrm{Aut}(\mathbb{F}_n)$ are isomorphic to a subgroup of $\mathrm{GL}(2,\mathbb{C})$ or $\mathrm{PGL}(2,\mathbb{C})\times\mathbb{C}^$. But $\mathrm{PGL}(2,\mathbb{C})$ can be embedded into $\mathrm{GL}(3,\mathbb{C})$ and $\mathrm{PGL}(3,\mathbb{C})$ into $\mathrm{GL}(8,\mathbb{C})$, and $\mathrm{Aut}(S_6)$ into $\mathrm{GL}(6,\mathbb{C})$ (Lemma \[lem:te2\]); the group $\mathrm{G}$ is thus isomorphic to a subgroup of $\mathrm{GL}(8,\mathbb{C})$. - Suppose that $\mathrm{G}$ is finite and not contained in an infinite bounded subgroup. Then $\mathrm{G}$ is contained in the automorphism group (Theorem \[thm:blanc11cases\]) - either of a <span style="font-variant:small-caps;">del Pezzo</span> surface, - or of an exceptional fibration, - or of a $(\faktor{\mathbb{Z}}{2\mathbb{Z}})^2$-fibration. In the first case we get from Lemma \[lem:ru4\] that $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{GL}(8,\mathbb{C})$. In the second case the group $\mathrm{G}$ can be embedded into $\mathrm{PGL}(2,\mathbb{C})\times\mathrm{PGL}(2,\mathbb{C})$ (Lemma \[lem:ru2\]). In the last case $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{GL}(48,\mathbb{C})$ according to [@Urech:phd Lemma 6.2.12]. <span style="font-variant:small-caps;">Tits</span> alternative and <span style="font-variant:small-caps;">Burnside</span> problem {#sec:tits} --------------------------------------------------------------------------------------------------------------------------------- A group $\mathrm{G}$ is virtually solvable* if $\mathrm{G}$ contains a finite index solvable subgroup. A group $\mathrm{G}$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative if every subgroup of $\mathrm{G}$ either is virtually solvable or contains a non-abelian free subgroup. A group $\mathrm{G}$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative for finitely generated subgroups if every finitely generated subgroup of $\mathrm{G}$ either is virtually solvable or contains a non-abelian free subgroup. <span style="font-variant:small-caps;">Tits</span> showed that linear groups over fields of characteristic zero satisfy the <span style="font-variant:small-caps;">Tits</span> alternative and that linear groups over fields of positive characteristic satisfy the <span style="font-variant:small-caps;">Tits</span> alternative for finitely generated subgroups ([@Tits]). Other well-known examples of groups that satisfy <span style="font-variant:small-caps;">Tits</span> alternative include mapping class groups of surfaces ([@Ivanov]), the outer automorphisms group of the free group of finite rank $n$ ([@BFH]), or hyperbolic groups in the sense of <span style="font-variant:small-caps;">Gromov</span> ([@Gromov:hypgroups]). <span style="font-variant:small-caps;">Lamy</span> studied the group $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$; in particular using its amalgamated product structure he showed that <span style="font-variant:small-caps;">Tits</span> alternative holds for $\mathrm{Aut}(\mathbb{A}^2_\mathbb{C})$ (see [@Lamy]). In [@Cantat:annals] <span style="font-variant:small-caps;">Cantat</span> established that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative for finitely generated subgroups. Then <span style="font-variant:small-caps;">Urech</span> proved that $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative ([@Urech:simplesubgroups]). On the contrary the group of $\mathcal{C}^\infty$-diffeomorphisms of the circle does not satisfy <span style="font-variant:small-caps;">Tits</span> alternative ([@BrinSquier; @GhysSergiescu]). Note that since solvable subgroups have either polynomial or exponential growth, if $G$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative, $G$ does not contain groups with intermediate growth. The main technique to prove that a group contains a non-abelian free group is the ping-pong Lemma (for instance [@delaHarpe]): \[lem:pingpong\] Let $S$ be a set. Let $g_1$ and $g_2$ be two bijections of $S$. Assume that $S$ contains two non-empty disjoint subsets $S_1$ and $S_2$ such that $$\begin{aligned} & g_1^m(S_2)\subset S_1&& g_2^m(S_1)\subset S_2&& \qquad\forall\,m\in\mathbb{Z}\smallsetminus\{0\}. \end{aligned}$$ Then $\langle g_1,\,g_2\rangle$ is a free group on two generators. Let $w=w(a,b)$ be a reduced word that represents a non-trivial element in the free group $\mathbb{F}_2=\langle a,\,b\rangle$. Let us prove that $w(g_1,g_2)$ is a non-trivial map of $S$. Up to conjugacy by a power of $g_1$ assume that $w(g_1,g_2)$ starts and ends with a power of $g_1$: $$w(g_1,g_2)=g_1^{\ell_n}g_2^{m_n}\ldots g_2^{m_1}g_1^{\ell_0}.$$ One checks that $g_1^{\ell_0}$ maps $S_2$ into $S_1$, then $g_2^{m_1}g_1^{\ell_0}$ maps $S_2$ into $S_2$, $\ldots$ and $w$ maps $S_2$ into $S_1$. As $S_2$ is disjoint from $S_1$ one gets that $w(g_1,g_2)$ is non-trivial. Consider a group $\Gamma$ that acts on a hyperbolic space $\mathbb{H}^\infty$ and that contains two loxodromic isometries $\psi_1$ and $\psi_2$ whose fixed points in $\partial\mathbb{H}^\infty$ form two disjoint pairs. Let us take disjoint neighborhoods $S_i\subset\overline{\mathbb{H}^\infty}$ of the fixed point sets of $\psi_i$, $i=1$, $2$. Then Lemma \[lem:pingpong\] applied to sufficiently high powers $\psi_1^n$ and $\psi_2^n$ of $\psi_1$ and $\psi_2$ respectively produces a free subgroup of $\Gamma$. This strategy can be used for the <span style="font-variant:small-caps;">Cremona</span> group acting by isometries on $\mathbb{H}^\infty(\mathbb{P}^2_\mathbb{C})$. More precisely <span style="font-variant:small-caps;">Cantat</span> obtained the following result: \[thm:cantattits\] Let $S$ be a projective surface $S$ over a field $\Bbbk$. The group $\mathrm{Bir}(S)$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative for finitely generated subgroups. Then <span style="font-variant:small-caps;">Urech</span> proves: \[thm:urechtits\] Let $S$ be a complex <span style="font-variant:small-caps;">Kähler</span> surface. Then $\mathrm{Bir}(S)$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative. Let us now give a sketch of the proof of this result in the case $S=\mathbb{P}^2_\mathbb{C}$. ### Subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that contain a loxodromic element Recall that - the subgroup of diagonal automorphisms $$\mathrm{D}_2=\big\{(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)\,\vert\,\alpha,\,\beta\in\mathbb{C}^\big\}\subset\mathrm{PGL}(3,\mathbb{C})=\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$$ is a real torus of rank $2$; - a matrix $A=(a_{ij})\in\mathrm{GL}(2,\mathbb{Z})$ determines a birational map of $\mathbb{P}^2_\mathbb{C}$ $$(z_0,z_1)\dashrightarrow\big(z_0^{a_{00}}z_1^{a_{01}},z_0^{a_{10}}z_1^{a_{11}}\big)$$ The normalizer of $\mathrm{D}_2$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is the semidirect product $$\mathrm{Norm}\big(\mathrm{D}_2,\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\big)=\big\{\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})\,\vert\,\phi\circ\mathrm{D}_2\circ\phi^{-1}=\mathrm{D}_2\big\}=\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2.$$ If $M\in\mathrm{GL}(2,\mathbb{Z})$ has spectral radius strictly larger than $1$, the associated birational map is loxodromic. In particular there exist loxodromic elements that normalize an infinite elliptic subgroup. Up to conjugacy these are the only examples with this property: \[thm:DelzantPy\] Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ containing at least one loxodromic element. Assume that there exists a short exact sequence $$1\longrightarrow \mathrm{A}\longrightarrow \mathrm{G}\longrightarrow\mathrm{B}\longrightarrow 1$$ where $\mathrm{A}$ is infinite and of bounded degree. Then $\mathrm{G}$ is conjugate to a subgroup of $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. <span style="font-variant:small-caps;">Urech</span> generalizes this result to the case where $\mathrm{A}$ is an infinite group of elliptic elements ([@Urech:ellipticsubgroups]): \[thm:urechnorm\] Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ containing at least one loxodromic element. Suppose that there exists a short sequence $$1\longrightarrow \mathrm{A}\longrightarrow \mathrm{G}\longrightarrow\mathrm{B}\longrightarrow 1$$ where $\mathrm{A}$ is an infinite group of elliptic elements. Then $\mathrm{G}$ is conjugate to a subgroup of $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. In order to give the proof of Theorem \[thm:urechnorm\] we need to establish some results. \[lem:densezar\] Let $\phi$ be a loxodromic monomial map of the complex projective plane. Let $\Delta_2$ be an infinite subgroup of $\mathrm{D}_2$ normalized by $\phi$. Then $\Delta_2$ is dense in $\mathrm{D}_2$ with respect to the <span style="font-variant:small-caps;">Zariski</span> topology. Denote by $\overline{\Delta_2}^{\,0}$ the neutral component of the <span style="font-variant:small-caps;">Zariski</span> closure of $\Delta_2$. If $\overline{\Delta_2}^{\,0}$ has a dense orbit on $\mathbb{P}^2_\mathbb{C}$, then $\Delta_2$ is dense in $\mathrm{D}_2$. Otherwise the dimension of the generic orbits of $\overline{\Delta_2}^{\,0}$ is $1$. But $\phi$ normalizes $\overline{\Delta_2}^{\,0}$ so preserves its orbits. In particular $\phi$ thus preserves a rational fibration: contradiction with the fact that $\phi$ is loxodromic. In [@ShepherdBarron] the classification of tight elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is given: \[thm:ShepherdBarron\] Every loxodromic element of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is rigid. Let $\phi$ be a loxodromic birational self map of the complex projective plane; then - if $\phi$ is conjugate to a monomial map, no power of $\phi$ is tight; - otherwise $\phi^n$ is tight for some integer $n$. Consider a subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let $\phi\in\mathrm{G}$ be a rigid element; then $\phi$ is also a rigid element in $\mathrm{G}$. The same holds for tight elements but the converse does not: there exist loxodromic maps $\phi\in\mathrm{G}$ such that $\phi$ is tight in $\mathrm{G}$ but not in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Proof of Theorem \[thm:ShepherdBarron\] and Lemma \[lem:densezar\] imply the following: \[thm:pfou\] Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let $\phi$ be a loxodromic element. The following assertions are equivalent: - no power of $\phi$ is tight in $\mathrm{G}$; - there is a subgroup $\Delta_2\subset\mathrm{G}$ that is normalized by $\phi$ and a birational self map $\psi$ of $\mathbb{P}^2_\mathbb{C}$ such that $\psi\circ\Delta_2\circ\psi^{-1}$ is a dense subgroup of $\mathrm{D}_2$ and $\psi\circ\varphi\circ\psi^{-1}$ belongs to $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. The group $\mathrm{A}$ fixes a point $p\in\partial\mathbb{H}^\infty\cup\mathbb{H}^\infty$ (Theorem \[thm:weak\]). Note that if $p$ belongs to $\mathbb{H}^\infty$, then $\mathrm{A}$ is bounded and Theorem \[thm:DelzantPy\] allows to conclude. Let us assume that $p$ belongs to $\partial\mathbb{H}^\infty$. Remark that if $\mathrm{A}$ fixes an other point $q$ on $\partial\mathbb{H}^\infty$, then $\mathrm{A}$ fixes the geodesic between $p$ and $q$ and so $\mathrm{A}$ would be bounded again. Suppose thus that $p$ is the only fixed point of $\mathrm{A}$ in $\partial\mathbb{H}^\infty$. Consider a loxodromic map $\phi$ of $\mathrm{N}$. It normalizes $\mathrm{A}$ and so $\phi$ fixes $p$. As $\phi$ is loxodromic, $\phi$ does not preserve any fibration; consequently $p$ does not correspond to the class of a fibration. From Lemma \[lem:Urechtecalt\] any finitely generated group of elliptic elements that fixes $p$ is bounded. Let $\mathrm{G}$ be the subgroup of birational self maps of $\mathbb{P}^2_\mathbb{C}$ that fix $p$. Denote by $L$ the $1$-dimensional subspace of $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ corresponding to $p$. The group $\mathrm{G}$ fixes $p$; hence its linear action on $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ acts on $L$ by automorphisms preserving the orientation. This implies the existence of a group homomorphism $\rho\colon\mathrm{G}\to\mathbb{R}^_+$. Note that $\mathrm{G}$ does not contain any parabolic element because $p$ does not correspond to the class of a fibration and that loxodromic elements do not fix any vector in $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$. As a result $\ker\rho$ consists of elliptic elements. But $1$ is the only eigenvalue of a map of $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ induced by an elliptic birational self map ([@Cantat:annals]); as a consequence any elliptic birational map of $\mathrm{G}$ is contained in $\ker\rho$. Take a loxodromic map $\phi$ in $\mathrm{G}$. Let us show by contradiction that no power of $\phi$ is tight in $\mathrm{G}$. So assume that there exists $n\in\mathbb{Z}$ such that $\phi^n$ is tight in $\mathrm{G}$. The subgroup $\mathrm{N}$ of $\mathrm{G}$ is infinite and $\langle\phi\rangle$ has finite index in $\mathrm{Cent}(\phi)$ (Theorem \[thm:cent\]); there thus exists $\psi\in\mathrm{G}$ that do not commute with $\phi^n$. Since all non trivial elements of $\ll\phi^n\gg$ are loxodromic ([@CantatLamy]) the map $\psi\circ\phi^n\circ\psi^{-1}\circ\phi^{-n}$ is loxodromic. But $\rho\big(\psi\circ\phi^n\circ\psi^{-1}\circ\phi^{-n}\big)=1$, [i.e.]{} $\psi\circ\phi^n\circ\psi^{-1}\circ\phi^{-n}$ is elliptic: contradiction. Finally no power of $\phi$ is tight in $\mathrm{G}$. According to Theorem \[thm:pfou\] there exist $\varphi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and $\Delta_2$ an algebraic subgroup of $\mathrm{G}$ such that - $\varphi\circ\phi\circ\varphi^{-1}$ is monomial; - $\varphi\circ\Delta_2\circ\varphi^{-1}=\mathrm{D}_2$. Consider a finitely generated subgroup $\Gamma$ of $\ker\rho$. The <span style="font-variant:small-caps;">Zariski</span> closure $\overline{\Gamma}$ of $\Gamma$ is an algebraic subgroup of $\mathrm{G}$ because $\Gamma$ is bounded. Set $$d=\sup\{\dim\overline{\Gamma}\,\vert\,\Gamma\subset\ker\rho\text{ finitely generated}\}$$ We will distinguish the cases $d$ is finite and $d$ is infinite. - First consider the case $d<\infty$. Note that $\ker\rho$ contains a subgroup conjugated to $\mathrm{D}_2$ so $d\geq 2$. Take $\Gamma$ a finitely generated subgroup of $\ker\rho$ such that $\dim\overline{\Gamma}=d$. Let $\overline{\Gamma}^0$ be the neutral component of the algebraic group $\overline{\Gamma}$. Let $\phi$ be an element of $\mathrm{G}$. The group $\phi\circ\overline{\Gamma}^0\circ\phi^{-1}$ is again an algebraic subgroup and $\langle\overline{\Gamma}^0,\,\phi\circ\overline{\Gamma}^0\circ\phi^{-1}\rangle$ is contained in $\overline{\langle\Gamma,\,\phi\circ\Gamma\circ\phi^{-1}\rangle}$. According to [@Humphreys] the group $\langle\overline{\Gamma}^0,\,\phi\circ\overline{\Gamma}^0\circ\phi^{-1}\rangle$ is closed and connected. On the one hand $\dim \langle\overline{\Gamma}^0,\,\phi\circ\overline{\Gamma}^0\circ\phi^{-1}\rangle\leq d$ and on the other hand $\overline{\Gamma}^0\subset\langle\overline{\Gamma}^0,\,\phi\circ\overline{\Gamma}^0\circ\phi^{-1}\rangle$. As a consequence $\langle\overline{\Gamma}^0,\,\phi\circ\overline{\Gamma}^0\circ\phi^{-1}\rangle=\overline{\Gamma}^0$. In other words $\phi$ normalizes $\overline{\Gamma}^0$. But $\Gamma\cap\overline{\Gamma}^0$ is infinite so there exists a birational self map $\psi$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ such that $\psi\circ\mathrm{G}\circ\psi^{-1}\subset\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$ (Theorem \[thm:DelzantPy\]) and hence $\psi\circ\mathrm{N}\circ\psi^{-1}\subset\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. - Now assume $d=\infty$. Let $\Gamma$ be a finitely generated subgroup of $\ker\rho$ such that $\dim\overline{\Gamma}\geq~9$. The closure $\overline{\Gamma}$ of $\Gamma$ preserves a unique rational fibration given by a rational map $\pi\colon\mathbb{P}^2_\mathbb{C}\dashrightarrow\mathbb{P}^1_\mathbb{C}$ (Lemma \[lem:Urechmoinsde9\]). Consider an element $\phi$ of $\ker\rho$. The algebraic group $\overline{\langle\Gamma,\,\phi\rangle}$ also preserves a unique rational fibration; since $\overline{\Gamma}\subset\langle\overline{\Gamma},\,\phi\rangle$ this fibration is given by $\pi$. As a result $\ker\rho$ preserves a rational fibration. Hence $\ker\rho$ is bounded and the group $\phi\circ\mathrm{G}\circ\phi^{-1}$ is contained in $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$ (Theorem \[thm:DelzantPy\]); in particular $\phi\circ\mathrm{N}\circ\phi^{-1}$ is a subgroup of $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. \[lem:withoutaxe\] Let $\phi$ and $\psi$ be two loxodromic elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ such that $\mathrm{Ax}(\phi)\not=\mathrm{Ax}(\psi)$. Then - either $\phi$ and $\psi$ have not a common fixed point on $\partial\mathbb{H}^\infty$, - or $\langle\phi,\,\psi\rangle$ contains a subgroup $\mathrm{G}$ and there exists a birational self map $\varphi$ of the complex projective plane such that - $\varphi\circ\langle\phi,\,\psi\rangle\circ\varphi^{-1}\subset\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$, - $\varphi\circ\mathrm{G}\circ\varphi^{-1}$ is a dense subgroup of $\mathrm{D}_2$. Suppose that $\phi$ and $\psi$ have a common fixed point $p\in\partial\mathbb{H}^\infty$. Denote by $L$ the $1$-dimensional subspace of $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ corresponding to $p$. The group $\langle\phi,\,\psi\rangle$ generated by $\phi$ and $\psi$ fixes $p$ so its linear action on $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ acts on $L$ by automorphisms preserving the orientation. A reasoning analogous to that of the proof of Theorem \[thm:urechnorm\] implies the existence of a group homomorphism $$\rho\colon\langle\phi,\,\psi\rangle\to\mathbb{R}^_+$$ whose kernel consists of elliptic birational maps (see* Proof of Theorem \[thm:urechnorm\]). Assume that $\phi^n$ is tight for some $n$. Since $\mathrm{Ax}(\phi)\not=\mathrm{Ax}(\psi)$ the maps $\phi^n$ and $\psi$ do not commute. According to [@CantatLamy] any non trivial element of $\ll\phi^n\gg$ is loxodromic. Therefore on the one hand $\psi\circ\phi^n\circ\psi^{-1}\circ\phi^{-n}$ is loxodromic, and on the other hand $\rho(\psi\circ\varphi^n\circ\psi^{-1})=1$ hence $\psi\circ\varphi^n\circ\psi^{-1}$ is elliptic: contradiction. As a result for any $k$ the map $\phi^k$ is not tight in $\langle\phi,\,\psi\rangle$. Theorem \[thm:pfou\] implies that there exist a birational self map $\varphi$ of $\mathbb{P}^2_\mathbb{C}$ and a bounded subgroup $\Delta_2\subset\langle\phi,\,\psi\rangle$ such that - $\varphi\circ\phi\circ\varphi^{-1}$ is monomial; - $\varphi\circ\Delta_2\circ\varphi^{-1}$ is a dense subgroup of $\mathrm{D}_2$. In particular $\ker\rho\supset\Delta_2$ is thus infinite. Theorem \[thm:urechnorm\] allows to conclude. \[Urech:without\] Let $\phi$ and $\psi$ be two loxodromic elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ such that $\mathrm{Ax}(\phi)\not=\mathrm{Ax}(\psi)$. Then $\phi$ and $\psi$ have not a common fixed point on $\partial\mathbb{H}^\infty$. Assume by contradiction that $\phi$ and $\psi$ have no common fixed point on $\partial\mathbb{H}^\infty$. Lemma \[lem:withoutaxe\] thus implies that up to birational conjugacy - $\varphi\circ\langle\phi,\,\psi\rangle\circ\varphi^{-1}\subset\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$, - $\varphi\circ\mathrm{G}\circ\varphi^{-1}\subset\mathrm{D}_2$ is a dense subgroup. Let us write $$\begin{aligned} & \phi=d_1\circ m_1 && \psi=d_2\circ m_2\end{aligned}$$ with $d_i\in\mathrm{D}_2$ and $m_i\in\mathrm{GL}(2,\mathbb{Z})$. The group $\mathrm{D}_2$ fixes the axes of all the monomial loxodromic elements; in particular $m_1$ and $m_2$ have the same fixed points on $\partial\mathbb{H}^\infty$ as $\phi$ and $\psi$. But the group $\langle m_1,\,m_2\rangle$ does not contain any infinite abelian group so according to Lemma \[lem:withoutaxe\] the birational maps $m_1$ and $m_2$ have not a common fixed point on $\partial\mathbb{H}^\infty$: contradiction. \[lem:Urechlox\] Let $\mathrm{G}$ be a subgroup of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group that contains a loxodromic element. Then one of the following holds: - $\mathrm{G}$ is conjugate to a subgroup of $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$; - $\mathrm{G}$ contains a subgroup of index at most $2$ that is isomorphic to $\mathbb{Z}\ltimes\mathrm{H}$ where $\mathrm{H}$ is a finite group; - $\mathrm{G}$ contains a non-abelian free subgroup. Let $\phi$ be a loxodromic map of $\mathrm{G}$. - Assume first that all elements in $\mathrm{G}$ preserve the axis $\mathrm{Ax}(\phi)$ of $\phi$. The group $\mathrm{G}$ contains a subgroup $\mathrm{H}$ of index at most $2$ with the following property: $\mathrm{H}$ preserves the orientation of the axis. Any element $\psi\in\mathrm{H}$ thus translates the points on $\mathrm{Ax}(\phi)$ by a constant $c_\psi$. This yields a group morphism $$\begin{aligned} &\pi\colon\mathrm{H}\to\mathbb{R}, && \psi\mapsto c_\psi\end{aligned}$$ such that $\ker\pi$ is a bounded group. From Theorem \[thm:DelzantPy\] either $\ker\pi$ is finite, or $\mathrm{G}$ is conjugate to a subgroup of $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. - Suppose that there is an element $\psi\in\mathrm{G}$ that does not preserve $\mathrm{Ax}(\phi)$. Denote by $\alpha(\phi)$ and $\omega(\phi)$ (resp. $\alpha(\psi)$ and $\omega(\psi)$) the attractive and repulsive fixed points of $\phi_\bullet$ (resp. $\psi_\bullet$). Let $\mathcal{U}_1^+$ (resp. $\mathcal{U}_1^-$, resp. $\mathcal{U}_2^+$, resp. $\mathcal{U}_2^-$) be a small neighborhood of $\alpha(\phi)$ (resp. $\omega(\phi)$, resp. $\alpha(\psi)$, resp. $\omega(\psi)$) in $\partial\mathbb{H}^\infty$. We can assume that $\mathcal{U}_1^+$, $\mathcal{U}_1^-$, $\mathcal{U}_2^+$ and $\mathcal{U}_2^-$ are pairwise disjoint. Set $\mathcal{U}_1=\mathcal{U}_1^+\cup\mathcal{U}_1^-$ and $\mathcal{U}_2=\mathcal{U}_2^+\cup\mathcal{U}_2^-$. There exist $n_1$, $n_2$, $n_3$, $n_4$ some positive integers such that $$\begin{aligned} &\phi^{n_1}(\mathcal{U}_2)\subset\mathcal{U}_1^+, && \phi^{-n_2}(\mathcal{U}_2)\subset\mathcal{U}_1^-, && \psi^{n_3}(\mathcal{U}_1)\subset\mathcal{U}_2^+,&&\psi^{-n_4}(\mathcal{U}_1)\subset\mathcal{U}_2^-.\end{aligned}$$ Set $n=\max(n_1,n_2,n_3,n_4)$. Since $$\begin{aligned} &\phi(\mathcal{U}_1^+)\subset\mathcal{U}_1^+ && \phi^{-1}(\mathcal{U}_1^-)\subset\mathcal{U}_1^-&& \psi(\mathcal{U}_2^+)\subset\mathcal{U}_2^+ && \psi^{-1}(\mathcal{U}_2^-)\subset\mathcal{U}_2^-\end{aligned}$$ one gets that for any $k\leq n$ $$\begin{aligned} &\phi^k(\mathcal{U}_2)\subset\mathcal{U}_1 &&\phi^{-k}(\mathcal{U}_2)\subset\mathcal{U}_1 &&\psi^k(\mathcal{U}_1)\subset\mathcal{U}_2 && \psi^{-k}(\mathcal{U}_1)\subset\mathcal{U}_2\end{aligned}$$ According to Ping Pong Lemma applied to $\phi^n$, $\psi^n$ together with $\mathcal{U}_1$ and $\mathcal{U}_2$ we get that $\langle\psi^n,\,\phi^n\rangle$ generates a non-abelian free subgroup of $\mathrm{G}$. ### <span style="font-variant:small-caps;">Tits</span> alternative for finitely generated subgroups for automorphisms groups and <span style="font-variant:small-caps;">Jonquières</span> group \[lem:extsolv\] Let $\mathrm{G}$ be a finitely generated group. Assume that $\mathrm{G}$ is an extension of a virtually solvable group $\mathrm{R}$ of length $r$ by an other virtually solvable group $\mathrm{Q}$ of length $q$ $$1\longrightarrow \mathrm{R}\longrightarrow\mathrm{G}\longrightarrow\mathrm{Q}\longrightarrow 1.$$ Then $\mathrm{G}$ is virtually solvable of length $\leq q+r+1$. Hence one has the following statement: \[pro:titsext\] Let $\mathrm{G}_1$ and $\mathrm{G}_2$ be two groups that satisfy <span style="font-variant:small-caps;">Tits</span> alternative. If $\mathrm{G}$ is an extension of $\mathrm{G}_1$ by $\mathrm{G}_2$, then $\mathrm{G}$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative. Let $\Gamma$ be a subgroup of $\mathrm{G}$ that does not contain a non abelian free subgroup. For $i\in\{1,\,2\}$ denote by $\mathrm{pr}_i\colon\mathrm{G}\to\mathrm{G}_i$ the canonical projection. Since $\mathrm{pr}_i(\mathrm{G})$ does not contain a non abelian free subgroup $\mathrm{pr}_i(\Gamma)=\Gamma\cap\mathrm{G}_i$ is virtually solvable ($\mathrm{G}_i$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative). Then according to Lemma \[lem:extsolv\] the group $\Gamma$ is virtually solvable. A first consequence of this result is the following one: \[thm:titsaut\] Let $V$ be a <span style="font-variant:small-caps;">Kähler</span> compact manifold. Its automorphism group satisfies <span style="font-variant:small-caps;">Tits</span> alternative. The group $\mathrm{Aut}(V)$ acts on the cohomology of $V$. This yields to a morphism $\rho$ from $\mathrm{Aut}(V)$ to $\mathrm{GL}(H^(V,\mathbb{Z}))$ where $H^(V,\mathbb{Z})$ denotes the direct sum of the cohomology groups of $V$. According to [@Lieberman] - the kernel of $\rho$ is a complex <span style="font-variant:small-caps;">Lie</span> group with a finite number of connected components; - its connected component $\mathrm{Aut}^0(V)$ is an extension of a compact complex torus by a complex algebraic group. We get the result from Proposition \[pro:titsext\] and classical <span style="font-variant:small-caps;">Tits</span> alternative. A direct consequence of Proposition \[pro:titsext\] and <span style="font-variant:small-caps;">Tits</span> alternative for linear groups is: \[pro:Titsjonq\] The <span style="font-variant:small-caps;">Jonquières</span> group $$\mathcal{J}\simeq\mathrm{PGL}(2,\mathbb{C})\rtimes\mathrm{PGL}(2,\mathbb{C}(z_0))$$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative. ### “Weak alternative” for isometries of $\mathbb{H}^\infty$ Let us recall some notations and definitions introduced in Chapter \[chap:hyperbolicspace\]. Let $\mathcal{H}$ be a seperable <span style="font-variant:small-caps;">Hilbert</span> space. Let us fix a <span style="font-variant:small-caps;">Hilbert</span> basis $\mathcal{B}=(\mathbf{e}_i)_i$ on $\mathcal{H}$. Consider the scalar product defined on $\mathcal{H}$ by $$\langle v,\,v\rangle=v_0^2-\displaystyle\sum_{i=1}^\infty v_i^2$$ where the coordinates $v_i$ are the coordinates of $v$ in $\mathcal{B}$. The light cone of $\mathcal{H}$ is the set $$\mathcal{L}(\mathcal{H})=\big\{v\in\mathcal{H}\,\vert\,\langle v,\,v\rangle=0\big\}.$$ Let $\mathbb{H}^\infty$ be the connected component of the hyperboloid $$\big\{v\in\mathcal{H}\,\vert\,\langle v,\,v\rangle=1\big\}$$ that contains $\mathbf{e}_0$. Consider the metric defined on $\mathbb{H}^\infty$ by $$d(u,v):=\mathrm{arccos}(\langle u,\,v\rangle).$$ The space $\mathbb{H}^\infty$ is a complete CAT$(-1)$ space so is hyperbolic (Chapter \[chap:hyperbolicspace\]). Its boundary $\partial\mathbb{H}^\infty$ can be identified to $\mathbb{P}(\mathcal{L}(\mathcal{H}))$. \[thm:weak\] Let $\Gamma$ be a subgroup of $\mathrm{O}(1,\infty)$. 1. If $\Gamma$ contains a loxodromic isometry $\psi$, then one of the following properties holds: - $\Gamma$ contains a non-abelian free group, - $\Gamma$ permutes the two fixed points of $\psi$ that lie on $\partial\mathbb{H}^\infty$. 2. If $\Gamma$ contains no loxodromic isometry, then $\Gamma$ fixes a point of $\mathbb{H}^\infty\cup\partial\mathbb{H}^\infty$. - Assume first that $\Gamma$ contains two loxodromic isometries $\phi$ and $\psi$ such that the fixed points of $\phi$ and $\psi$ on $\partial\mathbb{H}^\infty$ are pairwise distinct. According to the ping-pong Lemma (Lemma \[lem:pingpong\]) there are two integers $n$ and $m$ such that $\phi^n$ and $\psi^m$ generate a subgroup of $\Gamma$ isomorphic to the free group $\mathbb{F}_2$. - Suppose that $\Gamma$ contains at least one loxodromic isometry $\phi$. Let $\alpha(\phi)$ and $\omega(\phi)$ be the fixed points of $\phi$ on $\partial\mathbb{H}^\infty$. If $\Gamma$ contains an element $\psi$ such that $$\big\{\alpha(\phi),\,\omega(\phi)\big\}\cap\big\{\alpha(\psi),\,\omega(\psi)\big\}=\emptyset$$ then $\phi$ and $\psi\circ\phi\circ\psi^{-1}$ are two loxodromic isometries to which we can apply the previous argument. Otherwise $\Gamma$ fixes either $\big\{\alpha(\phi),\,\omega(\phi)\big\}$, or $\{\alpha(\phi)\}$, or $\{\omega(\phi)\}$. Then $\Gamma$ contains a subgroup of index $2$ that fixes $\alpha(\phi)$ of $\omega(\phi)$. - Assume that $\Gamma$ contains two parabolic isometries $\phi$ and $\psi$ whose fixed points $\alpha(\phi)\in\partial\mathbb{H}^\infty$ and $\alpha(\psi)\in\partial\mathbb{H}^\infty$ are distinct. Take two elements of $\mathcal{L}(\mathcal{H})$ still denoted $\alpha(\phi)$ and $\alpha(\psi)$ that represent these two points of $\partial\mathbb{H}^\infty$. Let $\ell$ be a point of $\mathcal{H}$ such that $$\begin{aligned} \langle\alpha(\phi),\,\ell\rangle<0 && \langle\alpha(\psi),\,\ell\rangle>0 .\end{aligned}$$ The hyperplane of $\mathcal{H}$ orthogonal to $\ell$ intersects $\mathbb{H}^\infty$ in a subspace $L$ that “separates” $\alpha(\phi)$ and $\alpha(\psi)$. As a result there exist integers $n$ and $m$ such that $\phi^m(L)$, $\phi^{-m}(L)$, $\psi^n(L)$ and $\psi^{-n}(L)$ don’t pairwise intersect. The isometry $\phi^m\circ\psi^n$ has thus two distinct fixed points on $\partial\mathbb{H}^\infty$; hence it is a loxodromic one. Applying the above argument we get that $\langle\phi,\,\psi\rangle$ contains a free group. Therefore if $\Gamma$ contains at least one parabolic isometry, then - either $\Gamma$ contains a non-abelian free group; - or $\Gamma$ fixes a point of $\partial\mathbb{H}^\infty$ that is the unique fixed point of the parabolic isometries of $\Gamma$. - Let us finish by assuming that all elements of $\Gamma$ are elliptic ones. According to [@delaHarpeGhys Chapter $8$, Lemma $35$ and Corollary $36$] - either the orbit of any point of $\mathbb{H}^\infty$ is bounded; - or the limit set of $\Gamma$ is a point. From [@delaHarpeValette Chapter $2$, §b.8] one gets the following alternative: $\Gamma$ fixes - either a point of $\mathbb{H}^\infty$; - or a point of $\partial\mathbb{H}^\infty$. ### Proof of Theorem \[thm:urechtits\] #### Assume that $\mathrm{G}$ contains a loxodromic element Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that contains a loxodromic element. According to Lemma \[lem:Urechlox\] we have to consider the three following cases: - $\mathrm{G}$ is conjugate to a subgroup of $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$ and <span style="font-variant:small-caps;">Tits</span> alternative holds by Proposition \[pro:titsext\]; - $\mathrm{G}$ contains a subgroup of index at most $2$ that is isomorphic to $\mathbb{Z}\ltimes\mathrm{H}$ where $\mathrm{H}$ is a finite group, in other words $\mathrm{G}$ is cyclic up to finite index so <span style="font-variant:small-caps;">Tits</span> alternative holds; - $\mathrm{G}$ contains a non-abelian free subgroup, and <span style="font-variant:small-caps;">Tits</span> alternative holds. We can thus state: Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that contains a loxodromic element. Then $\mathrm{G}$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative. #### Assume that $\mathrm{G}$ contains a parabolic element but no loxodromic element \[lem:jonqhalph\] Let $\mathrm{G}$ be a subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that does not contain any loxodromic element but contains a parabolic element. Then $\mathrm{G}$ is conjugate to a subgroup of $\mathcal{J}$ or $\mathrm{Aut}(S)$, where $S$ is a <span style="font-variant:small-caps;">Halphen</span> surface. By Theorem \[thm:weak\] the group $\mathrm{G}$ fixes a point $p\in\mathbb{H}^\infty\cup\partial\mathbb{H}^\infty$. Consider a parabolic element $\varphi$ of $\mathrm{G}$; then $\varphi$ has no fixed point in $\mathbb{H}^\infty$ and a unique fixed point $q$ in $\partial\mathbb{H}^\infty$. As a consequence $p=q$. According to Theorem \[thm:dilfav\] there exist a surface $S$, a birational map $\psi\colon\mathbb{P}^2_\mathbb{C}\dashrightarrow S$, a curve $C$, and a fibration $\pi\colon S\to C$ such that $\psi\circ\varphi\circ\psi^{-1}$ permutes the fibres of $\pi$. In particular $\psi\circ\varphi\circ\psi^{-1}$ preserves the divisor class of a fibre $F$ of $\pi$. Since $F$ is a class of a fibre, $F\cdot F=0$. The point $m\in\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ corresponding to $F$ thus satisfies $m\cdot m=0$. Therefore $q\in\partial\mathbb{H}^\infty$ corresponds to the line passing through the origin and $m$. It follows that any element in $\mathrm{G}$ fixes $m$ and so preserves the divisor class of $F$. In other words any element in $\varphi\circ\mathrm{G}\circ\varphi^{-1}$ permutes the fibres of the fibration $\pi\colon S\to C$. If the fibration is rational, then up to birational conjugacy $\mathrm{G}\subset\mathcal{J}$. If the fibration is a fibration of genus $1$ curves, there exists a <span style="font-variant:small-caps;">Halphen</span> surface $S'$ such that up to birational conjugacy $\mathrm{G}$ is contained in $\mathrm{Bir}(S')$ and preserves the <span style="font-variant:small-caps;">Halphen</span> fibration. By Lemma \[lem:UrechHalphen\] the group $\mathrm{G}$ is contained in $\mathrm{Aut}(S')$. Assume first that up to birational conjugacy $\mathrm{G}\subset\mathcal{J}\simeq\mathrm{PGL}(2,\mathbb{C}(z_1))\rtimes\mathrm{PGL}(2,\mathbb{C})$. <span style="font-variant:small-caps;">Tits</span> alternative for linear groups in characteristic $0$ and Proposition \[pro:titsext\] imply <span style="font-variant:small-caps;">Tits</span> alternative for $\mathrm{G}$. Finally suppose that $\mathrm{G}\subset\mathrm{Aut}(S)$ where $S$ is a <span style="font-variant:small-caps;">Halphen</span> surface. The automorphisms groups of <span style="font-variant:small-caps;">Halphen</span> surfaces have been studied ([@Gizatullin; @CantatDolgachev; @Grivaux]). In particular <span style="font-variant:small-caps;">Cantat</span> and <span style="font-variant:small-caps;">Dolgachev</span> prove \[thm:candol\] Let $S$ be a <span style="font-variant:small-caps;">Halphen</span> surface. There exists a homomorphism $\rho\colon\mathrm{Aut}(S)\to\mathrm{PGL}(2,\mathbb{C})$ with finite image such that $\ker\rho$ is an extension of an abelian group of rank $\leq 8$ by a cyclic group of order dividing $24$. In other words the automorphism group of a <span style="font-variant:small-caps;">Halphen</span> surface is virtually abelian hence $\mathrm{G}$ is solvable up to finite index. #### Assume that $\mathrm{G}$ is a group of elliptic elements According to Theorems \[thm:urechell1\] and \[thm:urechell2\] one of the following holds: - $\mathrm{G}$ is isomorphic to a bounded subgroup; - $\mathrm{G}$ preserves a rational fibration. Suppose that $\mathrm{G}$ is isomorphic to a bounded subgroup; in particular $\mathrm{G}$ is isomorphic to a subgroup of linear groups and so satisfies <span style="font-variant:small-caps;">Tits</span> alternative. If $\mathrm{G}$ preserves a rational fibration, then $\mathrm{G}$ satisfies <span style="font-variant:small-caps;">Tits</span> alternative (Proposition \[pro:Titsjonq\]). ### A consequence of <span style="font-variant:small-caps;">Tits</span> alternative: the <span style="font-variant:small-caps;">Burnside</span> problem The <span style="font-variant:small-caps;">Burnside</span> problem posed by <span style="font-variant:small-caps;">Burnside</span> in $1902$ asks whether a finitely generated torsion group is finite. <span style="font-variant:small-caps;">Schur</span> showed in $1911$ that any finitely generated torsion group that is a subgroup of invertible $n\times n$ complex matrices is finite ([@Schur]). One of the tool of the proof is the <span style="font-variant:small-caps;">Jordan</span>-<span style="font-variant:small-caps;">Schur</span> Theorem. In the $1930$’s <span style="font-variant:small-caps;">Burnside</span> asked another related question called the restricted <span style="font-variant:small-caps;">Burnside</span> problem: if it is known that a group $\mathrm{G}$ with $m$ generators and exponent $n$ is finite, can one conclude that the order of $\mathrm{G}$ is bounded by some constant depending only on $n$ and $m$ ? In other words are there only finitely many finite groups with $m$ generators of exponent $n$ up to isomorphism ? In $1958$ <span style="font-variant:small-caps;">Kostrikin</span> was able to prove that among the finite groups with a given number of generators and a given prime exponent, there exists a largest one: this provides a solution for the restricted <span style="font-variant:small-caps;">Burnside</span> problem for the case of prime exponent ([@Kostrikin]). Later <span style="font-variant:small-caps;">Zelmanov</span> solved the restricted <span style="font-variant:small-caps;">Burnside</span> problem for an arbitrary exponent ([@Zelmanov1991; @Zelmanov1992]). <span style="font-variant:small-caps;">Golod</span> gave a negative answer to the <span style="font-variant:small-caps;">Burnside</span> problem for groups that have a complete system of linear representations ([@Golod]). Later many examples of infinite, finitely generated and torsion groups with even bounded ordres were exhibited ([@AdjanNovikov1; @AdjanNovikov2; @AdjanNovikov3; @Olshanskii; @Ivanov:Burnside; @Lysenok]). The problem raised by <span style="font-variant:small-caps;">Burnside</span> is still open for homeomorphism (resp. diffeomorphism) groups on closed manifolds. Very few examples are known. <span style="font-variant:small-caps;">Cantat</span> gave a positive answer to the <span style="font-variant:small-caps;">Burnside</span> problem for the <span style="font-variant:small-caps;">Cremona</span> group: \[thm:burnside\] Every finitely generated torsion subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is finite. Let $\mathrm{G}$ be a finitely generated torsion subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. From <span style="font-variant:small-caps;">Tits</span> alternative (Theorem \[thm:cantattits\]) $\mathrm{G}$ is solvable up to finite index. Since any torsion, solvable, finitely generated group is finite, $\mathrm{G}$ is finite. Solvable subgroups of the <span style="font-variant:small-caps;">Cremona</span> group {#sec:solvable} ------------------------------------------------------------------------------------- The study of the solvable subgroups of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group starts in [@Deserti:resoluble] and goes on in [@Urech:ellipticsubgroups]. \[thm:resoluble\] Let $\mathrm{G}$ be a solvable subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Then one of the following holds: - $\mathrm{G}$ is a subgroup of elliptic elements, in particular $\mathrm{G}$ is isomorphic either to a solvable subgroup of $\mathcal{J}$, or to a solvable subgroup of a bounded group; - $\mathrm{G}$ is conjugate to a subgroup of $\mathcal{J}$; - $\mathrm{G}$ is conjugate to a subgroup of the automorphism group of a <span style="font-variant:small-caps;">Halphen</span> surface; - $\mathrm{G}$ is conjugate to a subgroup of $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$ where $$\mathrm{D}_2=\big\{(z_0,z_1)\mapsto(\alpha z_0,\beta z_1)\,\vert\,\alpha,\,\beta\in\mathbb{C}^\big\};$$ - $\mathrm{G}$ contains a loxodromic element and there exists a finite subgroup $\mathrm{H}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ such that $\mathrm{G}=\mathbb{Z}\ltimes H$. A solvable subgroup of a bounded group is a solvable subgroup from one of the groups that appear in Theorem \[thm:blanc11cases\]. The centralizer of a birational self map of $\mathbb{P}^2_\mathbb{C}$ that preserves a unique fibration that is rational is virtually solvable (§\[subsubsec:centrjonq\]); this example illustrates the second case. Before giving the proof let us state some consequences. The soluble length of a nilpotent subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ can be bounded by the dimension of $\mathbb{P}^2_\mathbb{C}$ as <span style="font-variant:small-caps;">Epstein</span> and <span style="font-variant:small-caps;">Thurston</span> did in the context of <span style="font-variant:small-caps;">Lie</span> algebras and rational vector fields on a connected complex manifold ([@EpsteinThurston]): Let $\mathrm{G}$ be a nilpotent subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that is not a torsion group. The soluble length of $\mathrm{G}$ is bounded by $2$. Theorem \[thm:blanc11cases\] allows to prove: The derived length of a bounded solvable subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is $\leq 5$. The derived length of a solvable subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is at most $8$. It decomposes into three parts: $\mathrm{G}$ contains a loxodromic element; $\mathrm{G}$ does not contain a loxodromic element but $\mathrm{G}$ contains a parabolic element; $\mathrm{G}$ is a group of elliptic elements. 1. Assume first that $\mathrm{G}$ contains a loxodromic element. Then <span style="font-variant:small-caps;">Tits</span> alternative and Lemma \[lem:Urechlox\] imply the following alternative - either $\mathrm{G}$ is conjugate to a subgroup of $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$, - or $\mathrm{G}$ contains a subgroup of index at most two that is isomorphic to $\mathbb{Z}\ltimes\mathrm{H}$ where $\mathrm{H}$ is a finite group. 2. Suppose now that $\mathrm{G}$ does not contain a loxodromic element but $\mathrm{G}$ contains a parabolic element $\phi$. The map $\phi$ preserves a unique fibration $\mathcal{F}$ that is elliptic or rational. Let us prove that any element of $\mathrm{G}$ preserves $\mathcal{F}$. Denote by $\alpha(\phi)\in\partial\mathbb{H}^\infty$ the fixed point of $\phi_$. Take one element in the light cone $$\mathcal{L}\mathrm{Z}(\mathbb{P}^2_\mathbb{C})=\big\{d\in\mathrm{Z}(\mathbb{P}^2_\mathbb{C})\,\vert\, d\cdot d=0\big\}$$ of $\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ still denoted by $\alpha(\phi)$ that represents $\alpha(\phi)$. Assume by contradiction that there exists $\varphi$ in $\mathrm{G}$ such that $\varphi(\alpha(\phi))\not=\alpha(\phi)$. The map $\psi=\varphi\circ\phi\circ\varphi^{-1}$ is parabolic and fixes the unique element $\alpha(\psi)$ of $\mathcal{L}\mathrm{Z}(\mathbb{P}^2_\mathbb{C})$ proportional to $\varphi(\alpha(\phi))$. If $\varepsilon>0$ let us denote by $\mathcal{U}\big(\alpha,\varepsilon\big)$ the set $$\mathcal{U}\big(\alpha,\varepsilon\big)=\big\{\ell\in\mathcal{L}\mathrm{Z}(\mathbb{P}^2_\mathbb{C})\,\vert\,\alpha\cdot\ell<\varepsilon\big\}.$$ Take $\varepsilon>0$ such that $\mathcal{U}\big(\alpha(\phi),\varepsilon\big)\cap\mathcal{U}\big(\alpha(\psi),\varepsilon\big)=\emptyset$. Since $\psi_$ is parabolic, $\psi_^n\big(\mathcal{U}\big(\alpha(\phi),\varepsilon\big)\big)$ is contained in $\mathcal{U}\big(\alpha(\psi),\varepsilon\big)$ for $n$ large enough. For $m$ sufficiently large the following inclusions hold $$\phi_^m\circ\psi_^n\big(\mathcal{U}\big(\alpha(\phi),\varepsilon\big)\big)\subset\mathcal{U}\left(\alpha(\phi),\frac{\varepsilon}{2}\right)\subsetneq\mathcal{U}\big(\alpha(\phi),\varepsilon\big).$$ This implies that $\phi_^m\circ\psi_^n$ is loxodromic: contradiction. So $\alpha(\phi_)=\alpha(\varphi_)$ for any $\varphi\in\mathrm{G}$. Finally $\mathrm{G}$ is a subgroup either of $\mathcal{J}$, or of the automorphism group of a <span style="font-variant:small-caps;">Halphen</span> surface. 3. If $\mathrm{G}$ is a group of elliptic elements, then according to Theorems \[thm:urechell1\] and \[thm:urechell2\] either $\mathrm{G}$ is a bounded subgroup, or $\mathrm{G}$ preserves a rational fibration. Normal subgroups of the <span style="font-variant:small-caps;">Cremona</span> group {#CantatLamy:passimple} ----------------------------------------------------------------------------------- The strategy of <span style="font-variant:small-caps;">Cantat</span> and <span style="font-variant:small-caps;">Lamy</span> to produce strict, non-trivial, normal subgroups of $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ is to let $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ act on the hyperbolic space $\mathbb{H}^\infty(\mathbb{P}^2_\Bbbk)$. In the first part of their paper they define the notion of tight element: an element $\phi$ of $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ is tight if it satisfies the following three properties: - $\phi_\in\mathrm{Isom}(\mathbb{H}^\infty)$ is hyperbolic; - there exists a positive number $\varepsilon$ such that: if $\psi$ belongs to $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ and $\psi_(\mathrm{Ax}(\phi))$ contains two points at distance $\varepsilon$ which are at distance at most $1$ from $\mathrm{Ax}(\phi)$, then $\psi_(\mathrm{Ax}(\phi))=\mathrm{Ax}(\phi)$; - if $\psi$ belongs to $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ and $\psi_(\mathrm{Ax}(\phi))=\mathrm{Ax}(\phi)$, then $\psi\circ\phi\circ\psi^{-1}=\phi$ or $\psi\circ\phi\circ\psi^{-1}=~\phi^{-1}$. The second property is a rigidity property of $\mathrm{Ax}(\phi)$ with respect to isometries $\psi_$ for $\psi\in\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$; we say that $\mathrm{Ax}(\phi)$ is rigid* under the action of $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$. The third property means that the stabilizer of $\mathrm{Ax}(\phi)$ coincides with the normalizer of the cyclic group $\langle\phi\rangle$. Here since there is no confusion we write $\ll \phi\gg$ for $\ll \phi\gg_{\mathrm{Bir}(\mathbb{P}^2_\Bbbk)}$. <span style="font-variant:small-caps;">Cantat</span> and <span style="font-variant:small-caps;">Lamy</span> established the following statement: \[thm:canlam\] Let $\Bbbk$ be an algebraically closed field. If $\phi\in\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ is tight, then there exists a non-zero integer $n$ such that for any non-trivial element $\psi$ of $\ll \phi^n\gg$ $$\deg \psi\geq\deg(\phi^n).$$ In particular $\ll \phi^n\gg$ is a proper subgroup of $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$. In the second part of their article <span style="font-variant:small-caps;">Cantat</span> and <span style="font-variant:small-caps;">Lamy</span> showed that $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ contains tight elements. They distinguished two cases: $\Bbbk=\mathbb{C}$ and $\Bbbk\not=\mathbb{C}$. Let us focus on the case $\Bbbk=\mathbb{C}$. They proved that an element $\phi$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ of the form $a\circ j$, where $a$ is a general element of $\mathrm{PGL}(3,\mathbb{C})$ and $j$ is a <span style="font-variant:small-caps;">Jonquières</span> twist, is tight. Let us explain what general means in this context: any element of $\mathrm{PGL}(3,\mathbb{C})$ suits after removing a countable number of <span style="font-variant:small-caps;">Zariski</span> closed subsets of $\mathrm{PGL}(3,\mathbb{C})$. More precisely they needed the two following conditions: - the base-points of $\phi$ and $\phi^{-1}$ belong to $\mathbb{P}^2_\mathbb{C}$; - $\mathrm{Base}(\phi^k)\cap\mathrm{Base}(\phi^{-i})=\emptyset$ for any $k$, $i>0$. In [@Lonjou] <span style="font-variant:small-caps;">Lonjou</span> proved the following statement: \[thm:lonjou\] For any field $\Bbbk$ the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ is not simple. She did not use the notion of tight element but uses the WPD (weakly properly discontinuous) property. This property was proposed in the context of the mapping class group in [@BestvinaFujiwara]. An element $g$ of a group $\mathrm{G}$ satisfies the WPD property if for any $\varepsilon \geq 0$ for any point $p\in\mathbb{H}^\infty$ there exists a positive integer $N$ such that the set $$S(\varepsilon,p;N)=\big\{h\in\mathrm{G}\,\vert\,\mathrm{dist}(h(p),p)\leq\varepsilon,\,\mathrm{dist}(h(g^N(p)),g^N(p))\leq\varepsilon\big\}$$ is finite. Since the elements studied by <span style="font-variant:small-caps;">Lonjou</span> have an axis she followed the terminology introduced in [@Coulon] and said that the group $\mathrm{G}$ acts discretely along the axis of $g$. In [@DahmaniGuirardelOsin] the authors generalized the small cancellation theory for groups acting by isometries on $\delta$-hyperbolic spaces. Small cancellation theory and the WPD property are connected: - in the normal group generated by a family satisfying the small cancellation property elements have a large translation length ([@Guirardel]); - if some element $g$ satisfies WPD property then the conjugates of $\langle g^n\rangle$ form a family satisfying the small cancellation property. Combining these two statements the following holds: Let $\varepsilon$ be a positive real number. Let $\mathrm{G}$ be a group acting by isometries on a $\delta$-hyperbolic space $X$. Let $g$ be a loxodromic element of $\mathrm{G}$. If $\mathrm{G}$ acts discretely along the axis of $g$, then there exists $n\in\mathbb{N}$ such that for any $h\in\ll g^n\gg\smallsetminus\{\mathrm{id}\}$ the translation length $L(h)$ of $h$ satisfies $L(h)>\varepsilon$. In particular, for $n$ big enough $\ll g^n\gg$ is a proper subgroup of $\mathrm{G}$. Furthermore this subgroup is free. As a result to prove Theorem \[thm:lonjou\] <span style="font-variant:small-caps;">Lonjou</span> needed to exhibit elements satisfying the WPD property: \[pro:lonjou\] Let $n\geq 2$ and let $\Bbbk$ be a field of characteristic which does not divide $n$. Consider the action of $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ on $\mathbb{H}^\infty(\mathbb{P}^2_{\overline{\Bbbk}})$ where $\overline{\Bbbk}$ is the algebraic closure of $\Bbbk$. The group $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ acts discretely along the axis of the loxodromic map $$h_n\colon(z_0:z_1:z_2)\dashrightarrow\big(z_1z_2^{n-1}:z_1^n-z_0z_2^{n-1}:z_2^n\big).$$ If $\Bbbk$ is an algebraically closed field of characteristic $p>0$, then for any $\ell\geq 1$ one has ([@CerveauDeserti:ptdegre]) $$\ll h_p^\ell \gg=\mathrm{Bir}(\mathbb{P}^2_\Bbbk).$$ Let us explain why when $\Bbbk=\mathbb{C}$. Let us first establish that $$\label{relation1} \ll\sigma_2\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ Let $\phi$ be a birational self map of the complex projective plane. According to the <span style="font-variant:small-caps;">Noether</span> and <span style="font-variant:small-caps;">Castelnuovo</span> Theorem $$\phi=(A_1)\circ\sigma_2\circ A_2\circ\sigma_2\circ A_3\circ\ldots\circ A_n\circ(\sigma_2)$$ where the $A_i$’s belong to $\mathrm{PGL}(3,\mathbb{C})$. The group $\mathrm{PGL}(3,\mathbb{C})$ is simple; as a result any $A_i$ can be written as $$B_1\circ\big((z_0,z_1)\mapsto(-z_0,-z_1)\big)\circ B_1^{-1}\circ B_2\circ\big((z_0,z_1)\mapsto(-z_0,-z_1)\big)\circ B_2^{-1}\circ\ldots\circ B_n\circ\big((z_0,z_1)\mapsto(-z_0,-z_1)\big)\circ B_n^{-1}$$ with $B_i$ in $\mathrm{PGL}(3,\mathbb{C})$. The involutions $(z_0,z_1)\mapsto(-z_0,-z_1)$ and $\sigma_2$ are conjugate; therefore $\phi$ can be written as a composition of conjugates of $\sigma_2$. Since $\mathrm{PGL}(3,\mathbb{C})$ is simple, for any non-trivial element $A$ of $\mathrm{PGL}(3,\mathbb{C})$ the involution $\iota\colon(z_0,z_1)\mapsto(-z_0,z_1)$ can be written as a composition of conjugates of $A$. The involutions $\iota$ and $\sigma_2$ being conjugate one has $$\sigma_2=\varphi_1\circ A\circ \varphi_1^{-1}\circ \varphi_2\circ A\circ \varphi_2^{-1}\circ\ldots\circ \varphi_n\circ A\circ \varphi_n^{-1}$$ where the $\varphi_i$’s are some elements of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. As a result $\ll\sigma_2\gg\subset\ll A\gg$. But $\ll\sigma_2\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ (see (\[relation1\])) so $$\label{relation2} \ll A\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$$ If $\phi$ belongs to $\mathrm{PGL}(2,\mathbb{C}(z_1))$, then $$\label{relation3} \ll\phi\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$$ Indeed since $\mathrm{PGL}(2,\mathbb{C}(z_1))$ is simple, the involution $\iota$ can be written as a composition of conjugates of $\phi$. But according to (\[relation2\]) one has $\ll\iota\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ hence $\ll\phi\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. If $\phi$ belongs to $\mathcal{J}$, then $$\label{relation4} \ll\phi\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$$ Indeed up to birational conjugacy $\phi\colon(z_0,z_1)\dashrightarrow\Big(\phi_1(z_0,z_1),\gamma(z_1)\Big)$ where $\gamma$ is an homothety or a translation. Consider an element $\psi\colon(z_0,z_1)\dashrightarrow\Big(\psi_1(z_0,z_1),z_1\Big)$ of $\mathrm{PGL}(2,\mathbb{C}(z_1))$. The map $\varphi=[\phi,\psi]$ belongs to $$\ll\phi\gg\cap\,\mathrm{PGL}(2,\mathbb{C}(z_1)).$$ If $\psi$ is well chosen, then $\varphi$ is non trivial and from (\[relation3\]) one gets $$\ll\phi\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ As a result if $\phi$ is a birational self map of the complex projective plane such that there exists $\psi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ for which $[\phi,\psi]$ preserves a rational fibration, then from (\[relation4\]) $$\label{relation5} \ll\phi\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$$ Let $\phi\colon(z_0,z_1)\mapsto(z_1,P(z_1)-\delta z_0)$, $\delta\in\mathbb{C}^$, $P\in\mathbb{C}[z_1]$, $\deg P\geq 2$, be a <span style="font-variant:small-caps;">Hénon map</span>. Then $\ll\phi\gg=\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Indeed if $\psi\colon(z_0,z_1)\mapsto(z_0,2z_1)$, then $[\phi,\psi]$ preserves the rational fibration $z_0=$cst; one concludes with (\[relation5\]). More generally over any infinite field of characteristic which does not divide $n$ the map $h_n$ does not satisfy the WPD property: this explains the assumptions of Proposition \[pro:lonjou\]. Let us mention that <span style="font-variant:small-caps;">Lonjou</span> got not only the non-simplicity of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group from [@DahmaniGuirardelOsin] but also the following result: Let $\Bbbk$ be a field. The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group - contains free normal subgroups; - is $SQ$-universal, that is any countable subgroup embeds in a quotient of $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$. In [@ShepherdBarron] the author proved that any loxodromic element in the <span style="font-variant:small-caps;">Cremona</span> group over any field $\Bbbk$ generates a proper normal subgroup; as a result the group $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ is not a simple group. He also gave a criterion in terms of the translation length of a loxodromic map $\phi$ to know if $\phi$ is tight and hence if $\ll\phi^n\gg$ is a proper subgroup of $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ for some $n$. Let us give the relationship between tight element and element that satisfies WPD property. When we study the action of the <span style="font-variant:small-caps;">Cremona</span> group on $\mathbb{H}^\infty(\mathbb{P}^2_\Bbbk)$ the axis of any loxodromic element $\phi$ is rigid and the stabiliser $$\mathrm{Stab}(\mathrm{Ax}(\phi))=\big\{\psi\in\mathrm{Bir}(\mathbb{P}^2_\Bbbk)\,\vert\,\psi(\mathrm{Ax}(\phi))=\mathrm{Ax}(\phi)\big\}$$ of the axis $\mathrm{Ax}(\phi)$ is virtually cyclic if and only if some positive iterate of $\phi$ is tight ([@CantatLamy; @Lonjou; @ShepherdBarron]). As a result for $N$ large the set $S(\varepsilon,p;N)$ is contained in $\mathrm{Stab}(\mathrm{Ax}(\phi))$. The map $\phi$ thus satisfies the WPD property if and only if some positive iterate of $\phi$ is tight. Let us recall that a subgroup $\mathrm{H}$ of a group $\mathrm{G}$ is called a characteristic subgroup* of $\mathrm{G}$ if for every automorphism $\varphi$ of $\mathrm{G}$ the inclusion $\varphi(\mathrm{H})\subset\mathrm{H}$ holds. Recall that the examples of elements having the WPD property given by <span style="font-variant:small-caps;">Lonjou</span> are the <span style="font-variant:small-caps;">Hénon</span> maps $$h_n\colon(z_0:z_1:z_2)\dashrightarrow\big(z_1z_2^{n-1}:z_1^n-z_0z_2^{n-1}:z_2^n\big)$$ of degree $n$ which is not divisible by the characteristic of $\Bbbk$. The group of automorphisms of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is generated by inner automorphisms and the action of $\mathrm{Aut}(\mathbb{C},+,\cdot)$ (see §\[sec:autbir\]). As $h_n$ is defined over $\mathbb{Z}$ the subgroup $\ll h^m\gg$ is a characteristic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. One has the following result: The $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group contains infinitely many characteristic subgroups. Simple groups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ -------------------------------------------------------- This section is devoted to the classification of simple subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ (Theorems \[thm:urechsimple1\] and \[thm:urechsimple2\]) but also to the proof of the following statement: Let $S$ be a complex surface. If $\mathrm{G}$ is a finitely generated simple subgroup of $\mathrm{Bir}(S)$, then $\mathrm{G}$ is finite. ### Simple subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ Let us first prove Theorem \[thm:urechsimple1\]. Consider a simple group acting non-trivially on a rational complex surface. Then according to Theorems \[thm:urechsimple2\] and \[thm:blanc11cases\] the group $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{PGL}(3,\mathbb{C})$. Conversely the group $\mathrm{PGL}(3,\mathbb{C})=\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ acts by birational maps on $S$. Let us now deal with the proof of Theorem \[thm:urechsimple2\]. Let $\mathrm{G}$ be a simple subgroup of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. We distinguish three cases: - $\mathrm{G}$ contains no loxodromic element but a parabolic one; - $\mathrm{G}$ is an elliptic group; - $\mathrm{G}$ contains a loxodromic element. <!-- --> - Assume that $\mathrm{G}$ contains no loxodromic element but a parabolic one. Consider a simple subgroup $\mathrm{G}$ of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ that contains no loxodromic element but a parabolic element. Then $\mathrm{G}$ is conjugate to a subgroup of $\mathcal{J}$ and is isomorphic to a subgroup of $\mathrm{PGL}(2,\mathbb{C})$. According to Lemma \[lem:jonqhalph\] one has the following alternative: $\mathrm{G}$ is conjugate - either to a subgroup of the automorphisms group of a <span style="font-variant:small-caps;">Halphen</span> surface, - or to a subgroup of $\mathcal{J}$. But automorphisms groups of <span style="font-variant:small-caps;">Halphen</span> surfaces are finite extensions of abelian subgroups (Theorem \[thm:candol\]) so do not contain infinite simple subgroups. As a result $\mathrm{G}$ is conjugate to a subgroup of $\mathcal{J}$. The short exact sequence from the semi-direct product of $\mathcal{J}$ is $$1\longrightarrow\mathrm{PGL}(2,\mathbb{C}(z_1))\longrightarrow\mathcal{J}\stackrel{f}{\longrightarrow}\mathrm{PGL}(2,\mathbb{C})\longrightarrow 1$$ The group $\mathrm{G}$ is simple thus contained in the kernel of the image of $f$. In both cases $\mathrm{G}$ is isomorphic to a subgroup of $\mathrm{PGL}(2,\mathbb{C})$. - Suppose that $\mathrm{G}$ is an elliptic group. Let $\mathrm{G}$ be a simple subgroup of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group of elliptic elements. Then - either $\mathrm{G}$ is a subgroup of an algebraic group of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, - or $\mathrm{G}$ is conjugate to a subgroup of $\mathcal{J}$. According to Theorems \[thm:urechell1\] and \[thm:urechell2\] one of the following holds: - $\mathrm{G}$ is conjugate to a subgroup of an algebraic group; - $\mathrm{G}$ preserves a rational fibration; - $\mathrm{G}$ is a torsion group and $\mathrm{G}$ is isomorphic to a subgroup of an algebraic group. In the first two cases we are done. Let us assume that we are in the third one. Then $\mathrm{G}$ is a linear group and according to the Theorem of <span style="font-variant:small-caps;">Jordan</span> and <span style="font-variant:small-caps;">Schur</span> $\mathrm{G}$ has a normal abelian subgroup of finite index. As a consequence $\mathrm{G}$ is finite and so algebraic. - Finally we give a sketch of the proof of A simple subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ does not contain any loxodromic element. Let $\mathrm{G}$ be a simple subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Assume by contradiction that $\mathrm{G}$ contains a loxodromic map $\phi$. Theorems \[thm:canlam\] and \[thm:pfou\] imply that $\phi$ is a monomial map up to birational conjugacy. Looking at the curves contracted by elements of $\mathrm{G}$ <span style="font-variant:small-caps;">Urech</span> gets that all loxodromic elements of $\mathrm{G}$ are contained in $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$ ([@Urech:ellipticsubgroups Lemmas 3.17. and 3.18.]). Consider $\psi$ in $\mathrm{G}$. As $\psi\circ\phi\circ\psi^{-1}$ is loxodromic it is monomial. The axis of $\psi\circ\phi\circ\psi^{-1}$ is fixed pointwise by both $\psi\circ\mathrm{D}_2\circ\psi^{-1}$ and $\mathrm{D}_2$. The group $\mathrm{H}$ generated by $\psi\circ\phi\circ\psi^{-1}$ and $\mathrm{D}_2$ is thus bounded and according to Theorem \[thm:urechnorm\] conjugate to a subgroup of $\mathrm{D}_2$. Hence $\psi\circ\mathrm{D}_2\circ\psi^{-1}$ is contained in $\mathrm{D}_2$ and $\psi$ belongs to $\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$. We thus have the inclusion $\mathrm{G}\subset\mathrm{GL}(2,\mathbb{Z})\ltimes\mathrm{D}_2$ and get a non trivial morphism $\upsilon\colon\mathrm{G}\to\mathrm{GL}(2,\mathbb{Z})$. The kernel of $\upsilon$ contains an infinite subgroup of $\mathrm{D}_2$ normalized by $\phi$ (Lemma \[lem:densezar\]): contradiction with the fact that $\mathrm{G}$ is simple. ### Finitely generated simple subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ We finish the chapter by giving a sketch of the proof of the following statement: \[thm:finitelysimple\] Any finitely generated simple subgroup of the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group is finite. This result and the classification of finite subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ (see [@DolgachevIskovskikh]) imply: A finitely generated simple subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is isomorphic to - either $\faktor{\mathbb{Z}}{p\mathbb{Z}}$ for some prime $p$; - or $\mathcal{A}_5$; - or $\mathcal{A}_6$; - or $\mathrm{PSL}(2,\mathbb{C})$. Note that the conjugacy classes of these finite groups are also described in [@DolgachevIskovskikh]. Theorem \[thm:finitelysimple\] also holds for the group of birational self maps of a surface over a field $\Bbbk$. Let $\mathrm{G}$ be a finitely generated subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Let first see that $\mathrm{G}$ does not contain loxodromic elements: \[pro:finitelysimple\] Let $\mathrm{G}$ be a finitely generated subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. If $\mathrm{G}$ contains a loxodromic element, then $\mathrm{G}$ is not simple. To prove it we need the following statement. \[pro:degxie\] Let $\mathrm{G}$ be a finitely generated subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. There exist a finite field $\Bbbk$ and a non trivial morphism $\upsilon\colon\mathrm{G}\to\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ such that for any $\phi$ in $\mathrm{G}$ the following inequality holds: $\deg\upsilon(\phi)\leq\deg\phi$. Let $\phi$ be a loxodromic element of $\mathrm{G}$. If $\phi^n$ is tight in $\mathrm{G}$ for some integer $n$, then Theorem \[thm:canlam\] allows to conclude. If no power of $\phi$ is tight, then $\mathrm{G}$ contains an infinite subgroup $\Delta_2$ that is normalized by $\phi$ and that is conjugate either to a subgroup of $\mathrm{D}_2$, or to a subgroup of $\mathbb{C}^2$ (Theorem \[thm:pfou\]). In particular the degrees of the elements of $\Delta_2$ are uniformly bounded by an integer $N$. According to Proposition \[pro:degxie\] there exist a finite field $\Bbbk$ and a non trivial morphism $\upsilon\colon\mathrm{G}\to\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ such that for all $\phi$ in $\mathrm{G}$ $$\deg\upsilon(\phi)\leq\deg\phi.$$ In $\mathrm{Bir}(\mathbb{P}^2_\Bbbk)$ there exist only finitely many elements of degree $\leq N$. As a result $\upsilon(\Delta_2)$ is finite. The morphism $\upsilon$ has thus a proper kernel and $\mathrm{G}$ is not simple: contradiction. We now have the following alternative - $\mathrm{G}$ contains a parabolic element, - $\mathrm{G}$ is an elliptic subgroup. Let us look at these two possibilities. - If $\mathrm{G}$ contains a parabolic element, then $\mathrm{G}$ is conjugate either to a subgroup of the automorphism group $\mathrm{Aut}(S)$ of a <span style="font-variant:small-caps;">Halphen</span> surface, or to a subgroup of the <span style="font-variant:small-caps;">Jonquières</span> group $\mathcal{J}$. - Assume first that, up to conjugacy, $\mathrm{G}\subset\mathrm{Aut}(S)$ where $S$ is a <span style="font-variant:small-caps;">Halphen</span> surface. Recall that a group $\mathrm{G}$ satisfies <span style="font-variant:small-caps;">Malcev</span> property if every finitely generated subgroup $\Gamma$ of $\mathrm{G}$ is residually finite, [i.e.]{} for any $g\in\Gamma$ there exist a finite group $\mathrm{H}$ and a morphism $\upsilon\colon\Gamma\to\mathrm{H}$ such that $g$ does not belong to $\ker\upsilon$. <span style="font-variant:small-caps;">Malcev</span> showed that linear groups satisfy this property ([@Malcev]). In [@BassLubotzky] the authors proved that automorphism groups of scheme over any commutative ring also satisfy this property. Consequently if $\mathrm{G}$ contains a parabolic element, then $\mathrm{G}$ is, up to conjugacy, a subgroup of $\mathcal{J}$. - Suppose that $\mathrm{G}\subset\mathcal{J}$ up to birational conjugacy. Then $\mathrm{G}$ is finite. Indeed: \[lem:mot\] Let $\mathcal{C}$ be a curve and let $\mathrm{G}\subset\mathrm{Bir}(\mathbb{P}^1_\mathbb{C}\times\mathcal{C})$ be a finitely generated simple subgroup that preserves the $\mathbb{P}^1_\mathbb{C}$-fibration given by the projection to $\mathcal{C}$. Then $\mathrm{G}$ is finite. The group $\mathrm{G}$ being simple, $\mathrm{G}$ is isomorphic either to a subgroup of $\mathrm{PGL}(2,\mathbb{C})$, or to a subgroup of $\mathrm{Aut}(\mathcal{C})$. But both $\mathrm{PGL}(2,\mathbb{C})$ and $\mathrm{Aut}(\mathcal{C})$ satisfy <span style="font-variant:small-caps;">Malcev</span> property so $\mathrm{G}$ is finite. - It remains to look at $\mathrm{G}$ when $\mathrm{G}$ is a finitely generated, simple, elliptic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. Proposition \[pro:etaumilieuCantat\] asserts that either $\mathrm{G}$ is conjugate to a subgroup of $\mathcal{J}$, or $\mathrm{G}$ is contained in an algebraic subgroup of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$. In the first case Lemma \[lem:mot\] allows to conclude. Let us focus on the last case: algebraic subgroups of $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ are linear hence $\mathrm{G}$ is linear and therefore finite since linear groups satisfy <span style="font-variant:small-caps;">Malcev</span> property. Big subgroups of automorphisms “of positive entropy” ==================================================== In this chapter we will construct embeddings of the group $\mathrm{SL}(2,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ such that every non-torsion element in the image has positive entropy. The elements of $\mathrm{SL}(2,\mathbb{Z})$ are classified into elliptic, parabolic and loxodromic elements, with respect to their action on the hyperbolic upper-plane, or similarly to their trace. As we have seen the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group naturally acts on a hyperbolic space of infinite dimension so there is a notion of elliptic, parabolic and loxodromic elements in this group. A morphism $\theta$ from $\mathrm{SL}(2,\mathbb{Z})$ to $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is loxodromic if each element of $\mathrm{im}\,\theta$ of infinite order is loxodromic. An embedding of $\mathrm{SL}(2,\mathbb{Z})$ into the <span style="font-variant:small-caps;">Cremona</span> group is regularizable if it comes from an embedding into the group of automorphisms of a projective rational surface. Note that the existence of loxodromic automorphisms preserving an elliptic curve was not clear. But [@BlancDeserti:embeddings] gives an answer to this question; more precisely the following statement yields. \[thm:embeddings\] There exist loxodromic embeddings $\theta_{h,1}$, $\theta_{h,2}$, $\theta_{h,3}$ of $\mathrm{SL}(2,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ such that: - for each $i$, the group $\theta_{h,i}(\mathrm{SL}(2,\mathbb{Z}))$ preserves a smooth cubic curve $\Gamma\subset\mathbb{P}^2_\mathbb{C}$; - the action of $\theta_{h,1}$ on $\Gamma$ is trivial, and the blow up $S_1\to\mathbb{P}^2_\mathbb{C}$ of $12$ points of $\Gamma$ conjugates $\theta_{h,1}\big(\mathrm{SL}(2,\mathbb{Z})\big)$ to a subgroup of $\mathrm{Aut}(S_1)$; - the action of $\theta_{h,2}$ on $\Gamma$ is generated by a translation of order $3$, and the blow up $S_2\to\mathbb{P}^2_\mathbb{C}$ of $10$ points of $\Gamma$ conjugates $\theta_{h,2}\big(\mathrm{SL}(2,\mathbb{Z})\big)$ to a subgroup of $\mathrm{Aut}(S_2)$; - the action of $\theta_{h,3}$ on $\Gamma$ is generated by an automorphism of order $3$ with fixed points, and the blow up $S_3\to\mathbb{P}^2_\mathbb{C}$ of $10$ points of $\Gamma$ conjugates $\theta_{h,3}\big(\mathrm{SL}(2,\mathbb{Z})\big)$ to a subgroup of $\mathrm{Aut}(S_3)$; - for $i=1$, $2$, $3$ the strict transform $\widetilde{\Gamma}$ of $\Gamma$ on $S_i$ is the only invariant curve. Furthermore in cases $i=1$, $2$ we can choose $\Gamma$ to be any smooth cubic curve and this yields infinitely many loxodromic embeddings of $\mathrm{SL}(2,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ up to conjugacy. This problem is an opportunity to evoke the following natural question: \[ques:biraut\] When is a birational self map of a complex projective surface birationally conjugate to an automorphism ? In the first section we introduce the notion of topological entropy, recall which complex compact surfaces can carry a biholomorphism of positive topological entropy and obstructions to the existence of such biholomorphisms on rational surfaces. The second section is devoted to some properties of automorphisms of <span style="font-variant:small-caps;">Del Pezzo</span> surfaces. The third section deals with the study of the image of the central involution by an embedding of $\mathrm{SL}(2,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$, the idea of the proof of the existence of infinitely many loxodromic embeddings of $\mathrm{SL}(2,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ and an outline of the proof of Theorem \[thm:embeddings\]. We finish the chapter by giving three answers to Question \[ques:biraut\]. Two are related with the first dynamical degree, the last one involves another birational invariant called the dynamical number of base points that we will introduce. Topological entropy ------------------- Let $V$ be a complex projective manifold, and let $\phi$ be a rational or holomorphic map on $V$. We can consider the dynamical system induced by $\phi$: the point $p$ moves to $\phi(p)$ that moves to $\phi^2(p)$ etc In the $60$’s <span style="font-variant:small-caps;">Adler</span>, <span style="font-variant:small-caps;">Konheim</span> and <span style="font-variant:small-caps;">Mc Andrew</span> introduce a new way to measure the complexity of a dynamical system: the topological entropy ([@AdlerKonheimMcAndrew]). Let $X$ be a compact metric space, and let $\phi$ be a continuous map from $X$ to itself. Take a strictly positive real number $\varepsilon$. For all integer $n$ consider the minimal cardinal $N(n,\varepsilon)$ of a part $X_n$ of $X$ such that for all $q$ in $X$ there exists $p$ in $X$ satisfying $$\mathrm{dist}\big(\phi^j(p),\phi^j(q)\big)\leq\varepsilon\qquad\forall\,0\leq j\leq n.$$ Set $$\mathrm{h_{top}}(\phi,\varepsilon)=\limsup_{n\to +\infty}\frac{1}{n}\log N(n,\varepsilon).$$ The topological entropy of $\phi$ is given by $$\mathrm{h_{top}}(\phi)=\lim_{\varepsilon\to 0}\mathrm{h_{top}}(\phi,\varepsilon).$$ For instance the topological entropy of an isometry of $X$ is zero. If $\phi$ is the multiplication by an integer $n>1$ on a complex Abelian variety of dimension $k$, then $\mathrm{h_{top}}(\phi)=2k\log n$. If $\phi$ is an endomorphism of $\mathbb{P}^k_\mathbb{C}$ defined by homogeneous polynomials of degree $d$ without common factor of positive degree, then $\mathrm{h_{top}}(\phi)=k\log d$ (see [@Gromov]). Let $V$ be a compact <span style="font-variant:small-caps;">Kähler</span> manifold; consider $\mathrm{Aut}(V)$ its automorphisms group, and $\mathrm{Aut}^0(V)$ the connected component of $\mathrm{Aut}(V)$ which contains the identity element. The topological entropy of any element of $\mathrm{Aut}^0(V)$ is zero. Let $V$ be a complex manifold. Let $\Omega^{n,m}$ be the vector bundle of complex differential forms of degree $(n,m)$ on $V$. The <span style="font-variant:small-caps;">Dolbeault</span> operator is defined as a differential operator on smooth sections $$\overline{\partial}\colon\Gamma(\Omega^{n,m},V)\to\Gamma(\Omega^{n,m+1}(V)).$$ The <span style="font-variant:small-caps;">Dolbeault</span> cohomology group $H^{n,m}(V)$ is the quotient space $$H^{n,m}(V)=\frac{\ker\big(\overline{\partial}\colon\Gamma(\Omega^{n,m},V)\to\Gamma(\Omega^{n,m+1},V)\big)}{\overline{\partial}\Gamma(\Omega^{n,m+1},V)}.$$ Let $H^k(V,\mathbb{Z})$, $H^k(V,\mathbb{R})$ and $H^k(V,\mathbb{C})$ be the cohomology groups associated to the constant sheaves $\mathbb{Z}$, $\mathbb{R}$ and $\mathbb{C}$ respectively. Let $\phi$ be a map of class $\mathcal{C}^\infty$ on a compact manifold $V$, the topological entropy is greater than the logarithm of the spectral radius $r(\phi^)$ of the linear map induced by $\phi$ on the direct sum of the cohomological groups of $V$: $$\mathrm{h_{top}}(\phi)\geq\log r(\phi^).$$ Assume $V$ is <span style="font-variant:small-caps;">Kähler</span>. For any $0\leq k\leq\dim_\mathbb{C}V$ we denote by $\lambda_p(\phi)$ the spectral radius of the map $\phi^$ acting on the <span style="font-variant:small-caps;">Dolbeault</span> chomological group $H^{k,k}(V,\mathbb{R})$. Let $\phi$ be a holomorphic map on a compact complex kähler manifold $V$. Then $$\mathrm{h_{top}}(\phi)=\max_{0\leq k\leq\dim_{\mathbb{C}}V}\log\lambda_k(\phi).$$ The topological entropy of $\phi$ is strictly greater than $1$ if and only if one of the $\lambda_k(\phi)$’s is, and in fact if and only if $\lambda_1(\phi)=\lambda(\phi)$ is. On which conditions do rational maps of $V$ with chaotic behavior exist ? The existence of such rational maps implies a lot of constraints on $V$; let us for instance mention the following result: Let $n\geq 3$ be an integer. A smooth hypersurface of degree $\geq 3$ in $\mathbb{P}^n_\mathbb{C}$ admits no endomorphism of degree greater than $1$. Since the case of quadrics was treated in [@ParanjapeSrinivas] this settles the question of endomorphisms of hypersurfaces. Let us focus on the $2$-dimensional case. Recall that a K$3$ surface[^6] is a complex, compact, simply connected surface $S$ with a trivial canonical bundle. Equivalently there exists a holomorphic $2$-form $\omega$ on $S$ which is never zero; $\omega$ is unique modulo multiplication by a scalar. Let $S$ be a K$3$ surface with a holomorphic involution $\iota$. If $\iota$ has no fixed point, the quotient of $S$ by $\langle\iota\rangle$ is an <span style="font-variant:small-caps;">Enriques</span> surface, otherwise it is a rational surface. Recall that every non-singular rational surface can be obtained by repeatedly blowing up a minimal rational surface. The minimal rational surfaces* are the complex projective plane, $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ and the <span style="font-variant:small-caps;">Hirzebruch</span> surfaces $\mathbb{F}_n$, $n\geq 2$. If $S$ is a complex, compact surface carrying a biholomorphism of positive topological entropy, then $S$ is either a complex torus, or a K$3$ surface, or an <span style="font-variant:small-caps;">Enriques</span> surface, or a non-minimal rational surface ([@Cantat2]). Although automorphisms of complex tori are easy to describe, it is rather difficult to construct automorphisms on K$3$ surfaces or rational surfaces. Constructions and dynamical properties of automorphisms of K$3$ surfaces can be found in [@Cantat3] and [@McMullen2]. The first examples of rational surfaces endowed with biholomorphisms of positive entropy are due to <span style="font-variant:small-caps;">Coble</span> and <span style="font-variant:small-caps;">Kummer</span> ([@Coble]): - the <span style="font-variant:small-caps;">Coble</span> surfaces are obtained by blowing up the ten nodes of a nodal sextic in $\mathbb{P}^2_\mathbb{C}$; - the <span style="font-variant:small-caps;">Kummer</span> surfaces are desingularizations of quotients of complex $2$-tori by involutions with fixed points. Obstructions to the existence of such biholomorphisms on rational surfaces are also known: if $\phi$ is a biholomorphism of a rational surface $S$ such that $\mathrm{h_{top}}(\phi)>0$, then the representation $$\begin{aligned} &\mathrm{Aut}(S)\to\mathrm{GL}(\mathrm{Pic}(S)) && g\mapsto g^\end{aligned}$$ has infinite image. Hence according to [@Harbourne] its kernel is finite, so that $S$ has no non-zero holomorphic vector field. A second obstruction follows from [@Nagata]: the surface $S$ has to be obtained by successive blowups from the complex projective plane and the number of blowups must be at least ten. The first infinite families of examples have been constructed independently in [@McMullen] and [@BedfordKim1] by different methods. In [@BedfordKim2] <span style="font-variant:small-caps;">Bedford</span> and <span style="font-variant:small-caps;">Kim</span> constructed arbitrary big holomorphic families of rational surfaces endowed with biholomorphisms of positive entropy: Let $n\geq 3$ and $k\geq 2$ be two integers such that $n$ is odd and $(n,k)\not=(3,2)$. There exists a non-empty set $C_k$ of $\mathbb{R}$ such that if $c\in C_k$ and $a=(a_2,a_4,\ldots,a_{[(n-3)/2]})\in\mathbb{C}^{[(n-3)/2]}$, then the map $$(z_0:z_1:z_2)\dashrightarrow\Big(z_0z_2^{n-1}:z_2^n:z_0^n-z_1z_2^{n-1}+cz_2^n+\displaystyle\sum_{\stackrel{\ell=2}{\ell\text{ even}}}^{n-3}a_\ell z_0^{k+1}z_2^{n-\ell-1}\Big)$$ can be lifted to an automorphism of a rational surface $S_a$ with positive topological entropy. The surfaces $S_a$ are obtained by blowing up $k$ infinitely near points of length $2n-1$ on the invariant line $\{z_0=0\}$ and form a holomorphic family over the parameters space given by the $a_j's$. If $k=2$ and $n\geq 4$ is even, then for $a$ and $b$ near $0$ the surfaces $S_a$ and $S_b$ are holomorphic if and only if $a=b$. Automorphisms of <span style="font-variant:small-caps;">del Pezzo</span> surfaces --------------------------------------------------------------------------------- Any <span style="font-variant:small-caps;">del Pezzo</span> surface $S$ contains a finite number of $(-1)$-curves (i.e.* smooth curves isomorphic to $\mathbb{P}^1_\mathbb{C}$ and of self-intersection $-1$). Each of them can be contracted to get another <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $(K_S)^2+1$. There are, moreover, the only reducible curves of $S$ of negative self-intersection. If $S\not=\mathbb{P}^2_\mathbb{C}$, then there is a finite number of conic bundles $S\to\mathbb{P}^1_\mathbb{C}$ up to automorphism of $\mathbb{P}^1_\mathbb{C}$ and each of them has exactly $8-(K_S)^2$ singular fibers. This latter fact can be found by contracting one component in each singular fiber which is the union of two $(-1)$-curves, obtaining a line bundle on a <span style="font-variant:small-caps;">del Pezzo</span> surface, isomorphic to $\mathbb{P}^1_\mathbb{C}\times\mathbb{P}^1_\mathbb{C}$ or to the <span style="font-variant:small-caps;">Hirzebruch</span> surface $\mathbb{F}_1$ and having degree $8$. For more details see [@Demazure:sousgroupesalgebriques; @Manin]. ### Automorphisms of <span style="font-variant:small-caps;">del Pezzo</span> surfaces of order $4$ {#automorphisms-of-del-pezzo-surfaces-of-order-4 .unnumbered} Set $$S=\big\{(z_0:z_1:z_2:z_3)\in\mathbb{P}(2,1,1,1)\,\vert\,z_0^2-z_1^4=z_2z_3(z_2+z_3)(z_2+\mu z_3)\big\}$$ where $\mu$ belongs to $\mathbb{C}\smallsetminus\{0,\,1\}$. The surface $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> one of degree $2$. The automorphism $\beta$ given by $$\beta\colon(z_0:z_1:z_2:z_3)\mapsto(z_0:\mathbf{i}z_1:z_2:z_3)$$ fixes pointwise the elliptic curve given by $z_0=0$. When $\mu$ varies all possible elliptic curves are obtained. Moreover $\mathrm{rk}\,\mathrm{Pic}(S)^{\beta}=1$. There are other automorphisms $\beta$ of order $4$ of rational surfaces $S$ such that $\beta^2$ fixes an elliptic curve but none for which $\mathrm{rk}\,\mathrm{Pic}(S)^{\beta}=1$ (see [@Blanc:cyclic]). ### Automorphisms of <span style="font-variant:small-caps;">del Pezzo</span> surfaces of order $6$ {#automorphisms-of-del-pezzo-surfaces-of-order-6 .unnumbered} Let us give explicit possibilities for automorphisms of order $6$. - Set $$S=\big\{(z_0:z_1:z_2:z_3)\in\mathbb{P}(3,1,1,2)\,\vert\,z_0^2=z_3^3+\mu z_1^4z_3+z_1^6+z_2^6\big\}$$ for some general $\mu\in\mathbb{C}$ such that $S$ is smooth. The surface $S$ is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $1$. Consider on $S$ $$\alpha\colon(z_0:z_1:z_2:z_3)\mapsto(z_0:z_1:-\mathbf{j}z_2:z_3)$$ where $\mathbf{j}=\mathrm{e}^{2\mathbf{i}\pi/3}$. The automorphism $\alpha$ fixes pointwise the elliptic curve given by $z_2=0$. When $\mu$ varies all possible elliptic curves are obtained. The equality $\mathrm{rk}\,\mathrm{Pic}(S)^\alpha=1$ holds (see [@DolgachevIskovskikh Corollary 6.11]). - Set $$S=\big\{(z_0:z_1:z_2:z_3)\in\mathbb{P}^3_\mathbb{C}\,\vert\,z_0z_1^2+z_0^3+z_2^3+z_3^3+\mu z_0z_2z_3=0\big\}$$ where $\mu$ is such that the cubic surface is smooth. The surface is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $3$. Consider on $S$ the automorphism $\alpha$ given by $$\alpha\colon(z_0:z_1:z_2:z_3)\mapsto(z_0:-z_1:\mathbf{j}z_2:\mathbf{j}^2z_3).$$ Remark that $\alpha^3$ fixes pointwise the elliptic curve $z_1=0$ and $\alpha$ acts on it via a translation of order $3$. When $\mu$ varies all possible elliptic curves are obtained. The equality $\mathrm{rk}\,\mathrm{Pic}(S)^{\alpha}=1$ holds ([@DolgachevIskovskikh]). - Set $$S=\big\{(z_0:z_1:z_2:z_3)\in\mathbb{P}^3_\mathbb{C}\,\vert\, z_0^3+z_1^3+z_2^3+(z_1+\mu z_2)z_3^2=0\big\}$$ where $\mu\in\mathbb{C}$ is such that the cubic surface is smooth. It is a <span style="font-variant:small-caps;">del Pezzo</span> surface of degree $3$. Consider $\alpha$ defined by $\alpha\colon(z_0:z_1:z_2:z_3)\mapsto(\mathbf{j}z_0:z_1:z_2:z_3)$. The automorphism $\alpha^3$ fixes pointwise the elliptic curve $z_3=0$ and $\alpha$ acts on it via an automorphism of order $3$ with three fixed points. When $\mu$ varies the birational class of $\alpha$ changes but not the isomorphism class of the elliptic curve fixed by $\alpha^3$. Outline of the construction --------------------------- ### The central involution of $\mathrm{SL}(2,\mathbb{Z})$ and its image into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ Set $A=\left( \begin{array}{cc} 1 & 1 \\ 0 & 1 \end{array} \right)$ and $B=\left( \begin{array}{cc} 0 & 1 \\ -1 & 0 \end{array} \right).$ A presentation of $\mathrm{SL}(2,\mathbb{Z})$ is given by (see [@Newman]) $$\langle A,\, B\,\vert\,B^4=(AB)^3=1,\,B^2(AB)=(AB)B^2\rangle.$$ As as result the quotient of $\mathrm{SL}(2,\mathbb{Z})$ by its center is a free product of $\faktor{\mathbb{Z}}{2\mathbb{Z}}$ and $\faktor{\mathbb{Z}}{3\mathbb{Z}}$ generated by the classes $[B]$ of $B$ and $[AB]$ of $AB$ $$\mathrm{PSL}(2,\mathbb{Z})=\langle [B],\, [AB]\,\vert\, [B]^2= [AB]^3=\mathrm{id}\rangle.$$ Recall that $\mathrm{SL}(2,\mathbb{R})$ acts on the upper half plane $$\mathbb{H}=\big\{x+\mathbf{i}y\in\mathbb{C}\,\vert\, x,\, y\in\mathbb{R},\, y>0\big\}$$ by <span style="font-variant:small-caps;">Möbius</span> transformations $$\begin{aligned} &\mathrm{SL}(2,\mathbb{R})\times\mathbb{H}\to\mathbb{H}, && \left(\left( \begin{array}{cc} a & b\\ c & d \end{array} \right),z\right)\mapsto\frac{az+b}{cz+d}\end{aligned}$$ the hyperbolic structure of $\mathbb{H}$ being preserved. This yields to a natural notion of elliptic, parabolic and loxodromic elements of $\mathrm{SL}(2,\mathbb{R})$. If $M$ belongs to $\mathrm{SL}(2,\mathbb{Z})$ one can be more precise and check the following observations: - $M$ is elliptic if and only if $M$ has finite order; - $M$ is parabolic if and only if $M$ has infinite order and its trace is $\pm 2$; - $M$ is loxodromic if and only if $M$ has infinite order and its trace is $\not=\pm 2$. Up to conjugacy the elliptic elements of $\mathrm{SL}(2,\mathbb{Z})$ are $$\begin{aligned} & \left( \begin{array}{cc} -1 & 0 \\ 0 & -1 \end{array} \right), && \left( \begin{array}{cc} 0 & 1\\ -1 & -1 \end{array} \right), && \left( \begin{array}{cc} 0 & 1 \\ -1 & 0 \end{array} \right), && \left( \begin{array}{cc} 0 & -1\\ 1 & 0 \end{array} \right), && \left( \begin{array}{cc} 0 & -1 \\ 1 & 1 \end{array} \right).\end{aligned}$$ In particular an element of finite order is of order $2$, $3$, $4$ or $6$. A parabolic element of $\mathrm{SL}(2,\mathbb{Z})$ is up to conjugacy one of the following one $$\begin{aligned} & \left( \begin{array}{cc} 1 & a \\ 0 & 1 \end{array} \right) && \left( \begin{array}{cc} -1 & a \\ 0 & -1 \end{array} \right)\end{aligned}$$ with $a\in\mathbb{Z}$. Since $B^2\in\mathrm{SL}(2,\mathbb{Z})$ is an involution its image by any embedding $\theta\colon\mathrm{SL}(2,\mathbb{Z})\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ is a birational involution. As we have seen in §\[sec:ordre2\] an element of order $2$ of the <span style="font-variant:small-caps;">Cremona</span> group is up to conjugacy one of the following - an automorphism of $\mathbb{P}^2_\mathbb{C}$, - a <span style="font-variant:small-caps;">Jonquières</span> involution of degree $\geq 2$, - a <span style="font-variant:small-caps;">Bertini</span> involution, - a <span style="font-variant:small-caps;">Geiser</span> involution. Since $B^2$ commutes with $\mathrm{SL}(2,\mathbb{Z})$ the group $\theta\big(\mathrm{SL}(2,\mathbb{Z})\big)$ is contained in the centralizer of $\theta(B^2)$. But if $\theta(B^2)$ is a <span style="font-variant:small-caps;">Bertini</span> involution or a <span style="font-variant:small-caps;">Geiser</span> involution, then the centralizer of $\theta(B^2)$ is finite ([@BlancPanVust]). As a result $\theta(B^2)$ is conjugate either to an automorphism of $\mathbb{P}^2_\mathbb{C}$, or to a <span style="font-variant:small-caps;">Jonquières</span> involution. Assume that $\theta(B^2)$ is not linearisable; $\theta(B^2)$ fixes thus pointwise a unique irreducible curve $\Gamma$ of genus $\geq 1$. Denote by $\mathrm{G}$ the image of $\theta$. The group $\mathrm{G}$ preserves $\Gamma$ and the action of $\mathrm{G}$ on $\Gamma$ gives the exact sequence $$1 \longrightarrow\mathrm{G}'\longrightarrow\mathrm{G}\longrightarrow \mathrm{H}\longrightarrow 1$$ where $\mathrm{H}$ is a subgroup of $\mathrm{Aut}(\Gamma)$ and $\mathrm{G}'$ contains $\theta(B^2)$ and fixes $\Gamma$. The genus of $\Gamma$ is positive; hence $\mathrm{H}$ cannot coincide with $\faktor{\mathrm{G}}{\langle\theta(B^2)\rangle}$, a free product of $\faktor{\mathbb{Z}}{2\mathbb{Z}}$ and $\faktor{\mathbb{Z}}{3\mathbb{Z}}$. As a consequence $\mathrm{G}'\triangleleft\mathrm{G}$ strictly contains $\langle\theta(B^2)\rangle$ and thus $\mathrm{G}'$ is infinite and not abelian. In particular the group of birational maps fixing pointwise $\Gamma$ is infinite and not abelian. So according to [@BlancPanVust2] the curve $\Gamma$ has genus $1$. One can now state: Let $\theta$ be an embedding of $\mathrm{SL}(2,\mathbb{Z})$ into the $2$-dimensional <span style="font-variant:small-caps;">Cremona</span> group. Then up to birational conjugacy - either $\theta(B^2)$ is an automorphism of $\mathbb{P}^2_\mathbb{C}$ of order $2$, - or $\theta(B^2)$ is a <span style="font-variant:small-caps;">Jonquières</span> involution of degree $3$ fixing $($pointwise$)$ an elliptic curve. ### Existence of infinitely many loxodromic embeddings of $\mathrm{SL}(2,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ Let us consider the standard embedding $$\begin{aligned} & \theta_e\colon\mathrm{SL}(2,\mathbb{Z})\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}) && \left( \begin{array}{cc} a & b \\ c & d \end{array} \right)\mapsto \Big((z_0:z_1:z_2)\mapsto(az_0+bz_1:cz_0+dz_1:z_2)\Big).\end{aligned}$$ Note that $\theta_e\big(\mathrm{SL}(2,\mathbb{Z})\big)$ is a subgroup of $\mathrm{PGL}(3,\mathbb{C})$ that preserves the line $L_{z_2}$ of equation $z_2=0$ and acts on it via the maps $$\mathrm{SL}(2,\mathbb{Z})\to\mathrm{PSL}(2,\mathbb{Z})\subset\mathrm{PSL}(2,\mathbb{C})=\mathrm{Aut}(L_{z_2}).$$ Pick $\mu\in\mathbb{C}^$ such that the point $p=(\mu:1:0)\in L_{z_2}$ has a trivial isotropy group under the action of $\mathrm{PSL}(2,\mathbb{Z})$. Fix an even integer $k>0$; consider $\psi$ the conjugation of $$\psi'\colon(z_0:z_1:z_2)\dashrightarrow(z_0^k:z_0^{k-1}z_1+z_2^k:z_0^{k-1}z_2)$$ by $(z_0:z_1:z_2)\mapsto(z_0+\mu z_1:z_1:z_2)$. Then define the morphism $\theta_k\colon\mathrm{SL}(2,\mathbb{Z})\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ as follows $$\begin{aligned} &\theta_k(B)=\theta_e(B)\colon(z_0:z_1:z_2)\mapsto(z_1:-z_0:z_2) &&\theta_k(AB)=\psi\circ\theta_e(AB)\circ\psi^{-1}.\end{aligned}$$ The map $\psi'$ restricts to an automorphism of the affine plane where $z_0\not=0$, commutes with $\theta_k(B^2)=\theta_e(B^2)=(z_0:z_1:-z_2)\in\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ and acts trivially on $L_{z_2}$. Since $\psi$ commutes with $\theta_k(B^2)$ the map $\theta_k(AB)$ commutes with $\theta_k(B^2)$. As a result $\theta_k$ is a well-defined morphism. As $\psi_{\vert L_{z_2}\smallsetminus\{p\}}=\mathrm{id}$ the actions of $\theta_e$ and $\theta_k$ on $L_{z_2}$ are the same; $\theta_k$ is thus an embedding. \[lem:dynamicaldegree\] Let $n$ be a positive integer. Let $a_1$, $\ldots$, $a_n$, $b_1$, $\ldots$, $b_n$ be $2n$ elements in $\{-1,\,1\}$. The birational self map of $\mathbb{P}^2_\mathbb{C}$ $$\theta_k\big(B^{b_n}(AB)^{a_n}B^{b_{n-1}}(AB)^{a_{n-1}}\ldots B^{b_1}(AB)^{a_1}\big)$$ has degree $k^{2n}$ and has exactly $2n$ proper base-points, all lying on $L_{z_2}$. More precisely the base-points are $$p,\,\big((AB)^{a_1}\big)^{-1}(p),\,\big(B^{b_1}(AB)^{a_1}\big)^{-1}(p),$$ $$\big((AB)^{a_2}B^{b_1}(AB)^{a_1}\big)^{-1}(p),\,\ldots,\,\big((AB)^{a_n}B^{b_{n-1}}(AB)^{a_{n-1}}\ldots B^{b_1}(AB)^{a_1}\big)^{-1}(p),$$ $$\big(B^{b_n}(AB)^{a_n}B^{b_{n-1}}(AB)^{a_{n-1}}\ldots B^{b_1}(AB)^{a_1}\big)^{-1}(p).$$ This result implies the existence of infinitely many loxodromic embeddings of $\mathrm{SL}(2,\mathbb{Z})$ into $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$: Let $n$ be a positive integer. Let $a_1$, $a_2$, $\ldots$, $a_n$, $b_1$, $b_2$, $\ldots$, $b_n$ be $2n$ elements in $\{-1,\,1\}$. The birational self map of $\mathbb{P}^2_\mathbb{C}$ $$\theta_k\big(B^{b_n}(AB)^{a_n}B^{b_{n-1}}(AB)^{a_{n-1}}\ldots B^{b_1}(AB)^{a_1}\big)$$ has dynamical degree $k^{2n}$. In particular, $\theta_k$ is a loxodromic embedding and $$\big\{\lambda(\phi)\,\vert\,\phi\in\theta_k\big(\mathrm{SL}(2,\mathbb{Z})\big)\big\}=\big\{1,\,k^2,\,k^4,\,k^6,\,\ldots\big\}.$$ Let us consider an element of infinite order of $\mathrm{SL}(2,\mathbb{Z})$; it is conjugate to $$\varphi=B^{b_n}(AB)^{a_n}B^{b_{n-1}}(AB)^{a_{n-1}}\ldots B^{b_1}(AB)^{a_1}$$ where $a_1$, $a_2$, $\ldots$, $a_n$, $b_1$, $b_2$, $\ldots$, $b_n\in\{-1,\,1\}$. According to Lemma \[lem:dynamicaldegree\] the degree of $\theta_k(\varphi^r)$ is equal to $k^{2nr}$. As a consequence $\lambda\big(\theta_k(\varphi)\big)=k^{2n}$. We proceed by induction on $n$. Let us detail the case $n=~1$. The birational map $\psi$ has degree $k$ and has a unique proper base-point $p=(\mu:1:0)\in L_{z_2}$. The same holds for $\psi^{-1}$. Moreover $\psi_{\vert L_{z_2}\smallsetminus\{p\}}=\psi^{-1}_{\vert L_{z_2}\smallsetminus\{p\}}=\mathrm{id}$. Since $\theta_e(AB)^{a_1}\in\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$ moves the point $p$ onto another point of $L_{z_2}$, the map $\theta_k\big((AB)^{a_1}\big)$ has degree $k^2$ and exactly two proper base-points which are $p$ and $\big((AB)^{a_1}\big)^{-1}(p)=(\psi\circ\theta_e)(AB)^{-a_1}$. As $\theta_k(S)$ belongs to $\mathrm{Aut}(\mathbb{P}^2_\mathbb{C})$, $\theta_k(B^{b_1}(AB)^{a_1})$ has also degree $k^2$ and two proper base-points which are $p$ and $\big((AB)^{a_1}\big)^{-1}(p)$. ### Description of loxodromic embeddings for which the central element fixes (pointwise) an elliptic curve Let us note that $$\mathrm{SL}(2,\mathbb{Z})=\langle\alpha,\,\beta\,\vert\,\beta^4=\mathrm{id},\,\alpha^3=\beta^2\rangle$$ (take the presentation we gave before and set $\alpha^2=AB$, $\beta=B$) and that $$\mathrm{SL}(2,\mathbb{Z})=\langle\alpha,\,\beta\,\vert\,\alpha^6=\beta^4=\alpha^3\beta^2=\mathrm{id}\rangle.$$ In this section we will use this last presentation. We say that a curve is fixed* by a birational map if it is pointwise fixed, and say that a curve is invariant or preserved if the map induces a birational action on the curve. All conjugacy classes of elements of order $4$ and $6$ in $\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ have been classified in [@Blanc:commentarii]. Many of them can act on <span style="font-variant:small-caps;">del Pezzo</span> surfaces of degree $1$, $2$, $3$ or $4$. <span style="font-variant:small-caps;">del Pezzo</span> surfaces $X$, $Y$ of degree $\leq 4$ and automorphisms $\alpha\in\mathrm{Aut}(X)$, resp. $\beta\in\mathrm{Aut}(Y)$ of order $6$, resp. $4$ so that - $\alpha^3$ and $\beta^2$ fix pointwise an elliptic curve, - and that $\mathrm{Pic}(X)^{\alpha}$, $\mathrm{Pic}(Y)^{\beta}$ both have rank $1$ are defined to create the embedding. Contracting $(-1)$-curves invariant by the involutions $\alpha^3$ and $\beta^2$ (but not by $\alpha$, $\beta$ which act minimally on $X$ and $Y$) we get rational morphisms $X\to X_4$ and $Y\to Y_4$ where $X_4$, $Y_4$ are <span style="font-variant:small-caps;">del Pezzo</span> surfaces on which $\alpha^3$ and $\beta^2$ act minimally. Furthermore $X_4$ and $Y_4$ are <span style="font-variant:small-caps;">del Pezzo</span> surfaces of degree $4$, both $\mathrm{Pic}(X_4)^{\alpha^3}$ and $\mathrm{Pic}(Y_4)^{\beta^2}$ have rank $2$ and are generated by the fibers of the two conic bundles on $X_4$ and $Y_4$. Choosing a birational map $X_4\dashrightarrow Y_4$ conjugating $\alpha^3$ to $\beta^2$ (which exists if and only if the elliptic curves are isomorphic), which is general enough, we obtain a loxodromic embedding $$\mathrm{SL}(2,\mathbb{Z})\to\mathrm{Bir}(\mathbb{P}^2_\mathbb{C}).$$ To prove that there is no other relation in $\langle\alpha,\,\beta\rangle$ and that all elements of infinite order are loxodromic the morphisms $X\to X_4$ and $Y\to Y_4$ and the actions of $\alpha$ and $\beta$ on $\mathrm{Pic}(X)^{\alpha^3}$ and $\mathrm{Pic}(Y)^{\beta^2}$ are described ; furthermore the composition of the elements does what is expected. Birational maps and automorphisms --------------------------------- ### Definitions Given a birational map $\phi\colon S\dashrightarrow S$ of a projective complex surface its dynamical degree $\lambda(\phi)$ is a positive real number that measures the complexity of the dynamics of $\phi$ (see §\[sec:degreegrowth\]). The neperian logarithm $\log\lambda(\phi)$ provides an upper bound for the topological entropy of $\phi\colon S\dashrightarrow S$ and is equal to it under natural assumptions ([@BedfordDiller; @DinhSibony]). Let us give an alternative but equivalent definition to that of §\[sec:degreegrowth\]. A birational map $\phi\colon S\dashrightarrow S$ of a projective complex surface determines an endomorphism $\phi_\colon\mathrm{NS}(S)\to\mathrm{NS}(S)$; the dynamical degree $\lambda(\phi)$ of $\phi$ is defined as the spectral radius of the sequence of endomorphisms $(\phi^n)_$ as $n$ goes to infinity: $$\lambda(\phi)=\displaystyle\lim_{n\to +\infty}\vert\vert(\phi^n)_\vert\vert^{1/n}$$ where $\vert\vert\cdot\vert\vert$ denotes a norm on the real vector space $\mathrm{End}(\mathrm{NS}(S))$. This limit exists and does not depend on the choice of the norm. For any ample divisor $D\subset S$ $$\lambda(\phi)=\displaystyle\lim_{n\to +\infty}(D\cdot (\phi^n)_D)^{1/n}.$$ The <span style="font-variant:small-caps;">Neron</span> <span style="font-variant:small-caps;">Severi</span> group of $\mathbb{P}^2_\mathbb{C}$ coincides with the <span style="font-variant:small-caps;">Picard</span> group of $\mathbb{P}^2_\mathbb{C}$, has rank $1$, and is generated by the class $\mathbf{e}_0$ of a line $$\mathrm{NS}(\mathbb{P}^2_\mathbb{C})=\mathrm{Pic}(\mathbb{P}^2_\mathbb{C})=\mathbb{Z}\mathbf{e}_0.$$ A map $\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ acts on $\mathrm{Pic}(\mathbb{P}^2_\mathbb{C})$ by multiplication by $\deg\phi$. ### <span style="font-variant:small-caps;">Pisot</span> and <span style="font-variant:small-caps;">Salem</span> numbers We will give the definitions of <span style="font-variant:small-caps;">Pisot</span> and <span style="font-variant:small-caps;">Salem</span> numbers, for more details see [@BDGGHPDS]. A <span style="font-variant:small-caps;">Pisot</span> number is an algebraic integer $\lambda\in]1,+\infty[$ whose other <span style="font-variant:small-caps;">Galois</span> conjugates lie in the unit disk. Let us denote by $\mathrm{Pis}$ the set of <span style="font-variant:small-caps;">Pisot</span> numbers. It includes all integers $\geq 2$ as well as all reciprocical quadratic integers $\lambda>1$. The set $\mathrm{Pis}$ is a closed subset of the real line; its infimum is equal to the unique root $\lambda_P>1$ of the cubic equation $x^3=x+1$. The smallest accumultation point of $\mathrm{Pis}$ is the golden mean $\lambda_G=\frac{1+\sqrt{5}}{2}$. Note that all <span style="font-variant:small-caps;">Pisot</span> numbers between $\lambda_P$ and $\lambda_G$ have been listed. A <span style="font-variant:small-caps;">Salem</span> number is an algebraic integer $\lambda\in]1,+\infty[$ whose other Galois conjugates are in the closed unit disk with at least one on the boundary. The minimal polynomial of $\lambda$ has thus at least two complex conjugate roots on the unit circle, its roots are permuted by the involution $z\mapsto \frac{1}{z}$ and has degree at least $4$. Let $\mathrm{Sal}$ be the set of <span style="font-variant:small-caps;">Salem</span> numbers. The unique root $\lambda_L>1$ of the irreducible polynomial $x^{10}+x^9-x^7-x^6-x^5-x^4-x^3+x+1$ is a <span style="font-variant:small-caps;">Salem</span> number. Conjecturally the infimum of $\mathrm{Sal}$ is larger than $1$ and should be equal to $\lambda_L$. Remark that $\mathrm{Pis}$ is contained in the closure of $\mathrm{Sal}$. ### Dynamical degrees and <span style="font-variant:small-caps;">Pisot</span> and <span style="font-variant:small-caps;">Salem</span> numbers Let us recall that a birational map $\phi\colon S\dashrightarrow S$ of a compact complex surface is algebraically stable if $(\phi^)^n=(\phi^n)^$ for all $n\geq 0$ (see § \[sec:degreegrowth\]). If $\phi$ is algebraically stable, then so does $\phi^{-1}$ and $\lambda(\phi)$ is an algebraic integer. Any birational map of a compact complex surface is conjugate by a birational morphism to an algebraically stable map (Proposition \[pro:DillerFavre\]). From this fact and the <span style="font-variant:small-caps;">Hodge</span> index theorem according to which the intersection form has signature $(1,r_S-1)$, where $r_S$ denotes the rank of $S$, <span style="font-variant:small-caps;">Diller</span> and <span style="font-variant:small-caps;">Favre</span> get the following statement: Let $\phi$ be a birational self map of a complex projective surface. If $\lambda(\phi)$ is distinct from $1$, [i.e.]{} if $\phi$ is loxodromic, then $\lambda(\phi)$ is a <span style="font-variant:small-caps;">Pisot</span> or a <span style="font-variant:small-caps;">Salem</span> number. ### When is a birational map conjugate to an automorphism ? A natural question is the following one; when is a birational self map of a complex projective surface birationally conjugate to an automorphism ? There are three answers to this question and we will detail it. #### A first answer <span style="font-variant:small-caps;">Diller</span> and <span style="font-variant:small-caps;">Favre</span> give the first characterization of loxodromic birational maps which are conjugate to an automorphism of a projective surface: \[thm:blabla\] Let $\phi\in\mathrm{Bir}(\mathbb{P}^2_\mathbb{C})$ be a loxodromic map. Assume that $\phi$ is algebraically stable. The action of $\phi$ on $\mathrm{H}^{1,1}(\mathbb{P}^2_\mathbb{C})$ admits the eigenvalue $\lambda(\phi)>1$ with eigenvector $\Theta(\phi)$. The map $\phi$ is birationally conjugate to an automorphism if and only if $\Theta(\phi)\cdot\Theta(\phi)=0$. When $\phi$ is an automorphism, it is easy to check that $\Theta(\phi)\cdot\Theta(\phi)=0$. We will thus deal with the reciprocical property. Let $\phi$ be a birational self map of a complex projective surface $S$. Assume that $\phi$ is algebraically stable. Hence $\lambda(\phi)$ is equal to the spectral radius of $\phi_\in\mathrm{End}(\mathrm{NS}(\mathbb{R},S))$ but also to the spectral radius of $\phi^=(\phi^{-1})_$; indeed these endomorphisms are adjoint for the intersection form: $$\phi_C\cdot D=C\cdot\phi^D$$ for all $C$, $D$ divisor classes. One can factorize $\phi$ as $\phi=\eta\circ\pi^{-1}$ where $\eta\colon Z\to S$ and $\pi=\pi_1\circ\ldots\circ\pi_m\colon Z\to S$ are two sequences of point blowups. Denote by $F_j\subset Z$ the total transform of the indeterminacy point of $\pi_j^{-1}$ under the map $\pi_j\circ\ldots\circ\pi_m$. For $1\leq j\leq m$ let $E_j$ be the direct image of $F_j$ by $\eta$. Each $E_j$, if not zero, is an effective divisor. According to [@DillerFavre] we get the following formula called push-pull formula $$\label{eq:square} \phi_\phi^C=C+\displaystyle\sum_{j=1}^m(C\cdot E_j)E_j$$ for all curves (resp. divisor classes) $C$ in $S$. Since $\phi_$ and $\phi^$ are adjoint endomorphisms of $\mathrm{NS}(\mathbb{R},S)$ for the intersection form we get $$\label{eq:losange} \phi^C\cdot\phi^C=C\cdot C\displaystyle\sum_{j=1}^m(E_j\cdot C)^2$$ This formula and the <span style="font-variant:small-caps;">Hodge</span> index theorem imply that $\lambda(\phi)$ is a <span style="font-variant:small-caps;">Pisot</span> number or <span style="font-variant:small-caps;">Salem</span> number. The endomorphisms $\phi^$ and $\phi_$ preserve both the pseudo effective and nef cones of $\mathrm{NS}(\mathbb{R},S)$. Suppose that $\lambda(\phi)>1$. According to the <span style="font-variant:small-caps;">Perron</span>-<span style="font-variant:small-caps;">Frobenius</span> theorem there exists an eigenvector $\Theta(\phi)$ for $\phi_$ in the nef cone of $\mathrm{NS}(S)$ such that $$\label{eq:doublelosange} \phi^\Theta(\phi)=\lambda(\phi)\Theta(\phi)$$ Note that furthermore this vector is unique up to scalar form ([@DillerFavre]). Both (\[eq:losange\]) and (\[eq:doublelosange\]) imply that $$(\lambda(\phi)^2-1)\Theta(\phi)\cdot\Theta(\phi)=\displaystyle\sum_{j=1}^m(E_j\cdot\Theta(\phi))^2.$$ As a result for all $E_j$ $$\Theta(\phi)\cdot\Theta(\phi)=0\qquad\Longleftrightarrow \qquad \Theta(\phi)\cdot E_j=0.$$ Assume now that $\Theta(\phi)\cdot\Theta(\phi)=0$; then $\Theta(\phi)\cdot E_j=0$ for all $E_j$. As the $E_j$’s are effective and $\Theta(\phi)$ is nef the $\mathbb{Q}$-vector subspace of $\mathrm{NS}(\mathbb{Q},S)$ generated by the irreducible components of the divisors $E_j$ is contained in $\Theta(\phi)^\perp$. On the orthogonal complement $\Theta(\phi)^\perp$ of the isotropic vector $\Theta(\phi)$ the intersection form is negative and its kernel is the line generated by $\Theta(\phi)$. Equation (\[eq:square\]) implies $$\phi_^k\Theta(\phi)=\frac{1}{\lambda(\phi)^k}\Theta(\phi).$$ But $\lambda(\phi)>1$ and $\phi_$ preserves the lattice $\mathrm{NS}(\mathbb{Z},S)$ so $\Theta(\phi)$ is irrational. Consequently the intersection form is negative definite on the $\mathbb{Q}$-vector space generated by all classes of irreducible components of the divisors $E_j$. According to the <span style="font-variant:small-caps;">Grauert</span>-<span style="font-variant:small-caps;">Mumford</span> contraction theorem ([@BHPV]) there exists a birational morphism $\eta\colon S\to Y$ that contracts simultaneously all these components. Set $\varphi=\eta\circ\phi\circ\eta^{-1}$. As $\Theta(\phi)$ does not intersect the curves contracted by $\eta$ the class $\eta_\Theta(\phi)\in\mathrm{NS}(\mathbb{R},Y)$ is - isotropic, and - an eigenvector for $\varphi_$ with eigenvalue $\lambda(\phi)$. Let us iterate this process until $\varphi^{-1}$ does not contract any curve, that is $\varphi\in\mathrm{Aut}(Y)$. If $Y$ is singular, then consider the minimal desingularization $\widetilde{Y}$ of $Y$; the automorphism $\varphi$ lifts to an automorphism $\widetilde{\varphi}$ of $\widetilde{Y}$. As a result one can state \[thm:diamond\] Let $S$ be a complex projective surface. Let $\phi$ be a loxodromic birational self map of $S$. Then - all divisors $E_j$ are orthogonal to $\Theta(\phi)$ if and only if $\Theta(\phi)$ is an isotropic vector; - if $\Theta(\phi)$ is an isotropic vector, then there exists a birational morphism $\eta\colon S\to Y$ such that $\eta\circ\phi\circ\eta^{-1}$ is an automorphism of $Y$. Then <span style="font-variant:small-caps;">Diller</span> and <span style="font-variant:small-caps;">Favre</span> prove the following statement: \[thm:blablabla\] Let $\phi\in\mathrm{Bir}(S)$ $($resp. $\psi\in\mathrm{Bir}(\widetilde{S}))$ be an algebraically stable map of a complex projective surface $S$ $($resp. $\widetilde{S})$. Assume that $\phi$ and $\psi$ are conjugate via a proper modification. Suppose that $\lambda(\phi)>1$ $($or equivalently that $\lambda(\psi)>1)$. Then $\Theta(\phi)\cdot\Theta(\phi)=0$ if and only if $\Theta(\psi)\cdot\Theta(\psi)=0$. Theorem \[thm:blabla\] follows from Theorems \[thm:diamond\] and \[thm:blablabla\]. #### A second answer The following statement gives another characterization of birational maps conjugate to an automorphism of a smooth projective rational surface: \[thm:fleur\] Let $\phi$ be a birational map of a complex projective surface $S$. - If $\lambda(\phi)$ is a <span style="font-variant:small-caps;">Salem</span> number, then there exists a birational map $\psi\colon\widetilde{S}\dashrightarrow S$ that conjugates $\phi$ to an automorphism of $\widetilde{S}$; - if $\phi$ is conjugate to an automorphism, then $\lambda(\phi)$ is a quadratic integer or a <span style="font-variant:small-caps;">Salem</span> number. Assume that $\lambda(\phi)$ is a <span style="font-variant:small-caps;">Salem</span> number. Denote by $P(t)\in\mathbb{Z}[t]$ the minimal polynomial of $\lambda(\phi)$. But $\lambda(\phi)$ is a <span style="font-variant:small-caps;">Salem</span> number so there exists a root of $P$ with modulus $1$, denote it $\alpha$. Hence fix an automorphism $\kappa$ of the field $\mathbb{C}$ such that $\kappa(\lambda(\phi))=\alpha$. According to Proposition \[pro:DillerFavre\] we can suppose that $\phi$ is algebraically stable up to birational conjugacy. The eigenvector $\Theta(\phi)$ thus corresponds to the eigenvalue $\lambda(\phi)$ and so may be taken in $\mathrm{NS}(L,S)$ where $L$ is the splitting field of $P$. The automorphism $\kappa$ acts on $\mathrm{NS}(\mathbb{C},S)$ preserving $\mathrm{NS}(S)$ pointwise. Since $\phi^$ is defined over $\mathbb{Z}$ and $\phi^\Theta(\phi)=\lambda(\phi)\Theta(\phi)$ one obtains $$\phi^(\kappa(\Theta(\phi))=\kappa(\lambda(\phi))\kappa(\Theta(\phi))=\alpha\kappa(\Theta(\phi))$$ that is $\phi^\widetilde{\Theta}=\alpha\widetilde{\Theta}$ where $\widetilde{\Theta}=\kappa(\Theta(\phi))$. The divisor classes of the $E_j$’s belong to $\mathrm{NS}(S)$ so they are $\kappa$-invariant. As a consequence (\[eq:square\]) implies $$\label{eq:doublesquare} \phi_\phi^\widetilde{\Theta}=\widetilde{\Theta}+\displaystyle\sum_{j=1}^m(\widetilde{\Theta}\cdot E_j)E_j$$ Denote by $\overline{\widetilde{\Theta}}$ the conjugate of $\widetilde{\Theta}$ and by $\overline{\alpha}$ the conjugate of $\alpha$; from (\[eq:doublesquare\]) one gets $$(\alpha\overline{\alpha})\widetilde{\theta}\cdot\overline{\widetilde{\theta}}=\phi^\widetilde{\Theta}\cdot\phi^\overline{\widetilde{\Theta}}$$ As $\vert\alpha\vert=\alpha\overline{\alpha}=1$ one gets that $E_j\cdot\widetilde{\Theta}=0$ for any $1\leq j\leq m$ and $E_j\cdot\Theta(\phi)=0$ for any $1\leq j\leq m$. Theorem \[thm:fleur\] follows from Theorem \[thm:diamond\]. Theorem \[thm:fleur\] does not extend to quadratic integers (see* [@BlancCantat]). #### A third answer As we have seen in §\[sec:firstdef\] if $S$ is a projective smooth surface, then every $\phi\in\mathrm{Bir}(S)$ admits a minimal resolution, [i.e.]{} there exist $\pi_1\colon Z\to S$, $\pi_2\colon Z\to S$ two sequences of blow ups such that - no $(-1)$-curves of $Z$ is contracted by both $\pi_1$ and $\pi_2$; - $\phi=\pi_2\circ\pi_1^{-1}$. Denote by $\mathfrak{b}(\phi)$ the number of base points of $\phi$; note that $\mathfrak{b}(\phi)$ is equal to the difference of the ranks of $\mathrm{Pic}(Z)$ and $\mathrm{Pic}(S)$ and thus equal to $\mathfrak{b}(\phi^{-1})$. Let us introduce the dynamical number of the base-points of $\phi$. Since $\mathfrak{b}(\phi\circ\psi)\leq\mathfrak{b}(\phi)+\mathfrak{b}(\psi)$ for any $\phi$, $\psi$ in $\mathrm{Bir}(S)$, $\mu(\phi)$ is a non-negative real number. As $\mathfrak{b}(\phi)=\mathfrak{b}(\phi^{-1})$ one gets $\mu(\phi^k)=\vert k\mu(\phi)\vert$ for any $k\in\mathbb{Z}$. Furthermore if $\psi\colon S\dashrightarrow Z$ is a birational map between smooth projective surfaces and if $\phi\in\mathrm{Bir}(S)$, then for all $n\in\mathbb{Z}$ $$-2\mathfrak{b}(\psi)+\mathfrak{b}(\phi^n)\leq\mathfrak{b}(\psi\circ\varphi^n\circ\psi^{-1})\leq 2\mathfrak{b}(\psi)+\mathfrak{b}(\phi^n);$$ hence $\mu(\phi)=\mu(\psi\circ\phi\circ\psi^{-1})$. One can thus state the following result: \[lem:losange\] The dynamical number of base-points is an invariant of conjugation. In particular if $\phi$ is conjugate to an automorphism of a smooth projective surface, then $\mu(\phi)=0$. A base-point $p$ of $\phi$ is a persistent base-point if there exists an integer $N$ such that for any $k\geq N$ $$\left\{ \begin{array}{ll} p\in\mathrm{Base}(\phi^k)\\ p\not\in\mathrm{Base}(\phi^{-k}) \end{array} \right.$$ Let $p$ be a point of $S$ or a point infinitely near $S$ such that $p\not\in\mathrm{Base}(\phi)$. Consider a minimal resolution of $\phi$ $$\xymatrix{ & Z\ar[rd]^{\pi_1}\ar[ld]_{\pi_2} & \\ S\ar@{-->}[rr]_\phi & & S }$$ Because $p$ is not a base-point of $\phi$ it corresponds via $\pi_1$ to a point of $Z$ or infinitely near; using $\pi_2$ we view this point on $S$ again maybe infinitely near and denote it $\phi^\bullet(p)$. For instance if $S=\mathbb{P}^2_\mathbb{C}$, $p=(1:0:0)$ and $\phi$ is the birational self map of $\mathbb{P}^2_\mathbb{C}$ given by $$(z_0:z_1:z_2)\dashrightarrow(z_1z_2+z_0^2:z_0z_2:z_2^2)$$ the point $\phi^\bullet(p)$ is not equal to $p=\phi(p)$ but is infinitely near to it. Note that if $\phi$, $\psi$ are two birational self maps of $S$ and $p$ is a point of $S$ such that $p\not\in\mathrm{Base}(\phi)$, $\phi(p)\not\in\mathrm{Base}(\psi)$, then $(\psi\circ\phi)^\bullet(p)=\psi^\bullet(\phi^\bullet(p))$. One can put an equivalence relation on the set of points of $S$ or infinitely near $S$: the point $p$ is equivalent to the point $q$ if there exists an integer $k$ such that $(\phi^k)^\bullet(p)=q$; in particular $p\not\in\mathrm{Base}(\phi^k)$ and $q\not\in\mathrm{Base}(\phi^{-k})$. Note that the equivalence class is the generalization of set of orbits for birational maps. A base-point is periodic if - either $(\phi^k)^\bullet(q)=q$ for some $k\geq 0$, - or $q\in\mathrm{Base}(\phi^k)$ for any $k\in\mathbb{Z}\smallsetminus\{0\}$ (in particular $(\phi^k)^\bullet(p)$ is never defined for $k\not=0$). Let $\mathcal{P}$ be the set of periodic base-points of $\phi$. Denote by $\widehat{\mathcal{P}}$ the finite set of points equivalent to a point of $\mathcal{P}$. Both $\mathfrak{b}(\phi)$ and $\mathfrak{b}(\phi^{-1})$ are finite so there exists $n\in\mathbb{N}$ such that for any $p\in\mathrm{Base}(\phi)$ non periodic and for any $j$, $\ell\geq N$ $$\left\{ \begin{array}{ll} p\in\mathrm{Base}(\phi^j)\quad\Longleftrightarrow \quad p\in\mathrm{Base}(\phi^\ell)\\ p\in\mathrm{Base}(\phi^{-j})\quad\Longleftrightarrow \quad p\in\mathrm{Base}(\phi^{-\ell}) \end{array} \right.$$ Let us decompose $\mathrm{Base}(\phi)$ into five disjoint sets: $$\begin{aligned} & & \mathcal{B}_{++}=\big\{p\,\vert\,p\not\in\mathcal{P},\,p\in\mathrm{Base}(\phi^j),\,p\in\mathrm{Base}(\phi^{-j})\quad\forall\,j\geq N\big\}\\ & & \mathcal{B}_{+-}=\big\{p\,\vert\,p\not\in\mathcal{P},\,p\in\mathrm{Base}(\phi^j),\,p\not\in\mathrm{Base}(\phi^{-j})\quad\forall\,j\geq N\big\}\\ & & \mathcal{B}_{-+}=\big\{p\,\vert\,p\not\in\mathcal{P},\,p\not\in\mathrm{Base}(\phi^j),\,p\in\mathrm{Base}(\phi^{-j})\quad\forall\,j\geq N\big\}\\ & & \mathcal{B}_{--}=\big\{p\,\vert\,p\not\in\mathcal{P},\,p\not\in\mathrm{Base}(\phi^j),\,p\not\in\mathrm{Base}(\phi^{-j})\quad\forall\,j\geq N\big\}\end{aligned}$$ and $\mathcal{P}$. Note that: - $\mathcal{B}_{+-}$ is the set of persistent base-points of $\phi$; - $\mathcal{B}_{-+}$ is the set of persistent base-points of $\phi^{-1}$; - two equivalent base-points of $\phi$ belong to the same subsets of $\mathrm{Base}(\phi)$. Take $k\geq 2N$ an integer. Let us compute $\mathfrak{b}(\phi^k)$. Any base-point of $\phi^k$ is equivalent to a base-point of $\phi$. Let us thus consider a base-point $p$ of $\phi$ and determine the number $m_{p,k}$ of base-points of $\phi^k$ which are equivalent to $p$. 1. If $p$ belongs to $\mathcal{P}$, then the number of points equivalent to $p$ is less than $\#\mathcal{P}$ and $m_{p,k}\leq\#\mathcal{P}$. 2. If $p$ does not belong to $\mathcal{P}$, then any point equivalent to $p$ is equal to $(\phi^i)^\bullet(p)$ for some $i$; furthermore these points all are distinct. Hence $m_{p,k}=\# I_{p,k}$ where $$I_{p,k}=\big\{i\in\mathbb{Z}\,\vert\,p\not\in\mathrm{Base}(\phi^i),\,p\in\mathrm{Base}(\phi^{i+k})\big\}.$$ 1. Suppose that $p$ belongs to $\mathcal{B}_{++}$. Since $p$ does not belong to $\mathrm{Base}(\phi^i)$, the following inequalities hold: $-N<i<N$ and so $m_{p,k}<2N$. 2. If $p$ belongs to $\mathcal{B}_{--}$, then $p$ belongs to $\mathrm{Base}(\phi^{i+k})$ hence $-N<i+k<N$ and $m_{p,k}<2N$. 3. Assume that $p$ belongs to $\mathcal{B}_{-+}$. As $p\not\in\mathrm{Base}(\phi^i)$ (resp. $p\in\mathrm{Base}(\phi^{i+k})$), one has $-N<i$ (resp. $i+k\leq N$). These two conditions imply $-N<i\leq N-k$. But $k>2N$ thus $m_{p,k}=0$. 4. Finally consider a point $p$ in $\mathcal{B}_{+-}$. The fact that $p\not\in\mathrm{Base}(\phi^i)$ (resp. $p\in\mathrm{Base}(\phi^{i+k})$) yields $i<N$ (resp. $-N<i+k$). As a result $-N-k<i<N$ and $m_{p,k}\leq 2N+k$. Conversely if $i\leq -N$ and $i+k\geq N$, then $p\not\in\mathrm{Base}(\phi^i)$ and $p\in\mathrm{Base}(\phi^{i+k})$, [i.e.]{} $i\in I_{p,k}$. As a consequence $m_{p,k}\geq \#[N-k,-N]=k-2N+1$. Finally $$-2N\leq m_{p,k}-k\leq 2N.$$ Consequently there exist two constants $\alpha$, $\beta$ (independent on $k$) such that for all $k\geq 2N$ $$\nu k+\alpha\leq \mathfrak{b}(\phi^k)\leq\nu k+\beta$$ where $\nu$ is the number of equivalence classes of persistent base-points of $\phi$ (recall that $\mathcal{B}_{+-}$ is the set of persistent base-points of $\phi$). But $\mu(\phi)=\displaystyle\lim_{k\to +\infty}\frac{\mathfrak{b}(\phi^k)}{k}$ so $\mu(\phi)=\nu$. One can thus state: Let $S$ be a smooth projective surface. Let $\phi$ be a birational self map of $S$. Then $\mu(\phi)$ coincides with the number of equivalence classes of persistent base-points of $\phi$. In particular $\mu(\phi)$ is an integer. The following statement gives another characterization of birational maps which are conjugate to an automorphism of a projective surface; contrary to the two previous one it works for all maps of $\mathrm{Bir}(S)$. \[thm:doublelosange\] Let $\phi$ be a birational self map of a smooth projective surface. Then $\phi$ is conjugate to an automorphism of a smooth projective surface if and only if $\mu(\phi)=0$. This characterization was implicitely used in [@BedfordKim1; @BedfordKim2; @BedfordKim3; @DesertiGrivaux]. Let us give an example of [@DesertiGrivaux]. Consider the birational self map of $\mathbb{P}^2_\mathbb{C}$ given by $$\psi\colon (z_0:z_1:z_2)\dashrightarrow(z_0z_2^2+z_1^3:z_1z_2^2:z_2^3);$$ it has five base-points: $p=(1:0:0)$ and four points infinitely near. Denote by $\widehat{P_1}$ the collection of these points. Similarly $\psi^{-1}$ has five base-points: $(1:0:0)$ and four points infinitely near; let $\widehat{P_2}$ be the collection of these points. Consider the automorphism $A$ given by $$A\colon(z_0:z_1:z_2)\mapsto\big(\alpha z_0+2(1-\alpha)z_1+(2+\alpha-\alpha^2)z_2:-z_0+(\alpha+1)z_2:z_0-2z_1+(1-\alpha)z_2\big)$$ with $\alpha\in\mathbb{C}\smallsetminus\{0,\,1\}$. Then - $\widehat{P_1}$, $A(\widehat{P_2})$, and $(A\circ\psi\circ A)(\widehat{P_2})$ have distinct supports; - $\widehat{P_1}=(A\circ\psi)^2\circ A(\widehat{P_2})$. As a result the base-points of $\phi=A\circ\psi$ are non-persistent so $\phi$ is conjugate to an automorphism of a rational surface; this rational surface is $\mathbb{P}^2_\mathbb{C}$ blown up in $\widehat{P_1}$, $A(\widehat{P_2})$, and $(A\circ\psi\circ A)(\widehat{P_2})$. Furthermore $\lambda(A\circ\psi)>1$. Lemma \[lem:losange\] shows that if $\phi$ is conjugate to an automorphism of a smooth projective surface, then $\mu(\phi)=0$. Let us prove the converse. Assume thus that $\mu(\phi)=0$. One can suppose that by blowing-up points $\phi$ is algebraically stable (Proposition \[pro:DillerFavre\]). Therefore $\phi$ has no periodic base point and $\mathcal{B}_{++}=\emptyset$. Furthermore $\mu(\phi)=0$ corresponds to $\mathcal{B}_{+-}=\mathcal{B}_{-+}=\emptyset$. All base-points thus belong to $\mathcal{B}_{--}$. Assume that $\phi$ is not an automorphism of $S$. Let $\tau\colon Z\to S$ be the blow-up of the base-points of $\phi$. The morphism $\chi=\phi\circ\tau\colon Z\to S$ is the blow-up of the base-points of $\phi^{-1}$. Consider a $(-1)$-curve $E\subset Z$ contracted by $\chi$. The image $\chi(E)$ of $E$ is a proper point of $S$ that belongs to $\mathrm{Base}(\phi^{-1})$. Since $\phi$ is algebraically stable, then for all $k\geq 0$ $$\chi(E)\not\in\mathrm{Base}(\phi^k).$$ As a result $\phi^k\circ\chi\colon Z\dashrightarrow S$ is well-defined at any point of $E$. The curve $C=\tau(E)$ is thus an irreducible curve of $S$ contracted by $\phi^{k+1}$; any base-point of $\phi^{k+1}$ that belongs to $C$ as proper of infinitely near point is also a base-point of $\phi$. This finite set of points is contained in $\mathcal{B}_{--}$; so there is $n>0$ such that no base-point of $\phi^n$ belongs to $C$. Since $C$ is blown down by $\phi^n$, $C$ is a $(-1)$-curve of $S$. Contracting $C$ conjugates $\phi$ to an algebraically stable birational map whose all base-points are in $\mathcal{B}_{--}$. The rank of the <span style="font-variant:small-caps;">Picard</span> group of this new surface is strictly less than the rank of $\mathrm{Pic}(S)$. Consequently if we repeat this process, it has to stop. In other words $\phi$ is conjugate to an automorphism of a smooth projective surface. Index {#index .unnumbered} ===== [^1]: The author was partially supported by the ANR grant Fatou ANR-17-CE40- 0002-01 and the ANR grant Foliage ANR-16-CE40-0008-01. [^2]: The terminology corresponds to the initials of E. <span style="font-variant:small-caps;">Cartan</span>, A. <span style="font-variant:small-caps;">Alexandrov</span> and V. <span style="font-variant:small-caps;">Toponogov</span>. [^3]: Let $(X,d)$ be a metric space. Given a ray $\gamma$ the <span style="font-variant:small-caps;">Busemann</span> function $B_\gamma\colon X\to\mathbb{R}$ is defined by $B_\gamma(x)=\displaystyle\lim_{t\to +\infty}\big(d(\gamma(t),x)-t\big)$. [^4]: Recall that an exceptional curve $E$ on a surface $S$ is a smooth rational curve with $E^2=-1$. [^5]: except monomial maps $z\mapsto z^k$, <span style="font-variant:small-caps;">Tchebychev</span> polynomials, <span style="font-variant:small-caps;">Lattès</span> examples ... [^6]: “so named in honor of <span style="font-variant:small-caps;">Kummer</span>, <span style="font-variant:small-caps;">Kähler</span>, <span style="font-variant:small-caps;">Kodaira</span> and of the beautiful mountain K$2$ in Kashmir” ([@Weil:collected]).	{ "pile_set_name": "ArXiv" }
--- abstract: 'In joint work with J. Rasmussen, we gave an interpretation of Heegaard Floer homology for manifolds with torus boundary in terms of immersed curves in a punctured torus [@HRW]. In particular, knot Floer homology is captured by this invariant; see [@HRW-companion]. Appealing to earlier work of the authors on bordered Floer homology [@HW], we give a formula for the behaviour of these immersed curves under cabling.' address: - 'Department of Mathematics, Princeton University.[E-mail address:]{} ' - 'Department of Mathematics, University of British Columbia.[E-mail address:]{} ' author: - Jonathan Hanselman - Liam Watson bibliography: - 'references/bibliography.bib' title: Cabling in terms of immersed curves --- Immersed curves and the merge operation {#sec:curves} ======================================= The proof of Theorem \[thm:main-cable\] {#sec:proof} =======================================	{ "pile_set_name": "ArXiv" }
--- abstract: 'A simulation study of negative thermal expansion in Y$_2$W$_3$O$_{12}$ was carried out using calculations of phonon dispersion curves through the application of density functional perturbation theory. The mode eigenvectors were mapped onto flexibility models and results compared with calculations of the mode Grüneisen parameters. It was found that many lower-frequency phonons contribute to negative thermal expansion in Y$_2$W$_3$O$_{12}$, all of which can be described in terms of rotations of effectively rigid WO$_4$ tetrahedra and Y–O rods. The results are strikingly different from previous phonon studies of higher-symmetry materials that show negative thermal expansion.' address: - '$^1$ School of Physics and Astronomy and Materials Research Institute, Queen Mary University of London, Mile End Road, London, E1 4NS, United Kingdom.' - '$^2$ CrystalMaker Software Ltd, Centre for Innovation & Enterprise, Oxford University Begbroke Science Park, Woodstock Road, Begbroke, Oxfordshire, OX5 1PF, United Kingdom.' - '$^3$ Department of Earth Sciences, University of Cambridge, Downing Street, Cambridge, CB2 3EQ, United Kingdom.' author: - 'Leila H N Rimmer$^{1,2,3}$ and Martin T Dove$^{1,3}$' title: 'Simulation study of negative thermal expansion in yttrium tungstate Y$_2$W$_3$O$_{12}$' --- Introduction ============ Atomic-scale origin of negative thermal expansion ------------------------------------------------- As has been highlighted in several recent reviews [@Barrera_2005_review; @Romao_2013_NTEReview; @Lind_2012_Mater_NTEReview; @Takenaka_2012_NTEReview; @Miller_2009_NTEReview], a growing number of materials have been identified as exhibiting negative thermal expansion (NTE): the phenomenon whereby a material shrinks, rather than expands, on heating. NTE is found in several framework material families, such as those represented by SiO$_2$ [@NTE_quartz_data; @NTE_quartz_theory], Cu$_2$O [@Cu2O_Tiano; @Cu2O_Sanson; @Cu2O_Gupta; @Leila_Cu2O], ScF$_3$ [@ScF3_discovery; @ScF3_phonons], ZrW$_2$O$_8$ [@ZrW2O8_Science_1996; @Pryde_1996; @ZrW2O8_structure_1999; @Cao_etal_2002_PRL_ZrW2O8NTEFrustratedModes; @Cao_etal_2003_PRB_ZrW2O8NTEXAFS], Sr$_2$W$_3$O$_{12}$ [@Sc2W3O12_Evans], Zn(CN)$_2$ [@Goodwin_Zn(CN)2; @Hong_Zn(CN)2] as well as in a number of hybrid metal-organic framework materials [@MOFs_Han_2007; @MOFs_Dubbeldam_2007; @MOFs_Wu_2008]. Our understanding of positive thermal expansion is based on the nature of interatomic bonding. The potential energy between a pair of bonded atoms is anharmonic and asymmetric about the position of the energy minimum, with the slope of the potential energy function being shallower in the direction of greater interatomic separation. As a result, when the material is heated and the energy of the atoms increases, the average separation of each bonded pair also increases. This implies an overall expansion of the bulk material on heating; that is positive thermal expansion, making negative thermal expansion a counter-intuitive and unexpected phenomenon. Our qualitative understanding of NTE in framework materials is based on the fact that, in NTE systems, the energy cost of stretching interatomic bonds is typically much higher than transverse vibrations of those same bonds. For example, we could label two cations M and M$^\prime$ that form a M–O–M$^\prime$ linkage via a bridging oxygen atom. If the M–O and M$^\prime$–O bonds are very stiff, transverse displacements of the O atom will pull the M and M$^\prime$ cations towards each other. Since the amplitude of a transverse displacement will increase on heating, this leads to negative thermal expansion. This mechanism is called the tension effect [@Barrera_2005_review]. The drive towards positive thermal expansion exists in all materials, including those with overall NTE. Therefore, for a material to show NTE, the amplitude of the oxygen displacement must be sufficiently large to outweigh those mechanisms that drive positive thermal expansion. Given that the amplitude of a phonon is inversely proportional to the square of its frequency, an NTE-driving phonon must also have low frequency to achieve this condition. All of the NTE-exhibiting material families mentioned above share one significant feature: their crystal structures can all be described in terms of an infinite three-dimensional network of corner-sharing structural polyhedra (such as SiO$_4$, WO$_4$, OCu$_4$ and ZnN$_4$ tetrahedra, and ScF$_6$, ZrO$_6$, and YO$_6$ octahedra). Deformation of coordination polyhedra is likely to require higher energies due to the electrostatic repulsion between oxygens at the polyhedral vertices or to the high sensitivity of d-orbital bonding around a metal coordination centre with respect to bonding geometry. Therefore, any tension effect that avoids or minimises deformation of those polyhedra is more likely to be acting at a low enough frequency to drive macroscopic NTE. The conditions outlined above describe a Rigid Unit Mode (RUM) [@Pryde_1996; @Giddy_RUMs; @Hammonds_RUMs; @Dove_RUMs]: a low frequency phonon with rotations and translations of effectively rigid coordination polyhedra. The RUM model can be used to explain why structures with stiff coordination polyhedra have tension effect vibrations with sufficiently large amplitude to drive NTE [@NTE_theory_Welche]. In cases such as ScF$_3$ [@ScF3_phonons], Zn(CN)$_2$ [@Hong_Zn(CN)2] and Cu$_2$O [@Leila_Cu2O], lattice dynamics calculations have shown that the phonons responsible for NTE are RUMs involving motion of ScF$_6$ octahedra, Zn(C/N)$_4$ tetrahedra and CuO$_2$ rods respectively moving as rigid units. The case of ZrW$_2$O$_8$ is one that we will consider in a separate publication but, as a prelude to this, we consider here the case of Y$_2$W$_3$O$_{12}$—a material which, intriguingly, cannot support RUMs [@Hammonds_octahedra] but does exhibit NTE [@Forster_Sleight_1999_IntJInorgMat_Y2W3O12NTEExpt; @Sumithra_2005]. ![The Y$_2$W$_3$O$_{12}$ structure [@Forster_Sleight_1999_IntJInorgMat_Y2W3O12NTEExpt]. (a) Shows the structure in ball-and-stick format: green atoms are Y, grey atoms are W and red atoms are O. (b) Shows the structure in terms of its corner-sharing YO$_6$ and WO$_4$ coordination polyhedra: green octahedra are YO$_6$ units, grey tetrahedra are WO$_4$ units. The unit cell is shown as the dashed black line.[]{data-label="Y2W3O12picture"}](figure_1.pdf){width="100.00000%"} Y$_2$W$_3$O$_{12}$: an NTE material with no RUMs ------------------------------------------------ Y$_2$W$_3$O$_{12}$ [@Forster_Sleight_1999_IntJInorgMat_Y2W3O12NTEExpt] is a member of the A$_2$M$_3$O$_{12}$ family of NTE materials, where A is a trivalent cation in octahedral coordination, and M is either W or Mo with tetrahedral coordination [@Evans_etal_1997_JSolStatChem_NTEinA2M3O12Family; @Mary_1999]. A$_2$M$_3$O$_{12}$ materials have an orthorhombic structure—Y$_2$W$_3$O$_{12}$ is shown in —and thus, whilst they can (and in the case of this family, do) have volumetric NTE, their expansion is anisotropic. The A$_2$M$_3$O$_{12}$ family is of interest because of its customisability: the A site can withstand a wide range of substitutions, from ions as small as Al$^{3+}$ to ions as large as Ho$^{3+}$ [@Lind_2012_Mater_NTEReview], providing the opportunity to tune thermal expansion behaviour. In general, the smaller the A cation, the smaller the NTE; in extremis Al$_3$W$_2$O$_{12}$ actually exhibits weak PTE [@Evans_etal_1997_JSolStatChem_NTEinA2M3O12Family]. Unlike most A$_2$M$_3$O$_{12}$ compounds, Y$_2$W$_3$O$_{12}$ shows NTE along all three axes [@Forster_Sleight_1999_IntJInorgMat_Y2W3O12NTEExpt]. This is unusual for a non-cubic NTE material as they typically undergo NTE along one or two axes coupled with PTE along the remaining axis or axes [@Sleight_1995_Endeavour_NTE]. Furthermore, Y$_2$W$_3$O$_{12}$ has some of the largest coefficients of NTE of the entire A$_2$M$_3$O$_{12}$ family; its average linear expansion coefficient has been measured as $-7.0$ MK$^{-1}$ [@Forster_Sleight_1999_IntJInorgMat_Y2W3O12NTEExpt], which equates to a volumetric coefficient of thermal expansion of $\alpha_{\mathrm{V}}\approx-21$ MK$^{-1}$. Y$_2$W$_3$O$_{12}$ is an NTE material of particular interest to us because, if thought of as a series of rigid YO$_6$ and WO$_4$ coordination polyhedra, then the crystal structure would be too over-constrained to support RUMs (this is discussed further in \[maxwell\_counting\]). The existence of NTE in this material therefore suggests a degree of flexibility of either or both of the YO$_6$ octahedra and WO$_4$ tetrahedra, and it is important to understand how this flexibility generates a tension effect of sufficient strength to generate NTE. Analysis of neutron diffraction data [@Forster_Sleight_1999_IntJInorgMat_Y2W3O12NTEExpt] suggests that NTE in Y$_2$W$_3$O$_{12}$ may be driven by transverse vibrations of oxygens within the Y–O–W linkages, as per the tension effect. However, this prediction has yet to be confirmed through analysis of the vibrational spectrum. No experimental study of the phonon spectrum of Y$_2$W$_3$O$_{12}$ has yet been published but parametrised interatomic potential calculations [@Sumithra_etal_2007_PRB_Y2W3O12NTEPhononsMechanism] have suggested possible NTE phonons between 1 and 2.5 THz that may correspond to the tension effect. However, this model was only fitted to the structure (sensitive to the first-order differentials of the interatomic potential) and not to any property that will give information about the important second-order differentials (essential for vibrational frequencies). Moreover, the model includes interactions that depend on the Y–O–W bonds but nothing dependent on the more useful O–Y–O or O–W–O bonds. In this paper we investigate exactly where in the Y$_2$W$_3$O$_{12}$ phonon spectrum the structure flexibility manifests itself and, since the RUM model is not applicable, how NTE phonons exist in enough of the spectrum (and at low enough frequencies) so as to give rise to an overall negative coefficient of thermal expansion. To do this, we use an approach similar to that used in our studies of Zn(CN)$_2$ [@Hong_Zn(CN)2], Cu$_2$O [@Leila_Cu2O] and MOF-5 [@Leila_MOF5], that is we perform ab initio calculations of the material’s phonon spectrum that are then compared to models of structural flexibility. Lattice dynamics calculations have recently been reported for the related material Y$_2$Mo$_2$O$_{12}$ [@Wang_Y2Mo3O12; @Romao_Y2Mo3O12] which, like Y$_2$W$_3$O$_{12}$, also shows NTE along all three axes [@Marinkovic_Y2Mo3O12]. However, the approach taken in this paper (examining the full set of dispersion curves) was not followed in [@Wang_Y2Mo3O12; @Romao_Y2Mo3O12]. Methodology =========== Ab-initio calculations ------------------------ ------------------ -------- -------- -------- -------- -------- -------- Space group: Cell parameters: $x$ $y$ $z$ $x$ $y$ $z$ Y 0.4697 0.3833 0.2496 0.4669 0.3813 0.2493 W 0.2500 0.0000 0.4731 0.2500 0.0000 0.4739 W 0.1112 0.3574 0.3884 0.1167 0.3562 0.3928 O 0.0839 0.1396 0.0615 0.0905 0.1395 0.0675 O 0.1402 0.0644 0.3719 0.1343 0.0614 0.3723 O 0.0217 0.2629 0.3175 0.0177 0.2657 0.3211 O 0.3387 0.4209 0.0749 0.3332 0.4155 0.0767 O 0.0583 0.4648 0.3179 0.0732 0.4690 0.3226 O 0.2812 0.3435 0.3563 0.2868 0.3340 0.3613 W–O Y–O W–Y ------------------ -------- -------- -------- -------- -------- -------- : Structural data for the Y$_2$W$_3$O$_{12}$ equilibrium cell, as optimised at 0 GPa via DFT in CASTEP. Experimental data obtained for Y$_2$W$_3$O$_{12}$ from neutron diffraction at 0 GPa and 15 K [@Forster_Sleight_1999_IntJInorgMat_Y2W3O12NTEExpt] are given alongside for comparison. The ab-initio results are very close to the experimentally-derived values. \[Y2W3O12EquilibriumCell\] Density Functional Theory (DFT) calculations were carried out using the CASTEP code [@Clark_etal_2005_ZKrist_CASTEP]. This uses a plane-wave basis set to describe the electronic wave functions together with the pseudopotential method to remove the need to handle core atomic electrons explicitly. We used the GGA-PBE functional [@Perdew_etal_1996_PRL_PBEFunctional; @Perdew_etal_1997_PRL_PBEFunctionalErrata] and CASTEP’s internal on-the-fly-generated pseudopotentials for Y, W and O, regenerated so as to be norm-conserving. The material was explicitly defined as an insulator for all calculations. Forces, stresses and dielectric constants were converged to within 0.005 eVÅ$^{-1}$, 0.01 GPa and 0.00025 respectively. This was achieved using a plane wave cutoff energy of 1200 eV and a $3\times 2\times 3$ Monkhorst-Pack [@Monkhorst_Pack_1976_PRB_MPGrids] grid of wave-vectors for the integration of electronic states across the Brillouin zone. Geometry optimisation at 0 GPa gave final structural parameters as detailed in . Agreement with experiment is reasonable although, as ever with the GGA model, the calculation slightly overestimates bond distances and lattice parameters. shows that the differences occur mostly in the range $\le1\%$ except for the $b$ lattice parameter (just under $2\%$ discrepancy). Phonon frequencies were calculated using Density Functional Perturbation Theory (DFPT) [@Baroni_DFPT; @Refson_etal_2006_PhysRevB_DFPTPaper] with a $3\times 2\times 3$ Monkhorst-Pack [@Monkhorst_Pack_1976_PRB_MPGrids] grid that was offset to place one of the wave-vectors at the origin of reciprocal space. Fourier interpolation was used to calculate phonons for wave vectors along high-symmetry directions [@Aroyo_etal_2011_BilbaoCrystServer1; @Aroyo_etal_2006_ZKrist_BilbaoCrystServer1; @Aroyo_etal_2006_ActaCrystA_BilbaoCrystServer2; @Tasci_etal_2012_Conference_BilbaoCrystServerKVEC; @BilbaoCrystServer] for the production of dispersion curves. In addition, phonons were calculated for 490 wave-vectors spaced randomly throughout the Brillouin zone for the production of the vibrational density of states. This level of sampling of wave vectors proved to be sufficient to achieve convergence of phonon properties. Lattice dynamics calculations and thermal expansion --------------------------------------------------- The role of individual phonons in thermal expansion is represented by the mode Grüneisen parameter $\gamma_{\mathbf{k},i}$ [@Barrera_2005_review], which quantifies the change in phonon frequency with volume through the relation $$\gamma_{\mathbf{k},i} = -\frac{\partial \ln \omega_{\mathbf{k},i}}{\partial \ln V}$$ where $V$ is the crystal volume and $\omega_{\mathbf{k},i}$ is the angular frequency of the phonon of branch $i$ and wave vector $\mathbf{k}$. Within the quasi-harmonic approximation (the approximation in which all temperature dependence of phonon frequencies occurs indirectly through changes in force constants arising from changes in volume) the coefficient of volumetric thermal expansion $\alpha_V$ is given as $$\alpha_V = \frac{\partial \ln V}{\partial T} = \frac{1}{BV} \sum_{\mathbf{k},i} c_{\mathbf{k},i} \gamma_{\mathbf{k},i}$$ where $$c_{\mathbf{k},i} = \hbar \omega_{\mathbf{k},i} \frac{\partial n(T,\omega_{\mathbf{k},i})}{\partial T}$$ $c_{\mathbf{k},i}$ is the contribution of the individual phonon mode to the overall heat capacity, $T$ is the temperature, $B$ is the bulk modulus, $\hbar$ is the reduced Planck’s constant, and $n(T,\omega)$ is the Bose–Einstein distribution for phonons. A negative value of $\gamma_{\mathbf{k},i}$ indicates that the phonon $i$ at wave vector $\mathbf{k}$ contributes to negative thermal expansion and, if a sufficient number of modes with a large enough negative value of $\gamma_{\mathbf{k},i}$ exist, the overall outcome for the system is a negative value of $\alpha_V$, that is, NTE. Application of lattice dynamics and Grüneisen theory ---------------------------------------------------- Since Y$_2$W$_3$O$_{12}$ has an orthorhombic lattice, changes in cell size ($\Delta V$) were achieved through variable-cell geometry optimisation at an applied hydrostatic pressure. Testing showed that an applied pressure of $-0.06$ GPa resulted in the desired cell volume change of approximately 0.1%. The structure was optimised at both this pressure and zero pressure, with phonon frequencies also calculated at both pressures for a set of random wave vectors and for wave vectors along the key symmetry directions in reciprocal space. In order to calculate the change in frequency for a particular mode with change in volume, $\Delta \omega_{\mathbf{k},i}$, we used an eigenvector matching algorithm to ensure that we calculated $\omega_{\mathbf{k},i}$ for equivalent modes (\[draw\]). From these results we generated the set of mode Grüneisen parameters numerically as $\gamma_{\mathbf{k},i} = (\Delta \omega_{\mathbf{k},i} / \omega_{\mathbf{k},i}) / (\Delta V/V)$. Flexibility models {#y2w3o12flexmodelsection} ------------------ Our approach was to identify the origin of negative $\gamma_{\mathbf{k},i}$ values in terms of the flexibility of the structure. We therefore constructed a number of flexibility models in which we enforced a high degree of rigidity for certain bonds or structural units, whilst maintaining force constants of zero value for other bond stretching or bending motions. This meant that any phonon calculated for a given flexibility model would have zero or near-zero frequency if that phonon did not violate the constraints of that model, and high frequency otherwise. The objective was to compare eigenvectors of phonons from the ab initio lattice dynamics calculations with eigenvectors of those zero or near-zero frequency modes calculated for the constructed flexibility models. This allowed us to pinpoint different types of atomic-scale motion in the highly complex ab initio dispersion curves (including those regions which drive NTE) with relative ease. The flexibility models considered are best described as combinations of separate components that represent different structural units. Hence, the YO$_6$ units were represented by one of 1. Rigid Y–O rods 2. Rigid YO$_6$ octahedra while the WO$_4$ units were represented by one of 1. Rigid W–O rods 2. Rigid WO$_4$ tetrahedra The construction and application of these models is discussed in \[flexibility\_models\]. It has been suggested that in ZrW$_2$O$_8$, a similar material to Y$_2$W$_3$O$_{12}$ (in that both are frameworks of corner-linked tetrahedral and polyhedral groups of atoms), the NTE phonons may exhibit strong correlations between the motions of adjacent metal sites, with the W–O–Zr bond angles remain undistorted [@Cao_etal_2002_PRL_ZrW2O8NTEFrustratedModes; @Cao_etal_2003_PRB_ZrW2O8NTEXAFS; @Figueiredo_Perottoni_2007_DFTZrW2O8BondStiffness]. In order to investigate the possibility of a similar correlation between Y and W atoms in Y$_2$W$_3$O$_{12}$, additional flexibility models were constructed such that Y…W correlations could be taken into account via one of 1. No Y…W bond 2. Rigid Y…W rod In total this yielded eight different Y$_2$W$_3$O$_{12}$ flexibility models. However, the two models containing both rigid WO$_4$ tetrahedra and rigid YO$_6$ octahedra are over-constrained (see \[maxwell\_counting\] and [@Hammonds_octahedra]) and are, therefore, not discussed here (although models were constructed and tested to confirm this was actually the case). The six remaining flexibility models were constructed as a set of simple force field models within the GULP lattice simulation program [@GULP_2003]. Simple bond stretching and bond angle bending forces were added to each model as discussed in \[flexibility\_models\], using the ambient pressure structure calculated by CASTEP at 0 GPa. In order to compare the eigenvectors of the ab initio simulation with those of the flexibility model, we define a dimensionless ‘match’ coefficient, $m_{\mathbf{k},i}$, as $$\label{mequation} m_{\mathbf{k},i} = \Omega^2 \sum_j\frac{\mathbf{e}_{\mathbf{k},i}^\mathrm{phonon}\cdot\mathbf{e}_{\mathbf{k},j}^\mathrm{model}}{\Omega^2+\tilde{\omega}^2_{\mathbf{k},j}} \label{eq:m}$$ where $\mathbf{e}_{\mathbf{k},i}^\mathrm{phonon}$ is the eigenvector of the ab initio phonon mode $i$ at wave vector $\mathbf{k}$, $\mathbf{e}_{\mathbf{k},j}^\mathrm{model}$ is the eigenvector of a mode $j$ in the flexibility model calculated at the same wave vector $\mathbf{k}$, and $\tilde{\omega}_{\mathbf{k},j}$ is the corresponding angular frequency in the flexibility model. $\Omega$ is a scale factor that avoids division-by-zero errors when $\tilde{\omega}_{\mathbf{k},j}=0$, giving instead a value of $m_{\mathbf{k},i} = \Omega^2/(\Omega^2+\tilde{\omega}^2_{\mathbf{k},j})=1$ when $\tilde{\omega}^2_{\mathbf{k},j}=0$ (we used a value $\Omega = 1$ THz). This sets the range of $m_{\mathbf{k},i}$ values from 0 to 1. A value of $m_{\mathbf{k},i}$ close to 1 occurs when the ab initio phonon $i$ is a close match to a flexibility model mode that leaves the defined ‘rigid’ units undistorted. A value of $m_{\mathbf{k},i}$ close to 0 arises when the ab initio phonon $i$ involves significant distortions of the rigid units described in the flexibility model. Therefore $m_{\mathbf{k},i}$ is a measure of the extent to which mode $i$ of wave vector $\mathbf{k}$ in the ab initio model matches any modes given in the flexibility model at the same wave vector. Our practical implementation of this mode matching is described in \[draw\] and \[flexibility\_models\]. Results ======= ![(a) Low energy dispersion curves and densities of states for Y$_2$W$_3$O$_{12}$. The notation of Bradley and Cracknell [@Bradley_Cracknell_1972_CrystalBZSymmetryBook] is used to denote special Brillouin zone points ($\Gamma \equiv [0,0,0]$, $\mathrm{X} \equiv [\frac{1}{2},0,0]$, $\mathrm{Y} \equiv [0,\frac{1}{2},0]$, $\mathrm{Z} \equiv [0,0,\frac{1}{2}]$). (b) Shows the same data shaded according to the value of $\gamma_{\mathbf{k},i}$ of each mode at each wave vector. The colour scale ranges from red ($\gamma_{\mathbf{k},i}\le-9$) to white ($\gamma_{\mathbf{k},i}=0$) through to blue ($\gamma_{\mathbf{k},i}\ge+9$). Bins that make up the density of states are shaded according to the average $\gamma_{\mathbf{k},i}$ for each bin using the same colour scale.[]{data-label="Y2W3O12Phonons1"}](figure_2.pdf){width="70.00000%"} Phonon dispersion and mode Grüneisen parameters in Y$_2$W$_3$O$_{12}$ {#ntephononsiny2w3o12section} --------------------------------------------------------------------- a shows the calculated phonon dispersion curves and density of states of Y$_2$W$_3$O$_{12}$ in the 0–12 THz frequency range. As can be seen, Y$_2$W$_3$O$_{12}$ dispersion curves are relatively complex due to many mode crossings and the existence of multiple phonons within a tight frequency range. Nevertheless, it can be seen that the density of states is a continuous function up to about 8 THz, with a gap before another band of phonons in the range 9–11.5 THz. Not shown in the diagram is a further tight band of higher-frequency vibrations around 24 THz and a final set of vibrations between 27–30 THz. There are 204 separate modes $i$ for any wave vector $\mathbf{k}$. The two highest-frequency bands (24 and 27–30 THz) consist of 48 individual modes, the band between 9–11.5 THz consists of 60 individual modes, and the broad band from 0–8 THz consists of the other 96 modes. We discuss eigenvectors associated with these bands in the next subsection. b shows the dispersion curves with colours used to represent the values of the associated mode Grüneisen parameters $\gamma_{\mathbf{k},i}$ (red for negative and blue for positive values respectively). It is interesting to note that the dispersion diagram is divided into large bands of negative or positive values of mode Grüneisen parameters. This feature stands in contrast to other recently studied systems such as Cu$_2$O [@Leila_Cu2O], ScF$_3$ [@ScF3_phonons], Zn(CN)$_2$ [@Hong_Zn(CN)2] and MOF-5 [@Leila_MOF5], where phonons with significantly different values of mode Grüneisen parameters can exist at similar frequencies due to the significant dependence of the NTE phonon eigenvectors on wave vector. The strongest contributions to NTE come from phonons around 0.7–1 THz ($\gamma_{\mathbf{k},i}\sim-9$ on average, with values of $\sim-15$ at wave vectors near the Brillouin zone points S and T). Acoustic modes also contribute to NTE, as is the case in some other systems recently studied [@Leila_Cu2O; @Hong_Zn(CN)2; @Leila_MOF5]. We note that recent ab initio calculations of the mode Grüneisen parameters at the $\Gamma$ point in Y$_2$Mo$_3$O$_{12}$ using the frozen-phonon method also show a large number of modes with negative values, but see smaller values except for the lowest-frequency mode [@Wang_Y2Mo3O12]. A second group of strongly-NTE phonons exists around 1.8 THz but, more generally, it can be seen that all the phonons up to around 5 THz have negative $\gamma_{\mathbf{k},i}$ (albeit approaching zero values as the frequency increases to 5 THz). There is also a higher frequency band of weak NTE modes around 9–10 THz $\gamma_{\mathbf{k},i}\sim0$. Weak, low frequency, PTE modes exist around 5–8 THz and for frequencies larger than 10 THz. Flexibility model mapping ------------------------- ![Flexibility analysis of Y$_2$W$_3$O$_{12}$ phonon dispersion curves and densities of states. All flexibility models shown here describe YO$_6$ units in terms of corner-sharing rigid YO$_6$ octahedra. The notation of Bradley and Cracknell denotes special Brillouin zone points [@Bradley_Cracknell_1972_CrystalBZSymmetryBook]. All data are shaded according to the value of $m_{\mathbf{k},i}$ at each mode for each wave vector. The shading ranges from white ($m_{\mathbf{k},i}=0$) through to black ($m_{\mathbf{k},i}=1$). Bins that make up the density of states are shaded according to the average $m_{\mathbf{k},i}$ for each bin using the same colour scale.[]{data-label="Y2W3O12Flexibility"}](figure_3.pdf){width="\textwidth"} shows six plots of the 156 Y$_2$W$_3$O$_{12}$ ab initio dispersion curves for frequencies between 0–12 THz (as in ) together with their corresponding densities of states. In each case the data are shaded in greyscale according to the degree to which one of the six flexibility models can reproduce the eigenvectors of each phonon in the ab initio calculations. Black shading indicates a perfect match between the ab inito phonons and the flexibility model in question; white shading indicates no relation between the two. This approach offers a convenient, visual sense of the atomic motions associated with different regions of the complex Y$_2$W$_3$O$_{12}$ phonon spectrum. Further details on how these ‘match’ values were calculated are given in \[flexibility\_models\]. In what follows we use the information presented in and to correlate the NTE phonons identified in with the vibrations permitted by the six flexibility models. ### Rigid W–O and Y–O rods with no rigid polyhedra. As discussed in \[maxwell\_counting\], a flexibility model of Y$_2$W$_3$O$_{12}$ consisting of rigid W–O and Y–O rods has 96 excess degrees of freedom. We would therefore expect many of the 156 lower-frequency Y$_2$W$_3$O$_{12}$ phonons to correspond to rigid W–O and Y–O rod motions. shows that this is largely the case. Much of the ab inito phonon spectrum in the 0–12 THz range is a strong match for rigid W–O/Y–O flexibility model modes. However, this motion is not confined to specific phonons in the spectrum but, instead, there is a high degree of eigenvector mixing and the flexibility of the model is spread over multiple phonons in the real material. This mixing follows from the relatively low symmetry and the complexity of the dispersion curves. Despite the high degree of eigenvector mixing present, it is still possible to observe general trends within the phonon spectrum. Specifically, phonons in the 0–5 THz (approximately 64 modes) and 9–10 THz (approximately 28 modes) ranges have the strongest match for rigid W–O and Y–O motions. These differences correspond well with the separation of PTE and NTE phonons in the Y$_2$W$_3$O$_{12}$ spectrum as seen in . NTE phonons have the strongest match with the flexibility model (corresponding to negligible W–O and Y–O bond stretching) while PTE phonons have the weakest match (corresponding to small, but non-trivial, stretching of W–O and Y–O bonds). This observation is a good quantification of the tension effect giving rise to NTE in Y$_2$W$_3$O$_{12}$. The addition of the constraint that keeps the Y…W distance fixed yields similar results. This flexibility model has 60 excess degrees of freedom (reduced from the original 96; see \[maxwell\_counting\]). shows that the same eigenvector mixing that was observed previously is still present, and that the reduced flexibility of this model manifests itself in a reduction in the match with the ab initio phonons (seen in the lighter greyscale shading in ) across the whole phonon spectrum. We can conclude that the eigenvectors of all phonons with frequencies in the range 0–12 THz in Y$_2$W$_3$O$_{12}$ involve some changes in Y…W distance. ### Rigid WO$_4$ tetrahedra and Y–O bonds. The analysis of \[maxwell\_counting\] shows that the flexibility model for Y$_2$W$_3$O$_{12}$ consisting of rigid WO$_4$ tetrahedra and Y–O rods has 48 excess degrees of freedom. This model is still highly flexible, albeit less flexible than the two rigid rod models discussed above. shows, once again, that a high degree of eigenvector mixing occurs for this system with each flexibility model phonon spread across multiple ab initio phonons. It can also be seen that the lower overall flexibility manifests itself in the lightening of the greyscale, indicating a weaker overall match with the ab inito phonons. This lightening is not uniform; match values are still relatively strong in the 0–5 THz region and approach a perfect match for the lowest frequency phonons. Above 5 THz the match between this model and the ab initio phonons decreases with increasing frequency. The largest reduction in the match with the flexibility model is in the 9–11.5 THz region. We can identify this region with the modes involving significant bending of the bonds within the WO$_4$ tetrahedra. The analysis of \[maxwell\_counting\] shows that the number of WO$_4$ bond-bending modes is 60, which is exactly the number of modes in the 9–11.5 THz region. The phonons in the 0–5 THz range also correspond to the strongest NTE-driving phonons. This, together with the above observations, suggests that these lowest frequency NTE modes correspond to motion of near-rigid WO$_4$ tetrahedra and rigid Y–O rods. The addition the Y…W distance constraint leads to a very weak match between the flexibility model phonons and those of the ab initio calculation. This can be seen in the considerable lightening of the greyscale shading for this model in . It is consistent with the analysis of \[maxwell\_counting\], which shows that adding the additional Y…W distance constraint greatly reduces the flexibility of this model from 48 to exactly zero excess degrees of freedom. In short, a model with rigid WO$_4$ tetrahedra is unable to support motions in which the Y…W distance is conserved. ### Rigid YO$_6$ octahedra and W–O bonds It is seen from \[maxwell\_counting\] that a flexibility model consisting of rigid YO$_6$ octahedra and W–O rods has 36 excess degrees of freedom, which is fewer than for the rigid WO$_4$ tetrahedron/Y–O rod model discussed above. Accordingly, shows that the overall match between this flexibility model and the ab initio dispersion curves is lower than for the rigid WO$_4$/Y–O model. Once again, a significant amount of eigenvector mixing is also apparent. The strength of the match between this model and the ab initio phonons is greatest for the lowest frequency NTE phonons in the 0–2 THz range. In general, the match is weaker than that found for the rigid WO$_4$ tetrahedra/Y–O rod model and, aside from the acoustic phonons at the $\Gamma$ point, a perfect match is never achieved. There is one exception to this; in the 9–10 THz range the match between this model and the ab initio phonons increases slightly and is slightly stronger than that for the rigid WO$_4$ tetrahedra/Y–O rod model, which is consistent with the identification of the frequency region above 9 THz corresponding to WO$_4$ bond-bending modes. At frequencies above 10 THz the match values decrease once again. This all suggests that the full phonon spectrum of Y$_2$W$_3$O$_{12}$ involves some degree of O–Y–O bond angle bending. The amount of bending is least in two NTE regions of the phonon spectrum: the strongest NTE phonons in the 0–5 THz frequency band and the weak NTE phonons in the 9–10 THz frequency band. The latter is also a region where significant O–W–O bond angle bending can been seen. Otherwise, the general trend observed is for the degree of O–Y–O bond angle bending to increase as a function of increasing frequency. The analysis of \[maxwell\_counting\] shows that the addition of the Y…W distance constraint leads to the flexibility model having more constraints than degrees of freedom. This result is confirmed by the corresponding data plotted in , which show a negligible match between the flexibility model and the ab initio phonons. Discussion ========== In materials where NTE has been studied through lattice dynamics calculations of dispersion curves and associated mode Grüneisen parameters (for example, Cu$_2$ [@Cu2O_Gupta; @Leila_Cu2O], ScF$_3$ [@ScF3_phonons], Zn(CN)$_2$ [@Hong_Zn(CN)2], MOF-5 [@Leila_MOF5]), it has usually been found that there is a well-defined set of low-frequency modes with negative mode Grüneisen parameters. In this study we have found that Y$_2$W$_3$O$_{12}$ is unusual in that a high degree of eigenvector mixing occurs, leading to the existence of broad bands of phonons with either negative or positive mode Grüneisen parameters. More specifically, this study has found that NTE in Y$_2$W$_3$O$_{12}$ is primarily driven by phonons in the 0–5 THz frequency range, with the strongest contribution coming from those phonons at the lowest frequencies. There is also some weaker contribution to NTE from another band of phonons in the 9–10 THz frequency range. All other phonons drive PTE. We had previously shown that any material containing octahedra within a corner-linked network with no non-bridging bonds cannot support RUMs [@Hammonds_octahedra] (the ScF$_3$ structure being the one exception, as discussed in reference [@Dove_RUMs]). Thus it follows that NTE in Y$_2$W$_3$O$_{12}$ does not arise from the ‘traditional’ RUM mechanism [@NTE_theory_Welche] based on rigid YO$_6$ octahedra and WO$_4$ tetrahedra. Our flexibility analysis reported here has found that the NTE phonons in this system involve motion of rigid Y–O and W–O rods as per the tension effect enabled by relatively stiff cation-oxygen bonds. However, the tension effect appears to act here as part of a more complex process. In the 0–5 THz range (which contributes strongly to NTE) the WO$_4$ tetrahedra remain effectively rigid with minimal O–W–O bond angle bending while the YO$_6$ octahedra undergo significant O–Y–O bond angle deformation. In the 9–10 THz range (which contributes less to NTE) both O–W–O and O–Y–O bond angles undergo non-trivial deformation although the O–Y–O angles actually deform less than the O–W–O. All PTE regions involve non-trivial levels of Y–O and W–O bond stretching. As noted earlier, the RUM model is only applicable to systems with stiff coordination polyhedra and is therefore not applicable to Y$_2$W$_3$O$_{12}$. The results of this study show that, in Y$_2$W$_3$O$_{12}$, such very low frequency deformations are made possible by the relatively small energy cost of O–Y–O bond angle bending in YO$_6$ units. The combination of the flexibility of the O–Y–O bond angle and the existence of effectively-rigid WO$_4$ tetrahedra and Y–O rods is such that macroscopic NTE is still achieved. By contrast, the low flexibility of the O–W–O bond angle forces the second band of NTE phonons (which resemble the ‘traditional’ tension effect) to a much higher frequency range of 9–10 THz and, thus, they cannot drive macroscopic NTE on their own. One issue that remains unexplained is the variation in the size of the negative mode Grüneisen parameter for the lowest frequency NTE phonon at the S and T points in reciprocal space. The flexibility models considered here do not highlight anything unusual happening at these points; nevertheless Grüneisen parameters do become more negative, changing from $\sim-9$ to $\sim-15$ as these points are approached. Summary ======= We have investigated the origins of negative thermal expansion in Y$_2$W$_3$O$_{12}$, a structure that is too over-constrained to support RUMs but does, nevertheless, exhibit strong NTE. Ab-initio lattice dynamics calculations were used to generate phonon data that were analysed using a variety of flexibility models that identified NTE as being driven primarily by vibrations of effectively rigid WO$_4$ tetrahedra and Y–O rods. Such a mechanism can exist at the low frequencies necessary for macroscopic NTE because deformation of the O–Y–O bond angle in Y$_2$W$_3$O$_{12}$ involves very little energy cost. We have also observed that, unlike other recently-studied NTE materials, there is a high degree of eigenvector mixing occurring in Y$_2$W$_3$O$_{12}$. Eigenvectors associated with motions such as bond-bending are mixed across phonons spanning the whole frequency range so completely that there are no modes significantly associated with just one type of motion. As a result, unlike many NTE systems where specific phonons drive NTE, in Y$_2$W$_3$O$_{12}$ NTE is characterised by large bands of phonons with the same sign—and similar value—of mode Grüneisen parameter. Acknowledgements {#acknowledgements .unnumbered} ================ The authors are grateful to Keith Refson for his helpful advice and input. We are grateful for support from NERC (NE/I528277/1, studentship for LHNR), Innovate UK (KTP009358), and CrystalMaker Software Ltd. Via our membership of the UK’s HPC Materials Chemistry Consortium, which is funded by EPSRC (EP/F067496), this work made use of the facilities of HECToR, the UK’s national high-performance computing service, which is provided by UoE HPCx Ltd at the University of Edinburgh, Cray Inc and NAG Ltd, and funded by the Office of Science and Technology through EPSRC’s High End Computing Programme. Flexibility analysis {#maxwell_counting} ==================== At several points in this paper we use a rigidity analysis following the approach originally developed by James Clerk Maxwell [@Maxwell_1864]. In a ‘Maxwell count’ we may consider a structure to be a set of points (in this case atoms) connected by rods (in this case bonds). Each point has three degrees of freedom ($x$, $y$ and $z$ translations) and each rod is taken to be rigid and thus represents a constraint. The number of excess degrees of freedom $N_\mathrm{e}$ is determined by the number of degrees of freedom $N_\mathrm{f}$ and the number of constraints $N_\mathrm{c}$ such that, $$N_\mathrm{e} = N_\mathrm{f} - N_\mathrm{c}$$ The larger the positive value of $N_\mathrm{e}$, the more flexible the structure. Conversely, if $N_\mathrm{e}$ is negative, then the structure is over-constrained and has no flexibiility. To illustrate this approach we can calculate the constraints and degrees of freedom of an isolated WO$_4$ tetrahedron. This has 5 atoms and therefore $N_\mathrm{f}=5\times3=15$ degrees of freedom, while the 4 W–O bonds contribute 4 constraints. To make the tetrahedron rigid requires an additional 5 constraints on the internal bond angles (these can be thought of in terms of adding O–O bonds but, whilst there are 6 such bonds in a WO$_4$ tetrahedron, only 5 are required to make the structure rigid with the sixth being redundant). $N_\mathrm{c} = 4 + 5 = 9$ and thus, the number of excess degrees of freedom of a rigid WO$_4$ tetrahedron is $N_\mathrm{e}=15-9 =6$, which corresponds to the number of degrees of freedom of a three-dimensional rigid body (i.e. 3 translational plus 3 rotational). Similarly, for an isolated YO$_6$ octahedron with 7 atoms, $N_\mathrm{f}=7\times3=21$. The octahedron also contains 6 Y–O bonds which act as constraints and, to make the octahedron completely rigid, it requires another 9 constraints on its internal bond angles. Thus $N_\mathrm{c} = 6 + 9 = 15$ and $N_\mathrm{e}=21-15=6$. Once again, this is the standard result for a three-dimensional rigid body. Applying Maxwell counting to the crystal structure of Y$_2$W$_3$O$_{12}$, we note that the system has 4 formula units per unit cell with Y–O–W linkages between WO$_4$ tetrahedra and YO$_6$ octahedra and no non-bridging bonds. Modelling Y$_2$W$_3$O$_{12}$ as a series of unconstrained atoms, we have $N_\mathrm{f} = 17 \times 3 \times 4 = 204$; this number corresponds to the number of normal modes for any wave vector for this crystal. Using the above, we can now add the constraints associated with the flexibility models of Y$_2$W$_3$O$_{12}$ introduced in . First, we consider the model with rigid Y–O and W–O rods but otherwise flexible polyhedra. A single unit cell contains 8 YO$_6$ octahedra and 12 WO$_4$ tetrahedra and, thus, 48 Y–O bonds and 48 W–O bonds. For this model, $N_\mathrm{c}=48+48=96$ and $N_\mathrm{e}=204-96 =108$. The model comprising rigid Y–O rods and WO$_4$ tetrahedra has 48 constraints from the rods plus $9 \times12=108$ constraints from the tetrahedra, giving $N_\mathrm{c}=108+48=156$ and $N_\mathrm{e} = 204-156=48$. The model comprising rigid W–O rods and YO$_6$ octahedra has 48 constraints from the rods plus $15 \times 8=120$ constraints from the octahedra, giving $N_\mathrm{c} = 120+48=168$ and $N_\mathrm{e} = 204-168=36$. Meanwhile, a ‘classic’ RUM model comprising rigid WO$_4$ tetrahedra and YO$_6$ octahedra has 108 constraints from the tetrahedra and 120 constraints from the octahedra, giving $N_\mathrm{c} = 108+120=228$ and $N_\mathrm{e} = 204-228=-24$. This result is consistent with the assertion that a system of corner-linked rigid tetrahedra and octahedra with no non-bridging bonds is over-constrained and cannot support any RUMs [@Hammonds_octahedra]. An additional constraint, in the form of a Y–W rod, can be incorporated to explore the possibility of the Y…W distance in Y$_2$W$_3$O$_{12}$ remaining unchanged. Each unit cell would contain 48 Y…W rods, therefore, if we consider an otherwise unconstrained system containing only 8 Y and 12 W atoms then $N_\mathrm{f} = (8+12)\times3=60$. The addition of Y…W rods gives $N_\mathrm{e} = 60-48=12$. Such a system can, therefore, support motion in which the structure flexes without changing any of the Y…W distances. Applying a Y…W rod to the flexibility models described above causes the number of excess degrees of freedom to fall to 60 for the Y–O/W–O rod model; 0 for the WO$_4$ tetrahedron/Y–O rod model and $-12$ for the YO$_6$ octahedron/W–O rod model. Thus, the rigid rod model remains flexible to some extent with a rigid Y…W distance; the WO$_4$ tetrahedron/Y–O rod model reaches the balance between flexibility and rigidity, and the YO$_6$ octahedra/W–O rod model becomes over-constrained. Plotting enhanced dispersion curves {#draw} =================================== All results in this paper were processed using a program written specifically for this work. To draw dispersion curves, the program takes data in the form of phonon frequencies and eigenvectors for a given set of wave vectors. It uses these eigenvectors to match modes from one wave vector to the next via a calculation of $\mathbf{e}_\mathbf{k} \cdot \mathbf{e}_{\mathbf{k} + \delta \mathbf{k}}$, where $\mathbf{e}_\mathbf{k}$ is any phonon eigenvector at wave vector $\mathbf{k}$ and neighbouring wave vectors in the calculation differ by $\delta \mathbf{k}$. For small $\delta \mathbf{k}$, only one of the products of eigenvectors will have a value close to unity, and this continuity of points between wave vectors is easy identified. Dispersion curves are drawn directly and saved as an encapsulated PostScript (EPS) file, rather than by printing out a list of ordered data for another plotting program. This allows for the plotting of dispersion curves for many directions in reciprocal space simultaneously. a is an example of this application. The same eigenvector matching approach is used when numerically evaluating the difference in frequencies between identical modes for lattice dynamics calculations from different volumes. Here, the program takes as input two sets of phonon frequencies and their eigenvectors. Given phonon data calculated for the same material and the same wave vectors but at different pressures/volumes, the program initially matches up equivalent modes for the two different pressures/volumes via the comparison of eigenvectors using the product $\mathbf{e}_\mathbf{k} \cdot \mathbf{e}^\prime_{\mathbf{k}}$, where $\mathbf{e}_\mathbf{k}$ and $\mathbf{e}^\prime_{\mathbf{k}}$ are eigenvectors of phonons at the same wave vector but different volumes. It then calculates the mode Grüneisen parameters using the two sets of mode frequencies. When dispersion curves are plotted using the above approach, each mode is shaded according to its corresponding value of the mode Grüneisen parameter: red is chosen for negative values, blue for positive values, with both colours graduating towards white as values approach zero. b is an example of this application. The third capability of the program is to compare the eigenvectors of two separate phonon calculations, calculating a ‘match’ parameter using and converting this match value to a grey-scale value for the small section of the plotted dispersion curve, with black representing a value of 1 and white representing a value of 0. is an example of this application. The program was written in a modern dialect of Fortran (2008). It is available from the authors of this paper (send email to the corresponding author), together with a detailed instruction document. Other examples of its use are to be found in references [@Leila_Cu2O], [@Hong_Zn(CN)2] and [@Leila_MOF5]. Implementation of flexibility models {#flexibility_models} ==================================== As in our work on Cu$_2$O [@Leila_Cu2O], we constructed simple models that represent an idealised flexibility of the Y$_2$W$_3$O$_{12}$ crystal structure. Bonds included in any flexibility model were simulated using a potential energy function of the form $$\label{bond_stretch_energy} E_\mathrm{bond} = \frac{1}{2} k (r-r_0)^2$$ where $E_\mathrm{bond}$ is the energy of the bond, $r$ is the length of a bond, $r_0$ represents the expected bond length, and k is a constant with a large value. Bond angles were represented by a similarly simple function of the form $$\label{angle_stretch_energy} E_\mathrm{angle} = \frac{1}{2} K (\theta-\theta_0)^2$$ where $E_\mathrm{angle}$ is the energy of the bond, $\theta$ is the angle of a bond, $\theta_0$ is an ‘ideal’ (typically of order $90^\circ$ or $109.47^\circ$), and $K$ is a constant. The intention of this model is to give zero or near-zero (relative to the value of $\Omega$ in ) frequencies for phonons that represent flexible modes of the model; namely modes whose eigenvectors do not involve stretching or bending of the specified bonds. The exact values of $k$ and $K$ used in and will not change the eigenvectors of these modes (which are fixed by geometry and topology) and will primarily change the frequencies only of the modes that involve stretching and bending of bonds and whose frequencies will be larger than $\Omega$. Tuning of these parameter values against frequencies of the real material is not desirable, since the models are only abstractions designed to identify the rigid and flexible regions of the crystal structure and are not intended reflect any other properties of the real system. The flexibility models used in this paper were implemented using the GULP lattice modelling program [@GULP_2003]. References {#references .unnumbered} ========== [10]{} Barrera G D, Bruno J A O, Barron T H K and Allan N L 2005 Negative thermal expansion. Journal of Physics: Condensed Matter 17 R217–R252 Romao C P, Miller K J, Whitman C A, White M A and Marinkovic B A 2013 Comprehensive inorganic chemistry II, (Elsevier, Amsterdam) Chap. 4.07, Negative Thermal Expansion (Thermomiotic) Materials, pp. 127–151 Lind C 2012 Two decades of negative thermal expansion research: Where do we stand? Materials 5 1125–1154 Takenaka K 2012 Negative thermal expansion materials: technological key for control of thermal expansion. Science and Technology of Advanced Materials 13 013001 Miller W, Smith C, Mackenzie D and Evans K 2009 Negative thermal expansion: a review. Journal of Materials Science 44 5441–5451 Carpenter M A, Salje E K H, Graeme-Barber A, Wruck B, Dove M T and Knight K S 1998 Calibration of excess thermodynamic properties and elastic constant variations due to the $\alpha \leftrightarrow \beta$ phase transition in quartz. American Mineralogist 83 2–22 Welche P R L, Heine V and Dove M T 1998 Negative thermal expansion in beta-quartz. Physics and Chemistry of Minerals 26 63–77 Tiano W, Dapiaggi M and Artioli G 2003 Thermal expansion in cuprite-type structures from 10 K to decomposition temperature: Cu$_2$O and Ag$_2$O. Journal of Applied Crystallography 36 1461–1463 Sanson A, Rocca F, Dalba G, Fornasini P and Grisenti R 2006 Negative thermal expansion and local dynamics in Cu$_2$O and Ag$_2$O. Physical Review B 73 214305 Gupta M K, Mittal R, Chaplot S L and Rols S 2014 Phonons, nature of bonding, and their relation to anomalous thermal expansion behavior of M$_2$O (M = Au, Ag, Cu). Journal of Applied Physics 115 093507 Rimmer L H N, Dove M T, Winkler B, Wilson D J, Refson K and Goodwin A L 2014 Framework flexibility and the negative thermal expansion mechanism of copper(I) oxide Cu$_2$O. Physical Review B 89 214115 Greve B K, Martin K L, Lee P L, Chupas P J, Chapman K W and Wilkinson A P 2010 Pronounced negative thermal expansion from a simple structure: cubic ScF$_3$. Journal of the American Chemical Society 132 15496–15498 Li C W, Tang X, Munöz J A, Keith J B, Tracy S J, Abernathy D L, Fultz B 2011 Structural relationship between negative thermal expansion and quartic anharmonicity of cubic ScF$_3$. Physical Review B 107 195504 Mary T A, Evans J S O, Vogt T and Sleight A W 1996 Negative thermal expansion from 0.3 to 1050 Kelvin in ZrW$_2$O$_8$. Science 272 90–92 Pryde A K A, Hammonds K D, Dove M T, Heine V, Gale J D and Warren M C 1996 Origin of the negative thermal expansion in ZrW$_2$O$_8$ and ZrV$_2$O$_7$. Journal of Physics: Condensed Matter 8 10973–10982 Evans J S O, David W I F and Sleight A W 1999 Structural investigation of the negative-thermal-expansion material ZrW$_2$O$_8$. Acta Crystallographica B 55 333–340 Cao D, Bridges F, Kowach G R and Ramirez A P 2002 Frustrated soft modes and negative thermal expansion in ZrW$_2$O$_8$. Physical Review Letters 89 215902 Cao D, Bridges F, Kowach G R and Ramirez A P 2003 Correlated atomic motions in the negative thermal expansion material ZrW$_2$O$_8$: A local structure study. Physical Review B 68 014303 Evans, J S O, Mary T A and Sleight A W 1998 Negative thermal expansion in Sc$_2$(WO$_4$)$_3$. Journal of Solid State Chemistry 137 148–160 Goodwin A L and Kepert C J 2005 Negative thermal expansion and low-frequency modes in cyanide-bridged framework materials. Physical Review B 71 140301(R) Fang H, Dove M T, Rimmer L H N and Misquitta A J 2013 Simulation study of pressure and temperature dependence of the negative thermal expansion in Zn(CN)$_2$. Physical Review B 88 104306 Han S S and Goddard III W A 2007 Metal-organic frameworks provide large negative thermal expansion behavior. Journal of Physical Chemistry C 111 15185–15191 Dubbeldam D, Walton K S, Ellis D E and Snurr R Q 2007 Exceptional negative thermal expansion in isoreticular metal-organic frameworks. Angewandte Chemie International Edition 46 4496–4499 Wu Y, Kobayashi A, Halder G J, Peterson V K, Chapman K W, Lock N, Southon P D and Kepert C J 2008 Negative thermal expansion in the metal-organic framework material Cu$_3$(1,3,5-benzenetricarboxylate)$_2$. Angewandte Chemie International Edition 47 8929–8932 Giddy A P, Dove M T, Pawley G S and Heine V 1993 The determination of rigid-unit modes as potential soft modes for displacive phase transitions in framework crystal structures. Acta Crystallographica A 49 697–703 Hammonds K D, Dove M T, Giddy A P, Heine V and Winkler B 1996 Rigid unit phonon modes and structural phase transitions in framework silicates. American Mineralogist 81 1057–1079 Dove M T, Gambhir M, Hammonds K D, Heine V and Pryde A K A 1996 Distortions of framework structures. Phase Transitions 58 121–143 Heine V, Welche P R L and Dove M T 1999 Geometric origin and theory of negative thermal expansion in framework structures. Journal of the American Ceramic Society 82 1793–1802 Hammonds K D, Bosenick A, Dove M T and Heine V 1998 Rigid Unit Modes in crystal structures with octahedrally-coordinated atoms. American Mineralogist 83, 476–479 Forster P M and Sleight A W 1999 Negative thermal expansion in Y$_2$W$_3$O$_{12}$. International Journal of Inorganic Materials 1 123–127 Sumithra A and Umarji A M 2005 Hygroscopicity and bulk thermal expansion in Y$_2$W$_3$O$_{12}$. Materials Research Bulletin 40 167–176 Evans J S O, Mary T A and Sleight A W 1997 Negative thermal expansion in a large molybdate and tungstate family. Journal of Solid State Chemistry 133 580–583 T. A. Mary and A. W. Sleight 1999 Bulk thermal expansion for tungstate and molybdates of the type A$_2$M$_3$O$_{12}$. Journal of Materials Research 14 912–915 Sleight A W 1995 Thermal contraction. Endeavour 19 64–68 Sumithra S, Waghmare U V and Umarji A M 2007 Anomalous dynamical charges, phonons, and the origin of negative thermal expansion in Y$_2$W$_3$O$_{12}$. Physical Review B 76 024307 Rimmer L H N, Dove M T, Goodwin A L and Palmer D C 2014 Acoustic phonons and negative thermal expansion in MOF-5. Physical Chemistry Chemical Physics 16 21144–21152 Wang L, Wang F, Yuan P-F, Sun Q, Liang E-J, Jia Y and Guo Z-X 2013 Negative thermal expansion correlated with polyhedral movements and distortions in orthorhombic Y$_2$Mo$_3$O$_{12}$. Materials Research Bulletin 48 2724–2729 Romao C P, Miller K J, Johnson M B, Zwanziger J W, Marinkovic B A and White M A 2014 Thermal, vibrational, and thermoelastic properties of Y$_2$Mo$_3$O$_{12}$ and their relations to negative thermal expansion. Physical Review B 90 024305 Marinkovic B A, Jardim P M, de Avillez R T and Rizzo F 2005 Negative thermal expansion in Y$_2$Mo$_3$O$_{12}$. Solid State Sciences 7 1377–1383 Clark S J, Segall M D, Pickard C J, Hasnip P J, Probert M I J, Refson K and Payne M C 2005 First principles methods using CASTEP. Zeitschrift für Kristallographie 220 567–570 Perdew J P, Burke K and Ernzerhof M 1996 Generalized gradient approximation made simple. Physical Review Letters 77 3865–3868 Perdew J P, Burke K and Ernzerhof M 1997 Generalized gradient approximation made simple, erratum \[Phys. Rev. Lett. 77, 3865 (1996)\]. Physical Review Letters 78 1396 Monkhorst H J and Pack J D 1976 Special points for Brillouin-zone integrations. Physical Review B 13 5188–5192 Baroni S, de Gironcoli S, and Dal Corso A 2001 Phonons and related crystal properties from density-functional perturbation theory. Reviews of Modern Physics 73 515–562 Refson K, Tulip P R and Clark S J (2006) Variational density-functional perturbation theory for dielectrics and lattice dynamics. Physical Review B 73 155114 Aroyo M I, Perez-Mato J M, Orobengoa D, Tasci E, de la Flor G and Kirov A 2011 Crystallography online: Bilbao Crystallographic Server. Bulgarian Chemical Communications 43 183–197 Aroyo M I, Perez-Mato J M, Capillas C, Kroumova E, Ivantchev S, Madariaga G, Kirov A and Wondratschek H 2006 Bilbao Crystallographic Server I: Databases and crystallographic computing programs. Zeitschrift für Kristallographie 221 15–27 Aroyo M I, Kirov A, Capillas C, Perez-Mato J M and Wondratschek H, 2006 Bilbao Crystallographic Server II: Representations of crystallographic point groups and space groups. Acta Crystallographica A 62 115–128 Tasci E S, de la Flor G, Orobengoa D, Capillas C, Perez-Mato J M and Aroyo M I 2002 An introduction to the tools hosted in the Bilbao Crystallographic Server, in EPJ Web of Conferences, Vol. 22 Bilbao Crystallographic Server, Last accessed 24th October 2013. http://www.cryst. ehu.es. Figueirêdo C A and Perottoni C A 2007 B3LYP density functional calculations on the ground-state structure, elastic properties, and compression mechanism of ZrW$_2$O$_8$. Physical Review B 75 184110 Gale J D and Rohl A L 2003 The General Utility Lattice Program (GULP). Molecular Simulation 29 291–341 Bradley C and Cracknell A 1972 The Mathematical Theory of Symmetry in Solids: Representation Theory for Point Groups and Space Groups (Oxford University Press). Maxwell J C 1864 On the calculation of the equilibrium and stiffness of frames. Philosophical Magazine 27 294–299	{ "pile_set_name": "ArXiv" }
--- abstract: 'Internet of Medical Things (IoMT) can connect many medical imaging equipments to the medical information network to facilitate the process of diagnosing and treating for doctors. As medical image contains sensitive information, it is of importance yet very challenging to safeguard the privacy or security of the patient. In this work, a deep learning based encryption and decryption network (DeepEDN) is proposed to fulfill the process of encrypting and decrypting the medical image. Specifically, in DeepEDN, the Cycle-Generative Adversarial Network (Cycle-GAN) is employed as the main learning network to transfer the medical image from its original domain into the target domain. Target domain is regarded as a “Hidden Factors" to guide the learning model for realizing the encryption. The encrypted image is restored to the original (plaintext) image through a reconstruction network to achieve an image decryption. In order to facilitate the data mining directly from the privacy-protected environment, a region of interest(ROI)-mining-network is proposed to extract the interested object from the encrypted image. The proposed DeepEDN is evaluated on the chest X-ray dataset. Extensive experimental results and security analysis show that the proposed method can achieve a high level of security with a good performance in efficiency.' author: - 'Yi Ding, , Guozheng Wu, Dajiang Chen, , Ning Zhang, , Linpeng Gong, Mingsheng Cao, , and Zhiguang Qin, [^1]' title: 'DeepEDN: A Deep Learning-based Image Encryption and Decryption Network for Internet of Medical Things' --- Image encryption, Deep learning, Medical image, Internet of Medical Things Introduction ============ The Internet of Medical Things (IoMT) is an interdisciplinary filed which adopts the Internet of Things (IoT) technologies in the domain of medicine [@Gatouillat2018; @Zhang2018; @Chen2017File]. With the development of IoMT, many medical imaging equipments are widely connected to facilitate the process of diagnosing and treating for doctors, e.g., the brain magnetic resonance imaging (MRI) for brain tumor diagnosis and the computed tomography (CT) of lung for lung nodule detection. In IoMT, medical images are usually managed by a system called Picture Archiving and Communication Systems (PACS) [@Liu2020]. When a patient is scanned by the medical imaging equipment, the generated medical images will be firstly stored into the PACS. When the doctor begins to examine the patient, the PACS will retrieve the needed images from the database and transfer the images to the doctors’ workstation which works with the patient information from the Hospital Information System (HIS). Although the PACS and HIS operate in an intranet environment, there are still some critical security issues when storing, transferring, and reviewing medical images, which preserve sensitive privacy information of patients. If an attacker, either an internal or an external attacker, has the ability to intrude the PACS or HIS, it becomes much easy to eavesdrop these medical images, resulting in severe privacy information leak of patients [@kzhang2018; @Chen2018Channel; @DChen2018]. ![image](fig3.png){width="96.00000%"} To safeguard the IoMT system and protect the patients’ privacy, encryption and decryption can be performed on medical images, e.g., Data Encryption Standard (DES), Advanced Encryption Standard (AES) and the Hash Function [@b1; @b2]. In addition, image encryption based on chaotic systems are also employed in the literature [@b3]. However, these methods are hard to achieve a good balance between the security performance and the encryption efficiency. Deep learning also holds great potential in dealing with this issue, where multi-layer neural networks extract a hierarchy of features from raw input images. The Convolutional Neural Network (CNN) [@b4; @Ale2019] has demonstrated the significant advantages in computer vision [@b5; @b6; @b7; @b8; @add1; @add2; @add3; @add4] as well as in image domain transfer [@b9; @b10]. Transferring the image from one domain onto another can be considered as a problem of texture transfer where the goal is to learn the mapping relationship between an input image and an output image from a set of aligned image pairs. One of the most popular image-to-image transformation method is the Cycle-Consistent Adversarial Networks [@b11], which introduces two cycle consistency losses that transform the image from one domain to the other, and then reconstruct back to the original image. In fact, the deep learning algorithm has also been adopted to solve the problem of image denoising [@b12]. Inspired by the above works, in this work, a deep learning based image encryption and decryption network (DeepEDN) is proposed for image-to-image transformation and image denoising. The novel idea is based on the following two important insights: (1) If the medical image can be transferred into other image domain which is greatly different from the original one, this medical image can be regarded as encrypted; and (2) the medical image decryption process can be implemented in the manner of image denoising or image reconstruction. In DeepEDN, the Cycle-GAN network is employed as the main learning network to implement the image-to-image transformation. There are two domains in the encryption process: the original medical image domain and the target domain, where the target domain is regarded as a “Hidden Factors” to guide the learning model to realize the encryption process. For the encryption network, it consists of a generation network and a discriminator network. The former will generate the image similar to the target domain, while the latter will promote the generation network to generate the same images as the target domain by identifying the generated images. Therefore, after processing using the encryption network, the original medical image can be converted into the target domain and becomes the ciphertext. The decryption process is similar to traditional encryption-decryption methods, which is the inverse operation of the encryption process. In practice, a reconstruction network, which is actually a decryption procedure, is used to restore the encrypted image to the original one. In DeepEDN, the parameters of generation network is regarded as the private key for encryption while the parameters of reconstruction network is regarded as the private key for decryption. Moreover, DeepEDN adopts the unsupervised learning to train the learning network and it doesn’t need much labeled samples. It overcomes the dataset issues in training and is beneficial to the application of deep learning in cryptography filed. Based on DeepEDN, the PACS system is improved by employing a key generation server. As shown in Fig.\[fig:DLEDNet\], the key generation server is in charge of training the encryption network and the decryption network. The PACS system can call the encryption network to encrypt the medical image and then store these ciphertext images into the image database. When reviewing, the HIS system will adopts the decryption network to decrypt the ciphertext image to the original one. The encryption network and the decryption network will be transferred over the secure channel. Moreover, a ROI-mining-network is proposed to directly extract the ROI (organ or tissue) from the encrypted medical image without decryption. More specific, when inputting an encrypted medical image into the ROI-mining-network, the interested segmented object can be directly extracted without revealing other parts of the patient’s information. In a nutshell, the main contributions of this work are summarized as follows: 1\. An novel medical image encryption and decryption network, DeepEDN, is developed to realize the encipherment process by applying the deep learning in the field of image-to-image transformation. The proposed encryption method is with the large key space, one-time pad, and be sensitive to key change. To the best of our knowledge, this work is the first work to attempt to adopt the deep learning method in the area of medical image encryption. 2\. A ROI-mining-network is proposed to directly extract the interested segmentation region from the encrypted medical image instead of decrypting the ciphertext image firstly. From to the experiments, it can be found that the proposed approach can realize the data mining process directly from the privacy-protected environment. 3\. Extensive experiments are conducted on the chest Xray dataset to evaluate the proposed DeepEDN. The results demonstrate that the medical image can be transmitted with a high level of security and efficiency, compared with existing medical image encryption methods. Moreover, the proposed encryption algorithm can resist various attacks, even if the attacker has known the complete process for key generation. The remainder of this paper is as follows. Section \[sec:related work\] gives an introduction of the image encryption and deep learning. Section \[sec:endecryptnet\] presents the details of the proposed DeepEDN. Section \[sec:securityAnalysis\] analyzes the security of proposed method. Section \[sec:experiment\] shows the encryption and decryption performance and evaluates the network efficiency. Section \[sec:conclusion\] gives an summarization. Related Work {#sec:related work} ============ Medical Image Encryption ------------------------ In the literature, there are many approaches proposed for image encryption [@b14; @b15; @b16]. Lima et al [@b14] propose a novel scheme for encryption of medical images, based on the cosine number transformation, which is a mathematical tool requiring modular arithmetic only. This property avoids rounding-off errors and allows that the image recovered after the encryption/decryption process is identical to the corresponding original image. The proposed scheme is flexible and can be applied to images complying with the DICOM standard, which is frequently employed in medical applications. Natsheh et al [@b15] propose a simple and effective encryption approach for multi-frame DICOM medical images. It can reduce the encryption and decryption time of these images by using AES. Mukhedkar et al [@b16] demonstrate that image encryption can be done using Blowfish Algorithm as it is faster and LSB technique is used for image hiding. Chaotic maps have the characteristics of pseudo-randomness, ergodicity and initial value sensitivity. Chaotic sequences generated by chaotic maps have good characteristics of security keys. Chaotic cryptography has gradually become a new research direction of cryptography. Medical image encryption can also be performed using chaotic maps. Kanso et al [@b17] propose a novel full and selective chaos-based image encryption scheme suitable for medical image encryption. The proposed approach consists of several rounds, where each round has two phases: a shuffling phase and a masking phase. Both phases are block-based and use chaotic cat maps to shuffle and mask an input image. Chong et al [@b18] present a novel chaos-based medical image encryption scheme which introduces a substitution mechanism in the permutation process through a bit-level shuffling algorithm in order to improve the efficiency. In order to protect image effectively, Yu et al [@b19] present an image encryption algorithm based on wavelet function and four-dimension chaotic system. The algorithm firstly uses Wavelet function chaotic maps to scramble the image pixel location, and then use four-dimension chaotic system disturbs image pixel value. Image-to-Image Transfer by Generative Adversarial Networks ---------------------------------------------------------- Since the seminal work by Goodfellow et al [@b20] in 2014, a series of GAN-based methods have been proposed for a wide variety of applications. The original GAN can learn a generator to capture the distribution of real data by introducing an adversarial discriminator that evolves to discriminate the difference between the real data and the fake data [@b20]. GAN-based methods can produce state-of-the-art results in many applications such as image generation [@b21], image segmentation [@b22], image super-resolution [@b23], and image-to-image transformation [@b24][@b25]. Yi et al [@b27] use a conditional generative adversarial network to learn a mapping from input to output images. It is proved that this approach is effective on synthesizing photos from label maps, reconstructing objects from edge maps, and colorizing images. DualGAN [@b26] mechanism enables image translators to be trained from two sets of unlabeled images. Taking two sets of unlabeled images as the input, DualGAN simultaneously learns two reliable image translators from one domain to the other and hence can facilitate a wide variety of image-to-image transformation tasks. Cycle-GAN proposed in [@b11] is a framework to perform image transformation with unpaired training data. To achieve this goal, it trains two sets of GAN models at the same time, mapping from class A to class B and from class B to class A, respectively. The loss is formulated based on the combined mapping that maps images to the same class. The key to GANs’ success is the idea of an adversarial loss that forces the generated images to be distinguishable from target images. The adversarial loss is used to learn the mapping of the original images to the “target domain images” which represents the Image-to-Image transformation. Regardless of their merits, these algorithms are difficult to achieve a good balance between the security and the efficiency. On the one hand, as there is plenty of information in a single medical image with high correlation among these information, when encrypting the medical image, the block encryption algorithm is with low efficiency and cannot meet the real-time requirement. On the other hand, the chaotic system usually adopts the one-dimensional chaotic map to generate pseudorandom sequences. Consequently, the chaotic system tends to be easy to analyze and predict through a nonlinear prediction method based on phase-space reconstruction[@b27]. The deep learning algorithm has been used in the security field [@b28; @Chen2018An; @b29]. However, there is no work on the medical image encryption and decryption. In this paper, deep learning techniques are used to encrypt and decrypt medical images, in which parameters of the deep learning network model are regarded as the encryption and decryption keys. Due to the large key space and the complex model structure, the proposed method can achieve a high level of security with a high efficiency. Encryption and Decryption Network {#sec:endecryptnet} ================================= ![The overall framework of DeepEDN.[]{data-label="fig:overallDLEDNet"}](fig17.png){width="48.00000%"} Architecture of DeepEDN ----------------------- As shown in Fig.\[fig:overallDLEDNet\], DeepEDN mainly consists of three subnetworks: the encryption network G, the discriminator network D, and the decryption network F. The encryption network G is used to encrypt the original input images, the decryption network F is responsible for restoring the encrypted images to the original one (decrypting the image), and the discriminator network D is mainly designed for improving the performance of the encryption network by distinguishing the generated images from the images in target domain (Hiding Factors). In deep learning methods, loss function is usually used to train the model. The overall loss L of the proposed model is given as follows: $$L = L_G+L_D+L_R\label{eq},$$ where the $L_G$ indicates the loss of the encryption network G, $L_D$ indicates the loss of the discriminator network D, and $L_R$ indicates the loss of the decryption network F. ### Encryption Network and Decryption Network The encryption network G is used to transform the original medical images into the target domain for medical image encryption. The G network begins with an initial convolution stage to spatially compress and encode the images, and the useful features extracted in this stage will be used for the following transformation. Then, nine residual blocks\[48\] are performed to construct the content and manifold features. The output images are reconstructed by two up-convolution blocks which contain a strided convolutional layer and the stride is set to 2. Finally, the prediction is exported by a 7$\times$7 convolution kernel. In addition, the structure of decryption network F is the same as the encryption network G. The proposed model includes two mappings $G : X\to Y$ and $F: Y\to X$. The goal of mapping function G is to learn how to transform the original medical images $X$ into the images $Y$ in target domain, and cheat the discriminator network D. When the discriminator network D cannot successfully distinguish whether an image is generated by the encryption network G or a real ciphertext image domain $Y$, it means that the encryption network G converts the original patient image domain $X$ into a ciphertext image domain $Y$ successfully. The loss $L_G$ of the encrypted network G is: $$L_G = min_{G}(E_{x \sim pdata(x)}{log(1-D(G(x))})\label{eq},$$ where G represents an encryption network, and D represents the discriminator network. The goal of $L_G$ is to minimize the success rate of the discriminator network D for detecting the ciphertext generated by the encryption network G. In addition to the encryption, another goal of the proposed method is to ensure that the restored image reserves the texture information of the original one even it is encrypted. As shown in Fig.\[fig:overallDLEDNet\], for each image $x$ from domain $X$, the reconstruction loss measures the difference between $G(x)$ and the original image, i.e., $x\rightarrow G(x)\rightarrow F(G(x))\approx x$. The reconstruction loss L is defined as: $$\begin{split} L_R &= E_{x \sim p_{data(x)}}{\|\|F(G(X) - X\|\|_1}\\ &= E_{x \sim p_{data(x)}}\sum_{i=1}^n\|F(G(x_i) - x_i\| %&= E_{x \sim p_{data(x)}}(\|F(G(x_1) - x_1\|\\ % &+\|F(G(x_2) - x_2\|+...+\|F(G(x_i) - x_i\|) \end{split}$$ ### Discriminator Network The discriminator network D is used to evaluate whether the output image of encryption network belongs to the target domain. For the discriminator network D, after processing with initial convolutional layers, the network employs two strided convolutional blocks to reduce the resolution of image and encode essential local features for subsequential discrimination. Then, a feature construction block and a 3$\times$3 convolutional layer are used to obtain the final result. In addition, for each convolutional layer, the Leaky ReLU (LReLU) with $\alpha=0.2$ is adopted and followed with a batch normalization (BN) layer. The training of the discriminator network D is to classify the images and check whether it comes from the ciphertext domain $Y$ or is generated by the encryption network G. The encryption network G attempts to generate an image $G(x)$ similar to the image in domain $Y$, while the discriminator network D aims to find the difference between transformed samples from $G(x)$ and real samples in $Y$. Minimizing loss $L_D$ of the discriminator network D is equivalent to maximizing the classification accuracy of the discriminator network D, which is opposite to the goal of the encryption network G. The loss $L_D$ given as follows: $${L_D = E_{x\sim pdata(x)}{log{D(x)}}+E_{x\sim pdata(x)}{log(1-D(G(x)))}}, \label{eq}$$ where G represents the encrypted network, and D represents the discriminator network. $L_D$ and $L_G$ in the GAN network form a adversarial relationship. When the two networks reach an equilibrium state, the discriminator network D can achieve $50\%$ classification accuracy for both the generated ciphertext image and the real ciphertext domain image $Y$. In other words, the ciphertext image generated by the encryption network G is very similar to the real ciphertext domain $Y$ so that the discriminator network D cannot distinguish them. ### The key generation process In DeepEDN, the final parameters of network G can be considered as the private key for encryption while the parameters of network F are regarded as the private key for decryption. ![image](fig6.png){width="96.00000%"} For encryption, the parameters for each convolutional layer are firstly randomly initialized as follows: $$W_{n} = random[w_{n, 1}, w_{n, 2},...,w_{n, j},...]\label{eq},$$ where $w_n$ is the $n^{th}$ convolutional layer and $w_{n,j}$ is the $j$-$th$ parameter of one convolutional layer. Therefore, the private key $W$ for encryption is actually composed of all the parameters of each convolutional layer, and is defined as follows: $$W = consist[W_1, W_2,...,W_n,...]\label{eq}$$ When training the encryption network, the private key for encryption is continuously updated and refined with different input images through forward propagation training process. The adversarial loss $L_{gan}$ is calculated to measure the difference between the predicted result and the target one in “Hidden Factors”, thereby guiding the network to train and update the private key for encryption. Except for the forward propagation, the back-propagation algorithm (BP) is also employed to pass loss of the entire network back to the convolutional layers. It is actually a gradient descent, which can further update the parameters in each layer to achieve better performance. The gradient descent can be described as: $$\begin{split} \theta_{j}=&\theta_{j} - \alpha \lor J(\theta)= \theta_{j} - \alpha\frac{\delta}{\theta_{j}}J(\theta)\\ = &\theta_{j} - \alpha\frac{\delta}{\theta_{j}}\frac{1}{2m}\sum_{i=1}^m(h_{\theta}(x^{i})-y^{i})^2\\ %&= \theta_{j} - \alpha\frac{1}{2m}\sum_{i=1}^m\frac{\delta}{\theta_{j}}(h_{\theta}(x^{i})-y^{i})^2\\ =& \theta_{j} - \alpha\frac{1}{2m}\sum_{i=1}^m2\frac{\delta}{\theta_{j}}(h_{\theta}(x^{i})-y^{i})(\frac{\delta}{\theta_{j}}(h_{\theta}(x^{i})-y^{i}))\\ =&\theta_{j} - \alpha\frac{1}{m}\sum_{i=1}^m(h_{\theta}(x^{i})-y^{i})(\sum_{i=0}^n\frac{\delta}{\theta_{i}}\theta_{i}x_i-\frac{\delta}{\theta_{i}}y^{i}) %=&\theta_{j} - \alpha\frac{1}{m}\sum_{i=1}^m(h_{\theta}(x^{i})-y^{i})(\frac{\delta}{\theta_{i}}(\theta_{0}x_0+\cdots+\theta_{n}x_n)-\frac{\delta}{\theta_{i}}y^{i}) \end{split}$$ where $\theta_{j}$ is the value of parameter $\theta$ in the $j^{th}$ training epoch. $\alpha$ is the learning rate and $\lor J(\theta)$ means the gradient that is passed back to the convolution layer $\theta$ in the $j^{th}$ training epoch. The generation process of the private key for decryption is similar to the process of generating the privacy key for encryption, except that the initial input of the decryption network is the predicted result of the encryption network. In addition, the loss of decryption network is the reconstruction loss, which is given in Equ. (8). $$\begin{split} L_{reconstruction} =& E_{x \sim p_{data(x)}}{\|\|F(P(X))-O(X)\|\|_1}\\ =& E_{x \sim p_{data(x)}}\sum_{i=1}^n\|F(P(x_i)) - O(x_i)\| %=& E_{x \sim p_{data(x)}}(\|F(P(x_0)) - O(x_0)\|+\cdots+\|F(P(x_i)) - O(x_i)\|) \end{split}$$ where $F()$ is the decryption network, $P(x)$ is the pixel $x$ in the predicted image, and $O(x)$ is the corresponding position pixel $x$ in the original image. The encryption network G and the decryption network F are trained in an alternative manner. When the loss becomes stable, the final parameters (privacy keys) for encryption and decryption network can be obtained. The complete privacy key generation process is shown in Fig.\[fig:keygeneration\]. After obtaining the key, the patient’s medical image can be encrypted by the encryption network G, and then decrypted by the decryption network F. The proposed medical image encryption/decryption algorithm is given in Alg. \[code:recentStart\]. \[code:recentStart\] Since the GAN model is highly nonlinear and randomly initialized, and the parameters of the learning network can be totally different at different training times. In other words, the GAN network is unstable, which is its weakness when used for computer vision tasks. However, this instability has advantages for cryptography. By utilizing this instability, the proposed deep learning based encryption method can be regarded as an one-Time Pad (OTP) method. Specifically, the parameters of encryption network are totally different after training the network at different times. Overall, due to the depth and complex structure of the learning encryption network, the proposed framework is with higher security. ROI Mining Network in Ciphertext Environments --------------------------------------------- Although various methods have achieved a good performance in protecting the image privacy, it is still a challenge to directly obtain the effective information in a ciphertext environment, e.g., extracting the desired ROI from the encrypted medical image. In DeepEDN, a ROI-mining-network is proposed to segment the region of interest from the encrypted medical image. In order to extract useful texture features in a ciphertext environment, a deeper network structure is adopted to learn semantic features to accurately segment the specific target. The input encrypted image will be processed with 5 blocks, and each block has a down-sampling convolution. In the first block, since the convolutional kernel size is set to 3$\times$3, each convolution operation can learn the local information from the input image. As the depth of the network increases, more abstracted semantic information can be obtained. Finally, by combining the output results from each convolution layer, the final prediction results can be achieved. Each block in ResNet-50 has two sub-blocks. One is the identity (ID) block in which the stride of each convolution layer is 1. The identity block is mainly used to extract abstract features through multi-layer convolution. Since the dimensions of the input and output are the same, these feature maps can be serially connected. The other basic block is Conv Block where the dimensions of input and output are different and it is used to change the dimension of the feature vector and to resize feature size through a strided convolutional layer. The CNN-based neural network commonly converts the image into a small feature map with many channels. However, by increasing the network layers, there will be a huge number of output channels and parameters, resulting in increased computational complexity and reduced network efficiency. Therefore, it is necessary to reduce the dimension of Conv Block before processing with the Identity Block. In DeepEDN, the proposed ROI-mining-network is used to implement the medical image segmentation task in the ciphertext environment. The medical image segmentation is a critical step in medical image analysis. Its purpose is to extract useful features and segment the doctors’ interested objects. The segmentation results can provide a reliable basis for clinical diagnosis and pathological research. When training the ROI-mining-network, the encrypted medical image is firstly used as the input of the network. Then, the pixel-level segmentation labels in the corresponding medical image are adopted to supervise the training process. Finally, the model parameters are updated by the mean square error (MSE). The loss function of this segmentation model is described as: $$L_{S} = \frac{1}{N}\sum_{i=0}^N (g_i - p_i)^2\label{eq}$$ where $g_i$ represents the value of the $i^{th}$ pixel in the label and $p_i$ is the predicted value of the $i^{th}$ pixel in the predicted result. $N$ represents the total number of pixels in this image. The final training result is a high-quality splitter that can segment the medical images without decryption. The usage of the ROI-mining-network is of great significance for medical image security. It can implement data mining in an untrusted environment to securely extract specific objects, which is also beneficial for protecting the privacy of patient. This network can further improve the security of medical image analysis and can be widely used in many medical applications. Adversary Model --------------- In DeepEDN, the most important factors of key generation process include the structure of the model and the chosen hidden factors. If the network structure or hidden factors leaks, the attacker can train a similar encryption network by imitating the private key generation process so as to crack the ciphertext image. This kind of attack is called as the imitation learning attack. This paper proposes three possible adversary models for imitation learning attack: the hidden factors leakage, the network architecture leakage, and both the hidden factors and the network architecture leakage. ### Hidden Factors Leakage Hidden factors leakage means that the attacker has known the hidden factors used for the encryption, and tries to employ the same hidden factors to train the attacking network with several different network architectures to decrypt the ciphertext image. There are two encryption and decryption networks with different network structures: the encryption/decryption network A and encryption/decryption network B. These two encryption and decryption networks are trained with the same hiding factor. If the decryption network B is able to recover the image encrypted by the encryption network A, it means that the attacker can crack the secure key by imitation learning attack. ### Network Architecture Leakage Network architecture leakage assumes that only the architecture of the encryption and decryption network is leaked, and the hidden factors remains confidential. In this adversary model, the attacker can decrypt the encrypted image by training the same network structure without knowing the hidden factors. The attacker can employ different hidden factors to train the same network structure to construct different decryption networks. If the attacker is able to recover the encrypted ciphertext image, the attack is successful. ### Both Hidden Factors and Network Architecture Leakage The strongest adversary model is that both the network architecture and hidden factors are leaked. In such a scenario, the attacker can train the network with the same network structure and hidden factor adopted for training the encryption/decryption network. To prevent such attacks, after each training of the network, the parameters of the encryption/decryption network representing the actual private key must be completely different. It means that the proposed encryption algorithm should be similar to the OTP and can be regarded as a chaotic encryption algorithm. Convolution Layer Name Number Size Input Channels Output Channels Parameters Total Parameters ---------------------------- ------------ ------------ -------------------- --------------------- ---------------- ---------------------- Down Convolution1 1 7$\times$7 3 32 4704 4704 Down Convolution2 1 3$\times$3 32 64 18432 23136 Down Convolution3 1 3$\times$3 64 128 73728 95864 Residual Blocks 18 3$\times$3 128 128 2564208 2661072 Up Convolution1 1 3$\times$3 128 64 73728 2734800 Up Convolution2 1 3$\times$3 64 32 18432 2753232 Up Convolution3 1 7$\times$7 32 3 4704 2757936 \[tab01\] Convolution Layer Name Number Size Input Channels Output Channels Parameters Total Parameters ---------------------------- ------------ ------------ -------------------- --------------------- ---------------- ---------------------- Block 1 2 7$\times$7 3 64 4704 4704 Block 2 3 3$\times$3 64 256 18432 23136 Block 3 12 3$\times$3 256 512 73728 95864 Block 4 18 3$\times$3 512 1024 2564208 2661072 Block 5 1 3$\times$3 1024 2048 73728 2734800 \[tab02\] Security Analysis {#sec:securityAnalysis} ================= In DeepEDN, both encryption and decryption network are constructed with 24 layers and the number of the parameters for each network is 2,757,936. The explicit specification of the network are represented in Table \[tab01\]. For the ROI-mining-network, a deeper resnet-50 architecture is adopted. The network structure of the ROI-mining-network is given in Table \[tab02\]. The dataset is the chest x-rays[@b31]. The proposed method is running on the Nvidia GTX 2080Ti. When training the network, it takes around 10 mins for each epoch of the model. Key Security Analysis --------------------- The ideal encryption scheme has the following characteristics: 1) the key space is large enough so that it can effectively resist the exhaustive attack under the premise of the existing computing power; 2) the key generated for each time should be different, i.e., the key generation should be uniform at random; and 3) the encrypted image must be highly sensitive to the key. The security of the key will be analyzed from these three characteristics in the following sections. ### Key Space Analysis The size of the key space determines the difficulty of an attacker encounters using an exhaustive attack. In this work, the key space of the proposed encryption algorithm is the number of parameters for the deep learning network, with a total of 2,757,936 parameters in the experiments. Each parameter or key is a floating point number between 0 and 1, which is 32 bits in the computer and can be expressed as a decimal number with 10 significant digits. Therefore, the key space of the encryption model can be expressed as the $(2^{32})^{2757936}$. It becomes very hard for attackers to break system and the proposed scheme can effectively resist attacks. ### Key Randomness Analysis ![The same image is encrypted with the key obtained from four networks.[]{data-label="fig:encryptedimage4"}](fig11.png){width="48.00000%"} The encryption network is trained four times with the same settings. Accordingly, the parameters of these four networks are adopted as encryption keys, i.e., Key A, Key B, Key C and Key D, respectively. The same image is encrypted with these four keys, and the encrypted images are shown in Fig. 11. Fig. \[fig:encryptedimage4\](a), Fig. \[fig:encryptedimage4\](b), Fig. \[fig:encryptedimage4\](c), and Fig. \[fig:encryptedimage4\](d) are the results obtained by encrypting the same original image from four networks. It is clear that these four images are different. The similarity among these four encrypted images (SSIM) are calculated, and the result can be found in Table \[tab03\]. The SSIM index between different images is mostly lower than 0.1 which indicates that the similarity between different images is very low. According to the experiment, it can be found that since parameters of the neural network are randomly initialized, the private keys for the medical image encryption network are totally different after every training. These difference results in different encrypted images which are processed with different encryption networks. The idea behind this is the training of the deep learning network is not stable. Different initialized parameters can lead to dramatic difference in final parameters in different training. It can be demonstrated that the proposed method is similar to OTP and can be regarded as one type of OTP method. Image A B C D ----------- ------- ------- ------- ------- a 1 0.07 0.11 0.09 b 0.07 1 0.08 0.04 c 0.11 0.08 1 0.05 d 0.09 0.04 0.05 1 : SSIM between two encrypted images. \[tab03\] ### Key Sensitivity Analysis Unlike traditional encryption algorithms, the error in deep learning models will be propagated among layers. In the convolution process, the $l^{th}$ pixel in the $N^{th}$ layer feature map is passed to a neighboring pixel of the $(N+1)^{th}$ layer via a $3\times3$ convolution kernel. When a feature point is erroneous, it will be passed to the $3\times3$ feature points in the next layer. As the depth of the convolutional network increases, the error of feature points will increase with two pixels for each laye. In the up-sampling process, this error increases exponentially with the superposition of the deconvolution operation. The experiment assumes the attacker knows the most private keys. And only about $5\%$ of key parameters are modified which is regarded as the unknown part. Then, the encrypted image is input to the network with new parameters, and the network cannot decrypt the ciphertext image to the original one. It means that even if only $5\%$ of the parameters is changed, the private key cannot encrypt or decrypt the medical image correctly. In other words, it becomes very hard for attackers to guess at least $95\%$ of the right key parameters in a key space with $(10^{10})^{2757936}$ so as to break the proposed algorithm. Ciphertext Security Analysis ---------------------------- ### Histogram Analysis To evaluate the performance of the proposed encryption network, The original image is shown in Fig. \[fig:pixeldistribution\](a) and the encrypted image is shown in Fig. \[fig:pixeldistribution\](c). Through the experiment, it can be found that the pixel distribution of the original image and the encrypted image is quite different. In Fig. \[fig:pixeldistribution\], the pixel histogram of the original chest X-ray image has a total of 57600\(240 \ 240) pixels (Fig. \[fig:pixeldistribution\](b)), in which more than 30,000 pixels have a value of 0, and more than 5000 pixels have a value of 255. The pixel distribution of original image is relatively concentrated. However, the distribution of encrypted medical images (Fig. \[fig:pixeldistribution\](d)) is more uniform, which is helpful for mitigating the statistical analysis. ### Entropy Analysis The information entropy of the encrypted image is regarded as an effective quantitative measurement for algorithms to against statistical attacks. The image information entropy represents the statistical feature of the grayscale distribution of the image. In an ideal case, the encrypted image should be similar to random noise, the grayscale distribution tends to be uniform, and the expected value should be 8. The information entropy formula is defined as follows: $$Entropy = -\sum_{l=0}^N p(l)log_2 (p(l))\label{eq}$$ where $N$ is the number of gray levels of the pixel value and $p (l)$ is the probability that the pixel value $l$ appears. The entropy metric is calculated on the encrypted medical image, and the results are given in Table \[tab04\]. It is clear that the image encrypted by the proposed method is close to the ideal value of 8 on information entropy. Experiments show that the images encrypted by the proposed method has the ability to resist the statistical attacks. Image Id 1 2 3 4 5 -------------- ------- ------- ------- ------- -------- Entropy 7.96 7.96 7.95 7.94 7.95 Image Id 6 7 8 9 10 Entropy 7.97 7.95 7.96 7.96 7.95 : Evaluation of the Entropy Effect of Our Network. \[tab04\] Security Analysis under Different Adversary Models -------------------------------------------------- The experiments are conducted to validate whether an attacker can generate a key under three different adversary models. ### ‘Hidden Factors Leakage In this experiment, four different network structures are considered, namely network A, network B, network C and network D. The training conditions are kept the same. The network structure of these four networks are shown in Table \[tab05\]. Convolution Layer Net. A Net. B Net. C Net. D ----------------------- ------------ ------------ ------------ ------------ Down Convolution1 1 1 1 1 Down Convolution2 1 1 1 1 Down Convolution3 1 1 1 1 Residual Blocks 18 15 12 9 Up Convolution1 1 1 1 1 Up Convolution2 1 1 1 1 Up Convolution3 1 1 1 1 : The network model of the different architectures. \[tab05\] The original image is encrypted by using the trained network A. The ciphertext image is then decrypted by the decryption network obtained from network A, network B, network C and network D respectively to restore the original image. As shown in Fig.\[fig:performanceonnetwork\], the original image (Fig.\[fig:performanceonnetwork\](a)) encrypted by network A (the encrypted image is shown in Fig.\[fig:performanceonnetwork\](b)), can only be correctly decrypted by the decryption network A as shown in Fig.\[fig:performanceonnetwork\](c). While the image decrypted by network B, network C and network D are visually unrecognizable and the result is shown in Fig.\[fig:performanceonnetwork\](d), Fig.\[fig:performanceonnetwork\](e), Fig.\[fig:performanceonnetwork\](f), respectively. Experiments show that even if the attacker knows the hidden factors, the “attack network” trained with different network structure still cannot be used to decrypt the ciphertext image. ![Pixel distribution of the original image and the encrypted image.[]{data-label="fig:pixeldistribution"}](fig13.png){width="48.00000%"} ![The decryption performance for different networks.[]{data-label="fig:performanceonnetwork"}](fig14.png){width="48.00000%"} ### Network Architecture Leakage In this experiment, different hidden factors are adopted to train the encryption network with the same network structure. All training conditions are kept the same. As shown in Fig.\[fig:mutualperformancenetwork\](a) and Fig.\[fig:mutualperformancenetwork\](b), two different domain images (“Hidden Factors A” and “Hidden Factors B”) are chosen as hidden factors to train the network with the same architecture. The Fig.\[fig:mutualperformancenetwork\](c) is original image, Fig.\[fig:mutualperformancenetwork\](d) is the image generated by the encrypted network which is trained by “Hidden Factors A”, and Fig.\[fig:mutualperformancenetwork\](e) presents the result of decrypting ciphertext image through the decryption network trained with “Hidden Factors B”. From the experiment, it can be found that the image generated by the encrypted network which is trained by “Hidden Factors A” cannot be decrypted by the network trained by “Hidden Factors B”. Therefore, it can be proven that the “attack network” with the same architecture trained by different hidden factors, cannot be used to decrypt the ciphertex image with each other. That is, even if attackers obtain the network architecture, they cannot train the decryption network to decrypt the encrypted image without knowing the hidden factors. ![The mutual decryption performance between networks under different hidden factors training.[]{data-label="fig:mutualperformancenetwork"}](fig15.png){width="50.00000%"} ### Both Hidden Factors and Network Architecture Leakage In this experiment, the network is trained with four times under the same hidden factors and training conditions to get the networks A, B, C and D, respectively. The experiment evaluates the decryption performance for these four networks on the same ciphertext image to verify whether the parameters generated for each network are different. As shown in Fig.11, the gray value distribution of the image decrypted by the decryption key B, the decryption key C, and the decryption key D is completely different from the image decrypted by the decryption key A. It can be clearly found that under the same training condition, the encrypted medical image encrypted by one network, cannot be decrypted by adopting the parameters in other network. Even if the model parameters are trained with the same network architecture and the same hidden factors, they cannot be used to decrypt the image with each other. Experiments show that even if both the network architecture and the hidden factors are leaked, and training the network under the same training conditions, the parameters of each network are totally different, i.e., the secure keys are different. It can be proven that DeepEDN is secure even if the network architecture and hidden factors are revealed. ![The decryption performance for these four networks on the same ciphertext image.[]{data-label="fig:decryptionperformance"}](fig20.png){width="50.00000%"} Security Analysis under Different Attack Models ----------------------------------------------- ### Ciphertext Only Attack In this type of attack, the attacker has access to a string of ciphertext, but cannot access to the corresponding plaintext. In DeepEDN, the key space of the encryption model can be expressed as $(2^{32})^{2757936}$ and it is very hard for attacker to break down. At the same time, the privacy key generated with multiple iterations and diffusions is complex. Therefore, it is difficult to crack the ciphertext through ciphertext only attacks. ### Known Plaintext Attack In this type of attack, the attacker knows a string of plaintext, and the corresponding ciphertext. The attacker will try to decrypt the rest of the ciphertext by using these known information. In traditional sequential pixel visiting pattern methods, concrete encryption factors are generally retrieved as equivalent keys for recovering the received ciphertexts. Taking XOR encryption as an example, the masks calculated directly from plaintext and ciphertext are sufficient to decode the ciphertext. Typically, masks sequentially correspond to the plain pixels and the retrieved masks by plaintext attack can be directly used to decrypt other ciphertexts. However, the proposed algorithm adopted the non-sequential encryption mechanism. Without the knowledge of the pixel visiting pattern, the privacy key cannot be obtained by the attacker, thus making plaintext attack infeasible. The proposed algorithm adopts the iteration and diffusion procedures to generate the privacy key. These kinds of producers can significantly promote the security performance and thus provide additional immunity of the cipher against known plaintext attack. ### Chosen Plaintext Attack In this type of attack, the attacker can access the encryption device, choose a string of plaintext and construct its corresponding ciphertext string. Generally, an attacker can observe the change of the ciphertext image by making small changes to the plaintext image, such as changing the value of only one pixel of the ciphertext, so as to obtain the connection between the plaintext image and the ciphertext image. This type of attack is called as differential attack which is a kind of chosen plaintext attack method. If a small change in the plaintext image can cause a huge change in the ciphertext image, this differential attack method usually fails to take effect. It indicates that the encryption algorithm can resist this chosen plaintext attack method. Here, the Number of Pixel Change Rate (NPCR) is adopted to measure the degree of image changing. NPCR refers to the rate of pixels change which indicates the ratio of different pixel values at the same position between two plaintext/ciphertext images. The definition of NPCR is as follows: $$NPCR = \frac{\sum_{i=0}^W \sum_{j=0}^H D(i, j)}{W \times H} \label{eq}$$ where $W$ and $H$ are the width and height of the image, respectively. T1 and T2 represent a ciphertext image obtained by encrypting two different plaintext images, respectively. If $T_1(i,j) = T_2(i,j)$, $D(i, j) = 1$. If $T_1(i,j) \neq T_2(i,j)$, $D(i, j) = 0$. In the experiment, there is only about $1\%$ different pixels between these two plaintext images. Both the original plaintext image and the plaintext image with $1\%$ pixel value changed are input to the proposed encryption model. Then, the NPCR is used to compare the differences between these two encrypted images. The calculated average NPCR value is $94.21\%$, which means that the information of the plaintext image is well diffused into the ciphertext image. Since DeepEDN has good diffusion performance and is with highly sensitive to the plaintext, it achieves a good performance to resist the chosen plaintext attack like the differential attack. ![image](fig8.png){width="95.00000%"} ### Chosen Ciphertext Attack In this type of attack, the attacker can access the decryption device, choose a string of ciphertext and construct its corresponding plaintext string. Since the structure of our decryption model is exactly the same as the encryption model, the experiment for chosen ciphertext attack is similar to that in chosen plaintext attack. In this experiment, the input of the decryption network is the ciphertext image and the NPCR is used to calculate the different between two decrypted images. According to the experiment, it is found that when the input ciphertext image changes slightly ( just $1\%$ pixels changed), the average NPCR value between two decrypted images is $94.87\%$. It means that if the input ciphertext image changes slightly, the decrypted image will change dramatically. This demonstrates that the proposed algorithm has good diffusion performance and is also highly sensitive to the ciphertext. It is effective to resist the chosen ciphertext attack. Experimental Results {#sec:experiment} ==================== Performance of Encryption and Decryption ---------------------------------------- As shown in Fig.\[fig:encryptionanddecryption\], the results of the proposed method for medical image encryption and decryption are presented in a visual way. It can be seen that the ciphertext image generated by the encryption network G, is totally different from the original medical image and the pathology information cannot be observed. In addition, the image in the third row is decrypted from the encrypted one through the decryption network F, can recover the detailed information of the original image and restore to the original one. In order to evaluate the effectiveness of the decryption network, the Peak Signal to Noise Ratio (PSNR) and structural similarity index (SSIM) are employed as evaluation metrics. The quantitative measure of the decryption error is PSNR, which is based on the root mean square error (RMSE) between the decrypted data and ground truth. It can be represented as: $$PSNR = 20log_{10}\frac{255}{RMSE}\label{eq}$$ To further evaluate the performance of encryption and decryption, the SSIM is used as another metric. $$SSIM(x, y) = [l(x, y)]^{\alpha}[c(x, y)]^{\beta}[s(x, y)]^{\gamma}\label{eq}$$ where $l(x, y)$ is the brightness comparison, $c(x, y)$ is the contrast comparison, and $s(x, y)$ is the structure comparison. The closer the SSIM is to 1, the more resemblance the two images are. And if this value approaches to 0, the two images are completely different. In an ideal case, the SSIM between the encrypted image and the original image is equal to 0, and the SSIM between the decrypted image and the original image is equal to 1. As shown in the second and third rows of Table VI, the SSIM between the encrypted image and the original image is close to 0, and the SSIM between the decrypted image and the original image is close to 1. For most of medical image processing tasks, the image can be compressed to one-half size of the original one to reduce the storage consumption and does not affect the doctor’s diagnosis. In order to ensure that the decrypted image do not affect the doctor’s diagnosis, the performance of the reconstructed image decrypted by the decryption network, is also compared with the one-half compressed image. According to the experiment, it is demonstrated that the performance of the reconstructed image is equivalent to that through directly compressing the original image to half and then restoring it. In Table \[tab06\], from line 3 to line 6, 2X means that the original image is compressed to one-half and then restored. At this level, human can accurately identify the patient’s organ contours and bone information from reconstructed images. Image Id 1 2 3 4 5 6 7 8 9 10 ----------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- -------- SSIM(Encrypted) 0.93 0.88 0.90 0.94 0.93 0.91 0.91 0.93 0.91 0.89 SSIM(Decrypted) 0.01 0.02 0.01 0.01 0.02 0.02 0.01 0.02 0.01 0.01 SSIM(2X) 0.90 0.92 0.90 0.92 0.89 0.91 0.88 0.90 0.91 0.90 PSNR 37.43 35.34 36.01 38.03 35.76 35.87 36.13 37.17 35.88 35.74 PSNR(2X) 35.48 35.74 35.03 35.28 34.87 36.73 34.75 34.61 36.17 34.80 \[tab06\] Performance of ROI-Mining-Network --------------------------------- Direct extraction of interested information under ciphertext conditions is of great significance for medical image security and also for the data mining with privacy protection. The proposed ROI-mining-network can segment the patient’s interested organ tissue from the ciphertext image without decrypting the image firstly. The proposed network has the ability to realize the data mining from the privacy environment by extracting the ROI from the encrypted image directly. In order to evaluate the proposed ROI-mining-network, the well-known evaluation metric Dice score is adopted in here and is defined as: $$Dice(GT, AT) = \frac{GT \cap AT}{(\|GT\|+\|AT\|)/2}\label{eq}$$ The GT represents the ground truth and the AT represents the model predictions. Fig.\[fig:performanceofROI\] shows the performance of the proposed ROI-mining-network on the patient’s left lung. It can be clearly seen that the prediction (grey ones) obtained from the model is almost as the same as the ground truth (red ones). In addition, the original medical images are also adopted as the experiment data for training the same ROI-mining-network, which is mainly used as the comparison. Under the same training conditions, the DICE of the segmentation network for plaintext is 0.967, while the DICE of the segmentation network for the ciphertext image is 0.962. It can be proven that the ROI-mining-network can achieve a good segmentation performance on both plaintext and ciphertext images. ![The performance of ROI-Mining-Network.[]{data-label="fig:performanceofROI"}](fig9.png){width="48.00000%"} As mentioned before, the privacy keys of the network are totally different when training the network at different times even if all the conditions are the same. Therefore, the attacker cannot obtain the same ROI-mining-network even if employing the same ciphertext image for training. The experiment can be found in Fig.\[fig:attackforROI\]. In this experiment, Fig.\[fig:attackforROI\] (a) is the original image and Fig.\[fig:attackforROI\] (b) is the ciphertext image of Fig.\[fig:attackforROI\] (a). Fig.\[fig:attackforROI\] (c) is the ground truth for the right lung segmentation. Fig.\[fig:attackforROI\] (d) is the correct extraction result segmented by the ROI-mining-network. Fig.\[fig:attackforROI\] (e) is the error extraction result segmented by the attacker. ![Attack experiment for the proposed ROI-Mining-Network.[]{data-label="fig:attackforROI"}](fig10.png){width="48.00000%"} Efficiency ---------- To evaluate the efficiency of the proposed network, the running speed of encryption and decryption process on different resolution medical images is evaluated. For 256\256 resolution, the proposed network can encrypt or decrypt 14.28 medical images per second, while the speed is 3.65 images/second for encrypting or decrypting 512\512 resolution image. This encryption/decryption speed can basically meet the efficiency requirement in clinical practice. In addition, some chaotic encryption algorithms have been adopted as the comparison method for evaluating the efficiency. For instance, Zhou et al[@b32] introduce a simple chaotic system, which employs a combination of two existing one-dimension (1D) chaotic maps (seed maps). Liao et al[@b33] introduce a novel image encryption algorithm based on self-adaptive wave transmission. Wu et al[@b35] introduce a wheel-switch chaotic system for image encryption. In [@b36], the two-dimensional logistic map with complicated basin structures and attractors is firstly used for image encryption. This method can encrypt an intelligible image into a random-like one both from the point of view of the statistical and the human visual system. Fig. \[fig:efficiency\] shows the comparison among aforementioned five chaotic encryption algorithms and the proposed method. The FPS represents the number of images that can be encrypted/decrypted in one second. It can be found that our methods achieve the fastest encryption speed both on 512$\times$512 resolution and 256$\times$256 resolution images. Although the number of keys in our method is greater than the number of keys used in chaotic encryption methods, the processing time of our method is still with higher efficiency. ![The efficiency comparison between our method and other existing methods.[]{data-label="fig:efficiency"}](fig21){width="45.00000%"} Conclusion {#sec:conclusion} ========== In this paper, a novel medical image encryption and decryption method (namely DeepEDN) is proposed by leveraging deep learning techniques, which is one of the early attempts to adopt the concept of “deep learning” for medical image encryption. The Cycle-GAN network is adopted as the learning network to encrypt and decrypt the medical image. A target domain is used to guide the learning model in the encryption process. The reconstruction network can decrypt the encrypted image to the original image (plaintext). Moreover, a ROI-mining-network is proposed to directly extract the ROI from the encrypted medical image, with which DeepEDN can segment the interested organ or tissue in the ciphertext environment without decrypting the medical image. We conduct experiments on the chest X-ray datasets and the results show that the proposed algorithm can protect the medical image with a high security level and can encrypt/decrypt the image in a more efficient way, compared with state-of-the-art medical image encryption methods. [00]{} A. Gatouillat, Y. Badr, B. Massot, E. Sejdi$\acute{c}$, “Internet of Medical Things: A Review of Recent Contributions Dealing With Cyber-Physical Systems in Medicine,” IEEE Internet of Things Journal, vol. 5, no. 5, pp. 3810-3822, Oct. 2018. N Zhang, P Yang, J Ren, et. al., “Synergy of big data and 5g wireless networks: opportunities, approaches, and challenges,” IEEE Wireless Communications, vol. 25, no.1, pp. 12-18, 2018. D. Chen, N. Zhang, Z. Qin, X. Mao, Z. Qin, X. Shen, and X. Y. Li, “S2M: A Lightweight Acoustic Fingerprints-Based Wireless Device Authentication Protocol,” IEEE Internet of Things Journal, vol. 4, no. 1, pp. 88-100, Feb. 2017. B. Liu, H. Huang, “Picture archiving and communication systems and electronic medical records for the healthcare enterprise,” Biomedical Information Technology, Academic Press, pp. 105-164, 2020. K. Zhang, J. Ni, K. Yang, X. Liang, J. Ren, X. S. Shen, “Security and privacy in smart city applications: Challenges and solutions,” IEEE Communications Magazine, vol. 55, no. 1, pp. 122-129, Jan. 2017. D. Chen, N. Zhang, et al., “Channel precoding based message authentication in wireless networks: Challenges and solutions", IEEE Network, vol. 33, no. 1, pp. 99-105, 2018. D. Chen, N. Zhang, N. Cheng, K. Zhang, Z. Qin, X. Shen, Physical Layer based Message Authentication with Secure Channel Codes, IEEE Transactions on Dependable and Secure Computing, DOI: 10.1109/TDSC.2018.2846258. Y. Zhang, W. Liu, S. Cao, Z. Zhai, X. Nie, and W. Dai, “Digital image encryption algorithm based on chaos and improved DES,” in Proc. 2009 IEEE International Conference on Systems, Man and Cybernetics, USA, pp. 474-479, Mar. 2009. K. Chang, Y. Chen, C. Hsieh, C. Huang and C. Chang, “Embedded a Low Area 32-bit AES for Image Encryption/decryption Application,” in Proc. 2009 IEEE International Symposium on Circuits and Systems, Taipei, pp. 1922-1925, Apr. 2009. M. Preishuber, T. Hutter, S. Katzenbeisser and A. Uhl, “Depreciating Motivation and Empirical Security Analysis of Chaos-Based Image and Video Encryption,” IEEE Transactions on Information Forensics and Security, vol. 13, no. 9, pp. 2137-2150, Sept. 2018. Y. LeCun, B. Boser, J. Denker, et. al., “Backpropagation Applied to Handwritten Zip Code Recognition,” Neural Computation, vol. 1, no. 4, pp. 541-551, Dec. 1989. L. Ale, N. Zhang, H. Wu, et. al., “Online Proactive Caching in Mobile Edge Computing Using Bidirectional Deep Recurrent Neural Network, IEEE Internet of Things Journal, vol. 6, no. 3, pp. 5520-5530, 2019. H. Chen, Z. Qin, Y. Ding, et.al., “Brain tumor segmentation with deep convolutional symmetric neural networ,” Neurocomputing, DOI:10.1016/j.neucom.2019.01.111. Y. Ding, C. Luo, et.al., “High-order correlation detecting in features for diagnosis of Alzheimer’s disease and mild cognitive impairment,” Biomedical Signal Processing and Control, vol. 53, Sept. 2019. K. H. Jin, M. T. McCann, E. Froustey, et.al., “Deep Convolutional Neural Network for Inverse Problems in Imaging,” IEEE Transactions on Image Processing, vol. 26, no. 9, pp. 4509-4522, Sept. 2017. C. Dong, C. C. Loy, K. He and X. Tang, “Image Super-Resolution Using Deep Convolutional Networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 38, no. 2, pp. 295-307, Feb. 2016. B. Xiao, G. Ou, H. Tang, et.al., “Multi-Focus Image Fusion by Hessian Matrix Based Decomposition,” IEEE Transactions on Multimedia, vol. 22, no. 2, pp. 285-297, Feb. 2020. B. Xiao, K. Wang, X. Bi,et.al., “2D-LBP: An Enhanced Local Binary Feature for Texture Image Classification,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 29, no. 9, pp. 2796-2808, Sept. 2019. H. Tang, B. Xiao, W. Li, et al.,“Pixel Convolutional Neural Network for Multi-Focus Image Fusion,” Information Sciences, vol.433, pp.125-141, Sept. 2018, DOI:0.1016/j.ins.2017.12.043. B. Xiao et al., “Follow the Sound of Children¡¯s Heart: A Deep-Learning-Based Computer-Aided Pediatric CHDs Diagnosis System,” IEEE Internet of Things Journal, vol. 7, no. 3, pp. 1994-2004, March 2020. P. Isola, J. Zhu, T. Zhou and A. A. Efros, “Image-to-Image Translation with Conditional Adversarial Networks,” in Proc. IEEE CVPR2017, USA, pp. 5967-5976, June. 2017. A. Cherian and A. Sullivan, “Sem-GAN: Semantically-Consistent Image-to-Image Translation,” in Proc. IEEE WACV2019, USA, pp. 1797-1806, Oct. 2019. J. Zhu, T. Park, P. Isola and A. A. Efros, “Unpaired Image-to-Image Translation using Cycle-Consistent adversarial networks,” in Proc. IEEE ICCV2017, Italy, pp. 2242-2251, Oct. 2017. K. Zhang, W. Zuo and L. Zhang, “FFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising,” IEEE Transactions on Image Processing, vol. 27, no. 9, pp. 4608-4622, Sept. 2018. J. Lima, E. Neto. “Audio encryption based on the cosine number transform,” Kluwer Academic Publishers, Oct. 2016, DOI:10.1007/s11042-015-2755-6. Q. N. Natsheh, B. Li and A. G. Gale, “Security of Multi-frame DICOM Images Using XOR Encryption Approach,” Procedia Computer Science, vol. 90, pp. 175-181, July. 2016. M. Mukhedkar, P. Powar and P. Gaikwad, “Secure non real time image encryption algorithm development using cryptography $\&$ steganography,” in Proc. IEEE INDICON 2015, India, pp. 1-6, July. 2015. A. Kanso and M. Ghebleh, “An efficient and robust image encryption scheme for medical applications,” Communications in Nonlinear Science and Numerical Simulation, vol. 24, pp. 98-116, Jan. 2015. C. Fu, W. Meng, Y. Zhan, et.al., “An efficient and secure medical image protection scheme based on chaotic maps,” Computers in Biology and Medicine, vol. 43, no. 8, pp. 1000-1010, May. 2013. W. Yu, C. Chi, X. Wei and X. Yang, “Image encryption algorithm based on high-dimensional chaotic systems,” in Proc. 2010 International Conference on Intelligent Control and Information Procession, China, pp. 463-467, Nov. 2010. I. Goodfellow, J. Pouget-Abadie, M. Mirza, et.al., “Generative adversarial nets,” in Proc. NIPS2015, Canada, pp. 2672-2680, Dec. 2015. J. Bao, D. Chen, F. Wen, H. Li and G. Hua, “CVAE-GAN: Fine-Grained Image Generation through Asymmetric Training,” in Proc. IEEE ICCV2017, Italy, pp. 2764-2773, Oct. 2017. Y. Li and L. Shen, “cC-GAN: A Robust Transfer-learning Framework for HEp-2 Specimen Image Segmentation,” IEEE Access, vol. 6, pp. 14048-14058, Oct. 2018. W. Liu, X. Liu, H. Ma and P. Cheng, “Beyond Human-level License Plate Super-resolution with Progressive Vehicle Search and Domain Priori GAN,” in Proc. the 25th ACM International Conference on Multimedia, USA, pp. 1618-1626, Oct. 2017. Y. Zhou, L. Bao and C. L. P. Chen, “A new 1D chaotic system for image encryption,” Signal Process, vol. 97, pp. 172-182, Oct. 2014. X. Liao, S. Lai and Q. Zhou, “A novel image encryption algorithm based on self-adaptive wave transmission,” Signal Process, vol. 90, pp. 2714-2722, Mar. 2010. Z. Hua, Y. Zhou, C. Pun and C. L. P. Chen, “2D Sine Logistic modulation map for image encryption,” Information Sciences, vol. 297, pp. 80-94, Feb. 2015. Y. Wu, J. Noonan and S. Agaian, “A wheel-switch chaotic system for image encryption,” in Proc. 2011 International Conference on System Science and Engineering, Macao, pp. 23-27, May 2011. Y. Wu, G. Yang, H. Jin and J. Noonan, “Image encryption using the two-dimensional logistic chaotic map,” Journal of Electronic Imaging, vol. 21, pp. 3014-3022, Jan. 2012. A. Cherian and A. Sullivan, “Sem-GAN: Semantically-Consistent Image-to-Image Translation,” in Proc. IEEE WACV2019, USA, pp. 1797-1806, Oct. 2019. N. Wang, W. Zha, J. Li and X. Gao, “Back projection: An effective postprocessing method for GAN-based face sketch synthesis,” Pattern Recognition Letters, vol. 107, pp. 59-65, June. 2018. Z. Yi, H. Zhang, P. Tan and M. Gong, “DualGAN: Unsupervised Dual Learning for Image-to-Image Translation,” in Proc. IEEE ICCV2017, Venice, Mar. 2017, pp. 2868-2876. D. Arroyo, R. Rhouma, G. Alvarez, et.al., “On the security of a new image encryption scheme based on chaotic map lattices,” Chaos (Woodbury, N.Y.), vol. 18, pp. 1-8, Aug. 2008. F. Jiang, Y. Fu, B. Gupta, et.al., “Deep Learning based Multi-channel intelligent attack detection for Data Security,” IEEE Transactions on Sustainable Computing, DOI: 10.1109/TSUSC.2018.2793284. D. Chen, N. Zhang, et. al., “An LDPC code based physical layer message authentication scheme with prefect security", IEEE Journal on Selected Areas in Communications, vol. 36, no. 4, pp. 748-761, 2018. A. Ferdowsi and W. Saad, “Deep Learning for signal authentication and security in massive Internet of Things Systems,” IEEE Transactions on Communications, vol. 67, no. 2, pp. 1371-1387, Feb. 2019. K. He, X. Zhang, S. Ren and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE CVPR2016, USA, pp. 770-778, Sept. 2016. S. Jaeger, S. Candemir, S. Antani, et.al. “Two public chest X-ray datasets for computer-aided screening of pulmonary diseases,” Quantitative imaging in medicine and surgery, vol. 4, pp. 475-7, Dec. 2014. [Yi Ding]{} received the B.S. degree in software engineering from the University of Electronic Science and Technology of China and the Ph.D. degree in computer science from the Dublin Institute of Technology, Ireland, in 2008 and 2012, respectively. From 2012 to 2016, he was a postdoc research fellow at University of Electronic Science and Technology of China. Since 2014, he has been an associate professor in the School of Information and Software Engineering, University of Electronic Science and Technology of China. He is also a researcher in Institute of Electronic and Information Engineering of UESTC in Guangdong, China. His research interested include machine learning, deep learning, medical image processing, and computer-aided diagnosis. [Guozheng Wu]{} received the B.S. degree in Information Management and Decision Science from the University of Science and Technology of China and the M.S. degree in Public Management from the Tsinghua University, China, in 1997 and 2002, respectively. He received the Ph.D degree from University of Electronic Science and Technology of China in 2010. He is currently a staff in National Natural Science Foundation of China. His research interested include computer networking, information security, cryptography, information management and software engineering. [Dajiang Chen]{} (M’15) is currently an Associate Professor in the School of Information and Software Engineering at University of Electronic Science and Technology of China (UESTC). He was a Post Doctoral Fellow with the Broadband Communications Research (BBCR) group, Department of Electrical and Computer Engineering, University of Waterloo, Canada, from 2015 to 2017. He received the Ph.D. degree in information and communication engineering from UESTC in 2014. Dr. Chen served as the workshop chair for BDEC-SmartCity’19 in conjunction with IEEE WiMob 2019 and the organizer for IoT track in conjunction with EAI CollaborateCom 2020. He also serves/served as a Technical Program Committee Member for IEEE Globecom, IEEE ICC, IEEE VTC, IEEE WPMC, and IEEE WF-5G. His current research interest includes Wireless Security, Physical Layer Security, Secure Channel Coding, and Machine Learning and its applications in Wireless Network Security and Wireless Communications. [Ning Zhang]{} (M’15-SM’18) received B.E. degree and M.S. degree from Beijing Jiaotong University and Beijing University of Posts and Telecommunications in 2007 and 2010, respectively. He received the Ph.D degree from University of Waterloo, Canada, in 2015. From 2015 to 2017, he was a postdoc research fellow at University of Waterloo and University of Toronto, Canada, respectively. Since 2017, He has been an Assistant Professor at Texas A$\&$M University-Corpus Christi, USA. He serves/served as an Associate Editor of IEEE Transactions on Cognitive Communications and Networking, IEEE Access and IET Communications, an Area Editor of Encyclopedia of Wireless Networks (Springer) and Cambridge Scholars. He also served as the workshop chair for MobiEdge’18 (in conjunction with IEEE WiMob 2018) and CoopEdge’18 (in conjunction with IEEE EDGE 2018), and 5G$\&$NTN’19 (in conjunction with IEEE ICCC 2019). He is a recipient of the Best Paper Awards from IEEE Globecom in 2014, IEEE WCSP in 2015, Journal of Communications and Information Networks in 2018, IEEE Technical Committee on Transmission Access and Optical Systems in 2019, and IEEE ICC in 2019, respectively. His current research interests include next generation mobile networks, physical layer security, machine learning, and mobile edge computing. [Linpeng Gong]{} received B.E. degree from University of Electronic Science and Technology of China in 2016. He is currently pursuing M.S. degree from the University of Electronic Science and Technology of China. His current research interests include deep learning, computer vision, digital image security and Internet of Things. [Mingsheng Cao]{} currently is a lecturer at Network and Data Security Key Laboratory of Sichuan Province, University of Electronic Science and Technology of China, ChengDu. He received the B.Sc. and M.Sc. degree from UESTC, 2008 and 2011, respectively. In 2019, he obtained the PhD. degree from UESTC. His research interests include Network Security, Pervasive Computing, and Machine Learning and its applications in Network Security and Pervasive Computing. [Zhiguang Qin]{} (S’95-A’96-M’14) is the Director of the Key Laboratory of New Computer Application Technology and Director of UESTC-IBM Technology Center. Dr. Qin was the Dean of the School of Software of University of Electronic Science and Technology of China (UESTC). His research interests include wireless sensor networks, mobile social networks, Information Security, Applied Cryptography, Information Management, Intelligent Traffic, Electronic Commerce, Distribution and middleware, etc. Dr. Qin served as the General Co-Chair for WASA 2011, Bigcom 2017, etc. [^1]: Corresponding authors: Dajiang Chen ([email protected]) and Ning Zhang ([email protected])	{ "pile_set_name": "ArXiv" }
--- abstract: 'Fuchsian equations provide a way of constructing large classes of spacetimes whose singularities can be described in detail. In some of the applications of this technique only the analytic case could be handled up to now. This paper develops a method of removing the undesirable hypothesis of analyticity. This is applied to the specific case of the Gowdy spacetimes in order to show that analogues of the results known in the analytic case hold in the smooth case. As far as possible the likely strengths and weaknesses of the method as applied to more general problems are displayed.' author: - \| Alan D. Rendall\ Max–Planck–Institut für Gravitationsphysik\ Am Mühlenberg 1, 14476 Golm, Germany title: Fuchsian analysis of singularities in Gowdy spacetimes beyond analyticity --- \#1\#2 Introduction ============ The theory of Fuchsian equations has been applied to analyse singularities in a variety of classes of spacetimes in general relativity. The term Fuchsian equations has not always been used in the literature on this subject and in this paper it denotes a certain class of singular differential equations in a generic way. The existing results in this area will be surveyed briefly below. In this approach spacetimes containing singularities are parametrized by some functions which play the role of data on the singularity. In some cases it was necessary to assume the analyticity of these functions. In other cases smoothness was sufficient. The aim of this paper is to develop ways of removing the analyticity requirement. These will be illustrated by the case of Gowdy spacetimes which represent an ideal laboratory for testing new ideas on the mathematical treatment of the Einstein equations. It may not be immediately obvious why the apparently technical distinction between analytic ($C^{\omega}$) functions and smooth ($C^\infty$) functions should be significant with a view to physical applications. There are at least two reasons why it is important. The first is that the physical notion of causality cannot be reasonably formulated within the class of analytic functions, since the unique continuation property of the latter means that the solution of an equation at one point influences its behaviour at all other points, and not only at causally related points. Connected with this is the fact that solutions of an equation do not depend continuously on initial data in any useful sense. For more discussion of these points see [@friedrich00], particularly section 2.4. The second, which is also a direct consequence of the unique continuation property of analytic functions, is that there is not the same freedom to construct solutions with certain interesting properties within the analytic class. An example of this will be given in section \[discussion\] below. There are a number of results on Fuchsian equations with smooth coefficients in the literature and to start with we need to understand why these do not apply directly to Gowdy spacetimes. The general form of a system of Fuchsian equations for a vector-valued unknown function $u$ is $$\label{fuchs} t\frac{\d u}{\d t}+N(x)u=tf(t,x,u,D_x u)$$ Here $x$ is a point in some Euclidean space and $D_x u$ is a shorthand for the first order derivatives of $u$ with respect to the spatial variables $x$. The function $f$ is required to extend continuously to $t=0$ while the matrix $N(x)$ is required to satisfy some positivity condition, which may depend on the particular context. An example of a condition of this type is that the eigenvalues of the matrix $N(x)$ should have non-negative real parts for all $x$. Note that the apparently more general system $$\label{fuchsalpha} t\frac{\d u}{\d t}+N(x)u=t^\alpha f(t,x,u,D_x u)$$ with $\alpha>0$ can be reduced to the form (\[fuchs\]) by introducing $t^\alpha$ as a new time variable. This results in the matrix $N(x)$ being rescaled by a factor $\alpha$, but this does not affect its positivity properties. One approach to proving existence theorems for Fuchsian systems which does not require any analyticity assumption is due to Claudel and Newman [@claudel98]. Of course, in the context of data which are merely smooth (or even finitely differentiable) the system must be hyperbolic. This is needed to prove existence for an equation without any singular behaviour in $t$ and the singularity cannot be expected to improve the situation. In the Claudel-Newman theorem it is required that $f$ have an asymptotic expansion about $t=0$ in integral powers of $t$ and this property is inherited by the solution. The positivity assumption on $N(x)$ is weaker than that mentioned above. It is only required that there be no eigenvalues which are negative integers. An important element of the proof of the theorem is to expand the candidate solution $u$ in powers of $t$, writing it in the form $u_0+u_1$ where $u_0$ is an appropriate polynomial in $t$ and $u_1$ is a remainder of higher order in $t$. Then $u_1$ solves a Fuchsian system where the eigenvalues of $N(x)$ are shifted by an integer in the positive direction. If it is possible to expand to a sufficiently high order then the shifted eigenvalues all have positive real parts. The condition which makes the expansion possible is that a polynomial $u_0$ can be found which satisfies the equation up to an error of sufficiently high order in $t$. The only obstruction to this is if the shifted eigenvalue becomes zero at some stage in the process and this is prevented by the assumption made on the eigenvalues of the original matrix. In cases such as the system arising in the analysis of Gowdy spacetimes in [@kichenassamy98] the solutions cannot be expanded in integral powers of $t$. Instead non-integral (and even $x$-dependent) powers of $t$ and logarithms occur. For this reason the method of [@claudel98] does not apply directly. One of the main methods of the present paper is to extend the technique of using expansions of the solution to shift the eigenvalues of $N(x)$ to cases where terms more complicated than integral powers of $t$ occur. Another approach to proving existence theorems for Fuchsian systems with smooth coefficients is due to Kichenassamy[@kichenassamy96a]. In that case there is no restriction that solutions have expansions in integral powers of $t$. It is, however, required that the matrix $N(x)$ be independent of $x$. This is not satisfied in the Gowdy case. Since the eigenvalues of the matrix $N(x)$ correspond to powers occurring in the expansions, this is a consequence of the dependence of these powers on $x$ in the system coming from the Gowdy spacetimes. Thus the result of [@kichenassamy96a] does not apply. It might be possible to extend the proofs in [@kichenassamy96a] to the case of non-constant $N$, but this will not be attempted here. Both the proofs of Claudel and Newman and of Kichenassamy involve the use of sophisticated techniques from functional analysis, namely semigroup theory and the Yosida approximation, respectively. These will be avoided in the approach developed in the following. In the paper [@anguige99d] Anguige treats the case of plane symmetric solutions of the Einstein equations coupled to a perfect fluid. He uses a direct energy argument of a type familiar in the theory of regular symmetric hyperbolic systems. It is important in his proof that $N(x)$ is independent of $x$ and positive semi-definite. This kind of argument will be generalized in the following. On the one hand the condition that $N(x)$ should be constant will be removed. On the other hand the positivity condition on the matrix will be relaxed by a method related to that used in [@claudel98]. The eigenvalue condition used in [@claudel98] will be replaced by a condition of formal solvability which abstracts its essential significance. Next a brief survey of the literature on applications of Fuchsian equations to general relativity will be given. It appears that the first paper containing an application of this kind was [@moncrief82], where Moncrief proved the existence of a large class of analytic spacetimes with analytic compact Cauchy horizons. Later Newman [@newman93a; @newman93b] based his work on isotropic singularities and Penrose’s Weyl curvature hypothesis on existence theorems for hyperbolic systems with singularities of Fuchsian type. These papers did not include proofs of the required theorems but the necessary proofs were provided in [@claudel98]. More recently, results on isotropic singularities for more general matter models were obtained by Anguige and Tod [@anguige99a; @anguige99b; @anguige99c]. Their theorems require no symmetry assumptions but are confined to a special type of singularities. On the other hand Anguige[@anguige99d] proved a theorem on the existence of non-isotropic singularities in plane symmetric spacetimes with perfect fluid. Another line of development of the applications of Fuchsian equations in general relativity starts with the work of Kichenassamy and Rendall [@kichenassamy98] on singularities in analytic Gowdy spacetimes. It builds on previous work of Kichenassamy outside general relativity (see [@kichenassamy96b] and references therein). This direction is continued in the papers [@isenberg99], [@andersson00] and [@narita00] which deal with vacuum models with two spacelike Killing vectors, solutions of the Einstein-scalar field equations and analogues of Gowdy models in string cosmology, respectively. In all these cases analyticity is assumed. A notable feature of the result of [@andersson00] is that it makes no symmetry assumptions and so, on the basis of function-counting arguments, concerns general solutions of the Einstein equations with the given matter model. The paper is organized as follows. In the second section the notion of a formal solution of a Fuchsian system is defined. Assumptions on the coefficients are exhibited which guarantee the existence of a formal solution. They are fulfilled by the first order form of the Gowdy equations introduced in [@kichenassamy98]. In the third section this form of the equations is modified slightly so as to obtain a symmetric hyperbolic system. Its formal solvability is shown to follow from that of the original system. The system satisfied by the remainder term which is the difference between a true solution of the system and an approximate solution is computed. The fourth section proves an existence theorem which is general enough to apply to the case of Gowdy spacetimes with sufficiently low velocity ($k<3/4$). In the fifth section yet another form of the equations is used to cover the remainder of the full low velocity case ($k<1$). The wider applicablity of the methods of the paper is discussed in the final section. Formal solutions {#formal} ================ If the function $f$ in (\[fuchs\]) is smooth at $t=0$ and hence admits an asymptotic expansion about $t=0$ in integral powers of $t$ then it can be useful to expand it in this way. In the following a generalization to less smooth functions $f$ will play an important role. This uses the notion of a formal solution of equation (\[fuchs\]) which will now be defined. It will also be convenient for the following to introduce a notion of regularity of functions adapted to the given situation. An analogous notion in an analytic context was introduced in [@andersson00]. [Definition 1]{} A function $z(t,x)$ defined on an open subset of $[0,\infty)\times R^N$ and taking values in $\R^K$ is called [regular]{} if it is $C^\infty$ for all $t>0$ and if its partial derivatives (defined for $t>0$) of any order with respect to the variables $x\in\R^K$ extend continuously to $t=0$. [Definition 2]{} A finite sequence $(u_1,u_2,\ldots,u_p)$ of functions defined on an open subset $U$ of $[0,\infty)\times\R^n$ containing $\{0\}\times\R^n$ is called a [formal solution]{} of order $p$ of (\[fuchs\]) if 1. each $u_i$ is regular 2. $t\d_t u_i+N(x)u_i-tf(t,x,u_i,D_x u_i)=O(t^i)$ for all $i$ as $t\to 0$ Here and in the following the $O$-symbols are taken in the sense of uniform convergence on compact subsets. In [@kichenassamy98] an iteration was defined which, in the case that the function $f$ has suitable analyticity properties, converges to a solution of (\[fuchs\]). It is doubtful if it converges in any useful sense when $f$ is only smooth. However it can be used to produce a formal solution of any desired order, as will now be shown. [Lemma 2.1]{} If the function $f$ is regular and the matrix $N(x)$ is smooth and satisfies an estimate of the form $\\|\sigma^{N(x)})\\|\le C$ with a constant $C$ for $\sigma$ in a neighbourhood of zero then for each $p$ the equation (\[fuchs\]) has a formal solution of order $p$ which vanishes at $t=0$. [Proof]{} First some definitions from [@kichenassamy98] will be recalled. For a regular function $u$ define $F[u]=tf(t,x,u,D_x u)$. Then $F[u]$ is also regular and is $O(t)$ as $t\to 0$ together with all its spatial derivatives. If $v$ is regular and $O(t)$ as $t\to 0$ together with all its spatial derivatives, define $u=H[v]=\int_0^1 \sigma^{N(x)-I}v(\sigma t)d\sigma$. Then $u$ is regular and $O(t)$ together with all its spatial derivatives and satisfies $(t\d_t+N)u=v$. Then, if $G$ is defined to be the composition $FH$, any fixed point $v$ of $G$ defines a solution $u$ of (\[fuchs\]) by $u=H[v]$. Let $u_1=0$. It defines a formal solution of (\[fuchs\]) of order one. It will be shown that defining $u_i=HG^{i-1}[u_1]$ defines a formal solution of order $p$ for each $p$. Note the relation $u_{i+1}=HFu_i$. The first defining property of a formal solution is easily proved by induction. The main point is to verify the second property. To do this it will first be shown that $u_{i+1}-u_i=O(t^i)$ for each $i$, and that spatial derivatives of all orders of the $u_i$ satisfy the corresponding estimates. For $i=1$ this follows directly from the properties already demonstrated. To prove the general case, consider the equation obtained by forming the difference of the equations satisfied by $u_{i+1}$ and $u_i$. This gives $$\label{udiff} t\d_t (u_{i+1}-u_i)+N(x)(u_{i+1}-u_i)=tM_1(u_i-u_{i-1})+tM_2 D_x(u_i-u_{i-1})$$ for regular functions $M_1$ and $M_2$ of the arguments $t$, $x$, $u_i$, $u_{i-1}$, $D_x u_i$, $D_x u_{i-1}$, obtained by applying the mean value theorem to differences. The right hand side of (\[udiff\]) is $O(t^i)$. Then the fact can be applied that the operator $H$ preserves the set of functions which are $O(t^j)$ for any $j$. Spatial derivatives can be handled in the same way. To complete the proof of the lemma, consider the relation: $$\begin{aligned} \label{remainder} &&t\d_t u_{i+1}+N(x)u_{i+1}-tf(t,x,u_{i+1},D_x u_{i+1})\nonumber \\ &&=-t[f(t,x,u_{i+1},D_x u_{i+1})-f(t,x,u_i,D_x u_i)]\end{aligned}$$ Using the mean value theorem and the estimates already obtained for $u_{i+1}-u_i$ shows that the right hand side of (\[remainder\]) is $O(t^{i+1})$. [Remark]{} A general criterion for checking the condition on $N(x)$ required to apply this lemma has been given in ([@andersson00]). Next some basic equations for the Gowdy spacetimes will be recalled. More details can be found in [@kichenassamy98]. The basic unknowns are two real-valued functions $X(t,x)$ and $Z(t,x)$ of two variables. New variables $u$ and $v$ are defined so that the relations \[zansatz\] Z(t,x)=k(x)t+(x)+t\^u(t,x) and \[xansatz\] X(t,x)=X\_0(x)+t\^[2k(x)]{} ((x)+v(t,x)) hold, where $k$, $X_0$, $\phi$ and $\psi$ are given functions. The positive constant $\epsilon$ will be restricted later. Next introduce further variables by setting (u\_0,u\_1,u\_2,v\_0,v\_1,v\_2)=(u,t\_t u,tu\_x,v,t\_t v,tv\_x) The Gowdy equations imply the following first order system: \[reduced1\] t\_t u\_0&=&u\_1\ t\_t u\_1&=&-2u\_1-\^2 u\_0+t\^[2-]{}(k\_[xx]{}t +\_[xx]{}) +t\_x u\_2\ &&-(-2-2t\^u\_0){t\^[2k-]{}(v\_1+2kv\_0+2k)\^2 -t\^[2-2k-]{}X\_[0x]{}\^2\ &&-2t\^[1-]{}X\_[0x]{}(v\_2+t\_x+k\_x(v\_0+)tt)\ &&-t\^[2k-]{}(v\_2+t\_x+2k\_x (v\_0+) tt)\^2}\ t\_t u\_2&=&t\_x(u\_0+u\_1)\ t\_t v\_0&=&v\_1\ t\_t v\_1&=&-2kv\_1+t\^[2-2k]{}X\_[0xx]{}+t\_x(v\_2+t\_x)+4k\_x(v\_2+t\_x)tt\ &&+(v\_0+)\[2k\_[xx]{}t\^2t+4(k\_x tt)\^2\]\ &&+2t\^(v\_1+2kv\_0+2k)(u\_1+u\_0)\ &&-2X\_[0x]{}t\^[2-2k]{}(k\_xt+\_x+t\^\_x u\_0)\ &&-2t(\_x(v\_0+)+2k\_x(v\_0+)t)(k\_xtt+t\_x+t\^u\_2)\ t\_t v\_2&=&t\_x(v\_0+v\_1) Here some minor errors in the equations given in [@kichenassamy98] have been corrected[^1]. This system is of the form (\[fuchsalpha\]) which implies, as indicated in the introduction, that it can be brought into the form (\[fuchs\]) by a change of time variable. This possibility will be used freely without further comment in the following. After the change of time coordinate the system arising from the Gowdy equations satisfies the hypotheses of Lemma 2.1 provided $\epsilon<2k$ and $\epsilon<2-2k$. In particular, the bound on $\sigma^{N(x)}$ was verified directly in [@kichenassamy98]. Alternatively, it follows easily from the criterion given in [@andersson00]. Hence the above system has a formal solution of any order which vanishes at $t=0$. The symmetric hyperbolic system {#symhyp} =============================== In the following the form of the Gowdy equations introduced in section \[formal\] will be called the first reduced system. Now it will be modified to get a system which, while less convenient for showing formal solvability, is symmetric hyperbolic and therefore appropriate for allowing the theory of hyperbolic equations to be applied. It is obtained from the first reduced system by making the substitutions $u_2=t\d_x u_0$ and $v_2=t\d_x v_0$ in some places. The result is: \[reduced2\] t\_t u\_0&=&u\_1\ t\_t u\_1&=&-2u\_1-\^2 u\_0+t\^[2-]{}(k\_[xx]{}t +\_[xx]{}) +t\_x u\_2\ &&-(-2-2t\^u\_0){t\^[2k-]{}(v\_1+2kv\_0+2k)\^2 -t\^[2-2k-]{}X\_[0x]{}\^2\ &&-2t\^[1-]{}X\_[0x]{}(v\_2+t\_x+k\_x(v\_0+)tt)\ &&-t\^[2k-]{}(v\_2+t\_x+2k\_x (v\_0+) tt)\^2}\ t\_t u\_2&=&u\_2+t\_x u\_1\ t\_t v\_0&=&v\_1\ t\_t v\_1&=&-2kv\_1+t\^[2-2k]{}X\_[0xx]{}+t\_x(v\_2+t\_x)+4k\_x(v\_2+t\_x)tt\ &&+(v\_0+)\[2k\_[xx]{}t\^2t+4(k\_x tt)\^2\]\ &&+2t\^(v\_1+2kv\_0+2k)(u\_1+u\_0)\ &&-2X\_[0x]{}t\^[2-2k]{}(k\_xt+\_x+t\^[-1]{}u\_2)\ &&-2(v\_2+t\_x+2tk\_x(v\_0+)t)(k\_xtt+t\_x+t\^u\_2)\ t\_t v\_2&=&v\_2+t\_x v\_1 This system, which will be referred to as the second reduced system, has the advantage of being symmetric hyperbolic but also has two potential disadvantages. The one is that the matrix $N(x)$ has been modified while the second is that possibly dangerous powers of $t$ have been introduced on the right hand side. The matrix $N(x)$ acquires two negative eigenvalues, which will have to be dealt with by appropriate methods in due course. As far as the other problem is concerned, the power $t^{1+\epsilon-2k}$ is introduced. This power should be positive. This can be achieved subject to the inequalities already assumed for $\epsilon$ if and only if $k<3/4$. This restriction appears unnatural, but will be assumed in this section and the next for the second reduced system. In fact, although the desired inequalities relating $\epsilon$ and $k(x)$ can be ensured by a suitable choice of $\epsilon$ at any given point $x$, this cannot be done at all points simultaneously by a single choice of the constant $\epsilon$. It can be ensured in a neighbourhood of any given point. Solutions will be constructed in local neighbourhoods of this kind and then put together using the domain of dependence to get a solution which is global in $x$. Any formal solution $\{u_1,\ldots,u_p\}$ of the first reduced system which vanishes at $t=0$ is also a formal solution of the second reduced system. For $t\d_t((u_2-t\d_x u_0)_i)=O(t^i)$ and using the fact that the formal solution vanishes at $t=0$ it follows that $(u_2-t\d_x u_0)_i=O(t^i)$. This can then be used to see that the difference terms arising when passing from the first to the second reduced systems are $O(t^i)$, assuming the condition $k<3/4$. Although it is not of significance for the following it is interesting to note that for $3/4<k<1$ the sequence whose element with index $i$ is the element of the formal solution of the first reduced system with index $i+1$ is a formal solution of the second reduced system. Given a formal solution it is possible to consider the difference between an actual solution and the formal solution and the equations which this difference satisfies. The hope is that this equation is more tractable analytically than the first and second reduced systems. This is a generalization of the procedure of subtracting a Taylor polynomial of finite order in the case that the solutions are smooth at $t=0$. Rather than doing this calculation in the specific case of the Gowdy system it will be done for the following more general symmetric hyperbolic Fuchsian system: $$\label{symhypeq} t\d_t u+N(x)u+tA^j(t,x,u)\d_j u=tf(t,x,u)$$ The condition for symmetric hyberbolicity is that the matrices $A^j$ should be symmetric. As before, all coefficients in the equation are assumed regular. If $\{u_i\}$ is a formal solution and $v_i=t^{-i}(u-u_i)$ then $$\label{diffeq} t\d_t v_i+(N(x)+iI)v_i+tA^j(t,x,u_i+t^iv_i)\d_j v_i=tg_i(t,x,v_i)$$ for some regular function $g_i$. Choosing $i$ large enough ensures that the eigenvalues of $N+iI$ have positive real parts, or even that the matrix is positive definite. In the Gowdy case this will be referred to as the third reduced system. The basic existence theorem {#existence} =========================== In this section a local existence theorem will be proved for solutions of the Gowdy equations in a neighbourhood of the initial singularity in the case that the data $k$, $X_0$, $\phi$ and $\psi$ are merely smooth. In the case of analytic data the problem was solved in [@kichenassamy98]. Thus if the data are approximated by a sequence of analytic data $(k_m,X_{0m}, \phi_m,\psi_m)$, a corresponding sequence of analytic solutions is obtained. At the same time formal solutions can be obtained for both the approximate data and the actual data. Denote the former by $u_{mi}$, where the first index corresponds to the sequence of data and the second to the enumeration of elements of an approximate solution. Denote the latter by $u_i$, as before. If the approximate solutions are constructed as in the proof of the Lemma 2.1 then $u_{mi}\to u_i$ as $m\to\infty$, uniformly on compact subsets. The same is true of the spatial derivatives of these functions of any order. It can be concluded that the sequence of coefficients obtained for the third reduced system is also convergent on compact sets as $m\to\infty$. The sequence of solutions of these equations which we have is only defined a priori on a time interval which depends on $m$. However, using the global existence theorem for the Gowdy equations [@moncrief80] it is possible to conclude that a sequence of smooth solutions exists on a common time interval. The aim now is to show that this is a Cauchy sequence in a sufficiently strong topology. If that can be done then it will follow that the sequence converges to a limit which is a solution corresponding to the smooth data originally prescribed. The tool to obtain convergence of the approximations is the technique of energy estimates. This requires some preliminary remarks on linear algebra. Consider the matrix-valued function $N(x)$ in the Fuchsian system. Spatial derivatives of a solution of this system also satisfy a system of the same form, but with a different matrix $N(x)$. Suppose, for instance, we consider the first derivative $D_xu$ of the unknown. The system for the pair $(u,D_xu)$ has a matrix in its singular term which has diagonal blocks $N(x)$ and an off-diagonal block involving $D_xN(x)$. This off-diagonal block does not affect the eigenvalues of the matrix but may well affect whether it is positive definite or not. Since the positive definiteness of matrices like this is important in what follows we adopt a strategy which avoids positivity being lost. In order to implement this it will be assumed that $N(x)$ is positive definite. Then use the variable $w=KD_xu$ for a positive constant $K$ instead of $D_x u$ itself. For the equation satisfied by $(u,w)$ the matrix of interest is positive definite provided $K$ is chosen sufficiently small. The same trick works for higher derivatives. It suffices to replace the collection of unknowns $\{D^\alpha u\}$ by $w^\alpha=K^{\|\alpha\|}D^\alpha u$. Let the matrix corresponding to $N$ arising in the system for all these derivatives up to order $s$ be denoted by $N^{(s)}$. By construction it is positive definite. The standard method of energy estimates (see e.g. [@taylor96] or, for a discussion aimed at relativists, [@friedrich00]) proceeds by estimating the Sobolev norms of solutions. The usual Sobolev norm is given by $\\|u\\|_{H^s}=(\sum_{\|\alpha \|\le s}\\|D^\alpha u\\|^2_{L^2})^{1/2}$. For the present purposes it is convenient to use the equivalent norm $\\|u\\|_{H^s, K}=(\sum_{\|\alpha \|\le s}K^{2\|\alpha\|} \\|D^\alpha u\\|^2_{L^2})^{1/2}$, where $K$ is, as before, a small enough positive constant. As a first application of energy estimates, a theorem on the domain of dependence will be proved. Let $u$ and $v$ be two regular solutions of (\[symhypeq\]) vanishing at $t=0$. Then their difference satisfies an equation of the following form: $$\label{difference} \d_t (u-v)+t^{-1}N(x)(u-v)+A^j(t,x,u)\d_j (u-v)=M(t,x) (u-v)$$ where $M$ is a regular function constructed from $u$ and $v$. Choose two times $t_1$ and $t_2$ with $0<t_1<t_2$ and let $G$ be the region defined by the inequalities $t_1\le t\le t_2$ and $\|x\|\le 2t_2-t$. Let $S_1$ and $S_2$ be its intersections with $t=t_1$ and $t=t_2$ respectively. Now multiply the equation (\[difference\]) by $e^{-\kappa t}(u-v)$, integrate over $G$ and integrate by parts in the way this is usually done in the derivation of energy estimates. Here $\kappa$ is a positive constant. The singular term containing $N$ can be discarded, due to its sign, giving an estimate of the form $e^{-\kappa t_2}I_2\le e^{-\kappa t_1}I_1+I_G$, where $I_1$ and $I_2$ are the $L^2$ norms of the restrictions of $u-v$ to $S_1$ and $S_2$ respectively, and $I_G$ is a volume integral over $G$ which for $\kappa$ sufficiently large is negative unless $u-v$ is identically zero on $G$. Letting $t_1$ tend to zero, so that $I_1\to 0$, gives a contradiction unless $u-v=0$ on $G$. Thus it has been proved that the solutions $u$ and $v$ agree on $G$. This proves a domain of dependence property for solutions of (\[symhypeq\]) which can be used for the purpose of gluing together solutions. This means that even if we are interested in producing solutions on manifolds it is enough to solve the problem on $\R^n$. Moreover, it is possible to consider without loss of generality the case of cut-off coefficients and data. By this we mean that there is a compact subset of $\R^n$ such that for $x$ outside this compact set the initial data vanish, the coefficients $A^j$ and $f$ vanish and $N$ is constant. The aim is now to construct solutions of the third reduced system for Gowdy in the case of smooth coefficients. As already indicated it is enough to do this under the assumption of a cut-off in space. A sequence of functions which is a candidate for a sequence converging to a solution of the third reduced system has already been produced. This sequence is obtained by fixing a value of $i$ sufficiently large that the matrix occurring in the singular term of the third reduced system is positive definite and forming the difference of the solution of the second reduced system corresponding to the data $(k_m,\phi_m,X_{0m},\psi_m)$ and the function $u_{mi}$. It will now be shown by using energy estimates that this sequence is bounded in suitable Sobolev spaces and in fact is a Cauchy sequence. To do this differentiate $\\|u\\|_{H^s,K}^2$ with respect to $t$ and substitute the third reduced system into the result. This gives $$\label{energy} d/dt(\\|u\\|_{H^s,K})^2)=-t^{-1}\langle N^{(s)}u^{(s)},u^{(s)} \rangle_{L^2,K}+R$$ Here $u^{(s)}$ is the collection of all derivatives of $u$ up order $s$ and $R$ is the sum of the terms which arise in the regular case, i.e. in the case where $N$ is identically zero. The first term on the right hand side is non-positive and may be discarded. The terms in $R$ can be estimated just as in the regular case and this gives a bound for the $H^s$ norm of $u$, provided $s>n/2+1$. Next the difference of successive approximants will be estimated. An attempt to apply the standard techniques in order to show that the sequence is Cauchy only meets one difficulty not present in the regular Cauchy problem. This is due to differences of the matrices $N$ for successive elements of the sequence. This can be overcome in a way similar to that used above where the norms were scaled. The trick is to consider the collection of the derivatives of the functions of the sequence up to order $s$ together with the derivatives of the differences of successive elements of the sequence up to order $s-1$, the derivatives of the differences being multiplied by an additional factor $K$. This once again ensures that the final matrix obtained is positive definite. It follows from the above discussions that the sequence of solutions $v_i$ of the third reduced system converges to a solution of the third reduced system corresponding to the original smooth data. This can then be used to define solutions of the second and first reduced systems and finally a solution of the Gowdy system itself corresponding to the data we started with. The construction is such that the interval on which convergence is obtained may depend on $s$. However standard results about symmetric hyperbolic systems with smooth coefficients show that the solutions for all values of $s$ can be extended to a common time interval. Hence a solution is obtained on that time interval which is $C^\infty$ for $t>0$. The results of this discussion are summed up in the following theorem, which may be compared with Theorem 1 of [@kichenassamy96a]. [Theorem 4.1]{} Let $k(x)$, $X_0(x)$, $\phi(x)$ and $\psi(x)$ be $C^\infty$ and assume that $0<k(x)<3/4$ for all $x$. Then there exists a solution of the Gowdy equations with following properties. For each spatial point $x$ there exists an open neighbourhood $U_x$ of $x$ and a number $\epsilon_x>0$ such that the restriction of the solution to $U_x$ satisfies (\[zansatz\]) and (\[xansatz\]) with $\epsilon=\epsilon_x$, where $u$ and $v$ are regular and tend to zero as $t\to 0$. The $U_x$ and $\epsilon_x$ can be chosen in such a way that the inequalities $2k(y)-1<\epsilon_x<\min\{2k(y),2-2k(y)\}$ are satisfied for all $x$ and all $y\in U_x$. Under these conditions the solution is unique. [Remark]{} A formulation of the theorem which is equivalent but cleaner can be obtained by replacing the constant $\epsilon$ by a function $\epsilon(x)$. Then it would not be necessary to introduce the $U_x$. In [@kichenassamy96a] a theorem was proved concerning high velocity analytic solutions in the case where $X_0$ is independent of $x$. The method used to prove Theorem 4.1 applies straightforwardly to the high velocity case to give an analogue of the result of [@kichenassamy96a] in the smooth case. The following theorem results. [Theorem 4.2]{} Let $k(x)$, $\phi(x)$ and $\psi(x)$ be $C^\infty$ and let $X_0$ be a constant. Assume that $k(x)>0$. Then there exists a solution of the Gowdy equations with the following properties. For each spatial point $x$ there exists an open neighbourhood $U_x$ of $x$ and a number $\epsilon_x>0$ such that the restriction of the solution to $U_x$ satisfies (\[zansatz\]) and (\[xansatz\]) with $\epsilon=\epsilon_x$, where $u$ and $v$ are regular and tend to zero as $t\to 0$. The $U_x$ and $\epsilon_x$ can be chosen in such a way that the inequality $\epsilon_x <2k(y)$ is satisfied for all $x$ and all $y\in U_x$. Under these conditions the solution is unique. Data with intermediate velocity =============================== In the previous section an existence theorem was proved for Gowdy spacetimes under the restriction $0<k<3/4$ on the function $k$. Next it will be shown that using a different ansatz allows the range $1/2<k<1$ to be treated. The two together then cover the whole range $0<k<1$ for which results were available in the analytic case. The new ansatz involves expanding the function $Z$ to a higher order in $t$. The ansatz for $X$ remains unchanged. Now $Z$ is of the form $$\label{zansatz2} Z=k\log t+\phi+\alpha t^{2-2k}+t^{2-2k+\epsilon}u$$ where $\alpha=(2-2k)^{-2}X_{0x}^2$. Reexpressing the Gowdy equations in terms of the new variables $u$ and $v$ and reducing to first order as before leads to an analogue of the second reduced system of section \[symhyp\]. The evolution equations for $u_0$, $u_2$, $v_0$ and $v_2$ are the same as before and will not be repeated. The modified equation for $u_1$ is \[u1new\] t\_t u\_1&=&-2(2-2k+) u\_1-(2-2k+)\^2 u\_0 +t\^[2k-]{}(k\_[xx]{}t+\_[xx]{}) +t\_x u\_2\ &&+t\^[1-]{}\[t\_[xx]{}-4k\_xt(t\_x+t\^u\_2) +4k\_x\^2t(t)\^2(+t\^u\_0)\]\ &&-(-2-2t\^[2-2k]{}-2t\^[2-2k+]{} u\_0) {t\^[4k-2-]{}(v\_1+2kv\_0+2k)\^2\ &&-2t\^[2k-1-]{}X\_[0x]{}(v\_2+t\_x+k\_x(v\_0+)tt)\ &&-t\^[4k-2-]{}(v\_2+t\_x+2k\_x (v\_0+) tt)\^2} and that for $v_1$ is \[v1new\] t\_t v\_1&=&-2kv\_1+t\^[2-2k]{}X\_[0xx]{}+t\_x(v\_2+t\_x)+4k\_x(v\_2+t\_x)tt\ &&+(v\_0+)\[2k\_[xx]{}t\^2t+4(k\_x tt)\^2\]\ &&+2t\^[2-2k+]{} (v\_1+2kv\_0+2k)(u\_1+(2-2k+) u\_0)\ &&+(4-4k)t\^[2-2k]{}(v\_1+2kv\_0+2k)\ &&-2X\_[0x]{}t\^[2-2k]{}(k\_xt+\_x+t\^[2-2k]{}\_x+t\^[1-2k+]{}u\_2\ &&-2k\_xt\^[2-2k]{}t(+t\^u\_0)) -2(v\_2+t\_x+2tk\_x(v\_0+)t)\ && (k\_xtt+t\_x+t\^[3-2k]{}\_x +t\^[2-2k+]{} u\_2-2k\_x t\^[3-2k]{}t (+t\^u\_0)There is also an obvious analogue of the first reduced system of section \[formal\]. The existence of formal solutions of the latter is guaranteed by Lemma 2.1 and these give rise to formal solutions of the second reduced system as in section \[symhyp\]. Consider now the sequence of analytic solutions of the Gowdy equations corresponding to a sequence of analytic approximations to the smooth data of interest. It will be shown that, under the condition that $1/2<k<1$, these define a sequence of regular solutions of the second reduced system of this section. To do this it is necessary to show that for each of these solutions the function $Z$ admits an asymptotic expansion of the form (\[zansatz2\]) and not just of the form (\[zansatz\]), which is known a priori. To do this it suffices to apply the existence theorem of [@kichenassamy98] to the first reduced system of this section with the analytic data. Once these facts are known, it is straightforward to prove an existence theorem for the second reduced system of this section using the same techniques as were applied to the second reduced system of section \[symhyp\] provided certain inequalities are satisfied. These are the inequalities which ensure that each term on the right hand side of the equation contains a positive power of $t$. Under the assumption that $1/2<k<1$ this can be achieved by choosing the positive real number $\epsilon$ to satisfy $4k-3<\epsilon<2k-1$. The following theorem is obtained. [Theorem 5.1]{} Let $k(x)$, $X_0(x)$, $\phi(x)$ and $\psi(x)$ be $C^\infty$ and assume that $1/2<k(x)<1$ for all $x$. Then there exists a solution of the Gowdy equations with the following properties. For each spatial point $x$ there exists an open neighbourhood $U_x$ of $x$ and a number $\epsilon_x>0$ such that the restriction of the solution to $U_x$ satisfies (\[zansatz2\]) and (\[xansatz\]) with $\epsilon=\epsilon_x$, where $u$ and $v$ are regular and tend to zero as $t\to 0$. The $U_x$ and $\epsilon_x$ can be chosen in such a way that the inequalities $4k(y)-3<\epsilon_x<2k(y)-1$ are satisfied for all $x$ and all $y\in U_x$. Under these conditions the solution is unique. Discussion ========== The theorems stated in this paper have all concerned Gowdy spacetimes. It is nevertheless clear that many of the arguments are much more generally applicable. At the same time some steps are essentially related to the specific Gowdy case. A general discussion of the procedure will now be given which separates the general from the particular as much as possible. The first step is to make a suitable ansatz for the solutions to be constructed as the sum of an explicit part and a remainder. There may be more than one useful way of doing this. For example, in the Gowdy case equations (\[zansatz\]) and (\[xansatz\]) were useful for proving one theorem while replacing (\[zansatz\]) by (\[zansatz2\]) allowed a different theorem to be proved. The second step is to reduce the equations to first order. The aim is to produce a system of Fuchsian form for which the theorem of [@kichenassamy96a] ensures the existence of solutions corresponding to the case where the free functions in the ansatz are analytic. If these free functions are merely smooth the lemma proved in section 2 may be used to show the existence of formal solutions. The third step is to produce a system which is symmetric hyperbolic and in Fuchsian form. At this stage the matrix $N(x)$ may have negative eigenvalues, as is the case in the Gowdy example. It needs to be shown that the formal solutions already produced define formal solutions of the symmetric hyperbolic system. From this point on the argument proceeds on a general level, with no more details of the Gowdy special case being used. It is instructive at this stage to consider what difficulties would be likely to arise in an attempt to generalize the results of [@andersson00] from the analytic to the smooth case. One problem is to bring the equations into a suitable hyperbolic form by the choice of coordinate or gauge conditions. There was no difficulty of this type in the Gowdy case, where a rather rigid preferred coordinate system is available. In more general cases it will be necessary to choose a form of the reduced Einstein equations carefully from the myriad on offer. If a symmetric hyperbolic system is obtained it is likely to involve a matrix $A^0$ multiplying the time derivative of the unknown which is not the identity, thus going beyond the case discussed above. Even worse, it may be difficult to ensure that $A^0$ remains bounded and uniformly positive definite as $t\to 0$. These conditions are very important for the use of energy estimates. To conclude the paper, an application will be presented where the flexibility of smooth functions is essential. Existence theorems have been proved for Gowdy spacetimes in the low velocity case and, under the condition that $X_0$ is constant, also in the high velocity case. Using the domain of dependence these can be combined to give a more general class of solutions. To do this consider a smooth function $X_0$ which is constant on a non-empty open interval $I$. Now complete this to data $(k,X_0,\phi,\psi)$ in such a way that $k<1$ on the closure of the complement of $I$. Then each point $x$ has a neighbourhood on which one of the existence theorems applies and the resulting local solutions can be put together to produce a solution corresponding to the chosen initial data globally in $x$. If we tried to do this construction with analytic data then $X_0$ would have to be globally constant and nothing new would be obtained. [12]{} Andersson, L., Rendall, A. D. 2000 Quiescent cosmological singularities. Preprint gr-qc/0001047. Anguige, K., Tod, K. P. 1999 Isotropic cosmological singularities 1: Polytropic perfect fluid spacetimes. Ann. Phys. (NY) 276, 257-293. Anguige, K., Tod, K. P. 1999 Isotropic cosmological singularities 2: The Einstein-Vlasov system. Ann. Phys. (NY) 276, 294-320. Anguige, K. 1999 Isotropic cosmological singularities 3: The Cauchy problem for the inhomogeneous conformal Einstein-Vlasov equations. Preprint gr-qc/9903018. Anguige, K. 1999 A class of plane symmetric perfect-fluid cosmologies with a Kasner-like singularity. Preprint gr-qc/9908072. Claudel, C. M., Newman, K. P. 1998 The Cauchy problem for quasi-linear hyperbolic evolution problems with a singularity in the time. Proc. R. Soc. London A 454, 1073-1107. Friedrich, H., Rendall, A. D. 2000 The Cauchy problem for the Einstein equations. In B. G. Schmidt (ed) Einstein’s Field Equations and Their Physical Implications. Lecture Notes in Physics 540. Springer, Berlin. Isenberg, J., Kichenassamy, S. 1999 Asymptotic behaviour in polarized $T^2$-symmetric vacuum spacetimes. J. Math. Phys. 40, 340-352 Kichenassamy, S. 1996 Fuchsian equations in Sobolev spaces and blow-up. J. Diff. Eq. 125, 299-327. Kichenassamy, S. 1996 Nonlinear Wave Equations. Marcel Dekker, New York. Kichenassamy, S., Rendall, A. D. 1998 Analytic description of singularities in Gowdy spacetimes. Class. Quantum Grav. 15, 1339-1355. Moncrief, V. 1980 Global properties of Gowdy spacetimes with $T^2\times \R$ topology. Ann. Phys. (NY) 132, 87-107. Moncrief, V. 1982 Neighbourhoods of Cauchy horizons in cosmological spacetimes with one Killing vector. Ann. Phys. (NY) 141, 83-103. Narita, M., Torii, T and Maeda, K. 2000 Asymptotic singular behaviour of Gowdy spacetimes in string theory. Preprint gr-qc/0003013. Newman, R. P. A. C. 1993 On the structure of conformal singularities in classical general relativity. Proc. R. Soc. London A 443, 473-492. Newman, R. P. A. C. 1993 On the structure of conformal singularities in classical general relativity II: evolution equations and a conjecture of K. P. Tod. Proc. R. Soc. London A 443, 493-515. Taylor, M. E. 1996 Partial Differential Equations III. Nonlinear Equations. Springer, Berlin. [^1]: I thank Aurore Cabet for pointing these out	{ "pile_set_name": "ArXiv" }
--- abstract: 'High redshift DLA systems suggest that the relative abundances of elements might be roughly solar, although with absolute abundances of more than two orders of magnitude below solar. The result comes from observations of the \[SII/ZnII\] ratio, which is a reliable diagnostic of the true abundance, and from DLA absorbers with small dust depletion and negligible HII contamination. In particular, in two DLA systems nitrogen is detected and at remarkably high levels (Vladilo et al. 1995, Molaro et al. 1995, Green et al. 1995, Kulkarni et al. 1996). Here we compare the predictions from chemical evolution models of galaxies of different morphological type with the abundances and abundance ratios derived for such systems. We conclude that solar ratios and relatively high nitrogen abundances can be obtained in the framework of a chemical evolution model assuming short but intense bursts of star formation, which in turn trigger enriched galactic winds, and a primary origin for nitrogen in massive stars. Such a model is the most successful in describing the chemical abundances of dwarf irregular galaxies and in particular of the peculiar galaxy IZw18. Thus, solar ratios at very low absolute abundances, if confirmed, seem to favour dwarf galaxies rather than spirals as the progenitors of at least some of the DLA systems.' author: - 'F. Matteucci , P. Molaro , G. Vladilo' date: 'Received date; accepted date' title: 'Chemical evolution of Damped Ly$\alpha$ systems' --- Introduction ============ Absorption line systems detected in the spectra of Quasi Stellar Objects (QSO) originate in intervening galaxies or protogalaxies. Among the different classes of absorbers, the damped Lyman $\alpha$ systems (DLA) are those characterized by N(HI) $\ge$ 10$^{20}$ cm$^{-2}$ and by showing many low ionization species such as FeII, SiII, CrII, ZnII and OI, which are in their dominant ionization stage for a HI region. The high accuracy in the hydrogen column density determination derived from the damped wings and the absence of significant ionization corrections allow accurate absolute abundance determinations of the gas phase elements. Since DLA systems are observable up to the highest redshift they provide a unique tool for the study of the cosmic chemical history of the early universe. Such observations are complementary to the Hubble Deep Field ones (Mobasher et al. 1996) which show that high redshift galaxies have a large variety of morphological types including many peculiar objects with traces of interaction and merging: interestingly, a large fraction are starburst galaxies, as suggested by their colors. The relative elemental abundances rather than absolute ones are a valuable diagnostic of the first elemental buildup. This is because absolute abundances are affected by all the model assumptions, whereas abundance ratios generally depend only on the assumed nucleosynthesis and stellar lifetimes. The way of using the information contained in abundance ratios is to compare the observed ratios with predictions from chemical evolution models which take into account detailed stellar nucleosynthesis and lifetimes. As in the Milky Way, we hope to distinguish halo-like abundances, produced by type II SNe and characterized by enhancements of $\alpha$ elements with respect to the iron-peak elements, from disk-like abundances, produced by the cumulative effect of both type Ia and type II SNe, where the abundance ratios become progressively solar. The halo abundance pattern follows from the fact that Fe arises mostly from type Ia supernovae with some contribution from type II, while the reverse is true for Si and O. The longer lifetimes of the SNIa, which are believed to have progenitors with masses of 1-8 M$_{\odot}$, result in a delayed iron enrichment compared to the major SNII products such as $\alpha$-elements (Tinsley, 1980; Greggio and Renzini, 1983a; Matteucci and Greggio, 1986). Therefore, the behaviour of the \[$\alpha$/Fe\] ratios is mainly dependent on the relative lifetimes of type Ia and II supernovae. Another key element is nitrogen and the ratio N/O. This element, in fact, is thought to originate mainly from low and intermediate mass stars and to be a “secondary” element, in the sense that it is produced proportionally to the initial stellar metallicity. As a consequence nitrogen is restored into the interstellar medium with a large temporal delay relative to oxygen, which is produced in massive stars and is a “primary” element, namely its production is independent of the initial stellar metallicity. For this reason, the $\alpha$/Fe and N/O ratios can be used as cosmic clocks and represent a clue in understanding the nature of high redshift objects such as QSO (Hamman and Ferland, 1993; Matteucci and Padovani, 1993) and DLA systems (Matteucci, 1995). The DLA systems are generally believed to be the progenitors of the present-day spiral galaxies (Wolfe et al. 1986). However, it has also been suggested that they could be dwarf galaxies, since they have a large amount of gas, low metallicities and low dust-to-gas ratios (Pettini et al. 1990;Meyer and York 1992; Steidel 1994). Whether DLA are proto-spirals or proto-dwarfs can be indicated by their observed abundance pattern. To this purpose in this paper we will present models of chemical evolution for galaxies of different morphological type and compare them with the observed abundances in DLA. Observed abundances in damped Ly$\alpha$ systems ================================================ So far chemical abundances have been measured in a set of DLA systems and with a variety of resolutions and accuracies by Black et al. (1987), Meyer & York (1987), Chaffee et al. (1988), Meyer et al. (1989), Meyer & Roth (1990), Rauch et al. (1990), Pettini et al. (1990), Meyer & York (1992), Pettini et al. (1994), Fan & Tytler (1994), Wolfe et al. (1994), Pettini et al. (1995), Lu et al. (1995a), Steidel et al. (1995), Lu et al (1995b). Here, we would like to focus on the abundances which could be used to compare with chemical evolution models and in particular with the few nitrogen detections so far reported in the literature. A summary of the abundances for the systems for which nitrogen detection or upper limits are available are reported in Table 1 and 2. All the original abundances have been renormalized to the solar abundances of Anders & Grevesse (1989), with the exception of $\log (Fe/H)_{\sun}$=$-4.52$ taken from Hannaford et al. (1992) (\[M/H\] $\equiv$ log(M/H)-log(M/H)$_{\sun}$). The abundances in the Tables are derived from column density ratios of different atomic and ionic species in the gas phase, and can be translated to elemental abundances only if ionization effects and dust depletion are negligible or can be accounted for. In the following we discuss shortly these two effects. QSO $z_{\rm abs}$ N(HI) \[N/H\] \[O/H\] \[Si/H\] \[S/H\] \[Fe/H\] Refs. ----------- --------------- ------- ------------------------------ ------------------------- ------------------------- ------------------------- ------------------------- --------- 0000-26 3.3901 21.30 $-2.77^{+0.17}_{-0.17}$ $-3.13^{+0.17}_{-0.17}$ $-2.48^{+0.19}_{-0.19}$ — $-2.38^{+0.16}_{-0.16}$ 1 1331+1700 1.7765 21.18 $-2.73^{+0.11}_{-0.11}$ $-2.81^{+0.21}_{-0.30}$ — $-1.35^{+0.11}_{-0.11}$ $-2.25^{+0.11}_{-0.12}$ 2$^a$ 2348-147 2.2794 20.57 $<-3.15$ $-2.13^{+2.26}_{-0.58}$ $-1.97^{+0.13}_{-0.10}$ $-1.91^{+0.14}_{-0.16}$ $-2.35^{+0.22}_{-0.10}$ 3 2344+124 2.5379 20.43 $-3.00^{+0.12}_{-0.22}$ $^b$ $-2.11^{+2.5}_{-0.36}$ $-1.72^{+0.27}_{-0.13}$ $<-1.20$ $-1.85^c$ 4 1946+7658 2.8443 20.30 $<-3.26$ $^d$ $-2.65^{+0.15}_{-0.15}$ $-2.16^{+0.10}_{-0.10}$ $<-0.89$ $-2.41^{+0.10}_{-0.10}$ 5$^e$ 0.4 truecm References for Table 1\ (1) Molaro et al. (1995) ; (2) Green et al. (1995) ; (3) Pettini et al.(1995);\ (4) Lipman, thesis (1995) ; (5) Lu et al. (1995a) 0.4 truecm $a$ : Hydrogen column density from Pettini et al. (1994); metal abundance error bars include propagation of the N(HI) error. $b$ : Nitrogen value derived by using only the main absorption component of the system; $c$ : Error bar not reported by the author. $d$ : Upper limit from Lipman (1995; thesis). $e$ : Column densities from line profile fitting where all the lines are fitted simultaneously; the final choice of Lu et al (1995a) for O is \[O/H\]$>$ -2.99; metal abundance error bars include propagation of the N(HI) error. QSO $z_{\rm abs}$ \[N/O\] \[N/Si\] \[N/S\] \[Si/O\] \[S/O\] \[O/Fe\] \[Si/Fe\] --------------- --------------- ------------ ------------- ------------ ------------- ------------ ------------- -------------- 0000-26 3.3901 +0.36 $-$0.29 — +0.65 — $-$0.75 $-0.10$ 1331+1700 1.7765 +0.08 — $-$1.38 — +1.46 $-$0.56 — 2348-147 2.2794 $<-1.02$ $<-1.18$ $<-1.24$ +0.16 +0.22 +0.22 +0.38 2344+124 $^a$ 2.5379 $-0.59$ $-1.14$ $>-1.32$ +0.55 $<+0.73$ $-0.59$ +0.13 1946+7658 2.8443 $<-0.61$ $<-1.10$ — +0.49 $<+1.76$ $-0.24$ +0.25 0.5 truecm $a$ Values derived by using only the main absorption component Ionization corrections and dust depletions ------------------------------------------ The neutral hydrogen in the DLA is optically thick to the ionizing radiation either from the intergalactic background or from starlight inside the DLA, and the abundances derived from species that have ionization potentials in excess of 13.6 eV do not require ionization corrections. This is the case of abundances derived from NI, OI, AlII, SiII and FeII, which are dominant ionization stages in the HI gas in our own Galaxy. Detailed ionization models for intergalactig radiation field at high redshift explored by Lu et al (1995) and Fan and Tytler (1994) show that ionization corrections are indeed minimal for systems with $\log$ N(HI)$>$ 20. Another effect which could affect the abundance determinations is the presence of HII regions within the DLA intercepted by the line of sight (Steigman et al 1975). These regions of warm ($T$ $\simeq$ 10$^4$ K) ionized gas would contribute to the column densities of singly ionized species, such as FeII, SiII and AlII, but not to the HI column density. In the interstellar gas of our Galaxy the column density contribution of HII regions is generally negligible when the HI column density is as high as it is in DLA systems (N(HI) $\ga$ 10$^{20}$ cm$^{-2}$). If HII regions were more important in DLA than in our Galaxy they would produce different line profiles between neutral and ionized lines, when observed at high resolution. This effect is not observed in the DLA towards QSO 0000-2169 where the $b$ values, velocities and profiles of singly ionized species are the same as those of the neutral species, suggesting that both form in the same slab of HI material. Extra contributions to the absorption from HII regions in this DLA are also constrained by the lack of detection of NII, resulting in NII/NI $<$ 0.4. Viceversa, Green et al. (1995) interpret the abundances of the DLA towards MC3 1331+170 invoking an extraordinary amount of ionized gas, namely six times the neutral gas, which makes rather uncertain the abundances of ionized species. Highly ionized gas in the form of SiIV and CIV is frequently observed in the DLA. Sometimes the highly ionized gas is found at the same velocities of the neutral one but often is found at different velocities and therefore is very probable that highly ionized gas arises either in regions disconnected from the neutral material, or in the interfaces between neutral material and the intergalactic medium. Presence of dust will selectively deplete the elements observed in the gas phase. In the dense clouds in the Galaxy the interstellar depletion of Al and Fe can reach $\simeq$ 2 dex, that of Si $\approx$ 1 dex, while, on the other extreme the O depletion is $\le$ 0.4 dex, that of N $\le$ 0.2 dex and S essentially undepleted (Jenkins 1987). Reddening measurements of QSO with DLA systems suggest a dust-to-gas ratio in the DLA of about 10% of that in our Galaxy (Fall and Pei 1989). In the survey of ZnII and CrII in DLA by Pettini et al. (1994) chromium is typically one dex below zinc, while in the Galaxy is about two dex, showing the presence of dust and a reduced dust depletion in the DLA systems. If some dust were present this would alter the abundance determinations of refractory elements, such as Fe, Si and Al. Much more reliable are the abundances of the non-refractory elements, such as Zn, S, O and N. $\alpha$ versus iron-peak elements ---------------------------------- Of particular importance for understanding galactic chemical evolution is the comparison between $\alpha$ and iron peak elements. In particular, the ratio between sulphur and zinc is an important diagnostic tool. Both elements show little affinity with dust and their ratio is also safe against possible contributions from HII regions, since these essentially cancel out. Considering that S and Zn have different nucleosynthetic origin with S mainly a product of type II SN and Zn of type Ia SN (Matteucci et al. 1993), the \[SII/ZnII\] ratio is an ideal diagnostic for understanding the character of the chemical evolution. So far, only few determinations of \[S/Zn\] ratio in DLA are present in the literature. Meyer et al. (1989) found \[S/Zn\]=$-$0.1 in the DLA at $z=2.8$ towards QSO PKS 0528-250 and Green et al. (1995) found \[S/Zn\] = +0.1 in the $z$=1.775 DLA in Q1331. Thus, in spite of the very poor metallicity of the gas, which in these systems is $\simeq$ -2.0, the ratios of elements non depleted in dust have solar values. In DLA systems the relative abundances of Si and Fe are often found to be consistent with a halo pattern. Lu et al. (1995a) found \[Si/Fe\]=+0.31 for the DLA at $z=2.844$ towards HS 1946=76 and Pettini et al. found \[Si/Fe\]=+0.4 for the DLA at $z=2.27$ towards QSO 2348-147. However, Si and Fe are differentially depleted from gas to dust in our Galaxy. The average value in the compilation of 11 clouds observed with HST made by Lu et al (1995a, their Table 8) is \[Si/Fe\]=+0.66$\pm0.26$. The observed enhancement of Si versus Fe in the DLA would reflect the overabundances of $\alpha$ elements with respect to the iron-peak elements only in complete absence of dust. Since DLA show some amount of dust, a moderate enhancement of Si over Fe cannot be considered as a clear-cut evidence of a halo-like pattern. Moreover, values of \[Si/Fe\]$\ge$0.6, that would be expected if differential depletion and intrinsic $\alpha$ enhancement are present, have not yet been observed. Nitrogen and oxygen abundances ------------------------------ Nitrogen has been detected in the DLA at z=3.39 towards QSO 0000-26, with \[N/H\] = $-$2.77 $\pm 0.17$ (Vladilo et al. 1995 and Molaro et al. 1995) and in the DLA at $z$=1.78 towards MC3 1331+170 with \[N/H\]=-2.7 (Green et al. 1995). A significant upper limit has been derived by Pettini et al. (1995) in the DLA systems towards QSO 2348-147 NI$<$-3.15. By means of co-addition technique of the lines of the 1200 Å triplet Lipman (1995) achieved a marginal detection of NI for the main component of the DLA towards QSO 2344+124. By using the data published by Fan and Tytler (1994), Lipman derived also an upper limit of \[N/H\] $<$ -3.26. These measurements show that a real dispersion in the nitrogen abundances may be present among the DLAs. In order to understand the nucleosynthetic origin of nitrogen the ideal would be to follow its abundance with respect to that of oxygen. Unfortunately, oxygen abundance is generally given with a large uncertainty. This is because oxygen abundance is derived from the OI 1302.1685 Å line which has log gf=1.804 and is generally saturated. The other line OI 1355.5977, has log gf=-2.772 and is generally too faint to place useful upper limits. The line saturation leaves the line broadening poorly constrained and leads to the large uncertainty in the oxygen abundance. However, the $b$ value can be inferred from other lines under the assumption that they are formed in the same material and that the origin of broadening is not thermal. Lu et al. (1995a) and Molaro et al. (1995) use the $\it b$ value from the other observed species thus removing the large uncertainty in the oxygen abundance (cfr. Table 1). In Molaro et al. (1995). the broadening value is taken not only from ionized species, which might be sensitive to extra contribution from HII gas along the line of sight, but also from NI which is forming in the same material as neutral oxygen. It is rather striking that in the cases with oxygen abundances with small associated errors, oxygen turns out to be remarkably deficient relatively to the other elements measured, such as Si, S and Fe as it is possible to see in Table 2. To circumvent the problem of oxygen uncertainty, Pettini et al. (1995) and Lipman (1995) take Si or S as a proxy for oxygen assuming \[O/Si\] = \[O/S\]=0. This assumption is about true for the halo stars of our own Galaxy (however see later for silicon), but is rather risky for primeval galaxies which might have experienced a different chemical evolution from that of our own Galaxy. In particular, Si and O do not have the same nucleosynthetic origin, silicon may be affected by dust depletion and SII and SiII may be also affected by HII contribution. Also by adopting the Pettini et al. approach for the case of QSO 0000-2169 we have \[N/Si\] = $-0.3$, which stands genuinely high when compared to the other determinations. What chemical evolution? ======================== Some DLA seem to show similarities to dwarf irregular and blue compact galaxies in terms of abundance pattern. These galaxies, in fact, exhibit relatively high N/O, although with a large spread, at a low overall metallicity Z ranging from 1/10 to 1/30 of the solar value (see Matteucci, 1995). In particular, IZw18 is the galaxy with the lowest metal content known locally, and a N/O ratio roughly solar. Recently, a single starburst chemical evolution model has been successfully developed by Kunth et al. (1995) to reproduce the observed abundances of IZw 18. Kunth et al. (1995) applied Marconi’s et al. (1994) model where star formation is assumed to proceed in short but intense bursts. The contribution to the chemical enrichment of SNe of different type (II, Ia and Ib) is taken into account. The novel feature is the introduction of differential galactic winds where some elements are preferentially lost from the galaxy relative to others. Recently, more and more observational evidence is growing about the existence of such galactic winds in dwarf irregulars as provided by Meurer et al. (1992), Lequeux et al. (1995) and Papaderos et al. (1994). These winds are found to be a crucial ingredient in the model of Kunth et al. (1995) in order to reproduce the high observed N/O ratio in IZw18. To explain the nitrogen abundance of IZw18 this same model assumes also a primary N production from massive stars, as described in the next section. In this paper we will show the predictions of models similar to that of Kunth et al. together with the predictions of models for the chemical evolution of the solar region, under different assumptions about the production and nature of nitrogen, and we will compare them with the DLA data. The chemical evolution model for dwarf irregulars ================================================= We are using here the same model adopted by Kunth et al. (1995) which is aimed at describing the evolution of IZw18, namely of an object presently having a strong burst of star formation which induces a galactic wind. The main features of this chemical evolution model are the following: 1\) one-zone, with instantaneous and complete mixing of gas inside this zone, 2\) no instantaneous recycling approximation; i.e. the stellar lifetimes are taken into account, 3\) only one intense burst of star formation is assumed to occur, 4\) the evolution of several chemical elements (He, C, N, O, Fe) due to stellar nucleosynthesis, stellar mass ejection, galactic wind powered by SNe and infall of primordial gas, is followed in detail. If G$_i$ is the fractional mass of the element i, its evolution is given by the equations described in Marconi et al. (1994). $$\dot G_i(t) = -\psi(t)(1+w_{i}) X_i(t)+$$ $$\int_{M_{L}}^{M_{Bm}}{\psi(t-\tau_m) Q_{mi}(t-\tau_m)\phi(m)dm}+$$ $$A\int_{M_{Bm}}^{M_{BM}}{\phi(m)}\bigl[\int_{\mu_{min}} ^{0.5}{f(\mu)\psi(t-\tau_{m1}) Q_{mi}(t-\tau_{m2})d\mu\bigr] dm}+$$ $$(1-A)\int_{M_{Bm}}^ {M_{BM}}{\psi(t-\tau_{m})Q_{mi}(t-\tau_m)\phi(m)dm}+$$ $$\int_{M_{BM}}^{M_U}{\psi(t-\tau_m)Q_{mi}(t-\tau_m)+ \phi(m)dm} + \dot G_{i}(t)_{inf} \eqno(1)$$ where $G_i(t)$=$M_{gas}(t)X_i(t)/M_{tot}(t_{G})$ is the mass density of gas in the form of an element [i]{} normalized to the total mass at the present time $t_{G}$. The quantity $X_i(t) = G_i(t)/G(t)$ represents the abundance by mass of the element $i$ and by definition the summation over all the abundances of the elements present in the gas mixture is equal to unity. The quantity $G(t) = M_{gas}/M_{tot}$ is the total fractionary mass of gas, and $M_{tot}$ refers only to the mass present in the form of gas in the star forming region. The possible presence of a dark matter halo is not considered here, given the simple treatment of the development of a galactic wind, as we will see in the following. The star formation rate we assume during the burst, $\psi(t)$, is defined as: $$\psi(t)\,=-\nu \, \eta(t) \, G(t)$$ where $\nu$ is the star formation efficiency (expressed in units of Gyr$^{-1}$), and represents the inverse of the timescale of star formation, namely the timescale necessary to consume all the gas in the star forming region; $\eta(t)$ takes into account the stochastic nature of the star formation processes as in Gerola et al. (1980) and Matteucci and Tosi (1985), where a detailed description can be found. The galactic wind is assumed to be simply proportional to the star formation rate (namely to the rate of explosion of type II SNe). This is a reasonable choice since the duration of the burst is so short that it does not allow the explosion of type Ia SNe, which occur all after the burst. The simplicity of the treatment of the galactic wind avoids also the introduction of other unknown parameters such as the efficiency of energy tranfer from stars and supernovae to the ISM. The rate of mass loss via a galactic wind is defined as follows: $$\dot G_{iw}(t)\,=\, -w_{i}\psi(t)X_{iw}(t)$$ where $X_{iw}(t)=X_{i}(t)$ and $w_{i}$ is a free parameter containing all the information about the energy released by SNe and the efficiency with which such energy is transformed into gas escape velocity (note that Pilyugin (1992;1993) defines a wind parameter which is the inverse of $w_{i}$, namely the ratio of the star formation rate to the wind rate). The value of w$_{i}$ has been assumed to be different for different elements. In particular, the assumption has been made that only the elements produced by type II SNe (mostly $\alpha$-elements and some iron) can escape the star-forming regions. We have made this choice following the conclusions of Marconi et al. (1994) and Kunth et al. (1995), who showed that models with differential wind can explain better the observational constraints of blue compact galaxies in general and IZw18 in particular. The justification for the existence of differential galactic winds can be found in the fact that during short starbursts type II SNe dominate. Since SNII explode in association, they are likely to produce chimneys which will eject metal enriched material (De Young and Gallagher, 1990). On the other hand, type Ia SNe are not likely to trigger a wind since they explode mostly during the interburst phase and have a large range of explosion times (from $3 \cdot 10^{7}$ to a Hubble time) inducing them to explode in isolation. The terms on the right side of equation (1) represent, respectively, the rate at which the gas is lost via astration and galactic wind and the rates at which the matter is restored to the interstellar medium (ISM) by: \(a) single stars with masses between $M_L=1.0M_{\odot}$, which is the lowest mass contributing to the galactic enrichment, and ${M_{B_m}}=3.0M_{\odot}$, which represents the minimum mass for which a binary system (with the minimum mass ratio $\mu_{min}$ defined as in Greggio and Renzini 1983b) produces a type Ia SN, \(b) binary systems producing type Ia SNe, within the range of masses ${M_{B_m}}$=${3 M_{\odot}}$ and ${M_{B_M}}$=${16 M_{\odot}}$ \[the assumed progenitors of type Ia SNe are binary systems of C-O white dwarfs(Whelan and Iben, 1973); the parameter $A$, defined in Greggio and Renzini (1983b), represents the fraction of binary systems in the IMF which can give rise to type Ia SNe; \(c) single stars in the mass range ${M_{B_m}}$ - ${M_{B_M}}$ which end their lives either like white dwarfs or type II SNe, \(d) single stars in the mass range ${M_{B_M}}$ and ${M_U}$, where ${M_U}$ is the maximum stellar mass contributing to the galactic enrichment (100 ${M_{\odot}}$ in our models). These stars can either end their lives as type II or type Ib SNe. The initial mass function (IMF) by mass, $\phi(m)$, is expressed as a power law with an exponent x=1.35 over the mass range $0.1-100M_{\odot}$. The chemical evolution equations include also an accretion term: $$\dot G_{iinf}(t)\,=\,{ {(X_{i})_{inf}e^{-t/\tau}} \over {\tau(1-e^{-t_{G}/\tau})} }$$ where (X$_{i})_{inf}$ is the abundance of the element [i]{} in the infalling gas, assumed to be primordial, $\tau$ is the time scale of mass accretion and $t_{G}$ is the galactic lifetime. This accretion term may simulate the formation of dwarfs as the result of mergers of smaller subunits. The parameter $\tau$ has been assumed to be the same for all blue compact galaxies and short enough to avoid unlikely high infall rates at the present time ($\tau=0.5\cdot 10^{9}$ years). It is worth noting that a shorter timescale would not produce a noticeable effect on the results. Nucleosynthesis prescription ---------------------------- The term ${{Q_m}_i(t-{\tau_m})}$ in equation (1), the so called [production matrix]{} (Talbot and Arnett, 1973), represents the fraction of mass ejected by a star of mass $m$ in the form of the element $i$. The quantity $\tau_{m}$ represents the lifetime of a star of mass m and $\tau_{m2}$ refers to the mass of the secondary component and $\tau_{m1}$ to the mass of the primary component of a binary system giving rise to a type Ia SN (see Matteucci and Greggio, 1986 for details). For the nucleosynthesis prescriptions we have assumed the following: a\) For low and intermediate mass stars (0.8 $\leq$ M/M$_{\odot} \leq$ M$_{up}$) we have used Renzini and Voli’s (1981) nucleosynthesis calculations for a value of the mass loss parameter $\eta$ = 0.33 (Reimers 1975), and mixing length $\alpha_{RV}$ = 1.5. The standard value for M$_{up}$ is 8 M$_{\odot}$. b\) For massive stars (M $>$ 8 M$_{\odot}$) we have used Woosley’s (1987) nucleosynthesis computations but adopting the relationship between the initial mass $M$ and the He-core mass M$_{He}$, from Maeder and Meynet (1989). It is worth noting that the adopted M(M$_{He}$) relationship does not substantially differ from the original relationship given by Arnett (1978) and from the new one by Maeder (1992) based on models with overshooting and Z=0.001. These new models show instead a very different behaviour of M(M$_{He}$) for stars more massive than 25 M$_{\odot}$ and Z=0.02, but the galaxies we are modelling never reach such a high metallicity. c\) For the explosive nucleosynthesis products, we have adopted the prescriptions by Nomoto et al. (1984), model W7, for type Ia SNe, which we assume to originate from C-O white dwarfs in binary systems (see Marconi et al. (1994) for details). More recent nucleosynthesis calculation by Thielemann et al. (1993) do not significantly differ from the Nomoto et al. (1984) ones. The nucleosynthesis of nitrogen and the N/O ratio ------------------------------------------------- Nitrogen is a key element to understand the evolution of galaxies with few star forming events since it needs relatively long timescales as well as relatively high underlying metallicity to be produced. The reason is that N is believed to be mostly a product of secondary nucleosynthesis, being produced by CNO processing of $^{12}$C or $^{16}$O from earlier generations of stars. However, a primary component can be obtained when the seed nuclei of $^{12}$C or $^{16}$O are produced in earlier helium burning stages of the same star. Generally, N is believed to be secondary in massive stars, and mostly secondary and probably partly primary in low and intermediate mass stars. However, some doubts exist at the moment on the amount of primary nitrogen which can be produced in intermediate mass stars due to the uncertainties related to the occurrence of the third dredge-up in asymptotic giant branch stars (AGB). In fact, if Blöcker and Schoenberner (1991) calculations are correct, the third dredge-up in massive AGB stars should not occur and therefore the amount of primary N produced in AGB stars should be strongly reduced (Renzini, private communication). The only possible way to produce a reasonable quantity of N during a short burst (no longer than 20 Myr), as discussed in Kunth et al. (1995), is to require that massive stars produce a substantial amount of primary nitrogen. This claim was already made by Matteucci (1986) in order to explain the \[N/O\] abundances in the solar neighbourhood. Recently, calculations by Woosley and Weaver (private communication) seem to support the possibility that N is produced by C and O synthesized inside massive metal poor stars as a primary element. Therefore, as in Kunth et al. (1995) we have taken this possibility into account in the present calculation. Theoretical prescriptions for the evolution of the solar vicinity ================================================================= The model adopted for the evolution of the solar neighbourhood is the same as in Matteucci and François (1992) and the basic equations are similar to Eq (1), the only difference being the absence of a galactic wind, namely all the $w_{i}=0$. The main differences between the model for dwarf irregulars and the model for the solar vicinity is that in the latter case the star formation rate is continuous and the IMF for massive stars is steeper than a Salpeter IMF (i.e. Scalo 1986). The time scale for the formation of the disk in the solar region is also different from the timescale assumed for assembling the dwarf irregulars (i.e. $\tau=0.5 $Gyr), namely is $\tau=3$ Gyr. This choice ensures that the majority of the observational constraints in the solar neighbourhood are reproduced. The nucleosynthesis prescriptions adopted for the solar neighbourhood are exactly the same as for the dwarfs. Results ======= Besides Model 1 and 5 of Kunth et al. (1995) (model 1 and 3 respectively, in Table 3), we run several models by varying star formation efficiency and/or wind efficiency, all the other parameters being left the same as in Model 5 of Kunth et al. The model parameters are presented in Table 3. In column 1 are the model numbers and in column 2 the nucleosynthesis prescriptions. In particular, “STANDARD” refers to the prescriptions described in section III, whereas “N PRIMARY” refers to the assumption of primary production of N in massive stars, leaving all the rest unchanged. \[\] ----- ------------- ------ -------------- --------- ---------- MOD yields wind $\nu$ $N_{B}$ duration ($Gyr^{-1}$) (Myrs) 1 STANDARD 80 10 1 50 2 N PRIMARY 700 50 1 20 3 N PRIMARY 80 50 1 20 4 N PRIMARY 700 1000 1 20 5 N PRIMARY 1000 50 1 20 6 N PRIMARY 80 50 4 20 7 NPRIMARY 700 50 4 20 ----- ------------- ------ -------------- --------- ---------- : Model parameters : IMF slope: 1.35 (Salpeter); upper mass limit:\ 100 M$_\odot$; lower mass limit: 0.1 M$_\odot$ In column 3 we show the wind parameter [$w_{i}$]{} as defined in the previous section. This parameter is different from zero only for the elements produced and dispersed during the explosion of SNe II. The value for this parameter given in Table 3 refers only to the $\alpha$-elements, which are the main outcome of SN II explosions. In particular, for the elements studied here $w_{i}$ is zero for H, He and N, whereas is different from zero, but smaller than for $\alpha$-elements, for C and Fe. The parameter $w_{i}$ for these elements is chosen in such a way to account for the fact that Fe and C are produced both in massive and in intermediate mass stars (type Ia SNe and AGB stars, respectively). In column 4 is shown the star formation efficiency $\nu$ (in units of Gyr$^{-1}$), as defined in the previous section. In column 5 is shown the number of bursts and in column 6 the duration of each burst in Myr. Such a duration is constrained by results of population synthesis models (Mas-Hesse and Kunth, 1991) suggesting a maximum duration for the present burst in IZw18 of 20 Myr. The starting time of the burst is not important for models 1-5 since we assume that it is the only event of star formation and it could happen at any time (i.e. at any redshift). It is worth noting that the assumed range of variation of $\nu$ and $w_{i}$ is quite large but reasonable. In order to reproduce the star formation rate of IZw18 a value of $\nu$=50 Gyr$^{-1}$ (note that for the solar neighbourhood $\nu$=0.5 Gyr$^{-1}$) is preferred since it predicts a star formation rate of $\simeq 0.03 M_{\odot}$yr$^{-1}$, in very good agreement with observational estimates (see Kunth et al., 1995). An efficiency of star formation of 1000 Gyr$^{-1}$ is also plausible for starburst galaxies since it predicts a star formation rate of $\simeq 0.1M_{\odot}$yr$^{-1}$. Such star formation rates are quite reasonable for objects suffering only few bursts of star formation (may be only one) and having a large amount of gas. The wind parameter also spans a quite large range of values and this is possible under the assumption of enriched galactic wind, when only metals are lost. On the other hand, in the case of normal wind the value of $w_{i}$ is constrained by the condition of not destroying the galaxy. However, it is worth noting that even a totally disruptive wind, could in principle be considered. The initial mass of gas involved in the burst is assumed to be $6 \cdot 10^{6}M_{\odot}$. \[\] ---------------- --------------- ------------- ---------------- QSO $z_{\rm abs}$ 12+log(O/H) log(N/O) 0000-26 3.3901 5.80 $-0.52$ 1331+1700 1.7765 6.12 $-0.80$ 2348-147 $^a$ 2.2794 6.80 $< -2.20$ $^a$ 2344+124 $^a$ 2.5379 6.82 $-2.0$ $^a$ 1946+7658 $^a$ 2.8443 6.28 $< -2.0$ $^a$ ---------------- --------------- ------------- ---------------- : Measured (O/H) and (N/O) column density ratios in DLA systems to be compared with the figures. $(a)$ Values taken from Lipman (1995). The N/O versus O/H distribution is shown in Fig. 1 where the models with only one burst together with the models for the solar neighbourhood are shown. The dot-dashed curve is from model N. 5 from Kunth et al. (1995) (model 3 of table 3) and it can reproduce the observations of IZw 18. Incidentally, we note that this model can fit very well the N/O upper limit by Pettini et al. (1995) but for a burst age shorter than 20 Myr. On the other hand, Pettini’s et al.’s point, as well as the other points from Lipman (1995), could be marginally reproduced also by the evolution of a disk galaxy in the earliest phases of its evolution; in particular during the halo phase, as it is shown in the figure. In fact, the continuous curves in Fig. 1 represent the predictions from the model of chemical evolution of the solar neighbourhood under different assumptions about the nucleosynthesis of N. In particular, the area delimited by those models goes from purely secondary N in stars of all masses (the straight line at the right end) to primary N in massive stars and secondary and primary N in low and intermediate mass stars. On the other hand, the N/O ratios observed by Molaro et al.(1995) and Green et al.(1995) are reproduced by models with a strong starburst and very strong galactic wind (models 4 and 5 in Table 2), so that models for the solar neighbourhood seem to be completely ruled out in explaining these two DLA systems. Thus we show that it is possible to have large N/O ratios even at low O abundances if the nitrogen produced by massive stars, restored on relatively short time scales, is primary and with a strong differential effect in the galactic wind. As shown in Fig. 2 the same model produces solar ratios of alpha-elements (such as S, O and Si) to iron peak elements (such as Fe and Zn), as it is observed at least in few DLA. In particular, in Fig. 2 we show the predictions of the models for O and Si relative to iron to be compared with the data of Table 4. The x-axis does not extend to abundances larger than \[Fe/H\]=-2.0 since in one-burst models the metallicity does not increase any further. Sulphur is not shown here but it should closely follow oxygen. On the other hand, the predicted \[Si/Fe\] is lower than \[O/Fe\] and the reason for this resides in the fact that more Si than O is produced in type Ia SNe (Nomoto et al. 1984; Thielemann et al. 1993) and that we assume the same $w_{i}$ parameter for Si and O. Studies of abundances in halo stars do not allow us to discriminate clearly on this point (François, 1986; Ryan et al. 1991; Primas et al. 1994; McWilliam et al. 1995). Therefore, it does not seem safe to assume \[Si/O\]=0 in order to derive oxygen abundances in DLA systems (see Lipman, 1995). However, as indicated in Figure 2 the observed \[O/Fe\] is lower than \[Si/Fe\] for the DLA observed by Molaro et al. (1995). A possible explanation for this is that, in the framework of the differential galactic wind, O, which originates mostly in type II SNe, should be lost from the galaxy in a larger percentage than silicon, which originates also from type Ia SNe. In other words, Si should be treated as we do with C and Fe. However, numerical experiments show that even in this case is difficult to invert the situation and have \[Si/Fe\] higher than \[O/Fe\], as indicated by the data taken at face values! In Figure 3 we show the predicted N/O vs. O/H from models with 4 short bursts (20 Myr each) all occurring inside the first Gyr from the start of star formation, so that they can be representative of high redshift objects. In particular, we show the predictions of model 6 and 7 which are similar to model 3 and 2, respectively, but with four bursts of star formation.These models show that in principle the DLA systems from Green et al. (1995) and from Molaro et al. (1995) could be explained by a galaxy like IZw18 experiencing more than one burst of star formation and observed during the interburst period, without invoking an extremely large wind parameter. In fact, during this phase the N/O ratio increases as due to the fact that oxygen is no more produced while N continues to be restored from low and intermediate mass stars. This is a well known effect as shown by Pilyugin (1992;1993). In Figs. 4 and 5 we show the \[O,Si/Fe\] ratios predicted by model 6 and 7, respectively. Here too we can see the oscillating behaviour due to the alternating burst and quiescent phases. In particular, in both models the \[O,Si/Fe\] ratios decrease during the interburst phase and increase again during the bursts. The difference between the two models resides in the efficiency of the wind in model 7, which reflects in a stronger variation of the abundance ratios between the burst and the interburst phases. The reason for this variation in model 7 is that the stronger wind acting during bursts is responsible for very low absolute abundances at the end of each burst, so the increase of Fe during the interburst is stronger relatively to model 6 where the absolute abundances at the end of each burst are higher. Finally, it is worth noting that current models for elliptical galaxies (see Matteucci and Padovani, 1993) would never reproduce such the high N/O ratio observed by Molaro et al. (1995) for such low metallicities. In Figure 1, for example, the predictions of models for an elliptical galaxy of initial luminous mass $10^{11} M_{\odot}$ would lie at the right side of the solar neighbourhood curve for only secondary nitrogen, and even in the case of primary N from massive stars would be completely outside of the metallicity region where the DLA systems are observed. Conclusions =========== The abundance pattern of non refractory elements in some primeval DLA systems at high redshift seems not to follow the pattern observed in the halo of the Milky Way. This suggests that these objects had a different chemical evolution than the Milky Way and the spirals in general, at variance with the general belief that DLA are the progenitors of present day spirals. We have shown that the most promising models to explain the observed abundances are those succesfully applied to the dwarf irregular galaxies such as IZw18 (Marconi et al. 1994, Kunth et al. 1995). The conclusion about these DLA systems being dwarf galaxies has also been independently suggested by Meyer and York (1992) and by Steidel et al. (1994) from the low abundances found in the few DLA observed at low redshifts. In summary, our conclusions are: - In order to explain the high N/O ratios observed in two DLA systems by Green et al .(1995) and Molaro et al.(1995) one has to assume that these systems are dwarf irregular galaxies experiencing their first or one of their first bursts of star formation. These galaxies should also experience strong enriched galactic winds carrying away mostly the products of SN II explosions such as oxygen and other $\alpha$-elements. In particular, the high N/O abundance ratios could represent either the situation of an interburst phase where N increases and O does not, or the situation of a burst triggering an extremely strong and enriched galactic wind. -Nitrogen in massive stars should have a primary origin as already suggested by Matteucci (1986). This seems to be possible but is strongly dependent on the assumed treatment of convection in stellar interiors. The nucleosynthesis of N in stars of all masses, especially in low and intermediate mass stars, which are the main producers of this element, needs revision and a homogeneous set of calculations for stars of all masses is required. New yields, but only for stars below 4 $M_{\sun}$, have been recently computed by Marigo et al. (1996). Unfortunately, the mass range above that mass limit is very important for nitrogen production and therefore firm conclusions cannot yet be drawn. -We can exclude, on the basis of current models for elliptical galaxies that any of the systems discussed in this paper could be a proto-elliptical. - Some of the differences in the abundances observed among different DLA systems could be due to the fact that some of them are proto-spirals (see the systems observed by Pettini et al. 1995 and Lipman, 1995) and some are proto-dwarfs (see the systems observed by Molaro et al. 1995 and Green et al. 1995) - Abundance ratios between elements produced from stars at different rates such as N/O and $\alpha$/Fe represent a very useful tool either to date a galaxy or to understand the nature of high red-shift objects. Anders E., Grevesse N., 1989, Geochim. Cosmochim. Acta 53, 197 Arnett D.W., 1978, ApJ 219, 1008 Black J.H., Chaffee F.H., Foltz C.B., 1987, ApJ 317, 442 Blöcker T., Schoenberner D., 1991, A&A 244, L43 Chaffee F.H., Black J.H., Foltz C.B., 1988, ApJ 335, 584 De Young D.S., Gallagher J.S. III, 1990, ApJ 356, L15 Dufour R.J., Garnett D.R., Shields G.A., 1988, ApJ 332, 752 Fall S.M., Pei Y.C., 1989, ApJ 337, 7 Fan X.-M., Tytler D., 1994, ApJ Suppl. 94, 17 François P., 1986, A&A 160, 264 Gerola H., Seiden P.E., Schulman L.S., 1980, Ap.J., 242, 517 Green R.F., York D., Huang K., Bechtold J., Welty D., Carlson M., Khare P., Kulkarni V., 1995, Proc. [ESO Workshop on QSO Absorption Lines]{}, ed. G. Meylan, Springer Verlag, 85 Greggio L., Renzini A., 1983a, in [The first stellar generations]{}, Mem. S.A.It. Vol. 54, 311 Greggio L., Renzini A., 1983b, A&A 118, 217 Hamman F., Ferland G., 1993, ApJ, 418, 11 Hannaford P., Lowe R.M., Grevesse N., Noels A., 1992, A&A 259, 301 Kulkarni V.,P., Huang K., Green R.F., Bechtold J., Welty D., York D.G., 1996, MNRAS 279, 197 Kunth D., Matteucci F., Marconi G., 1995, A&A 297, 634 Lequeux J., Kunth D., Mas-Hesse J.M., Sargent W.L.W., 1995, A&A 301, 18 Lipman K., 1995, PhD thesis, Cambridge University Lu L., Savage B.D., Tripp T.M., Meyer D.M. 1995a, ApJ 447, 597 Lu L., Sargent, W.L.W., Womble, D.S.,Barlow, T.A. 1995b, ApJ in press Maeder A., Meynet G., 1989, A&A, 210, 155 Maeder A., 1992, A&A, 264, 105 Marconi G., Matteucci F., Tosi M., 1994, MNRAS 270, 35 Marigo P., Bressan A., Chiosi C., 1996, preprint Mas-Hesse J.M., Kunth D., 1991, A&A Suppl. Ser. 88, 399 Matteucci F., 1995, in [The interplay between massive star formation, ISM and galaxy evolution]{}, Eds. Kunth et al., Edition Frontiers, in press Matteucci F., 1986, MNRAS, 221, 911 Matteucci F., François, P. 1992, A&A, 262, L1 Matteucci F., Greggio L., 1986, A&A, 154, 279 Matteucci F., Padovani P., 1993, Ap. J., 419, 485 Matteucci F., Tosi M., 1985, MNRAS., 217, 391 Matteucci F., Raiteri C.M., Busso M., Gallino R., Gratton R., 1993, A&A, 272, 421 McWilliam A., Preston G., Sneden C., Searle L., 1995, Astron J. 109, 2757 Meurer G.R., Freeman K.C., Dopita M.A., 1992, Astron. J. 103, 60 Meyer D.M., Roth K.C., 1990, ApJ 363, 57 Meyer D.M., Welty D.E., York D.G., 1989, ApJ 343 L37 Meyer D.M., York D.G., 1987, ApJ 319, L45 Meyer D.M., York D.G., 1992, ApJ 399, L121 Mobasher, B., Rowan-Robinson, M., Georgakakis, A., Eaton, N., 1996, MNRAS, in press Molaro P., D’ Odorico S., Fontana A., Savaglio S., Vladilo G., 1995, A&A in press (ESO sc. prepr. No. 1102) Nomoto K., Thielemann F.K., Yokoi, K., 1984, Ap.J., 286, 644 Papaderos P., Fricke K., Thuan T.X., Loose H., 1994, A&A, 291, L13 Pettini M., Boksenberg A., Hunstead R.W., 1990, ApJ, 348, 48 Pettini M., Lipman K., Hunstead R.W., 1995, ApJ 451, 100 Pettini M., Smith L.J., Hunstead R.W., King D.L., 1994, ApJ 426, 79 Pilyugin, L.S., 1992, A&A, 260, 58 Pilyugin, L.S., 1993, A&A, 277, 42 Primas, F., Molaro, P., Castelli, F., 1994, A&A, 290, 885 Rauch M., Carswell R.F., Robertson J.G., Shaver P.A., Webb J.K., 1990, MNRAS 242, 698 Reimers D., 1975, Mém. R. Sci. Liège 6ème Sér., 8, 369 Renzini A., Voli M., 1981, A&A, 94, 175 Ryan S.G., Norris J.E., Bessel M.S., 1991, Astron. J. 102, 303 Scalo J.M., 1986, Fund. Cosmic Phys. 11, 1 Skillman E.D., Kennicutt R.C., 1993, ApJ 411, 655 Steidel C.C., Bowen D.V., Blades J.C., Dickinson M., 1995, ApJ 440, L45 Steigman, G., Strittmatter, P.A., Williams, R.E. 1975, ApJ, 198, 575 Talbot R.J., Arnett D.W., 1973, ApJ 197, 551 Tinsley B.M., 1980, Fund. Cosmic Phys. 5, 287 Thielemann F.K., Nomoto K., Ashimoto M., 1993, in [Origin and evolution of the elements]{}, Eds. N. Prantzos et al., Cambridge Univ. Press, 297 Vladilo G., D’Odorico S., Molaro P., Savaglio S., 1995, Proc. [ESO Workshop on QSO absorption lines]{}, ed. G. Meylan, Springer Verlag, 103 Whelan J., Iben I. Jr., 1973, ApJ 186, 1007 Wolfe A.M., Turnshek D.A., Smith H.E., Cohen R. D., 1986, ApJS 61, 249 Woosley S.E., 1987, in “Nucleosynthesis and Chemical Evolution” 16th Saas-Fee Advanced Course, Geneva Observatory, p.1	{ "pile_set_name": "ArXiv" }
--- abstract: 'In this paper, we introduce the Variational Autoencoder (VAE) to an end-to-end speech synthesis model, to learn the latent representation of speaking styles in an unsupervised manner. The style representation learned through VAE shows good properties such as disentangling, scaling, and combination, which makes it easy for style control. Style transfer can be achieved in this framework by first inferring style representation through the recognition network of VAE, then feeding it into TTS network to guide the style in synthesizing speech. To avoid Kullback-Leibler (KL) divergence collapse in training, several techniques are adopted. Finally, the proposed model shows good performance of style control and outperforms Global Style Token (GST) model in ABX preference tests on style transfer.' address: \| $^1$National Engineering Laboratory for Speech and Language Information Processing,\ University of Science and Technology of China, Hefei, P.R.China\ $^2$ Microsoft China\ [[email protected], {peterpan, helei}@microsoft.com, [email protected]]{} bibliography: - 'strings.bib' - 'refs.bib' title: 'Learning Latent Representations for style control and transfer in end-to-end speech synthesis' --- unsupervised learning, variational autoencoder, style transfer, speech synthesis Introduction {#sec:intro} ============ End-to-end text-to-speech (TTS) models which generate speech directly from characters have made rapid progress in recent years, and achieved very high voice quality [@shen2018natural; @ping2018deep; @li2018close]. While the single style TTS, usually neutral speaking style, is approaching the extreme quality close to human expert recording [@shen2018natural; @li2018close], the interests in expressive speech synthesis also keep rising. Recently, there also published many promising works in this topic, such as transferring prosody and speaking style within or cross speakers based on end-to-end TTS model [@wang2018style; @skerry2018towards; @stanton2018predicting]. Deep generative models, such as Variational Autoencoder (VAE) [@kingma2013auto] and Generative Adversarial Network (GAN) [@goodfellow2014generative], are powerful architectures which can learn complicated distribution in an unsupervised manner. Particularly, VAE, which explicitly models latent variables, have become one of the most popular approaches and achieved significant success on text generation [@bowman2015generating], image generation [@higgins2016beta; @burgess2018understanding] and speech generation [@akuzawa2018expressive; @hsu2017learning] tasks. VAE has many merits, such as learning disentangled factors, smoothly interpolating or continuously sampling between latent representations which can obtain interpretable homotopies [@bowman2015generating]. Intuitively, in speech generation, the latent state of speaker, such as affect and intent, contributes to the prosody, emotion, or speaking style. For simplicity, we’ll hereafter use speaking style to represent these prosody related expressions. The latent state plays a pretty similar role as the latent variable does in VAE. Therefore, in this paper we intend to introduce VAE to Tacotron2 [@shen2018natural], a state-of-the-art end-to-end speech synthesis model, to learn the latent representation of speaker state in a continuous space, and further to control the speaking style in speech synthesis. To be specific, direct manipulation can be easily imposed on the disentangled latent variable, so as to control the speaking style. On the other hand, with variational inference the latent representation of speaking style can be inferred from a reference audio, which then controls the style of synthesized speech. Style transfer, from reference audio to synthesized speech, is thus achieved. Last but not least, directly sampling on prior of latent distribution can generate a lot of speech with various speaking style, which is very useful for data augmentation. Comprehensive evaluation shows the good performance of this method. We have become aware of recent work by Akuzawa et al. [@akuzawa2018expressive] which combines an autoregressive speech synthesis model with VAE for expressive speech synthesis. The proposed work differs from Akuzawa’s as follows: 1) their goal is to synthesize expressive speech, which is achieved by direct sampling from prior of latent distribution at inference stage, while our goal is to control the speaking style of synthesized speech through direct manipulate latent variable or variational inference from a reference audio; 2) the proposed work is on end-to-end TTS model while Akuzawa’s not. The rest of the paper is organized as follows: Section \[sec:format\] introduces VAE model, our proposed model architecture and tricks for solving KL-divergence collapse problem. Section \[sec:exper\] presents the experimental results. Finally, the paper will be concluded in Section \[sec:conc\]. Model {#sec:format} ===== In this section, we first review Variational Autoencoder. We then show the details of our proposed style transfer model. Variational Autoencoder {#ssec:vae} ----------------------- Variational Autoencoder was first defined by Kingma et al. [@kingma2013auto] which constructs a relationship between unobserved continuous random latent variables z and observed dataset x. The true posterior density $p_{\theta}(\textbf{z}\|\textbf{x})$ is intractable, which results in an indifferentiable marginal likelihood $p_{\theta}(\textbf{x})$. To address this, a recognition model $q_{\phi}(\textbf{z}\|\textbf{x})$ is introduced as an approximation to the intractable posterior. Following the variational principle, $\log p_{\theta}(\textbf{x})$ can be rewritten as shown in equation (\[f1\]), where $\mathcal{L}(\theta,\phi;\textbf{x})$ is the variational lower bound to optimize. $$\begin{aligned} \log p_{\theta}(\textbf{x}) &= KL[q_{\phi}(\textbf{z}\|\textbf{x})\|\|p_{\theta}(\textbf{z}\|\textbf{x})] + \mathcal{L}(\theta,\phi;\textbf{x}) \\& \geq \mathcal{L}(\theta,\phi;\textbf{x}) \\&=\mathbb{E}_{q_{\phi}(\textbf{z}\|\textbf{x})}[\log p_{\theta}(\textbf{x}\|\textbf{z})] - KL[q_{\phi}(\textbf{z}\|\textbf{x})\|\|p_{\theta}(\textbf{z})] \end{aligned} \label{f1}$$ Generally, the prior over latent variables $p_{\theta}(\textbf{z})$ is assumed to be centered isotropic multivariate Gaussian $\mathcal{N}(\textbf{z};\textbf{0},\textbf{I})$, where I is the identity matrix. The usual choice of $q_{\phi}(\textbf{z}\|\textbf{x})$ is $\mathcal{N}(\textbf{z};\boldsymbol{\mu}(\textbf{x}),\boldsymbol{\sigma}^2(\textbf{x})\textbf{I})$, so that $KL[q_{\phi}(\textbf{z}\|\textbf{x})\|\|p_{\theta}(\textbf{z})]$ can be calculated in closed form. In practice, $\boldsymbol{\mu}(\textbf{x})$ and $\boldsymbol{\sigma}^2(\textbf{x})$ are learned from observed dataset via neural networks which can be viewed as an encoder. The expectation term in equation (\[f1\]) plays the role of decoder which decodes latent variables z to reconstruct x. The decoder may produce the expected reconstruction if the output of decoder is averaged over many samples of x and z [@doersch2016tutorial]. In the rest of the paper, $-\mathbb{E}_{q_{\phi}(\textbf{z}\|\textbf{x})}[\log p_{\theta}(\textbf{x}\|\textbf{z})]$ is referred to as reconstruction loss and $KL[q_{\phi}(\textbf{z}\|\textbf{x})\|\|p_{\theta}(\textbf{z})]$ is referred to as KL loss. Stochastic inputs can be processed by stochastic gradient descent via backpropagation, but stochastic units within the network cannot be processed by backpropagation. Thus, in practice, “reparameterization trick” is introduced to VAE framework. Sampling z from distribution $\mathcal{N}(\boldsymbol{\mu},\boldsymbol{\sigma}^2\textbf{I})$ is decomposed to first sampling $\boldsymbol{\epsilon} \sim \mathcal{N}(\textbf{0},\textbf{I})$ and then computing $\textbf{z}=\boldsymbol{\mu}+\boldsymbol{\sigma}\odot\boldsymbol{\epsilon}$, where $\odot$ denotes an element-wise product. Proposed Model Architecture {#ssec:archi} --------------------------- ![\[fig:FIG1\] An architecture of the proposed style transfer TTS model. The dashed lines denote sampling z from parametric distribution.](framework.jpg){width="7.5cm"} In this work, we introduce VAE into end-to-end TTS model and propose a flexible model for style control and style transfer. The whole network consists of two components, as shown in Fig.\[fig:FIG1\]: (1) A recognition model or inference network which encodes reference audio into a fixed-length short vector of latent representation (or latent variables z which stand for style representation), and (2) an end-to-end TTS model based on Tacotron 2, which converts the combined encoder states (including latent representations and text encoder states) to generated target sentence with specific style. The input texts are character sequences and the acoustic features are mel-frequency spectrograms. One may use various powerful and complex neural networks for the recognition model. Here, we only adopt a recurrent reference encoder followed by two fully connected layers. We use the same architecture and hyperparameters for reference encoder as Wang et al. [@wang2018style] which consists of six 2-D convolutional layers followed by a GRU layer. The output, which denotes some embedding of the reference audio, is then passed through two separate fully connected (FC) layers with linear activation function to generate the mean and standard deviation of latent variables z. The prior and approximative posterior are Gaussian distribution mentioned Section \[ssec:vae\]. Then z is derived by reparameterization trick. The encoder which deals with character inputs consists of three 1-D convolutional layers with 5 width and 512 channels followed by a bidirectional [@schuster1997bidirectional] LSTM [@hochreiter1997long] layer using zoneout [@krueger2016zoneout] with probability 0.1. The output text encoder state is simply added by z and then is consumed by a location-sensitive attention network [@chorowski2015attention] which converts encoded sequence to a fixed-length context vector for each decoder output step. In addition, z should be first passed through a FC layer to make sure the dimension equal to text encoder state before add operation. The attention module and decoder have the same architecture as Tacotron 2 [@shen2018natural]. Then, WaveNet [@van2016wavenet] vocoder is utilized to reconstruct waveform. The total loss of proposed model is shown in equation (\[f2\]). $$\begin{aligned} Loss =KL[q_{\phi}(\textbf{z}\|\textbf{x})\|\|p_{\theta}(\textbf{z})]-\mathbb{E}_{q_{\phi}(\textbf{z}\|\textbf{x})}[\log p_{\theta}(\textbf{x}\|\textbf{z},\textbf{t})]+l_{stop} \end{aligned} \label{f2}$$ Compared with the lower bound in equation (\[f1\]), the reconstruction loss term is conditioned on both latent variable z and input text t and a stop token loss $l_{stop}$ is added. It is worth mentioning that, after comparing L2-loss with negative log likelihood of Gaussian distribution, we finally choose L2-loss of mel spectrograms as reconstruction loss. Resolve KL collapse problem {#ssec:kl} --------------------------- During training, we observe that the KL loss $KL[q_{\phi}(\textbf{z}\|\textbf{x})\|\|p_{\theta}(\textbf{z})]$ is always found collapsed before they learned a distinguishable representation, which is a common phenomenon but a crucial issue in training VAE models. In other words, the convergence speed of KL loss far surpasses that of the reconstruction loss and the KL loss quickly drops to nearly zero and never rises again, which means the encoder doesn’t work. Thus, KL annealing [@bowman2015generating] is introduced to our task to solve this problem. That is, during training, add a variable weight to the KL term. The weight is close to zero at the beginning of training and then gradually increase. In addition, KL loss is taken into account once every K steps. By combining these two tricks, the KL loss keeps nonzero and avoids to collapse. Experiments and Analysis {#sec:exper} ======================== ![\[fig:FIG2\] Spectrograms generated by interpolation between two z. The interpolation coefficient is : (a) $\textbf{z}_a$, (b) $\frac{1}{3}\textbf{z}_a+\frac{2}{3}\textbf{z}_d$, (c) $\frac{2}{3}\textbf{z}_a+\frac{1}{3}\textbf{z}_d$, (d) $\textbf{z}_d$ .](interpolate.jpg){width="8cm"} ![\[fig:FIG3\]Spectrograms generated to demonstrate disentangled factors. The first row exhibits the control of pitch height only by adjusting latent dimension 6 to be -0.9, -0.1, 0.7. The second row shows that the local pitch variation is gradually magnified by increasing the value of dimension 10, which is 0.1, 0.5, 0.9, respectively. ](disentangle.jpg){width="8cm"} experimental setup {#sec:confi} ------------------ An 105-hour audiobook recordings dataset read with various storytelling styles by a single English speaker (Blizzard Challenge 2013) was used in our experiments. The dataset contains 58453 utterances for training and 200 for test. 80-dimensional mel spectrograms were extracted with frame shift 12.5 ms and frame length 50 ms. GST model [@wang2018style] with character inputs was used as our baseline model. The hyperparameters are set according to [@wang2018style]. As for our proposed model, the dimension of latent variables is 32. The parameter K mentioned in \[ssec:kl\] is 100 before 15000 training steps and 400 after the threshold. At inference stage, in evaluation of style control, we directly manipulate z without going through the whole recognition model. With regard to evaluation of style transfer, we feed audio clips as reference and go through the recognition model. Both parallel and non-parallel style transfer audios are generated and evaluated[^1]. Parallel transfer means the target text information is the same as reference audio’s, vice versa. Style control ------------- ### Interpolation of latent variables As mentioned in [@bowman2015generating], VAE supports smoothly interpolation and continuous sampling between latent representations, which obtains interpretable homotopies. Thus, we did interpolation operation between two z[^2]. One can generate speech with high speaking rate and high-pitch, the other with low speaking rate and low-pitch. The mel spectrograms of generated speech are shown in Fig. \[fig:FIG2\]. As we can see, both pitch and speaking rate of generated speech gradually decrease along with the interpolating. The result shows that the learnt latent space is continuous in controlling the trend of spectrograms which will further reflect in the change of style. ### Disentangled factors A disentangled representation means that a latent variable completely controls a concept alone and is invariant to changes from other factors [@higgins2016beta]. In experiments, we found that several dimensions of z could independently control different style attributes, such as pitch-height, local pitch variation, speaking rate. Fig. \[fig:FIG3\] shows the alteration of spectrograms by manipulating single dimension while fixing others. Adjusting one of these dimensions, only one attribute of generated speech changes. This shows that, in our model, VAE has the ability of learning disentangled latent factors. Next, we combined two disentangled dimensions to verify the additivity of latent variables. Fig. \[fig:FIG4\] illustrates the combination results of pitch height and local pitch variation attributes. It shows that the audio generated with combined z inherits the characteristics of both disentangled dimensions. ![\[fig:FIG4\]Audio (a) and (b) are generated with z which setting a single dimension to be non-zero with other dimensions to be zero. The valued dimension in (a) controls pitch height, while in (b) controls pitch variation. (c) is generated with the summation of z from (a) and (b). ](combina.jpg){width="8cm"} Style transfer -------------- Fig. \[fig:FIG5\] shows mel spectrograms of the style transferred synthetic speech aligned with their corresponding references. The reference audios are chosen from test set with certain styles. The synthesized audios share the same input text. As we can see in Fig. \[fig:FIG5\], the mel spectrograms of generated speech and their reference audio have pattern similarities, such as in pitch-height, pause time, speaking rate and pitch variation. ![\[fig:FIG5\]The first row exhibits the mel spectrograms of three recordings with different styles, while the second row exhibits the synthesised audios referenced on those recordings separately. The synthesised audios have the same text “She went into the shop . It was warm and smelled deliciously.”](ref.jpg){width="8cm"} ![\[fig:FIG6\]ABX test results for parallel and non-parallel transfers.](abx.jpg){width="8cm"} Subjective test --------------- To subjectively evaluate the performance of style transfer, crowd-sourcing ABX preference tests on parallel and non-parallel transfer were conducted. For parallel transfer, 60 audio clips with their texts are randomly selected from test set. For non-parallel transfer, 60 sentences of text and 60 other reference audio clips are selected to generate speech. The baseline voice is generated from the best GST model we have built. Each case in ABX test is judged by 25 native judgers. The total number of judger is 56 for parallel test and 57 for non-parallel test. The criterion in rating is “which one’s speaking style is closer to the reference style” with three choices: (1) 1st is better; (2) 2nd is better; (3) neutral. Fig. \[fig:FIG6\] shows the ABX results. As we can see, the proposed model outperforms GST model on both parallel and non-parallel style transfer (at p-value $<$ 10$^{-5}$). It shows that VAE can better model the latent style representations, which results in better style transfer. What’s more, the performance of the proposed model on non-parallel transfer is much better than that on parallel transfer, which shows the better generalization capability of the proposed model. Conclusion {#sec:conc} ========== A VAE module is introduced to end-to-end TTS model, to learn the latent representation of speaking style in a continuous space in an unsupervised manner, which then can control the speaking style in synthesized speech. We have demonstrated that the latent space is continuous and explored the disentangled factors in learned latent variables. The proposed model shows good performance in style transfer, which outperforms GST model via ABX test, especially in non-parallel transfer. Future work will keep focusing on getting better disentangled and interpretable latent representations. In addition, the scope of style transfer research will further extend to multi-speakers, instead of single speaker. [^1]: The audio samples can be found at <http://home.ustc.edu.cn/~zyj008/ICASSP2019>. [^2]: These two z are derived by feeding two audios to the recognition model, one with high speaking rate and high-pitch, the other with low speaking rate and low-pitch.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We impose uniform electric fields both parallel and normal to $180^{\circ}$ ferroelectric domain walls in PbTiO$_3$ and obtain the equilibrium structures using the method of anharmonic lattice statics. In addition to Ti-centered and Pb-centered perfect domain walls, we also consider Ti-centered domain walls with oxygen vacancies. We observe that electric field can increase the thickness of the domain wall considerably. We also observe that increasing the magnitude of electric field we reach a critical electric field $E^c$; for $E > E^c$ there is no local equilibrium configuration. Therefore, $E^c$ can be considered as an estimate of the threshold field $E_h$ for domain wall motion. Our numerical results show that Oxygen vacancies decrease the value of $E^c$. As the defective domain walls are thicker than perfect walls, this result is in agreement with the recent experimental observations and continuum calculations that show thicker domain walls have lower threshold fields.' author: - Arzhang Angoshtari - Arash Yavari title: \| Effect of External Normal and Parallel Electric Fields\ on $180^{\circ}$ Ferroelectric Domain Walls in PbTiO$_3$ --- Introduction ============ Ferroelectric materials have been used in many important applications such as high strain actuators, electro-optical systems, non-volatile and high density memories, etc. [@Scott2007; @Kalinin2010]. The properties of domain walls in ferroelectric materials including their structure, thickness, and mobility are important parameters as they determine the performance of devices that use these materials [@Jia2008]. Theoretical calculations have predicted that ferroelectric domain walls are atomically sharp and their thickness is about a few angstroms [@MeyerVanderbilt2001; @Padilla1996; @YaOrBh2006b; @AngYa2010]. However, experimental measurements show the existence of domain walls with thicknesses of a few micrometers [@Iwata2003; @Lehnen2000]. It has been observed that such broadening of domain walls is due to the presence of extrinsic defects, charged walls, and surfaces [@Choudhury2008]. Shilo et al. [@Shilo2004] used atomic force microscopy to measure the surface profile close to emerging domain walls in PbTiO$_3$ and then fitted it to the soliton-type solution of GLD theory. They measured wall widths of $1.5nm$ and $4nm$ and observed a wide scatter in wall widths. They suggested that the presence of point defects is responsible for such wide variations. Lee et al. [@Lee2005] proposed a continuum model to investigate this proposal and reproduced the experimentally observed range of wall widths with their model. They mentioned that the interaction between the order parameter and point defects and interaction of point defects with each other are two important interactions that should be considered properly in such modelings. Jia et al. [@Jia2008] investigated the cation-oxygen dipoles near $180^{\circ}$ domain walls in PbZr$_{0.2}$Ti$_{0.8}$O$_{3}$ thin films. They measured the width and dipole distortion across domain walls using the negative spherical-aberration imaging technique in an aberration-corrected transmission electron microscope and observed a large difference in atomic details between charged and uncharged domain walls. External electric field can cause the motion of ferroelectric domain walls if the magnitude of the field reaches the threshold field $E_h$ for wall motion, i.e., the field at which a domain wall begins to move after overcoming the intrinsic Peierls friction of the ferroelectric lattice [@Choudhury2008]. It was observed that threshold fields that are predicted via thermodynamic calculations are usually much greater than the experimental values. For example, Bandyopadhyay and Ray [@BandyopadhyayRay2004] predicted an upper limit for $E_h$ of LiNbO$_3$ to be $30000kV/cm$ but experimental observations show that the threshold field for wall motion can be less than $15kV/cm$. Choudhury et al. [@Choudhury2008] suggested that the reason for such large differences between theoretical and experimental values of $E_h$ is broadening of the domain walls. Using microscopic phase-field modeling, they show that the threshold field for moving an antiparallel ferroelectric domain wall dramatically drops by two or three orders of magnitude if the wall was diffused by only about $1-2nm$. Su and Landis [@Landis2007] developed a continuum thermodynamics framework to model the evolution of ferroelectric domain structures and investigated the fields near $90^{\circ}$ and $180^{\circ}$ domain walls and the electromechanical pining strength of an array of line charges on these domain walls. In this work, we investigate the effect of external electric field $(E)$ on the perfect and defective $180^{\circ}$ domain walls in PbTiO$_3$ using the method of anharmonic lattice statics. We consider both Pb-centered and Ti-centered perfect domain walls and also defective domain walls with oxygen vacancies. In agreement with experimental results, our calculations show that such defective domain walls are thicker than perfect walls [@AngYavari2010]. By increasing $E$ we reach a critical value $E^c$ such that for $E>E^c$ the lattice statics iterations do not converge. Therefore, this critical value can be considered as a lower bound for the threshold field for wall motion. This paper is organized as follows. In §\[sec:Ferro\], we explain the geometry of the perfect and defective domain walls that we use throughout this work. In §\[sec:Calculation\], we describe the method of analysis used in our calculations. Our numerical results are presented in §\[sec:Results\]. The paper ends with some concluding remarks in §\[sec:Concluding\]. Ferroelectric Domain Walls {#sec:Ferro} ========================== Due to the relative displacements between the center of the positive and negative charges, each unit cell of a ferroelectric crystal has a net polarization below its Curie temperature. Fig.\[Geom\](a) shows the relaxed unit cell of tetragonal PbTiO$_3$. In this work, we consider $180^{\circ}$ domain walls in PbTiO$_3$ parallel to a $(100)$-plane. These domain walls are two dimensional defects across which the direction of the polarization vector switches. There are two types of perfect $180^{\circ}$ domain walls in PbTiO$_3$: Pb-centered and Ti-centered domain walls. Fig.\[Geom\](b) shows the geometry of a Ti-centered domain wall. In addition to perfect domain walls, we also consider $180^{\circ}$ domain walls with oxygen vacancies. It is known that oxygen vacancies tend to move toward domain walls and pin them [@HeVanderbilt2001; @Calleja2003; @Salje2010]. Therefore, we study domain walls with oxygen vacancies sitting on them. In order to be able to obtain a solution, we need to consider periodically arranged vacancies on the domain walls. Although in reality oxygen vacancies have lower densities, our results with the current assumption can still provide important insights on the effect of oxygen vacancies on $180^{\circ}$ domain walls. Depending on which oxygen in the PbTiO$_3$ unit cell sits on the domain wall, there would be three types of defective domain walls: (i) O2-defective, (ii) O1-defective, and (iii) O3-defective. Fig.\[Geom\](c) shows an O1-defective domain wall. Note that O1- and O3-defective domain walls are Ti-centered while O2-defective domain wall is Pb-centered. It has been observed that O2-defective domain walls are not stable [@AngYavari2010; @HeVanderbilt2001], i.e., the lattice statics iterations do not converge. Thus, we consider O1- and O3-defective domain walls in the following. Let x, y, and z denote coordinates along the $\langle100\rangle$, $\langle010\rangle$, and $\langle001\rangle$-directions, respectively. We assume a 1D symmetry reduction, which means that all the atoms with the same x-coordinates have the same displacements. Therefore, we partition the 3D lattice $\mathcal{L}$ as $\mathcal{L}=\bigsqcup_{I}^{}\bigsqcup_{\alpha\in \mathbbm{Z}}\mathcal{L}_{I\alpha}$, where $\mathcal{L}_{I\alpha}$ and $\mathbbm{Z}$ are 2D equivalence classes parallel to the $(100)$ plane and the set of integers, respectively. $j=J\beta$ is the atom in the $\beta$th equivalence class of the $J$th sublattice. See [@YaOrBh2006a; @YavariAngoshtari2010] for more details on the symmetry reduction. Method of Calculation {#sec:Calculation} ===================== We apply a uniform electric field on $180^{\circ}$ domain walls and obtain the equilibrium structure using the method of anharmonic lattice statics [@YaOrBh2006a]. We use a shell potential for PbTiO$_3$ [@Asthagiri2006] for modeling the atomic interactions. Each ion is represented by a core and a massless shell in this potential. Let $\mathcal{L}$ denote the collection of cores and shells, $i\in\mathcal{L}$ denotes a core or a shell in $\mathcal{L}$, and $\left\{\mathbf{x}^i \right\}_{i\in \mathcal{L}}$ represents the current position of cores and shells. In this shell potential, three different energies are assumed to exist due to the interactions of cores and shells: $\mathcal{E}_{\textrm{short}}$, $\mathcal{E}_{\textrm{long}}$, and $\mathcal{E}_{\textrm{core-shell}}$. $\mathcal{E}_{\textrm{short}}\left(\left\{\mathbf{x}^i \right\}_{i\in \mathcal{L}}\right)$ denotes the energy of short range interactions, which are assumed to be only between Pb-O, Ti-O, and O-O shells. The short range interactions are described by the Rydberg potential of the form $(A+Br)\exp(-r/C)$, where A, B, and C are potential parameters and $r$ is the distance between interacting elements. $\mathcal{E}_{\textrm{long}}\left(\left\{\mathbf{x}^i \right\}_{i\in \mathcal{L}} \right)$ denotes the Coulombic interactions between the core and shell of each ion with the cores and shells of all the other ions. For calculating the classical Coulombic energy and force, we use the damped Wolf method [@Wolf99]. Finally, $\mathcal{E}_{\textrm{core-shell}}\left(\left\{\mathbf{x}^i \right\}_{i\in \mathcal{L}}\right)$ represents the interaction of core and shell of an atom and is assumed to be an anharmonic spring of the form $(1/2)k_2 r^2 +(1/24)k_4 r^4$, where $k_2$ and $k_4$ are constants. The total static energy is written as $$\begin{aligned} && \mathcal{E}\left(\left\{\mathbf{x}^i \right\}_{i\in \mathcal{L}} \right) = \mathcal{E}_{\textrm{short}}\left(\left\{\mathbf{x}^i \right\}_{i\in \mathcal{L}} \right) + \mathcal{E}_{\textrm{long}}\left(\left\{\mathbf{x}^i \right\}_{i\in \mathcal{L}} \right) \nonumber \\ && ~~~~~~~~~~~~~~~~~~~+\mathcal{E}_{\textrm{core-shell}}\left(\left\{\mathbf{x}^i \right\}_{i\in \mathcal{L}} \right).\end{aligned}$$ Note that all the calculations are done for absolute zero temperature. At this temperature PbTiO$_3$ has a tetragonal unit cell with lattice parameters $a=3.843~{\AA}$ and $c=1.08a$ [@Asthagiri2006]. Assume that a uniform electric field $\mathbf{E}=(E_{x},E_{y},E_{z})$ is applied to a collection of atoms. Then for the relaxed configuration $\mathcal{B}=\left\{\mathbf{x}^i \right\}_{i\in\mathcal{L}}\subset\mathbbm{R}^3$, we have $$\label{equilibrium} \frac{\partial \mathcal{E}}{\partial \mathbf{x}^i}+ q_{i}\mathbf{E}=\mathbf{0}~~~~~~~\forall~i\in\mathcal{L},$$ where $q_i$ denotes the charge of the ith particle (core or shell). To obtain the solution of the above problem, we utilize the Newton method. Having a configuration $\mathcal{B}^{k}$ the next configuration $\mathcal{B}^{k+1}$ is calculated from the current configuration $\mathcal{B}^{k}$ as: $\mathcal{B}^{k+1}=\mathcal{B}^{k}+\tilde{\boldsymbol{\delta}}^k$, where $$\tilde{\delta}^k=-\mathbf{H}^{-1}\left(\mathcal{B}^{k}\right)\cdot\boldsymbol{\nabla}\mathcal{E}\left(\mathcal{B}^{k}\right),$$ with $\mathbf{H}$ denoting the Hessian matrix. The calculation of the Hessian becomes inefficient as the size of the problem increases and hence we use the quasi-Newton method. This method uses the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm to approximate the inverse of the Hessian [@PressTVF1989] instead of the direct calculation of the Hessian at each iteration. We start from a positive-definite matrix and use the BFGS algorithm to update the Hessian at each iteration as follows: $$\begin{aligned} \label{BFGS} && \mathbf{C}^{i+1} = \mathbf{C}^i + \frac{\tilde{\boldsymbol{\delta}}^k\otimes\tilde{\boldsymbol{\delta}}^k}{ (\tilde{\boldsymbol{\delta}}^k)^{\textsf{T}}\cdot\mathbf{\Delta}} -\frac{\left(\mathbf{C}^{i}\cdot\mathbf{\Delta}\right)\otimes\left(\mathbf{C}^{i}\cdot\mathbf{\Delta}\right)} {\mathbf{\Delta}^{\textsf{T}}\cdot\mathbf{C}^{i}\cdot\mathbf{\Delta}} \nonumber\\ && ~~~~~~~~+ \left(\mathbf{\Delta}^{\textsf{T}}\cdot\mathbf{C}^{i}\cdot\mathbf{\Delta}\right)\mathbf{u}\otimes\mathbf{u},\end{aligned}$$ where $\mathbf{C}^{i}=\left(\mathbf{H}^i\right)^{-1}$, $\mathbf{\Delta}=\boldsymbol{\nabla}\mathcal{E}^{i+1}-\boldsymbol{\nabla}\mathcal{E}^{i}$, and $$\begin{aligned} \mathbf{u}=\frac{\tilde{\boldsymbol{\delta}}^k}{(\tilde{\boldsymbol{\delta}}^k)^{\textsf{T}}\cdot\mathbf{\Delta}}- \frac{\mathbf{C}^{i}\cdot\mathbf{\Delta}}{\mathbf{\Delta}^{\textsf{T}}\cdot\mathbf{C}^{i}\cdot\mathbf{\Delta}}.\end{aligned}$$ Calculating $\mathbf{C}^{i+1}$, one then should use $\mathbf{C}^{i+1}$ instead of $\mathbf{H}^{-1}$ to update the current configuration. If $\mathbf{C}^{i+1}$ is a poor approximation, then one may need to perform a linear search to refine $\mathcal{B}^{k+1}$ before starting the next iteration [@PressTVF1989]. In the presence of oxygen vacancies on the domain wall, one needs to consider charge redistribution between some ions. To model an oxygen vacancy using a shell potential, we remove the core and shell of the oxygen atom and because we assume a charge neutral oxygen vacancy, there will be a charge redistribution in the neighboring shells [@AngYavari2010]. It is known that charge redistribution is highly localized and hence in our calculations we equally distribute the charge $\Delta Q=Q_s+Q_c$, where $Q_s$ and $Q_c$ are oxygen shell and core charges, between the (fourteen) first nearest neighbors of each oxygen vacancy. ![The polarization profiles of O1-defective domain walls under (a) zero, (b) normal critical field ($E^{c}_{x}$), and (c) parallel critical field ($E^{c}_{y}$). (d) Components of the polarization vector $\overline{\mathbf{P}}=(\bar{P}_x,\bar{P}_y)$ of O1-defective domain walls under zero, $E^{c}_{x}$, and $E^{c}_{y}$.[]{data-label="PolarO1"}](PolO1) To obtain the equilibrium configuration under an external electric field we need to start from an appropriate initial configuration. This initial configuration for perfect and defective domain walls is the equilibrium configuration of these domain walls under zero electric field (see [@AngYavari2010; @YaOrBh2006b] for discussions on how to calculate these configurations). As we mentioned earlier, we assume a 1D symmetry reduction for the lattice and hence as is shown in Fig.\[CB\], our computational box (CB) consists of a row of unit cells perpendicular to the domain wall. In this figure, the shaded region is the computational box. Note that because in general there is no symmetry in the problem, we need to relax all the atoms inside the CB. For removing the rigid body translation freedom of the atoms, one should fix the core of an atom and relax the other atoms. We fix Pb-core (Ti-core) of an atom located on the domain wall in Pb-centered (Ti-centered) domain walls. Thus, if there are $M$ unit cells in the CB, we would have $30M-3$ variables in our calculations. We should mention that to investigate the effect of the size of CB in the domain wall plane, we consider CBs with one, four, and sixteen unit cells in the domain wall plane and therefore the number of the unit cells in CB in each case is $M$, $4M$, and $16M$, respectively. We observe that the final relaxed structure does not depend on the size of CB in the domain wall plane. This suggests that the symmetry reduction that we use in our calculations is a reasonable assumption for this problem. Note that we consider a finite number of unit cells in the CB and do not assume any periodicity condition in our calculations. This means that we need to impose some proper boundary conditions to take into account the effect of the atoms located outside of CB. To this end, we rigidly move the unit cells outside of CB with displacements equal to those of the first or last unit cell of the CB (the unit cell on the boundary of the CB that is closer to the unit cell outside of the CB). This is a natural boundary condition as we expect the bulk configuration far from the domain wall. In our calculations we use $M=20$ as larger values for $M$ do not affect the results. Imposing an external electric field should be done step by step, i.e., one first needs to obtain the configuration for $\mathbf{E}=\Delta \mathbf{E}_{1}$ from the initial configuration and then use this configuration to obtain the equilibrium configuration for $\mathbf{E}=\Delta \mathbf{E}_{1}+\Delta \mathbf{E}_{2}$ and so on. We use the average step size of $20kV/cm$ for electric field. Using this step size and force tolerance of $0.005eV{\AA}^{-1}$, our solutions converge after about $30$ to $40$ iterations. Numerical Results {#sec:Results} ================== In this section we present our numerical results for perfect and defective domain walls. Note that as the coordinates of cores and shells are close to each other, we only report the results for cores. Also as we mentioned earlier, x, y, and z are coordinates along the $\langle100\rangle$, $\langle010\rangle$, and $\langle001\rangle$-directions, respectively. Perfect domain walls: We plot the y-coordinates of Ti-cores under external electric field normal to the Ti-centered domain wall, $E_x$, in Fig.\[Perfect\](a). As expected, we see that increasing the electric field, the atomic structure loses its symmetry. We observe that there exists an upper bound for $E_x$, i.e., there exists a critical electric field $E^{c}_{x}$ such that for $E_{x}>E^{c}_{x}$ there is no local equilibrium structure. The critical value of the normal electric field is about $E^{c}_{x}=1400kV/cm$. The thickness of the domain wall slightly increases as the normal electric field increases. Note that domain wall thickness cannot be defined uniquely very much like boundary layer thickness in fluid mechanics. Here, domain wall thickness is by definition the region that is affected by the domain wall, i.e. those layers of atoms that are distorted. One can use definitions like the $99\%$-thickness in fluid mechanics and define the domain wall thickness as the length of the region that has $99\%$ of the far field rigid translation displacement. What is important here is that no matter what definition is chosen, domain wall “thickness" increases as the normal electric field increases. For a Ti-centered domain wall, the domain wall thickness increases from $3$ atomic spacings ($1nm$) to about $5$ atomic spacings ($1.5nm$) for $E_{x}=E^{c}_{x}$. Fig.\[Perfect\](b) depicts the y-coordinates of Ti-cores under an external electric field $E_y$ parallel to a Ti-centered domain wall. It is observed that such electric fields do not alter the domain wall thickness. Note that similar to the atomic structure for normal fields, the atomic structure under parallel fields loses its symmetry as well. The critical value of the parallel electric field is about $E^{c}_{y}=5900kV/cm$, which is $4$ times larger than that of the normal electric field. Fig.\[Perfect\](c) shows the y-coordinates of Pb-cores of a Pb-centered domain wall under normal electric field $E_x$. We observe that the critical electric field is about $E^{c}_{x}=6300kV/cm$, which is about $4.5$ times greater than the critical normal electric field of Ti-centered walls. Also it is observed that domain wall thickness increases to about $11$ atomic spacings ($4nm$) under critical normal electric field. The y-coordinates of Pb-cores of a Pb-centered domain wall under parallel electric field $E_y$ are shown in Fig.\[Perfect\](d). Similar to perfect Ti-centered domain walls, we observe that parallel electric fields do not affect the domain wall thickness. The critical parallel electric field is about $E^{c}_{y}=6500kV/cm$. For Pb-centered domain walls we see that unlike Ti-centered domain walls, the critical normal electric field is close to the critical parallel electric field. Fig.\[PolarPerfect\] depicts the polarization profiles normal and parallel to the domain walls. For calculation of the cell-by-cell polarization, we follow Meyer and Vanderbilt [@MeyerVanderbilt2001]. We plot $\overline{\mathbf{P}}=(\bar{P}_{x},\bar{P}_{y})= \mathbf{P}/\|\mathbf{P}_{b}\|$, where $\mathbf{P}$ is the polarization and $\|\mathbf{P}_{b}\|=80.1\mu Ccm^{-2}$ is the norm of the bulk polarization [@AngYa2010_4]. Fig.\[PolarPerfect\](a) shows $\bar{P}_{x}$ and $\bar{P}_{y}$ for a Ti-centered domain walls under zero and critical electric fields. In agreement with Lee et al. [@Lee2009] and Angoshtari and Yavari [@AngYa2010], it is observed that $(100)$ Ti-centered domain walls have a mixed Ising-Néel character, i.e., polarization rotates normal to the $(100)$-plane near the domain wall. For $E=0$, the maximum normal component of the polarization is about $2\%$ of the bulk polarization. For $E=E^{c}_{x}$, as can be expected, normal electric field causes the positive and negative charges to have normal displacements that create a polarization in the x-direction. This normal component of the polarization ($\bar{P}_x$) reaches to about $13.5\%$ of the bulk polarization at $E^{c}_x$, but we observe that normal electric field $E^{c}_x$ does not have a remarkable effect on the parallel component of polarization, $\bar{P}_{y}$. On the other hand, we observe that under $E=E^{c}_{y}$, $\bar{P}_{x}$ does not change considerably but $\bar{P}_y$ has an unsymmetric profile with the maximum value of about $105\%$ of the bulk polarization. Fig.\[PolarPerfect\](b) presents similar results for Pb-centered domain walls. Similar to Ti-centered domain walls, we observe that Pb-centered domain walls have a mixed Ising-Néel character [@Lee2009; @AngYa2010] with $\bar{P}_x$ about $2\%$ of the bulk polarization for zero electric field. For $E=E^{c}_{x}$, $\bar{P}_{x}$ reaches to about $38\%$ of the bulk polarization. Also we observe that $E^{c}_{x}$ has more impact on $\bar{P}_{y}$ compared to Ti-centered walls. Finally, it is observed that similar to Ti-centered domain walls, $E^{c}_{y}$ does not have a significant effect on $\bar{P}_x$ but makes $\bar{P}_y$ unsymmetric with maximum value of about $107\%$ of the bulk polarization. Defective domain walls: In this part we report the structure of defective domain walls under normal and parallel external electric fields. Because the results for O1- and O3-defective domain walls are similar, we only present the results for O1-defective walls, which are Ti-centered. Note that as we mentioned earlier, O2-defective domain walls, which are Pb-centered, are not stable. Our calculations show that they are not stable even under external electric fields. We had earlier shown that they are not stable under strain as well [@AngYavari2010]. Fig.\[O1\](a) depicts the y-coordinates of Ti-cores in an O1-defective domain wall under normal external electric field $E_x$. The critical normal field is about $E^{c}_{x}=380kV/cm$. It is observed that domain wall thickness increases up to about $16$ atomic spacings ($6nm$) under the critical normal electric field. Comparing O1-defective atomic structure with the structure of perfect Ti-centered domain wall under normal field (Fig.\[Perfect\](a)), we observe that oxygen vacancies increase the thickness of the domain wall considerably. Also it is observed that critical normal electric field of defective domain walls is smaller than that of perfect Ti-centered wall. Fig.\[O1\](b) shows the y-coordinates of Ti-cores in an O1-defective domain walls under parallel electric field $E_y$. The value of the critical field is about $E^{c}_{y}=5100kV/cm$. Here we observe a major difference between the atomic structures of perfect and defective domain walls; unlike perfect domain walls, parallel electric fields increase the thickness of defective domain walls up to about $13$ atomic spacings ($5nm$) under the critical parallel electric field. Also similar to normal electric fields, we observe that critical electric field of defective domain walls is smaller than that of perfect domain walls. Defective domain walls are thicker than perfect domain walls. The observation that the defective domain walls have smaller critical electric fields is in agreement with the experimental observations of Choudhury et al. [@Choudhury2008]. They observed that the threshold field for domain wall motion exponentially decreases as the domain wall width increases. Fig.\[PolarO1\] shows the polarization profiles for O1-defective domain walls. It is observed that similar to perfect domain walls, defective domain walls have an Ising-Néel character with $\bar{P}_{x}$ of about $2.5\%$ of the bulk polarization for zero electrical field. For $E=E^{c}_x$, $\bar{P}_x$ reaches to about $55\%$ of the bulk polarization, which is greater than the corresponding values for perfect domain walls, and $\bar{P}_y$ shows more Ising-type character. As we mentioned earlier, for $E=E^{c}_y$ we observe a difference between perfect and defective domain walls; unlike perfect walls, parallel electric fields have considerable effects on $\bar{P}_x$: it reaches to about $55\%$ of the bulk polarization under $E^{c}_y$. Similar to perfect domain walls, $\bar{P}_y$ has an unsymmetric distribution and reaches to about $106\%$ of the bulk polarization. Concluding Remarks {#sec:Concluding} ================== In this work we obtained the atomic structure of perfect and defective $180^{\circ}$ domain walls in PbTiO$_3$ under both parallel and normal external electric fields using the method of anharmonic lattice statics. We observe that electric field can increase the thickness of a domain wall considerably (up to $5$ times thicker than domain walls under no external electric field). This can be one reason for the wide scatter of the domain wall thicknesses observed in experimental measurements. In agreement with previous works [@AngYavari2010; @Shilo2004], we observe that oxygen vacancies can increase the thickness of the domain walls. We also observe that by increasing the external electric field we reach a critical electric field $E^c$. For $E>E^c$ there is no local equilibrium configuration and hence $E^c$ can be considered as an estimate of the threshold field for the domain wall motion. We observe that defective domain walls, which are thicker than perfect domain walls, have smaller critical fields. This is in agreement with the experimental observations that show the threshold field decreases as the domain wall thickness increases [@Choudhury2008]. In practice, it has been observed that domain walls move or break down under electric fields in the order of a few $kV/cm$ [@Choudhury2008; @Roy2010], which are considerably smaller than the high-fields that we consider here. We do not consider break down of the domain walls in our model. Also as was mentioned earlier, the high density of oxygen vacancies that we assume is unrealistic. In practice, steps and other complex defects on domain walls can increase the thickness of the domain walls considerably [@Iwata2003; @Lehnen2000; @AngYa2010_4]. Therefore, as the threshold fields for domain walls decrease exponentially with the increase of the domain wall width [@Choudhury2008], one can obtain better estimates for the critical electric fields with more realistic models for defects on domain walls. Also as suggested by Roy et al. [@Roy2010] electric fields change the potential parameters. In this paper our aim is to demonstrate that even with our simple model, one can show that the threshold field has an inverse relation with the domain wall thickness. We benefited from a discussion with C.M. Landis. We thank an anonymous reviewer whose comments helped us improve the paper. J. F. Scott, [Science]{}, [315]{} , 954 (2007). S. V. Kalinin, A. N. Morozovska, L. Q. Chen and B. J. Rodriguez, [Rep. Prog. Phys.]{}, [73]{} , 056502 (2010). C-L. Jia, S-B. Mi, K. Urban, I. Vrejoiu, M. Alexe, D. Hesse, [Nature Mater.]{}, [7]{}, 57 (2008). B. Meyer and D. Vanderbilt [Phys. Rev. B]{}, [65]{}, 104111 (2002). J. Padilla, W. Zhong and D. Vanderbilt [Phys. Rev. B]{}, [53]{}, R5969 (1996). A. Yavari, M. Ortiz and K. Bhattacharya, [Philos. Mag.]{}, [87]{} , 3997 (2007). A. Angoshtari and A. Yavari, [EPL]{}, [90]{}, 27007 (2010). M. Iwata, K. Katsuraya, I. Suzuki, M. Maeda, N. Yasuda and Y. K. Ishibashi, [Jpn. J. Appl. Phys.]{}, [42]{}, 6201 (2003). P. Lehnen, J. Dec and W. Kleemann, [J. Phys. D: Appl. Phys.]{}, [33]{}, 1932 (2000). S. Choudhury et al., [J. Appl. Phys.]{}, [104]{}, 084107 (2008). D. Shilo, G. Ravichandran and K. Bhattacharya, [Nature Mater.]{}, [3]{}, 453 (2004). W. T. Lee, E. K. H. Salje and U. Bismayer, [Phys. Rev. B]{}, [72]{}, 104116 (2005). A. K. Bandyopadhyay and P. C. Ray, [J. Appl. Phys.]{}, [95]{}, 226 (2004). Y. Su and C. M. Landis, [J. Mech. Phys. Solids]{}, [55]{}, 280 (2007). L. X. He and D. Vanderbilt, [Phys. Rev. B]{}, [68]{}, 134103 (2001). M. Calleja, M. T. Dove and E. K. H. Salje, [J. Phys.: Condens. Matter]{}, [15]{}, 2301 (2003). L. Goncalves-Ferreira, S. A. T. Redfern, E. Artacho, E. Salje and W. T. Lee, [Phys. Rev. B]{}, [81]{}, 024109 (2010). A. Angoshtari and A. Yavari, [Comput. Mater. Sci.]{}, [48]{}, 258 (2010). A. Yavari, M. Ortiz and K. Bhattacharya, [J. Elasticity]{}, [86]{} , 41 (2007). A. Yavari and A. Angoshtari, [Inter. J. Solids Struct.]{}, [47]{}, 1807 (2010). A. Asthagiri, Z. Wu, N. Choudhury and R. E. Cohen, [Ferroelec.]{}, [333]{} , 69 (2006). D. P. Wolf, P. Keblinski, S. R. Phillpot and J. Eggebrecht, [J. Chem. Phys.]{}, [110]{}, 8254 (1999). W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, [Numerical recipes: the art of scientific computing]{} (Cambridge University Press, Cambridge, 1989). A. Angoshtari and A. Yavari, [J. Appl. Phys.]{}, [108]{}, 084112 (2010). D. Lee, R. K. Behera, P. Wu, H. Xu, Y. L. Li, S. B. Sinnott, S. R. Phillpot, L. Q. Chen and V. Gopalan, [Phys. Rev. B]{}, [80]{}, 060102(R) (2009). A. Roy, M. Stengel, and D. Vanderbilt, [Phys. Rev. B]{}, [81]{}, 014102 (2010).	{ "pile_set_name": "ArXiv" }
--- abstract: 'We investigate the $Q$-ball formation in the thermal logarithmic potential by means of the lattice simulation, and reconfirm qualitatively the relation between $Q$-ball charge and the amplitude of the Affleck-Dine field at the onset of its oscillation. We find time dependence of some properties of the $Q$ ball, such as its size and the field value at its center. Since the thermal logarithmic potential decreases as the temperature falls down, the gravity-mediation potential will affect the properties of the $Q$ ball. Even in the case when the gravity-mediation potential alone does not allow $Q$-ball solution, we find the transformation from the thick-wall type of the $Q$ ball to the thin-wall type, contrary to the naive expectation that the $Q$ balls will be destroyed immediately when the gravity-mediation potential becomes dominant at the center of the $Q$ ball.' author: - Shinta Kasuya date: 'February 22, 2010' title: Formation of the $Q$ ball in the thermal logarithmic potential and its properties --- Introduction ============ $Q$-ball formation is ubiquitous in the Affleck-Dine mechanism for baryogenesis [@KuSh; @EnMc; @KK1; @KK2; @KK3]. Soon after the Affleck-Dine field begins rotation in the potential which allows the $Q$-ball solution, the homogeneous field starts to fluctuate and transforms into lumps. The actual formation was investigated numerically on the lattices for the gauge- and gravity-mediated supersymmetry (SUSY) breaking scenarios [@KK1; @KK2; @KK3; @Qsim]. On the other hand, the properties of the $Q$ ball are quite different among the different forms of the potential. For example, for the flat potential, such as in the gauge-mediated SUSY breaking scenario, the $Q$-ball size, the field value at its center, and the field rotation speed depend on the charge $Q$ of the $Q$ ball nontrivially [@DKS]. The potential may be dominated by thermal effects after inflation. In the Affleck-Dine scenario, the field amplitude at the onset of the rotation is very large, and the two-loop thermal effects on the potential are crucial [@AnDi; @FHY]. This potential is given by $$V_T \sim T^4 \log \left(\frac{\|\Phi\|^2}{T^2}\right),$$ for large field values [^1] . We called it the thermal logarithmic potential [@KK3]. The energy density of the universe is dominated by the oscillation of the inflaton after inflation, but there exists dilute plasma [@KT] which would build up the thermal logarithmic potential. We considered in Ref. [@KK3] that the $Q$-ball formation in the thermal logarithmic potential would be more or less similar to the time-independent logarithmic potential because of the fast growth of the fluctuations of the field in spite of the time dependence of the temperature $T$. We thus borrowed the results from the time-independent logarithmic potential case. For example, the charge of the produced $Q$ ball is given by [@KK3] $$Q=\beta\left(\frac{\|\Phi\|}{T}\right)^4_{\rm osc},$$ where $\beta\approx 6\times 10^{-4}$, and the subscript “osc" denotes the values of the variables at the onset of the oscillation of the field. In this article we actually perform lattice simulations of the $Q$-ball formation in the thermal logarithmic potential, and see the evolution of the field and the distribution of the produced $Q$ balls. We also study of the properties of the formed $Q$ balls, especially about the evolution of the field amplitude at the center of the $Q$ ball. In addition to the thermal logarithmic term, there would also be a mass term due to the gravity-mediated SUSY breaking effects for both the gauge- and gravity-mediation scenarios. The mass term alone (including one-loop potential) does not allow the $Q$-ball solution in some cases. In such cases, one may naively consider those $Q$ balls, created when the thermal logarithmic potential dominates, to disappear when the mass term begins to dominate the potential at the field value of the $Q$-ball center. As shown below, however, the $Q$-ball solution still exists later. Actually, the thick-wall type (gauge-mediation type) $Q$ ball transforms to the thin-wall type. The structure of the article is as follows. In the next section, we show the results of lattice simulations for both time-independent and thermal logarithmic potentials. Some properties of the $Q$ ball with thermal logarithmic potential are shown in Sec.III, while, in Sec. IV, we focus on the transition from the gauge-mediation type to the thin-wall type Q balls when the mass term in the potential gradually dominates over the thermal logarithmic potential. We finally conclude in Sec.V. Q-ball formation ================ We investigate the $Q$-ball formation by means of three-dimensional lattice simulations. Interpolating the thermal mass term at smaller field amplitudes and the thermal two-loop potential growing logarithmically at larger field values, we take the following form of the potential: $$V_T(\Phi) = T^4 \log \left( 1+ \frac{\|\Phi\|^2}{T^2}\right).$$ In addition, we restudy the time-independent case where $T$ is replaced by the constant mass $M_F$ for comparison, which is nothing but the potential of the gauge-mediated SUSY breaking effects. Here we consider the inflaton-oscillation dominated universe after inflation before reheating, but a similar argument applies also to the radiation dominated era. The temperature decreases as the universe expands as $T \propto a^{-3/8}\propto t^{-1/4}$, where $a(t)$ is the scale factor of the universe. Since we are interested in the period after the Affleck-Dine field starts the oscillation (rotation) when $H^2 \simeq \|V''\|$, we take initial conditions as[^2] $$\begin{aligned} & & \varphi_1(0) = \varphi_0(1 + \delta_1), \quad \varphi'_1(0)=\delta_2, \nonumber \\ & & \varphi_2(0) = \delta_3, \quad \varphi'_2(0)=\sqrt{2}(1+\delta_4), \nonumber \\ & & \tau(0) = \frac{2}{3h} = \frac{\sqrt{2}}{3}\varphi_0,\end{aligned}$$ where all the variables are normalized by the temperature at the onset of the oscillation, $T_{\rm osc}$, such that $\varphi=\Phi/\sqrt{2}T_{\rm osc}$, $h=H/T_{\rm osc}$, $\xi=T_{\rm osc} x$ and $\tau=T_{\rm osc} t$. Here we decompose the field into real and imaginary parts as $\varphi = \varphi_1+i\varphi_2$. $\delta$’s represent the fluctuations of $O(10^{-7})$. We mostly use $256^3$ lattices, but in order to see any box size effects we also perform on $350^3$ lattices in some cases, but find no crucial differences between them. Figure \[fig\_charge\] shows the initial amplitude dependence of the largest charge of the $Q$ ball produced for the time-independent (lower) and thermal (upper) logarithmic potentials. Here we average the charge over five realizations of the initial fluctuations except for the smallest initial amplitude case ($\varphi_0=300$). The lower line corresponds to the relation obtained in Ref. [@KK3], $Q = \beta \varphi_0^4$ where $\beta\approx 6\times 10^{-4}$, thus we reconfirmed the previous results. ![Charge of the Q ball depending on the amplitude of the Affleck-Dine field at the onset of the oscillation. The lower (upper) points and line correspond to the time-independent (thermal) logarithmic potential.[]{data-label="fig_charge"}](form.eps){width="90mm"} On the other hand, in the thermal logarithmic potential, we have a similar relation $Q=\beta' \varphi_0^4$ with $\beta'\approx 2\times 10^{-3}$, although it might have a little tilt. Thus, qualitative features for the $Q$ balls in the thermal logarithmic potential can be captured by the case with the time-independent logarithmic potential with $M_F$ being replaced by $T_{osc}$. The difference between the time-independent and thermal logarithmic potentials can be qualitatively considered as follows. $Q$-ball charge can be estimated as $Q \sim q \ell_H^3$ at the formation time, where $q$ is the charge density and $\ell_H \sim t$ is the horizon scale. From Fig. \[fig\_fluct\], the formation times are $a_0 \sim 4.6$ and $a_T \sim 9.3$, respectively, for the time-independent and thermal logarithmic potential. Thus, the ratio of the charges would be $\sim (a_T/a_0)^{-3}(t_T/t_0)^3\sim (a_T/a_0)^{3/2}\sim 3$. Notice that the later rise and the slower growth of the amplitude of the fluctuations are due to the shrinking instability band $0<k/a<2T^2/\phi_0$ and the lowering growth rate $T^2/(\sqrt{2}\phi_0)$ due to the decreasing temperature $T$. ![Evolution of the fluctuations of the Affleck-Dine in the time-independent (green, the earlier rise) and thermal (red, the later rise) logarithmic potentials for $\varphi_0=10^3$. Notice that all the parameters including initial conditions are taken to be the same.[]{data-label="fig_fluct"}](fluct.eps){width="85mm"} The distribution of the $Q$ balls in example cases are shown in Table \[dist-th\] for the thermal logarithmic potential and in Table \[dist-log\] for the time-independent logarithmic potential, where we identified $Q$ balls with $Q > 10^3$ and $Q > 10^5$, respectively, in these cases. As one can see, the charge is dominated by those $Q$ balls with the charge of the largest magnitude. Therefore, it is fairly reasonable to estimate any relations among the $Q$-ball parameters by using the largest $Q$ ball, as we derived the charge and the initial field amplitude above. Since it is beyond the scope of the present paper to provide an analytical estimate of the distribution, we just provide a fitting formula of the form $$N_i(\tilde{Q}) = \alpha_i \varphi_0 \tilde{Q}^{-\eta} e^{-\left(\frac{\tilde{Q}}{Q_{{\rm max}, i}}\right)^2} \quad (i=0,T),$$ where $\tilde{Q}$ denotes the charge in terms of the order of magnitude. Here $\eta\approx0.3$. $Q_{\rm max} = \beta_i \varphi_0^4$ with $\beta_0 \approx 6\times 10^{-4}$ and $\beta_T=2\times 10^{-3}$ for the time-independent and thermal logarithmic potentials, respectively. $\alpha_0 \approx 1.6$ and $\alpha_T\approx 1.1$. Charge $\ $ Numbers $\ $ Sum of the charge $\ $ Fraction ----------- ------ --------- ------ -------------------- ------ ---------- $O(10^9)$ 2 $3.77 \times 10^9$ 0.7358 $O(10^8)$ 2 $6.33 \times 10^8$ 0.1235 $O(10^7)$ 15 $4.39 \times 10^8$ 0.0857 $O(10^6)$ 15 $5.82 \times 10^7$ 0.0114 $O(10^5)$ 12 $7.21 \times 10^6$ 0.0014 $O(10^4)$ 51 $1.05 \times 10^6$ 0.0002 $O(10^3)$ 195 $8.94 \times 10^6$ 0.0002 : \[dist-th\] Distribution of Q balls in the thermal logarithmic potential for $\varphi_0=10^3$. Charge $\ $ Numbers $\ $ Sum of the charge $\ $ Fraction -------------- ------ --------- ------ ----------------------- ------ ---------- $O(10^{10})$ 5 $1.42 \times 10^{11}$ 0.7807 $O(10^9)$ 10 $2.33 \times 10^{10}$ 0.1280 $O(10^8)$ 20 $5.92 \times 10^9$ 0.0326 $O(10^7)$ 44 $1.56 \times 10^9$ 0.0086 $O(10^6)$ 56 $2.27 \times 10^8$ 0.0012 $O(10^5)$ 163 $6.46 \times 10^7$ 0.0004 : \[dist-log\] Distribution of Q balls in the time independent logarithmic potential for $\varphi_0=3\times 10^3$. Q-ball properties ================= Let us investigate the evolution of the $D$-dimensional $Q$ ball formed in the thermal logarithmic potential, where for $D=1$ and 2 it is wall- and stringlike objects, respectively [@KK1]. The charge of the $Q$ ball is given by $$\label{constQ} Q \sim a^3 R^D q \sim {\rm const.},$$ where $q$ is the charge density and $R$ is the $Q$-ball size. Charge conservation implies that $Q$ is constant. If we write $\Phi({\bf x},t) = \phi({\bf x})e^{i\omega t}/\sqrt{2}$, the energy of the $Q$ ball is written as $$\begin{aligned} E & = & \int d^3x \left[\frac{1}{2}(\nabla\phi)^2+V(\phi)-\frac{1}{2}\omega^2\phi^2\right]+\omega Q \nonumber \\ & = & \int d^3x \left[ E_{\rm grad} + V_1 + V_2 \right] +\omega Q,\end{aligned}$$ where $$\begin{aligned} & & E_{\rm grad} \sim \frac{\phi^2}{a^2R^2}, \\ & & V_1 \sim T^4 \log\left( 1+\frac{\phi^2}{2T^2}\right) \sim T^4,\\ & & V_2 \sim \omega^2 \phi^2.\end{aligned}$$ When the energy takes the minimum value, the equipartition is achieved for gauge-mediation type $Q$ balls [@KK1]. From $E_{\rm grad} \sim V_1 \sim V_2$, together with the charge conservation (\[constQ\]), we obtain the evolution of the (comoving) $Q$-ball size $R$, the rotation speed of the field $\omega$, and the field amplitude at the center of the $Q$ ball $\phi_c$, respectively as $$\begin{aligned} R & \propto & a^{-(4-\gamma)/(D+1)},\\ \omega & \propto & a^{-(D-3+\gamma)/(D+1)},\\ \phi_c & \propto & a^{(D-3-\frac{D-1}{2}\gamma)/(D+1)},\end{aligned}$$ where we define $\gamma$ by $V_1 \sim T^4 \propto a^{-\gamma}$. Notice that $\gamma=3/2$ and 4 for the inflaton-oscillation and radiation dominated universe, respectively. These properties are observed in the lattice simulations which we perform for the $D=3$ case. Transformation of Q-ball types ============================== In addition to the thermal logarithmic term in the potential, there is a mass term which stems from the gravity-mediated SUSY breaking effects. This potential can be written as $$V_m = m_{3/2}^2 \|\Phi\|^2 \left[ 1+ K \log\left(\frac{\|\Phi\|^2}{M_^2}\right)\right],$$ where $m_{3/2}$ is the gravitino mass, and one-loop effects are included. $K$ is either a positive or negative constant of $O(0.01 -0.1)$, and $M_$ is a normalization scale. This potential alone allows the $Q$-ball solution only if $K<0$ [@EnMc; @New]. In the opposite case ($K>0$), $Q$-ball formation is prohibited. One may thus be apt to consider that the $Q$ ball created in the thermal logarithmic potential will disappear once the mass term with $K>0$ dominates over the thermal one at the field value of the $Q$-ball center, $V_T(\phi_c) < V_m(\phi_c)$. This condition can be written as $$\phi_c > \phi_{eq}\sim \frac{T^2}{m_{3/2}}.$$ Since $\phi_{eq} \propto a^{-3/4}$ and $\phi_c \propto a^{-3/8}$ for $D=3$ in the inflaton-oscillation domination, and $\phi_{eq} \propto a^{-2}$ and $\phi_c \propto a^{-1}$ for any $D$ in the radiation domination, it is true that $V_m(\phi_c)$ will eventually overcome $V_T(\phi_c)$. However, the $Q$-ball solution [does]{} exist for $V=V_T+V_m$ with $K>0$ for $T \gtrsim m_{3/2}$. Therefore, $Q$ balls are not destroyed, but the metamorphosis will take place in such situations. It might be best to simulate on the lattices to verify this phenomenon, but it is very time-consuming to perform. Here, instead, we take another approach, and leave the lattice simulations for future work. We seek the $Q$-ball solution for $V=V_T+V_m$ at some time snapshots. In order to obtain the solution, we just have to solve the equation $$\frac{d^2\phi}{dr^2}+\frac{2}{r}\frac{d\phi}{dr}+\left(\omega^2\phi-\frac{dV}{d\phi}\right)=0,$$ with boundary conditions $\phi(\infty)=0$ and $\phi'(0)=0$ [@small; @KK4]. Since we would like to compare the results also to the solution of the pure logarithmic potential case, we change the value of the mass $m_{3/2}$ in $V_m$, leaving the logarithmic potential time independent. Thus we use the following potential with various values for $m_{3/2}$: $$V=M_F^4\log\left(1+\frac{\phi^2}{2M_F^2}\right) +\frac{1}{2} m_{3/2}^2 \phi^2\left(1+K\log\frac{\phi^2}{2M_^2}\right).$$ Notice that it is practically the same if one varies $M_F$ while $m_{3/2}$ is fixed, since the results are derived and shown by variables normalized with respect to $M_F$. The profiles are shown in Fig. \[fig\_prof\] for $m_{3/2}/M_F= (0, 0.5, 1, 2, 3, 4, 5, 10, 20) \times 10^{-7}$ from the top to the bottom. Here we set $K=0.1$ and $M_/M_F=10^6$. It mimics the time evolution that the mass term eventually dominates over the logarithmic potential. Since the charge should be conserved, the angular velocity $\omega$ increases: $\omega/M_F=(2.0, 2.1, 2.4, 3.2, 4.1, 5.2, 6.2, 11, 21)\times 10^{-6}$ from the top to the bottom. One can see that the thick-wall type (gauge-mediation type) of $Q$ ball transforms into the thin-wall-like type as time goes on. Notice that $\phi_c > \phi_{eq}$ takes place for $m/M_F \gtrsim 10^{-7}$. ![Profiles of the Q balls with $Q\approx 7.1 \times 10^{26}$ for $m/M_F = (0, 0.5, 1, 2, 3, 4, 5, 10, 20) \times 10^{-7}$ from the top to the bottom. The $Q$-ball profile “evolves" from the top to the bottom.[]{data-label="fig_prof"}](profile.eps){width="90mm"} In the gravity-mediation, $Q$-ball solutions may exist until $T \sim m_{3/2}$ for large enough $\omega$. [^3] After that, $Q$ balls will disappear quickly, but the Affleck-Dine field can no longer be regarded as homogeneous because the field is localized near the place where destroyed $Q$ balls had existed; they were very much separated from each other. Typical separation length is estimated as $\ell_{H,{\rm formation}} (a_{\rm destruction}/a_{\rm formation})$. On the other hand, in the gauge-mediation, $Q$ balls will remain intact for $m_\phi > m_{3/2}$, where $m_\phi=\sqrt{V''(0)}$ is the curvature of the gauge-mediation potential at $\phi=0$, since the $Q$-ball solution will exist irrespective of the temperature. Conclusions =========== We have investigated the $Q$-ball formation in the thermal logarithmic potential by means of three dimensional lattice simulations. First of all, $Q$ balls are actually formed. This is because the growth of the field fluctuations is fast enough to create $Q$ balls, in spite of the shrinking instability band due to the decreasing temperature, and so on. We have found that the charge of the $Q$ ball in the thermal logarithmic potential has almost the same dependence on the initial amplitude of the Affleck-Dine field, $$Q=\beta' \left(\frac{\phi_0}{T_{\rm init}}\right)^4,$$ with $\beta \approx 2 \times 10^{-3}$, which is a factor of 3 larger than that in the time-independent logarithmic potential case. We have also estimated the evolutions of parameters, such as the $Q$-ball size $R$, the field rotation velocity $\omega$, and the field value at the center of the $Q$ ball $\phi_c$. Since the thermal logarithmic potential decreases as the temperature drops, the mass term would dominate over the thermal logarithmic one at the field value $\phi=\phi_c$. Even if the mass term $V_m$ with positive $K$ alone does not allow any $Q$-ball solution, the total potential $V=V_T+V_m$ [does]{} allow the solution. In fact, we have found that the thick-wall type Q ball will eventually transform into the thin-wall type. Finally, the $Q$ balls will disappear when $T \sim m_{3/2}$, with an inhomogeneous Affleck-Dine field being left afterward in the gravity-mediation scenario. Acknowledgments {#acknowledgments .unnumbered} =============== The author is grateful to Masahide Yamaguchi for useful discussion. [90]{} A. Kusenko and M. E. Shaposhnikov, Phys. Lett. B [418]{}, 46 (1998). K. Enqvist and J. McDonald, Phys. Lett. B [425]{}, 309 (1998); Nucl. Phys. B [538]{}, 321 (1999). S. Kasuya and M. Kawasaki, Phys. Rev. D [61]{}, 041301(R) (2000). S. Kasuya and M. Kawasaki, Phys. Rev. D [62]{}, 023512 (2000). S. Kasuya and M. Kawasaki, Phys. Rev. D [64]{}, 123515 (2001). K. Enqvist, A. Jokinen, T. Multamäki and I. Vilja, Phys. Rev. D [63]{}, 083501 (2001); T. Multamäki and I. Vilja, Phys. Lett. B [535]{}, 170 (2002); M. I. Tsumagari, Phys. Rev. D [80]{}, 085010 (2009). G. R. Dvali, A. Kusenko and M. E. Shaposhnikov, Phys. Lett. B [417]{}, 99 (1998). A. Anisimov and M. Dine, Nucl. Phys. B [619]{}, 729 (2001). M. Fujii, K. Hamaguchi and T. Yanagida, Phys. Rev. D [63]{}, 123513 (2001). S. Kasuya, M. Kawasaki and F. Takahashi, Phys. Rev. D [68]{}, 023501 (2003); S. Kasuya, M. Kawasaki and F. Takahashi, Phys. Lett. B [578]{}, 259 (2004). E. W. Kolb and M. S. Turner, [The Early universe]{} (Addison-Wesley, Redwood City, CA, 1990). S. Kasuya and M. Kawasaki, Phys. Rev. Lett. [85]{}, 2677 (2000). A. Kusenko, Phys. Lett. B [404]{}, 285 (1997). S. Kasuya and M. Kawasaki, Phys. Rev. D [80]{}, 123529 (2009). T. Chiba, K. Kamada and M. Yamaguchi, arXiv:0912.3585 \[astro-ph.CO\]. [^1]: We do not consider a negative thermal logarithmic potential [@negalog; @cg], because $Q$-ball formation does not occur in that potential. [^2]: It is implicitly assumed that the helical motion is dynamically achieved by so-called $A$ terms. [^3]: The fate of the $Q$ ball will depend on the cosmological situation. For thorough analysis, see Ref. [@CKY].	{ "pile_set_name": "ArXiv" }
--- abstract: '[In superconductors with unconventional pairing mechanisms, the energy gap in the excitation spectrum often has nodes, which allow quasiparticle excitations at low energies. In many cases, e.g. $d$-wave cuprate superconductors, the position and topology of nodes are imposed by the symmetry, and thus the presence of gapless excitations is protected against disorder. Here we report on the observation of distinct changes in the gap structure of iron-pnictide superconductors with increasing impurity scattering. By the successive introduction of nonmagnetic point defects into BaFe$_2$(As$_{1-x}$P$_x$)$_2$ crystals via electron irradiation, we find from the low-temperature penetration depth measurements that the nodal state changes to a nodeless state with fully gapped excitations. Moreover, under further irradiation the gapped state evolves into another gapless state, providing bulk evidence of unconventional sign-changing $s$-wave superconductivity. This demonstrates that the topology of the superconducting gap can be controlled by disorder, which is a strikingly unique feature of iron pnictides. ]{}' author: - 'Y. Mizukami$^{1,2}$' - 'M. Konczykowski$^{3}$' - 'Y. Kawamoto$^1$' - 'S. Kurata$^{1,2}$' - 'S. Kasahara$^1$' - 'K. Hashimoto$^{1,4}$' - 'V. Mishra$^5$' - 'A. Kreisel$^6$' - 'Y. Wang$^6$' - 'P.J. Hirschfeld$^6$' - 'Y. Matsuda$^1$' - 'T. Shibauchi$^{1,2}$' title: 'Disorder-induced topological change of the superconducting gap structure in iron pnictides' --- When repulsive electron-electron interactions are strong, a sign change in the superconducting order parameter (or the gap function) often leads to some energy gain for electron pairing [@Scalapino95; @Hirschfeld11]. The positions of the gap nodes in momentum ${\bf k}$ space, at which the order parameter changes sign, are determined by the superconducting pairing interactions. In the high transition temperature ($T_{\rm c}$) cuprates, it has been established that the lines of nodes are located along $k_x=\pm k_y$, as expected in the $d_{x^2-y^2}$ symmetry of the order parameter [@Tsuei00]. In iron pnictides, on the other hand, the multiband Fermi surface structure leads to several candidates for the superconducting symmetry, including sign-changing $s_\pm$ with and without nodes, sign-preserving $s_{++}$, $d$-wave and time-reversal symmetry breaking $s+{\rm i}d$ [@Hirschfeld11; @Mazin08; @Kuroki09; @Graser09; @Kontani10; @Thomale11]. Different symmetries and details of higher order momentum dependence of the gap yield topologically different nodal structures, such as vertical or horizontal lines of nodes, nodal loops or no nodes. The effects of nonmagnetic impurities on unconventional superconductivity has been one of the central issues in condensed matter physics [@Balatsky06; @Alloul09]. For the conventional BCS superconductors, it has been established that the effect is essentially null: the transition temperature is robust against nonmagnetic impurity scattering (the so-called Anderson theorem) and so is the superconducting gap. For the unconventional superconductors, the important aspect of the impurity effects is to mix gaps on different parts of the Fermi surface and thereby smear out the momentum dependence [@Mishra09]. In the case of superconducting gap with symmetry protected nodes such as $d$-wave, this averaging mechanism leads to the suppression of the gap amplitude, which enhances the low-lying quasiparticle excitations near the nodal positions. In addition to this, the sign change in the order parameter gives rise to impurity-induced Andreev bound states, which lead to additional quasiparticle excitations [@Hirschfeld93]. Such pair-breaking effects of impurities have been observed, e.g. in Zn-doped YBa$_2$Cu$_3$O$_7$ in the bulk measurements of magnetic penetration depth, where the $T$-linear temperature dependence in the clean-limit $d$-wave superconductivity gradually changes to a $T^2$ dependence at low temperatures with increasing Zn concentrations [@Bonn93]. In sharp contrast, when the nodal positions are not symmetry protected, as in the nodal $s$-wave case, the averaging mechanism of impurity scattering can displace the nodes, and at a certain critical impurity concentration the nodes may be lifted if intraband scattering dominates [@Mishra09], eliminating the low-energy quasiparticle excitations. In the fully gapped state after the node lifting, we have two cases in the multiband superconductors. If the signs of the order parameter on different bands are opposite, residual interband scattering can give rise to midgap Andreev bound states localized at nonmagnetic impurities that can contribute to the low-energy excitations, provided that the concentration of impurities is enough to create such states. If there is no sign change, Anderson’s theorem will be recovered, no Andreev states will be created, and thus no significant further change is expected. Indeed such a difference between nodal sign-changing $s_\pm$ and sign-preserving $s_{++}$ cases has been theoretically suggested by the recent calculations for multiband superconductivity considering the band structure of iron pnictides [@Wang13]. Therefore, the impurity effects on the gap nodes and low-energy excitations can be used as a powerful probe for the pairing symmetry of superconductors. ![[Particle irradiation and created defects.]{} [a]{}, Contour plot of maximum energy transferred to Fe atoms (recoil energy) for irradiated particles with the rest mass $m$ and incident energy $E$. Typical threshold energy $E_d$ for the displacement of Fe atoms from the lattice is marked by thick orange line. 2.5MeV electron irradiation used in this study (red square) has orders of magnitude smaller recoil energy than other particle irradiation owing to the small mass. Typical energies for neutron (green diamond), proton (purple circle), $\alpha$ particle (black triangle) and heavy-ion Pb (blue diamond) irradiation are indicated. [b-d]{}, Schematic illustrations for different types of defects created by particle irradiation. Columnar defects can be created by heavy-ion irradiation ([b]{}). Particle irradiation with relatively large recoil energies tends to have cascades of point defects due to successive collisions of atoms ([c]{}). Electron irradiation with a small recoil energy is the most reliable way to obtain uniform point defects ([d]{}). []{data-label="irradiation"}](Fig1.eps){width="1.0\linewidth"} In this study, we focus on isovalently substituted system BaFe$_2$(As$_{1-x}$P$_x$)$_2$ close to the optimum composition with $T_{\rm c}\sim30$K [@Kasahara10]. This system is particularly suitable for the study of the impurity effect on gap structure, because several experiments have indicated that the pristine crystals are very clean and exhibit nodes in the superconducting gap [@Shibauchi14]. To detect changes in the low-energy quasiparticle excitations, we measure the magnetic penetration depth $\lambda$ at low temperatures, a fundamental property of superconductors whose $T$ dependence is directly related to the excited quasiparticles. The tunnel diode oscillators (TDOs) in $^3$He and dilution refrigerators operating at 13MHz are used to measure the temperature dependence of $\lambda$ down to 0.4K and 80mK, respectively [@Hashimoto10; @Hashimoto12]. To introduce the impurity scattering in a controllable way, we employ electron irradiation with the incident energy of 2.5MeV, for which energy transfer from impinging electron to the lattice is above the threshold energy for the formation of vacancy-interstitial (Frenkel) pairs that act as point defects. The long attenuation length and the small recoil energy due to the small mass of electrons is important to create uniformly distributed point defects over the entire crystal with the width of $\sim 30\,\mu$m (Figs.\[irradiation\]a-d). If the recoil energy is too large, one may expect creation of complex defects such as clusters and cascades of point defects and, in the extreme case, columnar tracks, which has been realized by heavy-ion irradiation [@Nakajima09] (Figs.\[irradiation\]b and c). Another advantage of electron irradiation is that, unlike chemical substitutions, the defects can be introduced without changing lattice constants, which is quite important as the gap structure may be sensitive to the lattice parameters in iron-based superconductors [@Hirschfeld11]. ![[Effect of electron irradiation on the superconducting transition in BaFe$_2$(As$_{1-x}$P$_x$)$_2$ single crystals.]{} [a]{}, Temperature dependence of ac susceptibility at 13MHz for crystals (\#28A and \#28B) from a batch of $T_{\rm c0}=28$K with irradiated doses of 0, 1.0, 2.0, 4.0, 6.0 C/cm$^2$ with decreasing $T_{\rm c}$. [b]{}, Similar plot for crystals (\#29A, \#29B, \#29C, \#29D, and \#29E) from a batch of $T_{\rm c0}=29$K with irradiated doses of 0, 1.5, 2.7, 4.7, 4.9, 8.3C/cm$^2$ with decreasing $T_{\rm c}$. [c]{}, Temperature dependence of resistivity for a crystal (\#30A) with $T_{\rm c0}=30$K with irradiated doses of 0, 1.1, 2.3, 3.8, 4.5C/cm$^2$ with decreasing $T_{\rm c}$. Dotted lines are linear extrapolations to zero temperature to estimate changes in residual resistivity $\Delta\rho_0$. [d]{}, Transition temperature $T_{\rm c}$ normalized by the pristine value $T_{\rm c0}=30$K as a function of $\Delta\rho_0$, determined by the resistivity measurements in crystal \#30A. []{data-label="Tc_dose"}](Fig2.eps){width="0.75\linewidth"} By successive electron irradiation into clean BaFe$_2$(As$_{1-x}$P$_x$)$_2$ ($x=0.33-0.36$) single crystals with the initial transition temperatures $T_{c0}$ of 28, 29, and 30K, we observe a systematic downward shift of $T_{\rm c}$ with increasing defect dose (Figs.\[Tc\_dose\]a-d). The transition width in the ac susceptibility data measured by the TDO frequency change (Figs.\[Tc\_dose\]a and b) remains almost unchanged after irradiation, which implies a good homogeneity of the point defects introduced. The temperature dependence of in-plane resistivity $\rho(T)$ measured by the van der Pauw configuration (Fig.\[Tc\_dose\]c) shows parallel shifts of $\rho(T)$ after each irradiation in a single crystal. The parallel shifts imply that point defects increase impurity (elastic) scattering with little changes of carrier concentrations and inelastic scattering, which is closely related to the electron correlations. We estimate the change in the residual resistivity $\Delta\rho_0$ by extrapolating the normal state data linearly to the zero temperature. The $T_{\rm c}$ reduction rate with respect to $\Delta\rho_0$ is about $-0.3$K$\mu\Omega^{-1}$cm$^{-1}$ (Fig.\[Tc\_dose\]d), which is comparable to the similar electron irradiation measurements in Ru-substituted BaFe$_2$As$_2$ [@Prozorov14]. Previous studies of the $T_{\rm c}$ reduction rate in iron-pnictide superconductors by chemical substitutions [@Sato10; @Li12; @Kirshenbaum12] and by particle irradiation [@Prozorov14; @Tarantini10; @Nakajima10; @Taen13] focus on the comparisons with theoretical calculations for $s_\pm$ and $s_{++}$-wave superconductivity [@Onari09; @Wang13], but they report various values of the suppression rate. Here we instead focus on changes of low-energy excitations induced by disorder from penetration depth measurements. ![[Effect of electron irradiation on the low-temperature penetration depth in BaFe$_2$(As$_{1-x}$P$_x$)$_2$ single crystals.]{} [a,b]{}, Change in the magnetic penetration depth $\Delta\lambda$ plotted against $(T/T_{\rm c})^2$ for the same $T_{\rm c0}=28$K samples as Fig.\[Tc\_dose\]a ([a]{}) and for the $T_{\rm c0}=29$K samples as Fig.\[Tc\_dose\]a ([b]{}). The same colours are used as Figs.2a and b for the corresponding irradiated doses. Each curve is shifted vertically for clarity. Lines are the $T^2$ dependence fits at high temperatures. [c,d]{}, Temperature dependence of $\Delta\lambda$ for sample \#28B with 2.0C/cm$^2$ ([c]{}) and for sample \#29C with 2.7C/cm$^2$ ([d]{}). Red lines are the fits to the exponential dependence expected in fully gapped superconductors. [e]{}, Temperature dependence of $\Delta\lambda$ for sample \#29A before (red) and after irradiation (blue) of 4.9C/cm$^2$ dose. Inset is an expanded view at the lowest temperatures. []{data-label="lambda"}](Fig3.eps){width="1.0\linewidth"} In pristine crystals of BaFe$_2$(As$_{1-x}$P$_x$)$_2$, the penetration depth shows a strong temperature dependence at low temperatures (Figs.\[lambda\]a and b), as reported previously [@Hashimoto10; @Hashimoto12]. The temperature dependence of the change in the penetration depth $\Delta\lambda(T)=\lambda(T)-\lambda(0)$ can be fitted to a power law $T^n$ with the exponent $n<1.5$, indicating that this system has lines nodes in the energy gap. After irradiation, we first find that $\Delta\lambda(T)$ at the lowest temperatures becomes more gentle, or the exponent $n$ increases, and in $T_{\rm c0}=28$K crystal we have almost $T^2$ dependence (Fig.\[lambda\]a). Further increase of the defect density results in a flat temperature dependence below $T/T_{\rm c}\sim0.06-0.1$, indicating that the system is changed to a fully gapped state (Figs.\[lambda\]c and d). A fitting to the exponential dependence gives the gap size $2\Delta\gtrsim k_{\rm B}T_{\rm c}$, which is a substantial portion of the BCS value. Such flat temperature dependence of $\Delta\lambda$ has been reproduced in three crystals measured in this study, one of which has been measured down to 80mK (Fig.\[lambda\]e). The data represent completely temperature-independent behaviour below $T/T_{\rm c}\sim0.05$ within a precision of $\sim1$[Å]{} (Fig.\[lambda\]e, inset). The fact that we do not observe any Curie upturn at low temperatures is a strong indication that the point defects are essentially nonmagnetic [@Cooper96]. In fact, if we use the reported estimate of the defect density [@Beek13], our precision gives an upper limit of magnetic moment of $\sim0.2\mu_{\rm B}$ per Fe defect, which is much smaller than the moment expected in the magnetic spin states of Fe. Further increase of the defect density leads to another change of $\Delta\lambda(T)$. The $T$-dependence at the lowest temperatures gets steeper with increasing defect dose, which is an opposite trend to the initial change at low doses. This second stage of the changes clearly indicates that by irradiation we create the low-energy excitations again inside the formed gap. At the highest doses we measured, we observe the $T^2$ dependence, which is a manifestation of the formation of the Andreev bound states expected for the sign-changing order parameter with impurity scattering. The impurity band corresponding to these bound states must overlap the Fermi level in order to cause this effect. ![[Theoretical calculations of impurity effects in iron-based superconductors.]{} [a]{}, Schematic of possible $s_\pm$ and $s_{++}$ states. Large circles and small ellipsoids (black lines) are hole and electron Fermi surfaces, respectively. Red and blue represents the superconducting order parameter with different signs. [b]{}, Schematic of $s_\pm$ order parameter vs. azimuthal angle $\phi$ (top row) and density of states $N$ vs. energy $\omega$ (bottom row) with increasing irradiation dosage (from left to right). Dotted lines (top row) are zero lines. [c]{}, Density of states at Fermi level $N(0)$ for nodal band plotted as false color vs. inter/intraband scattering ratio $\alpha$ and irradiation-induced residual resistivity $\Delta \rho_0$ for the model defined in Ref.. [d]{}, Penetration depth change $\Delta\lambda$ vs. reduced temperature $T/T_{\rm c}$ for different values of $\Delta \rho_0$ as marked in legend, and inter/intraband scattering ratio $\alpha =0.6$ (arrows in [c]{}). Each curve is shifted vertically for clarity. []{data-label="theory"}](Fig4.eps){width="1.0\linewidth"} The observation of the impurity-induced fully gapped state provides one of the strongest pieces of evidence from bulk measurements that the nodes are not symmetry protected in this system. Moreover, the two-stage changes in the low energy excitations observed here most likely come from the peculiar band-dependent gap functions of multiband iron-pnictide superconductors. Owing to the well separated hole and electron Fermi surface sheets, the fully gapped $s_\pm$ superconductivity with opposite signs of the order parameter on different bands has been predicted by the theories based on spin fluctuations with the antiferromagnetic vector $\bf{Q}=(\pi,\pi)$ in the 2-Fe zone notation [@Mazin08; @Kuroki09]. The presence of nodes in the pristine samples implies some competing additional mechanisms, such as spin fluctuations with a different $\bf{Q}$ vector [@Kuroki09] or orbital fluctuations which prefer the same sign in both bands [@Kontani10]. To reproduce the observed two-stage changes, we have made calculations for the low-energy excitations and the low-temperature behaviour of $\lambda(T)$ by using a two-band model [@Wang13] (Figs.\[theory\]a-d). We assume that one band is fully gapped, and the other has accidental nodes whose positions are not symmetry protected. Two possible realizations with $s_\pm$ and $s_{++}$ structure are shown in Fig.\[theory\]a. Figure\[theory\]b shows schematically the evolution of the gap function in the $s_\pm$ case as a function of increasing disorder of dominant intraband and subdominant interband character. The corresponding densities of states are also plotted. Only in the last panels, where the subgap bound state appears, do the $s_\pm$ and $s_{++}$ cases differ qualitatively. Figures\[theory\]c and d show concrete theoretical calculations using the model of Ref. in support of this scenario. The variation of the density of states at the Fermi level with scattering rate and the ratio $\alpha$ of inter- and intraband scattering (Fig.\[theory\]c), clearly shows the nonmonotonic behaviour of the density of states with scattering rate at a fixed $\alpha$. The temperature dependence of $\Delta\lambda$ depends directly on the residual density of states at the Fermi level, and therefore changes with increasing scattering as follows: (1) $T$-linear, (2) $T^2$, (3) exponential, and (4) $T^2$ (Fig.\[theory\]d). It is the overlap of the bound state in the last panel – not present in the $s_{++}$ case – with the Fermi level that gives rise to the “re-entrant” final $T^2$ dependence of the penetration depth. Thus the observed sequential changes of $\Delta\lambda(T)$ in this system are fully consistent with a sign changing $s$-wave superconducting gap of $s_\pm$ type. Our results demonstrates that the gap topology and the low-energy excitations can be changed by controlling disorder. Such an impurity effect is unprecedented among superconductors, highlighting a unique aspect of iron-based superconductors. Our study also shows that the impurity effects on gap structure can provide phase information on the superconducting order parameter in bulk measurements, in contrast to other phase sensitive experiments [@Tsuei00; @Hanaguri10], most of which require excellent surfaces or interfaces. Methods {#methods .unnumbered} ======= The single crystals of BaFe$_2$(As$_{1-x}$P$_x$)$_2$ were grown by the self-flux method [@Kasahara10] and characterised by several techniques as reported previously [@Shibauchi14]. The observation of the quantum oscillations in this series of crystals and the sharp superconducting transition indicate the very high quality of our pristine crystals. We used several crystals from three batches, which exhibit slightly different $T_{\rm c0}$ values of 28, 29, and 30K. Electron irradiation experiments were performed on SIRIUS platform operated by LSI at Ecole Polytechnique, composed by Pelltron type NEC accelerator and closed cycle cryocooler maintaining sample immersed in liquid hydrogen at 20-22K during irradiation. The low-temperature environment is important to prevent defect migration and agglomeration. Partial annealing of introduced defects occurs upon warming to room temperature and sample transfer [@Prozorov14]. The resistivity measurements were performed at LSI, Ecole Polytechnique by the van der Pauw method with four contacts on corners of crystals to minimize the possible effect of the unirradiated area due to contacts. The penetration depth measurements by using 13MHz TDOs [@Hashimoto10; @Hashimoto12] were performed at Kyoto University before and after irradiation. [99]{} Scalapino, D.J. The case for $d_{x^2-y^2}$ pairing in the cuprate superconductors. [Physics Reports]{} [250]{}, 329-365 (1995). Hirschfeld, P.J., Korshunov, M.M., & Mazin, I.I. Gap symmetry and structure of Fe-based superconductors. [Rep. Prog. Phys.]{} [74]{}, 124508 (2011). Tsuei, C.C. & Kirtley, J.R. Pairing symmetry in cuprate superconductors. [Rev. Mod. Phys.]{} [72]{}, 969-1016 (2000). Mazin, I.I., Singh, D.J., Johannes, M.D., & Du, M.H. Unconventional superconductivity with a sign reversal in the order parameter of LaFeAsO$_{1-x}$F$_x$. [Phys. Rev. Lett.]{} [101]{}, 057003 (2008). Kuroki, K., Usui, H., Onari, S., Arita, R., & Aoki, H. Pnictogen height as a possible switch between high-$T_{\rm c}$ nodeless and low-$T_{\rm c}$ nodal pairings in the iron-based superconductors. [Phys. Rev. B]{} [79]{}, 224511 (2009). Graser, S., Maier, T.A., Hirschfeld, P.J., & Scalapino, D.J. Near-degeneracy of several pairing channels in multiorbital models for the Fe pnictides. [New J. Phys.]{} [11]{}, 025016 (2009). Kontani, H.& Onari, S. Orbital-fluctuation-mediated superconductivity in iron pnictides: Analysis of the five-orbital Hubbard-Holstein model. [Phys. Rev. Lett.]{} [104]{}, 157001 (2010). Thomale, R., Platt, C., Hanke, W., Hu, J., & Bernevig, B.A. Exotic $d$-wave superconducting state of strongly hole-doped K$_x$Ba$_{1-x}$Fe$_2$As$_2$. [Phys. Rev. Lett.]{} [107]{}, 117001 (2011). Balatsky, A.V., Vekhter, I., & Zhu, J.-X. Impurity-induced states in conventional and unconventional superconductors. [Rev. Mod. Phys.]{} [78]{}, 373-433 (2006). Alloul, H., Bobroff, J., Gabay, M. & Hirschfeld, P.J. Defects in correlated metals and superconductors. [Rev. Mod. Phys.]{} [81]{}, 45-108 (2009). Mishra, V., Boyd, G., Graser, S., Maier, T., Hirschfeld, P.J., & Scalapino, D.J. Lifting of nodes by disorder in extended-$s$-state superconductors: Application to ferropnictides. [Phys. Rev. B]{} [79]{}, 094512 (2009). Hirschfeld, P.J. & Goldenfeld, N. Effect of strong scattering on the low-temperature penetration depth of a $d$-wave superconductor. [Phys. Rev. B]{} [48]{}, 4219 (1993). Bonn, D.A. [et al.]{} Comparison of the imfluence of Ni and Zn impurities on the electromagnetic properties of YBa$_2$Cu$_3$O$_{6.95}$. [Phys. Rev. B]{} [50]{}, 4051 (1993). Wang, Y., Kreisel, A., & Hirschfeld, P.J. Using controlled disorder to distinguish $s_\pm$ and $s_{++}$ gap structure in Fe-based superconductors. [Phys. Rev. B]{} [87]{}, 094504 (2013). Kasahara, S. [et al.]{} Evolution from non-Fermi- to Fermi-liquid transport via isovalent doping in BaFe$_2$(As$_{1-x}$P$_x$)$_2$ superconductors. [Phys. Rev. B]{} [81]{}, 184519 (2010). Shibauchi, T., Carrington, A., & Matsuda, Y. A quantum critical point lying beneath the superconducting dome in iron-pnictides. [Annu. Rev. Condens. Matter Phys.]{} [5]{}, 113-135 (2014). Hashimoto, K. [et al.]{} Line nodes in the energy gap of superconducting BaFe$_2$(As$_{1-x}$P$_x$)$_2$ single crystals as seen via penetration depth and thermal conductivity. [Phys. Rev. B]{} [81]{}, 220501(R) (2010) Hashimoto, K. [et al.]{} A sharp peak of the zero-temperature penetration depth at optimal composition in the iron-based superconductor BaFe$_2$(As$_{1-x}$P$_x$)$_2$. [Science]{} [336]{}, 1554-1557 (2012). Nakajima, Y. [et al.]{} Enhancement of critical current density in Co-doped BaFe$_2$As$_2$ with columnar defects introduced by heavy-ion irradiation. [Phys. Rev. B]{} [80]{}, 012510 (2009). Prozorov, R. [et al.]{} Effect of electron irradiation on superconductivity in single crystals of Ba(Fe$_{1-x}$Ru$_x$)$_2$As$_2$ $(x =0.24)$. Preprint at http://arXiv.org/abs/1405.3255 (2014). Sato, M. [et al.]{} Studies on effects of impurity doping and NMR measurements of La 1111 and/or Nd 1111 Fe-pnictide superconductors. [J. Phys. Soc. Jpn.]{} [79]{}, 014710 (2010). Li, J. [et al.]{} Superconductivity suppression of Ba$_{0.5}$K$_{0.5}$Fe$_{2-2x}M_{2x}$As$_2$ single crystals by substitution of transition metal ($M = $ Mn, Ru, Co, Ni, Cu, and Zn). [Phys. Rev. B]{} [85]{}, 214509 (2012). Kirshenbaum, K., Saha, S.R., Ziemak, S., Drye, T. & Paglione, J. Universal pair-breaking in transition-metal-substituted iron-pnictide superconductors. [Phys. Rev. B]{} [86]{}, 140505(R) (2012). Tarantini, C. [et al.]{} Suppression of the critical temperature of superconducting NdFeAs(OF) single crystals by Kondo-like defect sites induced by $\alpha$-particle irradiation. [Phys. Rev. Lett.]{} [104]{}, 087002 (2010). Nakajima, Y. [et al.]{} Suppression of the critical temperature of superconducting Ba(Fe$_{1-x}$Co$_x$)$_2$As$_2$ by point defects from proton irradiation. [Phys. Rev. B]{} [82]{}, 220504(R) (2010). Taen, T. [et al.]{} Pair-breaking effects induced by 3-MeV proton irradiation in Ba$_{1-x}$K$_x$Fe$_2$As$_2$. [Phys. Rev. B]{} [88]{}, 224514 (2013). Onari, S. & Kontani, H. Violation of Anderson’s theorem for the sign-reversing $s$-wave state of iron-pnictide superconductors. [Phys. Rev. Lett.]{} [103]{}, 177001 (2009). Cooper, J.R. Power-law dependence of the ab-plane penetration depth in Nd$_{1.85}$Ce$_{0.15}$CuO$_{4-y}$. [Phys. Rev. B]{} [54]{}, R3753(R) (1996). van der Beek, C. J. [et al.]{} Electron irradiation of Co, Ni, and P-doped BaFe$_2$As$_2$–type iron-based superconductors. [J. Phys.: Conf. Ser.]{} [449]{}, 012023 (2013). Hanaguri, T., Niitaka, S., Kuroki, K., & Takagi, H. Unconventional $s$-wave superconductivity in Fe(Se,Te). [Science]{} [328]{}, 474-476 (2010). Acknowledgments {#acknowledgments .unnumbered} =============== We thank C.J. van der Beek, A. Carrington, H. Kontani, and R. Prozorov for fruitful discussion. We also thank B. Boizot, J. Losco, and V. Metayer for technical assistance. This work was supported by Grants-in-Aid for Scientific Research (KAKENHI) from Japan Society for the Promotion of Science (JSPS), and by the “Topological Quantum Phenomena” (No.25103713) Grant-in Aid for Scientific Research on Innovative Areas from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan. Irradiation experiments were supported by EMIR network, proposal No.11–10–8071. Author contributions {#author-contributions .unnumbered} ==================== T.S. conceived the project. S.Kasahara carried out sample preparation. Y.Mizukami, M.K., S.Kasahara, and T.S. performed irradiation experiments. M.K. performed resistivity measurements. Y.Mizukami, Y.K., S.Kurata, K.H., and T.S. performed penetration depth measurements and analysed data. V.M., A.K., Y.W., and P.J.H. performed theoretical calculations. Y.Matsuda and T.S. directed the research. T.S. wrote the manuscript with inputs from Y.Mizukami, M.K., K.H., V.M., A.K., Y.W., P.J.H., and Y.Matsuda.	{ "pile_set_name": "ArXiv" }
--- abstract: 'Transferring entangled states between photon pairs is essential for quantum communication technologies. Semiconductor quantum dots are the most promising candidate for generating polarization-entangled photons deterministically. Recent improvements in photonic quality and brightness now make them suited for complex quantum optical purposes in practical devices. Here we demonstrate for the first time swapping of entangled states between two pairs of photons emitted by a single quantum dot. A joint Bell measurement heralds the successful generation of the Bell state $\Psi^+$ with a fidelity of up to $0.81 \pm 0.04$. The state’s nonlocal nature is confirmed by violating the CHSH-Bell inequality. Our photon source is compatible with atom-based quantum memories, enabling implementation of hybrid quantum repeaters. This experiment thus is a major step forward for semiconductor based quantum communication technologies.' author: - Michael - Robert - Yan - Jingzhong - Disheng - Fei - 'Oliver G. ' title: 'Entanglement Swapping with Semiconductor-generated Photons' --- Semiconductor light sources have revolutionized science and technology since laser diodes [@Hall1962; @Kroemer1963] and vertical-cavity surface-emitting lasers (VCSELs) [@Melngailis1965; @Soda1979] arrived in the 1960’s. Quantum mechanics lies at the roots for these devices, yet quantum states of light have only in recent decades been studied extensively in their own right - sparking the “second quantum revolution”. Semiconductor sources can now emit single photons [@Somaschi2016] and entangled photons [@Salter2010] on demand (see Fig. \[fig1\]a), more reliably and intensely than non-linear crystals. They hold great potential for a range of applications in quantum communication [@Gisin2002], quantum metrology [@Dowling2008] and quantum computation [@Knill2001]. The next step towards building quantum networks is to transfer entangled states between distinct pairs of photons [@Kimble2008; @Bose1998; @Briegel1998]. This entails substituting the pairwise entanglement in two-photon states with entanglement between photons from different pairs [@Bennett1993; @Zukowski1993]. The first experiment to do this two decades ago [@Pan1998] used a technique based on spontaneous parametric down conversion in a nonlinear optical crystal [@Kwiat1995; @Shih1988]. Though such sources are widely used, for example to entangle multiple photons [@Wang2016], their brightness and therefore scalability is fundamentally limited owing to Poissonian emission statistics [@Scarani2005]. Semiconductor quantum dots (QDs), by contrast, are able to generate entangled photon pairs deterministically one by one [@Akopian2006]. However, until recently, QDs were too faint and of poor degree of entanglement and indistinguishability to use for advanced quantum applications. Improvements of the past three years have overcome these limitations. Highly coherent [@Jahn2015] and strongly entangled photons [@Keil2017; @Huber2017] can now be generated with high brightness [@Chen2018] and reproducibility [@Keil2017] from QDs. Here we demonstrate, for the first time, entanglement swapping between polarization-entangled photons emitted by a semiconductor QD. The Bell state $\Psi^+$ is generated with high fidelity and strong nonlocal characteristics, proven by violating the CHSH-Bell inequality [@Bell1964; @CHSH1969]. Our semiconductor sources are compatible with atom-based quantum memories. This opens up their use in devices such as quantum repeaters (the quantum equivalent of a classical amplifier) [@Rakher2013] which are essential for long distance quantum communication. Entangled state generation ========================== ![ (a) Historical development of integrated entangled photon sources, starting from semiconductor lasers and photonic entanglement based on nonlinear optical materials, to scalable quantum dot sources of entangled photons. (b) Principle of an entanglement swapping experiment using a quantum dot. Two distinct pairs of entangled photons are generated (emission 1 and 2). One photon from each pair is directed to a Bell state measurement (BSM). Upon success, the BSM establishes entanglement of the remaining photons sent to Alice and Bob.[]{data-label="fig1"}](fig1.pdf){width="100.00000%"} ![image](fig2.pdf){width="100.00000%"} The experimental concept is sketched in Fig. \[fig1\]b. Two pairs of polarization-entangled photons are consecutively emitted (emission 1 and 2) by a single semiconductor quantum dot. The polarization of one photon from each pair is measured by separate detectors, labeled Alice and Bob. Then, a joint Bell state measurement (BSM) is made on the remaining two photons; this swaps the entanglement of the original pairs to the photons that Alice and Bob receive. The source of entangled photons in our experiment are GaAs/AlGaAs QDs grown by local droplet etching, as they are reliable and reproducable to make with entanglement fidelities close to unity [@Keil2017] and highly indisitinguishable photons [@Huber2017]. The QDs are embedded in a nanomembrane which is sandwiched by a silver reflector and a spacing layer, attached to a gallium phosphide hemsipherical lens [@Chen2018]. This design provides a broadband optical antenna, offering photon extraction efficiencies up to 65% while preserving a high single photon purity and entanglement fidelity.\ The QD antenna’s operating temperature of $T=4$K is reached using a closed-cycle helium cryostat. A selected QD is first triggered by optically pumping the surrounding host semiconductor material. The emission spectrum in Fig. \[fig2\]b displays two prominent features: the exciton (X) emission at $780.0$nm and the bi-exciton (XX) emission at $781.6$nm. Here, the X photons reside near the optical $\mathrm{D_2}$ transitions of rubidium, a prominent quantum memory candidate [@Koerber2018]. To generate two consecutive polarization-entangled photon pairs (emission 1 and 2), we exploit the bi-exciton(XX)-exciton(X) radiative cascade [@Akopian2006]. Deterministic excitation of the XX state is ensured by resonant two-photon excitation [@MullerM2014]. A pair of photons is emitted in the successive decay via the intermediate X states to the ground state (left inset of Fig. \[fig2\]c). The photons share the polarization-entangled Bell state $\ket{\Phi^+}_{i}$ in the respective emission $i=1,2$: $$\ket{\Phi^+}_{i} = \ket{H_XH_{XX}} + \ket{V_XV_{XX}}.$$ with $H$ and $V$ representing horizontal and vertical polarization of the rectilinear basis. For efficient resonant excitation we use a pulsed Ti:sapphire laser operating at a $76$MHz repetition rate. Guiding the laser light into a tunable, unbalanced Mach-Zehnder interferometer yields two consecutive excitation pulses. The laser’s spectral width is reduced and the central wavelength adjusted by a successive diffraction grating and single-mode fiber. Thus the laser emission wavelength is fixed at the XX two-photon resonance between the X and XX emission. Notch filters are used to suppress the scattered laser in the QD emission signal. The signal intensity is enhanced further by exciting the QD by a weak continuous wave laser emitting at 650nm. Fig. \[fig2\]c shows the resulting emission spectrum of the XX cascade emission and a well-suppressed resonant laser. The right inset shows the intensity auto-correlation $g^{(2)}(\tau)$ of the X and XX emissions obtained in a Hanbury Brown and Twiss measurement [@HBT1956]. Vanishing coincidences at zero delay time bear witness to a high single photon purity, with values of $g^{(2)}_{X}(0)=0.0041 \pm 0.0003$ and $g^{(2)}_{XX}(0)=0.0050 \pm 0.0005$. We attribute the residual coincidences at zero delay to laser background, which can in principle be suppressed further using additional notch filters. So far the two photon pairs share the same light path upon emission. As laid out in Fig. \[fig2\]a, a non-polarizing beam splitter is used together with time-gated detection in order to separate emissions 1 and 2. The XX and X photons from each pair are split apart using dichroic optical filters, which transmit XX photons and reflect the X emission. At this stage, the four-photon state $\ket{\alpha}$ is a product of the states from emissions 1 and 2. It can be rewritten into products of Bell states between the X and XX photons: $$\begin{split} \ket{\alpha} &= \ket{\Phi^+}_1 \ket{\Phi^+}_2\\ &= \frac{1}{2}( \ket{\Phi^+}_{X} \ket{\Phi^+}_{XX} + \ket{\Phi^-}_{X} \ket{\Phi^-}_{XX}\\ &\ \ \ \ \ \ \ + \ket{\Psi^+}_{X} \ket{\Psi^+}_{XX} + \ket{\Psi^-}_{X} \ket{\Psi^-}_{XX} ) \end{split}$$ with the four polarization Bell states being $$\begin{split} \ket{\Phi^{\pm}} &= \ket{HH} \pm \ket{VV}\\ \ket{\Psi^{\pm}} &= \ket{HV} \pm \ket{VH} \end{split}$$ Projecting $\ket{\alpha}$ to a Bell state between photons XX1 and XX2 will in turn result in a Bell state shared by the previously uncorrelated X1 and X2. We project to the state $\ket{\Psi^{+}}$ by performing the following BSM: First, photons XX1 and XX2 are sent to interfere on a non-polarizing beam splitter. To ensure successful quantum interference, the arrival times of XX1 and XX2 have to be matched. Therefore the XX1 photons are delayed before the BSM, in order to compensate for the time difference between emission 1 and 2. After interference, the photons pass through an H- or V-oriented polarizer in each beam splitter output, respectively. Single-mode fibers then deliver the photons to superconducting nanowire single photon detectors (SNSPDs) with time resolutions of $50$ps.\ Successful coincidence detection at the BSM now leaves the two remaining photons X1 and X2 in the Bell state $$\ket{\Phi^{+}}_{AB} = \ket{HV} + \ket{VH}$$ sent to Alice and Bob for measurement. Subsequent arrangement of a quarter-wave plate, half-wave plate, polarizer and SNSPD allows for projection on any desired polarization state. In order to compensate for an accumulated phase and retardation in the setup we employ liquid crystal retarders and tilted quarter wave plates. Initial state characterization ============================== ![Degree of entanglement and photon indistinguishability. Two-photon density matrices of the photon pairs Xi–XXi from (a) emission i=1 and (b) emission i=2. Real (left) and imaginary part (right) closely resemble the Bell state $\ket{\Phi^+}$ with fidelities of $f_1=0.9369\pm0.0004$ (emission 1) and $f_2=0.9267\pm0.0004$ (emission 2). The shaded areas represent the difference to the ideally obtainable values. (c) The indistinguishability $I=0.569 \pm 0.009$ of photons XX1 and XX2 is derived from a Hong-Ou-Mandel measurement. Coincidences are detected at the output of an unbalanced Mach-Zehnder interferometer (inset) and binned according to their detection delay times. Using a half-wave plate (HWP), co-polarized photons yield reduced coincidences (red) compared with crossed polarizations (black).[]{data-label="fig3"}](fig3.pdf){width="100.00000%"} Successful entanglement swapping relies on high entanglement fidelities $f_i$ of the initial photon pairs (emission $i=1,2$) and on high photon indistinguishabilities $I$ of the XX photons sent to the BSM. We perform quantum state tomography [@James2001] to reconstruct the full two-photon density matrix $\rho_i$ of emissions $i=1,2$ as shown in Fig. \[fig3\]a and Fig. \[fig3\]b, respectively. The real (left) and imaginary parts (right) clearly resemble the Bell state $\ket{\Phi^+}$. We obtain fidelities of $f_1=0.9369\pm0.0004$ (emission 1) and $f_2=0.9267\pm0.0004$ (emission 2) indicating highly entangled photon emission. We attribute the deviation from unity fidelity to the slightly polarizing optical filters in the setup and the minor presence of a QD emission in the red-shifted vicinity of the XX photons (see Fig. \[fig2\]c). The marginally lifted exciton spin degeneracy by $(0.4\pm0.1)\,\mathrm\mu$eV, evanescent laser light background and polarization-dephasing during the QD’s emission process are expected to have only a small effect on the fidelity [@Keil2017].\ ![image](fig4.pdf){width="100.00000%"} Fig. \[fig3\]c shows a coincidence histogram obtained in an indistinguishability measurement [@Santori2002] based on Hong-Ou-Mandel interference [@Hong1987]. The two consecutive XX photons are guided into an unbalanced Mach-Zehnder interferometer featuring a time delay identical to that between XX1 and XX2. Indistinguishable XX photons will interfere at the second beam splitter and exit it pairwise, observable in the detection of reduced photon coincidences. Using a half-wave plate (HWP), the photon polarizations at the beam splitter can be made orthogonal. This renders the photons distinguishable which in turn gives rise to coincidences. At zero delay time between the detection events, the coincidences for parallel polarizations (red) show a significant reduction in comparison with those for perpendicular polarizations (black). We extract photon indistinguishabilities of $I=0.569 \pm 0.009$, which directly specifies the success probability of the BSM in the entanglement swapping experiment. The offset from unity arises most likely from internal dephasing processes and spectral jittering. Further spectral filtering or time-gating of detection events in the BSM can circumvent these effects at the expense of the BSM coincidence rate. Entanglement Swapping ===================== Having ensured high-fidelity emission of entangled photons we can focus on the execution of the entanglement swapping experiment. As a control measurement, the photon state shared by Alice and Bob is first investigated without considering the BSM. The density matrix $\rho_{mix}$ extracted from our observations via quantum state tomography is shown in Fig. \[fig4\]a. The signature of a statistical mixture of polarization states is evident, with a fidelity of $f_{mix}=0.9960 \pm 0.0004$ to the completely mixed state $\frac{1}{4}\mathbb{1}$. This is expected, since the photons X1 and X2 do not stem from the same emission cascade.\ Now the entanglement shall be swapped from the initial photon pairs to the photons received by Alice and Bob, as established by coincidences at the BSM. Each SNSPD in the setup detects approx. 0.5million QD photons per second. To increase the entanglement swapping fidelity we use time gating of BSM detection events (gate width: $47$ps) at the expense of the total rate of heralding events. Quantum state tomography is performed using sets of four-fold coincidences at different polarization settings for Alice and Bob. The determined density matrix shown in Fig. \[fig4\] closely resembles the Bell state $\ket{\Psi^+}$. The fidelity of $f_{AB}=0.81\pm0.04$ clearly surpasses the classical limit of $0.5$ and therefore testifies to the successful swapping of the entangled state. Fig. \[fig4\]c features the measurement of four-fold coincidences in the co- and cross-polarized diagonal bases as a function of the relative time delay between emission 1 and 2. In this fashion the temporal overlap of the XX photons at the beam splitter in the BSM is tuned. The highest XX photon indistinguishability is found at zero delay, resulting in a distinct coincidence offset for co- and cross-polarized bases. As the delay time departs from zero the BSM success starts to drop. This results in assimilating four-fold coincidences. The data, obtained without time gating at the BSM, can be well fitted to double-sided exponential functions denoted as solid lines. The main bottleneck in reaching higher $f_{AB}$ is the XX photon indistinguishability. This can also be seen from $Re(\rho_{AB})$ in Fig. \[fig4\]b: The well-fitting diagonal elements are mainly determined by the high initial fidelities $f_1$ and $f_2$. However, the more deviant off-diagonal elements depend on both the degree of entanglement and the XX photon indistinguishability. In Fig. \[fig4\]d the fidelity $f_{AB}$ and the Bell parameter $S$, as used in the CHSH-Bell inequality [@Bell1964; @CHSH1969], are plotted against the temporal gate width. For large gate widths the fidelity decreases to $f_{AB} = 0.71 \pm 0.03$. This is in perfect agreement with the calculated fidelity of $f_{AB} = 0.71$, given the observed entanglement fidelities $f_i$ and XX photon indistinguishability $I$ discussed above. The maximum achievable fidelity for our QD emission is $f_{max} = 0.89$, assuming unity indistinguishability (a gate width approaching zero). In reality this value cannot be approached due to the limited time resolution of the detectors. The Bell parameter $S=2.28 \pm 0.13$ at the $47$ps gate violates the CHSH-Bell inequality, $S\leq 2$, by more than two standard deviations. Assuming perfect indistinguishability it reaches $S_{max}=2.47$.\ In a final step we investigate the compatibility of our semiconductor entangled photon source with atomic transitions of rubidium, a prominent quantum memory candidate. Maintaining entangled photon emission, the emission frequency is tuned over the Rb $D_2$ transitions at $780.04$ nm by controlling the QD temperature [@Ding2016]. Fig. \[fig4\]e displays the $^{87}$Rb vapor cell transmission against the relative frequency detuning of a spectrally narrow laser (black). Two prominent absorption features are observed corresponding to the two $^{87}$Rb ground states split by the hyperfine interaction [@Penselin1962]. Residual $^{85}$Rb in the vapor cell results in the smaller absorption features visible at detunings of $-1$GHz and $2$GHz. The transmission of the QD photons (blue) shows two clear absorption dips, which are broadened due to the QD linewidth of $\Delta\nu = (4.9 \pm 0.2)$GHz. This opens the door for further experiments involving the storage of polarization-encoded qubits in atomic quantum memories. In addition, Rb atomic transitions could serve as a common and global reference at which the QD emission can be frequency-stabilized [@Zopf2018]. Thus the indistinguishability of photons from distant nodes in a quantum network could be ensured. Discussion and Outlook ====================== Demonstrating entanglement swapping between photon pairs emitted from semiconductor QDs marks a milestone for quantum photonics, since these sources surpass existing technologies in terms of on-demand photon emission and scalability. Compatibility with atom-based quantum memories paves the way for hybrid quantum repeater implementations. An efficient photon-atom interface requires the linewidths of both systems to be matched, e.g. by combining lifetime-limited QD emission and Purcell broadening of atomic lines [@Gallego2018].\ Further experiments that now become feasible with these sources are entanglement swapping with photons from distant emitters, multi-photon entanglement or entanglement distillation. The outcome will be dictated by the optical quality of these sources. Promising improvements include silicon-integrated strain-tuning platforms, which facilitates the emission of wavelength-tunable entangled photons [@Chen2016]. Integrating QDs into micro-cavities can increase their brightness and photon indistinguishability [@Senellart2017; @Sapienza2015]. Another key ingredient towards a scalable quantum photonic network is electrically triggered photon emission [@Zhang2015a; @Mueller2018]. Decoherence due to coupling to the solid-state environment can be controlled by electric fields in QD integrated diode structures [@Kuhlmann2013]. Overcoming the challenge of combinig these techniques in fabricated devices will be a next major step in realizing semiconductor based quantum networks. Acknowledgments =============== We acknowledge funding by the BMBF (Q.com) and the European Research Council (QD-NOMS). F.D. acknowledges support by IFW Excellence Program. We thank Wenjamin Rosenfeld (LMU München), Tobias Macha (Universität Bonn), Matthew Eiles (MPIPKS Dresden), and Franz Löchner (FSU Jena) for fruitful discussions. Author contributions ==================== The experiment was conceived by F.D. and O.G.S. who directed the research. The quantum dot samples were grown by R.K and integrated with the broadband optical antenna by Y.C. The optical setup was built and tested by M.Z., J.Y. and D.C. Optical measurements and data analysis were performed by M.Z. with help from J.Z. and R.K. The manuscript was written by M.Z., with input from all the authors. [43]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\ 12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty [**, ()](\doibase 10.1103/PhysRevLett.9.366) [, ()](\doibase 10.1109/PROC.1963.2706) [, ()](\doibase 10.1063/1.1754164) [, ()](\doibase 10.1143/JJAP.18.2329) [, ()](\doibase 10.1038/nphoton.2016.23) [, ()](\doibase 10.1038/nature09078) [, ()](\doibase 10.1103/RevModPhys.74.145) [, ()](\doibase 10.1080/00107510802091298) [, ()](\doibase 10.1038/35051009) [, ()](\doibase 10.1038/nature07127) [, ()](\doibase 10.1103/PhysRevA.57.822) [, ()](\doibase 10.1103/PhysRevLett.81.5932) [, ()](\doibase 10.1103/PhysRevLett.70.1895) [, ()](\doibase 10.1103/PhysRevLett.71.4287) [, ()](\doibase 10.1103/PhysRevLett.80.3891) [, ()](\doibase 10.1103/PhysRevLett.75.4337) [, ()](\doibase 10.1103/PhysRevLett.61.2921) [, ()](\doibase 10.1103/PhysRevLett.117.210502) [, ()](\doibase 10.1140/epjd/e2004-00170-7) [, ()](\doibase 10.1103/PhysRevLett.96.130501) [, ()](\doibase 10.1103/PhysRevB.92.245439) [, ()](http://dx.doi.org/10.1038/ncomms15501) [, ()](\doibase 10.1038/ncomms15506) [, ()](\doibase 10.1038/s41467-018-05456-2) [, ()](\doibase 10.1103/PhysicsPhysiqueFizika.1.195) [, ()](\doibase 10.1103/PhysRevLett.23.880) [, ()](\doibase 10.1103/PhysRevA.88.053834) [, ()](https://doi.org/10.1038/s41566-017-0050-y) [, ()](http://dx.doi.org/10.1038/nphoton.2013.377) [, ()](\doibase 10.1038/1781046a0) [, ()](\doibase 10.1103/PhysRevA.64.052312) [, ()](\doibase 10.1038/nature01086) [, ()](\doibase 10.1103/PhysRevLett.59.2044) [, ()](\doibase 10.1103/PhysRevLett.116.020401) [, ()](\doibase 10.1103/PhysRev.127.524) [, ()](\doibase 10.1103/PhysRevB.98.161302) [, ()](\doibase 10.1103/PhysRevLett.121.173603) [, ()](\doibase 10.1038/ncomms10387) [, ()](\doibase 10.1038/nnano.2017.218) [, ()](\doibase 10.1038/ncomms8833) [, ()](\doibase 10.1038/ncomms10067) [, ()](\doibase 10.1038/s41467-018-03251-7) [**, ()](\doibase 10.1038/nphys2688)	{ "pile_set_name": "ArXiv" }
--- abstract: 'We study the fundamental limitations of cooling to absolute zero for a qubit, interacting with a single mode of the electromagnetic field. Our results show that the dynamical Casimir effect, which is unavoidable in any finite-time thermodynamic cycle, forbids the attainability of the absolute zero of temperature, even in the limit of an infinite number of cycles.' author: - Giuliano Benenti - Giuliano Strini title: Dynamical Casimir effect and minimal temperature in quantum thermodynamics --- Introduction. Due to recent progress of nanofabrication technology, quantum effects in small heat engines have become an increasingly important subject. Concepts from quantum thermodynamics [@mahlerbook] have been applied to investigate questions such as the optimization of quantum thermal machines [@benenti2013], the fundamental dimensional limits to thermodynamic machines [@popescu], and the minimum temperature achievable in nanoscopic chillers [@kosloffEPL2009; @kosloffEPL2010; @eisert2012; @vandenbroeck2012; @kurizki2012; @kosloff2012; @kosloff2012b; @kosloff2013]. Cooling a system to the absolute zero of temperature ($T=0$) is prohibited by Nernst’s unattainability principle [@nerst], also known as the dynamical formulation of the third law of thermodynamics. Such principle states that it is impossible by any procedure to reduce any system to $T=0$ in finite time. Nernst’s principle has been recently challenged [@kurizki2012]. It this Letter, we investigate the unattainability principle in a minimal model: a qubit coupled to a single mode of the electromagnetic field, i.e. a harmonic oscillator. The oscillator is the working medium, shuttling heat from the qubit to a hot reservoir by means of a (quantum) Otto cycle. Although oversimplified, our model has several relevant features: - The matter-field interaction is treated at the fundamental level of quantum electrodynamics. - The equations of motion are solved by accurate numerical simulations, without resorting to the approximations necessarily involved in the master equations often used in the literature; - We do not use the Rotating Wave Approximations (RWA), that neglects the effects of rapidly rotating terms in the equations of motion [@micromaser]. The latter point is particularly relevant where addressing the fundamental limits to cooling, since the terms beyond the RWA lead to the generation of photons from the vacuum due to time-dependent boundary conditions for the electromagnetic field. Such quantum vacuum amplification effect, known as the Dynamical Casimir Effect (DCE) [@moore; @dodonov; @noriRMP], has been observed in recent experiments with superconducting circuits [@norinature; @lahteenmaki], and also investigated in the context of Bose-Einstein condensates [@jaskula], in excition-polariton condensates [@koghee], for multipartite entanglement generation in cavity networks [@solano2014] and for quantum communication protocols [@casimirqip]. Here, we show that the DCE is a fundamental, purely quantum, limitation to cooling. We point out that the DCE is unavoidable in the context of finite-time thermodynamics [@andresen] for cyclic quantum cooling machines, where the field’s boundary conditions are effectively changed in time due to the switching on/off of the matter-field coupling [@footnote]. In what follows, we show that, due to the DCE, the T=0 state of the qubit cannot be attained, even in the limit of infinite number of cooling cycles. The model. We consider a reciprocating refrigerator, operating by means of a working medium \[a single mode of the electromagnetic field, that is, a harmonic oscillator, with a time-dependent frequency $\omega(t)$\], shuttling heat from a cold finite-size “bath” (a single qubit) to a hot bath. The working medium undergoes a four-stroke Otto cycle: - Isochore A$\to$B: the working medium is in contact with the cold bath at temperature $T_c$; the work parameter, i.e. the oscillator frequency is maintained constant, $\omega(t)=\omega_c$. - Adiabatic compression B$\to$C: the frequency $\omega(t)$ of the working medium changes in time from $\omega_c$ to $\omega_h$. - Isochore C$\to$ D: the working medium is in contact with the hot bath at temperature $T_h$; the oscillator frequency is constant, $\omega(t)=\omega_h$. - Adiabatic expansion D$\to$A: the frequency $\omega(t)$ changes from $\omega_h$ to $\omega_c$. In order to elucidate the limitations imposed by the DCE on the lowest temperature that could be reached by a cooling protocol, we first consider the isochore stroke A$\to$B, with both the qubit and the oscillator prepared in their ground state. We wish to emphasize here that, even starting from these ideal conditions, due to the DCE both the oscillator and the qubit are excited, so that at the end of the isochore stroke the qubit is left in a state at a nonzero temperature. During the stroke A$\to$B the qubit-oscillator interaction is described by the time-dependent Rabi Hamiltonian [@micromaser] (we set the reduced Planck’s constant $\hbar=1$): $$\begin{array}{c} {\displaystyle H(t)=H_0+H_I(t), } \\ \\ {\displaystyle H_0=-\frac{1}{2}\,\omega_a \sigma_z + \omega\left(a^\dagger a +\frac{1}{2}\right), } \\ \\ {\displaystyle H_I(t)=f(t)\,\left[g \,\sigma_+\,(a^\dagger+a) +g^\star \sigma_-\,(a^\dagger+a)\right], } \end{array} \label{eq:noREWAquantum}$$ where $\sigma_i$ ($i=x,y,z$) are the Pauli matrices, $\sigma_\pm = \frac{1}{2}\,(\sigma_x\mp i \sigma_y)$ are the rising and lowering operators for the qubit: $\sigma_+ \|g\rangle = \|e\rangle$, $\sigma_+ \|e\rangle = 0$, $\sigma_- \|g\rangle = 0$, $\sigma_- \|e\rangle = \|g\rangle$; the operators $a^\dagger$ and $a$ for the field create and annihilate a photon: $a^\dagger \|n\rangle=\sqrt{n+1}\|n+1\rangle$, $a \|n\rangle=\sqrt{n}\|n-1\rangle$, $\|n\rangle$ being the Fock state with $n$ photons. We first assume sudden switch on/off of the coupling: $f(t)=1$ for $0\le t \le \tau$, $f(t)=0$ otherwise, with $\tau$ duration of the A$\to$B stroke. For simplicity’s sake, we consider the resonant case ($\omega=\omega_a$) and the coupling strength $g\in\mathbb{R}$. The RWA (exact only in the limit $g\to0$) is obtained when we neglect the term $\sigma_+ a^\dagger$, which simultaneously excites the qubit and creates a photon, and $\sigma_- a$, which de-excites the qubit and annihilates a photon. In this limit, Hamiltonian (\[eq:noREWAquantum\]) reduces to the Jaynes-Cummings Hamiltonian [@micromaser]. We set $\omega=1$, so that in the RWA the swap time needed to transfer an excitation from the qubit to the field or vice versa ($\|e\rangle \|0\rangle\leftrightarrow \|g\rangle \|1\rangle$) is $\tau_S=\pi/2g$. Results. If the qubit is prepared in its ground state, $\rho_q(0)=\|g\rangle\langle g\|$ and the oscillator is in the vacuum state, $\rho_o(0)=\|0\rangle\langle 0\|$, then, as we will discuss below, the qubit remains in a diagonal state in the basis of its eigenstates, $\rho_q(t)= p(t) \|g\rangle\langle g\|+ [1-p(t)] \|e\rangle\langle e\|$. Therefore, the qubit’s temperature reads as follows: $T(t)=\omega/\ln \{p(t)/[1-p(t)]\}$ (we set the Boltzmann constant $k_B=1$). We found more convenient for visualization to plot, rather than the temperature, the $z$-coordinate of the Bloch vector [@qcbook] ${\bf r}(t)=[x(t),y(t),z(t)]$ of the state $\rho_{q}(t)=\frac{1}{2}\left(I+{\bf r}(t)\cdot \bm{\sigma}\right)$, with $I$ identity operator and $\bm{\sigma}=(\sigma_x,\sigma_y,\sigma_z)$. In Fig. \[fig:gg\], we show $z(\tau)$ at the end of the isochore transformation A$\to$B, as a function of $\tau$ and of the qubit-oscillator coupling strength $g$. Within the RWA the initial tensor-product state $\rho_q(0)\otimes \rho_o(0)$ is the ground state of the overall qubit-oscillator system. Hence, $z(\tau)=1$, that is, the temperature $T(\tau)=0$ for any value of $\tau$. On the other hand, due to the DCE for any value of $g$ there exists a finite probability to generate photons and to excite the qubit, so that $z(\tau)<1$, namely the temperature $T(\tau)$ of the qubit is nonzero. ![(color online) Bloch coordinate $z$ of the qubit after the A$\to$B isochore, as a function of the qubit-oscillator interaction time $\tau$ (in units of the swap time $\tau_S$) and the interaction strength $g$.[]{data-label="fig:gg"}](fig1.eps){width="8cm"} Since we are interested in the fundamental limitations to cooling imposed by the DCE, for the remaining part of the Otto cycle we ideally consider the most favorable conditions for cooling. That is, we assume that the adiabatic transformations can be performed without friction in a finite time by utilizing “shortcuts to adiabaticity” [@torrontegui; @paternostro]. Furthermore, we assume that the overall cycle is engineered in such a way that the oscillator is left in its vacuum state at the end of the cycle. While a careful treatment of the Otto cycle should be performed to evaluate the cooling power of a refrigerator [@kosloffEPL2009; @kosloffEPL2010], our analysis, based on the most favorable instance, is sufficient to investigate the limitations set by the DCE to the lowest attainable temperature in finite-time thermodynamic cycles. With the above assumptions, the state of the qubit after $n$ Otto cycles is given by $$\rho_{q,n}={\rm Tr}_o\,[U(\rho_{q,n-1}\otimes \|0\rangle\langle 0)\|U^\dagger], \label{eq:rhoqn}$$ with $U$ the unitary time evolution operator describing the A$\to$B evolution for the qubit and the oscillator and $\rho_{q,0}\equiv \rho_q(0)$ initial state of the qubit. The quantum channel $\mathcal{E}$ mapping $\rho_{q,n-1}$ into $\rho_{q,n}$ can be conveniently described in the Fano-Bloch representation [@fano; @eberly; @mahler; @striniqpt; @strinidiamondnorm; @casimirqip]. If ${\bf r}_n=(x_n,y_n,z_n)$ denotes the Bloch vector of the state $\rho_{q,n}$, it follows from the linearity of quantum mechanics that ${\bf r}_{n-1}$ and ${\bf r}_{n}$ are connected through an affine map $\mathcal{M}$: $$\left[ \begin{array}{c} {\bf r}_{n} \\ \hline 1 \end{array} \right] = \mathcal{M} \left[ \begin{array}{c} {\bf r}_{n-1} \\ \hline 1 \end{array} \right] = \left[ \begin{array}{ccc} {\bf M} & \Big \lvert & {\bf a} \\ \hline {\bf 0}^T & \Big \lvert & 1 \end{array} \right] \left[ \begin{array}{c} {\bf r}_{n-1} \\ \hline 1 \end{array} \right], \label{eq:affine}$$ where ${\bf M}$ is a $3\times 3$ real matrix, ${\bf r}_{n-1}$, ${\bf r}_{n}$ and ${\bf a}$ real column vectors of dimension $3$ and ${\bf 0}$ the null vector of the same dimension. The Fano-Bloch representation of quantum operations is also physically transparent since the Bloch vector directly provides the expectation values of polarization measurements. While in general an affine map for a qubit depends on twelve parameter [@qcbook], we found from the numerical simulation of the quantum map $\mathcal{E}$ the following structure of ${\bf M}$ and ${\bf a}$: $${\bf M} = \left( \begin{array}{ccc} m_{xx} & m_{xy} & 0 \\ m_{yx} & m_{yy} & 0 \\ 0 & 0 & m_{zz} \\ \end{array} \right), \qquad {\bf a} = \left( \begin{array}{c} 0 \\ 0 \\ a_z \\ \end{array} \right). \label{eq:Memoryless-Kraus-Operators}$$ It follows that, whatever the initial state $\rho_{q,0}$ is, for $n\to\infty$ we have $x_n,y_n\to 0$ and $z_n\to a_z/(1-m_{zz})$. Therefore the asymptotic state of the qubit, $\rho_{\infty}=\frac{1}{2}\,(1+z_{\infty})\|g\rangle\langle g\|+ \frac{1}{2}\,(1-z_{\infty})\|e\rangle\langle e\|$, is diagonal and we can readily derive the asymptotic temperature $T_\infty$. In the particular case in which the state is diagonal from the beginning, then, as observed above, it remains diagonal and the temperature $T_n$ can be computed as a function of the number $n$ of Otto cycles. Examples of the evolution of the Bloch ball coordinates $x_n,y_n$ and $z_n$ are shown in Fig. \[fig:bloch\], for different initial states: the qubit’s ground state $\|g\rangle\langle g\|$, the maximally mixed state $\frac{1}{2}\,I$, a thermal state $p\|g\rangle\langle g\|+ (1-p)\|e\rangle\langle e\|$, with $p=0.6$, and the superposition state $\frac{1}{\sqrt{2}}\,(\|g\rangle+\|e\rangle)$. As expected, the $z$-coordinate converges to a limiting value depending only on the channel’s control parameters (here, $g=0.5$ and $\tau=\pi/2g$) but not on the initial state of the qubit. For the superposition state we also show the asymptotic decay of $x$ and $y$ (in the other cases $x=y=0$). ![(color online) Evolution of the Bloch ball coordinates as a function of the number of Otto cycles. Top: $z$-coordinate for the initial state of the qubit $\|g\rangle\langle g\|$ (black full curve), $\frac{1}{2}\,I$ (red dashed curve), $p\|g\rangle\langle g\|+ (1-p)\|e\rangle\langle e\|$, with $p=0.6$ (blue dotted curve). In the bottom panel we show, for the state $\frac{1}{\sqrt{2}}\,(\|g\rangle+\|e\rangle)$ the $x$- (black full curve) and the $y$-coordinate (red dashed curve). Note that for this state $z$ evolves exactly as for the maximally mixed input state. In the inset, for the same initial state, we show the exponential decay of $\|x\|$ and $\|y\|$.[]{data-label="fig:bloch"}](fig2.eps){width="8cm"} In Fig. \[fig:Tinfty\] we show the asymptotic value $z_\infty=\lim_{n\to\infty} z_n$ as a function of the time $\tau$ and of the interaction strength $g$. It is clear from this plot that in our model only in the limit $g\to 0$ the temperature asymptotically vanishes. In the same figure, it might appear at first sight surprising that for a given $g$ and $\tau\to 0$ we have $z_\infty\to 0^+$, namely $T_\infty\to +\infty$. In this limit, we have $a_z\to 0^+$ and $m_{zz}\to 1^-$, while the ratio $z_\infty=a_z/(1-m_{zz})\to 0^+$ [@footnote2]. Note that from the affine map (\[eq:affine\]) we have $z_{n}=m_{zz} z_{n-1}+ a_z$, and therefore a value of $m_{zz}$ close to $1$ implies a slow convergence to the asymptotic temperature. ![(color online) Asymptotic value $z_\infty$ of the $z$-coordinate for the qubit, as a function of the time $\tau$ and the interaction strength $g$.[]{data-label="fig:Tinfty"}](fig3.eps){width="8cm"} Finally, we investigate the effects of a smooth switch on/off of the interaction, by substituting in Eq. (\[eq:noREWAquantum\]) the rectangular window so far considered with the Hamming window: $f(t)=\frac{1}{\alpha}\,[1-\cos(2\pi t/\alpha\tau)]$ if $0\le t\le \alpha\tau$, $f(t)=0$ otherwise. In particular, we consider the values $\alpha=1$, for which the interaction time (equal to $\tau$) is the same as for the rectangular window and the peak value (equal to $2g$) is doubled, and $\alpha=2$, for which the interaction time ($2 \tau$) is doubled and the peak value ($g$) unchanged. In both cases, the area below the Hamming window is the same as for the rectangular window previously considered, so that such a window does not affect the cooling within the RWA, while significant differences appear in the ultra-strong coupling regime. In Fig. \[fig:window\] we consider the time $\tau=\tau_S=\pi/2g$, so that within the RWA the asymptotic temperature $T_\infty=0$ \[in the affine map (\[eq:affine\]), $m_{zz}=0$ and $a_z=1$\]. More precisely, the qubit achieves the zero-temperature state in a single Otto cycle, since for this value of $\tau$ and for the oscillator prepared in the vacuum state, the qubit and the oscillator swap their state, leaving the qubit in its ground state. On the other hand, if the terms beyond the RWA are taken into account, the asymptotic temperature changes with respect to the rectangular window. The oscillations of $T_\infty$ with $g$ are smoothed; the temperature is raised for $\alpha=1$ and lowered for $\alpha=2$. This result can be understood from the adiabatic theorem: for $\alpha=2$ the interaction is switched on/off more gradually than for $\alpha=1$ and therefore, if we start from an eigenstate of the unperturbed Hamiltonian $H_0$, for $\alpha=2$ we end up, after the $A\to B$ isochore, on the same eigenstate with higher probability than for the case $\alpha=1$. In particular, if we start from the $T=0$ state, for the Hamming window such state is exactly preserved only in the limit $\alpha\to\infty$, that is, the qubit-field interaction time $\tau_I=\alpha\tau\to\infty$. In this limit the DCE vanishes since it is an effect due to the change of the boundary conditions for the field when it interacts with the qubit and such change becomes infinitely slow. We should like to stress that the two limits of interaction time $\tau_I\to\infty$ and number of cycles $n\to\infty$ do not commute. If we let before $\tau_I\to\infty$, then we can achieve (in infinite time) the $T=0$ limit, as allowed from Nernst’s principle. If on the other hand we set a large as desired but finite value of $\tau_I$ and then let $n\to\infty$, then the $T=0$ limit is not attained. This latter case is the one relevant for cyclic quantum chillers in finite-time thermodynamics. ![(color online) Asymptotic value $z_\infty$ of the $z$-coordinate for the qubit, as a function of the interaction strength $g$, for the rectangular (black full curve) and the Hamming window for $\alpha=1$ (red dashed curve) or $\alpha=2$ (blue dotted curve). The interaction time $\tau=\tau_S=\pi/2g$.[]{data-label="fig:window"}](fig4.eps){width="8cm"} Discussion and conclusions. In this paper, we have investigated the limitations imposed by the DCE on the minimal temperature achievable in a cooling cycle. We have considered a minimal model, where the purpose is to cool a single qubit and the working medium is a single mode of the electromagnetic field, undergoing a (quantum) Otto cycle. Since we are interested in the minimum attainable temperature, we have considered the most favorable instance in which the field is reset to its vacuum state at the beginning of each cycle. Our model can be also interpreted as a collision model [@gisin2002; @gisin2002b; @buzek2005; @buzek2008; @palma2009; @palma2012; @palma2013] of irreversible quantum dynamics. Such kind of models were used in the literature to analyze the process of thermalization of a system in contact with a bath. In our case, the qubit undergoes a sequence of identical collisions, described by the unitary evolution operator $U$ of Eq. (\[eq:rhoqn\]), with the bath composed of an arbitrarily large number of oscillators initially in their ground state. Our results show that, due to the DCE, the qubit thermalizes to a nonzero temperature, different from the initial $T=0$ temperature of the bath of oscillators. We emphasize that our results, though obtained for a very simple model, are exact, in that the coupled matter-field equations are numerically integrated, without (i) the RWA approximation and (ii) the master equation approximation often used in the literature. Since the DCE is a generic feature of finite-time quantum electrodynamics and therefore also of finite-time quantum thermodynamics, we conjecture that the unattainability of the zero temperature limit, even in the limit of infinite number of cycles, would remain valid for any cyclic quantum cooling machine. At any rate, our results call for a deeper understanding of the relevance of the dynamical Casimir effect in quantum thermodynamics. G.B. acknowledges the support by MIUR-PRIN project “Collective quantum phenomena: From strongly correlated systems to quantum simulators”. [0]{} G. Gemma, M. Michel, and G. Mahler, G., Quantum thermodynamics (second Ed.) (Springer–Verlag, Berlin, 2009). G. Benenti, G. Casati, T. Prosen, and K. Saito, preprint arXiv:1311.4430. N. Linden, S. Popescu, and P. Skrzypczyk, P. Phys. Rev. Lett. 105, 130401 (2010). Y. Rezek, P. Salamon, K. H. Hoffmann, and R. Kosloff, Europhys. Lett. 85, 30008 (2009). T. Feldmann, and R. Kosloff, Europhys. Lett. 89, 20004 (2010). A. Mari and J. Eisert, Phys. Rev. Lett. 108, 120602 (2012). B. Cleuren, B. Rutten, and C. Van den Broeck, Phys. Rev. Lett. 108, 120603 (2012). M. Kolář, D. Gelbwaser–Klimovsky, R. Alicki, and G. Kurizki, Phys. Rev. Lett. 109, 090601 (2012). A. Levy and R. Kosloff, Phys. Rev. Lett. 108, 070604 (2012). A. Levy, R. Alicki, and R. Kosloff, Phys. Rev. E 85, 061126 (2012). E. Torrontegui and R. Kosloff, Phys. Rev. E 88, 032103 (2013). W. Nernst, The new heat theorem, its foundations in theory and experiment (Dutton, New York, 1926). P. Meystre and M. Sargent III, Elements of quantum optics (4th Ed.) (Springer–Verlag, Berlin, 2007). G. T. Moore, J. Math. Phys. (N.Y.) 11, 2679 (1970). V. V. Dodonov, Phys. Scripta 82, 038105 (2010). P. D. Nation, J. R. Johansson, M. P. Blencowe, and F. Nori, Rev. Mod. Phys. 84, 1 (2012). C. M. Wilson, G. Johansson, A. Pourkabirian, M. Simoen, J. R. Johansson, T. Duty, F. Nori, and P. Delsing, Nature (London) 479, 376 (2011). P. Lähteenmäki, G. S. Paraoanu, J. Hassel, and P. J. Hakonen, PNAS 110, 4234 (2013). J.-C. Jaskula, G. B. Partridge, M. Bonneau, R. Lopes, J. Ruaudel, D. Boiron, and C. I. Westbrook, Phys. Rev. Lett. 109, 220401 (2012). S. Koghee and M. Wouters, Phys. Rev. Lett. 112, 036406 (2014). S. Felicetti, M. Sanz, L. Lamata, G. Romero, G. Johansson, P. Delsing, and E. Solano, Phys. Rev. Lett. 113, 093602 (2014). G. Benenti, A. D’Arrigo, S. Siccardi, and G. Strini, Phys. Rev. A 90, 052313 (2014). B. Andresen, Angew. Chem. Int. Ed. 50, 2690 (2011). We will consider a model without moving boundaries, but with a time-dependent matter-field coupling $g(t)$. Since the coupling constant $g\propto 1/\sqrt{V}$, with $V$ the quantization volume for the field, changing $g$ in time is equivalent to changing the volume $V$, see Ref. [@casimirepjd]. G. Benenti, S. Siccardi, and G. Strini, Eur. Phys. J. D 68, 139 (2014). G. Benenti, G. Casati, and G. Strini, [Principles of Quantum Computation and Information]{}, Vol. I: Basic concepts (World Scientific, Singapore, 2004); Vol. II: Basic tools and special topics (World Scientific, Singapore, 2007). E. Torrontegui, S. Ibánez, S. Martínez-Garaot, M. Modugno, A. del Campo, D. Guéry-Odelin, A. Ruschhaupt, X. Chen, and J. G. Muga, Adv. At. Mol. Opt. Phys. 62, 117 (2013). A. del Campo, J. Goold, and M. Paternostro, Sci. Rep. 4, 6208 (2014). U. Fano, Rev. Mod. Phys. 29, 74 (1957); ibid. 55, 855 (1983). F. T. Hioe and J. H. Eberly, Phys. Rev. Lett. 47, 838 (1981). J. Schlienz and G. Mahler, Phys. Rev. A 52, 4396 (1995). G. Benenti and G. Strini, Phys. Rev. A 80, 022318 (2009). G. Benenti and G. Strini, J. Phys. B: At. Mol. Opt. Phys. 43, 215508 (2010). G. Benenti, S. Felloni, and G. Strini, Eur. Phys. J. D 38, 389 (2006). More precisely, from numerical data as well as from a second order (in the interaction Hamiltonian $H_I$) perturbative treatment of finite-time quantum electrodynamics, outlined in Ref. [@casimirepjd], we find that $$z_\infty=\frac{1}{6}\,(\omega\tau)^2\,\left[1+ \frac{9g^2}{4\omega^2}\right].$$ V. Scarani, M. Ziman, P. Štelmachovič, N. Gisin, and V. Bužek, Phys. Rev. Lett. 88, 097905 (2002). M. Ziman, P. Štelmachovič, V. Bužek, M. Hillery, V. Scarani, and N. Gisin, Phys. Rev. A 65, 042105, (2002). M. Ziman and V. Bužek, Phys. Rev. A 72, 022110, (2005). M. Koniorczyk, A. Varga, P. Rapčan, and V. Bužek, Phys.Rev A 77, 052106 (2008). G. Gennaro, G. Benenti, and G. M. Palma, Phys. Rev. A 79, 022105 (2009). V. Giovannetti and G. M. Palma, Phys. Rev. Lett. 108, 040401 (2012). F. Ciccarello, G. M. Palma, and V. Giovannetti, Phys. Rev. A 87, 040103(R) (2013).	{ "pile_set_name": "ArXiv" }
--- abstract: 'In this report a detailed derivation of the dynamical equations for an $n$–dimensional heterotic string theory of the Horowitz type is carried out in the string frame and in the Einstein frame too. In particular, the dynamical equations of the three dimensional string theory are explicitly given. The relation of the Horowitz–Welch and Horne–Horowitz string black hole solution is exhibited. The Chan–Mann charged dilaton solution is derived and the subclass of string solutions field is explicitly identified. The stationary generalization, via $SL(2,R)$ transformations, of the static (2+1) Horne–Horowitz string black hole solution is given.' author: - 'A.A. Garcia–Diaz' - 'G. Gutierrez–Cano' title: 'Low energy 2+1 string gravity; black hole solutions' --- -2cm $n$–dimensional heterotic string dynamical equations {#StringEqHOR} ==================================================== Following Horowitz [@Horowitz92], in this contribution we reproduce the field equations “for a part of the low energy action” to a $n$–dimensional heterotic string theory\[heteroticstring\] described by a metric $g_{\mu\nu}$, a scalar field $\Phi$, a Maxwell field $F_{\mu\nu}$, and a three–form $H_{\mu\nu\lambda}$. The three–form $\bm H$\[heteroticH\] is related to the two–form potential $\bm B$ and a gauge field $A_{\mu}$ through $\bm {H=dB} -a\bm{\,A\wedge dF},$ where $a$ is a constant to be adjusted at the end for final results. In this text to denote the number of dimensions is used $n$ instead of $D$. Moreover $\Lambda$ is reserved for the standard cosmological constant, whereas $\Lambda_{H}$, and $\Lambda_{CM}=-\Lambda$, denote the $\Lambda'$s used by Horowitz [@Horowitz92] and Chan and Mann [@Chan:1995wj] respectively.\ String frame ------------ The corresponding heterotic string action\[actionHor\] for dimension $n$, $S=\int{d^{\,n}x\mathfrak {L}},$ is given by $$\begin{aligned} \label{actionHor} S=\int{d^{\,n}x\sqrt{- g}{\rm e}^{-2\Phi}\left[{ R} -2 \Lambda+U(\Phi)+4 (\nabla\Phi)^2 -F^{2} -\frac{1}{12}H^{2}\right]}.\end{aligned}$$ Variations with respect to the metric\[Variations with\] give: \[actionHorEINS\] $$\begin{aligned} \frac{\delta \sqrt{- g}}{\delta { g}^{\mu\nu}} =-\frac{1}{2}\sqrt{- g}{ g}_{\mu\nu},\end{aligned}$$ $$\begin{aligned} \label{VariationRHor1} \frac{\delta \sqrt{- g}{ R}} {\delta { g}^{\mu\nu}}=\frac{\delta \sqrt{- g}{ g }^{\alpha\beta} { R}_{\alpha\beta}}{\delta { g}^{\mu\nu}}&& =\sqrt{- g}({ R}_{\mu\nu}-\frac{1}{2}{ g}_{\mu\nu}{ R}) +\sqrt{- g}{ g }^{\alpha\beta} \frac{\delta { R}_{\alpha\beta}}{\delta { g}^{\mu\nu}}\nonumber\\&& =\sqrt{- g}\,{ G}_{\mu\nu}+\sqrt{- g}{ g }^{\alpha\beta}\frac{\delta { R}_{\alpha\beta}}{\delta { g}^{\mu\nu}},\end{aligned}$$ $$\begin{aligned} \label{VariationRHor2} \frac{\delta }{\delta { g}^{\mu\nu}} \sqrt{- g}(\nabla\Phi)^2=\frac{\delta } {\delta { g}^{\mu\nu}}\sqrt{- g}{ g}^{\alpha\beta} \nabla_{\alpha}\,\Phi\,\nabla_{\beta}\,\Phi =\sqrt{- g}(\nabla_{\mu}\Phi\,\nabla_{\nu}\Phi -\frac{1}{2}{ g}_{\mu\nu} (\nabla\Phi)^2),\end{aligned}$$ $$\begin{aligned} \label{VariationRHor3} \frac{\delta }{\delta { g}^{\mu\nu}} \sqrt{- g}\,F^2=\frac{\delta }{\delta { g}^{\mu\nu}} \sqrt{- g}{{ g}^{\alpha\lambda}{ g}^{\beta\rho} F_{\alpha\beta}F_{\lambda\rho}}=2\, \sqrt{- g}({ g}^{\beta\rho} F_{\mu\beta}F_{\nu\rho} -\frac{1}{4}{ g}_{\mu\nu}F^2),\end{aligned}$$ $$\begin{aligned} \label{VariationRHor4} \frac{\delta}{\delta { g}^{\mu\nu}} \sqrt{- g}H^{2} =\frac{\delta }{\delta { g}^{\mu\nu}} \left(\sqrt{- g}{{ g}^{\alpha\sigma}{ g}^{\beta\rho} { g}^{\gamma\lambda}H_{\alpha\beta\gamma} H_{\sigma\rho\lambda}}\right)=3\,\sqrt{- g} (H_{\mu\alpha\beta}{H_{\nu}}^{\alpha\beta} -\frac{1}{6}{ g}_{\mu\nu}\,H^{2}).\end{aligned}$$ As far as to the term $\sqrt{- g}{ g }^{\alpha\beta} {\delta { R}_{\alpha\beta}}$ in Eq. (\[VariationRHor1\]) is concerned, it is easy to establish its proportionality to a divergence, ${\it v^\mu}_{;\mu}$: in standard textbooks one find the variation of the Riemann tensor in terms of the variation of the Christoffel symbols $$\begin{aligned} \delta{R^\lambda}_{\mu\kappa\nu}=(\delta{\Gamma^{\lambda}}_{\mu\nu})_{;\kappa}-(\delta {\Gamma^{\lambda}}_{\mu\kappa})_{;\nu}\end{aligned}$$ then, since ${ R}_{\mu\nu}= { R^\kappa}_{\mu\kappa\nu}$, one has $$\begin{aligned} && { g}^{\mu\nu}\delta{{ R}_{\mu\nu}}=({ g}^{\mu\lambda}\delta {{\Gamma}^{\nu}}_{\mu\lambda})_{;\nu}-({ g}^{\mu\nu} \delta{{ \Gamma}^{\lambda}}_{\mu\lambda})_{;\nu} =({ g}^{\mu\lambda}\delta {{\Gamma}^{\nu}}_{\mu\lambda} -{ g}^{\mu\nu}\delta{{\Gamma}^{\lambda}}_{\mu\lambda} )_{;\nu}=:{\it v^\nu}_{;\nu}= \nonumber\\ && =\frac{1}{\sqrt{-{ g}}} \left[\sqrt{-{ g}}\left({ g}^{\mu\lambda}\delta {{\Gamma}^{\nu}}_{\mu\lambda}-{ g}^{\mu \nu}\delta {{\Gamma}^{\lambda}}_{\mu\lambda}\right)\right]_{,\nu}.\end{aligned}$$ This term, taking into account the multiplicative factor $\exp{(-2\Phi)}$, will give rise in the Einstein’s equation to second order derivatives of $\Phi$ due first to the covariant derivative acting on $\delta{\Gamma}$ when forming a new divergence, and second to the covariant derivative acting on variations of the metric $\delta{ g}$ for $\delta{\Gamma}$ expressed in terms of $\delta{ g}$ according to [^1], $$\begin{aligned} \label{VarGammaVarmet} \delta{{\Gamma}^\mu}_{\alpha\beta} =\frac{1}{2}{ g}^{\mu\nu}\left(\nabla_{\alpha}\delta{ g}_{\nu\beta} +\nabla_{\beta}\delta{ g}_{\alpha\nu} -\nabla_{\nu}\delta{ g}_{\alpha\beta}\right),\end{aligned}$$ which implies $$\begin{aligned} { \it v^\nu}={ g}^{\nu\mu}{ g}^{\alpha\beta} \left(\nabla_{\alpha}\delta{ g}_{\mu\beta} -\nabla_{\mu}\delta{ g}_{\alpha\beta}\right) =[\delta{ g}_{\alpha\beta} ({ g}^{\nu\alpha}{ g}^{\mu\beta} -{ g}^{\nu\mu}{ g}^{\alpha\beta})]_{;\mu}, \end{aligned}$$ therefore, $$\begin{aligned} && e^{-2\Phi}\sqrt{-{ g}}{ g}^{\mu\nu} \delta{{ R}_{\mu\nu}}=e^{-2\Phi} \sqrt{-{ g}}{\it v^\nu}_{;\nu} \nonumber\\&&=e^{-2\Phi} \left(\sqrt{-{ g}}{\it v^\nu}\right)_{,\nu}=\left(e^{-2\Phi}\sqrt{-{ g}}{\it v^\nu}\right)_{,\nu}+2\,\Phi_{,\nu}e^{-2\Phi}\,\sqrt{-{ g}}{\it v^\nu},\end{aligned}$$ consequently $$\begin{aligned} \label{HORphicontribution} 2\Phi_{,\nu}&&e^{-2\Phi}\, \sqrt{-{ g}}\,{\it v^\nu} =2\sqrt{-{ g}}\left[e^{-2\Phi}\Phi_{,\nu}\delta{ g}_{\alpha\beta} \left({ g}^{\nu\alpha}{ g}^{\mu\beta} -{ g}^{\nu\mu}{ g}^{\alpha\beta}\right)\right]_{;\mu} \nonumber\\&&-2\left(e^{-2\Phi}\Phi_{,\nu} \right)_{;\mu}\sqrt{-{ g}}{ g}^{\nu\alpha} { g}^{\mu\beta}\delta{ g}_{\alpha\beta} +2\left(e^{-2\Phi}\Phi_{,\nu} \right)_{;\mu}\sqrt{-{ g}}{ g}^{\nu\mu} { g}^{\alpha\beta}\delta{ g}_{\alpha\beta}\nonumber\\&& =2\,\sqrt{-{ g}}{\rm {Div}} + 2\sqrt{-{ g}}\left[\left(e^{-2\Phi}\Phi_{,\nu} \right)_{;\mu}-{ g}_{\mu\nu}{ g}^{\alpha\beta}\left(e^{-2\Phi}\Phi_{,\alpha} \right)_{;\beta}\right]\delta{ g}^{\mu\nu},\end{aligned}$$ where it has been used $\delta g_{\mu\nu} =-g_{\mu\alpha}g_{\nu\beta}\delta g^{\alpha\beta}.$ Gathering all the divergence terms, by applying the Stokes theorem\[Stokes theorem\], the integral of the total derivative terms becomes an integral over the boundary; imposing conditions on the variations $\delta{ g}_{\alpha\nu}$ and on the covariant derivatives of these variations, $\nabla_{\mu}\delta{ g}_{\alpha\nu}$, one can avoid the contributions of the boundary terms. On this respect see Wald [@Wald:1984book], Appendix E, and also Ort[í]{}n [@Ortin:2004book], Part I, Chapter 4. The second term in the last line of Eq.(\[HORphicontribution\]) contributes to Einstein equations. Therefore $\frac{\delta \mathfrak {L} } {\delta { g}^{\mu\nu}}=0$ gives rise to Einstein equations in the form $$\begin{aligned} \label{HOREins1} &&{ G}_{\mu\nu}+\frac{1}{2}\, \left(2\Lambda-U(\Phi)\right){ g}_{\mu\nu}+ 2\,\Phi_{,\nu;\mu}-2{ g}_{\mu\nu}{ g}^{\alpha\beta} \Phi_{,\alpha;\beta}+ 2\,{ g}_{\mu\nu} { g}^{\alpha\beta}\,\Phi_{,\alpha}\,\Phi_{,\beta} \nonumber\\&& -2\,\left({ g}^{\beta\rho}F_{\mu\beta}F_{\nu\rho} -\frac{1}{4}{ g}_{\mu\nu}{ F}^2\right)-\frac{1}{4} \left(H_{\mu\alpha\beta}{H_{\nu}}^{\alpha\beta} -\frac{1}{6}{ g}_{\mu\nu}\,H^{2}\right)=0.\end{aligned}$$ The variation with respect to the scalar field $\Phi$, $\frac{\delta \mathfrak {L} }{\delta {\Phi}} =(\frac{\partial}{\partial \Phi} -\frac{\partial}{\partial x^\alpha} \frac{\partial}{\partial \Phi_{,\alpha}})\mathfrak {L} ,$ considering that $$\begin{aligned} &&\frac{\partial}{\partial x^\alpha}\left({\rm e}^{-2\Phi}\sqrt{- g} \frac{\partial}{\partial \Phi_{,\alpha}}({ g}^{\mu\nu} \Phi_{,\mu}\Phi_{,\nu})\right)={2} \frac{\partial}{\partial x^\alpha} \left({\rm e}^{-2\Phi}\sqrt{- g}\,{ g}^{\alpha\nu} \Phi_{,\nu}\right) \nonumber\\&& =2{\rm e}^{-2\Phi}\left(-2\sqrt{- g}\, \Phi_{,\alpha}\,{ g}^{\alpha\nu}\Phi_{,\nu} +\frac{\partial}{\partial x^\alpha} (\sqrt{- g}\,{ g}^{\alpha\nu}\Phi_{,\nu})\right) \nonumber\\&& =-4{\rm e}^{-2\Phi}\sqrt{- g}\, \Phi_{,\alpha}\,{ g}^{\alpha\nu} \Phi_{,\nu}+2{\rm e}^{-2\Phi} \sqrt{- g}({ g}^{\alpha\nu}\Phi_{;\nu})_{;\alpha},\end{aligned}$$ yields the dynamical equation $$\begin{aligned} \label{HOREins2} 4{ g}^{\alpha\nu}\Phi_{;\nu;\alpha} -4 { g}^{\mu\nu}\Phi_{,\mu}\Phi_{,\nu}+{ R} -2\Lambda-{ F}^{2}-\frac{1}{12}H^2 +U(\Phi)-\frac{1}{2}\frac{d\,U}{d\Phi}=0.\end{aligned}$$ The field equation for $B_{\mu\nu}$ arises from the variation of $H^2$, where $$H_{\mu\nu\lambda}=3\,B_{[\mu\nu,\lambda]} -a\,A_{[\mu}\,F_{\nu\lambda]} \rightarrow\frac{\partial H_{\alpha\beta\gamma}} {B_{\mu\nu,\sigma}}=3\,\delta^\mu_{[\alpha} \delta^\nu_{\beta}\delta^\sigma_{\gamma]},$$ consequently $$\begin{aligned} \label{HOREins3} \frac{\partial }{\partial B_{\mu\nu,\sigma}}H^2 =\frac{\partial H_{\alpha\beta\gamma}} {\partial B_{\mu\nu,\sigma}} \frac{\partial}{\partial H_{\alpha\beta\gamma}}{{ g}^{\tau\sigma} { g}^{\kappa\rho}{ g}^{\lambda\xi}H_{\tau\kappa\xi} H_{\sigma\rho\lambda}} =2\frac{\partial H_{\alpha\beta\gamma}}{B_{\mu\nu,\sigma}} H^{\alpha\beta\gamma}=3!H^{\mu\nu\sigma}\end{aligned}$$ therefore, for $\frac{\delta }{\delta { B_{\mu\nu}}}\mathfrak {L} =(\frac{\partial }{\partial { B_{\mu\nu}}} -\frac{\partial}{\partial{x^{\sigma}}} \frac{\partial }{\partial { B_{\mu\nu,\sigma}}}) \mathfrak {L},$ one gets $$\begin{aligned} \label{DynEDqforB} \frac{\delta }{\delta { B_{\mu\nu}}} \sqrt{- g}{\rm e}^{-2\Phi}H^2=-3!\frac{\partial}{\partial x^\sigma } \left(\sqrt{- g}{\rm e}^{-2\Phi}H^{\mu\nu\sigma}\right) \rightarrow\nabla_{\sigma}({\rm e}^{-2\Phi} H^{\mu\nu\sigma})=0.\end{aligned}$$ Finally, the electromagnetic Maxwell equations are derived for $A_{\alpha}$ related with the electromagnetic field through $F_{\mu\nu} =F_{\mu\nu}(A_{\alpha})=A_{\nu,\mu}-A_{\mu,\nu} =2\partial_{[\mu}A_{\nu]},$ $$\begin{aligned} \label{HOREinsMax} \frac{\delta}{\delta A_{\epsilon}}\mathfrak {L}&& =-(\frac{\partial}{\partial A_{\epsilon}} -\frac{\partial}{\partial x^\sigma} \frac{\partial}{\partial A_{\epsilon,\sigma}}) [\sqrt{- g}{\rm e}^{-2\Phi}(F^2+\frac{1}{12}H^2)] =-\frac{1}{12}\,\sqrt{- g}{\rm e}^{-2\Phi} \frac{\partial H_{\alpha\beta\gamma}} {\partial A_{\epsilon}}\frac{\partial}{\partial H_{\alpha\beta\gamma}}H^2\nonumber\\ && +\frac{\partial}{\partial x^\sigma} (\sqrt{- g}{\rm e}^{-2\Phi} \frac{\partial F_{\alpha\beta} }{\partial A_{\epsilon,\sigma}} \frac{\partial}{\partial F_{\alpha\beta}} F^2)+\frac{1}{12} \frac{\partial}{\partial x^\sigma}(\sqrt{- g}{\rm e}^{-2\Phi} \frac{\partial H_{\alpha\beta\gamma}}{\partial A_{\epsilon,\sigma}} \frac{\partial}{\partial H_{\alpha\beta\gamma}}H^2)\end{aligned}$$ taking into account that $$\begin{aligned} \label{HOREinsMax2} \frac{\partial}{\partial F_{\alpha\beta}} F^2 =2F^{\alpha\beta},\, \frac{\partial F_{\alpha\beta}}{\partial A_{\epsilon,\lambda}} =-2\delta^{\epsilon}_{[\alpha} \delta^{\lambda}_{\beta]},\,\frac{\partial}{\partial H_{\alpha\beta\gamma}}H^2 =2H^{\alpha\beta\gamma}\nonumber\\\end{aligned}$$ together with $$\frac{\partial}{\partial A_{\epsilon}}H_{\mu\nu\lambda} =-a\delta^\epsilon_{[\mu}\,F_{\nu\lambda]}$$ $$\frac{\partial}{\partial A_{\epsilon,\sigma}}H_{\alpha\beta\gamma} =-a\,\frac{\partial}{\partial A_{\epsilon,\sigma}}A_{[\alpha}\,F_{\beta\gamma]}= 2\,a\,(A_{[\alpha}\,\delta^\epsilon_{\beta}\delta^\sigma_{\gamma]})$$ one has for $\frac{\delta \mathfrak {L} }{\delta A_{\epsilon}}$ the expression $$\begin{aligned} \label{HOREinsMax3} \frac{\delta \mathfrak {L} }{\delta A_{\epsilon}} &&=-4\frac{\partial}{\partial x^\lambda} (\sqrt{- g}\,{\rm e}^{-2\Phi}{ F}^{\epsilon\lambda}) +a\frac{1}{6}\,\sqrt{- g}{\rm e}^{-2\Phi}F_{\nu\lambda}\, { H}^{\epsilon\nu\lambda} \nonumber\\&& +a\frac{1}{3}\frac{\partial}{\partial x^\sigma} \left(A_{[\alpha}\,\delta^\epsilon_{\beta}\delta^\sigma_{\gamma]} \sqrt{- g}{\rm e}^{-2\Phi}{ H}^{\alpha\beta\gamma}\right) \nonumber\\&&=4\,\sqrt{- g}\left({\rm e}^{-2\Phi}{ F}^{\lambda\epsilon}\right)_{;\lambda} +a\frac{1}{3}\sqrt{- g}{\rm e}^{-2\Phi}F_{\nu\lambda}\,{ H}^{\epsilon\nu\lambda}=0,\end{aligned}$$ where the equation (\[DynEDqforB\]) has been used.\ Summarizing, the field dynamical equations\[field dynam\] are \[einsteinHORdyn\] $$\begin{aligned} \label{einsteinHOR1} {G}_{\mu\nu}&&-\frac{1}{2}{ g}_{\mu\nu}\, (-2\Lambda+U)+2\nabla_{\mu}\nabla_{\nu}\Phi -2{ g}_{\mu\nu}{\nabla}^{2}\Phi +2{ g}_{\mu\nu} (\nabla\,\Phi)^2\nonumber\\&& -2\,(F_{\mu\alpha}F_{\nu\beta}{ g}^{\alpha\beta} -\frac{1}{4}{ g}_{\mu\nu}{ F}^2) -\frac{1}{4}(H_{\mu\alpha\beta}{H_{\nu}}^{\alpha\beta} -\frac{1}{6}{g}_{\mu\nu}{ H}^2)=0,\end{aligned}$$ $$\begin{aligned} \label{scalarHORscaS} 4\nabla^{2}\,\Phi -4(\nabla\,\Phi)^2+R-2\,\Lambda-{ F}^{2}-\frac{1}{6}H^2 +U(\Phi)-\frac{1}{2}\frac{d\,U}{d\,\Phi}=0,\end{aligned}$$ $$\begin{aligned} \label{BfieldHOR} \nabla_{\sigma}({\rm e}^{-2\Phi}H^{\mu\nu\sigma})=0,\end{aligned}$$ $$\begin{aligned} \label{AfieldHOR} \nabla_{\lambda}\left({\rm e}^{-2\Phi}{ F}^{\lambda\epsilon}\right) +a\frac{1}{12}{\rm e}^{-2\Phi}F_{\nu\lambda}\,{ H}^{\epsilon\nu\lambda}=0.\end{aligned}$$ The last equation differs in sign from the corresponding Horowitz’s equation (2.10b).\ By constructing (\[einsteinHOR1\]) one evaluates ${ R}$, $$\begin{aligned} \label{einsteinHORscal} &&R=\frac{2}{n-2}\left(n \Lambda-\frac{n}{2}U-2(n-1)\nabla^2\Phi +2n({\nabla}\Phi)^2 +\frac{n-4}{2}{ F}^2+\frac{n-6}{24}{ H}^2\right),\end{aligned}$$ which replaced again into (\[einsteinHOR1\]) and (\[scalarHORscaS\]) allows one to rewrite the set of dynamical equations as \[FieldEQstring\] $$\begin{aligned} \label{einsteinHORsumm} &&{R}_{\mu\nu}=-2\nabla_{\mu}\nabla_{\nu}\Phi+2F_{\mu\alpha}F_{\nu\beta}{ g}^{\alpha\beta }+\frac{1}{4}\,H_{\mu\alpha\beta}{H_{\nu}}^{\alpha\beta} \nonumber\\&&+ \frac{2}{n-2}{ g}_{\mu\nu}\,\left(\Lambda-\frac{1}{2}U-\nabla^2\Phi+2({\nabla}\Phi)^2 -\frac{1}{2}F^2-\frac{1}{12}{ H}^2\right),\end{aligned}$$ $$\begin{aligned} \label{tildeEinsHorScal} {2}\nabla^2\Phi -4 (\nabla\Phi)^2 -2\Lambda+\,{ F}^{2}+\frac{1}{6}H^2+U +\frac{n-2}{4}\frac{d\,U}{d\,\Phi} =0,\end{aligned}$$ $$\begin{aligned} \label{BfieldHOR} \nabla_{\sigma}({\rm e}^{-2\Phi}H^{\mu\nu\sigma})=0,\end{aligned}$$ $$\begin{aligned} \label{AfieldHOR} \nabla_{\lambda}\left({\rm e}^{-2\Phi}{ F}^{\lambda\epsilon}\right) +a\frac{1}{12}{\rm e}^{-2\Phi}F_{\nu\lambda} \,{ H}^{\epsilon\nu\lambda}=0.\end{aligned}$$ Einstein frame -------------- On the other hand, to pass to the Einstein frame description\[Einsteinframe\] of this low energy string theory, one accomplishes a conformal transformation\[conformaltran\] of the form $$\begin{aligned} {\tilde g}_{\mu\nu}={\rm e}^{2\sigma}{g}_{\mu\nu}, {\tilde g}^{\mu\nu}={\rm e}^{-2\sigma}{g}^{\mu\nu}\rightarrow {\tilde g}={\rm e}^{2\,n\sigma}{g}, \rm dim=n,\end{aligned}$$ which transforms the action $S$ (\[actionHor\]), considered as the barred one, taking into account that ${\tilde F}^2={\tilde g}^{\mu\alpha}{\tilde g}^{\nu\beta} F_{\mu\nu}F_{\alpha\beta}={\rm e}^{-4\sigma}{ g}^{\mu\alpha}{ g}^{\nu\beta} F_{\mu\nu}F_{\alpha\beta}={\rm e}^{-4\sigma}F^2,$ and ${\tilde H}^2={\rm e}^{-6\sigma}H^2,$ to the form $$\begin{aligned} \label{actioMiddle} S=\int{d^{\,n}x\sqrt{-g}{\rm e}^{(n\,\sigma-2\,\sigma -2\Phi)}\left[{\rm e}^{2\sigma}{\tilde R}+{\rm e}^{2\, \sigma}(-2\Lambda+U)+4(\nabla\Phi)^2\, -{\rm e}^{-2\sigma}F^{2} -\frac{1}{12}{\rm e}^{-4\sigma}H^2\right]}.\nonumber\\\end{aligned}$$ Thus, one may use the conformal transformed curvature scalar $$\begin{aligned} {\rm e}^{2\,\sigma}{\tilde R}= { R}-2(n-1)g^{\mu\nu} \nabla_{\nu}\nabla_{\mu}\sigma-(n-1)(n-2)g^{\mu\nu} \nabla_{\mu}\sigma\nabla_{\nu}\sigma,\end{aligned}$$ and chose $$\begin{aligned} \label{conditionCT} n\,\sigma-2\,\sigma-2\Phi=0 \rightarrow{\sigma =\frac{2}{n-2}\Phi},\,\Phi=\frac{n-2}{2}\sigma.\end{aligned}$$ Substituting these relations in the action above (\[actioMiddle\]), one arrives at $$\begin{aligned} \label{actionEISNforString} S=&&\int\,d^{n}x\sqrt{-g}\left[{R} -\frac{4}{n-2}(\nabla\Phi)^2 -2{\rm e}^{4\,\Phi/(n-2)}\Lambda+V(\Phi)\right. \nonumber\\&&\left.-{\rm e}^{-4\,\Phi/(n-2)}F^{2} -\frac{1}{12}{\rm e}^{-8\,\Phi/(n-2)}H^2 \right],\end{aligned}$$ where it has been dropped from this action the divergence $$- 2(n-1)\sqrt{-g}g^{\mu\nu}\nabla_{\nu}\nabla_{\mu}\sigma=- 2(n-1)(\sqrt{-g}\sigma^{,\nu})_{,\nu},$$ and denoted $${\rm e}^{2\sigma}U(\Phi)=V(\Phi).$$ The extremum of $S$ is achieved along the dynamical equations: \[EinsteinStCMgHor\] $$\begin{aligned} \label{EinsteinStCM1Hor} &&{ G}_{\mu\nu}=-g_{\mu\nu}\,{\rm e}^{4\Phi/(n-2)}\Lambda +\frac{1}{2}g_{\mu\nu}\,V(\Phi)+\frac{4}{n-2} \left(\nabla_{\mu}\Phi\,\nabla_{\nu}\Phi -\frac{1}{2} g_{\mu\nu}\nabla_{\alpha}\Phi\,\nabla^{\alpha}\Phi\right) \nonumber\\ && +2\,{\rm e}^{-4\Phi/(n-2)}(F_{\mu\sigma} {F_{\nu}}^{\sigma}-\frac{1}{4} g_{\mu\nu}{F^{2}}) -\frac{1}{4}\,{\rm e}^{-8\Phi/(n-2)}(H_{\mu\alpha\beta} {H_{\nu}}^{\alpha\beta}-\frac{1}{6} g_{\mu\nu}H^2),\end{aligned}$$ $$\begin{aligned} 8\,\nabla^{\nu}\nabla_{\nu}\Phi-8\,\Lambda {\rm e}^{4\Phi/(n-2)} +4\, {\rm e}^{-4\Phi/(n-2)}\,F^{2}+\frac{2}{3}\, {\rm e}^{-8\Phi/(n-2)}\,H^{2}+\frac{n-2}{2} \frac{d\,V}{d\,\Phi}=0,\end{aligned}$$ $$\begin{aligned} \nabla_{\sigma}\left({\rm e}^{-8\Phi/(n-2)}{H}^{\alpha\beta\sigma}\right)=0,\end{aligned}$$ $$\begin{aligned} \nabla_{\lambda}\left({\rm e}^{-4\Phi/(n-2)} { F}^{\lambda\epsilon}\right)+\frac{a}{12}\, {\rm e}^{-8\Phi/(n-2)}F_{\alpha\beta}{ H}^{\alpha\beta\epsilon}=0,\end{aligned}$$ Replacing in (\[EinsteinStCM1Hor\]) the scalar curvature $R$, $$R=\frac{4}{n-2}(\nabla\Phi)^2 +\frac{n-4}{n-2}\,{\rm e}^{-4\Phi/(n-2)}{ F}^2 +\frac{1}{12}\frac{n-6}{n-2}\,{\rm e}^{-8\Phi/(n-2)}{ H}^2 -\frac{n}{n-2}\,(-2\Lambda\,{\rm e}^{4\Phi/(n-2)}+V),$$ one rewrites (\[EinsteinStCM1Hor\]) as $$\begin{aligned} \label{EinsteinStCMRHor} {R}_{\mu\nu} &=&\frac{2 \Lambda}{n-2} \,g_{\mu\nu}\,{\rm e}^{4\Phi/(n-2)}+\frac{4}{n-2} \,\nabla_{\mu}\Phi\,\nabla_{\nu}\Phi +{\rm e}^{-4\Phi/(n-2)}\left(2\,F_{\mu\sigma} {F_{\nu}}^{\sigma}-\frac{1}{n-2} g_{\mu\nu}\, F^{2}\right)\nonumber\\&& +\frac{1}{4}{\rm e}^{-8\Phi/(n-2)} \left(2\,H_{\mu\alpha\beta}{H_{\nu}}^{\alpha\beta} -\frac{2}{3}\frac{1}{n-2} g_{\mu\nu}\,H^{2}\right) -\frac{1}{n-2} g_{\mu\nu}\,V(\Phi).\end{aligned}$$ Dynamical equations in $2+1$ string gravity {#twoplusonedynString} =========================================== In the three dimensional case, the above Einstein action (\[actionEISNforString\]) reduces to $$\begin{aligned} &&S=\int{d^3x\sqrt{-g}\left[{{R}}-2{\rm e}^{4\Phi}\Lambda -4(\nabla\Phi)^2-{\rm e}^{-4\Phi}F^{2} -\frac{1}{12}{\rm e}^{-8\Phi}H^{2}+V(\phi)\right]},\nonumber\\\end{aligned}$$ and the Einstein frame dynamical equations (\[EinsteinStCMgHor\]) become \[EinsteinStCMgHor3\] $$\begin{aligned} \label{EinsteinStCMRHor3} {R}_{\mu\nu} &=&{2 \Lambda} \,g_{\mu\nu}\,{\rm e}^{4\Phi}+{4} \,\nabla_{\mu}\Phi\,\nabla_{\nu}\Phi +{\rm e}^{-4\Phi}\left(2\,F_{\mu\sigma} {F_{\nu}}^{\sigma}- g_{\mu\nu}\, F^{2}\right)\nonumber\\&& +\frac{1}{2}{\rm e}^{-8\Phi} \left(\,H_{\mu\alpha\beta}{H_{\nu}}^{\alpha\beta} -\frac{1}{3} g_{\mu\nu}\,H^{2}\right) - g_{\mu\nu}\,V(\Phi).\end{aligned}$$ $$\begin{aligned} 8\,\nabla^{2}\Phi-8\,\Lambda {\rm e}^{4\Phi} +4\, {\rm e}^{-4\Phi}\,F^{2}+\frac{2}{3}\, {\rm e}^{-8\Phi}\,H^{2}+\frac{1}{2} \frac{d\,V}{d\,\Phi}=0,\end{aligned}$$ $$\begin{aligned} \nabla_{\sigma}\left({\rm e}^{-8\Phi}{H}^{\alpha\beta\sigma}\right)=0,\end{aligned}$$ $$\begin{aligned} \nabla_{\lambda}\left({\rm e}^{-4\Phi} { F}^{\lambda\epsilon}\right)+\frac{a}{12}\, {\rm e}^{-8\Phi}F_{\alpha\beta}{ H}^{\alpha\beta\epsilon}=0,\end{aligned}$$ One can recover the string dynamical equations\[stringdynamical\] by using the conformal inverse relations, see [@Eisenhart:1966book], $$\begin{aligned} && {{\sigma}}_{,i;j}={{\tilde\sigma}}_{,i;j} +2\sigma_{,i}\sigma_{,j} -{\tilde g}_{ij}({\tilde \nabla \sigma})^2, \,{\rm e}^{-2\,\sigma}{{\tilde\sigma}^{,k}}_{\,;k} ={{\sigma}^{,k}}_{\,;k} -(n-2)({\tilde \nabla \sigma})^2,\nonumber\\ && {\stackrel{W}{{ R}_{ij}}} ={\tilde{\stackrel{W} {R}}}_{ij} +{\tilde g}_{ij} {{\tilde\sigma}^{,k}}_{;k} + (n-2)\left({{\tilde\sigma}}_{,i;j}+\sigma_{,i}\sigma_{,j} -{\tilde g}_{ij}({\tilde \nabla \sigma})^2\right),\nonumber\\ && {\rm e}^{-2\,\sigma}{\stackrel{W}{{ R}}} = {\tilde{\stackrel{W} {R}}}+2(n-1){{\tilde\sigma}^{,k}}_{;k} -(n-1)(n-2)({\tilde \nabla \sigma})^2, \end{aligned}$$ where tilde is used to denote that covariant differentials are constructed with $\tilde \Gamma$’s or contravariant tensor components are build with ${\tilde g}^{\mu\nu}$. For $\sigma=2\Phi$ and $n=3$ one gets $$\begin{aligned} \label{EinsteinStCMRn} && { \Phi}_{,\mu;\nu}={\tilde \Phi}_{,\mu;\nu} +4{\Phi}_{,\mu}{\Phi}_{,\nu} -2{\tilde g}_{\mu\nu}({\tilde \nabla}\Phi)^2,\nonumber\\ && {{ \Phi}^{;\alpha}}_{;\alpha}={\rm e}^{4\Phi}\, \left({{\tilde \Phi}^{\,;\alpha}}_{;\alpha} -2({\tilde \nabla}\Phi)^2\right), \nonumber\\ &&{R}_{\mu\nu}={\tilde R}_{\mu\nu} +2{\tilde \Phi}_{,\mu;\nu} +2{\tilde g}_{\mu\nu}{{\tilde \Phi}^{;\alpha}}_{;\alpha} +4{\Phi}_{,\mu}{\Phi}_{,\nu}- 4{\tilde g}_{\mu\nu}({\tilde \nabla}\Phi)^2,\nonumber\\ && R={\rm e}^{4\Phi}\left({\tilde R} +8\,{{\tilde \Phi}^{;\alpha}}_{\,;\alpha}-8\,({\tilde \nabla}\Phi)^2\right),\end{aligned}$$ and conformally transforming the above dynamical equations (\[EinsteinStCMgHor3\]), using relations (\[EinsteinStCMRn\]), one gets the barred dynamical equations of the $2+1$ string theory under consideration \[FieldEQstring\] $$\begin{aligned} &&S=\int{d^3x\sqrt{-{\tilde g}}{\rm e}^{-2\Phi}\left[{\tilde {R}}-2\Lambda +4({\tilde \nabla}\Phi)^2-{\tilde F}^{2}-\frac{1}{12}{\tilde H}^{2}+U(\phi)\right]},\nonumber\\\end{aligned}$$ $$\begin{aligned} \label{tildeEinsHor} &&{\tilde R}_{\mu\nu}+2\nabla_{\mu}\nabla_{\nu}\Phi -2F_{\mu\alpha}F_{\nu\beta}{\tilde g}^{\alpha\beta}- \frac{1}{4}{ H}_{\mu\alpha\beta}{ H_{\nu\gamma\lambda}}{\tilde g}^{\gamma\alpha}{\tilde g}^{\lambda\beta}\nonumber\\&& +{\tilde g}_{\mu\nu}\left(-2\Lambda +U+{\tilde F}^{2}+\frac{{\tilde H}^{2}}{6} +2{\tilde \nabla}^2\Phi-4({\tilde \nabla}\Phi)^2\right)=0,\end{aligned}$$ $$\begin{aligned} \label{tildeEinsHorScal} {2}{\tilde\nabla}^2\Phi -4 ({\tilde\nabla}\Phi)^2+\,{ \tilde F}^{2}-2\Lambda+\frac{1}{6}{\tilde H}^2+U(\Phi) +\frac{1}{4}\frac{d\,U}{d\,\Phi} =0,\end{aligned}$$ $$\begin{aligned} \label{BfieldHOR} {\tilde\nabla}_{\sigma}({\rm e}^{-2\Phi}{\tilde H}^{\mu\nu\sigma})=0,\end{aligned}$$ $$\begin{aligned} \label{AfieldHOR} {\tilde\nabla}_{\lambda}\left({\rm e}^{-2\Phi}{\tilde F}^{\lambda\epsilon}\right) +a\frac{1}{12}{\rm e}^{-2\Phi}{\tilde F}_{\nu\lambda} \,{\tilde H}^{\epsilon\nu\lambda}=0.\end{aligned}$$ Horne-Horowitz black string {#Horne-HorowitzSTR} =========================== In 1991, Horne and Horowitz published an exact string black hole solution in three dimensions [@HorneHorowitzNPB91] for the full string theory of Sec. \[StringEqHOR\], endowed with mass, axion charge per unit length, the asymptotic value of the dilaton, and a cosmological constant. This string solution is given by $$\begin{aligned} \label{HorHor} &&{\stackrel{S}{g}}=-(1-\frac{M}{r})dt^2 +(1-\frac{Q^2}{M\,r})dx^2 +(1-\frac{M}{r})^{-1}(1-\frac{Q^2}{M\,r})^{-1}\frac{k\,dr^2}{8r^2}, \nonumber\\&& H_{rtx}=\frac{Q}{r^2},\,\,\phi=\ln{r}+\frac{1}{2}\ln{\frac{k}{2}}.\end{aligned}$$ The identification of the functions appearing in the action (\[actionHor\]) and equations (\[einsteinHORdyn\]) with the ones of [@HorneHorowitzNPB91] corresponds to $\phi=-2\Phi,$ and $\frac{8}{k}=-2\Lambda=\frac{4}{l^2},\, k=2\,l^2$, where $l$ has dimension of length.\ Accomplishing in the above–mentioned metric a conformal transformation ${\stackrel{E}{g}}_{\mu\nu}={\rm e}^{-2\sigma} {\stackrel{S}{g}}_{\mu\nu}=\frac{k\,r^2}{2}{\stackrel{S}{g}}_{\mu\nu},$ where superscripts $E$ and $S$ stand for Einstein and String respectively, one arrives at the corresponding solution in the Einstein frame, namely $$\begin{aligned} \label{HorHorEF} &&{\stackrel{E}{g}} =-\frac{k\,r^2}{2}(1-\frac{M}{r})dt^2 +\frac{k\,r^2}{2}(1-\frac{Q^2}{M\,r})dx^2+ (1-\frac{M}{r})^{-1}(1-\frac{Q^2}{M\,r})^{-1}\frac{k^2\,dr^2}{16}, \nonumber\\&& H_{rtx}=\frac{Q}{r^2},\,\Phi =-\frac{1}{2}\ln{r}-\frac{1}{4}\ln{\frac{k}{2}},\end{aligned}$$ fulfilling the dynamical equations (\[EinsteinStCMgHor\]) for dimension $n=3$. By accomplishing a $SL(2,R)$ transformation of the Killing coordinates $$\begin{aligned} \label{GENmetricHH} &&t=\alpha\,T+\beta\,\phi,\,\,\Delta:=\alpha\delta-\beta\gamma. \nonumber\\&& x=\gamma\,T+\delta\,\phi,\end{aligned}$$ one arrives at a stationary HH–string solution\[stationaryHH–string\] $$\begin{aligned} \label{HorHorEF} &&{\stackrel{E}{g}}=-\frac{k\,r^2}{2} \left[(1-\frac{M}{r})\alpha^2 -(1-\frac{Q^2}{M\,r})\gamma^2\right]dT^2 +{k\,r^2}\left[(1-\frac{Q^2}{M\,r}) \gamma\delta-(1-\frac{M}{r})\alpha\beta\right]dT\,d\phi\nonumber\\&& +\frac{k\,r^2}{2}\left[(1-\frac{Q^2}{M\,r})\delta^2 -(1-\frac{M}{r})\beta^2\right]d\phi^2+ (1-\frac{M}{r})^{-1}(1-\frac{Q^2}{M\,r})^{-1}\frac{k^2\,dr^2}{16}, \nonumber\\&& H_{rtx}=\frac{Q}{r^2},\,\Phi =-\frac{1}{2}\ln{r}-\frac{1}{4}\ln{\frac{k}{2}},\frac{8}{k}=-2\Lambda==\frac{4}{l^2}.\end{aligned}$$ In the literature one frequently encounters the $SL(2,R)$ transformation $$\begin{aligned} t={\frac {T}{\sqrt {1-{\frac {{\omega}^{2}}{{l}^{2}}}}}}-\omega\, {\frac {\phi}{\sqrt {1-{\frac {{\omega}^{2}}{{l}^{2}}}}}},\, x=- \frac{\omega}{{l}^ {2}}\,{\frac {T}{\sqrt {1-{\frac {{\omega}^{2}}{{l}^{2}}}}}}+ {\frac {\phi}{\sqrt {1-{\frac {{\omega}^{2}}{{l}^{2}}}}}},\end{aligned}$$ which yields $$\begin{aligned} &&g=-{\frac { \left( r-M \right) \left( rM-{Q}^{2} \right) rk \left( {l}^{2}-{\omega}^{2} \right) }{2\,{l}^{2} \left( -{\omega}^{2}rM+{\omega} ^{2}{M}^{2}+rM-{Q}^{2} \right) }}dT^2+ (1-\frac{M}{r})^{-1}(1-\frac{Q^2}{M\,r})^{-1}\frac{k^2\,dr^2}{16} \nonumber\\&& +{\frac {{l}^{2}kr \left( -{\omega}^{2}rM+{\omega}^{2}{M}^{2}+rM-{ Q}^{2} \right) }{2M \left( {l}^{2}-{\omega}^{2} \right) }} \left(d\phi-{\frac {\omega\, \left( -{l}^{2}Mr+{l}^{2}{M}^{2}+rM-{Q}^{2} \right) }{{l}^{2} \left( -{\omega}^{2}rM+{\omega}^{2}{M}^{2}+rM-{Q}^{2} \right) }} dT\right)^2, \nonumber\\&& H_{rtx}=\frac{Q}{r^2},\,\Phi=-\frac{1}{2}\ln{r} -\frac{1}{4}\ln{\frac{k}{2}}.\end{aligned}$$ The evaluation of the quasilocal mass, energy and momentum is done using the Brown–York approach [@BrownY93], this yields $$\begin{aligned} J(r\rightarrow \infty)=2\omega\,{l}^{2} \,{\frac { \left( {M}^{2} -{Q}^{2}\right) }{ \left( { l}^{2}-{\omega}^{2} \right) M}},\end{aligned}$$ $$\begin{aligned} \epsilon(r\rightarrow \infty) =-1/2\,{\frac {1}{r\pi\,{l}^{2}}}-1/4\,{\frac { \left( M-\omega\,Q \right) \left( M+\omega\,Q \right) }{M{l}^{2}\pi\, \left( \omega-1 \right) \left( \omega+1 \right) {r}^{2}}}-\epsilon_0,\end{aligned}$$ $$\begin{aligned} E(r\rightarrow \infty)={\frac {2\,\sqrt {1-{\omega}^{2}}}{\sqrt {{l}^{2} -{\omega}^{2}}}}-2\pi\,K(r)\epsilon_0,\end{aligned}$$ $$\begin{aligned} M_{BY}(r\rightarrow \infty)=4r+4\,l\,(-2r^2+M_0l^2),\end{aligned}$$ $$\begin{aligned} \epsilon_0=-{\frac {1}{\pi\,l}} +\frac{1}{2}\,{\frac {{\it M_0\l}}{\pi\,{r}^{2} }},\end{aligned}$$ $\omega$ is interpreted as a rotating parameter, the mass function increases as the radial coordinate approach spatial infinity. Moreover various pathologies take place at this location. Horowitz–Welch black string {#Horne-HorowitzSTR} =========================== In 1993, Horowitz and Welch published an exact string black hole solution in three dimensions [@Horowitz:1993jc] for the low energy string theory $$\begin{aligned} \label{actionHorWel} S=\int{d^{\,3}x\sqrt{- g}{\rm e}^{-2\Phi}\left[{ R} -2 \Lambda+4 (\nabla\Phi)^2 -\frac{1}{12}H^{2}\right]},\,\Lambda=-2/ k_{HW},\end{aligned}$$ of Sec. \[StringEqHOR\], endowed with mass, angular momentum, axion charge per unit length, and a negative cosmological constant. This string solution is given by a modified BTZ black hole to a $2+1$ string theory with vanishing scalar $\Phi$ and electromagnetic $F_{\alpha\beta}$ fields; in this last case the field equations (\[FieldEQstring\]) become \[FieldEQstringW\] $$\begin{aligned} \label{tildeEinsHorW} { R}_{\mu\nu}- \frac{1}{4}{ H}_{\mu\alpha\beta}{ H_\nu}^{\alpha\beta}=0,\end{aligned}$$ $$\begin{aligned} \label{tildeEinsHorScalW} -2\Lambda+\frac{1}{6}H^2 =0,\end{aligned}$$ $$\begin{aligned} \label{BfieldHORW} \nabla_{\sigma}(H^{\mu\nu\sigma})=0.\end{aligned}$$ The totally anti–symmetric tensor $H_{\mu\nu\alpha}$ has to be proportional to the volume three form $\epsilon_{\mu\nu\alpha}$, because of the equation (\[BfieldHORW\]), the proportionality factor ought to be a constant, hence one may chose $$\begin{aligned} \label{BfieldHORW1} H_{\mu\nu\sigma}=\alpha\frac{1}{l}\epsilon_{\mu\nu\sigma}.\end{aligned}$$ Taking into account the properties of $\epsilon$ $$\begin{aligned} \label{BfieldHORW2} \epsilon_{\alpha\nu\sigma}\epsilon^{\beta\nu\sigma}=-2{\delta^\beta}_{\alpha},\end{aligned}$$ therefore $$\begin{aligned} \label{BfieldHORW2} H_{\alpha\nu\sigma}H^{\beta\nu\sigma} =-2\frac{\alpha^2}{l^2}{\delta^\beta}_{\alpha},\,H^2=-6\frac{\alpha^2}{l^2}.\end{aligned}$$ Consequently, from (\[tildeEinsHorW\]) one has $$\begin{aligned} \label{tildeEinsHorW} { R}_{\mu\nu}- \frac{1}{4}{ H}_{\mu\alpha\beta}{ H_\nu}^{\alpha\beta}=0\rightarrow{{ R}_{\mu\nu}=-\frac{2}{l^2}{ g}_{\mu\nu},\,\alpha=2}\end{aligned}$$ $$\begin{aligned} \label{tildeEinsHorScalW} -2\Lambda+\frac{1}{6}H^2 =0\rightarrow{\Lambda=-\frac{\alpha^2}{2l^2}=-\frac{2}{l^2},\, k_{HW}=l^2}\end{aligned}$$ On the other hand, $\bm H=\bm{dB}$, thus $$\begin{aligned} \label{tildeEinsHorScalWx} H_{\alpha\beta\gamma}=3B_{[\alpha\beta,\gamma]} =\frac{2}{l}\epsilon_{\alpha\beta\gamma}.\end{aligned}$$ As concluded in [@Horowitz:1993jc]: “Thus every solution to three dimensional general relativity with negative cosmological constant is a solution to low energy string theory with: $\Phi=0$, $H_{\alpha\beta\gamma}=\frac{2}{l}\epsilon_{\alpha\beta\gamma}$, and $\Lambda=-\frac{2}{l^2}$.” In particular in [@Horowitz:1993jc] is established that the BTZ black hole metric $$\begin{aligned} \label{HorWELCHmetBTZ} &&{g}=-(\frac{r^2}{l^2}-{M})dt^2-J\,dt\,d\phi +{r}^2d\phi^2 +(\frac{r^2}{l^2}-{M}+\frac{J^2}{4l^2})^{-1}dr^2,\end{aligned}$$ in the presence of an anti–symmetric $B$ field $$\begin{aligned} \label{HorWELCHmetB} B_{\phi\,t}=\frac{r^2}{l},\,\bm H=\bm{dB},\end{aligned}$$ is a solution of the string theory with a zero scalar field $\Phi$.\ By a target space duality transformation (13) of [@Horowitz:1993jc], which referred to [@Buscher88], which means that from a given solution $(g_{\mu\nu},\,B_{\mu\nu},\,\Phi)$ independent on one variable, say $x$, one generates a new solution $(\tilde{g}_{\mu\nu},\,\tilde{B}_{\mu\nu},\,\tilde{\Phi})$ with $$\begin{aligned} \label{HorWELCHmetBT} &&\tilde{g}_{xx}=\frac{1}{{g}_{xx}},\,\tilde{g}_{x\alpha} =\frac{B_{x\alpha}}{{g}_{xx}},\nonumber\\&& \tilde{g}_{\alpha\beta}={g}_{\alpha\beta} -\frac{1}{{g}_{xx}}({g}_{x\alpha}\,{g}_{x\beta} -{B}_{x\alpha}\,{B}_{x\beta}),\nonumber\\&& \tilde{B}_{x\alpha}=\frac{{g}_{x\alpha}}{{g}_{xx}},\, \tilde{B}_{\alpha\beta}={B}_{\alpha\beta} -2\frac{{g}_{x[\alpha}{B}_{\beta]\,x}}{{g}_{xx}},\nonumber\\&& \tilde{\Phi}={\Phi}-\frac{1}{2}\ln{{g}_{xx}},\end{aligned}$$ where $\alpha$ and $\beta$ run over all directions except $x$. Applying this transformation to expressions (\[HorWELCHmetBTZ\]) and (\[HorWELCHmetB\]), along the coordinate symmetry $\phi$, one gets (14) of [@Horowitz:1993jc], namely $$\begin{aligned} \label{HorWELCHmet} &&{\stackrel{S}{g}}=(M-\frac{J^2}{4r^2})dt^2 +\frac{2}{l}dt\,d\phi+\frac{1}{r^2}\,d\phi^2 +(\frac{r^2}{l^2}-M+\frac{J^2}{4r^2})^{-1}dr^2,\end{aligned}$$ which, once diagonalized by means of a $SL(2,R)$ coordinate transformation $$\begin{aligned} \label{HorWELCHtrnscoor} &&t=-\frac{l}{\sqrt{r_{+}^2-r_{-}^2}}\tilde{t} +\frac{l}{\sqrt{r_{+}^2-r_{-}^2}}\tilde{x},\nonumber\\&& \phi=\frac{r_{+}^2}{\sqrt{r_{+}^2-r_{-}^2}}\tilde{t} -\frac{r_{-}^2}{\sqrt{r_{+}^2-r_{-}^2}}\tilde{x},\end{aligned}$$ and the $r$–coordinate transformation $$\begin{aligned} \label{HorWELCHtrnscoorRad} &&r^2=l\tilde{r}\end{aligned}$$ yields the string solution derived in [@HorneHorowitzNPB91], see also Sec. \[Horne-HorowitzSTR\]. Dropping primes, it becomes $$\begin{aligned} \label{HorHor} &&{\stackrel{S}{g}}=-(1-\frac{\mathcal{M}}{r})dt^2 +(1-\frac{\mathcal{Q}^2}{\mathcal{M}\,r})dx^2 +(1-\frac{\mathcal{M}}{r})^{-1}(1-\frac{\mathcal{Q}^2} {\mathcal{M}\,r})^{-1}\frac{l^2\,dr^2}{4r^2},\, \nonumber\\&& B_{xt}=\frac{\mathcal{Q}}{r},\,\,\phi=-\frac{1}{2}\ln{(r\,l)},\end{aligned}$$ where $\mathcal{M}={r_{+}^2}/l$ and $\mathcal{Q}=J/2$.\ The identification of the functions appearing in the action (\[actionHor\]) and equations (\[einsteinHORdyn\]) with the ones of [@HorneHorowitzNPB91], $$\begin{aligned} \label{actionHorHorcom} S=\int{d^{\,3}x\sqrt{- g}{\rm e}^{\phi}\left[{ R} -2 \Lambda+ (\nabla\Phi)^2 -\frac{1}{12}H^{2}\right]},\, \Lambda=-4/k_{HH},\end{aligned}$$ requires that $\phi=-2\Phi_{HW},$ and $k_{HH}=2\,k_{HW}$, $k_{HW}=l^2$, $\Lambda=-2/l^2$, where the subscripts are in correspondence with the initial of the author’s family name. Chan–Mann string solution {#Chan--Mannstring} ========================= Chan and Mann [@Chan:1994qa], see also [@Chan:1995wj], derived a class of solutions to dilaton minimally coupled to $2+1$ Einstein–Maxwell gravity. There is a subclass of solutions allowing an interpretation from the viewpoint of the low energy $2+1$ string theory for specific values of the charged dilaton solution. First, we derive the dilaton solution. Next, assigning specific values to constant characterizing the charged dilaton, the correspondence with the string theory developed in the previous section is established. Einstein–Maxwell–scalar field equations {#sec:cyclic} --------------------------------------- The action considered in [@Chan:1994qa], CM (1), for a (2+1)-dimensional gravity is given by $$\label{action} \mathcal{S}=\int{d^3x}\sqrt{-g}\left[ R-\frac{B}{2}\nabla_{\mu}{\Psi}\,\nabla^{\mu}{\Psi} +2\,{\rm e}^{b\Psi}\Lambda_{CM}-{\rm e}^{-4\,a\,\Psi}\,{ F^2}\right],$$ where $\Lambda_{CM}, b$ are arbitrary at this stage parameters, ${\Psi}$ is the massless minimally coupled scalar field, $R$ is the scalar curvature, and $F^2=F_{\mu\,\nu}\,F^{\mu\,\nu}$ the electromagnetic invariant. The variations of this action yield the dynamical equations $$\begin{aligned} \label{Einsteinfield} &&{R_{\mu\nu}}= \frac{B}{2} \nabla_{\mu}{\Psi}\,\nabla_{\nu}{\Psi}-2{g_{\mu\nu}}{\rm e}^{\,b\,\Psi}\Lambda_{CM} +2\,{\rm e}^{-4\,a\,\Psi}\left({F_\mu}^\alpha\,{F_\nu}_\alpha-g_{\mu\,\nu}\,F^2\right), \nonumber\\ && \frac{B}{2}{\nabla}^{\mu}{\nabla}_{\mu}{{\Psi}} +b\,{\rm e}^{\,b\,\Psi}\Lambda_{CM}+2\,a\,{\rm e}^{-4\,a\,\Psi}\,{ F^2} =0, \nonumber\\ && \nabla^{\mu}\left({\rm e}^{-4\,a\,\Psi}\,F_{\mu\,\nu}\right)=0. \end{aligned}$$ General static cyclic symmetric black hole solution coupled to a scalar field $\Psi(r)=k\,{\rm \ln}(r)$ {#ScalarStatic} -------------------------------------------------------------------------------------------------------- The static cyclic symmetric metric in the $2+1$ Schwarzschild coordinate frame is given by $$\label{metricscLNK_rW_0} \bm{g}=-N (r)^2\bm{dt}^2+\frac{\bm{dr}^2}{L(r)^2} +r^2\bm{d\phi}^2.$$ The electromagnetic field equations for the tensor field $F_{\mu\nu}=2F_{tr}\delta^t_{[\mu}\delta^r_{\nu]}$, and the dilaton $\Phi(r)=k\,{\rm \ln}(r)$ becomes $$\begin{aligned} \label{scalarfeqr} EQ_F=\frac{d}{dr}\frac {F_{tr} \, L \, r^{-4\,a\,k+1}}{N} \rightarrow F_{tr} = {\it Q}\frac{N}{L}{r}^{4\,ak-1}. \end{aligned}$$ The simplest Einstein equations occurs to be ${\it R_{tt}}+{\it R_{rr}}\,L^2\,N^2$, which yields $$\begin{aligned} {\frac {1}{N }}{\frac {d}{dr}}N-{ \frac {1}{L }}{\frac {d}{dr}}L -\frac{1}{2} \,{\frac {B{k}^{2}}{{r}}}=0, \end{aligned}$$ thus one gets $$\begin{aligned} N(r)={\it C_N}\,L \left( r \right) {r}^{B\,{k}^{2}/2}. \end{aligned}$$ On the other hand, the equation ${\it R_{\phi\phi}}$ gives a first order equation for $Y \left( r \right)=L^2$, namely $$\begin{aligned} \label{eqL2sol} && {\frac {d}{dr}}Y \left( r \right) +\frac{1}{2}\,{\frac {B{k}^{2}Y \left( r \right) }{r}}+2\,{\frac { {r}^{4\,ak} {{\it Q}}^{2}} {r}}-2\,\Lambda_{CM}\,{r}^{bk+1}=0 \end{aligned}$$ integrating one obtains $$\begin{aligned} \label{L2sol} L(r)^2=Y \left( r \right) =-4\,{\frac {{r}^{4\,ak}{{\it Q}}^{2}}{B{k}^{ 2}+8\,ak}}+4\,{\frac {\Lambda_{CM}\,{r}^{2+bk}}{4+\,B{k}^{2}+2bk}}+{r}^{- 1/2\,B{k}^{2}}{\it C_1}\end{aligned}$$ Substituting this expression of $Y \left( r \right)$ into the remaining scalar field equation $$\begin{aligned} {\frac {d}{dr}}Y \left( r \right) +\frac{1}{2}\,{\frac {B{k}^{2}Y \left( r \right) }{r}}-8\,{\frac { {r}^{4\,ak}\,a{{\it Q}}^{2}}{B\,k\,r}}+2\,{\frac {b\Lambda_{CM}\,{r}^{bk+1}}{Bk}}=0, \end{aligned}$$ one arrives at relationships between constants, namely $$\begin{aligned} a=-\frac{1}{4\,B\,k},\,b=-B\,k. \end{aligned}$$\ Therefore, the general charged dilaton static solution can be given as $$\begin{aligned} \label{staticGADxe} && \b{g}=-{\it C_N}^{2}\,{r}^{B\,{k}^{2}} \,L(r)^2{dt}^2+\frac{{dr}^2}{L(r)^2}+r^2{d\phi}^2, \nonumber\\ &&L(r)^2=\left({r}^{B\,{k}^{2}/2}{\it C_1} +4\,{\frac {{r}^{2}\Lambda_{CM}}{4-B\,{k}^{2}}}+4\,{\frac {{{\it Q}}^{2}}{B{ k}^{2}}} \right) {r}^{-B\,{k}^{2}},\,{k}^{2}\neq\frac{4}{B}, \nonumber\\ && F_{\mu\nu}=2F_{tr}\delta^t_{[\mu}\delta^r_{\nu]}, \,F_{tr}={\it Q}{\it C_N}\,{r}^{-1/2\,B{k}^{2}-1}=-A_{t,r} \rightarrow\,A_{t} =2\frac{{\it Q}{\it C_N}}{B\,k^2} \,{r}^{-B\,{k}^{2}/2},\nonumber\\ && \Psi(r)=k\,{\rm \ln}{(r)}, \end{aligned}$$ endowed with four relevant parameters: in particular, one may identify the mass $M=-{\it C_1}$, cosmological constant $\Lambda_{CM}\rightarrow\pm\frac{1}{l^2}$, dilaton parameter $k$, and the charge $ {\it Q}$. The constant ${\it C_N}$ can be absorbed by scaling the coordinate $t$, thus it can be equated to unit, ${\it C_N}\rightarrow 1$. Moreover, one has to set the charge $Q$ to zero, $Q=0$, when looking for the limiting solutions for vanishing dilaton $k=0$, which are just the dS and AdS soutions with parameters ${\it C_1}=\pm M$ respectively, and ${\it C_N}=1$. There is no static electrically charged limit of this solution for vanishing dilaton field. The constant $\Lambda_{CM} $ can be equated to minus the standard cosmological constant $\Lambda_{s}=\pm \frac{1}{l^2}$; indeed, by setting in (\[staticGADxe\]) $$\begin{aligned} &&\Lambda_{CM}=\pm \frac{1}{l^2}\alpha^2, \,r\rightarrow r\alpha^{2/(B\,k^2)},\,\phi \rightarrow \phi\,\alpha^{-2/(B\,k^2)}, {\it Q}\rightarrow {\it Q}\,\alpha^{(1+2/(B\,k^2))}, \nonumber\\ && {\it C_1}\rightarrow {\it C_1}\,\alpha^{1+4/(B\,k^2)}, {\it C_N}\rightarrow {\it C_N}\,\alpha^{-(1+2/(B\,k^2))},\end{aligned}$$ one arrives at the metric (\[staticGADxe\]) with $\Lambda_{CM}=\pm\frac{1}{l^2}$. Notice that the $\Lambda_{CM}$ used by in Chan–Mann works, when considered as a cosmological constant, differs from the standard cosmological constant $\Lambda_{s}=\pm \frac{1}{l^2}=-\Lambda_{CM}$, where $+$ and $-$ stand correspondingly for de Sitter and Anti de Sitter (AdS) in $\Lambda_{s}$.\ The string solution derived in [@Chan:1994qa], for $B=8,k=-1/2, a=1,b=4$, which fulfills the Einstein string equations (\[EinsteinStCMgHor3\]) for $\Lambda =\Lambda_{s}=\pm\frac{1}{l^2}=-\Lambda_{CM}$ , in the case of vanishing $H$, can be given as $$\begin{aligned} \label{staticGADx} && {g}_{E}=-{r}^{{2}} \,L(r)^2{dt}^2+\frac{{dr}^2}{L(r)^2}+r^2{d\phi}^2, \nonumber\\ &&L(r)^2=\left({r}\,{\it C_1} -2\,{r}^{2}\Lambda_{s}+2\,{\it Q}^{2} \right) {r}^{-{2}}, \nonumber\\ && F_{\mu\nu}=2F_{tr}\delta^t_{[\mu}\delta^r_{\nu]}, \,F_{tr}=\frac{{\it Q}}{r^2}=-A_{t,r} \rightarrow\,A_{t}=\frac{{\it Q}}{r},\nonumber\\ && \Psi(r)=-1/2\,{\rm \ln}{(r)}. \end{aligned}$$ Under the conformal transformation $$\begin{aligned} {\tilde g}_{\mu\nu} ={\rm e}^{4\,\Psi(r)}{g}_{\mu\nu} =\frac{1}{r^2}{g}_{\mu\nu}\end{aligned}$$ it becomes $$\begin{aligned} \label{staticGADxs} && {g}_{S}=-\left({r}\,{\it C_1} -2\,{r}^{2}\Lambda_{s}+2\,{\it Q}^{2} \right){r}^{{-2}} \,{{dt}^2}+\frac{{dr}^2}{\left({r}\,{\it C_1} -2\,{r}^{2}\Lambda_{s}+2\,{\it Q}^{2} \right) }+{d\phi}^2, \nonumber\\ && F_{\mu\nu}=2F_{tr}\delta^t_{[\mu}\delta^r_{\nu]}, \,F_{tr}=\frac{{\it Q}}{r^2}=-A_{t,r} \rightarrow\,A_{t}=\frac{{\it Q}}{r},\nonumber\\ && \Psi(r)=-1/2\,{\rm \ln}{(r)}, \end{aligned}$$ this is a solution of the equations (\[FieldEQstring\]) of the $2+1$ string theory. Moreover, subjecting the metric (\[staticGADxs\]) to $SL(2,R)$ transformations of the Killing coordinates $$\begin{aligned} t=\alpha\,\tau+\beta\,\theta,\,\, \phi=\gamma\,\tau+\delta\,\theta\end{aligned}$$ one arrives at a rotating charged string solution\[CMrotatingstring\], namely $$\begin{aligned} \label{staticGADx} {g_{S}} &&=-\left(\alpha^2\, \mathcal{L}^2/{r}^{2}-\gamma^2\right) {{d\tau}^2} -2\left(\alpha\beta\,\mathcal{L}^2/{r}^{2} -\gamma\,\delta\right) {d\theta}\, {d\tau} +\left(\delta^2-\beta^2\,\,\mathcal{L}^2/{r}^{2}\right) {d\theta}^2+\frac{ {dr}^2}{\mathcal{L}^2 }, \nonumber\\ && F_{\mu\nu}=2\alpha\,F_{tr}\delta^\tau_{[\mu}\delta^r_{\nu]} -2\beta\,F_{tr}\delta^\theta_{[\mu}\delta^r_{\nu]}, \,F_{tr}:=\frac{{\it Q}}{r^2}=-A_{t,r} \rightarrow\,A_{t}=\frac{{\it Q}}{r},\nonumber\\ && \Psi(r)=-1/2\,{\rm \ln}{(r)},\,\, \mathcal{L}^2:={r}\,{\it C_1} -2\,{r}^{2}\Lambda_{s}+2\,{\it Q}^{2}. \end{aligned}$$ In particular when $\Lambda_{CM}=1/l^2$, AdS branch, a usual choice of the $SL(2,R)$ transformations is given by $$\begin{aligned} t=\frac{\tau}{\sqrt{1-\omega^2/l^2}} -\omega\frac{\theta}{\sqrt{1-\omega^2/l^2}}, \,\, \phi=-\frac{\omega}{l^2}\frac{\tau}{\sqrt{1 -\omega^2/l^2}}+\frac{\theta}{\sqrt{1-\omega^2/l^2}},\end{aligned}$$ where $\omega$ stands for the rotation parameter. This metric can be written as $$\begin{aligned} {g_{S}} &&=-\frac{\mathcal{L}^2}{r^2} \frac{1-{\omega^2}/{l^2}}{1-\omega^2\, \mathcal{L}^2/r^2} { {d\tau}^2} +\frac{1-\omega^2\,\mathcal{L}^2/r^2}{1-{\omega^2}/ {l^2}}\left( {d\theta}-\frac{\omega} {l^2}\frac{1-{\omega^2}/{l^2}} {1-\omega^2\,\mathcal{L}^2/r^2}\, {d\tau}\right)^2 +\frac{ {dr}^2}{\mathcal{L}^2},\nonumber\\\end{aligned}$$ and is endowed with four parameters: the mass $M=-{\it C_1}$, charge ${\it Q}$, rotation $\omega$, and cosmological constant $\Lambda_{s}=\pm1/l^2.$ It is worthy to point out that string solutions one can find in various dimensions, for instance, the works [@WittenPRD91] [@MandalSW_MPLA91], and [@MakiSH_CQG93], among others. Acknowledgments =============== This work has been partially carried out during a sabbatical year at the Physics Department–UAMI, and has been supported by Grant CONACyT 178346. [xxxx]{} G. T. Horowitz, “ The dark side of string theory; Black holes and black strings," \[arXiv:hep-th/9210119\], (1992). J. H. Horne and G. T. Horowitz“Exact string solution in three dimensions,” [Nucl. Phys. [ B368]{}]{} 444 (1992). J.D. Brown and J.W. York, Jr. “Quasilocal energy and conserved charges derived from the gravitational action”, [Phys. Rev.]{} [D 47]{} 1407 (1993). K .C . K. Chan, and R. B. Mann “ Static charged black holes in $(2+1)$–dimensional dilaton gravity", [Phys. Rev.[ D 50]{}]{} 6385 (1994). K. C. K. Chan, and R. B. Mann “ Spinning black holes in $(2+1)$–dimensional string and dilaton gravity", [Phys. Lett. [ B371]{}]{}, 199 (1996). Horowitz, G.T., and Welch, D.L. “Exact three-dimensional black holes in string theory,” [Phys. Rev. Lett.]{}[71]{}, 328 (1993) \[hep-th/9302126\]. Buscher, T.“Path integral derivation of quantum duality in nonlinear sigma models”, [Phys. Lett.]{} [B 201]{}, 466 (199388). L. .P. Eisenhart, [Riemannian Geometry,]{} [ Princeton, Princeton University Press. (1966)]{}. R. M. Wald, [General Relativity,]{} [Chicago, USA: University Chicago Press (1984).]{} T. Ort[í]{}n, [Gravity and Strings,]{} Cambridge University Press (2004). E. Witten, “String theory and black holes," [Phys. Rev.]{} [D44]{} 314 (1991). G. Mandal, A. M. Sengupta, and S. R. Wadia “ Classical solutions of 2–dimensional string theory," [Mod. Phys. Lett.]{} [A6]{} 1685 (1991). T. Maki and K. Shiraishi, “Multi - black hole solutions in cosmological Einstein-Maxwell dilaton theory,” [Class. Quant. Grav.]{} [10]{}, 2171 (1993). [^1]: Acting with the variation operation $\delta$ on the metric tensor and its derivatives in the expression ${{\Gamma}^\mu}_{\alpha\beta},$ taking into account the commutativity of $\delta$ and partial derivatives ${\partial_{,i}},$ and ${ g}^{\mu\nu} {\delta g}_{\sigma\nu}=-{ \delta g}^{\mu\nu}{ g}_{\sigma\nu},$ one has $$\begin{aligned} &&{{\Gamma}^\mu}_{\alpha\beta} =\frac{1}{2}{ g}^{\mu\nu}\left[{ g}_{\nu\beta,\alpha} +{ g}_{\alpha\nu,\beta}-{ g}_{\alpha\beta,\nu}\right] =\frac{1}{2}{ g}^{\mu\nu}\left[\alpha\beta,\nu\right], \left[\alpha\beta,\nu\right]=2{ g}_{\mu\nu}{{\Gamma}^\mu}_{\alpha\beta},\nonumber\\ && \delta{{\Gamma}^\mu}_{\alpha\beta}=\frac{1}{2}{\delta g}^{\mu\nu} \left[\alpha\beta,\nu\right] +\frac{1}{2}{ g}^{\mu\nu}\delta\left[\alpha\beta,\nu\right] ={\delta g}^{\mu\nu}{ g}_{\sigma\nu}{{\Gamma}^\sigma}_{\alpha\beta} +\frac{1}{2}{ g}^{\mu\nu}\delta\left[\alpha\beta,\nu\right]\nonumber\\ && =-{ g}^{\mu\nu}{\delta g}_{\sigma\nu} {{\Gamma}^\sigma}_{\alpha\beta} +\frac{1}{2}{ g}^{\mu\nu}\left[(\delta{ g}_{\nu\beta})_{,\alpha} +(\delta{ g}_{\alpha\nu})_{,\beta}-(\delta{ g}_{\alpha\beta})_{,\nu}\right] =\delta{{\Gamma}^\mu}_{\alpha\beta}(\ref{VarGammaVarmet}).\end{aligned}$$	{ "pile_set_name": "ArXiv" }
--- abstract: 'In a network, a node is said to incur a delay if its encoding of each transmitted symbol involves only its received symbols obtained before the time slot in which the transmitted symbol is sent (hence the transmitted symbol sent in a time slot cannot depend on the received symbol obtained in the same time slot). A node is said to incur no delay if its received symbol obtained in a time slot is available for encoding its transmitted symbol sent in the same time slot. Under the classical model, every node in the network incurs a delay. In this paper, we investigate the multimessage multicast network (MMN) under a generalized-delay model which allows some nodes to incur no delay. We obtain the capacity regions for three classes of MMNs with zero-delay nodes, namely the deterministic network dominated by product distribution, the MMN consisting of independent DMCs and the wireless erasure network. In addition, we show that for any MMN belonging to one of the above three classes, the set of achievable rate tuples under the generalized-delay model and under the classical model are the same, which implies that the set of achievable rate tuples for the MMN does not depend on the delay amounts incurred by the nodes in the network.' author: - 'Silas L. Fong, [^1] [^2]' title: 'Classes of Delay-Independent Multimessage Multicast Networks with Zero-Delay Nodes' --- Multimessage multicast network (MMN), zero-delay nodes, capacity region, cut-set bound, delay-independent. Introduction ============ a multimessage multicast network (MMN), each source sends a message and each destination wants to decode all the messages. The set of source nodes and the set of destination nodes may not be disjoint. A node in the network is said to incur a delay if its encoding of each transmitted symbol involves only its received symbols obtained before the time slot in which the transmitted symbol is sent. In contrast, a node is said to incur no delay if its received symbol obtained in a time slot is available for encoding its transmitted symbol sent in the same time slot. Similarly, the network is said to contain zero-delay nodes if there exists a node that incurs zero delay on another node; the network is said to contain no zero-delay node if every node incurs a delay on all the nodes. In [@fongYeung15], the capacity region of the MMN with zero-delay nodes is defined to be the set of rate tuples achievable by all feasible schemes that do not include deadlock loops, and the positive-delay region is defined to be the set of rate tuples achievable by all classical schemes under the constraint that each node incurs a delay (and hence deadlock loops are automatically excluded). By this definition, the positive-delay region is always a subset of the capacity region. It is easy to construct a network with zero-delay nodes whose capacity region is strictly larger than the positive-delay region. One such network is the binary symmetric channel with correlated feedback (BSC-CF) considered in [@fongYeung15 Section VII], which will be introduced in the next section. A Motivating Example {#sectionMotivating} -------------------- Consider a network that consists of two nodes denoted by $1$ and $2$ respectively. Node 1 and node 2 want to transmit a message to each other. This is a two-way channel [@database:bib17]. Since we can assume without loss of generality that both nodes want to decode both messages, this network can be regarded as a MMN. In each time slot, node 1 and node 2 transmit $X_1$ and $X_2$ respectively, and they receive $Y_1$ and $Y_2$ respectively. All the input and output alphabets are binary, and the channel that carries $X_1$ to node 2 is a binary symmetric channel (BSC) while the channel that carries $X_2$ to node 1 is a discrete memoryless channel (DMC) whose output may depend on the output of the BSC. In this network, node 1 incurs zero delay on node 2, i.e., node $2$ can receive $Y_2$ before encoding and transmitting $X_2$. We call this network the BSC with DMC feedback (BSC-DMCF), which is illustrated in Figure \[BSCFB\]. The BSC-CF is a special case of the BSC-DMCF when $Y_1=X_2+Y_2$ [@fongYeung15 Section VII], where $+$ denotes the XOR operation. It is shown in [@fongYeung15 Section IX] that the capacity region of the BSC-CF is strictly larger than the positive-delay region (recall that the positive-delay region is obtained under the assumption that the network contains no zero-delay node while the capacity region of the BSC-CF is achieved when node 1 incurs zero delay on node 2). Other MMNs whose capacity regions are strictly larger than their positive-delay regions include the relay-without-delay channel studied by El Gamal et al. [@AbbasRelayNetwork] and the causal relay network studied by Baik and Chung [@causalRelayNetwork]. In other words, for some MMNs with zero-delay nodes, their capacity regions can be strictly larger than their positive-delay regions, which motivates us to classify the set of MMNs with zero-delay nodes into the following two categories: (i) : Delay-independent MMNs whose capacity regions coincide with their positive-delay regions. (ii) : Delay-dependent MMNs whose capacity regions are strictly larger than their positive-delay regions. For each MMN in Category (i), the set of achievable rate tuples does not depend on the delay amounts incurred by the nodes in the network. On the other hand, for each MMN in Category (ii), the set of achievable rate tuples shrinks if we impose the additional constraint that each node incurs a positive delay. It is important to decide which category a given MMN belongs to because the category of the MMN affects how the delays should be handled and how the transmissions in the network should be synchronized to achieve optimal performance. Main Contribution ----------------- The main contributions of this work are identification of three classes of delay-independent MMNs and complete characterizations of their capacity regions. The first class is called the deterministic MMN dominated by product distribution. Being a subclass of MMNs consisting of deterministic channels, the deterministic MMN dominated by product distribution is a generalization of the deterministic relay network with no interference in [@multicastCapacityRelayNetworks] and the finite-field linear deterministic network in [@AvestimehrDeterministic; @linearFiniteField09]. The second class is the MMN consisting of independent DMCs [@networkEquivalencePartI]. The third class is the wireless erasure network [@dana06]. We successfully evaluate the capacity regions for the above classes of MMNs with zero-delay nodes and show that their capacity regions coincide with their positive-delay regions, which implies that the above classes of MMNs belong to the category of delay-independent MMNs. A natural consequence of our result is that for any MMN belonging to one of the above three classes, using different methods for handling delays and synchronization in the network does not affect the capacity region. Given a MMN with zero-delay nodes belonging to one of the above three classes, in order to show its delay-independence, we first evaluate an achievable rate region for the MMN by invoking the noisy network coding (NNC) inner bound [@noisyNetworkCoding Theorem 1] (which was also discovered in [@NNCv2]). The achievable rate region is contained in the positive-delay region because the NNC inner bound was proved in [@noisyNetworkCoding] for classical MMNs. Then, we evaluate an outer bound on the capacity region of the MMN with zero-delay nodes by simplifying the cut-set outer bound in [@fongYeung15 Theorem 1] and show that the cut-set outer bound coincides with the NNC inner bound (which is within the positive-delay region), implying that the MMN is delay-independent. This work should not be confused with the work by Effros [@EffrosIndependentDelay], which shows that under the classical model which assumes a positive delay at every node, the set of achievable rate tuples for any MMN does not depend on the amount of positive delay incurred by each node. Here, we prove a different result for the above classes of MMNs with zero-delay nodes that their capacity regions and positive-delay regions are the same. Our result is meaningful given the fact that for some MMNs with zero-delay nodes, their capacity regions are strictly larger than their positive-delay regions (see Section \[sectionMotivating\]). Paper Outline ------------- This paper is organized as follows. Section \[notation\] presents the notation used in this paper. Section \[sectionDefinition\] presents the formulation of the MMN with zero-delay nodes. Section \[sectionInnerOuterBound\] recapitulates the NNC inner bound and the cut-set outer bound for the capacity region of the MMN with zero-delay nodes. In Section \[sectionClassesDelayIndependent\], we use the two bounds obtained in Section \[sectionInnerOuterBound\] to identify the three classes of delay-independent MMNs – the deterministic MMN dominated by product distributions, the MMN consisting of independent DMCs and the wireless erasure network, whose problem formulations and proofs for delay-independence are contained in Section \[sectionDeterministicNetwork\], Section \[sectionDM-MMNconsistingOfDMCs\] and Section \[sectionErasureNetworks\] respectively. Concluding remarks are given in Section \[conclusion\]. Notation ======== We use $\Pr\{\mathcal{E}\}$ to represent the probability of an event $\mathcal{E}$, and use $\mathbf{1}\{\mathcal{E}\}$ to denote the characteristic function of $\mathcal{E}$. We use a capital letter $X$ to denote a random variable with alphabet $\mathcal{X}$, and use the small letter $x$ to denote the realization of $X$. We use $X^n$ to denote a random tuple $(X_1, X_2, \ldots, X_n)$, where the components $X_k$ have the same alphabet $\mathcal{X}$. We let $p_X$ and $p_{Y\|X}$ denote the probability mass distribution of $X$ and the conditional probability mass distribution of $Y$ given $X$ respectively for any discrete random variables $X$ and $Y$. We let $p_X(x)\triangleq\Pr\{X=x\}$ and $p_{Y\|X}(y\|x)\triangleq\Pr\{Y=y\|X=x\}$ be the evaluations of $p_X$ and $p_{Y\|X}$ respectively at $X=x$ and $Y=y$. We let $p_Xp_{Y\|X}$ denote the joint distribution of $(X,Y)$, i.e., $p_Xp_{Y\|X}(x,y)=p_X(x)p_{Y\|X}(y\|x)$ for all $x$ and $y$. If $X$ and $Y$ are independent, their joint distribution is simply $p_X p_Y$. We will take all logarithms to base 2. For any discrete random variable $(X,Y,Z)$ distributed according to $p_{X,Y,Z}$, we let $H_{p_{X,Z}}(X\|Z)$ and $I_{p_{X,Y,Z}}(X;Y\|Z)$ be the entropy of $X$ given $Z$ and mutual information between $X$ and $Y$ given $Z$ respectively. For simplicity, we drop the subscript of a notation if there is no ambiguity. If $X$, $Y$ and $Z$ are distributed according to $p_{X,Y,Z}$ and they form a Markov chain, we write $(X\rightarrow Y\rightarrow Z)_{p_{X,Y,Z}}$ or more simply, $(X\rightarrow Y\rightarrow Z)_p$. The sets of natural and real numbers are denoted by $\mathbb{N}$ and $\mathbb{R}$ respectively. The closure of a set $S$ is denoted by $\cl(S)$ Discrete Memoryless Multimessage Multicast Network with Zero-Delay Nodes {#sectionDefinition} ======================================================================== We consider a multimessage multicast network (MMN) that consists of $N$ nodes. Let $$\mathcal{I}\triangleq \{1, 2, \ldots, N\}$$ be the index set of the nodes, and let $\mathcal{V}\subseteq \mathcal{I}$ and $\mathcal{D}\subseteq \mathcal{I}$ be the sets of sources and destinations respectively. We call $(\mathcal{V}, \mathcal{D})$ the multicast demand on the network. The sources in $\mathcal{V}$ transmit information to the destinations in $\mathcal{D}$ in $n$ time slots (channel uses) as follows. Each node $i\in \mathcal{V}$ transmits message $$W_{i}\in \{1, 2, \ldots, M_i\}$$ and each node $j \in\mathcal{D}$ wants to decode all the messages $\{W_{i}: i\in \mathcal{V}\}$. We assume that each message $W_{i}$ is uniformly distributed over $\{1, 2, \ldots, M_i\}$ and all the messages are independent. For each $k\in \{1, 2, \ldots, n\}$ and each $i\in \mathcal{I}$, node $i$ transmits $X_{i,k} \in \mathcal{X}_i$ and receives $Y_{i,k} \in \mathcal{Y}_i$ in the $k^{\text{th}}$ time slot where $\mathcal{X}_i$ and $\mathcal{Y}_i$ are some alphabets that depend on $i$. After $n$ time slots, node $j$ declares $\hat W_{i,j}$ to be the transmitted $W_{i}$ based on $(W_{j},Y_j^n)$ for each $(i, j)\in \mathcal{V}\times \mathcal{D}$. To simplify notation, we use the following conventions for each $T\subseteq \mathcal{I}$: For any random tuple $$(X_{1}, X_{2}, \ldots, X_{N}) \in \mathcal{X}_1\times \mathcal{X}_2 \times \ldots \times \mathcal{X}_N,$$ we let $$X_T \triangleq (X_{i} : i\in T)$$ be a subtuple of $(X_{1}, X_{2}, \ldots, X_{N})$. Similarly, for any $k\in \{1, 2, \ldots, n\}$ and any random tuple $$(X_{1,k}, X_{2,k}, \ldots, X_{N, k}) \in \mathcal{X}_1\times \mathcal{X}_2 \times \ldots \times \mathcal{X}_N,$$ we let $$X_{T,k}\triangleq(X_{i,k} : i\in T)$$ be a subtuple of $(X_{1,k}, X_{2,k}, \ldots, X_{N, k})$. For any $N^2$-dimensional random tuple $(\hat W_{1,1}, \hat W_{1,2}, \ldots, \hat W_{N,N})$, we let $$\hat W_{T\times T^c}\triangleq(\hat W_{i,j} : (i,j)\in T\times T^c)$$ be a subtuple of $(\hat W_{1,1}, \hat W_{1,2}, \ldots, \hat W_{N,N})$. We follow the formulation of the discrete memoryless network with zero-delay nodes in [@fongYeung15], which includes the following six definitions. The definitions are given here for completeness, and the detailed motivations behind them can be found in [@fongYeung15]. \[defOrderedPartition\] An $\alpha$-dimensional tuple $(\mathcal{S}_1, \mathcal{S}_2, \ldots \mathcal{S}_\alpha)$ consisting of subsets of $\mathcal{I}$ is called an $\alpha$-partition of $\mathcal{I}$ if $\cup_{h=1}^\alpha \mathcal{S}_h = \mathcal{I}$ and $\mathcal{S}_i\cap \mathcal{S}_j = \emptyset$ for all $i\ne j$. For any $(\mathcal{S}_1, \mathcal{S}_2, \ldots \mathcal{S}_\alpha)$ which is an $\alpha$-partition of $\mathcal{I}$, we let $$\mathcal{S}^h \triangleq \cup_{i=1}^h \mathcal{S}_i$$ for each $h\in\{1, 2, \ldots, \alpha\}$ to facilitate discussion. \[defDiscreteNetwork\] The discrete network consists of $N$ finite input sets $\mathcal{X}_1, \mathcal{X}_2, \ldots, \mathcal{X}_N$, $N$ finite output sets $\mathcal{Y}_1, \mathcal{Y}_2, \ldots, \mathcal{Y}_N$ and $\alpha$ channels characterized by conditional distributions $q_{Y_{\mathcal{G}_1}\|X_{\mathcal{S}^1}}^{(1)}$, $q_{Y_{\mathcal{G}_2}\|X_{\mathcal{S}^2}, Y_{\mathcal{G}^1}}^{(2)},\ldots,q_{Y_{\mathcal{G}_\alpha}\|X_{\mathcal{S}^\alpha},Y_{\mathcal{G}^{\alpha-1}}}^{(\alpha)}$, where $$\boldsymbol{\mathcal{S}} \triangleq (\mathcal{S}_1, \mathcal{S}_2, \ldots \mathcal{S}_\alpha)$$and $$\boldsymbol{\mathcal{G}} \triangleq (\mathcal{G}_1, \mathcal{G}_2, \ldots \mathcal{G}_\alpha)$$ are two $\alpha$-dimensional partitions of $\mathcal{I}$. We call $\boldsymbol{\mathcal{S}}$ and $\boldsymbol{\mathcal{G}}$ the input partition and the output partition of the network respectively. The discrete network is denoted by $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$ where $$\boldsymbol q \triangleq (q^{(1)}, q^{(2)}, \ldots, q^{(\alpha)}).$$ \[defDelayProfile\] A delay profile is an $N$-dimensional tuple $(b_1, b_2, \ldots , b_N)$ where $b_i \in \{0, 1\}$ for each $i\in \mathcal{I}$. The delay profile is said to be positive if its elements are all 1. When we formally define a code on the discrete network later, a delay profile $B=(b_1, b_2, \ldots , b_N)$ will be associated with the code and $b_i$ represents the amount of delay incurred by node $i$ for the code. Under the classical model, $B$ can only be positive, meaning that the amount of delay incurred by each node is positive. In contrast, under our generalized-delay model some elements of $B$ can take $0$ as long as deadlock loops do not occur. Therefore our model is a generalization of the classical model. The essence of the following definition is to characterize delay profiles which will not cause deadlock loops for the transmissions in the network. \[defFeasible\] Let $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$ be a discrete network. For each $i\in \mathcal{I}$, let $h_i$ and $m_i$ be the two unique integers such that $i\in \mathcal{S}_{h_i}$ and $i\in\mathcal{G}_{m_i}$. Then, a delay profile $(b_1, b_2, \ldots, b_N)$ is said to be feasible for the network if the following holds for each $i\in \mathcal{I}$: If $b_i=0$, then $h_i > m_i$. Under the classical model, Definition \[defFeasible\] is trivial because any delay profile is positive and hence always feasible for the network. We are ready to define codes that use the network $n$ times as follows. \[defCode\] Let $B\triangleq (b_1, b_2, \ldots , b_N)$ be a delay profile feasible for $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$, and let $(\mathcal{V}, \mathcal{D})$ be the multicast demand on the network. A $(B, n, M_{\mathcal{I}})$-code, where $M_\mathcal{I}\triangleq (M_1, M_2, \ldots, M_N)$ denotes the tuple of message alphabets, for $n$ uses of the network consists of the following: 1. A message set $$\mathcal{W}_{i}\triangleq \{1, 2, \ldots, M_i\}$$ at node $i$ for each $i\in \mathcal{I}$, where $M_i=1$ for each $i\in \mathcal{V}^c$. Message $W_i$ is uniformly distributed on $\mathcal{W}_i$. 2. An encoding function $ f_{i,k} : \mathcal{W}_i \times \mathcal{Y}_i^{k-b_i} \rightarrow \mathcal{X}_i $ for each $i\in \mathcal{I}$ and each $k\in\{1, 2, \ldots, n\}$, where $f_{i,k}$ is the encoding function at node $i$ in the $k^{\text{th}}$ time slot such that $ X_{i,k}=f_{i,k} (W_{i}, Y_i^{k-b_i})$. 3. A decoding function $ g_{i,j} : \mathcal{W}_{j} \times \mathcal{Y}_j^{n} \rightarrow \mathcal{W}_{i} $ for each $(i, j) \in \mathcal{V}\times \mathcal{D}$, where $g_{i,j}$ is the decoding function for $W_{i}$ at node $j$ such that $$\hat W_{i, j} \triangleq g_{i,j}(W_{j}, Y_j^{n}).$$ Given a $(B, n, M_{\mathcal{I}})$-code, it follows from Definition \[defCode\] that for each $i\in\mathcal{I}$, node $i$ incurs a delay if $b_i>0$, where $b_i$ is the amount of delay incurred by node $i$. If $b_i=0$, node $i$ incurs no delay, i.e., for each $k\in \{1, 2, \ldots, n\}$, node $i$ needs to receive $Y_{i,k}$ before encoding $X_{i,k}$. The feasibility condition of $B$ in Definition \[defFeasible\] ensures that the operations of any $(B, n, M_{\mathcal{I}})$-code are well-defined for the subsequently defined discrete memoryless network; the associated coding scheme is described after the network is defined. \[defDiscreteMemoryless\] A discrete network $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$ with multicast demand $(\mathcal{V}, \mathcal{D})$, when used multiple times, is called a discrete memoryless multimessage multicast network (DM-MMN) if the following holds for any $(B, n, M_{\mathcal{I}})$-code: Let $U^{k-1}\triangleq (W_{\mathcal{I}}, X_{\mathcal{I}}^{k-1}, Y_{\mathcal{I}}^{k-1})$ be the collection of random variables that are generated before the $k^{\text{th}}$ time slot. Then, for each $k\in\{1, 2, \ldots, n\}$ and each $h\in \{1, 2, \ldots, \alpha\}$, $$\begin{aligned} & \Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{S}^h,k} =x_{\mathcal{S}^h,k}, Y_{\mathcal{G}^{h},k}=y_{\mathcal{G}^{h},k} \} \notag\\ & = \Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{S}^h,k} =x_{\mathcal{S}^h,k}, Y_{\mathcal{G}^{h-1},k}=y_{\mathcal{G}^{h-1},k} \} q_{Y_{\mathcal{G}_h}\|X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)}(y_{\mathcal{G}_{h},k}\|x_{\mathcal{S}^h,k}, y_{\mathcal{G}^{h-1},k}) \label{memorylessStatement}\end{aligned}$$ for all $u^{k-1}\in \mathcal{U}^{k-1}$, $x_{\mathcal{S}^h,k}\in \mathcal{X}_{\mathcal{S}^h}$ and $y_{\mathcal{G}^h,k}\in \mathcal{Y}_{\mathcal{G}^h}$. Following the notation in Definition \[defDiscreteMemoryless\], consider any $(B, n, M_{\mathcal{I}})$-code on the DM-MMN. In the $k^{\text{th}}$ time slot, $X_{\mathcal{I},k}$ and $Y_{\mathcal{I},k}$ are generated in the order $$X_{\mathcal{S}_1,k}, Y_{\mathcal{G}_1,k}, X_{\mathcal{S}_2,k}, Y_{\mathcal{G}_2,k}, \ldots, X_{\mathcal{S}_\alpha,k}, Y_{\mathcal{G}_\alpha,k} \label{orderExplanation}$$ by transmitting on the channels in this order $q^{(1)},q^{(2)}, \ldots, q^{(\alpha)}$ using the $(B, n, M_{\mathcal{I}})$-code (as prescribed in Definition \[defCode\]). Specifically, $X_{\mathcal{S}^h, k}$, $Y_{\mathcal{G}^{h-1},k}$ and channel $q^{(h)}$ together define $Y_{\mathcal{G}_h,k}$ for each $h\in\{1, 2, \ldots, \alpha\}$. It is shown in [@fongYeung15 Section IV] that the encoding of $X_{\mathcal{S}_h,k}$ before the transmission on $q^{(h)}$ and the generation of $Y_{\mathcal{G}_h,k}$ after the transmission on $q^{(h)}$ for each $h\in\{1, 2, \ldots, \alpha\}$ are well-defined. After defining the DM-MMN with zero-delay nodes in the above six definitions, we are now ready to formally define the capacity region and the positive-delay region through the following three intuitive definitions. \[defErrorProbability\] For a $(B, n, M_{\mathcal{I}})$-code on the DM-MMN, the average probability of decoding error $P_{\text{err}}^{n}$ is defined as $$P_{\text{err}}^{n} \triangleq \Pr\bigg\{\bigcup_{(i,j)\in \mathcal{V}\times \mathcal{D}} \big\{\hat{W}_{i,j} \ne W_{i } \big\}\bigg\}.$$ \[defAchievableRate\] Let $B$ be a feasible delay profile for the network. A rate tuple $(R_{1}, R_{2}, \ldots, R_{N})$, denoted by $R_{\mathcal{I}}$, is $B$-achievable for the DM-MMN if there exists a sequence of $(B, n, M_{\mathcal{I}})$-codes such that $$\lim\limits_{n\rightarrow \infty} \frac{\log M_{i}}{n} \ge R_{i}$$ for each $i\in\mathcal{I}$ and $$\lim\limits_{n\rightarrow \infty} P_{\text{err}}^{n} = 0.$$ \[defCapacityRegion\] The $B$-capacity region, denoted by $\mathcal{C}_B$, of the DM-MMN is the set consisting of every $B$-achievable rate tuple $R_{\mathcal{I}}$ with $R_{i}=0$ for all $i\in \mathcal{V}^c$. The capacity region $\mathcal{C}$ is defined as $$\mathcal{C}\triangleq \bigcup_{B: B\text{ is feasible}}\mathcal{C}_B$$ and the positive-delay region $\mathcal{C}_+$ is defined as $$\mathcal{C}_+ \triangleq \bigcup_{B: B\text{ is positive}}\mathcal{C}_B.$$ If $\mathcal{C} = \mathcal{C}_+$, the DM-MMN is said to be delay-independent. If $\mathcal{C} \supsetneq \mathcal{C}_+$, the DM-MMN is said to be delay-dependent. Roughly speaking, the capacity region is the set of rate tuples which are achievable by codes that do not incur a deadlock loop, and the positive-delay region is the set of rate tuples which are achievable by codes under the constraint that every node incurs a delay. Definitions \[defDelayProfile\], \[defFeasible\] and \[defCapacityRegion\] imply that $\mathcal{C} \supseteq \mathcal{C}_+$, which implies that each DM-MMN is either delay-independent (i.e., $\mathcal{C} = \mathcal{C}_+$) or delay-dependent (i.e., $\mathcal{C} \supsetneq \mathcal{C}_+$). Inner and Outer Bounds for the Capacity Region {#sectionInnerOuterBound} ============================================== We start this section by stating an achievability result for classical DM-MMNs in the following theorem, which is a specialization of the main result of [noisy network coding (NNC)]{} inner bound by Lim, Kim, El Gamal and Chung [@noisyNetworkCoding] (the NNC inner bound was also discovered by Yassaee and Aref [@NNCv2]). \[theoremNNC\] Let $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$ be a DM-MMN, and let $$\begin{aligned} \mathcal{R}_{\text{in}} & \triangleq \bigcup_{\substack{p_{X_\mathcal{I}, Y_\mathcal{I}}: p_{X_\mathcal{I}, Y_\mathcal{I}}= \\ (\prod_{i=1}^N p_{X_i})(\prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h}\|X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)})}} \bigcap_{T\subseteq \mathcal{I}: T^c \cap \mathcal{D} \ne \emptyset } \notag\\* & \hspace{0.5 in}\left\{ R_\mathcal{I}\left\| \: \parbox[c]{3.6 in}{$ \sum_{ i\in T} R_{i} \le I_{p_{X_\mathcal{I},Y_\mathcal{I}}}(X_T; Y_{T^c}\|X_{T^c})-H_{p_{X_\mathcal{I},Y_\mathcal{I}}}(Y_T\|X_\mathcal{I}, Y_{T^c}), \\ R_i=0 \text{ for all }i\in\mathcal{V}^c$} \right.\right\}. \label{Rin}\end{aligned}$$ Then, $$\mathcal{R}_{\text{in}} \subseteq \mathcal{C}_+. \label{theoremSt1theoremNNC}$$ For every (classical) $((1, 1, \ldots, 1), n, M_\mathcal{I})$-code, the MMN $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$ is equivalent to the MMN $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, 1, \mathcal{I}, \mathcal{I} , \prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h}\|X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)})$ by Theorem 3 in [@fongYeung15]. The intuition behind the above equivalence can be reasoned as follows: If every node incurs a delay, then the outputs of the $\alpha$ channels in $\boldsymbol q$ will be independent given their inputs, and hence the relationship between the inputs and outputs of the network can be characterized simply by a product of the $\alpha$ channels, which is $\prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h}\|X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)}$. On the other hand, $\mathcal{R}_{\text{in}}$ is a specialization of the NNC inner bound in [@noisyNetworkCoding Theorem 1] by taking $\hat Y=Y$ for the MMN $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, 1, \mathcal{I}, \mathcal{I} , \prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h}\|X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)})$. Since the NNC inner bound was developed under the classical model where each node incurs a delay, any rate tuple in $\mathcal{R}_{\text{in}}$ is achievable by some sequence of $((1, 1, \ldots, 1), n, M_\mathcal{I})$-codes, which then implies . Following similar procedures for proving Theorem 1 in [@fongYeung15], we can prove an outer bound on $\mathcal{C}$ stated in the following theorem. Since networks with zero-delay nodes can be viewed as networks with in-block memory formulated by Kramer [@kramer14 Section VII-D], the following theorem can be seen as the multicast version of Theorem 1 in [@kramer14]. \[thmCapacityRegionMulticast\] Let $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$ be a DM-MMN, and let $$\begin{aligned} \mathcal{R}_{\text{out}}& \triangleq \bigcup_{\substack{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}:p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}=\\ \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q^{(h)}_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}) }} \bigcap_{T\subseteq \mathcal{I}: T^c \cap \mathcal{D} \ne \emptyset } \notag\\* & \quad \qquad \left\{ R_\mathcal{I}\left\| \: \parbox[c]{4.25 in}{$ \sum_{ i\in T} R_{i} \le \sum_{h=1}^\alpha I_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}}(X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{T^c\cap \mathcal{G}_h}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}),\\ R_i=0 \text{ for all }i\in\mathcal{V}^c$} \right.\right\}. \label{Rout}\end{aligned}$$ Then, $$\mathcal{C}\subseteq\mathcal{R}_{\text{out}}.$$ Let $R_{\mathcal{I}}$ be an achievable rate tuple for the DM-MMN denoted by $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$. By Definitions \[defAchievableRate\] and \[defCapacityRegion\], there exists a sequence of $(B, n, M_{\mathcal{I}})$-codes on the DM-MMN such that $$\lim_{n\rightarrow \infty} \frac{\log M_{i}}{n} \ge R_{i} \label{thmTempEq1}$$ for each $i\in \mathcal{I}$ and $$\lim_{n\rightarrow \infty} P_{\text{err}}^n = 0. \label{thmTempEq2}$$ Fix any $T\subseteq \mathcal{I}$ such that $T^c\cap \mathcal{D}\ne \emptyset$, and let $d$ denote a node in $T^c\cap \mathcal{D}$. For each $(B, n, M_{\mathcal{I}})$-code, since the $N$ messages $W_{1}, W_{2}, \ldots, W_{N}$ are independent, we have $$\begin{aligned} \sum_{i\in T} \log M_{i} & = H(W_{T}\|W_{T^c})\notag \\ &= I(W_{T}; Y_{T^c}^n\|W_{T^c}) + H(W_{T}\|Y_{T^c}^n,W_{T^c}) \notag\\ &\le I(W_{T}; Y_{T^c}^n\|W_{T^c}) + H(W_{T}\|Y_d^n, W_{d})\notag \\ &\le I(W_{T}; Y_{T^c}^n\|W_{T^c}) + 1+ P_{\text{err}}^n \sum_{i\in T} \log M_{i}, \label{cutseteqnSet1}\end{aligned}$$ where the last inequality follows from Fano’s inequality (cf. Definition \[defErrorProbability\]). Following similar procedures for proving Theorem 1 in [@fongYeung15], we can show by using , and that there exists a joint distribution $p_{X_\mathcal{I}, Y_\mathcal{I}}$ which depends on the sequence of $(B, n, M_{\mathcal{I}})$-codes but not on $T$ such that $$p_{X_{\mathcal{I}}, Y_{\mathcal{I}}} = \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q^{(h)}_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}})$$ and $$\sum_{i\in T} R_i\le \sum_{h=1}^\alpha I_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}}(X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{T^c\cap \mathcal{G}_h}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}). \label{stInTheorem}$$ Since $p_{X_\mathcal{I}, Y_\mathcal{I}}$ depends on only the sequence of $(B, n, M_{\mathcal{I}})$-codes but not on $T$, holds for all $T\subseteq \mathcal{I}$ such that $T^c\cap \mathcal{D}\ne \emptyset$. This completes the proof. Classes of Delay-Independent MMNs {#sectionClassesDelayIndependent} ================================= In this section, we will use our inner and outer bounds developed in the previous section to calculate the capacity regions for some classes of MMNs with zero-delay nodes and then show that the MMNs are delay-independent, i.e., $\mathcal{C}=\mathcal{C}_+$. In the process of calculating their capacity regions, we will use the following proposition extensively to characterize an important property of Markov chains. \[propositionMCsimplification\] Suppose there exist two probability distributions $r_{X,Y}$ and $q_{Z\|Y}$ such that $$p_{X, Y, Z} = r_{X,Y}q_{Z\|Y}. \label{statement1CorollaryMC}$$ Then $$(X\rightarrow Y\rightarrow Z)_{p_{X,Y,Z}} \label{statement3CorollaryMC}$$ forms a Markov chain. In addition, $$p_{Z\|Y}=q_{Z\|Y} \label{statement2CorollaryMC}.$$ The proof of is contained [@Yeung08Book Proposition 2.5]. In addition, follows from . Deterministic MMN Dominated by Product Distribution {#sectionDeterministicNetwork} --------------------------------------------------- ### Problem Formulation and Main Result \[definitionDeterministicMatrix\] A conditional distribution $q_{Y\|X}$ is said to be deterministic if for each $x^\in \mathcal{X}$, there exists a $y^\in \mathcal{Y}$ such that $q_{Y\|X}(y^\|x^)=1$. \[definitionDeterministicMMN\] The MMN $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ is said to be deterministic if $q^{(h)}_{Y_{\mathcal{G}_h}\|X_{\mathcal{S}^{h}}, Y_{\mathcal{S}^{h-1}}}$ is deterministic for each $h\in\{1, 2, \ldots, \alpha\}$. With the help of the following definition, we can completely characterize the capacity region for a class of deterministic MMNs with zero-delay nodes. \[defMMNdominatedByProduct\] The deterministic MMN $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ is said to be dominated by product distributions if the following holds for each distribution $p_{X_{\mathcal{I}}}$:\ Define $s_{X_i}$ to be the marginal distribution of $p_{X_{\mathcal{I}}}$ for each $i\in\mathcal{I}$, i.e., $s_{X_i}(x_i) = \sum_{x_j:j\in \mathcal{I} \setminus \{i\}}p_{X_{\mathcal{I}}}(x_{\mathcal{I}})$ for all $x_i$. In addition, define $ p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}\triangleq p_{X_{\mathcal{I}}} \prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)} $ and $ s_{X_{\mathcal{I}}, Y_{\mathcal{I}}}\triangleq (\prod_{i=1}^N s_{X_i})( \prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)})$. Then for any $T\subseteq \mathcal{I}$, $H_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{T^c}\| X_{T^c}) \le H_{s_{X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{T^c}\| X_{T^c})$. The following is our main result in this section. \[theoremDeterministicChannels\] Let $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ be a deterministic MMN dominated by product distributions, and let $$\begin{aligned} \mathcal{R}_{\text{in}}^{\text{det}} & \triangleq \bigcup_{\substack{p_{X_\mathcal{I}, Y_\mathcal{I}}: p_{X_\mathcal{I}, Y_\mathcal{I}}= \\ (\prod_{i=1}^N p_{X_i})(\prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h}\|X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)})}} \bigcap_{T\subseteq \mathcal{I}: T^c \cap \mathcal{D} \ne \emptyset } \left\{ R_\mathcal{I}\left\| \: \parbox[c]{2 in}{$ \sum_{ i\in T} R_{i} \le H_{p_{X_\mathcal{I},Y_\mathcal{I}}}(Y_{T^c}\|X_{T^c}), \\ R_i=0 \text{ for all }i\in\mathcal{V}^c$} \right.\right\}. \label{RinDet}\end{aligned}$$ Then, $$\mathcal{C}= \mathcal{C}_+ = \mathcal{R}_{\text{in}}^{\text{det}} \label{theoremDeterministicChannelsSt1}$$ and hence the network is delay-independent. In particular, holds for the deterministic relay network with no interference in [@multicastCapacityRelayNetworks] and the finite-field linear deterministic network in [@AvestimehrDeterministic; @linearFiniteField09], which implies that they are delay-independent. It has been shown in [@multicastCapacityRelayNetworks] that the capacity region of the deterministic relay network with no interference is contained in the classical cut-set bound even though the network contains zero-delay nodes. Therefore, it is intuitive that the capacity region of any deterministic MMN with zero-delay nodes should be contained in the classical cut-set bound. In addition, the cut-set bound can be achieved if the deterministic MMN is dominated by product distributions. Combining the intuition and the fact provided above, it is intuitive that Theorem \[theoremDeterministicChannels\] should hold. Consider a relay channel that consists of three nodes, where node 1 wants to transmit information to node 3 via a relay node 2. In each time slot, node $i$ transmits $X_i$ and receives $Y_i$ for each $i\in\{1,2,3\}$. All the alphabets are assumed to be binary, and we assume that $Y_2=X_1$ and $Y_3=X_1+X_2$. This relay channel is illustrated in Figure \[RClinearDet\]. The relay channel is a finite-field linear deterministic network [@AvestimehrDeterministic], and it can be formulated as a deterministic MMN with zero-delay nodes by setting $\boldsymbol{\mathcal{S}}\triangleq (\{1\},\{2,3\})$ and $\boldsymbol{\mathcal{G}}\triangleq (\{2\},\{1,3\})$ and choosing appropriate $q^{(1)}_{Y_2\|X_1}$ and $q^{(2)}_{Y_1, Y_3\|X_1, X_2, X_3, Y_2}$ such that $Y_2=X_1$ and $Y_3=X_1+X_2$ with probability one. Since node 2 incurs no delay under this formulation, we cannot characterize the capacity region by applying the classical cut-set bound. However, since every finite-field linear deterministic network is dominated by product distributions [@noisyNetworkCoding Section II-A], Theorem \[theoremDeterministicChannels\] implies that this relay channel with a zero-delay node is delay-independent and its capacity region coincides with the classical cut-set bound. $\blacksquare$ In the following, we provide the proof of Theorem \[theoremDeterministicChannels\]. Since the last statement of the theorem follows from the fact that the deterministic relay network with no interference and the finite-field linear deterministic network are dominated by product distributions [@noisyNetworkCoding Section II-A], it suffices to prove . To this end, it suffices to prove the achievability statement $$\mathcal{R}_{\text{in}}^{\text{det}} \subseteq \mathcal{C}_+ \label{achievabilityStatementDet}$$ and the converse statement $$\mathcal{C}\subseteq \mathcal{R}_{\text{in}}^{\text{det}}. \label{converseStatementDet}$$ ### Achievability In this subsection, we would like to show by using Theorem \[theoremNNC\] and Definition \[definitionDeterministicMMN\]. Since $\mathcal{R}_{\text{in}} \subseteq \mathcal{C}_+$ by Theorem \[theoremNNC\] (cf. ), it suffices to prove that $$\mathcal{R}_{\text{in}}=\mathcal{R}_{\text{in}}^{\text{det}}. \label{theoremSt1lemmaDeterministicRin}$$ Fix any $p_{X_\mathcal{I}, Y_\mathcal{I}}$ that satisfies $$p_{X_\mathcal{I}, Y_\mathcal{I}}= \left(\prod_{i=1}^N p_{X_i}\right)\left(\prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h}\|X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)}\right). \label{theoremSt2lemmaDeterministicRin}$$ Since $$p_{Y_\mathcal{I}\|X_\mathcal{I}}= \prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h}\|X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)} %\label{theoremSt4lemmaDeterministicRin}$$ by and $q_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)}$ is deterministic for each $h\in\{1, 2, \ldots, \alpha\}$, it follows that $p_{Y_{\mathcal{I}}\|X_{\mathcal{I}}}$ is deterministic and hence $$H_{p_{Y_{\mathcal{I}}\|X_{\mathcal{I}}}}(Y_\mathcal{I}\|X_\mathcal{I})=0,$$ which then implies that $$I_{p_{X_\mathcal{I},Y_\mathcal{I}}}(X_T; Y_{T^c}\|X_{T^c})-H_{p_{X_\mathcal{I},Y_\mathcal{I}}}(Y_T\|X_\mathcal{I}, Y_{T^c}) = H_{p_{X_\mathcal{I},Y_\mathcal{I}}}(Y_{T^c}\|X_{T^c}). \label{lemmaCutsetDetinProofEq1}$$ Consequently, follows from , and . ### Converse In this subsection, we will show . Given a $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code on the deterministic MMN and the messages $W_{\mathcal{I}}$, a careful inspection of Definitions \[defCode\], \[defDiscreteMemoryless\] and \[definitionDeterministicMMN\] will reveal that $(X_\mathcal{I}^n, Y_\mathcal{I}^n)$ is just a function of $W_{\mathcal{I}}$, which is formally stated in the following lemma. Since the proof of the lemma is straightforward, it is relegated to Appendix \[appendixA\]. \[cutsetMCLemma\\] Let $(\mathcal{X}_\mathcal{I}, \alpha, \mathcal{Y}_\mathcal{I}, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ be a deterministic MMN. For any $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code on the network, $$H_{p_{ W_{\mathcal{I}},X_{\mathcal{I}}^n, Y_{\mathcal{I}}^n }}(X_{\mathcal{I}}^n, Y_{\mathcal{I}}^n \| W_{\mathcal{I}})=0, \label{lemmaStatement1MCLemma}$$ where $p_{W_{\mathcal{I}},X_{\mathcal{I}}^n, Y_{\mathcal{I}}^n }$ is the joint distribution induced by the code according to Definitions \[defCode\] and \[defDiscreteMemoryless\]. In order to show that the capacity region of the deterministic MMN with zero-delay nodes lies within the classical cut-set bound, we will prove in Theorem \[equivalentNetwork\], the theorem following the proposition below, that the deterministic MMN with zero-delay nodes is equivalent to some classical deterministic MMN. The following proposition is an important step for proving Theorem \[equivalentNetwork\]. \[PropositionequivalentNetwork\] Let $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ be a deterministic MMN. For any $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code on the network, if some $u$, $x_{\mathcal{I}}$ and $y_{\mathcal{I}}$ satisfy $$\Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{I}, k} = x_\mathcal{I}\}>0 \label{assumption0}$$ and $$\Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{I}, k} = x_{\mathcal{I}}, Y_{\mathcal{I},k} = y_{\mathcal{I}} \}=0, \label{contradiction1}$$ then there exists some $h\in\{1, 2, \ldots, \alpha\}$ such that $ q^{(h)}(y_{\mathcal{G}_h} \| x_{\mathcal{S}^h}, y_{\mathcal{G}^{h-1}})=0 $ (where $x_{\mathcal{S}^h}$ is a subtuple of $x_{\mathcal{I}}$, and $y_{\mathcal{G}_h}$ is a subtuple of $y_{\mathcal{I}}$). Suppose there exist $u$, $x_{\mathcal{I}}$ and $y_{\mathcal{I}}$ that satisfy and . We prove the proposition by assuming the contrary. Assume $$q^{(h)}(y_{\mathcal{G}_h} \| x_{\mathcal{S}^h}, y_{\mathcal{G}^{h-1}} )>0 \label{assumption1}$$ for all $h\in\{1, 2, \ldots, \alpha\}$. We now prove by induction on $h$ that $$\Pr\{U^{k-1}\! = u^{k-1}, X_{\mathcal{S}^{h},k} \! = x_{\mathcal{S}^{h}}, \! Y_{\mathcal{G}^{h},k}\!=y_{\mathcal{G}^{h}} \}>0 \label{assumption2}$$ for each $h\in\{1, 2, \ldots, \alpha\}$. For $h=1$, the LHS of is $$\begin{aligned} & \Pr\{U^{k-1}\! = u^{k-1}, X_{\mathcal{S}^{1},k} \! = x_{\mathcal{S}^{1}}, \! Y_{\mathcal{G}^{1},k}\!=y_{\mathcal{G}^{1}} \} \notag \\ & \quad \stackrel{\text(a)}{=} p_{U^{k-1}, X_{\mathcal{S}^1,k}}(u^{k-1}, x_{\mathcal{S}^1}) q^{(1)}(y_{\mathcal{G}_1} \| x_{\mathcal{S}^1}) \notag \\ & \quad \stackrel{\text{(b)}}{>} 0, \label{inductionFirstStatement}\end{aligned}$$ where 1. follows from Definitions \[defDiscreteMemoryless\] and \[definitionDeterministicMMN\]. 2. follows from and . If holds for $h=m$, i.e., $$\Pr\{U^{k-1}\! = u^{k-1}, X_{\mathcal{S}^{m},k} \! = x_{\mathcal{S}^{m}}, \! Y_{\mathcal{G}^{m},k}\!=y_{\mathcal{G}^{m}} \}>0, \label{assumption2}$$ then for $h=m+1$ such that $m+1 \le \alpha$, $$\begin{aligned} & \Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{S}^{m+1},k} =x_{\mathcal{S}^{m+1}}, Y_{\mathcal{G}^{m+1},k}=y_{\mathcal{G}^{m+1}} \}\notag \\ &\stackrel{\text{(a)}}{=} p_{U^{k-1}, X_{\mathcal{S}^{m+1},k},Y_{\mathcal{G}^{m},k}}(u^{k-1}, x_{\mathcal{S}^{m+1}}) q^{(m+1)}(y_{\mathcal{G}_{m+1}} \| x_{\mathcal{S}^{m+1}},y_{\mathcal{G}^{m}})\notag \\ &=p_{U^{k-1}, X_{\mathcal{S}^{m},k}, Y_{\mathcal{G}^{m},k}}(u^{k-1}, x_{\mathcal{S}^{m}}, y_{\mathcal{G}^{m}}) q^{(m+1)}(y_{\mathcal{G}_{m+1}} \| x_{\mathcal{S}^{m+1}},y_{\mathcal{G}^{m}}) \notag \\ & \quad \:\: p_{X_{\mathcal{S}_{m+1},k}\|U^{k-1},X_{\mathcal{S}^{m},k}, Y_{\mathcal{G}^{m},k}}(x_{\mathcal{S}_{m+1}}\|u^{k-1}, x_{\mathcal{S}^{m}},y_{\mathcal{G}^{m}})\notag \\ &\stackrel{\text{(b)}}{=} p_{U^{k-1}, X_{\mathcal{S}^{m},k}, Y_{\mathcal{G}^{m},k}}(u^{k-1}, x_{\mathcal{S}^{m}}, y_{\mathcal{G}^{m}}) q^{(m+1)}(y_{\mathcal{G}_{m+1}} \| x_{\mathcal{S}^{m+1}},y_{\mathcal{G}^{m}}) \notag \\ & \quad \:\: p_{X_{\mathcal{S}_{m+1},k}\|U^{k-1}}(x_{\mathcal{S}_{m+1}}\|u^{k-1}) \notag \\ &\stackrel{\text{(c)}}{>} 0, \label{inductionStatement}\end{aligned}$$ where 1. follows from Definitions \[defDiscreteMemoryless\] and \[definitionDeterministicMMN\]. 2. follows from Lemma \[cutsetMCLemma\\] that $(X_{\mathcal{S}^{m},k}, Y_{\mathcal{G}^{m},k})$ is a function of $U^{k-1}$. 3. follows from , and . Consequently, it follows from , and that holds for $h=\alpha$ by mathematical induction, which then implies that $ \Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{S}^{\alpha},k} = x_{\mathcal{S}^{\alpha}}, Y_{\mathcal{G}^{\alpha},k}=y_{\mathcal{G}^{\alpha}} \}>0$, which contradicts . Surprisingly, each deterministic MMN with zero-delay nodes is equivalent to some classical deterministic MMN, which is proved as follows. \[equivalentNetwork\] For any $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code, the deterministic MMN specified by $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ is equivalent to the deterministic MMN specified by $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, 1, \mathcal{I}, \mathcal{I}, q^{(1)} q^{(2)} \ldots q^{(\alpha)})$. Fix a $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code, and let $U^{k-1}\triangleq (W_{\mathcal{I}}, X_{\mathcal{I}}^{k-1}, Y_{\mathcal{I}}^{k-1})$ be the collection of random variables that are generated before the $k^{\text{th}}$ time slot. To prove the theorem statement for this code, it suffices to show that the following two statements are equivalent for each $k\in\{1, 2, \ldots, n\}$ (cf. ):\ Statement 1: For each $h\in\{1, 2, \ldots, \alpha\}$, $$\begin{aligned} & \Pr\{U^{k-1} = u^{k-1} , X_{\mathcal{S}^h, k} = x_{\mathcal{S}^h}, Y_{\mathcal{G}^h, k} = y_{\mathcal{G}^h}\} \notag\\ &\quad = \Pr\{U^{k-1} = u^{k-1} , X_{\mathcal{S}^h, k} = x_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}, k} = y_{\mathcal{G}^{h-1}}\} q^{(h)}(y_{\mathcal{G}_h} \| x_{\mathcal{S}^h}, y_{\mathcal{G}^{h-1}}). \label{memorylessStatement1}\end{aligned}$$ Statement 2: $$\begin{aligned} & \Pr\{U^{k-1} = u^{k-1} , X_{\mathcal{I}, k} = x_\mathcal{I}, Y_{\mathcal{I}, k} = y_\mathcal{I}\} \notag\\* & \:= \Pr\{U^{k-1}\! = u^{k-1} , X_{\mathcal{I}, k}\! = x_\mathcal{I}\}\prod_{h=1}^\alpha q^{(h)}(y_{\mathcal{G}_h} \| x_{\mathcal{S}^h}, y_{\mathcal{G}^{h-1}}). \label{memorylessStatement2}\end{aligned}$$ Fix a $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code and a $k\in\{1, 2, \ldots, n\}$. We first show that implies . Suppose holds for each $h\in\{1, 2, \ldots, \alpha\}$. Consider the following three mutually exclusive cases:\ Case [$\Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{I}, k} = x_\mathcal{I}\}=0$]{}:\ Both the LHS and the RHS of equal zero.\ Case [$\Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{I}, k} = x_\mathcal{I}\}>0$]{} and\ :\ For this case, the LHS of equals zero. By Proposition \[PropositionequivalentNetwork\], there exists some $h\in\{1, 2, \ldots, \alpha\}$ such that $q^{(h)}(y_{\mathcal{G}_h} \| x_{\mathcal{S}^h}, y_{\mathcal{G}^{h-1}})=0$, which implies that the RHS of equals zero.\ Case [$\Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{I}, k} = x_{\mathcal{I}}, Y_{\mathcal{I}, k} = y_{\mathcal{I}}\}>0$]{}:\ For this case, $$\begin{aligned} & \Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{I}, k} = x_\mathcal{I}, Y_{\mathcal{I}, k} = y_\mathcal{I}\} \\ %&\quad = p_{U^{k-1}, X_{\mathcal{I},k}}(u, x_{\mathcal{I}}) p_{Y_{\mathcal{I},k}\|U^{k-1},X_{\mathcal{I},k}}(y_{\mathcal{I}} \|u , x_{\mathcal{I}})\\ &\quad= p_{U^{k-1}, X_{\mathcal{I},k}}(u^{k-1}, x_{\mathcal{I}})\prod_{h=1}^\alpha p_{Y_{\mathcal{G}_h,k} \|U^{k-1} , X_{\mathcal{I},k}, Y_{\mathcal{G}^{h-1},k}}(y_{\mathcal{G}_h} \|u^{k-1} , x_{\mathcal{I}}, y_{\mathcal{G}^{h-1}}) \\ &\quad \stackrel{\text{(a)}}{=} p_{U^{k-1}, X_{\mathcal{I},k}}(u^{k-1}, x_{\mathcal{I}}) \prod_{h=1}^\alpha p_{Y_{\mathcal{G}_h,k} \|U^{k-1} , X_{\mathcal{S}^h,k}, Y_{\mathcal{G}^{h-1},k}}(y_{\mathcal{G}_h} \|u^{k-1} , x_{\mathcal{S}^h}, y_{\mathcal{G}^{h-1}}) \\ &\quad \stackrel{\eqref{memorylessStatement1}}{=} p_{U^{k-1}, X_{\mathcal{I},k}}(u^{k-1}, x_{\mathcal{I}}) \prod_{h=1}^\alpha q^{(h)}(y_{\mathcal{G}_h} \| x_{\mathcal{S}^h}, y_{\mathcal{G}^{h-1}}),\end{aligned}$$ where (a) follows from Lemma \[cutsetMCLemma\\] that for the $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code, $X_{\mathcal{I},k}$ is a function of $U^{k-1}$. Therefore, the LHS and the RHS of are equal. Combining the three mutually exclusive cases, we obtain that implies . We now show that implies . Suppose holds. Then for each $h\in\{1, 2, \ldots, \alpha\}$ and each $m\in\{1, 2, \ldots, h\}$, $$\begin{aligned} & \Pr\{U^{k-1} = u^{k-1}, X_{\mathcal{S}^h,k} =x_{\mathcal{S}^h}, Y_{\mathcal{G}^{m},k}=y_{\mathcal{G}^{m}} \} \notag \\ &\quad = \sum\limits_{\substack{x_{\mathcal{S}_{h+1}},\ldots, x_{\mathcal{S}_{\alpha}}\\ y_{\mathcal{G}_{m+1}}, \ldots, y_{\mathcal{G}_{\alpha}}}} p_{U^{k-1}, X_{\mathcal{I}, k}, Y_{\mathcal{I},k}}(u^{k-1}, x_{\mathcal{I}}, y_{\mathcal{I}}) \notag \\ &\quad \stackrel{\eqref{memorylessStatement2}}{=} \sum\limits_{\substack{x_{\mathcal{S}_{h+1}},\ldots, x_{\mathcal{S}_{\alpha}}\\ y_{\mathcal{G}_{m+1}}, \ldots, y_{\mathcal{G}_{\alpha}}}} p_{U^{k-1}, X_{\mathcal{I}, k}}(u^{k-1}, x_{\mathcal{I}})\prod_{\ell=1}^\alpha q^{(\ell)}(y_{\mathcal{G}_\ell} \| x_{\mathcal{S}^\ell}, y_{\mathcal{G}^{\ell-1}} )\notag \\ &\quad = \sum_{x_{\mathcal{S}_{h+1}},\ldots, x_{\mathcal{S}_{\alpha}}} p_{U^{k-1}, X_{\mathcal{I}, k}}(u^{k-1}, x_{\mathcal{I}}) \prod_{\ell=1}^{m} q^{(\ell)}(y_{\mathcal{G}_\ell} \| x_{\mathcal{S}^\ell}, y_{\mathcal{G}^{\ell-1}} ) \notag \\ &\quad \stackrel{\text{(a)}}{=} p_{U^{k-1}, X_{\mathcal{S}^h, k}}(u^{k-1}, x_{\mathcal{S}^h}) \prod_{\ell=1}^{m} q^{(\ell)}(y_{\mathcal{G}_\ell} \| x_{\mathcal{S}^\ell}, y_{\mathcal{G}^{\ell-1}} ), \label{equivalentNetworkEqn1}\end{aligned}$$ where (a) follow from the fact that $m\le h$. Then, for each $h\in\{1, 2, \ldots, \alpha\}$, the equality in can be verified by substituting into the LHS and the RHS. The following lemma simplifies the outer bound in Theorem \[thmCapacityRegionMulticast\] for the deterministic MMN. \[lemmaCutsetDeterministicChannelsRout\] Let $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ be a deterministic MMN. Define $$\begin{aligned} \mathcal{R}_{\text{out}}^{\text{det}} & \triangleq \bigcup_{\substack{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}:p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}=\\ p_{X_\mathcal{I}}\prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)}}} \bigcap_{T\subseteq \mathcal{I}: T^c \cap \mathcal{D} \ne \emptyset } & \left\{ R_\mathcal{I}\left\| \: \parbox[c]{2 in}{$ \sum_{ i\in T} R_{i} \le H_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{T^c}\| X_{T^c}),\\ R_i=0 \text{ for all }i\in\mathcal{V}^c$} \right.\right\}. \label{RoutDet}\end{aligned}$$ Then, $$\mathcal{C}\subseteq \mathcal{R}_{\text{out}}^{\text{det}}.$$ Suppose $R_{\mathcal{I}}$ is an achievable rate tuple for $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$. It follows from Definition \[defAchievableRate\] that there exists a sequence of $(B, n, M_{\mathcal{I}})$-codes on $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ such that $ \lim\limits_{n\rightarrow \infty} \frac{\log M_{i}}{n} \ge R_{i} $ for each $i\in\mathcal{I}$ and $ \lim\limits_{n\rightarrow \infty} P_{\text{err}}^{n} = 0$. Since $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ is equivalent to $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, 1, \mathcal{I}, \mathcal{I}, q^{(1)} q^{(2)} \ldots q^{(\alpha)})$ for any $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code on the deterministic DMN by Theorem \[equivalentNetwork\], it follows from Definition \[defCode\] that $R_{\mathcal{I}}$ is achievable for $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, 1, \mathcal{I}, \mathcal{I}, q^{(1)} q^{(2)} \ldots q^{(\alpha)})$, which then implies from Theorem \[thmCapacityRegionMulticast\] that there exists some $p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}^$ satisfying $$p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}^=p_{X_\mathcal{I}}^\prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)} \label{lemmaStatement1CutsetDeterministic}$$ such that for any $T\subseteq\mathcal{I}$ such that $T^c\cap \mathcal{D} \ne \emptyset$, $$\begin{aligned} \sum_{i\in T} R_{i} & \le I_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}^}(X_T;Y_{T^c}\| X_{T^c}) \notag\\ & = H_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}^}(Y_{T^c}\| X_{T^c}) - H_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}^}(Y_{T^c}\| X_{\mathcal{I}}) \notag\\ & \stackrel{\text{(a)}}{=} H_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}^}(Y_{T^c}\| X_{T^c}) \label{lemmaStatement2CutsetDeterministic}\end{aligned}$$ where (a) follows from the fact that $$p_{ Y_{\mathcal{I}}\|X_{\mathcal{I}}}^* \stackrel{\eqref{lemmaStatement1CutsetDeterministic}}{=} \prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)}$$ is deterministic. Consequently, it follows from , and that $R_{\mathcal{I}}\in\mathcal{R}_{\text{out}}^{\text{det}}$. We are now ready to prove as follows. Using Lemma \[lemmaCutsetDeterministicChannelsRout\], we obtain $\mathcal{C}\subseteq \mathcal{R}_{\text{out}}^{\text{det}}$ where $\mathcal{R}_{\text{out}}^{\text{det}}$ is defined in . In addition, it follows from , and Definition \[defMMNdominatedByProduct\] that $\mathcal{R}_{\text{out}}^{\text{det}}\subseteq \mathcal{R}_{\text{in}}^{\text{det}}$. Consequently, $\mathcal{C}\subseteq \mathcal{R}_{\text{in}}^{\text{det}}$. MMN Consisting of Independent DMCs {#sectionDM-MMNconsistingOfDMCs} ---------------------------------- ### Problem Formulation and Main Result Consider a DM-MMN $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ defined as follows: The edge set of the network is characterized by $$\Omega \triangleq \bigcup_{h=1}^\alpha \mathcal{S}^h\times \mathcal{G}_h, \label{defOmega}$$ and a DMC denoted by $q_{Y_{i,j}\|X_{i,j}}$ is associated with every edge $(i,j)\in \Omega$, where $\mathcal{X}_{i,j}$ and $\mathcal{Y}_{i,j}$ are the input and output alphabets of the DMC carrying information from node $i$ to node $j$. The definition of $\Omega$ in ensures that $q_{ Y_{\mathcal{G}_h}\|X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)}$ can be well-defined for each $h\in\{1, 2,\ldots, \alpha\}$. For each $(i,j)\in \Omega$, the capacity of channel $q_{Y_{i,j}\|X_{i,j}}$, denoted by $C_{i,j}$, is attained by some $\bar p_{X_{i,j}}$ due to the channel coding theorem, i.e., $$\begin{aligned} C_{i,j}& \triangleq \max_{p_{X_{i,j}}}I_{p_{X_{i,j}} q_{Y_{i,j}\|X_{i,j}}}(X_{i,j};Y_{i,j}) \ \notag\\ &= I_{ \bar p_{X_{i,j}} q_{Y_{i,j}\|X_{i,j}}}(X_{i,j};Y_{i,j}). \label{defPTPCapacity} \end{aligned}$$ For all the other $(\tilde i, \tilde j)\in \Omega^c$, we assume without loss of generality that $$\mathcal{X}_{\tilde i,\tilde j}=\mathcal{Y}_{\tilde i,\tilde j}=\{0\} \label{alphabetSize=1}$$ and $C_{\tilde i,\tilde j}=0$. Then, we define the input and output alphabets for each node $i$ in the following natural way: $$\mathcal{X}_i \triangleq \mathcal{X}_{i,1}\times \mathcal{X}_{i,2} \times \ldots \times \mathcal{X}_{i,N} \label{alphabetXSequence}$$ and $$\mathcal{Y}_i \triangleq \mathcal{Y}_{1,i}\times \mathcal{Y}_{2,i} \times \ldots \times \mathcal{Y}_{N,i}\label{alphabetYSequence}$$ for each $i\in\mathcal{I}$. In addition, we define $$q_{Y_{\mathcal{G}_h}\|X_{\mathcal{S}^h},Y_{\mathcal{G}^{h-1}}}^{(h)} \triangleq \prod_{(i,j)\in\mathcal{S}^h \times \mathcal{G}_h}q_{Y_{i,j}\|X_{i,j}} \label{defPTPnetworkChannel}$$ for each $h\in\{1, 2, \ldots, \alpha\}$, i.e., the random transformations (noises) from $X_{i,j}$ to $Y_{i,j}$ are independent and each channel $q_{Y_{\mathcal{G}_h}\|X_{\mathcal{S}^h},Y_{\mathcal{G}^{h-1}}}^{(h)}$ is in a product form. We call the network described above the DM-MMN consisting of independent DMCs. The classical MMN consisting of independent DMCs studied in [@networkEquivalencePartI] is a special case of this network model when $\alpha=1$ and $\Omega=\mathcal{I}\times \mathcal{I}$. The following is the main result in this section, and the proof will be presented in the next two subsections. \[thmCapacityPTP\] For the DM-MMN consisting of independent DMCs, define $$\mathcal{R}_{\text{in}}^{\text{DMCs}} \triangleq \bigcap_{T\subseteq \mathcal{I}: T^c \cap \mathcal{D} \ne \emptyset } \left\{ R_\mathcal{I}\left\| \: \parbox[c]{1.8 in}{$ \sum_{ i\in T} R_{i} \le \sum_{ (i,j)\in T \times T^c} C_{i,j},\\ R_i=0 \text{ for all }i\in\mathcal{V}^c$} \right.\right\}. \label{RDMCs}$$ Then, $$\mathcal{C} = \mathcal{C}_+ = \mathcal{R}_{\text{in}}^{\text{DMCs}}$$ and hence the network is delay-independent. In network coding theory, it is well-known that the classical cut-set bound (also called max-flow bound) always holds for networks consisting of noiseless bit-pipes with zero-delay nodes [@Yeung08Book Chapter 18]. Therefore, it is intuitive that by replacing the noiseless bit-pipes by independent DMCs, the cut-set bound still serves as an outer bound on the capacity region. On the other hand, the cut-set bound can be achieved for MMNs consisting of independent DMCs. Combining the intuition and the fact provided above, it is intuitive that Theorem \[thmCapacityPTP\] should hold. \[exampleRelayIndDMCs\] Consider a relay channel that consists of three nodes and three edges connecting the nodes, where node 1 wants to transmit information to node 3 via a relay node 2 through edges $(1,2)$, $(1,3)$ and $(2,3)$. In each time slot, node $i$ transmits $X_{i,j}$ to node $j$ through edge $(i,j)$ and receives $Y_{\ell,i}$ from node $\ell$ through edge $(\ell, i)$. Each edge is associated with a DMC. The three DMCs associated with the three edges, denoted by $q_{Y_{1,2}\| X_{1,2}}$, $q_{Y_{1,3}\| X_{1,3}}$ and $q_{Y_{2,3}\| X_{2,3}}$ respectively, are assumed to be independent, i.e., $$p_{Y_{1,2}, Y_{1,3}, Y_{2,3}\| X_{1,2}, X_{1,3}, X_{2,3}} = q_{Y_{1,2}\| X_{1,2}} q_{Y_{1,3}\| X_{1,3}} q_{Y_{2,3}\| X_{2,3}} %\label{indChannelInExample}$$ regardless of the distribution of $(X_{1,2}, X_{1,3}, X_{2,3})$. This relay channel is illustrated in Figure \[RCindDMCs\]. The relay channel can be formulated as a DM-MMN consisting of independent DMCs by setting $\boldsymbol{\mathcal{S}}\triangleq (\{1\},\{2,3\})$, $\boldsymbol{\mathcal{G}}\triangleq (\{2\},\{1,3\})$, $\Omega \triangleq (\mathcal{S}_1\times \mathcal{G}_1) \cup (\mathcal{S}^2 \times \mathcal{G}_2)$, $X_1\triangleq (X_{1,2}, X_{1,3})$, $X_2\triangleq X_{2,3}$, $Y_2\triangleq Y_{1,2}$, $Y_3\triangleq (Y_{1,3}, Y_{2,3})$, $q_{Y_2\|X_1}^{(1)}\triangleq q_{Y_{1,2}\|X_{1,2}}$ and $q_{Y_3\|X_1, X_2}^{(2)}\triangleq q_{Y_{1,3}\|X_{1,3}}q_{Y_{2,3}\|X_{2,3}}$. The set of non-trivial edges $\{(1,2), (1,3), (2,3)\}$ is inside $\Omega$ by . Since node 2 incurs no delay under this formulation, we cannot characterize the capacity region by applying the classical cut-set bound. Surprisingly, Theorem \[thmCapacityPTP\] implies that this relay channel with a zero-delay node is delay-independent and its capacity region coincides with the classical cut-set bound. $\blacksquare$ In the following, we provide the proof of Theorem \[thmCapacityPTP\]. To this end, it suffices to prove the achievability statement $$\mathcal{R}_{\text{in}}^{\text{DMCs}} \subseteq \mathcal{C}_+ \label{achievabilityStatementPTP}$$ and the converse statement $$\mathcal{C}\subseteq \mathcal{R}_{\text{in}}^{\text{DMCs}}. \label{converseStatementPTP}$$ ### Achievability In this subsection, we would like to prove . Since the DMCs $q_{Y_{i,j}\|X_{i,j}}$ are all independent and each of the DMC can carry information at a rate arbitrarily close to the capacity, it is intuitive that $\mathcal{R}_{\text{in}}^{\text{DMCs}}$ lies in the positive-delay region of the DM-MMN consisting of independent DMCs, which is proved as follows. Let $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ be the DM-MMN consisting of independent DMCs whose positive-delay region is denoted by $\mathcal{C}_+$, and construct a counterpart of the channel $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, q_{Y_\mathcal{I}\|X_\mathcal{I}})$ as follows: Let $(\bar {\mathcal{X}}_\mathcal{I}, \bar {\mathcal{Y}}_\mathcal{I}, \alpha, \bar{\boldsymbol{\mathcal{S}}}, \bar {\boldsymbol{\mathcal{G}}}, \bar {\boldsymbol{q}})$ be a noiseless DM-MMN consisting of independent DMCs with multicast demand $(\mathcal{V}, \mathcal{D})$ such that for each $(i,j)\in \mathcal{I}\times \mathcal{I}$, the DMC carrying information from node $i$ to node $j$ is an error-free (noiseless) channel, denoted by $\bar q_{\bar X_{i,j} \| \bar X_{i,j}}$, with capacity $C_{i,j}$ (cf. ). To be more precise, $\bar q_{\bar X_{i,j} \| \bar X_{i,j}}$ can carry $\lfloor n C_{i,j} \rfloor$ error-free bits for each $(i,j)\in \mathcal{I}\times \mathcal{I}$ for $n$ uses of $(\bar{\mathcal{X}}_\mathcal{I}, \bar{\mathcal{X}}_\mathcal{I}, \bar q_{\bar X_\mathcal{I}\|\bar X_\mathcal{I}})$. Let $\bar {\mathcal{C}}_+$ denote the positive-delay region of $(\bar {\mathcal{X}}_\mathcal{I}, \bar {\mathcal{Y}}_\mathcal{I}, \alpha, \bar{\boldsymbol{\mathcal{S}}}, \bar {\boldsymbol{\mathcal{G}}}, \bar {\boldsymbol{q}})$. Since the original as well as the counterpart DM-MMNs consist of independent DMCs, it follows from the network equivalence theory [@networkEquivalencePartI] that $\mathcal{C}_+ = \bar {\mathcal{C}}_+$. In addition, it has been shown in [@noisyNetworkCoding Section II-A] that $\bar {\mathcal{C}}_+ =\mathcal{R}_{\text{in}}^{\text{DMCs}}$. Consequently, $\mathcal{C}_+ = \bar {\mathcal{C}}_+=\mathcal{R}_{\text{in}}^{\text{DMCs}}$ and holds. ### Converse of Theorem \[thmCapacityPTP\] In this subsection, we would like to prove . Define $$\begin{aligned} & \mathcal{R}_{\text{out}}^{\text{DMCs}} \triangleq \notag\\* & \bigcup_{\substack{p_{ X_\mathcal{I}, Y_\mathcal{I}}: p_{ X_\mathcal{I}, Y_\mathcal{I}}= \\ \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q^{(h)}_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}) }} \!\!\!\! \bigcap_{T\subseteq \mathcal{I}: T^c \cap \mathcal{D} \ne \emptyset } \left\{ R_\mathcal{I}\left\| \: \parbox[c]{3 in}{$\sum_{i\in T}R_i\le\sum_{h=1}^\alpha I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; \\ { }\hspace{1.2 in} Y_{T^c\cap \mathcal{G}_h}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}),\\ R_i=0 \text{ for all }i\in\mathcal{V}^c$} \right.\right\}. \label{RoutDMCs}\end{aligned}$$ It follows form Theorem \[thmCapacityRegionMulticast\] that $\mathcal{C}\subseteq \mathcal{R}_{\text{out}}^{\text{DMCs}}$. Therefore, it remains to show that $$\mathcal{R}_{\text{out}}^{\text{DMCs}} \subseteq \mathcal{R}_{\text{in}}^{\text{DMCs}}. \label{RoutSubsetRinDMC}$$ For any $p_{ X_\mathcal{I}, Y_\mathcal{I}} = \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q^{(h)}_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}) $, it follows from that $$p_{ X_\mathcal{I}, Y_\mathcal{I}} = \prod_{h=1}^\alpha \left( p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} \prod_{(i,j)\in\mathcal{S}^h \times \mathcal{G}_h}q_{Y_{i,j}\|X_{i,j}}\right). \label{distributionToBeMarginalizedDMCs}$$ Marginalizing , we have $$p_{ X_{\mathcal{S}^m}, Y_{\mathcal{G}^m}} = \prod_{h=1}^m \left( p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} \prod_{(i,j)\in\mathcal{S}^h \times \mathcal{G}_h}q_{Y_{i,j}\|X_{i,j}}\right) \label{distributionToBeMarginalizedDMCs**}$$ for each $m\in\{1, 2, \ldots, \alpha\}$. Relabeling symbols in and using and , we have $$p_{X_{\mathcal{S}^h\times \mathcal{I}}, Y_{\mathcal{I}\times \mathcal{G}^h}}=\prod_{\ell=1}^h \left( p_{X_{\mathcal{S}_\ell \times \mathcal{I}}\|X_{\mathcal{S}^{\ell-1}\times \mathcal{I}}, Y_{\mathcal{I}\times \mathcal{G}^{\ell-1}}} \prod_{(i,j)\in\mathcal{S}^\ell \times \mathcal{G}_\ell}q_{Y_{i,j}\|X_{i,j}}\right)$$ for each $h\in\{1, 2, \ldots, \alpha\}$, which implies from Proposition \[propositionMCsimplification\] that for each $h\in\{1, 2, \ldots, \alpha\}$ and each $(i,j)\in \mathcal{S}^h \times \mathcal{G}_h$ (cf. ), $$( \{(X_{k, \ell}, Y_{k, \ell}): (k, \ell)\in \mathcal{S}^h\times \mathcal{G}^h, (k, \ell)\ne (i,j)\} \rightarrow X_{i,j} \rightarrow Y_{i,j})_{p_{ X_\mathcal{I}, Y_\mathcal{I}}} \label{markovChainPTPNetwork}$$ forms a Markov chain. Following , we consider the following chain of inequalities for a fixed $p_{ X_\mathcal{I}, Y_\mathcal{I}} = \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q_{ Y_{\mathcal{G}_h}\| X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)})$ and a fixed $T\subseteq \mathcal{I}:$ $$\begin{aligned} & \sum_{h=1}^\alpha I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{T^c\cap \mathcal{G}_h}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}) \notag\\ & =\sum_{h=1}^\alpha \sum_{j\in T^c\cap \mathcal{G}_h} I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_j\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}) \notag\\ & \stackrel{\eqref{alphabetYSequence}}{=} \sum_{h=1}^\alpha \hspace{-0.2 in}\sum_{\qquad i\in \mathcal{I}, j\in T^c\cap \mathcal{G}_h} \hspace{-0.2 in}I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{i,j}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}) \notag\\ & \stackrel{\text{(a)}}{=} \sum_{h=1}^\alpha \hspace{-0.2 in} \sum_{\qquad i\in \mathcal{S}^h, j\in T^c\cap \mathcal{G}_h} \hspace{-0.2 in}I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{i,j}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}) \notag\\ & \stackrel{\text{(b)}}{=} \sum_{h=1}^\alpha \hspace{-0.2 in} \sum_{\qquad i\in T\cap\mathcal{S}^h, j\in T^c\cap \mathcal{G}_h} \hspace{-0.2 in} I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{i,j}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}) \notag\\ & \stackrel{\text{(c)}}{\le} \sum_{h=1}^\alpha \hspace{-0.2 in} \sum_{\qquad i\in T\cap\mathcal{S}^h, j\in T^c\cap \mathcal{G}_h} \hspace{-0.2 in} I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{i,j} ; Y_{i,j}) \notag\\ &\stackrel{\eqref{defPTPCapacity}}{\le} \sum_{h=1}^\alpha \hspace{-0.2 in} \sum_{\qquad i\in T\cap\mathcal{S}^h, j\in T^c\cap \mathcal{G}_h} \hspace{-0.2 in} C_{i,j} \notag\\ & \le \sum_{h=1}^\alpha \hspace{-0.2 in}\sum_{\qquad i\in T, j\in T^c\cap \mathcal{G}_h} \hspace{-0.2 in} C_{i,j} \notag\\ & = \sum_{(i,j)\in T\times T^c} C_{i,j}, \label{st1corollaryOuterBoundPTP}\end{aligned}$$ where 1. follows from the fact that for each $h\in\{1, 2, \ldots, \alpha\}$ and each $(i,j)\in (\mathcal{I}\setminus\mathcal{S}^h) \times \mathcal{G}_h$, $(i,j)$ lies in $\Omega^c$ (cf. ) and hence $$\begin{aligned} & I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{i,j}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}) \notag\\ &\le H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{i,j}) \notag\\ & \stackrel{\eqref{alphabetSize=1}}{=} 0.\end{aligned}$$ 2. follows from the fact that for each $h\in\{1, 2, \ldots, \alpha\}$ and each $(i,j)\in (T^c\cap \mathcal{S}^h) \times (T^c\cap \mathcal{G}_h)\subseteq \Omega$, $$\begin{aligned} & I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{i,j}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}) \notag\\ & = H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{i,j}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}) \notag\\ &\qquad - H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( Y_{i,j}\| X_{\mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}, Y_{T\cap \mathcal{G}^{h-1}}) \notag\\ &\stackrel{\eqref{markovChainPTPNetwork}}{=} H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{i,j}\|X_{i,j}) - H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{i,j}\|X_{i,j})\notag\\ & = 0.\end{aligned}$$ 3. follows from the fact that for each $h\in\{1, 2, \ldots, \alpha\}$ and each $(i,j)\in (T\cap \mathcal{S}^h) \times (T^c\cap \mathcal{G}_h)\subseteq \Omega$, $$\begin{aligned} & I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{i,j}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}) \notag\\ & \le H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{i,j}) - H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( Y_{i,j}\| X_{\mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_\ell\}_{\ell \in T^c\cap \mathcal{G}_h, \ell<j}, \{ Y_{m,j}\}_{m \in \mathcal{I}, m<i}, Y_{T\cap \mathcal{G}^{h-1}}) \notag\\ &\stackrel{\eqref{markovChainPTPNetwork}}{=} H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{i,j}) - H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{i,j}\|X_{i,j})\notag\\ & = I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(X_{i,j};Y_{i,j}).\end{aligned}$$ Consequently, it follows from , and that holds. Wireless Erasure Network {#sectionErasureNetworks} ------------------------ ### Problem Formulation and Main Result Consider a DM-MMN $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol{q})$ defined as follows: Similar to the MMN consisting of independent DMCs discussed in the previous section, we let $$\Omega \triangleq \bigcup_{h=1}^\alpha \mathcal{S}^h \times \mathcal{G}_h \label{defOmegaWEN}$$ characterize the edge set of the network so that $q_{ Y_{\mathcal{G}_h}\|X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}^{(h)}$ can be well-defined for each $h\in\{1, 2,\ldots, \alpha\}$. To simulate the broadcast nature of wireless networks, we assume that in every time slot, each node $i$ broadcasts a symbol $X_i$ for each $i\in\mathcal{I}$ and we let $\mathcal{X}_i$ denote the finite alphabet of $X_{i}$. For each $(i,j)\in \Omega$, we assume that node $j$ receives $X_{i}$ with erasure probability $\varepsilon_{i,j}\in [0,1]$, and we let $Y_{i,j}$ and $ \mathcal{Y}_{i,j}\triangleq\mathcal{X}_i \cup \{\varepsilon\} $ denote the received symbol and its alphabet respectively where $\varepsilon$ denotes the erasure symbol. For every edge $(i^\prime, j^\prime)$ that is not in $\Omega$, we set its erasure probability $\varepsilon_{i^\prime, j^\prime}$ to be $1$ and $$\mathcal{Y}_{i^\prime, j^\prime} \triangleq \{\varepsilon\}, \label{alphabetSize=1WEN}$$ indicating that no information can be transmitted from $i^\prime$ to $j^\prime$. We let $q_{Y_{i,j}\|X_{i}}$ characterize the channel corresponding to edge $(i,j)$ such that for each $x_i\in \mathcal{X}_i$ and each $y_{i,j}\in \mathcal{Y}_{i,j}$, $$q_{Y_{i,j}\|X_{i}}(y_{i,j}\|x_{i}) = \begin{cases} 1-\varepsilon_{i,j} &\text{if $y_{i,j}=x_{i}$,}\\ \varepsilon_{i,j}& \text{if $y_{i,j}=\varepsilon$.}\end{cases} \label{PrYijGivenXij}$$ The symbols transmitted on the edges in $\Omega$ are assumed to be erased independently, i.e., $$\Pr\left\{\left.\bigcap_{(i,j)\in \Omega}\{Y_{i,j}=y_{i,j}\}\right\|X_{\mathcal{I}}=x_{\mathcal{I}}\right\} = \prod_{(i,j)\in \Omega}q_{Y_{i,j}\|X_{i}}(y_{i,j}\|x_{i}) \label{indErasureOmega}$$ for each $x_{\mathcal{I}}\in \mathcal{X}_\mathcal{I}$ and each $\|\Omega\|$-dimensional tuple $(y_{i,j}: (i,j)\in \Omega)\in \prod_{(i,j)\in \Omega}\mathcal{Y}_{i,j}$. For each $(i,j)\in \mathcal{I}\times \mathcal{I}$, let $E_{i,j}$ and $\mathcal{E}_{i,j}$ be the indicator random variable for the erasure occurred on edge $(i,j)$ and its alphabet respectively such that $$E_{i,j} \triangleq \mathbf{1}\left(\left\{ Y_{i,j}= \varepsilon \right\}\right) = \begin{cases}1 & \text{if $Y_{i,j}=\varepsilon$,} \\0 & \text{if $Y_{i,j}\ne \varepsilon$.} \end{cases} \label{defEij}$$ Let $$E_{\mathcal{I}\times\mathcal{I}}\triangleq (E_{1,1}, E_{1,2}, \ldots, E_{N,N})$$ be the $N^2$-dimensional random tuple containing all the $E_{i,j}$’s so that $E_{\mathcal{I}\times\mathcal{I}}$ characterizes the network erasure pattern, and let $\mathcal{E}_{\mathcal{I}\times\mathcal{I}}$ denote the alphabet of $E_{\mathcal{I}\times\mathcal{I}}$. Recalling that $(\mathcal{V}, \mathcal{D})$ is the multicast demand, we assume that the following two statements hold:\ (i) All the destinations are contained in $\mathcal{G}_\alpha$, i.e., $\mathcal{D}\subseteq \mathcal{G}_\alpha$.\ (ii) The network erasure pattern $E_{\mathcal{I}\times\mathcal{I}}$ in each time slot is accessible by each destination node in $\mathcal{D}$.\ Since $E_{\mathcal{I}\times\mathcal{I}}$ is a function of $Y_{\mathcal{I}}$ by , there exists some conditional distribution $\chi_{E_{\mathcal{I}\times\mathcal{I}}\|Y_{\mathcal{I}}}$ such that $$\chi_{E_{\mathcal{I}\times \mathcal{I}}\|Y_{\mathcal{I}}}(e_{\mathcal{I}\times \mathcal{I}}\|y_{\mathcal{I}}) = \begin{cases} 1 & \text{if $e_{i,j}=\mathbf{1}(\{y_{i,j}=\varepsilon\})$ for all $(i,j)\in \mathcal{I}\times \mathcal{I}$,} \\ 0& \text{otherwise} \end{cases} \label{defChi}$$ for all $y_{\mathcal{I}}$ and $e_{\mathcal{I}\times \mathcal{I}}$. We are now ready to formally define $\mathcal{X}_\mathcal{I}$, $\mathcal{Y}_\mathcal{I} $ and $\boldsymbol{q}$ as follows: For each $i\in\mathcal{I}$, recalling that $\mathcal{X}_i$ is the finite alphabet of the symbol $X_{i,k}$ broadcast by node $i$, we define $$\mathcal{X}_\mathcal{I} \triangleq \mathcal{X}_1\times \mathcal{X}_2 \times \ldots \times \mathcal{X}_N. \label{defWENalphabetX}$$ Recalling that $\mathcal{Y}_{i,j}= \mathcal{X}_i \cup \{\varepsilon\}$ is the alphabet of the noisy version of $X_{i,k}$ that is received by node $j$ in each time slot for each $(i,j)\in \Omega$, we define for each $h\in \{1, 2, \ldots, \alpha\}$ and each $m\in\mathcal{G}_h$ $$\mathcal{Y}_m \triangleq \begin{cases} \left(\prod_{i\in \mathcal{I}} \mathcal{Y}_{i,m}\right)\times \mathcal{E}_{\mathcal{I}\times \mathcal{I}}& \text{if $m$ is an element in $\mathcal{D} \subseteq\mathcal{G}_\alpha$,} \\ \prod_{i\in \mathcal{S}^h} \mathcal{Y}_{i,m} & \text{otherwise.}\end{cases} \label{defWENalphabetYm}$$ The definition of $\mathcal{Y}_m$ in is divided into two cases because we assume according to Statements (i) and (ii) that the destination nodes in $\mathcal{D}\subseteq\mathcal{G}_\alpha$ have access to the network erasure pattern. After defining $\mathcal{Y}_m$ for each $m\in \mathcal{I}$ in , we define $$\mathcal{Y}_\mathcal{I} \triangleq \mathcal{Y}_1\times \mathcal{Y}_2 \times \ldots \times \mathcal{Y}_N. \label{defWENalphabetY}$$ Based on the definitions of $\mathcal{X}_\mathcal{I}$ and $\mathcal{Y}_\mathcal{I}$ in and respectively and recalling and , we define $q_{Y_{\mathcal{G}_h}\|X_{\mathcal{S}^h}, Y_{\mathcal{S}^{h-1}}}^{(h)}$ for each $h\in \{1,2, \ldots, \alpha\}$ as $$\begin{aligned} & q_{Y_{\mathcal{G}_h}\|X_{\mathcal{S}^h}, Y_{\mathcal{S}^{h-1}}}^{(h)}(y_{\mathcal{G}_h}\|x_{\mathcal{S}^h},y_{\mathcal{S}^{h-1}})\notag\\ &\triangleq \begin{cases} \left(\prod_{i\in \mathcal{I}} \prod_{m\in \mathcal{G}_h} q_{Y_{i,m}\|X_{i}}(y_{i,m}\|x_{i})\right)\chi_{E_{\mathcal{I} \times \mathcal{I}}\|Y_{\mathcal{I}}}(e_{\mathcal{I} \times \mathcal{I}}\|y_{\mathcal{I}}) & \text{if $h=\alpha$,} \\ \prod_{i\in \mathcal{S}^h} \prod_{m\in \mathcal{G}_h} q_{Y_{i,m}\|X_{i}}(y_{i,m}\|x_{i}) & \text{otherwise} \end{cases} \label{defYmGivenX} \end{aligned}$$ for all $x_{\mathcal{S}^h}\in \mathcal{X}_{\mathcal{S}^h}$, $y_{\mathcal{G}^h}\in\mathcal{Y}_{\mathcal{G}^h}$ and $e_{\mathcal{I} \times \mathcal{I}}\in \mathcal{E}_{\mathcal{I} \times \mathcal{I}}$, where for each $m\in\mathcal{G}_h$ $$y_m = \begin{cases} ((y_{i,m}: i\in\mathcal{I}),e_{\mathcal{I} \times \mathcal{I}}) & \text{if $m$ is an element in $\mathcal{D} \subseteq\mathcal{G}_\alpha$,}\\ (y_{i,m}: i\in\mathcal{S}^h) & \text{otherwise.}\end{cases}$$ We call the network described above the wireless erasure network. The random variables $X_{\mathcal{I}}$ and $Y_{\mathcal{I}}$ in the wireless erasure network are generated according to this order $$X_{\mathcal{S}_1}, Y_{\mathcal{G}_1}, X_{\mathcal{S}_2}, Y_{\mathcal{G}_2}, \ldots, X_{\mathcal{S}_\alpha}, Y_{\mathcal{G}_\alpha}$$ (cf. ), which implies from , and that $X_\mathcal{I}$, $\{Y_{i,j}\}_{(i,j)\in \Omega}$ and $E_{\mathcal{I}\times \mathcal{I}}$ are generated according to this order $$\begin{aligned} X_{\mathcal{S}_1}, \{Y_{i,j}\}_{(i,j)\in \mathcal{S}^1\times \mathcal{G}_1}, X_{\mathcal{S}_2}, \{Y_{i,j}\}_{(i,j)\in \mathcal{S}^2\times \mathcal{G}_2}, \ldots,X_{\mathcal{S}_\alpha}, \{Y_{i,j}\}_{(i,j)\in \mathcal{S}^\alpha\times \mathcal{G}_\alpha}, E_{\mathcal{I}\times \mathcal{I}}\,. \label{randomVariableSequenceWEN} \end{aligned}$$ It may not be obvious from that $X_\mathcal{I}$ and $E_{\mathcal{I}\times \mathcal{I}}$ are always independent, but it follows from , and that for any $e_{\mathcal{I}\times \mathcal{I}}\in \mathcal{E}_{\mathcal{I}\times \mathcal{I}}$ and $x_\mathcal{I}\in \mathcal{X}_\mathcal{I}$, $$\Pr\{E_{\mathcal{I}\times \mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}\|X_\mathcal{I}=x_\mathcal{I}\} = \prod_{(i,j)\in \Omega} \left(\varepsilon_{i,j}^{\mathbf{1}\{e_{i,j}=1\}}(1-\varepsilon_{i,j})^{\mathbf{1}\{e_{i,j}=0\}}\right), \label{EIIindependetOfXI} %\Pr\{E_{\mathcal{I}\times \mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}\}=$$ which implies the independence between $X_\mathcal{I}$ and $E_{\mathcal{I}\times \mathcal{I}}$, i.e., $$\Pr\{E_{\mathcal{I}\times \mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}\|X_\mathcal{I}=x_\mathcal{I}\} = \Pr\{E_{\mathcal{I}\times \mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}\} \label{EIIindependetOfXI} %=$$ for any $e_{\mathcal{I}\times \mathcal{I}}\in \mathcal{E}_{\mathcal{I}\times \mathcal{I}}$ and $x_\mathcal{I}\in \mathcal{X}_\mathcal{I}$. The classical wireless erasure network studied in [@dana06] is a special case of our model when $\alpha=1$ and $\Omega = \mathcal{I}\times \mathcal{I}$. The following theorem is the main result in this section, and the proof will be provided in the next two subsections. \[thmWEN\] For the wireless erasure network, let $$\begin{aligned} \mathcal{R}_{\text{in}}^{\text{WEN}}& \triangleq \bigcap_{T\subseteq \mathcal{I}: T^c \cap \mathcal{D} \ne \emptyset } \left\{ R_\mathcal{I}\left\| \: \parbox[c]{2.3 in}{$ \sum\limits_{i\in T} R_i \le \sum_{i\in T} \left(1-\prod_{j\in T^c}e_{i,j} \right)\|\mathcal{X}_i\|, \vspace{0.04 in}\\ R_i=0 \text{ for all }i\in\mathcal{V}^c$} \right.\right\}. \label{defRWEN}\end{aligned}$$ Then, $$\mathcal{C} = \mathcal{C}_+ =\mathcal{R}_{\text{in}}^{\text{WEN}}$$ and hence the network is delay-independent. For the wireless erasure network with zero-delay nodes, due to the independence nature among the erasures, the network can be intuitively viewed as a MMN consisting of independent erasure channels, whose capacity region is contained in the classical cut-set bound by Theorem \[thmCapacityPTP\]. On the other hand, it has been shown in [@dana06] that the cut-set bound can be achieved for the wireless erasure network. Combining the intuition and the fact provided above, it is intuitive that Theorem \[thmWEN\] should hold. \[exampleRelayWEN\] Consider a relay channel that consists of three nodes where node 1 wants to transmit information to node 3 via a relay node 2. In each time slot, node $i$ transmits $X_i$ for each $i\in\{1,2,3\}$, while node 2 receives an erased version of $X_1$ denoted by $Y_{1,2}$ and node 3 receives erased versions of $X_1$ and $X_2$ denoted by $Y_{1,3}$ and $Y_{2,3}$ respectively. Let $E_{i,j}$ denote the erasure random variable for $(i,j)$ where $$E_{i,j} = \begin{cases} 0 & \text{if $X_i$ is not erased at node~$j$, i.e., $Y_{i,j}=X_i$,} \\ 1 & \text{otherwise, i.e., $X_i$ is erased at node~$j$.} \end{cases} \label{defEijInExample}$$ The erasures are assumed to be independent, i.e., $p_{E_{1,2}, E_{1,3}, E_{2,3}}=p_{E_{1,2}} p_{E_{1,3}} p_{E_{2,3}}$ regardless of the distribution of $(X_1, X_2)$. In addition, node 3 is assumed to have access of the network erasure pattern $(E_{1,2}, E_{1,3}, E_{2,3})$ (note that $E_{1,3}$ and $E_{2,3}$ can be deduced from $Y_{1,3}$ and $Y_{2,3}$ respectively by , but $E_{1,2}$ is an extra information provided for node 3 for decoding). This relay channel is illustrated in Figure \[RCWEN\]. The relay channel can be formulated as a wireless erasure network by setting $\boldsymbol{\mathcal{S}}\triangleq (\{1\},\{2,3\})$, $\boldsymbol{\mathcal{G}}\triangleq (\{2\},\{1,3\})$, $\Omega \triangleq (\mathcal{S}_1\times \mathcal{G}_1) \cup (\mathcal{S}^2 \times \mathcal{G}_2)$, $Y_2\triangleq Y_{1,2}$ and $Y_3\triangleq (Y_{1,3}, Y_{2,3}, E_{1,2}, E_{1,3}, E_{2,3})$. The set of non-trivial edges $\{(1,2), (1,3), (2,3)\}$ is contained in $\Omega$ by . Since node 2 incurs no delay under this formulation, we cannot characterize the capacity region by applying the classical cut-set bound. Surprisingly, Theorem \[thmWEN\] implies that this three-node wireless erasure network with a zero-delay node is delay-independent. $\blacksquare$ In the following, we provide the proof of Theorem \[thmWEN\]. To this end, it suffices to prove the achievability statement $$\mathcal{R}_{\text{in}}^{\text{WEN}} \subseteq \mathcal{C}_+ \label{achievabilityStatementWEN}$$ and the converse statement $$\mathcal{C}\subseteq \mathcal{R}_{\text{in}}^{\text{WEN}}. \label{converseStatementWEN}$$ ### Achievability {#sectionAchThmWEN} In this subsection, we would like to prove . Since the achievability statement has been shown in [@dana06] under the classical model which considers no zero-delay nodes, holds naturally under our generalized-delay model. For completeness, the proof of under our generalized-delay model is provided in Appendix \[appendixB\]. ### Converse {#sectionCovThmWEN} In this subsection, we would like to prove . We will first prove the following counterpart of Theorem \[thmCapacityRegionMulticast\] to show an outer bound on $\mathcal{C}$, and then show that the outer bound is contained in $\mathcal{R}_{\text{in}}^{\text{WEN}}$. \[thmCapacityRegionMulticastWEN\] Let $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$ be a wireless erasure network, and let $$\begin{aligned} & \mathcal{R}_{\text{out}}^{\text{WEN}} \triangleq \notag\\* & \bigcup_{\substack{p_{ X_\mathcal{I}, Y_\mathcal{I}}: p_{ X_\mathcal{I}, Y_\mathcal{I}}= \\ \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q^{(h)}_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}}) }}\bigcap_{T\subseteq \mathcal{I}: T^c \cap \mathcal{D} \ne \emptyset } \left\{ R_\mathcal{I}\left\| \: \parbox[c]{3 in}{$\sum_{i\in T}R_i\le\sum_{h=1}^\alpha I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; \\ { }\hspace{1 in} Y_{T^c\cap \mathcal{G}_h}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, E_{\mathcal{I}\times \mathcal{I}}),\\ R_i=0 \text{ for all }i\in\mathcal{V}^c$} \right.\right\} \label{RoutWEN}\end{aligned}$$ where $E_{\mathcal{I}\times \mathcal{I}}$, the network erasure pattern, is a function of $Y_{\mathcal{I}}$ defined by . Then, $$\mathcal{C}\subseteq \mathcal{R}_{\text{out}}^{\text{WEN}}.$$ Let $R_{\mathcal{I}}$ be an achievable rate tuple for the wireless erasure network denoted by $(\mathcal{X}_\mathcal{I}, \mathcal{Y}_\mathcal{I}, \alpha, \boldsymbol{\mathcal{S}}, \boldsymbol{\mathcal{G}}, \boldsymbol q)$. Then, there exists a sequence of $(B, n, M_{\mathcal{I}})$-codes on the network such that $$\lim_{n\rightarrow \infty} \frac{\log M_{i}}{n} \ge R_{i} \label{thmTempEq1WEN}$$ and $$\lim_{n\rightarrow \infty} P_{\text{err}}^n = 0 \label{thmTempEq2WEN}$$ for each $i\in \mathcal{I}$. Fix any $T\subseteq \mathcal{I}$ such that $T^c\cap \mathcal{D}\ne \emptyset$, and let $d$ denote a node in $T^c\cap \mathcal{D}$. Fix a $(B, n, M_{\mathcal{I}})$-code and let $E_{\mathcal{I}\times \mathcal{I},k}$ denote the network erasure pattern occurred in time slot $k$ for each $k\in\{1, 2, \ldots, n\}$. Then, we consider the following chain of inequalities: $$\begin{aligned} \sum_{i\in T} \log M_{i} & \stackrel{\text{(a)}}{=} H(W_{T}\|W_{T^c})\notag \\* & \stackrel{\text{(b)}}{=} H(W_{T}\|W_{T^c}, E_{\mathcal{I}\times \mathcal{I}}^n)\notag \\ &= I(W_{T}; Y_{T^c}^n\|W_{T^c},E_{\mathcal{I}\times \mathcal{I}}^n) + H(W_{T}\|Y_{T^c}^n,W_{T^c},E_{\mathcal{I}\times \mathcal{I}}^n) \notag\\ &\le I(W_{T}; Y_{T^c}^n\|W_{T^c},E_{\mathcal{I}\times \mathcal{I}}^n) + H(W_{T}\|Y_d^n, W_{d},E_{\mathcal{I}\times \mathcal{I}}^n)\notag \\ &\stackrel{\text{(c)}}{\le} I(W_{T}; Y_{T^c}^n\|W_{T^c},E_{\mathcal{I}\times \mathcal{I}}^n) + 1+ P_{\text{err}}^n \sum_{i\in T} \log M_{i}, \label{cutseteqnSet1WEN}\end{aligned}$$ where 1. follows from the fact that the $N$ messages $W_{1}, W_{2}, \ldots, W_{N}$ are independent. 2. follows from the fact that $W_{\mathcal{I}}$ and $E_{\mathcal{I}\times \mathcal{I}}^n$ are independent. 3. follows from Fano’s inequality. Following similar procedures for proving Theorem 1 in [@fongYeung15], we can show by using , and that there exists a joint distribution $p_{X_\mathcal{I}, Y_\mathcal{I}}$ which depends on the sequence of $(B, n, M_{\mathcal{I}})$-codes but not on $T$ such that $$p_{X_{\mathcal{I}}, Y_{\mathcal{I}}} = \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q^{(h)}_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}})$$ and $$\sum_{i\in T} R_i\le \sum_{h=1}^\alpha I_{p_{X_{\mathcal{I}}, Y_{\mathcal{I}}}}(X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{T^c\cap \mathcal{G}_h}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, E_{\mathcal{I}\times \mathcal{I}}). \label{stInTheoremCon}$$ Since $p_{X_\mathcal{I}, Y_\mathcal{I}}$ depends on only the sequence of $(B, n, M_{\mathcal{I}})$-codes but not on $T$, holds for all $T\subseteq \mathcal{I}$ such that $T^c\cap \mathcal{D}\ne \emptyset$. This completes the proof. Since $$\mathcal{C}\subseteq \mathcal{R}_{\text{out}}^{\text{WEN}} \label{converseWENByTheorem2}$$ by Lemma \[thmCapacityRegionMulticastWEN\] and our goal is to prove $\mathcal{C}\subseteq \mathcal{R}_{\text{in}}^{\text{WEN}} $, it remains to show that $$\mathcal{R}_{\text{out}}^{\text{WEN}} \subseteq \mathcal{R}_{\text{in}}^{\text{WEN}}. \label{RoutSubsetRinWEN}$$ For any $p_{ X_\mathcal{I}, Y_\mathcal{I}} = \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q^{(h)}_{Y_{\mathcal{G}_h} \| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}})$, it follows from and that for each $h\in\{1, 2, \ldots, \alpha\}$ and each $(i,j)\in \mathcal{S}^h \times \mathcal{G}_h$, $Y_{i,j}$ is a function of $(X_{i}, E_{\mathcal{I}\times \mathcal{I}})$ and hence $$\left(\{(X_{k}, Y_{k, \ell}): (k, \ell)\in (\mathcal{S}^h \setminus \{i\})\times \mathcal{G}^h\} \rightarrow (X_{i}, E_{\mathcal{I}\times \mathcal{I}}) \rightarrow Y_{i,j}\right)_{p_{ X_\mathcal{I}, Y_\mathcal{I}}} \label{markovChainWENNetwork}$$ forms a Markov chain. Following and , we fix $p_{ X_\mathcal{I}, Y_\mathcal{I}} = \prod_{h=1}^\alpha (p_{X_{\mathcal{S}_h}\|X_{\mathcal{S}^{h-1}}, Y_{\mathcal{G}^{h-1}}} q_{ Y_{\mathcal{G}_h}\| X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)})$ and $T\subseteq \mathcal{I}$ such that $T^c \cap \mathcal{D} \ne \emptyset$, and we consider $$\begin{aligned} & \sum_{h=1}^\alpha I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{T^c\cap \mathcal{G}_h}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, E_{\mathcal{I}\times \mathcal{I}}) \notag\\* & \stackrel{\eqref{defWENalphabetYm}}{=}\sum_{h=1}^\alpha \sum_{i\in \mathcal{S}^h} I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_{\{\ell\}\times (T^c\cap \mathcal{G}_h})\}_{\ell<i},E_{\mathcal{I}\times \mathcal{I}}) \notag\\ & \stackrel{\text{(a)}}{=} \sum_{h=1}^\alpha \sum_{i\in T\cap \mathcal{S}^h} I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_{\{\ell\}\times (T^c\cap \mathcal{G}_h)}\}_{\ell<i}, E_{\mathcal{I}\times \mathcal{I}}) \notag\\ & \stackrel{\text{(b)}}{\le} \sum_{h=1}^\alpha \sum_{i\in T\cap\mathcal{S}^h} I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{i} ; Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| Y_{\{i\}\times (T^c\cap \mathcal{G}^{h-1})}, E_{\mathcal{I}\times \mathcal{I}}) \notag\\ & \stackrel{\text{(c)}}{=} \sum_{h=1}^\alpha \sum_{i\in T\cap\mathcal{S}^h} H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| Y_{\{i\}\times (T^c\cap \mathcal{G}^{h-1})}, E_{\mathcal{I}\times \mathcal{I}})\notag\\ & = \sum_{i\in T\cap\mathcal{S}^h} H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times T^c}\|E_{\mathcal{I}\times \mathcal{I}})\notag\\ & \le \sum_{i\in T\cap\mathcal{S}^h} H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times T^c}\|E_{\{i\}\times T^c}) \label{eqn4WENCon}\end{aligned}$$ where 1. follows from the fact that for each $h\in\{1, 2, \ldots, \alpha\}$ and each $i\in T^c\cap \mathcal{S}^h$, $$\begin{aligned} & I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_{\{\ell\}\times T^c\cap \mathcal{G}_h}\}_{\ell<i},E_{\mathcal{I}\times \mathcal{I}}) \notag\\ & = H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_{\{\ell\}\times T^c\cap \mathcal{G}_h}\}_{\ell<i},E_{\mathcal{I}\times \mathcal{I}}) \notag\\ &\qquad - H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}, \{ Y_{\{\ell\}\times T^c\cap \mathcal{G}_h}\}_{\ell<i},E_{\mathcal{I}\times \mathcal{I}}) \notag\\ &\stackrel{\eqref{markovChainWENNetwork}}{=} H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\|X_i,E_{\mathcal{I}\times \mathcal{I}}) - H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\|X_i,E_{\mathcal{I}\times \mathcal{I}})\notag\\ & = 0.\end{aligned}$$ 2. follows from the fact that for each $h\in\{1, 2, \ldots, \alpha\}$ and each $i\in T\cap \mathcal{S}^h$, $$\begin{aligned} & I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, \{ Y_{\{\ell\}\times T^c\cap \mathcal{G}_h}\}_{\ell<i}, E_{\mathcal{I}\times \mathcal{I}}) \notag\\ & \le H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| Y_{\{i\}\times T^c\cap \mathcal{G}^{h-1}}, E_{\mathcal{I}\times \mathcal{I}}) \notag\\ &\qquad -H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| X_{\mathcal{S}^h}, Y_{\mathcal{G}^{h-1}}, \{ Y_{\{\ell\}\times T^c\cap \mathcal{G}_h}\}_{\ell<i}, E_{\mathcal{I}\times \mathcal{I}}) \notag\\ &\stackrel{\eqref{markovChainWENNetwork}}{=} H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| Y_{\{i\}\times T^c\cap \mathcal{G}^{h-1}}, E_{\mathcal{I}\times \mathcal{I}}) - H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\|X_{i}, E_{\mathcal{I}\times \mathcal{I}})\notag\\ & \le I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(X_{i};Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}\| Y_{\{i\}\times T^c\cap \mathcal{G}^{h-1}}, E_{\mathcal{I}\times \mathcal{I}}).\end{aligned}$$ 3. follows from and that $Y_{\{i\}\times (T^c\cap \mathcal{G}_h)}$ is a function of $(X_i, E_{\mathcal{I}\times \mathcal{I}})$. Following and letting $\mathbf{1}^{T^c}$ denote the $\|T^c\|$-dimensional all-1 tuple, we consider the following chain of inequalities for each $h\in\{1, 2, \ldots, \alpha\}$ and each $i\in T\cap\mathcal{S}^h$: $$\begin{aligned} &H_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}(Y_{\{i\}\times T^c}\| E_{\{i\}\times T^c}) \notag\\ & = \Pr\{E_{\{i\} \times T^c} =\mathbf{1}^{T^c} \}H_{p_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\{i\} \times T^c}=\mathbf{1}^{T^c}) \notag\\ & \qquad + \Pr\{E_{\{i\} \times T^c} \ne \mathbf{1}^{T^c} \}H_{p_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\{i\} \times T^c}\ne\mathbf{1}^{T^c})\notag\\ & \stackrel{\text{(a)}}{=} \Pr\{E_{\{i\} \times T^c} \ne \mathbf{1}^{T^c} \}H_{p_{X_\mathcal{I}, Y_\mathcal{I}}}(X_i \|E_{\{i\} \times T^c}\ne\mathbf{1}^{T^c})\notag\\ & \le \Pr\{E_{\{i\} \times T^c} \ne \mathbf{1}^{T^c} \}\|\mathcal{X}_i\|\notag\\ &\stackrel{\text{(b)}}{=} \left(1-\prod_{j\in T^c}e_{i,j} \right)\|\mathcal{X}_i\| \label{eqnWENconditionalH=0eq5Con}\end{aligned}$$ where 1. follows from that for each $j\in \mathcal{I}$, $$Y_{i,j}= \begin{cases} \varepsilon & \text{if $E_{i,j}=1$,} \\ X_i & \text{otherwise.} \end{cases}$$ 2. follows from and . Combining and , we have $$\begin{aligned} & \sum_{h=1}^\alpha I_{p_{ X_{\mathcal{I}}, Y_{\mathcal{I}}}}( X_{T\cap \mathcal{S}^h}, Y_{T\cap \mathcal{G}^{h-1}} ; Y_{T^c}\| X_{T^c \cap \mathcal{S}^h}, Y_{T^c\cap \mathcal{G}^{h-1}}, E_{\mathcal{I}\times \mathcal{I}}) \notag\\ & \le \sum_{i\in T\cap \mathcal{S}^h} \left(1-\prod_{j\in T^c}e_{i,j} \right)\|\mathcal{X}_i\| \notag\\ & \le \sum_{i\in T} \left(1-\prod_{j\in T^c}e_{i,j} \right)\|\mathcal{X}_i\|. \label{eqnWENconditionalH=0eq6Con}\end{aligned}$$ Consequently, it follows from , and that holds, which implies from that $\mathcal{C}\subseteq \mathcal{R}_{\text{in}}^{\text{WEN}} $. Concluding Remarks {#conclusion} ================== We have investigated under the generalized-delay model three classes of delay-independent multimessage multicast networks (MMNs), namely the deterministic MMN dominated by product distributions, the MMN consisting of independent DMCs and the wireless erasure network respectively. We are able to evaluate the capacity regions for the above classes of MMNs with zero-delay nodes and demonstrate that their capacity regions coincide with the positive-delay regions, which implies that the above classes of MMNs with zero-delay nodes belong to the category of delay-independent MMNs. In other words, for each MMN with zero-delay nodes which belongs to one of the above three classes, the set of achievable rate tuples does not depend on the delay amounts incurred by the nodes in the network. This is in contrast to the fact that for some MMNs with zero-delay nodes, the set of achievable rate tuples shrinks if we impose the additional constraint that each node incurs a positive delay. An important implication of our result is that for each MMN belonging to one of the above three classes, using different methods for handling delay and synchronization does not affect the network capacity. Future research may continue the theme of this work – to identify other important classes of delay-independent and delay-dependent MMNs under the generalized-delay model. This work is limited to identifying delay-independent MMNs whose capacity regions lie in the corresponding cut-set bounds and at the same time the cut-set bounds can be achieved. The search of delay-independent and delay-dependent MMNs whose capacity regions are strictly smaller than the classical cut-set bounds is an interesting research direction. Another direction is exploring delay-dependent MMNs whose capacity regions are strictly larger than the classical cut-set bounds. Proof of Lemma \[cutsetMCLemma\\] {#appendixA} ================================== Fix a $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code, and let $p_{W_{\mathcal{I}},X_{\mathcal{I}}^n, Y_{\mathcal{I}}^n }$ be the joint distribution induced by the code according to Definitions \[defCode\] and \[defDiscreteMemoryless\]. For each $k\in\{1, 2, \ldots, n\}$, let $U^{k-1}\triangleq (W_{\mathcal{I}}, X_{\mathcal{I}}^{k-1}, Y_{\mathcal{I}}^{k-1})$ be the collection of random variables that are generated before the $k^{\text{th}}$ time slot for the $(B, n, M_{\mathcal{I}\times\mathcal{I}})$-code. In order to prove , it suffices to show that $$H_{p_{U^{k-1},X_{\mathcal{S}^h,k}, Y_{\mathcal{G}^h,k} }}(X_{\mathcal{S}^h,k}, Y_{\mathcal{G}^h,k} \| U^{k-1})=0 \label{cutsetstatement2}$$ holds for each $k\in\{1, 2, \ldots, n\}$ and each $h\in\{1, 2, \ldots, \alpha\}$, which will then imply that $$\begin{aligned} H_{p_{W_{\mathcal{I}}, X_{\mathcal{I}}^n, Y_{\mathcal{I}}^n}}(X_{\mathcal{I}}^n, Y_{\mathcal{I}}^n \| W_{\mathcal{I}}) &= \sum_{k=1}^n H_{p_{U^{k-1}, X_{\mathcal{I},k}, Y_{\mathcal{I},k}}}(X_{\mathcal{I},k}, Y_{\mathcal{I},k}\|U^{k-1}) \notag\\ &\stackrel{\eqref{cutsetstatement2}}{=}0.\end{aligned}$$ Fix a $k\in\{1, 2, \ldots, n\}$. We prove by induction on $h$ as follows. For $h=1$, the LHS of is $$\begin{aligned} &H_{p_{U^{k-1},X_{\mathcal{S}^1,k}, Y_{\mathcal{G}^1,k} }}(X_{\mathcal{S}^1,k}, Y_{\mathcal{G}^1,k} \| U^{k-1}) \notag\\ & = H_{p_{U^{k-1},X_{\mathcal{S}^1,k}}}(X_{\mathcal{S}^1,k}\| U^{k-1}) + H_{p_{U^{k-1},X_{\mathcal{S}^1,k},Y_{\mathcal{G}^1,k}}}(Y_{\mathcal{G}^1,k}\|U^{k-1},X_{\mathcal{S}^1,k})\notag\\ & \stackrel{\text{(a)}}{=} H_{p_{U^{k-1},X_{\mathcal{S}^1,k},Y_{\mathcal{G}^1,k}}}(Y_{\mathcal{G}^1,k}\|U^{k-1},X_{\mathcal{S}^1,k})\notag \\ & \le H_{p_{X_{\mathcal{S}^1,k}, Y_{\mathcal{G}^1,k}}}(Y_{\mathcal{G}^1,k}\|X_{\mathcal{S}^1,k}) \notag\\ & \stackrel{\eqref{memorylessStatement}}{=} H_{p_{X_{\mathcal{S}^1,k}}q_{Y_{\mathcal{G}^1,k}\|X_{\mathcal{S}^1,k}}^{(1)}}(Y_{\mathcal{G}^1,k}\|X_{\mathcal{S}^1,k}) \notag\\ & \stackrel{\text{(b)}}{=} 0,\label{MIstatement1}\end{aligned}$$ where 1. follows from Definitions \[defFeasible\] and \[defCode\] that $X_{\mathcal{S}^1,k}$ is a function of $U^{k-1}$ for the code. 2. follows from the fact that $q^{(1)}$ is deterministic (cf. Definition \[definitionDeterministicMMN\]). If holds for $h=m$, i.e., $$H_{p_{U^{k-1},X_{\mathcal{S}^m,k}, Y_{\mathcal{G}^m,k}}}(X_{\mathcal{S}^m,k}, Y_{\mathcal{G}^m,k} \| U^{k-1})=0 \label{assumption2}$$ then for $h=m+1$ such that $m+1\le \alpha$, the LHS of is $$\begin{aligned} &H_{p_{U^{k-1},X_{\mathcal{S}^{m+1},k}, Y_{\mathcal{G}^{m+1},k}}}(X_{\mathcal{S}^{m+1},k}, Y_{\mathcal{G}^{m+1},k} \| U^{k-1})\notag \\ & = H_{p_{U^{k-1},X_{\mathcal{S}^m,k},Y_{\mathcal{G}^{m},k}}}(X_{\mathcal{S}^m,k},Y_{\mathcal{G}^{m},k} \|U^{k-1} )+ H_{p_{U^{k-1},X_{\mathcal{S}^m,k},Y_{\mathcal{G}^{m},k} ,X_{\mathcal{S}_{m+1},k}}}(X_{\mathcal{S}_{m+1},k}\|U^{k-1},X_{\mathcal{S}^m,k},Y_{\mathcal{G}^{m},k} ) \notag\\ &\qquad + H_{p_{U^{k-1},X_{\mathcal{S}^{m+1},k},Y_{\mathcal{G}^{m+1},k}}}(Y_{\mathcal{G}_{m+1},k}\|U^{k-1},X_{\mathcal{S}^{m+1},k}, Y_{\mathcal{G}^{m},k}) \notag\\ & \stackrel{\eqref{assumption2}}{=} H_{p_{U^{k-1},X_{\mathcal{S}^m,k},Y_{\mathcal{G}^{m},k} ,X_{\mathcal{S}_{m+1},k}}}(X_{\mathcal{S}_{m+1},k}\|U^{k-1},X_{\mathcal{S}^m,k},Y_{\mathcal{G}^{m},k} ) \notag\\ &\qquad + H_{p_{U^{k-1},X_{\mathcal{S}^{m+1},k},Y_{\mathcal{G}^{m+1},k}}}(Y_{\mathcal{G}_{m+1},k}\|U^{k-1},X_{\mathcal{S}^{m+1},k}, Y_{\mathcal{G}^{m},k})\notag\\ & \stackrel{\text{(a)}}{=} H_{p_{U^{k-1},X_{\mathcal{S}^{m+1},k},Y_{\mathcal{G}^{m+1},k}}}(Y_{\mathcal{G}_{m+1},k}\|U^{k-1},X_{\mathcal{S}^{m+1},k}, Y_{\mathcal{G}^{m},k})\notag \\ & \le H_{p_{X_{\mathcal{S}^{m+1},k},Y_{\mathcal{G}^{m+1},k}}}(Y_{\mathcal{G}_{m+1},k}\|X_{\mathcal{S}^{m+1},k}, Y_{\mathcal{G}^{m},k}) \notag\\ &\stackrel{\eqref{memorylessStatement}}{=} H_{p_{X_{\mathcal{S}^{m+1},k},Y_{\mathcal{G}_{m},k}}q_{Y_{\mathcal{G}_{m+1},k}\|X_{\mathcal{S}^{m+1},k},Y_{\mathcal{G}_{m},k}}^{(m+1)}}(Y_{\mathcal{G}_{m+1},k}\|X_{\mathcal{S}^{m+1},k}, Y_{\mathcal{G}^{m},k})\notag\\ & \stackrel{\text{(b)}}{=} 0, \label{MIstatement2}\end{aligned}$$ where 1. follows from Definitions \[defFeasible\] and \[defCode\] that $X_{\mathcal{S}_{m+1},k}$ is a function of $(U^{k-1}, Y_{\mathcal{G}^{m},k})$ for the code. 2. follows from the fact that $q^{(m+1)}$ is deterministic (cf. Definition \[definitionDeterministicMMN\]). For $h=1$, it follows from that holds. For all $1\le m \le \alpha-1$, it follows from and that if is assumed to be true for $h=m$, then is also true for $h=m+1$. Consequently, it follows by mathematical induction that holds for all $1\le h\le \alpha$. Proof of the Achievability of Theorem \[thmWEN\] {#appendixB} ================================================ Our goal is to prove . Let $u_{X_i}$ be the uniform distribution on $\mathcal{X}_i$ for each $i\in\mathcal{I}$ and let $$u_{X_\mathcal{I}, Y_\mathcal{I}} = \left(\prod_{i=1}^N u_{X_i}\right)\left(\prod_{h=1}^\alpha q_{Y_{\mathcal{G}_h}\|X_\mathcal{S}^{h}, Y_\mathcal{G}^{h-1}}^{(h)}\right). \label{productDistributionWEN}$$ Fix any $T\subseteq\mathcal{I}$ such that $$T^c\cap \mathcal{D}\ne \emptyset. \label{eqnWENTCintersectD}$$ In order to apply Theorem \[theoremNNC\], we consider $$\begin{aligned} & H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_T\|X_\mathcal{I},Y_{T^c}) \notag\\ & \stackrel{\text{(a)}}{=} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_T\|X_\mathcal{I},Y_{T^c}, E_{\mathcal{I}\times\mathcal{I}}) \notag\\ & \stackrel{\text{(b)}}{=} 0 \label{eqnWENconditionalH=0eq1} \end{aligned}$$ and $$\begin{aligned} & I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_T;Y_{T^c}\|X_{T^c}) \notag\\* & \stackrel{\text{(c)}}{=} I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_T;Y_{T^c},E_{\mathcal{I}\times\mathcal{I}} \|X_{T^c}) \notag\\ & \stackrel{\text{(d)}}{=} I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_T;Y_{T^c} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}) \notag\\ & = H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{T^c} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}) - H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{T^c} \|X_{\mathcal{I}},E_{\mathcal{I}\times\mathcal{I}}) \notag\\ & \stackrel{\eqref{eqnWENconditionalH=0eq1}}{=} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{T^c} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}) \notag\\ & \stackrel{\eqref{defWENalphabetYm}}{=} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in \mathcal{I}\times T^c} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}) \notag\\ &\stackrel{\text{(e)}}{=} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\times T^c} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}) \label{eqnWENconditionalH=0eq2} % % &\stackrel{\text{(f)}}{=} \sum_{e_{\mathcal{I}\times \mathcal{I}}\in \mathcal{E}_{\mathcal{I}\times \mathcal{I}}} \Pr\{E_{\mathcal{I}\times \mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}\} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\times T^c} \|E_{\mathcal{I}\times\mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}) \notag\\ %% &\stackrel{\text{(f)}}{=} \sum_{e_{\mathcal{I}\times \mathcal{I}}\in \mathcal{E}_{\mathcal{I}\times \mathcal{I}}} \Pr\{E_{\mathcal{I}\times \mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}\} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\times T^c} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}) \notag\\ %&\stackrel{\text{(f)}}{=} \sum_{e_{\mathcal{I}\times \mathcal{I}}\in \mathcal{E}_{\mathcal{I}\times \mathcal{I}}} \Pr\{E_{\mathcal{I}\times \mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}\} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in \Omega\cap(T\times T^c)} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}), % \notag\\ % & \stackrel{\text{(g)}}{=} \sum_{h=1}^\alpha \sum_{e_{\mathcal{S}_h\times \mathcal{G}_h}\in \mathcal{E}_{\mathcal{S}_h\times \mathcal{G}_h}} \Pr\{E_{\mathcal{S}_h\times \mathcal{G}_h}=e_{\mathcal{S}_h\times \mathcal{G}_h}\} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\cap \mathcal{S}_h\times T^c \cap \mathcal{G}_h} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}) \notag\\ % & \stackrel{\text{(h)}}{=} \sum_{h=1}^\alpha \sum_{(i,j)\in T\cap \mathcal{S}_h\times T^c \cap \mathcal{G}_h} \Pr\{E_{\mathcal{S}_h\times \mathcal{G}_h}=e_{\mathcal{S}_h\times \mathcal{G}_h}\} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\cap \mathcal{S}_h\times T^c \cap \mathcal{G}_h} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}=e_{\mathcal{I}\times \mathcal{I}}) \end{aligned}$$ where 1. follows from Statements (i) and (ii) in the previous subsection and that $Y_{T^c}$ contains the random variable $E_{\mathcal{I}\times\mathcal{I}}$. 2. follows from and that $Y_T$ is a function of $(X_\mathcal{I}, E_{\mathcal{I}\times\mathcal{I}})$. 3. follows from Statements (i) and (ii) in the previous subsection and that $Y_{T^c}$ contains the random variable $E_{\mathcal{I}\times\mathcal{I}}$. 4. follows from that $X_\mathcal{I}$ and $E_{\mathcal{I}\times\mathcal{I}}$ are independent, i.e., $$I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_\mathcal{I};E_{\mathcal{I}\times\mathcal{I}})=0. \label{XEindependentDistU}$$ 5. follows from and that $\{Y_{i,j}\}_{(i,j)\in T^c\times T^c}$ is a function of $(X_{T^c}, E_{\mathcal{I}\times\mathcal{I}})$. In order to further simplify , consider the following chain of inequalities for any $T_1, T_2\subseteq\mathcal{I}$ such that $T_1\cap T_2 = \emptyset$: $$\begin{aligned} &I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T_1\times T_2}; X_{T_2}\|E_{\mathcal{I}\times \mathcal{I}}) \notag\\ & \le I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_{T_1}, \{Y_{i,j}\}_{(i,j)\in T_1\times T_2}; X_{T_2}\|E_{\mathcal{I}\times \mathcal{I}}) \notag\\ & \stackrel{\text{(a)}}{=} I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_{T_1}; X_{T_2}\|E_{\mathcal{I}\times \mathcal{I}}) \notag\\ &\stackrel{\text{(b)}}{=} I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_{T_1}; X_{T_2}) \notag\\ & \stackrel{\text{(c)}}{=} 0 \label{eqnWENconditionalH=0eq3}\end{aligned}$$ where 1. follows from the fact that $\{Y_{i,j}\}_{(i,j)\in T_1\times T_2}$ is a function of $(X_{T_1}, E_{\mathcal{I}\times \mathcal{I}})$. 2. follows from that $X_\mathcal{I}$ and $E_{\mathcal{I}\times \mathcal{I}}$ are independent. 3. follows from that $X_{T_1}$ and $X_{T_2}$ are independent. Following , consider the following chain of inequalities: $$\begin{aligned} &H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\times T^c} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}) \notag\\ &\stackrel{\text{(a)}}{=}H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\times T^c} \|X_{T^c},E_{\mathcal{I}\times\mathcal{I}}) + I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_{T^c};\{Y_{i,j}\}_{(i,j)\in T\times T^c} \|E_{\mathcal{I}\times\mathcal{I}}) \notag\\ & = H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\times T^c} \|E_{\mathcal{I}\times\mathcal{I}}) \notag\\ % &\stackrel{\text{(a)}}{=} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(\{Y_{i,j}\}_{(i,j)\in T\times T^c} \|E_{\mathcal{I}\times\mathcal{I}}) \notag\\ & = \sum_{i\in T} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\mathcal{I}\times\mathcal{I}}, \{Y_{m,\ell}\}_{m\in T, m<i, \ell\in T^c}) \notag\\ & \ge \sum_{i\in T} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\mathcal{I}\times\mathcal{I}}, \{Y_{m,\ell}\}_{m\in T, m<i, \ell\in T^c}, \{X_m\}_{m\in T, m<i}) \notag\\ & \stackrel{\text{(b)}}{=} \sum_{i\in T} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\mathcal{I}\times\mathcal{I}}, \{X_m\}_{m\in T, m<i})\notag\\ &\stackrel{\eqref{eqnWENconditionalH=0eq3}}{=}\sum_{i\in T} H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\mathcal{I}\times\mathcal{I}})\notag\\ & \stackrel{\text{(c)}}{\ge} \sum_{i\in T} (H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\{i\}\times T^c}) - I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(E_{\{i\}\times T^c};Y_{\{i\}\times T^c} \|X_{i}, E_{\{i\}\times T^c}))\notag\\ &\stackrel{\text{(d)}}{=}H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\{i\}\times T^c}) \label{eqnWENconditionalH=0eq4} \end{aligned}$$ where 1. follows from by letting $T_1 = T$ and $T_2=T^c$. 2. follows from and that $\{Y_{m,\ell}\}_{m\in T, m<i, \ell\in T^c}$ is a function of $(\{X_m\}_{m\in T, m<i}, E_{\mathcal{I}\times\mathcal{I}})$ . 3. follows from the fact that $$\begin{aligned} &I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(E_{\{i\}\times T^c};Y_{\{i\}\times T^c} \|E_{\{i\}\times T^c}) \notag\\ & \le I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(E_{\{i\}\times T^c};X_{i}, Y_{\{i\}\times T^c} \|E_{\{i\}\times T^c}) \notag\\ & \stackrel{\eqref{XEindependentDistU}}{=} I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(E_{\{i\}\times T^c};Y_{\{i\}\times T^c} \|X_{i}, E_{\{i\}\times T^c}). \end{aligned}$$ 4. follows from and that $Y_{\{i\}\times T^c}$ is a function of $(X_{i}, E_{\{i\}\times T^c})$. Following and letting $\mathbf{1}^{T^c}$ denote the $\|T^c\|$-dimensional all-1 tuple, we consider the following chain of equalities for each $i\in T$: $$\begin{aligned} &H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\{i\}\times T^c}) \notag\\ & = \Pr\{E_{\{i\}\times T^c} =\mathbf{1}^{T^c} \}H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\{i\}\times T^c}=\mathbf{1}^{T^c}) \notag\\ & \qquad + \Pr\{E_{\{i\}\times T^c} \ne \mathbf{1}^{T^c} \}H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(Y_{\{i\}\times T^c} \|E_{\{i\}\times T^c}\ne\mathbf{1}^{T^c})\notag\\ & \stackrel{\text{(a)}}{=} \Pr\{E_{\{i\}\times T^c} \ne \mathbf{1}^{T^c} \}H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_i \|E_{\{i\}\times T^c}\ne\mathbf{1}^{T^c})\notag\\ & \stackrel{\text{(b)}}{=} \Pr\{E_{\{i\}\times T^c} \ne \mathbf{1}^{T^c} \}H_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_i)\notag\\ & \stackrel{\text{(c)}}{=} \Pr\{E_{\{i\}\times T^c} \ne \mathbf{1}^{T^c} \}\|\mathcal{X}_i\|\notag\\ &\stackrel{\text{(d)}}{=} \left(1-\prod_{j\in T^c}e_{i,j} \right)\|\mathcal{X}_i\| \label{eqnWENconditionalH=0eq5}\end{aligned}$$ where 1. follows from that for each $j\in \mathcal{I}$, $$Y_{i,j}= \begin{cases} \varepsilon & \text{if $E_{i,j}=1$,} \\ X_i & \text{otherwise.} \end{cases}$$ 2. follows from that $X_\mathcal{I}$ and $E_{\mathcal{I}\times \mathcal{I}}$ are independent. 3. follows from that $X_i$ is uniform on $\|\mathcal{X}_i\|$. 4. follows from and . Combining , and , we have $$I_{u_{X_\mathcal{I}, Y_\mathcal{I}}}(X_T;Y_{T^c}\|X_{T^c}) \ge \sum_{i\in T} \left(1-\prod_{j\in T^c}e_{i,j} \right)\|\mathcal{X}_i\|. \label{eqnWENconditionalH=0eq6}$$ Using Theorem \[theoremNNC\], , and , we have $\mathcal{R}_{\text{in}}^{\text{WEN}} \subseteq \mathcal{C}_+$. Acknowledgment {#acknowledgment .unnumbered} ============== The author would also like to thank Associate Editor Sae-Young Chung and the two anonymous reviewers for their useful comments that improve the presentation of this work. [10]{} \[1\][\#1]{} url@samestyle \[2\][\#2]{} \[2\][[l@\#1=l@\#1\#2]{}]{} , “[Cut-set bounds for networks with zero-delay nodes]{},” [IEEE]{} Trans. Inf. Theory, vol. 61, no. 7, pp. 3837–3850, 2015. , “[Two-way communication channels]{},” in [Proc. the 4th Berkeley Symposium on Mathematical Statistics and Probability]{}, vol. 1, 1961, pp. 611–644. , “[Relay networks with delays]{},” [IEEE]{} Trans. Inf. Theory, vol. 53, pp. 3413–3431, Oct. 2007. , “[Causal relay networks and new cut-set bounds]{},” in [Proc. Allerton Conference on Communication, Control and Computing]{}, Sep. 2011, pp. 247–252. N. Ratnakar and G. Kramer, “The multicast capacity of deterministic relay networks with no interference,” [IEEE]{} Trans. Inf. Theory, vol. 52, no. 6, pp. 2425–2432, 2006. A. S. Avestimehr, S. N. Diggavi, and D. N. Tse, “Wireless network information flow: A deterministic approach,” [IEEE]{} Trans. Inf. Theory, vol. 57, no. 4, pp. 1872–1905, 2011. E. Perron, “[Information-theoretic secrecy for wireless networks]{},” Ph.D. dissertation, École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, 2009. , “[A theory of network equivalence — [Part I]{}: [Point-to-point]{} channels]{},” [IEEE]{} Trans. Inf. Theory, vol. 57, no. 2, pp. 972––995, 2011. , “Capacity of wireless erasure networks,” [IEEE]{} Trans. Inf. Theory, vol. 52, no. 3, pp. 789–804, 2006. , “[Noisy network coding]{},” [IEEE]{} Trans. Inf. Theory, vol. 57, no. 5, pp. 3132–3152, 2011. M. Yassaee and M. R. Aref, “[Slepian]{}–-[Wolf]{} coding over cooperative relay networks,” [IEEE]{} Trans. Inf. Theory, vol. 57, no. 6, pp. 3462–3482, 2011. , “[On dependence and delay: Capacity bounds for wireless networks]{},” in [Proc. [IEEE]{} Wireless Communications and Networking Conference (WCNC)]{}, [Shanghai, China]{}, Apr. 2012, pp. 550–554. G. Kramer, “Information networks with in-block memory,” [IEEE]{} Trans. Inf. Theory, vol. 60, no. 4, pp. 2105 – 2120, 2014. , [Information Theory and Network Coding]{}.1em plus 0.5em minus 0.4emSpringer, 2008. [^1]: Silas L. Fong is with the Department of Electrical and Computer Engineering, National University of Singapore, Singapore (e-mail: `[email protected]`). [^2]: This paper was presented in part at the International Symposium on Information Theory and Its Applications (ISITA), Melbourne, Australia, Oct., 2014.	{ "pile_set_name": "ArXiv" }
--- abstract: 'For the size of the largest component in a supercritical random geometric graph, this paper estimates its expectation which tends to a polynomial on a rate of exponential decay, and sharpens its asymptotic result with a central limit theory. Similar results can be obtained for the size of biggest open cluster, and for the number of open clusters of percolation on a box, and so on.' title: The Asymptotic Size of The Largest Component in Random Geometric Graphs with some applications --- Introduction ============ The size of the largest component is a basic property for random geometric graphs (RGGs) and has attracted much interest during the past years, including both theoretical studies [@penrose1995][@penrose1996][@penrose2001][@p10] and various applications [@Glauche2003][@p17][@p16][@Pishro2009]. This paper firstly investigates the asymptotic size of the largest component of RGG in the supercritical case. Given a set $\mathcal {X}\subset\mathbb{R}^d$, let $G(\mathcal {X};r)$ denote the undirected graph with vertex set $\mathcal {X}$ and with undirected edges which connect all those pairs $\{X,Y\}$ with $\parallel Y-X\parallel\leq r$, where $\\|\cdot\\|$ denotes the Euclidean norm ($l_2-norm$). The basic model of RGGs can be formulated as $G(\mathcal {X}_n;r_n)$, where $\mathcal{X}_n$ denotes $n$ points which are independently and uniformly distributed in a $d$-dimensional unit cube. To overcome the lack of spatial independence for the binomial point process $\mathcal{X}_n$, the model of continuum percolation must be introduced. Following Section 1.7 in [@p10], let $\mathcal{H}_{\lambda}$ be a homogeneous Poisson process of intensity $\lambda$ on $\mathbb{R}^{d}$. For $s>0$, define $B(s):=[0,s]^{d}$ and $\mathcal{H}_{\lambda,s}:=\mathcal{H}_{\lambda}\cap B(s).$ Following [@p10], we write the Poisson Boolean model as $G(\mathcal {H}_{\lambda,s};1)$. There exist some notations related to percolation must be introduced. Following Section 9.6 in [@p10], let $\mathcal{H}_{\lambda,0}$ denote the point process $\mathcal{H}_{\lambda}\cup\{\textbf{0}\}$, where $\textbf{0}$ is the origin in $\mathbb{R}^d$, and for $k\in\mathds{N}$, let $p_{k}(\lambda)$ denote the probability that the order of the component in $G(\mathcal{H}_{\lambda,0};1)$ containing the origin is equal to $k$. The $percolation$ $probability$ $p_{\infty}(\lambda)$ is defined to be the probability that $\textbf{0}$ lies in an infinite component of the graph $G(\mathcal{H}_{\lambda,0};1)$. Therefore, we have $p_{\infty}(\lambda)=1-\sum\limits_{k=1}^{\infty}p_{k}(\lambda)$. Let $$\begin{aligned} \label{lc} \lambda_c=inf\{\lambda>0:p_{\infty}(\lambda)>0\}\end{aligned}$$ denote the critical intensity of continuum percolation. It is well known that $0<\lambda_c<\infty$ for $d\geq 2$ [@p12][@p13][@p14]. Following Section 9.6 in [@p10], let $L_j(G)$ denote the order of its $j$th-largest component for any graph $G$. Then $L_1(G(\mathcal {H}_{\lambda,s};1))$ denotes the order of the largest component of $G(\mathcal {H}_{\lambda,s};1)$. The asymptotic properties of $L_1(G(\mathcal{H}_{\lambda,s};1))$ have been well studied by Penrose. The basic asymptotic result about $L_1 (G(\mathcal{H}_{\lambda,s};1))$ is provided by Penrose (Theorem 10.9 in [@p10]), that if $\lambda\neq\lambda_c$ then $$\begin{aligned} \label{o1} s^{-d}L_1(G(\mathcal{H}_{\lambda,s};1))\xrightarrow{P}\lambda p_{\infty}(\lambda)\quad as \quad s\rightarrow \infty.\end{aligned}$$ Also, Penrose has given a central limit theorem for $L_1(G(\mathcal{H}_{\lambda,s};1))$ in the supercritical case $\lambda>\lambda_c$ (Theorem 10.22 in [@p10]), that $$\begin{aligned} \label{clt_1} s^{-d/2}(L_1(G(\mathcal{H}_{\lambda,s};1))-E[L_1(G(\mathcal{H}_{\lambda,s};1))]) \xrightarrow{D} \mathcal {N}(0,\sigma^2).\end{aligned}$$ However, the question as how large $E[L_1(G(\mathcal{H}_{\lambda,s};1))]$ should be still remains unsolved. By (\[o1\]) it can be deduced that $E[L_1(G(\mathcal{H}_{\lambda,s};1))]=\lambda p_{\infty}(\lambda) s^d +o(s^d)$, where $f(s)=o(g(s))$ indicates that $\lim_{s\rightarrow\infty}\frac{f(s)}{g(s)}=0$. This result is not precise enough for some theoretic analysis and practical applications. The corresponding asymptotic results and central limit theorem for $G(\mathcal {X}_n;r_n)$ have also been established by Peorose (Theorems 11.9 and 11.16 in [@p10]), but we may ask similar questions. This paper will study the problem and give a more precise description for the asymptotic sizes of $L_1(G(\mathcal{H}_{\lambda,s};1))$ and $L_1(G(\mathcal {X}_n;r_n))$. Our method can be adapted to study some other models and problems. Main Results ============ Our main results can be formulated as the following two theorems. \[t1\] Suppose $d\geq 2$ and $\lambda>\lambda_c$. Then there exist constants $c=c(d,\lambda)>0$ and $\tau_i=\tau_i(d,\lambda)$, $1\leq i \leq d$, with $\tau_1>0$, such that for all $s$ large enough, $$\begin{aligned} \label{order_t1_00} E[L_1(G(\mathcal{H}_{\lambda,s};1))]=\lambda p_{\infty}(\lambda) s^d-\sum_{i=1}^d \tau_i s^{d-i}+o\left(e^{-c s}\right).\end{aligned}$$ Also, there exists a constant $\sigma=\sigma(d,\lambda) > 0$, such that $$\begin{aligned} \label{order_t1_01} L_1(G(\mathcal{H}_{\lambda,s};1))s^{-d/2}- \lambda p_{\infty}(\lambda) s^{d/2}+ \sum_{i=1}^{\lfloor\frac{d}{2} \rfloor}\tau_i s^{d/2-i} \xrightarrow{D} \mathcal {N}(0,\sigma^2)\end{aligned}$$ as $s\rightarrow\infty$. \[t2\] Suppose $d\geq 2$ and $\lambda>\lambda_c$. Let $\sigma$ and $\tau_i$ be the same constants appearing in Theorem \[t1\]. There exists a constant $\delta=\delta (d,\lambda)$, with $0<\delta\leq\sigma$, such that $$\begin{aligned} L_1\left(G\left(\mathcal{X}_n;(n/\lambda)^{-1/d}\right)\right)\left(n/\lambda\right)^{-1/2}- p_{\infty}(\lambda) \left(\lambda n\right)^{1/2}+\sum_{i=1}^{\lfloor\frac{d}{2} \rfloor} \tau_i \left(n/\lambda\right)^{\frac{1}{2}-\frac{i}{d}} \xrightarrow{D} \mathcal {N}(0,\delta^2)\end{aligned}$$ as $n\rightarrow\infty$. To prove the two theorems, we estimate the value of $E[L_1(G(\mathcal{H}_{\lambda,s};1))]$ firstly, and then using the central limit theorems for $L_1(G(\mathcal{H}_{\lambda,s};1))$ and $L_1(G(\mathcal{X}_n;(n/\lambda)^{-1/d}))$, we can prove (\[order\_t1\_01\]) and Theorem \[t2\]. Some notations must be stated before the proof of our results. For any $x\in \mathbb{R}^d$, we write its $l_{\infty}$ norm with $\\|x\\|_{\infty}$ given by the maximum absolute value of its coordinates. For any finite set $A\subset \mathbb{R} ^d$, we set the diameter of $A$ by diam$(A)=\sup_{x,y\in A}\\|x-y\\|_{\infty}.$ Also, let $\|A\|$ denote the cardinality of $A$. Let $\oplus$ denote the Minkowski addition of sets. Let $Leb(\cdot)$ denote the Lebesgue measure. For $s \geq 0$, let $\lfloor s \rfloor$ denote the smallest integer not smaller than $s$. To simplify the expression, we will omit the dependence of all constants on $d$ and $\lambda$, for example, the constant $c$ stands for $c(d,\lambda)$. Given $\lambda>\lambda_c$, by the uniqueness of the infinite component in continuum percolation (Theorem 9.19 in [@p10]), the infinite graph $G(\mathcal{H}_{\lambda};1)$ has precisely one infinite component $\mathcal {C}_{\infty}$ with probability $1$. Let $C_1,C_2,...,C_M$ denote the components of $G(\mathcal {C}_{\infty}\cap B(s);1)$, taken in a decreasing order. We give a result on the rate of sub-exponential decay of the difference between $E[L_1(G(\mathcal{H}_{\lambda,s};1))]$ and $E[\|C_1\|]$. \[temp1\] Suppose $d\geq 2$ and $\lambda>\lambda_c$. The exists a constant $c>0$, such that for large enough $s$, $$\begin{aligned} \label{temp1_0} 0\leq E[L_{1}(G(\mathcal{H}_{\lambda,s};1))]-E[\|C_1\|] \leq e^{-cs}.\end{aligned}$$ By the definition of $L_{1}(G(\mathcal{H}_{\lambda,s};1))$ and $C_1$, obviously $E[L_{1}(G(\mathcal{H}_{\lambda,s};1))]\geq E[\|C_1\|]$. Thus it just remains to prove the second inequality of (\[temp1\_0\]). Given any $x\in\mathbb{R}^d$, let $C_{\infty}(x)$ denote the infinite connected component of $G(\mathcal{H}_{\lambda}\cup\{x\};1)$. By Palm theorem for Poisson processes (Theorem 1.6 in [@p10]), we have $$\begin{aligned} \label{order_t1_6} E[L_{1}(G(\mathcal{H}_{\lambda,s};1))]=\lambda\int_{B(s)}P[x\in V_{1}(x)]dx,\end{aligned}$$ where $V_{1}(x)$ denotes the largest component of $G(\mathcal{H}_{\lambda,s}\cup\{x\};1)$, and $$\begin{aligned} \label{order_t1_7} E[\|C_1\|]=\lambda\int_{B(s)}P[x\in C_{1}(x)]dx,\end{aligned}$$ where $C_{1}(x)$ denotes the largest component of $C_{\infty}(x) \cap B(s)$. Therefore, $$\begin{aligned} \label{order_t1_8} \begin{aligned} E[L_{1}(G(\mathcal{H}_{\lambda,s};1))]-E[\|C_1\|]&=\lambda\int_{B(s)}(P[x\in V_{1}(x)]-P[x\in C_{1}(x)])dx\\ &\leq \lambda\int_{B(s)} P[\{x\in V_{1}(x)\} \cap \{ x\notin C_1(x)\}]dx\\ &= \lambda\int_{B(s)} P[\{x\in V_{1}(x)\} \cap \{ x\notin C_{\infty}(x)\}]dx. \end{aligned}\end{aligned}$$ Suppose $0<\varepsilon<\frac{1}{2}$. By Theorem 10.19 in [@p10], there exist constants $c_{1}>0$ and $s_{1}>0$, such that if $s>s_1$ then $$\begin{aligned} \label{order_t1_9} \begin{aligned} P\left [\|V_{1}(x)\|<(1-\varepsilon)\lambda s^{d}p_{\infty}(\lambda) \right ] &\leq P\left [L_{1}(G(\mathcal{H}_{\lambda,s};1))<(1-\varepsilon)\lambda s^{d}p_{\infty}(\lambda)\right ]\\ &\leq \exp \left (-c_{1}s^{d-1} \right ). \end{aligned}\end{aligned}$$ Also, by Theorem 10.15 in [@p10], there exists a constant $c_2>0$ such that for $s$ large enough, $$\begin{aligned} \label{order_t1_10} \sum\limits_{k\geq \lceil(1-\varepsilon)\lambda s^{d}p_{\infty}(\lambda)\rceil}p_{k}(\lambda)<\exp \left (-c_2[(1-\varepsilon)\lambda s^{d}p_{\infty}(\lambda)]^{(d-1)/d} \right ).\end{aligned}$$ Therefore, from (\[order\_t1\_9\]) and (\[order\_t1\_10\]) we can obtain $$\begin{aligned} \label{order_t1_11} &&P[\{x\in V_{1}(x)\}\cap\{x\not\in C_{\infty}(x)\}]\nonumber\\ &&~~~~\leq P[\|V_{1}(x)\|<(1-\varepsilon)\lambda s^{d}p_{\infty}(\lambda)]\nonumber\\ &&~~~~~~~~~+P[\{x\in V_{1}(x)\}\cap\{x\not\in C_{\infty}(x)\}\cap\{\|V_{1}(x)\|\geq(1-\varepsilon)\lambda s^{d}p_{\infty}(\lambda)\}]\nonumber\\ &&~~~~\leq \exp \left (-c_{1}s^{d-1} \right )+\sum\limits_{k\geq \lceil(1-\varepsilon)\lambda s^{d}p_{\infty}(\lambda)\rceil}p_{k}(\lambda)\nonumber\\ &&~~~~< \exp \left (-c_{1}s^{d-1} \right )+ \exp \left (-c_2[(1-\varepsilon)\lambda p_{\infty}(\lambda)]^{(d-1)/d} s^{d-1} \right )~~as~s\rightarrow\infty.\end{aligned}$$ Combined with (\[order\_t1\_8\]) this yields our result. To estimate the value of $E[L_1(G(\mathcal{H}_{\lambda,s};1))]$, by Lemma \[temp1\] we just need to get the value of $E[\|C_1\|]$ instead. Actually, by Palm theory for infinite Poisson process (Theorem 9.22 in [@p10]), $$\begin{aligned} \label{comm_6} E\left [\sum_{i=1}^M \|C_i\| \right ]=E[\left\|\mathcal {C}_{\infty}\cap B(s)\right\|]=\lambda p_{\infty}(\lambda)s^d,\end{aligned}$$ so we just need to estimate the value of $E[\sum_{i=2}^M \|C_i\|]$. Let $L(s):=B(s)\backslash[1,s-1]^d$. For any $2\leq i \leq M$, since $C_i\subset \mathcal {C}_{\infty}$, therefore there exists at least one point in $L(s)\cap C_i$ which connects to $\mathcal {C}_{\infty} \setminus B(s)$ directly; we choose the nearest one to the boundary of $B(s)$ as the $out-connect$ $point$. We can see that each component of $C_2,...,C_M$ contains exactly one out-connect point. For any region $R\subseteq B(s)$ and $2\leq i \leq M$, define $$\label{chidef} \chi_{i}(R):=\left\{ \begin{array}{ll} 1, & \mbox{if the out-connect point of $C_i$ is contained by } R, \\ 0, & \mbox{otherwise}, \end{array} \right.$$ and define $$\label{xidef} \xi(R)=\xi(R,s):= \sum\limits_{i=2}^{M}\chi_{i}(R)\|C_i\|.$$ By the definition of $\xi(\cdot)$, it is easy to see that for any $R,\widetilde{R}\subset B(s)$, if $Leb(R\cap \widetilde{R})=0$, then $E[\xi(R\cap \widetilde{R})]=0$ and $E[\xi(R\cup \widetilde{R})]=E[\xi(R)]+E[\xi(\widetilde{R})].$ For $0 \leq i \leq d-1$, define $$R_i=R_i(s):=[0,1]\times\underbrace{[0,s/2]\times\cdots\times[0,s/2]}\limits_{d-1-i} \times \underbrace{[1,s/2]\times\cdots\times[1,s/2]}\limits_{i}.$$ Noted that $[1,s/2]^d \cap L(s)=\emptyset$, then by symmetry, $$\begin{aligned} \label{total_1} \begin{aligned} &E\left [\sum\limits_{i=2}^M \|C_i\| \right ]=E\left[\xi(B(s))\right]=2^dE\left[\xi\left(\left[0,\frac{s}{2}\right]^d\right)\right]\\ &=2^d\left\{E\left[\xi(R_0)\right]+E\left[\xi\left(\left[1,\frac{s}{2}\right]\times\left[0,\frac{s}{2}\right]^{d-1}\right)\right]\right\} = 2^d\sum\limits_{i=0}^{d-1} E\left[\xi(R_i)\right]. \end{aligned}\end{aligned}$$ Thus, we just need to estimate the value of $E\left[ \xi\left(R_i \right) \right]$. The following Lemmas \[exponent2\]-\[limit\] are introduced to get the desired estimation. \[exponent2\] Suppose $d\geq 2$ and $\lambda>\lambda_c$. Let $V_x=V_x(s)$ denote the connected component containing $x$ of $G(\mathcal{H}_{\lambda,s}\cup \{x\};1)$. There exist constants $c>0$ and $n_0>0$, such that if $n>n_0$ and $s>2n$ then for any point $x \in B(s)$, $$\begin{aligned} \label{exp00} P \left [ n \leq \mbox{diam}(V_x) \leq s/2 \right ] < e^{ -c n },\end{aligned}$$ and $$\begin{aligned} \label{exp01} P \left [ \left \{ \|V_x\| \geq n \right \} \cap \left \{ \mbox{diam}(V_x) \leq s/2 \right \} \right ] < \exp \left ( -c n^{(d-1)/d} \right ).\end{aligned}$$ The proof uses ideas from the latter part of the proof of Theorem 10.18 in [@p10]. Given $x\in \mathbb{R}^d$, let $\widetilde{z}$ denote the point in $B_{\mathbb{Z} }'(n(s))$ satisfying $x\in B_{\widetilde{z}}$, where the definition of $B_{\mathbb{Z} }'(n(s))$ and $ B_{\widetilde{z}}$ is given in pp.216 and pp.217 of [@p10] respectively. Also, $C_x$, $D_{ext}C_x$, $M_0$, $n(s)$ and $M(s)$ are defined as same as those appearing in pp.218-219 of [@p10]. Penrose has proved that $D_{ext}C_x$ is $-$connected and if $\|C_x\|<n(s)^d/2$ then $$\begin{aligned} \label{exp22_1} \begin{aligned} \|D_{ext}C_x\| \geq (2d)^{-1} (1- ({\textstyle \frac{2}{3}})^{1/d})\|C_x\|^{(d-1)/d}, \end{aligned}\end{aligned}$$ see pp.219 of [@p10]. Let $\mathcal{A}_{m,s}$ denote the collection of $-$connected subsets of cardinality $m$ which disconnects the point $\widetilde{z}$ from the giant component of $B_{\mathbb{Z} }'(n(s))$. Then $\mathcal{A}_{m,s}$ is restricted by the box of $B_{\mathbb{Z} }'(n(s))\cap ([-m,m]^d \oplus \widetilde{z})$ and $D_{ext}C_x \in \mathcal{A}_{\|D_{ext}C_x\|,s}$. By a Peierls argument (Corollary 9.4 in [@p10]), the cardinality $\|\mathcal{A}_{m,s}\|$ is bounded by $(2m+1)^d \gamma^m$, with $\gamma:=2^{3^d}$. Therefore, there exists a constant $k_0$ such that for any integer $k>k_0$, $$\begin{aligned} \label{exp23} \begin{aligned} P \left [\|D_{ext}C_x\|\geq k \right ] &\leq P \left [ \bigcup\limits_{m \geq k} \bigcup\limits_{\sigma \in \mathcal{A}_{m,s} } \{X_z=0, \forall z\in\sigma\} \right ]\\ &\leq \sum\limits_{m \geq k } (2m+1)^d \gamma^m (1-p_1)^m < (\frac{2}{3})^{k}. \end{aligned}\end{aligned}$$ By the definition of $C_x$ and $D_{ext}C_x$, if $n\leq$diam$(V_x)\leq s/2$ then $$\frac{n}{M(s)}-1\leq\mbox{diam}(C_x)\leq \frac{n(s)}{2}+2,$$ and therefore we can get $\|C_x\| < n(s)^d/2$ and $\|D_{ext}C_x\|\geq \frac{n}{M(s)}-1$ for large $s$. Therefore, by (\[exp23\]), there exists a constant $n_0>0$, such that if $n>n_0$ then, $$\begin{aligned} P \left [ n\leq \mbox{diam}(V_x) \leq \frac{s}{2}\right ] \leq P \left [\|D_{ext}C_x\|\geq \frac{n}{M(s)}-1 \right ] <\left (\frac{2}{3} \right )^{\frac{n}{2M_0}-1}.\end{aligned}$$ This yields (\[exp00\]). It remains to consider the case of $\|V_x\|>n$. Since $C_x$ is a $-$connected component containing $\widetilde{z}$ in $B_{\mathbb{Z} }'(n(s))$, by a Peierls argument (Lemma 9.3 in [@p10]), for all $k$, the number of $-$ connected subsets of $B_{\mathbb{Z} }'(n(s))$ of cardinality $k$ containing $\widetilde{z}$ is at most $\gamma^k$. Let $c_2\geq e^2 (2M_0)^d \lambda$. If $\|C_x\|<k$ and $\|V_x\|\geq c_2 k+1$, then for at least one of these subsets of $B_{\mathbb{Z} }'(n(s))$ the union of the associated boxes $B_z$ contains at least $c_2 k$ points of $\mathcal{H}_{\lambda}$. Therefore, by Lemma 1.2 in [@p10], we have $$\begin{aligned} \label{exp24} \begin{aligned} P[\{\|C_x\|<k\}\cap\{\|V_x\|\geq c_2 k+1\}] &< \gamma^k P \left [Po \left (k(2M_0)^d\lambda \right )\geq c_2 k \right ]\\ &\leq \gamma^k \exp \left \{-\left (\frac{c_2 k}{2} \right ) \log \left ( \frac{c_2}{(2M_0)^d \lambda } \right ) \right \}. \end{aligned}\end{aligned}$$ So if $c_2$ is chosen large enough, this probability decays exponentially in $k$. Set $\beta:=(2d)^{-1} (1- (\frac{2}{3})^{1/d})$. By (\[exp22\_1\]) and (\[exp23\]), we have $$\begin{aligned} P[\{\mbox{diam}(V_x) \leq s/2 \} \cap \{ \|C_x\| \geq k \} ] \leq P \left [\|D_{ext}C_x\| \geq \beta k^{(d-1)/d} \right ]< (\frac{2}{3})^{\beta k^{(d-1)/d}}.\end{aligned}$$ Combined with (\[exp24\]), this gives (\[exp01\]). \[expectation\] Suppose $d\geq 2$ and $\lambda>\lambda_c$. Then $$\begin{aligned} 0<\sum\limits_{n=1}^{\infty}nP(\|V_0\|=n)<\infty.\end{aligned}$$ By Lemma \[exponent\], there exist two constants $c>0$ and $n_0>0$, such that for all $n>n_0$, $$\begin{aligned} n^{-(d-1)/d}\log P( n\leq \|V_0\|<\infty)<-c.\end{aligned}$$ From this it can be deduced that $$\begin{aligned} \sum\limits_{n=n_0+1}^{\infty}P(n\leq \|V_0\|<\infty)< \sum\limits_{n=n_0+1}^{\infty}e^{-cn^{(d-1)/d}}<\infty.\end{aligned}$$ Therefore, we have $$\begin{aligned} \begin{aligned} \sum\limits_{n=1}^{\infty}nP(\|V_0\|=n)&=\sum\limits_{n=1}^{\infty}P(n\leq \|V_0\|<\infty)\\ &\leq n_0+\sum\limits_{n=n_0+1}^{\infty}P(n\leq \|V_0\|<\infty)\\ &< \infty. \end{aligned}\end{aligned}$$ In the following we prove that $P(1\leq \|V_0\|<\infty)>0$. Define $\widetilde{\tau}=\widetilde{\tau} (\lambda):=\sum\limits_{n=1}^{\infty}nP(\|V_0\|=n)$. For $x\in B(s)$ and $0<a\leq 1$, define the box $$\begin{aligned} B_i(x,a):=x\oplus \big(\underbrace{[0,1]\times\cdots\times[0,1]}\limits_i \times \underbrace{[0,a]\times\cdots\times[0,a]}\limits_{d-i}\big).\end{aligned}$$ Also, for any region $R\subseteq B(s)$, define $$\begin{aligned} \label{prop1_1} D(R)=D(R,s):=\max_{2\leq j \leq M,\chi_{j}(R)=1} \mbox{diam} (C_j).\end{aligned}$$ \[prop1\] Suppose $d\geq 2$ and $\lambda>\lambda_c$. There exist constants $c>0$ and $n_0>0$, such that if $x\in B(s)$, $a\in(0,1]$ and $n>n_0$ then $$\begin{aligned} \label{prop1_000} P[D(B_i(x,a))\geq n]<e^{-c n},\end{aligned}$$ and $$\begin{aligned} \label{prop1_00} P[\xi(B_i(x,a))\geq n]<\exp \left (-c n^{(d-1)/d} \right )+e^{-c s}.\end{aligned}$$ Let $W_1$ denote the number of the connected components which intersect with $B_i(x,a)$, and have metric diameter not greater than $s/2$ but not smaller than $n$. By Markov’s inequality, $$\begin{aligned} \label{prop1_02} P \left [\left \{D(B_i(x,a)) \geq n \right \} \cap \left \{ D(B_i(x,a)) \leq s/2 \right \} \right ]\leq P[W_1>0] \leq E[W_1].\end{aligned}$$ By Palm theory for Poisson process and Lemma \[exponent2\], if $n>n_0$ then $$\begin{aligned} \label{prop1_03} \begin{aligned} E[W_1]&=\lambda\int_{B_i(x,a)} P \left [ \left \{ \mbox{diam}(V_x(s)) \geq n \right \} \cap \left \{ \mbox{diam}(V_x(s)) \leq s/2 \right \} \right ]dx\\ &<\lambda a^{d-i} e^{-c n}. \end{aligned}\end{aligned}$$ Also, $C_i$ ($2\leq i \leq M$) is not the largest component of $G(\mathcal{H}_{\lambda,s};1)$, then by Proposition 10.13 in [@p10], there exist constants $c_1>0$ and $s_1>0$, such that if $s>s_1$ then $$\begin{aligned} \label{prop1_04} && P \left [ D(B_i(x,a)) > s/2 \right ] <e^{-c_1 s} .\end{aligned}$$ Together with (\[prop1\_02\]), (\[prop1\_03\]) and (\[prop1\_04\]), we obtain $$\begin{aligned} P[D(B_i(x,a))\geq n]<e^{-c n}+e^{-c_1 s}.\end{aligned}$$ Since $P[D(B_i(x,a))>s]=0$, thus (\[prop1\_000\]) follows. Note that $B_i(x,a)$ contains at most $2^d$ connected components. Thus, if $\xi(B_i(x,a)) \geq n$, by the definition of $\xi(\cdot)$, there exists at least one component intersecting with $B_i(x,a)$ such that it contains no less than $2^{-d}n$ points. Let $W_2$ be the number of the connected components which intersect with $B_i(x,a)$, and have more than $2^{-d}n$ elements and not larger than $s/2$ metric diameter. With the similar argument as (\[prop1\_02\]) and (\[prop1\_03\]), we get if $n>n_0$ then $$\begin{aligned} &&P \left [\left \{ \xi(B_i(x,a)) \geq n \right \} \cap \left \{ D(B_i(x,a)) \leq s/2 \right \} \right ] \leq E[W_2]\\ &&~~=\lambda\int_{B_i(x,a)} P \left [ \left \{ \|V_x(s)\| \geq 2^{-d}n \right \} \cap \left \{ \mbox{diam}(V_x(s)) \leq s/2 \right \} \right ]dx\\ &&~~<\lambda a^{d-i} \exp\left(-c2^{-d} n\right),\end{aligned}$$ together with (\[prop1\_04\]) this gives (\[prop1\_00\]). Let real numbers $s_1>2$ and $s_2>2$ be given. Let points $x=(x_1,x_2,\ldots,x_d)\in [0,s_1/2]^d$ and $\widetilde{x}=(\widetilde{x}_1,\widetilde{x}_2,\ldots,\widetilde{x}_d)\in [0,s_2/2]^d$ be given. For all $1\leq j \leq d$, define $$\begin{aligned} N_{x,\widetilde{x}}^j(s_1,s_2):=\left\{ \begin{array}{ll} \min(s_1,s_2)-x_j-1, & \mbox{if } x_j=\widetilde{x}_j, \\ \min(x_j,\widetilde{x}_j,s_1-x_j-1,s_2-\widetilde{x}_j-1), & \mbox{otherwise}, \end{array} \right.\end{aligned}$$ and let $$\begin{aligned} \label{prop1_1} N_{x,\widetilde{x}}(s_1,s_2):=\min_{1\leq j\leq d}\lfloor N_{x,\widetilde{x}}^j(s_1,s_2)\rfloor.\end{aligned}$$ \[argument\] Let us assume $d \geq 2$, $\lambda>\lambda_c$, $1\leq i \leq d$ and $0<a\leq 1$. There exist constants $c>0$ and $n_0>0$, such that if $x\in [0,s_1/2]^d$, $\widetilde{x}\in [0,s_2/2]^d$ and $N_{x,\widetilde{x}}(s_1,s_2)>n_0$ then $$\begin{aligned} &&\left\| E\left[\xi(B_i(x,a),s_1)\right]- E\left[\xi(B_i(\widetilde{x},a),s_2)\right]\right\|<\exp\left(-c N_{x,\widetilde{x}}(s_1,s_2) \right).\end{aligned}$$ Let $B'(s_2):=B(s_2)\oplus\{x-\widetilde{x}\},$ and let $\widetilde{C}_1,\widetilde{C}_2,\ldots,\widetilde{C}_{\widetilde{M}}$ denote the components of $G(\mathcal{C}_{\infty}\cap B'(s_2);1),$ taking in order of decreasing order. For any region $R\subseteq B'(s_2)$ and $2\leq i \leq \widetilde{M}$, define $$\widetilde{\chi_{i}}(R):=\left\{ \begin{array}{ll} 1, & \mbox{if the out-connect point of $\widetilde{C}_i$ is contained by } R, \\ 0, & \mbox{otherwise}. \end{array} \right.$$ Let $\widetilde{\xi}(R,s_2):= \sum_{i=2}^{\widetilde{M}}\widetilde{\chi}_{i}(R)\|\widetilde{C}_i\|$ and define $$\begin{aligned} \widetilde{D}(R,s_2):=\max_{2\leq j \leq \widetilde{M},\widetilde{\chi}_{j}(R)=1} \mbox{diam} (\widetilde{C}_j).\end{aligned}$$ According to the ergodicity of Poisson point processes, we can get $$\begin{aligned} \label{argu_1} P\left[\widetilde{\xi}\left(B_i(x,a),s_2\right)=k\right]=P\left[\xi\left(B_i(\widetilde{x},a),s_2\right)=k\right],~~~~\forall~k\geq 1.\end{aligned}$$ ![If $C_k$ connects with $\mathcal{H}_{\lambda}\cap \Delta$, the event of $\xi(B_i(x,a),s_1)\neq \widetilde{\xi}(B_i(x,a),s_2)$ may happen.[]{data-label="Lemmagraph"}](13295fig1.eps){width="3in"} Let $\Delta:=B(s_1)\cup B'(s_2)-B(s_1)\cap B'(s_2).$ If $\xi(B_i(x,a),s_1)\neq \widetilde{\xi}(B_i(x,a),s_2)$, then there exists at least one component among $C_2,\ldots,C_M,\widetilde{C}_2,\ldots,\widetilde{C}_{\widetilde{M}}$ which connects directly with $\mathcal{H}_{\lambda}\cap \Delta$, see Figure \[Lemmagraph\]. For simplicity of exposition, we take $N=N_{x,\widetilde{x}}(s_1,s_2)$, $\xi_1=\xi(B_i(x,a),s_1)$ and $\xi_2=\widetilde{\xi}(B_i(x,a),s_2)$. Therefore, by (\[prop1\_000\]), if $N>n_0+1$ then $$\begin{aligned} \label{argu_2} \begin{aligned} P\left[\xi_1 \neq \xi_2\right]&\leq P\left[ \left\{D(B_i(x,a),s_1)\geq N-1 \right\} \cup \left\{\widetilde{D}(B_i(x,a),s_2)\geq N-1 \right\} \right]\\ &< 2e^{-c(N-1)}. \end{aligned}\end{aligned}$$ Also, $$\begin{aligned} \label{argu_2_1} \begin{aligned} &P\left[\left\{\xi_1=k\right\} \cap \left\{\xi_2\neq k \right\} \right]+P\left[\left\{\xi_1\neq k\right\} \cap \left\{\xi_2=k \right\} \right]\\ &~~=P\left[\left\{\xi_1=k\right\}\right]+P\left[\left\{\xi_2=k\right\}\right]-2P\left[\left\{\xi_1=k\right\} \cap \left\{\xi_2=k \right\} \right]\\ &~~\geq \|P\left[\left\{\xi_1=k\right\}\right]-P\left[\left\{\xi_2=k\right\}\right]\|, \end{aligned}\end{aligned}$$ so by (\[argu\_2\]) and (\[argu\_2\_1\]) we have $$\begin{aligned} \label{argu_3} \begin{aligned} &\sum_{k=1}^{\infty} \left\|P\left[\left\{\xi_1=k\right\}\right]-P\left[\left\{\xi_2=k\right\}\right]\right\|\\ &~~\leq \sum_{k=1}^{\infty} \left( P\left[\left\{\xi_1=k\right\} \cap \left\{\xi_1\neq \xi_2 \right\} \right]+P\left[\left\{\xi_2=k \right\} \cap \left\{\xi_1 \neq \xi_2 \right\} \right] \right)\\ &~~=P\left[\left\{\xi_1\geq 1 \right\} \cap \left\{\xi_1\neq \xi_2 \right\} \right]+P\left[\left\{\xi_2\geq 1 \right\} \cap \left\{\xi_1 \neq \xi_2 \right\} \right]<4e^{-c(N-1)}. \end{aligned}\end{aligned}$$ Thus, by (\[argu\_1\]) and (\[argu\_3\]) we can get $$\begin{aligned} \label{argu_4} \begin{aligned} &\left\| E\left[\xi_1\right]-E\left[ \xi(B_i(\widetilde{x},a),s_2)\right]\right\|=\left\| \sum_{n=1}^{\infty}\sum_{k=n}^{\infty} \left(P\left[\xi_1=k\right]-P\left[ \xi_2=k \right]\right) \right\|\\ &< 4N^{d/(d-1)}e^{-c(N-1)}+ \sum_{n=N^{d/(d-1)}}^{\infty}\left( P\left[\xi_1\geq n\right]+P\left[ \xi_2\geq n) \right]\right). \end{aligned}\end{aligned}$$ In the following we estimate the upper bound of $\sum_{n=N^{d/(d-1)}}^{\infty} P[\xi_1 \geq n]$. Firstly, by (\[prop1\_00\]), for $N$ large enough, we can obtain $$\begin{aligned} \label{argu_5} \begin{aligned} \sum\limits_{n=N^{d/(d-1)}}^{e^2 \lambda s_1^d} P[\xi_1 \geq n]<\sum\limits_{n=N^{d/(d-1)}}^{e^2 \lambda s_1^d} \exp \left (-c n^{(d-1)/d} \right )+ e^2 \lambda s_1^d e^{-c s_1 }. \end{aligned}\end{aligned}$$ Set $\alpha:=\exp (-c N)$, then $$\begin{aligned} \label{argu_6} \begin{aligned} &\sum\limits_{n=N^{d/(d-1)}}^{e^2 \lambda s_1^d} \exp \left (-c n^{(d-1)/d} \right )=\sum\limits_{n=N^{d/(d-1)}}^{e^2 \lambda s_1^d} \alpha^{(nN^{-d/(d-1)})^{(d-1)/d}}\\ &~~<N^{d/(d-1)}\sum\limits_{k=1}^{\infty} \alpha ^{k^{(d-1)/d}}= N^{d/(d-1)} \alpha \sum\limits_{k=1}^{\infty} \alpha^{k^{(d-1)/d}-1}<M N^{d/(d-1)} \alpha, \end{aligned}\end{aligned}$$ where $M=\sum_{k=1}^{\infty} \exp(-c (k^{(d-1)/d}-1))<\infty$ is a constant. Secondly, by Lemma 1.2 in [@p10], $$\begin{aligned} \label{argu_7} \begin{aligned} &\sum\limits_{n=e^2 \lambda s_1^d+1}^{\infty} P[\xi_1 \geq n] < \sum\limits_{n=e^2 \lambda s_1^d+1}^{\infty} P[Po(\lambda s_1^d)\geq n]\\ &~~ \leq \sum\limits_{n=e^2 \lambda s_1^d+1}^{\infty} \exp \left ( -\left (\frac{n}{2} \right ) \log \left ( \frac{n}{\lambda s_1^d} \right ) \right ) < \frac{e^{-(e^2 \lambda s_1^d +1)}}{1-e^{-1}}. \end{aligned}\end{aligned}$$ Thus, by (\[argu\_5\]), (\[argu\_6\]) and (\[argu\_7\]), there exists a constant $c_1>0,$ such that for large $N$, $$\begin{aligned} \label{argu_8} \sum\limits_{n=N^{d/(d-1)}}^{\infty} P[\xi_1 \geq n]< e^{-c_1 N}.\end{aligned}$$ Using the ergodicity of Poisson point processes, similarly, we can get $$\begin{aligned} \label{argu_9} \sum\limits_{n=N^{d/(d-1)}}^{\infty} P[\xi_2 \geq n]< e^{-c_1 N}.\end{aligned}$$ Combining (\[argu\_4\]), (\[argu\_8\]) and (\[argu\_9\]) gives us the result. \[limit\] Suppose $d \geq 2$ and $\lambda>\lambda_c$. Let integer $i\in [1,d]$, and constants $a\in(0,1]$ and $x_j\in [0,\infty)$, $1\leq j \leq i$. Define the point $$\widetilde{x}_{s,a}=\widetilde{x}_{s,a}(x_1,\ldots,x_i):=\left(x_1,\ldots,x_i,\frac{s}{2}-a,\ldots,\frac{s}{2}-a\right)\in \mathbb{R}^d,$$ then the limit of $E[\xi(B_i(\widetilde{x}_{s,a},a))]$ exists and $$\begin{aligned} \label{lim_new_00} \lim_{s\rightarrow\infty}E[\xi(B_i(\widetilde{x}_{s,a},a))]=a^{d-i}\lim_{s\rightarrow\infty}E[\xi(B_i(\widetilde{x}_{s,1},1))].\end{aligned}$$ Also, if $\min_{1\leq j \leq i}\{x_j\}=0$, then $\lim_{s\rightarrow\infty}E[\xi(B_i(\widetilde{x}_{s,a},a))]>0$. For $s_1$ and $s_2$ large enough, suppose $s_2>s_1$. By (\[prop1\_1\]), it is easy to get $N_{\widetilde{x}_{s_1,a},\widetilde{x}_{s_2,a}}(s_1,s_2)>s_1/2-2.$ Therefore by Lemma \[argument\] and Cauchy’s criterion for convergence, the limit of $E[\xi(B_i(\widetilde{x}_{s,a},a))]$ exists as $s\rightarrow\infty$. For any constant $b\in[0,1]$, let $$\begin{aligned} y_{s,b}=y_{s,b}(x_1,\ldots,x_i):=\left(x_1,\ldots,x_i,\frac{s}{2}-1,\ldots,\frac{s}{2}-1,\frac{s}{2}-b\right)\in\mathbb{R}^d.\end{aligned}$$ Similarly, by Lemma \[argument\] and the Cauchy’s criterion we have the limit of $E[\xi(B_{d-1}(y_{s,b},b))]$ exists. Define $$f_{x_1,\ldots,x_i}(b):=\lim_{s\rightarrow\infty}E[\xi(B_{d-1}(y_{s,b},b))].$$ Since $Leb(B_{d-1}(y_{s,b},b)\cap B_{d-1}(y_{s,1},1-b))=0$, then by the definition of $\xi$ we have $$\begin{aligned} \label{lim_new_01} E\left[\xi(B_{d-1}(y_{s,1},1))\right]=E[\xi(B_{d-1}(y_{s,1},1-b))]+E[\xi(B_{d-1}(y_{s,b},b))].\end{aligned}$$ By (\[prop1\_1\]), $N_{y_{s,1},y_{s,1-b}}(s,s)>s/2-2$. Using Lemma \[argument\] and Cauchy’s criterion we have $$\begin{aligned} \lim_{s\rightarrow\infty}E[\xi(B_{d-1}(y_{s,1},1-b))]=\lim_{s\rightarrow\infty}E[\xi(B_{d-1}(y_{s,1-b},1-b))]=f_{x_1,\ldots,x_i}(1-b).\end{aligned}$$ Therefore, taking the limits of the both sides on (\[lim\_new\_01\]), we can get $$\begin{aligned} f_{x_1,\ldots,x_i}(1)=f_{x_1,\ldots,x_i}(1-b)+f_{x_1,\ldots,x_i}(b),\end{aligned}$$ which indicates that $f_{x_1,\ldots,x_i}(b)=bf_{x_1,\ldots,x_i}(1)$. With the similar method, we can get $$\begin{aligned} \lim_{s\rightarrow\infty}E[\xi(B_i(\widetilde{x}_{s,a},a))]=a^{d-i}f_{x_1,\ldots,x_i}(1),\end{aligned}$$ which gives (\[lim\_new\_00\]). It remains to prove that $\lim_{s\rightarrow\infty}E[\xi(B_i(\widetilde{x}_{s,a},a))]>0$ if $\min_{1\leq j \leq i}\{x_j\}=0$. For simplicity of exposition, we restrict ourselves to the case of $d=2$, and the proof of this result has no essential difficulty when $d\geq 3$. Let $\partial B(s)$ denote the boundary of $B(s)$. If $\min_{1\leq j \leq i}\{x_j\}=0$, then $\widetilde{x}_{s,a}\in\partial B(s)$. For $x\in B_i(\widetilde{x}_{s,a},a)$, let $d_x$ to be the Euclid distance from $x$ to $\partial B(s)$, then $0\leq d_x\leq 1$. Let $V_x$ denote the connected component containing $x$ of $G(\mathcal{H}_{\lambda,s}\cup \{x\};1)$. Firstly, we will show that there exists a constant $c>0$, such that $$\begin{aligned} \label{bound_2} \begin{aligned} &P\left[\{\|V_x\|=1\} \cap \{x\in \mathcal{C}_{\infty}\}\right]\\ &~~\geq c\left[1-\exp\left(\lambda\left(d_x\sqrt{1-d_x^2}-\arccos d_x\right)\right)\right]p_{\infty}(\lambda). \end{aligned}\end{aligned}$$ Define ![The placements of $B_x^-,B_x^+,R_1$ and $R_2$ are shown.[]{data-label="box2"}](13295fig2.eps){width="2in"} $B_x^-$ to be the rectangle of $(1+d_x)\times 2$ centred at $x$ and $B_x^+$ to be the rectangle of $(\frac{7}{3}+d_x)\times \frac{10}{3}$ centred at $x$. Divide the region of $B_x^+\backslash B_x^-$ into 64 small rectangles with two diffrent sizes: one size recorded $R_1$ is $\frac{1}{3}\times \frac{1}{3}$, and the other size recorded $R_2$ is $\frac{1+d_x}{6}\times \frac{1}{3}$, see Figure \[box2\]. The number of small rectangles with size $R_1$ is $40$, and the number of small rectangles with size $R_2$ is $24$. Define $A_1$ to be the event that each of these 64 small rectangles includes at least one point of $\mathcal{H}_{\lambda}$. By the properties of Poisson point processes, we have $$\begin{aligned} \label{bound_3} \begin{aligned} P(A_1)&=\left(1-e^{-\lambda/9}\right )^{40}\cdot\left(1-e^{-\lambda(1+d_x)/18}\right )^{24}\\ &\geq \left(1-e^{-\lambda/9}\right )^{40}\cdot\left(1-e^{-\lambda/18}\right )^{24}. \end{aligned}\end{aligned}$$ If $A_1$ happens, there exists a connected component in $B_x^+\backslash B_x^-$ which contains all the points in these small rectangles. Also, for any point in $\mathbb{R}^d \backslash B_x^-$ which can connect directly with a point in $B_x^-$, it must connect directly with this connected component. Let $A_2$ denote the event that there exists at least one point in $B_x^+\backslash B_x^-$ contained by $\mathcal{C}_{\infty}$. So according to above discussion, the event $A_1\cap A_2$ is independent with the distribution of the points of $\mathcal{H}_{\lambda}$ in $B_x^-$. Therefore, $$\begin{aligned} \label{bound_4} P(A_1\cap A_2)=P(A_1)P(A_2\|A_1) \geq P(A_1) p_{\infty}(\lambda).\end{aligned}$$ Denote $A_3$ to be the event that there exists at least one point of $\mathcal{H}_{\lambda}$ in $B(x;1) \cap B(s)^c$, where $B(x;1)$ denotes the $d-dimensional$ unit ball centred at point $x$. By the properties of Poisson point processes it can be computed that $$\begin{aligned} \label{bound_5} P(A_3)=1-\exp\left(\lambda\left(d_x\sqrt{1-d_x^2}-\arccos d_x\right)\right).\end{aligned}$$ Because $A_3$ and $A_1\cap A_2$ are both increasing events in $G(\mathcal{H}_{\lambda};1)$, by FKG inequality (Theorem 2.2 in [@p12]) we have $$\begin{aligned} \label{bound_6} P(A_3\cap A_1 \cap A_2) \geq P(A_3)P(A_1 \cap A_2).\end{aligned}$$ If the event $A_3\cap A_1 \cap A_2$ happens, it must be true that $x\in \mathcal{C}_{\infty}$. Also, the event $A_3$ is independent with the distribution of the points of $\mathcal{H}_{\lambda}$ in $B_x^-$, so we have $$\begin{aligned} \label{bound_6_temp} \begin{aligned} P\left[\{\|V_x\|=1\} \cap \{x\in \mathcal{C}_{\infty}\}\right] &\geq P[A_3\cap A_1 \cap A_2 \cap \{\mathcal{H}_{\lambda} \cap B_x^- = \emptyset\}]\\ &=e^{-2(1+x)\lambda}P(A_3\cap A_1 \cap A_2)\\ &\geq e^{-4\lambda}P(A_3\cap A_1 \cap A_2). \end{aligned}\end{aligned}$$ Set $c:=e^{-4\lambda}\cdot \left(1-e^{-\lambda/9}\right )^{40}\cdot\left(1-e^{-\lambda/18}\right )^{24}$, together with (\[bound\_3\]), (\[bound\_4\]), (\[bound\_5\]), (\[bound\_6\]) and (\[bound\_6\_temp\]) we can get (\[bound\_2\]). Let $W$ denote the number of the points of $\mathcal{H}_{\lambda} \cap B_i(\widetilde{x}_{s,a},a)$ which belong to $\mathcal{C}_{\infty}$ but are isolated in $B(s)$. By the definition of $\xi(B_i(\widetilde{x}_{s,a},a))$ and Palm theory for Poisson processes, we have $$\begin{aligned} E[\xi(B_i(\widetilde{x}_{s,a},a))]\geq E[W]=\lambda \int_{B_i(\widetilde{x}_{s,a},a))} P\left[\{\|V_x\|=1\} \cap \{x\in \mathcal{C}_{\infty}\}\right] dx.\end{aligned}$$ Combining this with (\[bound\_2\]), we can get $E[\xi(B_i(\widetilde{x}_{s,a},a))]>\frac{1}{2}c\left(1-e^{(1-\pi)\lambda/4}\right)\lambda p_{\infty}(\lambda).$ Our result follows. For simplicity of exposition, we shall prove (\[order\_t1\_00\]) only in the case of $d=3$, and this proof has no essential difficulty in the case of $d=2$ or $d\geq 4$. Let $\eta_{ij}(s):=E\left[\xi\left([0,1]\times [i,i+1]\times [j,j+1],s \right)\right]$ and take $n=\lfloor \frac{s}{2} \rfloor$. By symmetry we have $\eta_{ij}(s)=\eta_{ji}(s)$, and therefore $$\begin{aligned} \label{t1_1} \begin{aligned} E\left[\xi\left([0,1]\times[0,n]^2\right)\right]&=\sum_{i=0}^{n-1}\sum_{j=0}^{n-1}\eta_{ij}(s)\\ &=\eta_{00}(s)+\sum_{k=1}^{n-1}\left(2\sum_{i=0}^{k-1}\eta_{ik}(s)+\eta_{kk}(s) \right). \end{aligned}\end{aligned}$$ Set $$a_1(s):=\eta_{00}(s)+\sum_{k=1}^{n-1}\left(2\sum_{i=0}^{k-1}\left(\eta_{ik}(s)-\eta_{i,n-1}(s)\right)+\eta_{kk}(s)-\eta_{k,n-1}(s) \right),$$ then for large $s$ and $s_2$ satisfying $s_2>s$, by Lemma \[argument\] we have $$\begin{aligned} \label{t1_2} \begin{aligned} &\|a_1(s)-a_1(s_2)\|<2n^2 e^{-cs/2}\\ &~~~~~~~~+\sum_{k=n}^{n_2-1}\left(2\sum_{i=0}^{k-1}\|\eta_{ik}(s_2)-\eta_{i,n_2-1}(s_2)\|+\|\eta_{kk}(s_2)-\eta_{k,n_2-1}(s_2)\| \right)\\ &<2n^2 e^{-cs/2}+ \sum_{k=n}^{n_2-1}\left(2\sum_{i=0}^{k-1}e^{-ck}+e^{-ck}\right)=o\left(e^{-cs/3}\right), \end{aligned}\end{aligned}$$ where $n_2=\lfloor\frac{s_2}{2} \rfloor$ and $c$ is the same constant appearing in Lemma \[argument\]. Then by Cauchy’s criterion the limit of $a_1(s)$ exists. Define the point $y_i=(0,i,n)\in \mathbb{R}^3$. For any $i\in [0,n-1]$ and large $s$, using Lemmas \[argument\] and \[limit\] we can get $$\begin{aligned} \label{t1_3} \begin{aligned} &\big\|E\left[\xi\left(B_{2}\left(y_{i},{\textstyle \frac{s}{2}}-n\right)\right)\right]-({\textstyle \frac{s}{2}}-n)\eta_{i,n-1}(s)\big\|\\ &\leq \big\|E\left[\xi\left(B_{2}\left(y_{i},{\textstyle \frac{s}{2}}-n\right)\right)\right]-({\textstyle \frac{s}{2}}-n)E\left[\xi\left([0,1]\times[i,i+1]\times[{\textstyle \frac{s}{2}}-1,{\textstyle \frac{s}{2}}]\right)\right]\big\|\\ &~~~~~~~~+({\textstyle \frac{s}{2}}-n)\big\|E\left[\xi\left([0,1]\times[i,i+1]\times[{\textstyle \frac{s}{2}}-1,{\textstyle \frac{s}{2}}]\right)\right]-\eta_{i,n-1}(s)\big\|\\ &=o\left(e^{-cs/3}\right). \end{aligned}\end{aligned}$$ Similarly, we can get $$\begin{aligned} \label{t1_4} E\left[\xi\left([0,1]\times\left[n,\frac{s}{2}\right]^2\right)\right]=\left(\frac{s}{2}-n\right)^2\eta_{n-1,n-1}(s)+o\left(e^{-cs/3}\right).\end{aligned}$$ We recall that $R_0=[0,1]\times [0,s/2]^2$, then together with (\[t1\_1\]), (\[t1\_2\]), (\[t1\_3\]) and (\[t1\_4\]), $$\begin{aligned} \label{t1_5} \begin{aligned} E\left[\xi\left(R_0\right)\right]&=E\left[\xi\left([0,1]\times[0,n]^2\right)\right]+2\sum_{i=0}^{n-1}E\left[\xi\left(B_{2}\left(y_{i},\frac{s}{2}-n\right)\right)\right]\\ &~~~~~~~~~~~~+E\left[\xi\left([0,1]\times\left[n,\frac{s}{2}\right]^2\right)\right]\\ &=\sum_{k=1}^{n-1}\left(2\sum_{i=0}^{k-1}\eta_{i,n-1}(s)+\eta_{k,n-1}(s) \right)+\left(s-2n\right)\sum_{i=0}^{n-1}\eta_{i,n-1}(s)\\ &~~~~~~~~~~~~+\left(\frac{s}{2}-n\right)^2\eta_{n-1,n-1}(s)+a_1+o\left(e^{-cs/3}\right), \end{aligned}\end{aligned}$$ where $a_1:=\lim_{s\rightarrow\infty}a_1(s)$. Let $b_i(s):=\eta_{i,n-1}(s)-\eta_{n-1,n-1}(s)$, then by (\[t1\_5\]) we have $$\begin{aligned} \label{t1_6} \begin{aligned} E\left[\xi\left(R_0\right)\right] &=\left(\frac{s^2}{4}-1\right)\eta_{n-1,n-1}(s)+\sum_{k=1}^{n-1}\left(2\sum_{i=0}^{k-1}b_i(s)+b_k(s) \right)\\ &~~~~~~~~~~~~+\left(s-2n\right)\sum_{i=0}^{n-1}b_i(s)+a_1+o\left(e^{-cs/3}\right)\\ &=\left(\frac{s^2}{4}-1\right)\eta_{n-1,n-1}(s)+s\sum_{i=0}^{n-2}b_i(s)-2b_0(s)-\sum_{i=1}^{n-2}(2i+1)b_i(s)\\ &~~~~~~~~~~~~+a_1+o\left(e^{-cs/3}\right). \end{aligned}\end{aligned}$$ Set $$\begin{aligned} a_2(s):=\sum_{i=0}^{n-2}b_i(s)~~~\mbox{and}~~~a_3(s):=2b_0(s)+\sum_{i=1}^{n-2}(2i+1)b_i(s).\end{aligned}$$ With the similar argument as (\[t1\_3\]), we can get that the exist constants $a_2$ and $a_3$ such that $$\begin{aligned} \|a_2(s)-a_2\|<3ne^{-cs/2}~~~\mbox{and}~~~\|a_3(s)-a_3\|=o\left(e^{-cs/3}\right).\end{aligned}$$ Also, by Lemmas \[argument\] and the Cauchy’s criterion, there exists a constant $a_0>0$ such that $$\|\eta_{n-1,n-1}(s)-a_0\|<e^{-c(n-1)}.$$ Taking $a_0,a_2$ and $a_3$ into (\[t1\_6\]) we have $$\begin{aligned} E\left[\xi\left(R_0\right)\right]=\left(\frac{s^2}{4}-1\right)a_0+s a_2-a_3+a_1+o\left(e^{-cs/3}\right).\end{aligned}$$ with the similar argument as above, there exist constants $a_4, a_5, a_6$ and $a_7$, such that $$\begin{aligned} E\left[\xi\left(R_1\right)\right]=\frac{s^2}{4}a_0+s a_4+a_5+o\left(e^{-cs/3}\right),\end{aligned}$$ and $$\begin{aligned} E\left[\xi\left(R_2\right)\right]=\frac{s^2}{4}a_0+s a_6+a_7+o\left(e^{-cs/3}\right).\end{aligned}$$ Combined these with (\[comm\_6\]), (\[total\_1\]) and Lemma \[temp1\], (\[order\_t1\_00\]) has been deduced, where $\tau_1=6a_0>0$. With the results of Theorem 10.22 and Theorem 11.16 (which shows that $\delta>0$) in [@p10], (\[order\_t1\_00\]) is followed by (\[order\_t1\_01\]). Given the discussion in the proof of Theorem 11.16 in [@p10], (2.45) in [@p10] is followed by $$\begin{aligned} \left(n/\lambda\right)^{-1/2} \left(L_1\left(G\left(\mathcal{X}_n;(n/\lambda)^{-1/d}\right)\right)-E[L_1(G(\mathcal{H}_{\lambda,s};1))]\right) \xrightarrow{D} \mathcal {N}(0,\delta^2),\end{aligned}$$ where $s=(n/\lambda)^{1/d}$. Combining this and (\[order\_t1\_00\]) our result follows. Some Applications ================= Our method used in the proof of Theorem \[t1\] can be applied to estimate the expectation of many other random variables restricted to a box $B$ as $B$ becomes large, for example, the size of the biggest open cluster for percolation, the coverage area of the largest component for Poisson Boolean model, the number of open clusters or connected components for percolation and Poisson Boolean model, the number of open clusters or connected components with order $k$ for percolation and Poisson Boolean model, the final size of a spatial epidemic mentioned in [@p10] and so on. We will give the similar results as Theorem \[t1\] for the size of the biggest open cluster and the number of open clusters for site percolation but the method can be adapted to bond percolation. Following Chapter 1 of [@p13], let $\mathbb{L}^d=(\mathbb{Z}^d,\mathbb{E}^d)$ denote the integer lattice with vertex set $\mathbb{Z}^d$ and edges $\mathbb{E}^d$ between all vertex pairs at an $l_1$-distance of 1. For $d\geq 2$ we take $X =(X_x,x\in \mathbb{Z}^d)$ to be a family of i.i.d. Bernoulli random variables with parameter $p\in (0,1)$. Sites $x\in\mathbb{Z}^d$ with $X_x = 1(0)$ are denoted open (closed). The corresponding probability measure of on $\{0,1\}^{\mathbb{Z}^d}$ is denoted by $P_p$. The open clusters are denoted by the connected components of the subgraph of $\mathbb{L}^d$ induced by the set of open vertices. Let $C_{\textbf{0}}$ denote the open cluster containing the origin. The percolation probability is $\theta(p)=P_p(\|C_{\textbf{0}}\|=\infty)$ and the critical probability is $p_c=p_c(d):=\sup\{p:\theta(p)=0\}.$ It is well known [@p13] that $p_c\in(0,1)$. If $p>p_c$, by Theorem 8.1 in [@p13], with probability $1$ there exists exactly one infinite open cluster $\mathcal{C}_{\infty}$. Given integer $n>0$, we denote by open clusters in $B(n)$ the connected components of the subgraph of the integer lattice $\mathbb{L}^d$ induced by the set of open vertices lying in $B(n)$. Similar results as Theorem \[t1\] concerned with the order of the biggest open cluster in $B(n)$ can be given as follows. \[t3\] Suppose $d\geq 2$ and $p\in (p_c,1)$. Let $H(X;B(n))$ be the order of the biggest open cluster in $B(n)$. Then there exist constants $c=c(d,p)>0$ and $\tau_i=\tau_i(d,p)$, $1\leq i \leq d$, with $\tau_1>0$, such that for all large enough $n$, $$\begin{aligned} \label{t3_00} E_p[H(X;B(n-1))]=\theta(p)n^d-\sum_{i=1}^d\tau_i n^{d-i}+o\left(e^{-c n}\right).\end{aligned}$$ Also, there exists a constant $\sigma=\sigma(d,p) > 0$, such that $$\begin{aligned} \label{t3_01} H(X;B(n-1))n^{-d/2}- \theta(p) n^{d/2}+\sum_{i=1}^{\lfloor\frac{d}{2}\rfloor}\tau_i n^{d/2-i} \xrightarrow{D} \mathcal {N}(0,\sigma^2)\end{aligned}$$ as $n\rightarrow\infty$. Similar to the above, $E_p[\|\mathcal {C}_{\infty}\cap B(n-1)\|]=\theta(p)n^d$. Let $C_1,C_2,...,C_M$ denote the components of $\mathcal {C}_{\infty}\cap B(n-1)$, taken in a decreasing order. Let $L(n-1)=B(n-1)\backslash[1,n-2]^d$. For any $2\leq i \leq M$, since $C_i\subset \mathcal {C}_{\infty}$, therefore there exists at least one point in $L(n-1)\cap C_i$ which connects to $\mathcal {C}_{\infty}$ directly; we choose the smallest one according to the lexicographic ordering on $\mathbb{Z}^d$ as the $out-connect$ $point$. For any $x\in \mathbb{Z}^d \cap L(n-1)$, define $$\xi(x):=\left\{ \begin{array}{ll} \|C_i\|, &\mbox{if there exists } i\in[2,M] \mbox{ such that $x$ is the out-connect point of $C_i$}, \\ 0, & \mbox{otherwise}, \end{array} \right.$$ Also, for integer $j\in[0,d-1]$, let $$\begin{aligned} R_j:=\left([0,1]\times [0,n-1]^{d-1-j} \times [1,n-2]^j \right) \cap \mathbb{Z}^d,\end{aligned}$$ then $$E\left[\sum_{i=2}^M \|C_i\|\right]=\sum_{x\in \mathbb{Z}^d \cap L(n-1) }E[\xi(x)]=2\sum_{j=0}^{d-1}\sum_{x\in R_j }E[\xi(x)].$$ With the similar process as the proof of Theorem \[t1\], (\[t3\_00\]) can be deduced, where $$\tau_1=2d\lim_{n\rightarrow\infty}E\left[\xi\left(\left(0,\lfloor\frac{n}{2}\rfloor,\ldots,\lfloor\frac{n}{2}\rfloor\right)\right)\right]>0.$$ Using Theorem 3.2 in [@penrose2001], (\[t3\_01\]) follows. Following Chapter 1.5 of [@p13], we define the number of open clusters per vertex by $$\kappa(p)=E_p(\|C_{\textbf{0}}\|^{-1})=\sum_{n=1}^{\infty}\frac{1}{n}P_p(\|C_{\textbf{0}}\|=n),$$ with the convention that $1/\infty=0$. Similar results as Theorem \[t1\] concerning with the number of the open clusters in $B(n)$ can also be given as follows. \[t4\] Suppose $d\geq 2$ and $p\in (0,p_c)\cup (p_c,1)$. Let $H(X;B(n))$ be the number of the open clusters in $B(n)$. Then there exist constants $c=c(d,p)>0$ and $\tau_i=\tau_i(d,p)>0$, $1\leq i \leq d$, with $\tau_1>0$, such that for all large enough $n$, $$\begin{aligned} \label{t4_00} E_p[H(X;B(n-1))]=\kappa(p)n^d+\sum_{i=1}^d \tau_i n^{d-i}+o\left(e^{-cn}\right).\end{aligned}$$ Also, there exists a constant $\sigma=\sigma(d,p) > 0$, such that $$\begin{aligned} \label{t4_01} H(X;B(n-1))n^{-d/2}- \kappa(p) n^{d/2}- \sum_{i=1}^{\lfloor \frac{d}{2} \rfloor}\tau_i n^{d/2-i} \xrightarrow{D} \mathcal {N}(0,\sigma^2)\end{aligned}$$ as $n\rightarrow\infty$. Moreover, for any constant $\varepsilon\in(0,d/2)$, $$\begin{aligned} \label{t4_03} \begin{aligned} &P_p\left(\frac{H(X;B(n-1))-\kappa(p)n^d-\sum_{i=1}^d \tau_i n^{d-i}}{\mbox{Var}(H(X;B(n-1)))}\leq x \right)\\ &=\int_{-\infty}^x \frac{1}{\sqrt{2\pi}} e^{-y^2/2}dy+o\left(n^{-\frac{d}{2}+\varepsilon} \right), \end{aligned}\end{aligned}$$ where Var$(\cdot)$ denotes the variance. Let $L(n-1)=B(n-1)\backslash[1,n-2]^d$. For any $x\in B(n-1)\cap \mathbb{Z}^d$, let $C_x$ denote the open cluster including $x$, and let $C_x(B(n-1))$ denote the open cluster including $x$ in $B(n-1)$. Then $C_x(B(n-1))\subseteq C_x$. For all open clusters $C$ in $B(n-1)$, if $C\cap L(n-1)\neq \emptyset,$ according to the lexicographic ordering on $\mathbb{Z}^d$ we choose the smallest element of $C\cap L(n-1)$ as the indicated vertex of $C$. For any $x\in \mathbb{Z}^d \cap L(n-1)$, define $$\xi(x,B(n-1)):=\left\{ \begin{array}{ll} 1-\frac{\|C_x(B(n-1))\|}{\|C_x\|}, &\mbox{if $x$ is the idicated vertex of $C_x(B(n-1))$}, \\ 0, & \mbox{otherwise}. \end{array} \right.$$ Noted that for any $y\in \mathbb{Z}^d \cap B(n-1)$, $$\sum_{x\in C_y(B(n-1))} \left(\|C_y(B(n-1))\|^{-1}-\|C_y\|^{-1}\right)=1- \frac{\|C_y(B(n-1))\|}{\|C_y\|},$$ then by (4.7) in [@p13], we have $$\begin{aligned} \label{t4_1} \begin{aligned} H(X;B(n-1))&=\sum_{x\in \mathbb{Z}^d \cap B(n-1)} \|C_x(B(n-1))\|^{-1}\\ &= \sum_{x\in \mathbb{Z}^d \cap B(n-1)} \|C_x\|^{-1}+ \sum_{x\in \mathbb{Z}^d \cap L(n-1)} \xi(x,B(n-1)). \end{aligned}\end{aligned}$$ Therefore, take the expectation for the both sides of (\[t4\_1\]), we can get $$E_p[H(X;B(n-1))]=\kappa(p)n^d + \sum_{x\in \mathbb{Z}^d \cap L(n-1)} E_p[\xi(x,B(n-1))].$$ Suppose $1\leq i \leq d$ and $x_j\in [0,K/2-1]\cap \mathbb{Z}$ for $1\leq j \leq i$. For large integers $n_1,n_2$, let $x=(x_1,\ldots,x_i,\lfloor \frac{n_1}{2} \rfloor, \ldots,\lfloor \frac{n_1}{2} \rfloor) \in \mathbb{Z}^d$ and $\widetilde{x}=(x_1,\ldots,x_i,\lfloor \frac{n_2}{2} \rfloor, \ldots,\lfloor \frac{n_2}{2} \rfloor ) \in \mathbb{Z}^d$. Set $\widetilde{B}(n_2):=B(n_2)\oplus\{x-\widetilde{x}\}.$ Since $\xi$ is stationary under translations of the lattice $\mathbb{L}^d$, then $\xi(\widetilde{x},B(n_2))$ and $\xi(x,\widetilde{B}(n_2))$ have the same distribution function. However, let $n_0=\min\{\lfloor\frac{n_1}{2}\rfloor,\lfloor\frac{n_2}{2}\rfloor\}$, by the definition of $\xi$ we have $$\begin{aligned} &&P_p\left[\xi(x,B(n_1))\neq \xi(x,\widetilde{B}(n_2)) \right]=P_p\left[\xi(x,B(n_1))\neq \xi(x,\widetilde{B}(n_2)),C_x\neq C_\infty \right]\nonumber\\ &&~~\leq P_p\left[\mbox{diam}(C_x)\geq n_0,C_x\neq C_\infty\right]< e^{-cn_0},\end{aligned}$$ where the last inequality follows from Theorem 6.1 of [@p13] for $p<p_c$ and Theorem 8.18 of [@p13] for $p>p_c$ respectively. Thus, $$\begin{aligned} &&\left\|E_p\left[\xi(x,B(n_1))\right]-E_p\left[\xi(\widetilde{x},B(n_2))\right]\right\|\\ &&~~\leq \sum_{t} t \left\| P_p\left[\xi(x,B(n_1))=t\right]-P_p\left[\xi(x,\widetilde{B}(n_2))=t\right] \right\|\\ &&~~\leq \sum_{t} \left( P_p\left[\xi(x,B(n_1))=t,\xi(x,B(n_1))\neq \xi(x,\widetilde{B}(n_2)) \right] \right.\\ &&~~~~~~~~~~\left.+P_p\left[\xi(x,\widetilde{B}(n_2))=t,\xi(x,B(n_1))\neq \xi(x,\widetilde{B}(n_2))\right] \right)\\ &&~~=2P_p\left[\xi(x,B(n_1))\neq \xi(x,\widetilde{B}(n_2)) \right]<2e^{-c n_0}.\end{aligned}$$ Therefore, $\lim_{n\rightarrow\infty}E_p[\xi(x,B(n)]$ exists. In fact, a similar result as Theorem \[argument\] can be deduced. Let $$\begin{aligned} \widetilde{\tau}_i(K)={{d}\choose{i}}\sum_{x_j\in [0,K-1]\cup [n-K,n-1],1\leq j\leq i } \lim_{n\rightarrow\infty}E_p\left[\xi\left(\left(x_1,\ldots,x_i,\left\lfloor \frac{n}{2} \right\rfloor,\ldots,\left\lfloor \frac{n}{2} \right\rfloor\right)\right)\right],\end{aligned}$$ and let $\tau_i(K)=\sum_{j=1}^i \widetilde{\tau}_j(K) {{d-j}\choose{i-j}} (-2K)^{i-j}.$ In a similar way, (\[t4\_00\]) is obtained. Combining (\[t4\_00\]) with Theorem 3.1 in [@penrose2001], (\[t4\_01\]) follows immediately. By Theorem 2.1 in [@JJP2010], Theorem 3.1 in [@penrose2001] and (\[t4\_00\]), (\[t4\_03\]) can be deduced. It is worth noting that our results do have significance for some practical applications. In fact, the initial motivation of this paper is to provide theoretical foundation and guidance for the design of wireless multihop networks. The wireless multihop networks, e.g., vehicular ad hoc networks, mobile ad hoc networks, and wireless sensor networks, typically consists of a group of decentralized and self-organized nodes that communicate with each other in a peer-to-peer manner over wireless channels, and are increasingly being used in military and civilian applications [@p16]. The large scale wireless multihop networks are usually formulated by the random geometric graphs, and the size of the largest component is a fundamental variable for a network, which plays a key role for the topology control in wireless multihop networks. However, this variable can not be described very precisely by both former theoretic results and even computer simulations as the scale of the network grows to very large. Theorem \[t1\] and Theorem \[t2\] provides a precise estimation for this variable respectively. Using simulations the approximative values of the parameters $p_{\infty}(\lambda)$, $\tau_i$, $\sigma$ and $\delta$ can be obtained, and thus the expression of the asymptotic size of the largest component can be well established, which has guiding significance to the topology control in wireless multihop networks. This research was Supported by the National Natural Science Foundation of China under Grants No. 61203141 and 71271204, and the Innovation Program of the Chinese Academy of Sciences under Grant No. kjcx-yw-s7. [99]{} (2003). Continuum percolation of wireless ad hoc communication networks. [Physica A: Statistical Mechanics and its Applications]{} [325,]{} 577–600. (1999). [Percolation]{}, 2nd edn. Springer, Berlin. (2006). Connectivity in Wireless Ad Hoc Networks with a Log-normal Radio Model. [Mobile Networks and Applications]{} [11,]{} 351–360. (1996). [Continuum Percolation]{}. Cambridge University Press, New York. (2010). A convergence rate in a martingale CLT for percolation clusters. [Journal of the graduate school of the chinese academy of sciences]{} [27(5),]{} 577-583. (1991). On a continuum percolation model. [Advances in Applied Probability]{} [30,]{} 628–639. (1995). Single linkage clustering and continuum percolation. [Journal of Multivariate Analysis]{} [53,]{} 94–109. (2001). A central limit theorem with applications to percolation, epidemics and Boolean models. [Annals of Probability]{} [29,]{} 1515-1546. (2003). [Random Geometric Graphs]{}. Oxford University Press, New York. (1996). Large deviations for discrete and continuous percolation. [Advances in Applied Probability]{} [28,]{} 29–52. (2009). Connectivity properties of large-scale sensor networks. [Wireless Networks]{} [15,]{} 945–964. (2009).On the giant component of wireless multihop networks in the presence of shadowing. [IEEE Transactions on Vehicular Technology]{} [58,]{} 5152–5163.	{ "pile_set_name": "ArXiv" }
--- abstract: \| The purpose of this paper is to study the weak solutions of the fractional elliptic problem $$\label{000} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+\epsilon g(u)=k\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}\quad &{\rm in}\quad\ \ \bar\Omega,\\[3mm] \phantom{(-\Delta)^\alpha +\epsilon g(u)} u=0\quad &{\rm in}\quad\ \ \bar\Omega^c, \end{array}$$ where $k>0$, $\epsilon=1$ or $-1$, $(-\Delta)^\alpha$ with $\alpha\in(0,1)$ is the fractional Laplacian defined in the principle value sense, $\Omega$ is a bounded $C^2$ open set in ${\mathbb{R}}^N$ with $N\ge 2$, $\nu$ is a bounded Radon measure supported in $\partial\Omega$ and $\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}$ is defined in the distribution sense, i.e. $$\langle\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha},\zeta\rangle=\int_{\partial\Omega}\frac{\partial^\alpha\zeta(x)}{\partial \vec{n}_x^\alpha}d\nu(x), \qquad \forall\zeta\in C^\alpha({\mathbb{R}}^N),$$ here $\vec{n}_x$ denotes the unit inward normal vector at $x\in\partial\Omega$. In this paper, we prove that (\[000\]) with $\epsilon=1$ admits a unique weak solution when $g$ is a continuous nondecreasing function satisfying $$\int_1^\infty (g(s)-g(-s))s^{-1-\frac{N+\alpha}{N-\alpha}}ds<+\infty.$$ Our interest then is to analyse the properties of weak solution when $\nu=\delta_{x_0}$ with $x_0\in\partial\Omega$, including the asymptotic behavior near $x_0$ and the limit of weak solutions as $k\to+\infty$. Furthermore, we show the optimality of the critical value $\frac{N+\alpha}{N-\alpha}$ in a certain sense, by proving the non-existence of weak solutions when $g(s)=s^{\frac{N+\alpha}{N-\alpha}}$. The final part of this article is devoted to the study of existence for positive weak solutions to (\[000\]) when $\epsilon=-1$ and $\nu$ is a bounded nonnegative Radon measure supported in $\partial\Omega$. We employ the Schauder’s fixed point theorem to obtain positive solution under the hypothesis that $g$ is a continuous function satisfying $$\int_1^\infty g(s)s^{-1-\frac{N+\alpha}{N-\alpha}}ds<+\infty.$$ --- [Existence, Non-existence, Uniqueness of solutions\ for semilinear elliptic equations involving\ measures concentrated on boundary]{} Huyuan Chen[^1] Hichem Hajaiej[^2] [: Fractional Laplacian; Radon measure; Dirac mass; Green kernel; Schauder’s fixed point theorem.]{} [: 35R11, 35J61, 35R06. ]{} Introduction {#sec:intro} ============ Motivation ---------- In 1991, a fundamental contribution of semilinear elliptic equations involving measures as boundary data is due to Gmira and Véron in [@GV], which studied the weak solutions for $$\label{1.1.1} \arraycolsep=1pt \begin{array}{lll} -\Delta u+g(u)=0\quad &{\rm in}\quad \Omega, \\[2mm]\phantom{-----} u=\mu\quad&{\rm on}\quad \partial\Omega, \end{array}$$ where $\Omega$ is a bounded $C^2$ domain in ${\mathbb{R}}^N$ and $\mu$ is a bounded Radon measure defined in $\partial\Omega$. A function $u$ is said to be a weak solution of (\[1.1.1\]) [if $u\in L^1(\Omega)$, $g(u)\in L^1(\Omega,\rho_{\partial\Omega} dx)$ and $$\label{1.1.1.0} \int_\Omega [u(-\Delta)\xi+ g(u)\xi]dx=\int_{\partial\Omega}\frac{\partial\xi(x)}{\partial\vec{n}_x}d\mu(x),\quad \forall\xi\in C^{1.1}_0(\Omega),$$ where $\rho_{\partial\Omega}(x)={\rm dist}(x,\partial\Omega)$ and $\vec{n}_x$ denotes the unit inward normal vector at point $x$.]{} Gmira and V´eron proved that problem (\[1.1.1\]) admits a unique weak solution when $g$ is a continuous and nondecreasing function satisfying $$\label{14.04} \int_1^\infty [g(s)-g(-s)]s^{-1-\frac{N+1}{N-1}}ds<+\infty.$$ Furthermore, the weak solution of (\[1.1.1\]) is approached by the classical solutions of (\[1.1.1\]) replacing $\mu$ by a sequence of regular functions $\{\mu_n\}$, which converge to $\mu$ in the distribution sense. Then this subject has been vastly expanded in recent works, see the papers of Marcus and Véron [@MV1; @MV2; @MV3; @MV4], Bidaut-Véron and Vivier [@BV] and reference therein. A very challenging question consists in studying the analogue elliptic problem involving fractional Laplacian defined by $$(-\Delta)^\alpha u(x)=\lim_{\varepsilon\to0^+} (-\Delta)_\varepsilon^\alpha u(x),$$ where $$(-\Delta)_\varepsilon^\alpha u(x)=-\int_{{\mathbb{R}}^N\setminus B_\varepsilon(x)}\frac{ u(z)- u(x)}{\|z-x\|^{N+2\alpha}} dz$$ for $\varepsilon>0$. The main difficulty comes from how to define the boundary type data. Given a Radon measure $\mu$ defined in $\partial\Omega$, it is ill-posed that $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=0\quad &{\rm in}\quad \Omega, \\[2mm]\phantom{------\ } u=\mu\quad&{\rm on}\quad \partial\Omega, \\[2mm]\phantom{------\ } u=0\quad&{\rm in}\quad \bar\Omega^c. \end{array}$$ Indeed, let $\{\mu_n\}$ be a sequence of regular functions defined in $\partial\Omega$ converging to the measure $\mu$ and a surprising result is that there is just zero solution for $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=0\quad &{\rm in}\quad \Omega, \\[2mm]\phantom{------\ } u=\mu_n\quad&{\rm on}\quad \partial\Omega, \\[2mm]\phantom{------\ } u=0\quad&{\rm in}\quad \bar\Omega^c, \end{array}$$ which is in sharp contrast with Laplacian case, where (\[1.1.1\]) replacing $\mu$ by $\mu_n$ admits a unique nontrivial solution. On the other hand, it is also not proper to pose $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=0\quad &{\rm in}\quad \Omega, \\[2mm]\phantom{------\ } u=\mu\quad&{\rm in}\quad \Omega^c \end{array}$$ with $\mu$ being a Radon measure in $\Omega^c$ concentrated on $\partial\Omega$. In fact, letting functions $\{\mu_n\}\subset C^1_0(\Omega^c)$ converging to $\mu$, the solution $u_n$ of $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=0\quad &{\rm in}\quad \Omega, \\[2mm]\phantom{------\ } u=\mu_n\quad&{\rm in}\quad \Omega^c, \end{array}$$ is equivalent to the solution of $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=G_{\mu_n}\quad &{\rm in}\quad \Omega, \\[2mm]\phantom{------\ } u=0\quad&{\rm in}\quad \Omega^c, \end{array}$$ where $$G_{\mu_n}(x)=\int_{\Omega^c}\frac{\mu_n(y) }{\|x-y\|^{N+2\alpha}}dy,\qquad x\in\Omega,$$ see [@CFQ]. It could be seen that $$\int_\Omega [u_n(-\Delta)^\alpha\xi+ g(u_n)\xi]dx=\int_{\Omega}G_{\mu_n}\xi dx,\quad \forall\xi\in C^2_0(\Omega),$$ Then the limit of $\{u_n\}$ as $n\to\infty$ wouldn’t be a weak solution as we desired, similar to (\[1.1.1.0\]). Therefore, a totally different point of view has to be found to propose the fractional elliptic problem involving measure concentrated on boundary. Our idea is inspired by the study of elliptic equations with fractional Laplacian and Radon measure inside of $\Omega$ in [@CV1], where the authors considered the equations $$\label{1.22} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+h(u)=\nu\quad & \rm{in}\quad\Omega,\\[2mm] \phantom{ (-\Delta)^\alpha +h(u)} u=0\quad & \rm{in}\quad \Omega^c \end{array}$$ for $\nu\in\mathfrak{M}(\Omega,\rho_{\partial\Omega}^\beta)$ with $ \beta\in[0,\alpha]$ the space of Radon measure $\nu$ in $\Omega$ satisfying $$\int_\Omega\rho_{\partial\Omega}^\beta(x) d\|\nu(x)\|<+\infty.$$ A function $u$ is said to be a weak solution of (\[1.22\]), if $u\in L^1(\Omega)$, $h(u)\in L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)$ and $$\int_\Omega [u(-\Delta)^\alpha\xi+ h(u)\xi]dx=\int_{\Omega}\xi(x)d\nu(x),\qquad \forall\xi\in \mathbb{X}_\alpha,$$ where $\mathbb{X}_{\alpha}\subset C({\mathbb{R}}^N)$ with $\alpha\in(0,1)$ denotes the space of functions $\xi$ satisfying: - 1. ${\rm supp}(\xi)\subset\bar\Omega$; <!-- --> 1. $(-\Delta)^\alpha\xi(x)$ exists for all $x\in \Omega$ and $\|(-\Delta)^\alpha\xi(x)\|\leq C$ for some $C>0$; <!-- --> 1. there exist $\varphi\in L^1(\Omega,\rho^\alpha_{\partial\Omega} dx)$ and $\varepsilon_0>0$ such that $\|(-\Delta)_\varepsilon^\alpha\xi\|\le \varphi$ a.e. in $\Omega$, for all $\varepsilon\in(0,\varepsilon_0]$. A unique weak solution of (\[1.22\]) is obtained when the function $h$ is continuous, nondecreasing and satisfies $$\int_1^{\infty}(h(s)-h(-s))s^{-1-k_{\alpha,\beta}}ds<+\infty,$$ where $$k_{\alpha,\beta}=\left\{ \arraycolsep=1pt \begin{array}{lll} \frac{N}{N-2\alpha},\quad &{\rm if}\quad \beta\in[0,\frac{N-2\alpha}N\alpha],\\[2mm] \frac{N+\alpha}{N-2\alpha+\beta},\qquad &{\rm if}\quad \beta\in(\frac{N-2\alpha}N\alpha,\alpha]. \end{array} \right.$$ Motivated by the above results, we may approximate $\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}$ by a sequence measures defined in $\Omega$ and consider the limit of corresponding weak solutions. To this end, for a bounded Radon measure defined in $\bar\Omega$ with support in $\partial\Omega$, we observe that $$\langle\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha},\xi\rangle=\int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial \vec{n}_x^\alpha}d\nu(x), \quad \xi\in \mathbb{X}_\alpha,$$ and $$\frac{\partial^\alpha\xi(x)}{\partial \vec{n}_x^\alpha}=\lim_{s\to0^+}\frac{\xi(x+s\vec{n}_x)-\xi(x)}{s^\alpha} =\lim_{s\to0^+}\xi(x+s\vec{n}_x)s^{-\alpha},$$ so $\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}$ could be approximated by measures $\{t^{-\alpha}\nu_t\}$ with support in $\{x\in\Omega: \rho_{\partial\Omega}(x)=t\}$ generated by $\nu$, see Section 2 for details. Then we consider the limit of weak solutions as $t\to0^+$ for the problem: $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=t^{-\alpha}\nu_t\quad &{\rm in}\quad \Omega, \\[2mm]\phantom{------\ } u=0\quad&{\rm in}\quad \Omega^c. \end{array}$$ Here the limit of these weak solutions (if it exists) is called a weak solution of the following fractional elliptic problem with measure concentrated on boundary $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}\quad &{\rm in}\quad\ \ \bar\Omega,\\[3mm] \phantom{------\ } u=0\quad &{\rm in}\quad\ \ \bar\Omega^c. \end{array}$$ This will be our main focus in this paper. Statement of our problem and main results ----------------------------------------- Let $\alpha\in(0,1)$, $g:{\mathbb{R}}\to{\mathbb{R}}$ be a continuous function, $\Omega$ be a bounded smooth domain in ${\mathbb{R}}^N$ with $N\ge 2$ and denote by $\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$ the bounded Radon measure in $\bar\Omega$ with the support in $\partial\Omega$. Our purpose in this article is to investigate the existence, non-existence and uniqueness of weak solutions to semilinear fractional elliptic problem $$\label{eq 1.1} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+\epsilon g(u)=k\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}\quad &{\rm in}\quad\ \ \bar\Omega,\\[3mm] \phantom{------\ \ } u=0\quad &{\rm in}\quad\ \ \bar\Omega^c, \end{array}$$ where $\epsilon=1$ or $-1$, $k>0$, $(-\Delta)^\alpha$ is the fractional Laplacian and denote $\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}$ with $\nu\in\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$ by $$\langle\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha},\xi\rangle=\int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial \vec{n}_x^\alpha}d\nu(x), \qquad \xi\in \mathbb{X}_\alpha,$$ with $\vec{n}_x$ being the unit inward normal vector at $x$. We call $g$ the absorption nonlinearity if $\epsilon=1$, otherwise it is called as source nonlinearity. Before starting our main theorems we make precise the notion of weak solution used in this article. \[weak solution GV\] We say that $u$ is a weak solution of (\[eq 1.1\]), if $u\in L^1(\Omega)$, $g(u)\in L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)$ and $$\int_\Omega [u(-\Delta)^\alpha\xi+\epsilon g(u)\xi]dx=k\int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial\vec{n}_x^\alpha}d\nu(x),\qquad \forall\xi\in \mathbb{X}_\alpha.$$ We notice that $\mathbb{X}_\alpha\supset C_0^2(\Omega)$ is the test functions space when we study semilinear fractional elliptic equations involving measures, which plays the same role as $C^{1,1}_0(\Omega)$ for dealing with second order elliptic equations with measures, see [@CV1; @CV2; @CV3; @CY]. Moreover, it follows from [@RS Proposition 1.1] that $\xi $ is $C^\alpha$ ($\alpha$-Hölder continuous) in ${\mathbb{R}}^N$ if $\xi\in \mathbb{X}_\alpha$. Denote by $G_\alpha$ the Green kernel of $(-\Delta)^\alpha$ in $\Omega\times\Omega$ and by $\mathbb{G}_\alpha[\cdot]$ the Green operator defined as $$\mathbb{G}_\alpha[\frac{\partial^\alpha\nu}{\partial\vec{n}^\alpha}](x)=\lim_{t\to0^+}\int_{\partial\Omega} G_\alpha(x,y+t\vec{n}_y)t^{-\alpha}d\nu(y).$$ Now we are ready to state our first result for problem (\[eq 1.1\]). \[teo 1\] Assume that $\epsilon=1$, $k>0$, $\nu\in\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$ and $g$ is a continuous nondecreasing function satisfying $g(0)\ge0$ and $$\label{g1} \int_1^\infty [g(s)-g(-s)]s^{-1-\frac{N+\alpha}{N-\alpha}}ds<+\infty.$$ Then $(i)$ problem (\[eq 1.1\]) admits a unique weak solution $u_\nu$; $(ii)$ the mapping $\nu\to u_\nu$ is increasing and $$\label{1.33} -k\mathbb{G}_\alpha[\frac{\partial^\alpha\nu_-}{\partial\vec{n}^\alpha}](x)\le u_\nu(x)\le k\mathbb{G}_\alpha[\frac{\partial^\alpha\nu_+}{\partial\vec{n}^\alpha}](x),\qquad x\in\Omega,$$ where $\nu_+,\nu_-$ are the positive and negative decomposition of $\nu$ such that $\nu=\nu_+-\nu_-$; $(iii)$ if we assume additionally that $g$ is $C^{\beta}$ locally in ${\mathbb{R}}$ with $\beta>0$, then $u_\nu$ is a classical solution of $$\label{1.1} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=0\quad & {\rm in}\quad \Omega,\\[2mm] \phantom{------\ } u=0\quad & {\rm in}\quad \Omega^c\setminus {\rm supp}(\nu). \end{array}$$ We remark that\ $(i)$ the second equality in (\[1.1\]) is understood in the sense that $u=0$ in $\Omega^c\setminus {\rm supp}(\nu)$ and $u$ is continuous at every point in $\partial\Omega\setminus {\rm supp}(\nu)$;\ $(ii)$ the uniqueness requires the nondecreasing assumption on nonlinearity $g$, while the existence also holds without the nondecreasing assumption on $g$;\ $(iii)$ (\[g1\]) is called as integral subcritical condition with critical value $\frac{N+\alpha}{N-\alpha}$, similar integral subcritical conditions see the references [@BV; @CV1; @CV2; @V]. Applied Theorem \[teo 1\] when $\nu=\delta_{x_0}$ with $x_0\in\partial\Omega$, problem (\[eq 1.1\]) admits a unique nonnegative weak solution when $g$ satisfies the hypotheses in Theorem \[teo 1\]. Our second goal is to study the further properties of the weak solution. \[teo 2\] Assume that $\epsilon=1$, $k>0$, $\nu=\delta_{x_0}$ with $x_0\in\partial\Omega$, $g$ is a nondecreasing function in $C^\beta$ locally in ${\mathbb{R}}$ with $\beta>0$ satisfying $g(0)\ge0$ and (\[g1\]). Let $u_k$ be the weak solution of (\[eq 1.1\]), then $(i)$ $$\label{b k} \lim_{t\to0^+} \frac{u_k(x_0+t\vec{n}_{x_0})}{ \mathbb{G}_\alpha[\frac{\partial^\alpha\delta_{x_0}}{\partial\vec{n}^\alpha}](x_0+t\vec{n}_{x_0})}=k.$$ $(ii)$ if additionally $g(s)=s^p$ with $p\in(1+\frac{2\alpha}{N},\frac{N+\alpha}{N-\alpha})$, then the limit of $\{u_k\}$ as $k\to\infty$ exists in ${\mathbb{R}}^N\setminus\{x_0\}$, denoting $u_\infty$. Moreover, $u_\infty$ is a classical solution of $$\label{1.3} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^p=0\quad & {\rm in}\quad \Omega,\\[2mm] \phantom{(-\Delta)^\alpha +u^p} u=0\quad & {\rm in}\quad \Omega^c\setminus \{x_0\} \end{array}$$ and satisfies $$\label{b k 01} c_1 \le u_\infty(x_0+t\vec{n}_{x_0})t^{\frac{2\alpha}{p-1}}\le c_2,\qquad \forall t\in(0,\sigma_0),$$ where $c_2>c_1>0$ and $\sigma_0>0$ small enough. $(iii)$ if we assume more that $g(s)=s^p$ with $p\in(0,1+\frac{2\alpha}{N}]$, then $$\lim_{k\to\infty}u_k(x)=+\infty,\qquad \forall x\in \Omega.$$ We notice that the limit of $\{u_k\}$ as $k\to\infty$ blows up every where in $\Omega$ when $g(u)=u^p$ with $1<p\le1+ \frac{2\alpha}{N}$. This phenomena is different from the Laplacian case, which is caused by the nonlocal characteristic of the fractional Laplacian. Theorem \[teo 1\] and Theorem \[teo 2\] show the existence and properties of weak solutions to (\[eq 1.1\]) in the subcritical case. One natural question is what happens in the critical case, i.e., $g(s)=s^p$ with $p\ge \frac{N+\alpha}{N-\alpha}$. The results are given by: \[teo 4\] Assume that $\epsilon=1$, $k>0$, $\Omega=B_1(e_N)$ with $e_N=(0,\cdots, 0,1)$, $\nu=\delta_{0}$ and $g(s)=s^p$ with $p= \frac{N+\alpha}{N-\alpha}$. Then problem (\[eq 1.1\]) doesn’t admit any weak solution. In general, the nonexistence of weak solution is obtained by capacity analysis for second order differential elliptic equations involving measures, see [@V] and references therein. However, it is a very tough job to attain the nonexistence in the capacity framework by the nonlocal characteristic and the weak sense of $\frac{\partial^\alpha\delta_0}{\partial \vec{n}^\alpha}$, which is weaker than Radon measure. In the proof of Theorem \[teo 4\], we make use of the self-similar property in the half space. The last goal of this paper is to consider the fractional elliptic problem (\[eq 1.1\]) with source nonlinearity, that is, $\epsilon=-1$. In the last decades, semilinear elliptic problems with source nonlinearity and measure data $$\label{eq01} \arraycolsep=1pt \begin{array}{lll} -\Delta u=g(u)+k\nu\quad &{\rm in}\quad\ \ \Omega,\\[3mm] \phantom{-\Delta } u=\mu\quad &{\rm on}\quad\ \ \partial\Omega, \end{array}$$ have attracted numerous interests. There are three basic methods to obtain weak solutions. The first one is to iterate $$u_{n+1}=\mathbb{G}_1[g(u_n)]+k\mathbb{G}_1[\nu],\quad \forall n\in{\mathbb{N}}$$ and look for a function $v$ satisfying $$v\ge \mathbb{G}_1[g(v)]+k\mathbb{G}_1[\nu].$$ When $g$ is a pure power source, the existence results could be found in the references [@BC; @BV; @BY; @KV; @V]. The second method is to apply duality argument to derive weak solution when the mapping $r\mapsto g(r)$ is nondecreasing, convex and continuous, see Baras-Pierre [@BP]. These two methods are very difficult to deal with for a general source nonlinearity. Recently, Chen-Felmer-Véron in [@CFV] introduced a new method to solve problem (\[eq01\]) when $g$ is a general nonlinearity, where the authors employed Schauder’s fixed point theorem to obtain the uniform bound and then to approach the weak solution. Here we develop the latter method to attain weak solution of (\[eq 1.1\]) with $\epsilon=-1$ and the main results state as follows. \[teo 3\] Let $\epsilon=-1$, $k>0$ and $\nu\in\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$ nonnegative with ${\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)}=1$. $(i)$ Suppose that $$\label{06-08-2} g(s)\le c_3s^{p_0}+\epsilon,\quad \forall s\ge0,$$ for some $p_0\in(0,1]$, $c_3>0$ and $\epsilon>0$. Assume more that $c_3$ is small enough when $p_0=1$. Then problem (\[eq 1.1\]) admits a nonnegative weak solution $u_\nu$ satisfying $$\label{1.5} u_\nu(x)\ge \mathbb{G}_\alpha[\frac{\partial^\alpha\nu}{\partial\vec{n}^\alpha}](x),\qquad \forall x\in\Omega.$$ $(ii)$ Suppose that $$\label{1.1+++} g(s)\le c_4s^{p_}+\epsilon,\quad \forall s\in[0,1]$$ and $$\label{1.4} g_\infty:=\int_1^{\infty} g(s)s^{-1-\frac{N+\alpha}{N-\alpha}}ds<+\infty,$$ where $c_4,\epsilon>0$ and $p_>1$. Then there exist $k_0,\epsilon_0>0$ depending on $c_4, p_* $ and $ g_\infty$ such that for $k\in[0,k_0)$ and $\epsilon\in(0,\epsilon_0)$, problem (\[eq 1.1\]) admits a nonnegative weak solution $u_\nu$ satisfying (\[1.5\]). We remark that $(i)$ it does not require any restrictions on parameters $c_3, \epsilon, k$ when $p_0\in(0,1)$ or on parameters $ \epsilon, \sigma$ when $p_0=1$; $(ii)$ the integral subcritical condition (\[1.4\]) has the same critical value with (\[g1\]). The rest of the paper is organized as follows. In Section 2 we study the properties of $\frac{\partial^\alpha \nu}{\partial\vec{n}^\alpha}$. Section 3 is devoted to prove Theorem \[teo 1\]. In Section 4 we analyse the properties of the weak solution for problem (\[eq 1.1\]) when $\nu$ is Dirac mass. The nonexistence of weak solution in the critical case is addressed in Section 5. Finally we give the proof of Theorem \[teo 3\] in Section 6. General measure concentrated on boundary ======================================== In this section, we first build the one-to-one connection between the Radon measure space $\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$ and the bounded Radon measure space $\mathfrak{M}^b(\partial\Omega)$. On the one hand, for any $\mu\in \mathfrak{M}^b(\partial\Omega)$, we denote by $\tilde \mu$ the measure generated by $\mu$ extending inside $\Omega$ by zero, that is, $$\tilde \mu(E):=\mu(E\cap\partial\Omega),\qquad \forall E\subset \bar\Omega\ {\rm Borel\ set},$$ then $\tilde\mu\in \mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$. On the other hand, let $\tilde \mu\in \mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$, we see that $$\tilde \mu(E)=\tilde \mu(E\cap\partial\Omega),\qquad \forall E\subset \bar\Omega\ {\rm Borel\ set}.$$ Denote by $\mu$ a Radon measure such that $\mu(F):=\tilde \mu(F),\ F\subset \partial\Omega\ {\rm Borel\ set}$. Then $\tilde \mu(E)=\mu(E\cap\partial\Omega)$ for any Borel set $E\subset\bar\Omega$ and $${\\|\tilde \mu\\|}_{\mathfrak{M}^b(\bar\Omega)}={\\|\mu\\|}_{\mathfrak{M}^b(\partial\Omega)}.$$ Now we make an approximation of $\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}$ by a sequence Radon measure concentrated on one type of manifolds inside of $\Omega$. Indeed, we observe that there exists $\sigma_0>0$ small such that $$\Omega_t:=\{x\in \Omega,\ \rho_{\partial\Omega}(x)>t\}$$ is a $C^2$ domain in ${\mathbb{R}}^N$ for $t\in[0,\sigma_0]$ and for any $x\in\partial\Omega_t$, there exists a unique $x_\partial\in\partial\Omega$ such that $\|x-x_\partial\|=\rho_{\partial\Omega}(x)$. Conversely, for any $x\in\partial\Omega$, there exists a unique point $x_t\in\partial\Omega_t$ such that $\|x-x_t\|=\rho_{\partial\Omega_t}(x)$, where $t\in(0,\sigma_0)$ and $\rho_{\partial\Omega_t}(x)={\rm dist}(x,\partial\Omega_t)$. Then for any Borel set $E\subset \partial\Omega$, there exists unique $E_t\subset \partial\Omega_t$ such that $E_t=\{x_t: x\in E\}$. In what follows, we always assume that $t\in[0,\sigma_0]$. Denote by $\nu_t$ a Radon measure generated by $\nu$ as $$\nu_t(E_t)=\nu(E),$$ and then $\nu_t$ is a bounded Radon measure with support in $\partial\Omega_t$ and $$\nu_t(E)=\nu_t(E\cap\partial\Omega_t),\qquad \forall E\subset\bar\Omega\ {\rm Borel\ set}.$$ In the distribution sense, we have that $$\label{2.1.2} \langle\nu_t,f\rangle=\int_{\partial\Omega_t}f(x)d\nu_t(x)=\int_{\partial\Omega}f(x+t\vec{n}_x)d\nu(x),\qquad \forall f\in C_0(\Omega).$$ Then we observe that $$\label{2.1.0} \{x_t: x\in{\rm supp}(\nu)\}={\rm supp}(\nu_t) \quad{\rm and}\quad{\\|\nu_t\\|}_{\mathfrak{M}^b(\bar\Omega)}={\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)}.$$ Now we are able to show an approximation of $\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}$. \[pr 2.1\] The sequence of Radon measures $\{t^{-\alpha}\nu_t\}_t$ converges to $\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}$ as $t\to0^+$ in the following distribution sense: $$\lim_{t\to0^+}\int_{\partial\Omega_t}\xi(x)t^{-\alpha}d\nu_t(x)=\int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial \vec{n}^\alpha_x}d\nu(x),\qquad \forall\xi\in \mathbb{X}_\alpha.$$ [Proof.]{} It follows from [@RS Proposition 1.1], that $\xi\in C^\alpha({\mathbb{R}}^N)$ if $\xi\in \mathbb{X}_\alpha$. This together with the fact that supp$(\xi)\subset\bar\Omega$, $\frac{\partial^\alpha\xi(x)}{\partial \vec{n}_x^\alpha}$ is well-defined for any $x\in\partial\Omega$ and for $x_t\in\partial\Omega_t$, implies that there exists a unique $x\in\partial\Omega$ such that $$x_t=x+t\vec{n}_x\quad{\rm and}\quad \|x-x_t\|=\rho_{\partial\Omega}(x_t),$$ then $$\xi(x+t\vec{n}_x)t^{-\alpha}=\frac{\xi(x+t\vec{n}_x)-\xi(x)}{t^\alpha},$$ which implies that $$\xi(\cdot+t\vec{n})t^{-\alpha}\to \frac{\partial^\alpha\xi(\cdot)}{\partial \vec{n}^\alpha}\quad {\rm as}\quad t\to0^+\quad {\rm in}\quad C(\bar\Omega).$$ Along with (\[2.1.2\]), we have that $$\arraycolsep=1pt \begin{array}{lll} \|\int_{\partial\Omega_t}\xi(x)t^{-\alpha}d\nu_t(x)-\int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial \vec{n}_x^\alpha}d\nu(x)\| \\[3mm]\phantom{-----} =\|\int_{\partial\Omega}\xi(x+t\vec{n}_x)t^{-\alpha}d\nu(x)-\int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial \vec{n}_x^\alpha}d\nu(x)\| \\[3mm]\phantom{-----} \le\int_{\partial\Omega}\|\xi(x+t\vec{n}_x)t^{-\alpha}-\frac{\partial^\alpha\xi(x)}{\partial \vec{n}_x^\alpha}\|d\|\nu(x)\| \\[3mm]\phantom{-----}\to0\quad{\rm as}\ t\to0^+, \end{array}$$ which ends the proof. $\Box$ We note that Proposition \[pr 2.1\] shows that $\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}$ is approximated by a sequence Radon measure with support in $\Omega$ in the distribution sense and this provides a new method to derive weak solution of (\[eq 1.1\]) by considering the limit of the weak solutions to $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+\epsilon g(u)=kt^{-\alpha}\nu_t\quad & {\rm in}\quad \Omega,\\[2mm] \phantom{(-\Delta)^\alpha +\epsilon g(u)} u=0\quad &{\rm in}\quad \Omega^c. \end{array}$$ To end this section, we give a upper bound for $\mathbb{G}_\alpha[\frac{\partial^\alpha\|\nu\|}{\partial \vec{n}^\alpha}]$ . \[lm 2.1\] Let $\nu\in\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$, then there exists $c_5>0$ such that $$\mathbb{G}_\alpha[\frac{\partial^\alpha\|\nu\|}{\partial \vec{n}^\alpha}](x)\le \int_{\partial\Omega}\frac{c_5}{\|x-y\|^{N-\alpha}}d\|\nu\|(y),\qquad x\in\Omega.$$ [Proof.]{} From [@BV Theorem 1.1], there exists $c_5>0$ independent of $t$ such that for any $(x,y)\in \Omega\times\partial\Omega_t$, $x\neq y$, $$\label{annex 01} G_\alpha(x,y)\le c_5\frac{\rho_{\partial\Omega}^\alpha(y)}{\|x-y\|^{N-\alpha}}=\frac{c_5 t^{\alpha}}{\|x-y\|^{N-\alpha}}.$$ Then for $x\in\Omega$, $$\begin{aligned} \mathbb{G}_\alpha[\frac{\partial^\alpha\|\nu\|}{\partial \vec{n}^\alpha}](x) &=& \lim_{t\to0^+}\int_{\partial\Omega_t} G_\alpha(x,y)t^{-\alpha}d\|\nu_t\|(y)\\ &\le & \lim_{t\to0^+}\int_{\partial\Omega_t} \frac{c_5}{\|x-y\|^{N-\alpha}}d\|\nu_t\|(y) \\&=&\int_{\partial\Omega} \frac{c_5}{\|x-y\|^{N-\alpha}}d\|\nu\|(y).\end{aligned}$$ We complete the proof. $\Box$ Absorption Nonlinearity {#sec:existence} ======================= In this section, our goal is to prove the existence and uniqueness of weak solution for fractional elliptic problem (\[eq 1.1\]) with $\epsilon=1$. To this end, we first consider the properties of weak solution of $$\label{2.0.6} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g_n(u)=kt^{-\alpha}\nu_t\quad & {\rm in}\quad \Omega,\\[2mm] \phantom{(-\Delta)^\alpha +g_n(u)} u=0\quad &{\rm in}\quad \Omega^c, \end{array}$$ where $t\in(0,\sigma_0)$, $\nu_t$ is given in (\[2.1.2\]) and $\{g_n\}$ are a sequence of $C^1$ nondecreasing functions defined on ${\mathbb{R}}$ such that $g_n(0)=g(0)\ge 0$, $$\label{06-08-0} \| g_n\|\le g,\quad \sup_{s\in{\mathbb{R}}}\|g_n(s)\|=n\quad{\rm and}\quad \lim_{n\to\infty}{\\|g_n-g\\|}_{L^\infty_{loc}({\mathbb{R}})}=0.$$ The existence and uniqueness of weak solution to (\[2.0.6\]) is stated as follows. \[pr 1\] Assume that $k>0$, $\alpha\in(0,1)$, $g_n$ is a $C^1$ nondecreasing function satisfying $g_n(0)\ge 0$ and (\[06-08-0\]). Then for $t\in(0,\sigma_0)$, problem (\[2.0.6\]) admits a unique weak solution $u_{n,k\nu_t}$ such that $$\|u_{n,k\nu_t}\|\le k\mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|]\quad{\rm a.e.\ in}\quad \Omega$$ and $$\label{12-08-0} {\\|g_n(u_{n,k\nu_t})\\|}_{L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)}\le ck{\\|\mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|]\\|}_{L^1(\Omega)},$$ where $c>0$ independent of $t$, $k$ and $n$. Furthermore, for any fixed $n\in{\mathbb{N}}$, $t\in(0,\sigma_0)$ and $k>0$, the mapping $\nu\mapsto u_{n,k\nu_t}$ is increasing. [Proof.]{} For any $t>0$, we observe that $kt^{-\alpha}\nu_t$ is a bounded Radon measure in $\Omega$ and $g_n$ is bounded, then it follows from [@CV1 Theorem 1.1] that problem (\[2.0.6\]) admits a unique weak solution $u_{n,k\nu_t}$. Moreover, $kt^{-\alpha}\nu_t$ is increasing with respect to $\nu_t$ and $\nu_t$ is increasing with respect to $\nu$ by the definition of $\nu_t$, then applying [@CV1 Theorem 1.1], we have that for any fixed $t\in(0,\sigma_0)$ and $k>0$, the mapping $\nu\mapsto u_{n,k\nu_t}$ is increasing. $\Box$ To simplify the notation, we always write $u_{n,k\nu_t}$ by $u_{n,t}$ in this section. In order to consider the limit of $\{u_{n,t}\}$ as $t\to0^+$, we introduce some auxiliary lemmas which are the key steps to obtain $\{g_n(u_{n,t})\}$ uniformly integrable with respect to $t$. For $\lambda>0$, let us set $$\label{Slambda} S_\lambda=\{x\in \Omega:\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|](x)>\lambda\}\quad{\rm and}\quad m(\lambda)=\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx.$$ \[lm 0\] For $\nu\in\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$ and any $t\in(0, \sigma_0)$, there exists $c_6>0$ independent of $t$ such that $$\label{annex -0} m(\lambda)\le c_6\lambda^{-\frac{N}{N-\alpha}}.$$ [Proof.]{} For $\Lambda>0$ and $y\in\partial\Omega_t$ with $t\in(0, \sigma_0/2)$, we denote $$A_\Lambda(y)=\{x\in\Omega\setminus\{y\}: G_\alpha(x,y)>\Lambda\}\ \ {\rm {and}}\quad m_\Lambda(y)=\int_{A_\Lambda(y)}\rho_{\partial\Omega}^\alpha(x) dx.$$ For any $(x,y)\in\Omega\times\partial\Omega_t$, $x\neq y$, it infers by (\[annex 01\]) that $$\begin{aligned} A_\Lambda(y)\subset \left\{x\in\Omega\setminus\{y\}: \frac{c_5 t^\alpha}{\|x-y\|^{N-\alpha}}>\Lambda\right\}\subset B_r(y),\end{aligned}$$ where $r=(\frac{c_5t^\alpha}{\Lambda})^{\frac1{N-\alpha}}$. Thus, $\rho_{\partial\Omega}(x)\le R_0$ for some $R_0>0$ such that $\Omega\subset B_{R_0}(0)$ and $$\label{annex 1xhw} m_\Lambda(y)\le R_0^\alpha\int_{B_r(y)}dx\le c_7t^{\frac{N\alpha}{N-\alpha}}\Lambda^{-\frac{N}{N-\alpha}},$$ where $c_7>0$ independent of $t$. For $y\in\partial\Omega_t$, we have that $$\begin{aligned} \int_{S_\lambda} G_\alpha(x,y)\rho_{\partial\Omega}^\alpha(x)dx\le \int_{A_\Lambda(y)}G_\alpha(x,y)\rho_{\partial\Omega}^\alpha(x)dx+\Lambda\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx.\end{aligned}$$ By integration by parts, we obtain $$\begin{aligned} \int_{A_\Lambda(y)}G_\alpha(x,y)\rho_{\partial\Omega}^\alpha(x)dx=\Lambda m_\Lambda(y)+ \int_\Lambda^\infty m_s(y)ds\le c_8t^{\frac{N\alpha}{N-\alpha}}\Lambda^{1-\frac{N}{N-\alpha}},\end{aligned}$$ where $c_8>0$ independent of $t$. Thus, $$\begin{aligned} \int_{S_\lambda} G_\alpha(x,y)\rho_{\partial\Omega}^\alpha(x)dx\le c_8t^{\frac{N\alpha}{N-\alpha}}\Lambda^{1-\frac{N}{N-\alpha}}+\Lambda \int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx.\end{aligned}$$ Choose $\Lambda= t^\alpha(\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx)^{-\frac{N-\alpha}{N}}$ and then $$\begin{aligned} \int_{S_\lambda} G_\alpha(x,y)\rho_{\partial\Omega}^\alpha(x)dx\le c_9t^\alpha (\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx)^{\frac{\alpha}{N}},\end{aligned}$$ where $c_9=c_8+1$. Therefore, $$\begin{aligned} \int_{S_\lambda} \mathbb{G}_\alpha [t^{-\alpha}\|\nu_t\|](x)\rho_{\partial\Omega}^\alpha(x)dx&=&\int_\Omega\int_{S_\lambda} G_\alpha(x,y)\rho_{\partial\Omega}^\alpha(x)dxt^{-\alpha} d\|\nu_t(y)\| \\&\le &c_9\int_\Omega d\|\nu_t(y)\|(\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx)^{\frac{\alpha}{N}} \\&\le& c_9\\|\nu\\|_{\mathfrak{M}^b(\bar\Omega)} (\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx)^{\frac{\alpha}{N}}.\end{aligned}$$ As a consequence, $$\begin{aligned} \lambda m(\lambda)\le c_9\\|\nu\\|_{\mathfrak{M}^b(\bar\Omega)}m(\lambda)^{\frac{\alpha}{N}},\end{aligned}$$ which implies (\[annex -0\]). This ends the proof. $\Box$ From Lemma \[lm 0\], it implies that $$\label{24-08-0} {\\|\mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|]\\|}_{M^{\frac{N}{N-\alpha}}(\Omega,\rho_{\partial\Omega}^\alpha dx)}\le c_9\\|\nu\\|_{\mathfrak{M}^b(\bar\Omega)},$$ where $M^{\frac{N}{N-\alpha}}(\Omega,\rho_{\partial\Omega}^\alpha dx)$ is Marcinkiewicz space with exponent $\frac{N}{N-\alpha}$. The definition and properties of Marcinkiewicz space see the references [@BBC; @CC; @CV1; @V]. In next lemma, the uniformly regularity plays an important role in our approximation of weak solution. \[lm 1\] Assume that $u_{t}$ is a weak solution of (\[2.0.6\]) replacing $g_n$ by $g$, a continuous nondecreasing function satisfying $g(0)\ge 0$. Then for any compact subsets $\mathcal{K}\subset\Omega$, there exist $t_0>0$, $\beta>0$ small and $c_{10}>0$ independent of $t$ such that for $t\in(0,t_0]$, $$\label{2.0.8} {\\|u_{t}\\|}_{C^{\beta}(\mathcal{K})}\le c_{10} {\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)}.$$ Moreover, if $g$ is $C^\beta$ locally in ${\mathbb{R}}$, then there exists $c_{11}>0$ independent of $t$ such that $$\label{2.0.9} {\\|u_{t}\\|}_{C^{2\alpha+\beta}(\mathcal{K})}\le c_{11} {\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)}.$$ [Proof.]{} We observe from Proposition \[pr 1\] that $$\label{3--1} \| u_t\|\le \mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|]\quad{\rm a.e.\ in}\quad \Omega.$$ For compact set $\mathcal{K}$ in $\Omega$, there exists $t_0>0$ such that $$\mathcal{K}_{5t_0}\subset \Omega,$$ where $\mathcal{K}_r:=\{x\in{\mathbb{R}}^N:{\rm dist}(x,\mathcal{K})<r\}$ with $r>0$. Then $\mathcal{K}_{4t_0}\cap \partial\Omega_t=\O$ for any $t\in(0,t_0]$ and $${\\|g(u_t)\\|}_{L^\infty(\mathcal{K}_{3t_0})}\le {\\|g(\mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|])\\|}_{L^\infty(\mathcal{K}_{3t_0})}.$$ Since $t^{-\alpha}\nu_t$ is a bounded Radon measure in $\Omega$, there exists a sequence $\{f_n\}\subset C^2_0(\Omega)$ such that $f_n$ converges to $t^{-\alpha}\nu_t$ in the distribution sense and for some $N_{t_0}>0$ such that for $n\ge N_{t_0}$, supp$(f_n)\cap \mathcal{K}_{3t_0}=\O$. We may assume that $g$ is $C^\beta$ locally in ${\mathbb{R}}$. (In fact, we can choose a sequence of nondecreasing functions $\{g_n\}\subset C^\beta({\mathbb{R}})$ such that $g_n(0)\ge0$, $\|g_n(s)\|\le \|g(s)\|$ for $s\in{\mathbb{R}}$ and $g_n\to g$ locally in ${\mathbb{R}}$ as $n\to\infty$.) Let $w_n$ be the classical solution of $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g_n(u)=f_n\quad & {\rm in}\quad\Omega,\\[2mm] \phantom{ (-\Delta)^\alpha + g_n(u)} u=0\quad & {\rm in}\quad \Omega^c. \end{array}$$ By the uniqueness of weak solution to (\[2.0.6\]), we obtain that, up to some subsequence, $$\label{12-08.2} u_t=\lim_{n\to\infty} w_n\quad {\rm a.e.\ in}\quad \Omega.$$ We observe that $0\le w_n= \mathbb{G}_\alpha[f_n]-\mathbb{G}_\alpha[g(w_n)]\le \mathbb{G}_\alpha[f_n]$ and $\mathbb{G}_\alpha[f_n]$ converges to $\mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|]$ uniformly in any compact set of $\Omega\setminus \partial\Omega_t$ and in $L^1(\Omega)$, then there exists $c_{11}>0$ independent of $n$ and $t$ such that $${\\|w_n\\|}_{L^\infty(\mathcal{K}_{3t_0})}\le {c_{11}}{\\|\mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|]\\|}_{L^\infty(\mathcal{K}_{3t_0})},\quad {\\|w_n\\|}_{L^1(\Omega)}\le {c_{11}}{\\|\mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|]\\|}_{L^1(\Omega)}.$$ By [@CV3 Lemma 3.1], for $\beta\in(0,2\alpha)$, there exists $c_{12}>0$ independent of $n$ and $t$, such that $$\arraycolsep=1pt \begin{array}{lll} {\\|w_n\\|}_{C^{\beta}(\mathcal{K}_{2t_0})} \le c_8[{\\|w_n\\|}_{L^1(\Omega)}+{\\|g(w_n)\\|}_{L^{\infty}(\mathcal{K}_{3t_0})}+{\\|w_n\\|}_{L^{\infty}(\mathcal{K}_{3t_0})}]\\[3mm] \phantom{} \le c_{12}[\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|_{L^1(\Omega)}+{\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|}_{L^\infty( \mathcal{K}_{3t_0})}+\\|g(\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|])\\|_{L^\infty( \mathcal{K}_{3t_0})}]. \end{array}$$ It follows by [@RS Corollary 2.4] that there exist $c_{13},c_{14}>0$ such that $$\label{2.0.10.0} \arraycolsep=1pt \begin{array}{lll} {\\|w_n\\|}_{C^{2\alpha+\beta}(\mathcal{K})} \le {c_{13}}[{\\|w_n\\|}_{L^1(\Omega)}+{\\|g(w_n)\\|}_{C^{\beta}(\mathcal{K}_{2t_0})} +{\\|w_n\\|}_{C^{\beta}(\mathcal{K}_{2t_0})}]\\[3mm] \phantom{------\ } \le c_{14}[\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|_{L^1(\Omega)}+{\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|}_{L^\infty( \mathcal{K}_{3t_0})}\\[3mm] \phantom{-------\ }+\\|g\\|_{C^\beta([0,\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|_{L^\infty( \mathcal{K}_{3t_0})}])} \\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|_{C^\beta( \mathcal{K}_{3t_0})}]. \end{array}$$ Therefore, together with (\[12-08.2\]) and the Arzela-Ascoli Theorem, it follows that $u_t\in C^{2\alpha+\epsilon}(\mathcal{K})$ for $\epsilon\in(0,\beta)$. Then $w_n\to u_t$ and $f_n\to 0$ uniformly in any compact subset of $\Omega\setminus\partial\Omega_t$ as $n\to\infty$. It infers by [@CV3 Lemma 3.1] that $$\arraycolsep=1pt \begin{array}{lll} {\\|u_t\\|}_{C^{\beta}(\mathcal{K})} \le c_{8}[{\\|u_t\\|}_{L^1(\Omega)}+{\\|g(u_t)\\|}_{L^{\infty}(\mathcal{K}_{3t_0})}+{\\|u_k\\|}_{L^{\infty}(\mathcal{K}_{3t_0})}] \\[2mm]\phantom{-}\le c_{12}[\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|_{L^1(\Omega)}+{\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|}_{L^\infty( \mathcal{K}_{3t_0})}+\\|g(\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|)]\\|_{L^\infty( \mathcal{K}_{3t_0})}]. \end{array}$$ We next claim that $\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|_{L^1(\Omega)}$, ${\\|\mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|]\\|}_{L^\infty( \mathcal{K}_{3t_0})}$ are uniformly bounded. In fact, for $x\in\mathcal{K}$ and $y\in\partial\Omega_t$ with $t\in(0,t_0)$, we have that $\|x-y\|\ge 3t_0$. By (\[annex 01\]), it implies that $$\label{2.1.3} \arraycolsep=1pt \begin{array}{lll} \mathbb{G}_{\alpha}[t^{-\alpha}\|\nu_t\|](x)\le \int_{\partial\Omega_t}\frac{c_5}{\|x-y\|^{N-\alpha}}d\|\nu_t(y)\| \\[3mm]\phantom{------} \le c_5t_0^{\alpha-N} {\\|\nu_t\\|}_{\mathfrak{M}^b(\bar\Omega)} = c_5t_0^{\alpha-N} {\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)} \end{array}$$ and $$\label{2.1.4} \arraycolsep=1pt \begin{array}{lll} {\\|\mathbb{G}_{\alpha}[t^{-\alpha}\nu_t]\\|}_{L^1(\Omega)}\le \int_{\Omega}\int_{\partial\Omega_t}\frac{c_5}{\|x-y\|^{N-\alpha}}d\|\nu_t(y)\|dx \\[3mm]\phantom{------}=\int_{\partial\Omega_t}\int_{\Omega}\frac{c_5}{\|x-y\|^{N-\alpha}}dxd\|\nu_t(y)\| \le c_{15} {\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)}, \end{array}$$ which implies that $${\\|u_t\\|}_{C^{\beta}(\mathcal{K})}\le c_{15}{\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)},$$ where $c_{15}>0$ independent of $t$. Moreover, if $g$ is $C^\beta$ locally in ${\mathbb{R}}$, similar to (\[2.0.10.0\]) it implies by (\[2.1.3\]) and (\[2.1.4\]) that $${\\|u_t\\|}_{C^{2\alpha+\beta}(\mathcal{K})}\le c_{16}{\\|\nu\\|}_{\mathfrak{M}^b(\bar \Omega)},$$ where $c_{16}>0$ independent of $t$. We conclude by Theorem 2.2 in [@CFQ] that $u_t$ is a classical solution of $$\label{2.2.1} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+g(u)=0\quad & {\rm in} \quad\Omega\setminus\partial\Omega_t,\\[2mm] \phantom{ (-\Delta)^\alpha + g(u)} u=0\quad & {\rm in} \quad \Omega^c. \end{array}$$ This ends the proof.$\Box$ \[pr 2.01\] Assume that $k>0$ and $\{g_n\}$ are a sequence of $C^1$ nondecreasing functions defined on ${\mathbb{R}}$ such that $g_n(0)=g(0)$ and (\[06-08-0\]). Then problem $$\label{10-08-0} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+ g_n(u)=k\frac{\partial^\alpha\nu}{\partial \vec{n}^\alpha}\quad &{\rm in}\quad\ \ \bar\Omega,\\[3mm] \phantom{(-\Delta)^\alpha + g_n(u)} u=0\quad &{\rm in}\quad\ \ \bar\Omega^c \end{array}$$ admits a unique weak solution $u_{n}$ satisfying $$\label{10-08-1} -k\mathbb{G}_\alpha[\frac{\partial^\alpha\nu_-}{\partial\vec{n}^\alpha}](x)\le u_n(x)\le k\mathbb{G}_\alpha[\frac{\partial^\alpha\nu_+}{\partial\vec{n}^\alpha}](x),\qquad x\in\Omega,$$ where $\nu_+,\nu_-$ are the positive and negative decomposition of $\nu$ such that $\nu=\nu_+-\nu_-$.\ Furthermore, $$\label{12-08-1} {\\|g_n(u_{n})\\|}_{L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)}\le k{\\|\mathbb{G}_\alpha[\frac{\partial^\alpha\|\nu\|}{\partial\vec{n}^\alpha}]\\|}_{L^1(\Omega)}$$ and $u_n$ is a classical solution of (\[1.1\]) replacing $g$ by $g_n$. [Proof.]{} [To prove the existence of weak solution.]{} Since $\nu_t$ is a bounded Radon measure with supp$(\nu_t)\subset\partial \Omega_t$ for $t\in(0,\sigma_0)$, then by Proposition \[pr 1\], we have that problem (\[2.0.6\]) admits a unique weak solution $u_{n,t}$ such that $$\label{24-08-1} \|u_{n,t}\|\le \mathbb{G}_\alpha [t^{-\alpha}\|\nu_t\|]\quad{\rm a.e.\ in}\quad \Omega,\qquad \int_{\Omega}\|g_n(u_{n,t})\| \rho_{\partial\Omega}^\alpha dx\le k{\\|\mathbb{G}_\alpha [t^{-\alpha}\|\nu_t\|]\\|}_{L^1(\Omega)}$$ and $$\label{2.1.1--} \int_\Omega [u_{n,t}(-\Delta)^\alpha\xi+g_n(u_{n,t})\xi]dx=\int_{\partial\Omega_t}t^{-\alpha}\xi(x)d\nu_t(x),\quad \forall\xi\in \mathbb{X}_\alpha.$$ For any compact set $\mathcal{K} \subset \Omega$, there exists $t_0\in(0,\sigma_0)$ such that $$\mathcal{K} \subset \Omega_t\quad {\rm and}\quad {\rm dist}(\mathcal{K},\partial\Omega_{t})\ge t_0,\quad \forall t\in(0,t_0].$$ By Lemma \[lm 1\], we observe that for some $\beta\in(0,\alpha)$ $${\\|u_{n,t}\\|}_{C^\beta(\mathcal{K})}\le c_5t_0^{-N+2\alpha} {\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)}.$$ Therefore, up to some subsequence, there exists $u_n$ such that $$\lim_{t\to0^+}u_{n,t}=u_n\quad{\rm a.e.\ in}\quad \Omega.$$ Then $g_n(u_{n,t})$ converges to $g_n(u_n)$ almost every in $\Omega$ as $t\to0^+$. By (\[24-08-1\]) and (\[24-08-0\]), we have that $\{u_{n,t}\}_t$ is relatively compact in $L^1(\Omega)$, up to subsequence, $$u_{n,t}\to u_n\quad {\rm in}\ \ L^1(\Omega)\quad {\rm as}\ t\to0^+$$ and then $$g_n(u_{n,t})\to g_n(u_n)\quad {\rm in}\ \ L^1(\Omega,\rho^\alpha_{\partial\Omega}dx)\quad {\rm as}\ t\to0^+.$$ By Proposition \[pr 2.1\], $$\begin{aligned} \int_{\partial\Omega_t}t^{-\alpha}\xi(x)d\nu_t(x) \to \int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial\vec{n}_x^\alpha}d\nu(x)\quad {\rm as}\ t\to0^+,\end{aligned}$$ Passing to the limit as $t\to 0^+$ in the identity (\[2.1.1–\]), it implies that $$\int_\Omega [u_n(-\Delta)^\alpha\xi+g_n(u_n)\xi]dx=k\int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial\vec{n}_x^\alpha}d\nu(x),\quad \forall\xi\in\mathbb{ X}_\alpha.$$ This implies that $u_n$ is a weak solution of (\[10-08-0\]). We see that (\[12-08-1\]) follows by (\[12-08-0\]) and Lemma \[lm 0\]. Moreover, by the facts that $u_n=\lim_{t\to0^+} u_{n,t}$ and $$-k\mathbb{G}_\alpha[t^{-\alpha}\nu_-]\le u_{n,t}\le k\mathbb{G}_\alpha[t^{-\alpha}\nu_+]\quad {\rm in}\quad \Omega,$$ we have that (\[10-08-1\]) holds. [To prove that $u_n=0$ in $\Omega^c\setminus{\rm supp}(\nu)$. ]{} Let $x_0\in\partial\Omega\setminus{\rm supp}(\nu)$ and $x_s=x_0+s\vec{n}_{x_0}$ with $s\in(0,\sigma_0)$. We only have to prove that $\lim_{s\to0^+}u(x_s)=0$. From [@BV Theorem 1.1], for any $(x,y)\in \Omega\times \partial\Omega_t$, $x\neq y$, $$\label{annex 010} G_\alpha(x,y)\le c_5\frac{\rho^\alpha_{\partial\Omega}(y)\rho^\alpha_{\partial\Omega}(x)}{\|x-y\|^N }=c_5\frac{\rho^\alpha_{\partial\Omega}(x)t^{\alpha}}{\|x-y\|^N }.$$ For some $s_0>0$ and any $s\in(0,s_0)$, we observe that ${\rm dist}(x_s,{\rm supp}(\nu))\ge \frac{1}{2}{\rm dist}(x_0,{\rm supp}(\nu))$ and $$\begin{aligned} \mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|](x_s)&\le& c_5\int_{\partial\Omega}\frac{\rho^\alpha_{\partial\Omega}(x_s)}{\|x_s-y\|^N}d\|\nu\|(y) \\&=& c_5s^\alpha\int_{\partial\Omega\setminus{\rm supp}(\nu)}\frac{1}{\|x_s-y\|^N}d\|\nu\|(y) \\&\le& c_52^N s^\alpha{\rm dist}(x_0,{\rm supp}(\nu))^{-N}{\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)} \\&\to&0\quad {\rm as}\quad s\to0^+.\end{aligned}$$ Together with the facts that $$\label{facts} u_n=\lim_{t\to0^+}u_{n,t}\quad {\rm and}\quad \|u_{n,t}\|\le \mathbb{G}_\alpha[t^{-\alpha}\|\nu_t\|],$$ we derive that $u_n=0$ in $\Omega^c\setminus{\rm supp}(\nu)$. [To prove the uniqueness of weak solution.]{} Let $u_1,u_2$ be two weak solutions of (\[10-08-0\]) and $w=u_1-u_2$. Then $(-\Delta)^\alpha w=g_n(u_2)-g_n(u_1)$ and $g_n(u_2)-g_n(u_1)\in L^1(\Omega,\rho^\alpha_{\partial\Omega}dx)$. By Kato’s inequatlity, see Proposition 2.4 in [@CV1], for $\xi\in\mathbb{X}_\alpha$, $\xi\ge0$, we have that $$\begin{aligned} \int_\Omega \|w\|(-\Delta)^\alpha \xi dx+\int_\Omega[g_n(u_1)-g_n(u_2)]{\rm sign}(w)\xi dx\le0.\end{aligned}$$ Combining with $\int_\Omega[g_n(u_1)-g_n(u_2)]{\rm sign}(w)\xi dx\ge0$, then we have $$w=0\quad {\rm a.e.\ in}\ \ \Omega.$$ [Regularity of $u_n$.]{} Since $g_n$ is $C^1$ in ${\mathbb{R}}$, then by (\[2.0.9\]), we have $$\label{2.0.10} {\\|u_n\\|}_{C^{2\alpha+\beta}(\mathcal{K})}\le c_{17} {\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)},$$ for any compact set $\mathcal{K}$ and some $\beta\in(0,\alpha)$. Then $u_n$ is $C^{2\alpha+\beta}$ locally in $\Omega$. Together with the fact that $u_{n,t}$ is classical solution of (\[2.2.1\]), we derive by Theorem 2.2 in [@CFQ] that $u_n$ is a classical solution of (\[1.1\]). $\Box$ For $\lambda>0$, let us define $$\label{Slambda} \tilde S_\lambda=\{x\in \Omega:\mathbb{G}_{\alpha}[\frac{\partial^\alpha \|\nu\|}{\partial \vec{n}^\alpha}](x)>\lambda\}\quad{\rm and}\quad \tilde m(\lambda)=\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx.$$ \[lm 00\] For $\nu\in\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$, then there exist $\lambda_0>1$ and $c_{18}>0$ such that for any $\lambda\ge \lambda_0$, $$\label{annex 0} \tilde m(\lambda)\le c_{18}\lambda^{-\frac{N+\alpha}{N-\alpha}}.$$ [Proof.]{} From Lemma \[lm 2.1\], we see that $$\mathbb{G}_\alpha[\frac{\partial^\alpha\|\nu\|}{\partial \vec{n}^\alpha}](x)\le \int_{\partial\Omega}\frac{c_5}{\|x-y\|^{N-\alpha}}d\|\nu(y)\|,\qquad x\in\Omega.$$ For $\Lambda>0$ and $y\in\partial\Omega$, we denote $$\tilde A_\Lambda(y)=\{x\in\Omega: \frac{c_5}{\|x-y\|^{N-\alpha}}>\Lambda\}\ \ {\rm {and}}\quad \tilde m_\Lambda(y)=\int_{\tilde A_\Lambda(y)}\rho_{\partial\Omega}^\alpha(x) dx.$$ For any $(x,y)\in\Omega\times\partial\Omega$, it infers by(\[annex 01\]) that $$\begin{aligned} \tilde A_\Lambda(y)\subset B_{r_0}(y),\end{aligned}$$ where $r_0=(\frac{c_5}{\Lambda})^{\frac1{N-\alpha}}$. Since $\Omega$ is $C^2$, there exists $\Lambda_0>1$ such that for $\Lambda>\Lambda_0$ such that $$\rho_{\partial\Omega}(x)\le \|x-y\|,\quad \forall x\in \tilde A_\Lambda(y)$$ and $$\label{annex 1xhw} \tilde m_\Lambda(y)\le \int_{ B_{r_0}(y)}\|x-y\|^\alpha dx\le c_{19}\Lambda^{-\frac{N+\alpha}{N-\alpha}}.$$ For $y\in\partial\Omega$, we have that $$\begin{aligned} \int_{\tilde S_\lambda} \frac{c_5}{\|x-y\|^{N-\alpha}}\rho_{\partial\Omega}^\alpha(x)dx\le \int_{\tilde A_\Lambda(y)}\frac{c_5}{\|x-y\|^{N-\alpha}}\rho_{\partial\Omega}^\alpha(x)dx+\Lambda\int_{\tilde S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx.\end{aligned}$$ By integration by parts, we obtain $$\begin{aligned} \int_{\tilde A_\Lambda(y)}\frac{c_5}{\|x-y\|^{N-\alpha}}\rho_{\partial\Omega}^\alpha(x)dx&=&\Lambda \tilde m_\Lambda(y)+ \int_\Lambda^\infty\tilde m_s(y)ds \\&\le& c_{20} \Lambda^{1-\frac{N+\alpha}{N-\alpha}},\end{aligned}$$ where $c_{20}>0$. Thus, $$\begin{aligned} \int_{\tilde S_\lambda}\frac{c_5}{\|x-y\|^{N-\alpha}}\rho_{\partial\Omega}^\alpha(x)dx\le c_{20}\Lambda^{1-\frac{N+\alpha}{N-\alpha}}+\Lambda \int_{\tilde S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx.\end{aligned}$$ Since $S_{\tilde \lambda_1}\subset S_{\tilde \lambda_2}$ if $\lambda_1\ge \lambda_2$ and $$\lim_{\lambda\to0^+}\int_{\tilde S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx=0,$$ then there exists $\lambda_0>0$ such that $$\left(\int_{\tilde S_{\lambda_0}} \rho_{\partial\Omega}^\alpha(x)dx\right)^{-\frac{N-\alpha}{N+\alpha}}\ge\Lambda_0$$ and for $\lambda\ge \lambda_0$, we may choose $\Lambda= (\int_{\tilde S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx)^{-\frac{N-\alpha}{N+\alpha}}\ge \Lambda_0$ and then $$\begin{aligned} \int_{\tilde S_\lambda} \frac{c_5}{\|x-y\|^{N-\alpha}}\rho_{\partial\Omega}^\alpha(x)dx\le c_{21}(\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx)^{\frac{2\alpha}{N+\alpha}},\end{aligned}$$ where $c_{21}=c_{20}+1$. Therefore, $$\begin{aligned} \int_{\tilde S_\lambda}\mathbb{G}_{\alpha}[\frac{\partial^\alpha \|\nu\|}{\partial \vec{n}^\alpha}](x)\rho_{\partial\Omega}^\alpha(x)dx&\le &\int_{\partial\Omega}\int_{\tilde S_\lambda} \frac{c_5}{\|x-y\|^{N-\alpha}}\rho_{\partial\Omega}^\alpha(x)dx d\|\nu(y)\| \\&\le &c_{21}\int_{\partial\Omega} d\|\nu(y)\|(\int_{\tilde S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx)^{\frac{2\alpha}{N+\alpha}} \\&\le& c_{21}\\|\nu\\|_{\mathfrak{M}^b(\bar\Omega)} (\int_{S_\lambda} \rho_{\partial\Omega}^\alpha(x)dx)^{\frac{2\alpha}{N+\alpha}}.\end{aligned}$$ As a consequence, $$\begin{aligned} \lambda \tilde m(\lambda)\le c_{21}\\|\nu\\|_{\mathfrak{M}^b(\bar\Omega)}\tilde m(\lambda)^{\frac{2\alpha}{N+\alpha}},\end{aligned}$$ which implies (\[annex 0\]). This ends the proof. $\Box$ To estimate the nonlinearity in $L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)$, we have to introduce an auxiliary lemma as follows. \[lm 08-09\] Assume that $g:{\mathbb{R}}_+\mapsto{\mathbb{R}}_+$ is a continuous function satisfying $$\label{p} \int_1^{\infty} g(s)s^{-1-p}ds<+\infty$$ for some $p>0$. Then there is a sequence real positive numbers $\{T_n\}$ such that $$\lim_{n\to\infty}T_n=\infty\quad{\rm and}\quad \lim_{n\to\infty}g(T_n)T_n^{-p}=0.$$ Assume additionally that $g$ is nondecreasing, then $$\lim_{T\to\infty} g(T)T^{-p}=0.$$ [Proof.]{} The first argument see [@CV2 Lemma 3.1] and second see [@CV1 Lemma 3.1].$\Box$ Now we are ready to prove Theorem \[teo 1\]. [Proof of Theorem \[teo 1\].]{} [To prove the existence of weak solution.]{} Take $\{g_n\}$ a sequence of $C^1$ nondecreasing functions defined on ${\mathbb{R}}$ satisfying $g_n(0)=g(0)$ and (\[06-08-0\]). By Proposition \[pr 2.01\], problem (\[10-08-0\]) admits a unique weak solution $u_{n}$ such that $$\|u_{n}\|\le \mathbb{G}_{\alpha}[\frac{\partial^\alpha \|\nu\|}{\partial \vec{n}^\alpha}]\quad{\rm a.e.\ in}\quad \Omega$$ and $$\label{2.1.1000} \int_\Omega [u_n(-\Delta)^\alpha\xi+g_n(u_n)\xi]dx=k\int_{\partial\Omega}\frac{\partial^\alpha \xi(x)}{\partial \vec{n}_x^\alpha}d\nu(x),\quad \forall\xi\in \mathbb{X}_\alpha.$$ For any compact set $\mathcal{K} \subset \Omega$, we observe from Lemma \[lm 1\] that for some $\beta\in(0,\alpha)$, $${\\|u_n\\|}_{C^\beta(\mathcal{K})}\le c_{22} {\\|\nu\\|}_{\mathfrak{M}^b(\bar\Omega)}.$$ Therefore, up to some subsequence, there exists $u_\nu$ such that $$\lim_{n\to\infty}u_n=u_\nu\quad{\rm a.e.\ in}\ \Omega.$$ Then $ g_n(u_n)$ converge to $g(u_\nu)$ a.e. in $\Omega$ as $n\to\infty$. By Lemma \[lm 00\] and (\[12-08-1\]), we have that $$u_n\to u_\nu\ {\rm in}\ L^1(\Omega),\quad {\\|g_n(u_n)\\|}_{L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)}\le c_{23}{\\|\mathbb{G}_{\alpha}[\frac{\partial^\alpha \|\nu\|}{\partial \vec{n}^\alpha}]\\|}_{L^1(\Omega)}$$ and $$\tilde m(\lambda)\leq c_{18}\lambda^{-\frac{N+\alpha}{N-\alpha}} \ \quad {\rm for}\ \ \ \lambda>\lambda_0,$$ where $$\tilde m(\lambda)=\int_{\tilde S_\lambda}\rho_{\partial\Omega}^{\alpha}(x)dx \quad{\rm with}\quad \tilde S_\lambda=\{x\in\Omega: \mathbb{G}_{\alpha}[\frac{\partial^\alpha \|\nu\|}{\partial \vec{n}^\alpha}]>\lambda\}.$$ For any Borel set $E\subset\Omega$, we have that $$\displaystyle\begin{array}{lll} \displaystyle\int_{E}\|g_n(u_n)\|\rho_{\partial\Omega}^\alpha(x) dx\le \int_{E\cap\tilde S^c_{\frac{\lambda}{k}}}g\left(k\mathbb{G}_{\alpha}[\frac{\partial^\alpha \|\nu\|}{\partial \vec{n}^\alpha}]\right)\rho_{\partial\Omega}^\alpha(x) dx+\int_{E\cap \tilde S_{\frac{\lambda}{k}}}g\left(k\mathbb{G}_{\alpha}[\frac{\partial^\alpha \|\nu\|}{\partial \vec{n}^\alpha}]\right)\rho_{\partial\Omega}^\alpha(x) dx \\[4mm]\phantom{\int_{E}\|g(u_t)\|\rho^{\alpha}_{\partial\Omega}(x)dx} \displaystyle\leq \tilde g\left(\frac{\lambda}{k}\right)\int_E\rho^{\alpha}_{\partial\Omega}(x)dx+\int_{\tilde S_{\frac{\lambda}{k}}}\tilde g\left(k\mathbb{G}_{\alpha}[\frac{\partial^\alpha \|\nu\|}{\partial \vec{n}^\alpha}]\right)\rho^{\alpha}_{\partial\Omega}(x)dx \\[4mm]\phantom{\int_{E}\|g(u_t)\|\rho^{\alpha}_{\partial\Omega}(x)dx} \displaystyle\leq \tilde g\left(\frac{\lambda}{k}\right)\int_E\rho^{\alpha}_{\partial\Omega}(x)dx+\tilde m\left(\frac{\lambda}{k}\right) \tilde g\left(\frac{\lambda}{k}\right)+\int_{\frac{\lambda}{k}}^\infty\tilde m(s)d\tilde g(s), \end{array}$$ where $\tilde g(r)=g(\|r\|)-g(-\|r\|)$. On the other hand, $$\int_{\frac{\lambda}{k}}^\infty \tilde g(s)d\tilde m(s)=\lim_{T\to\infty}\int_{\frac{\lambda}{k}}^T \tilde g(s)d \tilde m(s).$$ Thus, $$\displaystyle\begin{array}{lll} \displaystyle \tilde m\left(\frac{\lambda}{k}\right) \tilde g\left(\frac{\lambda}{k}\right)+ \int_{\frac{\lambda}{k}}^T \tilde m(s)d\tilde g(s) \le c_{24}\tilde g\left(\frac{\lambda}{k}\right)\left(\frac{\lambda}{k}\right)^{-\frac{N+\alpha}{N-\alpha}}+c_{24}\int_{\frac{\lambda}{k}}^T s^{-\frac{N+\alpha}{N-\alpha}}d\tilde g(s) \\[4mm]\phantom{-----\ \int_{\lambda}^T \tilde g(s)d\omega(s)}\displaystyle \leq c_{25}T^{-\frac{N+\alpha}{N-\alpha}}\tilde g(T)+\frac{c_{24}}{\frac{N+\alpha}{N-\alpha}+1}\int_{\frac{\lambda}{k}}^T s^{-1-\frac{N+\alpha}{N-\alpha}}\tilde g(s)ds. \end{array}$$ By assumption (\[g1\]) and Lemma \[lm 08-09\] with $p=\frac{N+\alpha}{N-\alpha}$, $T^{-\frac{N+\alpha}{N-\alpha}}\tilde g(T)\to 0$ when $T\to\infty$, therefore, $$\tilde m\left(\frac{\lambda}{k}\right) \tilde g\left(\frac{\lambda}{k}\right)+ \int_{\frac{\lambda}{k}}^\infty \tilde m(s)\ d\tilde g(s)\leq \frac{c_{24}}{\frac{N+\alpha}{N-\alpha}+1}\int_{\frac{\lambda}{k}}^\infty s^{-1-\frac{N+\alpha}{N-\alpha}}\tilde g(s)ds.$$ Notice that the above quantity on the right-hand side tends to $0$ when $\lambda\to\infty$. The conclusion follows: for any $\epsilon>0$ there exists $\lambda>0$ such that $$\frac{c_{24}}{\frac{N+\alpha}{N-\alpha}+1}\int_{\frac{\lambda}{k}}^\infty s^{-1-\frac{N+\alpha}{N-\alpha}}\tilde g(s)ds\leq \frac{\epsilon}{2}.$$ For $\lambda$ fixed, there exists $\delta>0$ such that $$\int_E\rho_{\partial\Omega}^\alpha(x) dx\leq \delta\Longrightarrow \tilde g\left(\frac{\lambda}{k}\right)\int_E\rho_{\partial\Omega}^\alpha(x) dx\leq\frac{\epsilon}{2},$$ which implies that $\{g_n\circ u_n\}$ is uniformly integrable in $L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)$. Then $g_n\circ u_n\to g\circ u_\nu$ in $L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)$ by Vitali convergence theorem. Passing to the limit as $n\to +\infty$ in the identity (\[2.1.1000\]), it implies that $$\int_\Omega [u_\nu(-\Delta)^\alpha\xi+g(u_\nu)\xi]dx=\int_{\partial\Omega}\frac{\partial^\alpha\xi(x)}{\partial\vec{n}^\alpha_x}d\nu(x),\quad \forall\xi\in\mathbb{ X}_\alpha.$$ Then $u_\nu$ is a weak solution of (\[eq 1.1\]). Moreover, it follows by the fact $$-k\mathbb{G}_\alpha[\frac{\partial^\alpha \nu_-}{\partial \vec{n}^\alpha}]\le u_n\le k\mathbb{G}_\alpha[\frac{\partial^\alpha \nu_+}{\partial \vec{n}^\alpha}]\quad {\rm in}\ \Omega.$$ which, together with $u_\nu=\lim_{n\to+\infty} u_n$, implies (\[1.33\]). The arguments including $u_n=0$ in $\Omega^c\setminus{\rm supp}(\nu)$, uniqueness and regularity follow the proof of Proposition \[pr 2.01\]. $\Box$ The proof of the existence of weak solution is divided into two steps: the first step is to get weak solution $u_n$ to (\[eq 1.1\]) with truncated nonlinearity $g_n$ and then to prove the limit of $\{u_n\}$ as $n\to\infty$ is our desired weak solution. This is due to the estimate in Lemma \[lm 0\] where we only could get exponent $\frac{N}{N-\alpha}$ and in the second step, we make use of Lemma \[lm 00\], the critical exponent of the nonlinearity $g$ could be up to $\frac{N+\alpha}{N-\alpha}$. Isolated singularity on boundary {#sec:Dirac Mass} ================================ For simplicity, we assume that $x_0=0$ and $\vec{n}_0$ is the unit inward normal vector at the origin in what follows and $u_k$ is the weak solution of (\[eq 1.1\]). Weak singularity ---------------- In this subsection, we prove Theorem \[teo 2\] part $(i)$. The regularity refers to Theorem \[teo 1\] in the case that $\nu=\delta_{0}$ with $0\in\partial\Omega$ and our main work is to prove (\[b k\]). We start our analysis with an auxiliary lemma. \[lm 3.1\] Under the hypotheses of Theorem \[teo 2\] part $(i)$, we assume more that $x_s=s\vec{n}_{0}\in\Omega$ for $s>0$ small, then there exists $c_{26}>1$ such that $$\frac1{c_{26}}s^{-N+\alpha}\le \mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x_s)\le c_{26}s^{-N+\alpha}$$ and $$\lim_{s\to0^+}\mathbb{G}_\alpha[g(\mathbb{G}_\alpha[k\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x_s))]s^{N-\alpha}=0.$$ [Proof.]{} It follows by Lemma \[lm 2.1\] with $\nu=\delta_0$ that $$\label{4.3-1} \mathbb{G}_\alpha[\frac{\partial^\alpha \delta_0}{\partial \vec{n}^\alpha}](x)\le \frac{c_5}{\|x\|^{N-\alpha}}, \qquad \forall x\in\Omega,$$ in particular, $$\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_0}{\partial \vec{n}^\alpha}](x_s)\le \frac{c_5}{s^{N-\alpha}}.$$ Let $y_t=t\vec{n}_0$ with $t\in(0,s/2)$, then $$\|y_t-x_s\|=s-t> \frac s2=\frac12\max\{\rho_{\partial\Omega}(y_t),\rho_{\partial\Omega}(x_s)\}$$ and apply [@BV Theorem 1.2] to derive that there exists $c_{27}>0$ such that $$\label{11.04.2} G_\alpha(x_s,y_t) \ge c_{27}\frac{\rho^\alpha_{\partial\Omega}(y_t)\rho^\alpha_{\partial\Omega}(x_s)}{\|x_s-y_t\|^{N}}.$$ Thus, $$\mathbb{G}_\alpha[t^{-\alpha}\delta_{y_t}](x_s)\ge \frac{c_{27} s^{\alpha}}{\|s-t\|^{N}},$$ which implies that $$\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x_s)\ge \frac{c_{27}}{s^{N-\alpha}}.$$ $(ii)$ By (\[4.3-1\]) and monotonicity of $g$, we have that $$\begin{aligned} \mathbb{G}_\alpha[g(\mathbb{G}_\alpha[k\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])](x_s)s^{N-\alpha} &\le& \int_\Omega G_\alpha(x_s,y)g\left(\frac{c_5k}{\|y\|^{N-\alpha}}\right)dy s^{N-\alpha} \\ &\le & \int_\Omega\frac{c_5}{\|x_s-y\|^{N-\alpha}} g\left(\frac{c_5k}{\|y\|^{N-\alpha}}\right)dy s^{N-\alpha} \\ &=& c_5 s^{N-\alpha}\left[\int_{ B_{\frac s2}(x_s)}\frac{\|y\|^\alpha}{\|x_s-y\|^{N-\alpha}} g\left(\frac{c_5k}{\|y\|^{N-\alpha}}\right)dy\right. \\&& \left.+ \int_{ \Omega\setminus B_{\frac s2}(x_s)}\frac{\|y\|^\alpha}{\|x_s-y\|^{N-\alpha}} g\left(\frac{c_5k}{\|y\|^{N-\alpha}}\right)dy\right] \\&:=&A_1(s)+A_2(s).\end{aligned}$$ For $y\in B_{\frac s2}(x_s)$, we have $\frac s2\le \|y\|\le \frac{3s}2$ and by applying Lemma \[lm 08-09\], we derive that $$\begin{aligned} A_1(s)&\le&c_5s^{N+\alpha}g\left(\frac{2^{N-\alpha}c_5k}{s^{N-\alpha}}\right)\int_{ B_{1/2}(\vec{n}_0)} \frac{\|z\|^\alpha}{\|\vec{n}_0-z\|^{N-\alpha}}dz \\&=&c_5r^{-\frac{N+\alpha}{N-\alpha}}g\left(2^{N-\alpha}c_5rk\right)\int_{ B_{1/2}(\vec{n}_0)} \frac{\|z\|^\alpha}{\|\vec{n}_0-z\|^{N-\alpha}}dz \\&\to&0\quad{{\rm as}}\ \ r\to+\infty,\end{aligned}$$ where $r=s^{\alpha-N}$. We next claim that $A_2(s)\to0$ as $s\to0^+$. In fact, for $y\in B_{\frac s2}(0)$, we see that $\|x_s-y\|> s/2$ and $$\begin{aligned} s^{N-\alpha} \int_{ B_{\frac s2}(0)}\frac{\|y\|^\alpha}{\|x_s-y\|^{N-\alpha}} g\left(\frac{c_5k}{\|y\|^{N-\alpha}}\right)dy&\le& 2^{N-\alpha}\int_{ B_{\frac s2}(0)} \|y\|^\alpha g\left(\frac{c_5k}{\|y\|^{N-\alpha}}\right) dy \\&=&c_{28}\int_0^{\frac s2} r^\alpha g\left(\frac{c_5k}{r^{N-\alpha}}\right)r^{N-1} dr \\&=&\frac{c_{28}}{N-\alpha}\int_{s^{-\frac1{N-\alpha}}}^\infty \tau^{-1-\frac{N+\alpha}{N-\alpha}} g\left( c_5k\tau\right) d\tau \\&\to&0\quad{{\rm as}}\ \ s\to0^+,\end{aligned}$$ where the converging used (\[g1\]). For $y\in \Omega\setminus \left(B_{\frac s2}(0)\cup B_{\frac s2}(x_s)\right)$, we have that $\|y-x_s\|>\frac14 \|y\|$ and $$\begin{aligned} &&s^{N-\alpha} \int_{\Omega\setminus \left(B_{\frac s2}(0)\cup B_{\frac s2}(x_s)\right)}\frac{\|y\|^\alpha}{\|x_s-y\|^{N-\alpha}} g\left(\frac{c_5k}{\|y\|^{N-\alpha}}\right)dy \\&&\qquad\le s^{N-\alpha}\int_{B_R(0)\setminus B_s(0)} \|y\|^{2\alpha-N} g\left(\frac{c_5k}{\|y\|^{N-\alpha}}\right) dy \\&&\qquad=c_{29}s^{N-\alpha}\int_s^R \tau^{2\alpha-1} g( c_5k\tau^{\alpha-N}) d\tau \\&&\qquad=c_{29}\frac{s^{2\alpha-1} g(c_5ks^{\alpha-N})}{(N-\alpha)s^{\alpha-N-1}}\qquad \quad{\rm (L'Hospital's\ Rule) } \\&&\qquad=\frac{c_{29}}{N-\alpha}s^{N+\alpha} g(c_5ks^{\alpha-N}) \\&&\qquad\to0\quad{{\rm as}}\ \ s\to0^+,\end{aligned}$$ for some $R>0$ such that $\Omega \subset B_R(0)$ and $c_{29}>0$. Then $$\begin{aligned} A_2(s)\to 0\quad{{\rm as}}\ \ s\to0^+.\end{aligned}$$ Therefore, $$\label{4.4} \lim_{s\to0^+}\mathbb{G}_\alpha[g(\mathbb{G}_\alpha[k\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])](x_s)s^{N-\alpha}=0.$$ The proof ends. $\Box$ [Proof of Theorem \[teo 2\] $(i)$.]{} The existence, uniqueness and regularity follow by Theorem \[teo 1\]. We only need to prove (\[b k\]) to complete the proof. We observe that $$\begin{aligned} k\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x_s) \ge u_k(x_s) &\ge& k\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x_s)- \mathbb{G}_\alpha[g(u_k)](x_s) \\&\ge& k\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x_s)- \mathbb{G}_\alpha[g(k\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])](x_s),\end{aligned}$$ where $s>0$ small. Together with Lemma \[lm 3.1\], (\[b k\]) holds.$\Box$ Strong singularity for $p\in(1+\frac{2\alpha}{N},\frac{N+\alpha}{N-\alpha})$ ---------------------------------------------------------------------------- In this subsection, we consider the limit of $\{u_k\}$ as $k\to\infty$, where $u_k$ is the weak solution of $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^p=k\frac{\partial^\alpha\delta_0}{\partial \vec{n}^\alpha}\quad &{\rm in}\quad\ \ \bar\Omega,\\[3mm] \phantom{-----\ } u=0\quad &{\rm in}\quad\ \ \bar\Omega^c, \end{array}$$ here $0\in\partial\Omega$ and $p\in(1+\frac{2\alpha}{N},\frac{N+\alpha}{N-\alpha})$. From Theorem \[teo 1\] $(iii)$, we know that $u_k$ is a classical solution of $$\label{4.101} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^p=0\quad &{\rm in}\quad\ \ \Omega,\\[3mm] \phantom{-----\ } u=0\quad &{\rm in}\quad\ \ \Omega^c\setminus\{0\}. \end{array}$$ In order to study the limit of $\{u_k\}$ as $k\to\infty$, we have to obtain a super solution of (\[4.101\]). To this end, we consider the function $$\label{4.2} w_p(x)=\|x\|^{-\frac{2\alpha}{p-1}},\qquad x\in{\mathbb{R}}^N\setminus\{0\}.$$ \[lm 4.1\] Assume that $p\in(1+\frac{2\alpha}{N},\frac{N+\alpha}{N-\alpha})$ and $w_p$ is defined in (\[4.2\]). Then there exists $\lambda_0>0$ such that $\lambda_0 w_p$ is a super solution of (\[4.101\]). [Proof.]{} For $p\in(1+\frac{2\alpha}{N},\frac{N+\alpha}{N-\alpha})$, we have that $-\frac{2\alpha}{p-1}\in (-N,-N+2\alpha)$ and from [@FQ], it shows that there exists $c(p)<0$ such that $$(-\Delta)^\alpha w_p(x)=c(p)\|x\|^{-\frac{2\alpha}{p-1}-2\alpha},\quad x\in{\mathbb{R}}^N\setminus\{0\},$$ thus, taking $\lambda_0=\|c(p)\|^{\frac{1}{p-1}}$, we derive that $$(-\Delta)^\alpha (\lambda_0 w_p)+ (\lambda_0w_p)^p=0\quad{\rm in}\quad {\mathbb{R}}^N\setminus\{0\}.$$ Together with $\lambda_0 w_p>0$ in $\Omega^c$, $\lambda_0 w_p$ is a super solution of (\[4.101\]). The proof ends. $\Box$ We observe that the super solution $\lambda_0w_p$ constructed in Lemma \[lm 4.1\] could control the asymptotic behavior of $u_\infty$ near the origin, but for $\partial\Omega\setminus\{0\}$, $\lambda_0w_p$ does not provide enough information for us. To control the behavior of $u_\infty$ on $\partial\Omega\setminus\{0\}$, we have to construct new super solutions. For any given $y_0\in\partial\Omega\setminus\{0\}$, we denote $\eta_0:{\mathbb{R}}^N\to[0,1]$ a $C^2$ functions such that $$\label{4.5} \eta_0(x)=\left\{\arraycolsep=1pt \begin{array}{lll} 0,\quad &x\in B_r(y_0),\\[2mm] 1,\quad &x\in {\mathbb{R}}^N\setminus B_{2r}(y_0), \end{array} \right.$$ where $r=\frac{\|y_0\|}{8}$. \[lm 4.2\] Assume that $p\in(1+\frac{2\alpha}{N},\frac{N+\alpha}{N-\alpha})$ and $w_{\lambda,j}=\lambda\tilde w_p+j \eta_1$, where $\lambda,j>0$, $\tilde w_p=w_p\eta_0$ in ${\mathbb{R}}^N$ and $\eta_1=\mathbb{G}_\alpha[1]$. Then there exist $\lambda_1>0$ and $j_1>0$ depending on $\|y_0\|$ such that $w_{\lambda_1,j_1}$ is a super solution of (\[4.101\]). [Proof.]{} For $x\in \Omega\setminus B_{4r}(y_0)$, we have that $\tilde w_p(x)=w_p(x)$ and $$\begin{aligned} (-\Delta)^\alpha \tilde w_p(x)&=& -\lim_{\epsilon\to0^+}\int_{{\mathbb{R}}^N\setminus B_\epsilon(x)}\frac{\tilde w_p(z)-w_p(x)}{\|z-x\|^{N+2\alpha}}dz \\&=& (-\Delta)^\alpha w_p(x)-\lim_{\epsilon\to0^+}\int_{{\mathbb{R}}^N\setminus B_\epsilon(x)}\frac{\tilde w_p(z)-w_p(z)}{\|z-x\|^{N+2\alpha}}dz \\&\ge& (-\Delta)^\alpha w_p(x)-\int_{B_{2r}(y_0)}\frac{w_p(z)}{\|z-x\|^{N+2\alpha}}dz \\&\ge& c(p)\|x\|^{-\frac{2\alpha}{p-1}-2\alpha} -c_{30}r^{-\frac{2\alpha}{p-1}-2\alpha},\end{aligned}$$ where $c_{30}>0$ and the last inequality used the facts $\|z-x\|\ge 2r$ and $w_p(z)\le r^{-\frac{2\alpha}{p-1}}$. For $x\in B_{2r}(0)\setminus\{0\}$, take $\lambda=\lambda_0$ from Lemma \[lm 4.1\] and $j\ge c_{30}\lambda_0r^{-\frac{2\alpha}{p-1}-2\alpha}$, then we have $$\begin{aligned} (-\Delta)^\alpha w_{\lambda,j}(x)+ w_{\lambda,j}^p(x) &\ge & -c(p)\lambda_0\|x\|^{-\frac{2\alpha}{p-1}-2\alpha} +w_p^p(x)\ge 0.\end{aligned}$$ We observe that there exists $c_{31}>0$ dependent of $r$ such that $$\|(-\Delta)^\alpha \tilde w_p\|\le c_{31}\quad{\rm in}\quad \Omega\setminus B_{2r}(0),$$ then take $j\ge c_{31}\lambda_0$, we have that $$(-\Delta)^\alpha w_{\lambda_0, j}\ge 0,\quad \forall x\in \Omega\setminus B_{2r}(0).$$ Therefore, letting $\lambda_1=\lambda_0$ and $j_1=\max\{c_{31}\lambda_0, c_{30}\lambda_0r^{-\frac{2\alpha}{p-1}-2\alpha}\}$, we have that $$(-\Delta)^\alpha w_{\lambda_1,j_1}+ w_{\lambda_1,j_1}^p\ge 0\quad{\rm in}\quad \Omega.$$ The proof ends. $\Box$ Let $x_s=s\vec{n}_0\in\Omega$ and a set $$A_r=\bigcup_{s\in (0,r)}B_{\frac s8}(x_s).$$ It is obvious that $A_r$ is a cone with the vertex at the origin. \[lm 4.3\] Assume that $p\in(0,\frac{N+\alpha}{N-\alpha})$, then there exists $c_{32}>0$ such that for any $x\in A_{r_0}$, $$\label{4.6} \mathbb{G}_\alpha[(\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])^p](x) \le \left\{\arraycolsep=1pt \begin{array}{lll} c_{32} \|x\|^{-(N-\alpha)p+2\alpha}\quad &{\rm if}\quad p\in (\frac{2\alpha}{N-\alpha}, \frac{N+\alpha}{N-\alpha}),\\[1.5mm] -c_{32} \ln \|x\| \quad &{\rm if}\quad p= \frac{2\alpha}{N-\alpha},\\[1.5mm] c_{32}\quad &{\rm if}\quad p\in(0,\frac{2\alpha}{N-\alpha}). \end{array} \right.$$ [Proof.]{} Since $\partial\Omega$ is $C^2$, then for $r_0\in(0, 1/2)$ small enough, we observe that for any $x\in B_{\frac s8}(x_s)$ with $s\in(0,r_0)$, $$\frac {3s}{4}\le \rho_{\partial\Omega}(x)\le \frac {5s}{4}$$ and for any $t\in(0,\frac s8)$, $$\|x-x_t\|\ge \frac{5s}8 \ge \frac12\max\{\rho_{\partial\Omega}(x),\rho_{\partial\Omega}(x_t)\}.$$ Then it follows by [@BV Theorem 1.1, Theorem 1.2] that there exists $c_{33}>1$ such that $$\frac1{c_{33}} s^{\alpha-N} t^\alpha\le G_\alpha(x,x_t)\le c_{33}s^{\alpha-N} t^\alpha, \quad \forall x\in B_{\frac s8}(x_s).$$ Thus, there exists $c_{34}>0$ independent of $s,t$ such that $$\frac1{c_{34}}s^{-N+\alpha}\le \mathbb{G}_\alpha[t^{-\alpha}\delta_{x_t}](x)\le c_{34}s^{-N+\alpha}, \quad \forall x\in B_{\frac s8}(x_s),$$ which implies that $$\label{11.04.1} \frac1{c_{34}}s^{-N+\alpha}\le \mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x)\le c_{34}s^{-N+\alpha},\quad \forall x\in B_{\frac s8}(x_s).$$ From Lemma \[lm 2.1\], it shows that for any $x\in\Omega$, $$\label{11.04.3} \mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x)\le c_5 \|x\|^{-N+\alpha},\qquad \forall x\in \Omega.$$ It follows by (\[annex 01\]) and (\[11.04.3\]) that $$\arraycolsep=1pt \begin{array}{lll} \mathbb{G}_\alpha[(\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])^p](x_s)\le c_5^p\int_\Omega G_\alpha(x_s,y)\frac{1}{\|y\|^{(N-\alpha)p}}dy \\[3mm]\phantom{--------\ } \le c_5^{p+1} \int_\Omega\frac{\|y\|^\alpha}{\|x_s-y\|^{N-\alpha}} \frac{1}{\|y\|^{(N-\alpha)p}}dy \\[3mm]\phantom{--------\ } = c_5^{p+1} s^{2\alpha-(N-\alpha)p}\int_{\tilde \Omega_s} \frac1{\|\vec{n}_0-z\|^{N-\alpha}} \frac{1}{\|z\|^{(N-\alpha)p-\alpha}}dz \\[3mm]\phantom{--------\ } = c_5^{p+1}s^{2\alpha-(N-\alpha)p}\left[\int_{\tilde \Omega_s\cap B_{1/2}(\vec{n}_0)} \frac1{\|\vec{n}_0-z\|^{N-\alpha}} \frac{1}{\|z\|^{(N-\alpha)p-\alpha}}dz\right. \\[3mm]\phantom{----------------} \left.+ \int_{\tilde\Omega_s\cap B_{\frac12}^c(\vec{n}_0)} \frac1{\|\vec{n}_0-z\|^{N-\alpha}} \frac{1}{ \|z\|^{(N-\alpha)p-\alpha}}dz\right] \\[3mm]\phantom{--------\ } :=c_5^{p+1}s^{2\alpha-(N-\alpha)p}[I_1(s)+I_2(s)], \end{array}$$ where $\Omega_s=\{sz:\ z\in\Omega\}$. We observe that $$I_1(s)\le c_{35}\int_{ B_{1/2}(\vec{n}_0)}\frac1{\|\vec{n}_0-z\|^{N-\alpha}} dz\le c_{36}$$ and since $(N-\alpha)p-\alpha<N$ by $p\in(0,\frac{N+\alpha}{N-\alpha})$, then $$\begin{aligned} I_2(s)&\le& c_{37}\int_{\tilde \Omega_s}\frac{1}{\|z\|^{(N-\alpha)p-\alpha}(1+\|z\|)^{N-\alpha}} dz \\&\le& c_{37}\int_{B_{\frac Rs}(0)\setminus B_{\frac12}(0)}\frac{1}{\|z\|^{(N-\alpha)p-2\alpha+N}} dz \\&\le& \left\{\arraycolsep=1pt \begin{array}{lll} c_{38} s^{(N-\alpha)p-2\alpha}\quad &{\rm if}\quad p\in (\frac{2\alpha}{N-\alpha}, \frac{N+\alpha}{N-\alpha}),\\[1.5mm] -c_{38} \ln s \quad &{\rm if}\quad p= \frac{2\alpha}{N-\alpha},\\[1.5mm] c_{38}\quad &{\rm if}\quad p\in(0,\frac{2\alpha}{N-\alpha}), \end{array} \right.\end{aligned}$$ where $c_{35},c_{36},c_{37},c_{38}>0$ and $R>0$ such that $\Omega \subset B_R(0)$. Then (\[4.6\]) holds. $\Box$ [Proof of Theorem \[teo 2\] $(ii)$.]{} For $p\in(1+\frac{2\alpha}{N},\frac{N+\alpha}{N-\alpha})$, we have that $$-\frac{2\alpha}{p-1}\in(-N,-N+\alpha)$$ and it follows by Lemma \[lm 2.1\] that $$u_k(x)\le k\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_0}{\partial \vec{n}^\alpha}](x)\le \frac{c_5k}{\|x\|^{N-\alpha}},\quad x\in\Omega.$$ Then $\lim_{x\in\Omega,\|x\|\to0}\frac{u_k(x)}{ w_p(x)}=0$ and we claim that $$u_k\le \lambda_0w_p\quad{\rm in}\quad\Omega.$$ In fact, if it fails, then there exists $z_0\in\Omega$ such that $$(u_k-\lambda_0w_p)(z_0)=\inf_{\Omega}(u_k-\lambda_0w_p)={\rm ess}\inf_{{\mathbb{R}}^N}(u_k-\lambda_0w_p)<0.$$ Then we have $(-\Delta)^\alpha(u_k-\lambda_0w_p)(z_0)<0$, which contradicts the fact that $$(-\Delta)^\alpha(u_k-\lambda_0w_p)(z_0)=\lambda_0w_p^p(z_0)-u_k^p(z_0)>0.$$ By monotonicity of the mapping $k\to u_k$, there holds $$u_\infty(x):=\lim_{k\to\infty} u_k(x),\quad x\in{\mathbb{R}}^N\setminus\{0\},$$ which is a classical solution of (\[4.2\]) and $$u_\infty(x)\le \lambda_0w_p(x)= \lambda_0 \|x\|^{-\frac{2\alpha}{p-1}},\quad \forall x\in\Omega.$$ By applying Lemma \[lm 4.2\], we obtain that $u_\infty$ is continuous up to the boundary except the origin. Finally, we claim that there exists $c_{39}>0$ and $t_0<\sigma_0$ such that $$\label{13-08-0} u_\infty(x_t)\ge c_{39}t^{-\frac{2\alpha}{p-1}},\quad \forall t\in(0,t_0),$$ where $x_t=t\vec{n}_0\in\Omega$. Indeed, let $r_k=(\sigma^{-1} k)^{\frac{p-1}{(N-\alpha)p-N-\alpha}}$, where $\sigma>0$ will be chosen later, then $k=\sigma r_k^{\frac{(N-\alpha)p-N-\alpha}{p-1}}$ and for $x\in A_{r_0}\cap \left[B_{r_k}(0)\setminus B_{\frac{r_k}{2}}(0)\right]$, we apply Lemma \[lm 4.3\] with $p\in(1+\frac{2\alpha}{N},\frac{N+\alpha}{N-\alpha})$ that $$\begin{aligned} u_k(x)&\ge & k\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x)-k^p\mathbb{G}_\alpha[(\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])^p](x) \\ &\ge& c_5k\|x\|^{\alpha-N}[1-c_{40}k^{p-1}\|x\|^{(\alpha-N)p+\alpha+N}] \\&\ge& c_5\sigma r_k^{-\frac{2\alpha }{p-1}}[1-c_{40}\sigma^{p-1} r_k^{p-1}(r_k/2)^{(\alpha-N)p+\alpha+N}] \\&\ge& c_5\sigma r_k^{-\frac{2\alpha }{p-1}}[1-c_{40}\sigma^{p-1} 2^{(N-\alpha)p-\alpha-N}] \\&\ge& \frac{c_5\sigma}{2}\|x\|^{-\frac{2\alpha }{p-1}},\end{aligned}$$ where we choose $\sigma$ such that $c_{40}\sigma^{p-1} 2^{(N-\alpha)p-\alpha-N}=\frac12$. Then for any $x\in A_{r_0}\cap B_{r_k}^c(0)$, there exists $k>0$ such that $x\in A_{r_0}\cap [B_{r_k}(0)\setminus B_{\frac{r_k}2}(0)]$ and then $$u_\infty(x)\ge u_k(x)\ge \frac{c_5\sigma}{2}\|x\|^{-\frac{2\alpha }{p-1}},\qquad \forall x\in A_{r_0}\cap B_{r_k}^c(0).$$ This ends the proof. $\Box$ The limit of $\{u_k\}$ blows up when $p\in(0,1+\frac{2\alpha}{N}]$ ------------------------------------------------------------------ In this subsection, we derive the blow-up behavior of the limit of $\{u_k\}$ when $p\in(0,1+\frac{2\alpha}{N}]$. To this end, we first do precise estimate for $u_k$. \[lm 3.2\] Assume that $g(s)=s^p$ with $p\in(1,\frac{N}{N-\alpha}]$ and $u_k$ is the solution of (\[eq 1.1\]) obtained by Theorem \[teo 1\]. Then there exist $c_{41}>0$, $r_0\in(0,\frac14)$ and $\{r_k\}_k\subset(0,r_0)$ satisfying $r_k\to0$ as $k\to\infty$ such that $$\label{4.3.1} u_k(x)\ge\frac{c_{41}\|x\|^{-N}}{-\ln(\|x\|)},\qquad \forall x\in A_{r_0}\cap B_{r_k}^c(0).$$ [Proof.]{} [To prove (\[4.3.1\]) in the case of $p\in (\frac{2\alpha}{N-\alpha},1+\frac{2\alpha}{N})$.]{} Let $r_j=j^{-\frac1\alpha}$ with $j\in(k_0,k)$, then $j=r_j^{-\alpha}$. Applying Lemma \[lm 4.3\] with $p\in (\frac{2\alpha}{N-\alpha},1+\frac{2\alpha}{N})$ and (\[11.04.1\]), we have that for $x\in A_{r_0}\cap \left[B_{r_j}(0)\setminus B_{\frac{r_j}2}(0)\right]$, $$\begin{aligned} u_j(x)&\ge & j\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x)-j^p\mathbb{G}_\alpha[(\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])^p](x) \\ &\ge& c_{34}^{-1}jr_j^{\alpha-N}-c_{32}j^p\|x\|^{(\alpha-N)p+2\alpha} \\&\ge& c_{34}^{-1}r_j^{-N}-c_{32}r_j^{-\alpha p-(N-\alpha)p+2\alpha} \\&\ge& \frac{1}{2c_{34}}\|x\|^{-N},\end{aligned}$$ where the last inequality holds since $-\alpha p-(N-\alpha)p+2\alpha>-N$ and $r_j\to0$ as $j\to\infty$. Then for any $x\in A_{r_0}\cap B_{r_k}^c(0)$, there exists $j\in (k_0,k)$ such that $x\in A_{r_0}\cap [B_{r_j}(0)\setminus B_{\frac{r_j}2}(0)]$ and then $$u_k(x)\ge u_j(x)\ge \frac{1}{2c_{34}}\|x\|^{-N},\qquad \forall x\in A_{r_0}\cap B_{r_k}^c(0).$$ [To prove (\[4.3.1\]) in the case of $p\in(0,\frac{2\alpha}{N-\alpha}]$.]{} Let $r_j=j^{-\frac1\alpha}$ with $j\in(k_0,k)$, then $j=r_j^{-\alpha}$ and for $x\in A_{r_0}\cap \left[B_{r_j}(0)\setminus B_{\frac{r_j}2}(0)\right]$, we have that $$\begin{aligned} u_j(x)&\ge & j\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x)-j^p\mathbb{G}_\alpha[(\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])^p](x) \\ &\ge& c_{34}^{-1}j\|x\|^{\alpha-N}-c_{32}j^p \\&\ge& c_{34}^{-1}r_j^{-N}-c_{32}r_j^{-\alpha p} \\&\ge& \frac{1}{2c_{34}}\|x\|^{-N},\end{aligned}$$ where the last inequality holds since $-\alpha p>-N$ and $r_j\to0$ as $j\to\infty$. For any $x\in A_{r_0}\cap B_{r_k}^c(0)$, there exists $j\in (k_0,k)$ such that $x\in A_{r_0}\cap [B_{r_j}(0)\setminus B_{\frac{r_j}2}(0)]$ and then $$u_k(x)\ge u_j(x)\ge \frac{1}{2c_{34}}\|x\|^{-N},\qquad \forall x\in A_{r_0}\cap B_{r_k}^c(0).$$ [To prove (\[4.3.1\]) in the case of $p=1+\frac{2\alpha}{N}$.]{} Let $\rho_j=j^{-\frac1\alpha}$ and $r_j=\frac{\rho_j}{[-\log(\rho_j)]^{\frac1\alpha}}$, then $j=\rho_j^{-\alpha}$ and applied Lemma \[lm 4.3\] for $x\in A_{r_0}\cap \left[B_{r_j}(0)\setminus B_{\frac{r_j}{2}}(0)\right]$, $$\begin{aligned} u_j(x)&\ge & j\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}](x)-j^p\mathbb{G}_\alpha[(\mathbb{G}_\alpha[\frac{\partial^\alpha \delta_{0}}{\partial \vec{n}^\alpha}])^p](x) \\ &\ge& c_{34}^{-1}j\|x\|^{\alpha-N}-c_{32}j^p\|x\|^{(\alpha-N)p+2\alpha} \\&\ge& c_{34}^{-1}\rho_j^{-N} (-\log \rho_j)^{\frac{N-\alpha}{\alpha}}-c_{42} \rho_j^{-N} (-\log \rho_j)^{\frac{(N-\alpha)p-2\alpha}{\alpha}} \\&=& c_{34}^{-1}\rho_j^{-N} (-\log \rho_j)^{\frac{N-\alpha}{\alpha}}\left[1-c_{42}(-\log \rho_j)^{\frac{(N-\alpha)p-2\alpha}{\alpha}-\frac{N-\alpha}{\alpha}}\right] \\&\ge& c_{34}^{-1}\frac{r_j^{-N}}{-\log \rho_j}[1-c_{42}(-\log \rho_j)^{\frac{(N-\alpha)p-N-\alpha}{\alpha}}] \\&\ge& \frac{c_{34}\|x\|^{-N}}{-2\log \|x\|},\end{aligned}$$ where $c_{42}>0$ and we used the facts that $\log(\rho_j)\le c\log r_j\le c\log \|x\|$ and $\frac{(N-\alpha)p-N-\alpha}{\alpha}<0$. Then for any $x\in A_{r_0}\cap B_{r_k}^c(0)$, there exists $j\in (k_0,k)$ such that $x\in A_{r_0}\cap [B_{r_j}(0)\setminus B_{\frac{r_j}2}(0)]$ and then $$u_k(x)\ge \frac{c_{34}\|x\|^{-N}}{-2\log \|x\|},\qquad x\in A_{r_0}\cap B_{r_k}^c(0).$$ The proof ends. $\Box$ [Proof of Theorem \[teo 2\] $(iii)$.]{} It derives by Lemma \[lm 3.2\] that $$\label{3.2.3} \pi_k:=\int_{B_{r_0}(0)}u_k(x)\ge c_{41}\int_{A_{r_0}\cap B_{r_k}^c(0)}\frac{\|x\|^{-N}}{-\log\|x\|}dx\to\infty\quad {\rm as}\ k\to\infty.$$ Fix $y_0\in \Omega\setminus \bar B_{r_0}(0)$, it follows by Lemma 2.4 in [@CY] that problem $$\label{5.1} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^p=0 \quad & {\rm in}\quad B_{\varrho_0}(y_0),\\[2mm] \phantom{ (-\Delta)^\alpha +u^p} u=0 \quad & {\rm in}\quad {\mathbb{R}}^N \setminus (B_{\varrho_0}(y_0)\cup B_{r_0}(0)),\\[2mm] \phantom{ (-\Delta)^\alpha +u^p} u=u_k \quad & {\rm in}\quad B_{r_0}(0) \end{array}$$ admits a unique solution $w_k$, where $\varrho_0=\min\{\rho_{\partial\Omega}(y_0),\|y_0\|-r_0\}$. By Lemma 2.2 in [@CY], $$\label{4.1.3} u_{k}\ge w_k\quad {\rm in}\quad B_{\varrho_0}(y_0).$$ Let $\tilde w_k=w_k-u_k\chi_{B_{r_0}(0)},$ then $\tilde w_k=w_k$ in $B_{\varrho_0}(y_0)$ and for $x\in B_{\varrho_0}(y_0)$, $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha \tilde w_k(x) = -\lim_{\epsilon\to0^+}\int_{B_{\varrho_0}(y_0)\setminus B_\epsilon(x)}\frac{w_k(z)-w_k(x)}{\|z-x\|^{N+2\alpha}}dz \\[3mm]\phantom{--------}+\lim_{\epsilon\to0^+}\int_{B_{\varrho_0}^c(y_0)\setminus B_\epsilon(x)}\frac{w_k(x)}{\|z-x\|^{N+2\alpha}}dz \\[3mm]\phantom{------} =-\lim_{\epsilon\to0^+}\int_{{\mathbb{R}}^N\setminus B_\epsilon(x)}\frac{w_k(z)-w_k(x)}{\|z-x\|^{N+2\alpha}}dz +\int_{B_{r_0}(0)}\frac{u_k(z)}{\|z-x\|^{N+2\alpha}}dz \\[3mm]\phantom{------}\ge(-\Delta)^\alpha w_k(x)+c_{42}\pi_k, \end{array}$$ where $c_{42}=(\|y_0\|+r_0)^{-N-2\alpha}$ and the last inequality follows by the fact of $$\|z-x\|\le \|x\|+\|z\|\le \|y_0\|+r_0\quad {\rm for}\ z\in B_{\frac14}(0),\ x\in B_{\frac14}(y_0).$$ Therefore, $$\begin{aligned} (-\Delta)^\alpha \tilde w_k(x)+\tilde w_k^p(x) &\ge& (-\Delta)^\alpha w_k(x)+w_k^p(x)+ c_{42}\pi_k \\ &=&c_{42}\pi_k, \qquad x\in B_{\varrho_0}(y_0), \end{aligned}$$ that is, $\tilde w_k$ is a super solution of $$\label{4.1.2} \arraycolsep=1pt \begin{array}{lll} \displaystyle (-\Delta)^\alpha u+u^p=c_{42}\pi_k \quad & {\rm in}\quad B_{\varrho_0}(y_0),\\[2mm] \phantom{ (-\Delta)^\alpha +u^{p,}} u=0 \quad & {\rm in}\quad B_{\varrho_0}^c(y_0). \end{array}$$ Let $\eta_1$ be the solution of $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u=1 \quad & {\rm in}\quad B_{\varrho_0}(y_0),\\[2mm] \phantom{ (-\Delta)^\alpha } u=0 \quad & {\rm in}\quad B^c_{\varrho_0}(y_0). \end{array}$$ Then $(c_{42}\pi_k)^{\frac1p} \frac{\eta_1}{2\max_{{\mathbb{R}}^N}\eta_1}$ is sub solution of (\[4.1.2\]) for $k $ large enough. By Lemma 2.2 in [@CY], we have that $$\tilde w_k(x)\ge (c_{42}\pi_k)^{\frac1p} \frac{\eta_1(x)}{2\max_{{\mathbb{R}}^N}\eta_1},\quad \forall x\in B_{\varrho_0}(y_0),$$ which implies that $$w_k(y)\ge c_{43} (c_{42}\pi_k)^{\frac1p},\qquad \forall y\in B_{\frac{\varrho_0}{2}}(y_0),$$ where $c_{43}=\min_{x\in B_{\varrho_0}(y_0)}\frac{\eta_1(x)}{2\max_{{\mathbb{R}}^N}\eta_1}$. Therefore, (\[4.1.3\]) and (\[3.2.3\]) imply that $$\lim_{k\to\infty}u_{k}(y)\ge \lim_{k\to\infty}w_k(y)=\infty,\qquad\forall y\in B_{\frac{\varrho_0}{2}}(y_0).$$ Similarly, we can prove $$\lim_{k\to\infty}u_{k}(y)\ge \lim_{k\to\infty}w_k(y)=\infty,\qquad\forall y\in \Omega.$$ The proof ends.$\Box$ Nonexistence in the critical case ================================= In this section, we prove the nonexistence in the critical case. To this end, we consider the weak solution to elliptic problem $$\label{14-08-0} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^{\frac{N+\alpha}{N-\alpha}}=k\frac{\partial^\alpha\delta_0}{\partial e_N^\alpha}\quad &{\rm in}\quad\ \ \overline{{\mathbb{R}}^N_+},\\[3mm] \phantom{(-\Delta)^\alpha +u^p} u=0\quad &{\rm in}\quad\ \ {\mathbb{R}}^N_-, \end{array}$$ where ${\mathbb{R}}^N_+={\mathbb{R}}^{N-1}\times{\mathbb{R}}_+$ and $e_N=(0,\cdots,0,1)$. \[weak definition\] A function $u\in L^1({\mathbb{R}}^N,\mu dx)$ is a weak solution of (\[14-08-0\]) if $u^p\in L^1({\mathbb{R}}^N,\rho^\alpha \mu dx)$ and $$\label{weak sense} \int_{{\mathbb{R}}^N_+} [u(-\Delta)^\alpha\xi+u^{\frac{N+\alpha}{N-\alpha}}\xi]dx=\frac{\partial^\alpha \xi(0)}{\partial e_N^\alpha},\quad \forall\xi\in \mathbb{X}_{\alpha,{\mathbb{R}}^N_+},$$ where $\mu(x)=\frac1{1+\|x\|^{N+2\alpha}}$, $\rho(x)=\min\{1,\rho_{\partial\Omega}(x)\}$ and $\mathbb{X}_{\alpha,{\mathbb{R}}^N_+}\subset C({\mathbb{R}}^N)$ is the space of functions $\xi$ satisfying: \(i) the support of $\xi$ is a compact set in $\bar{\mathbb{R}}^N_+$; \(ii) $(-\Delta)^\alpha\xi(x)$ exists for any $x\in{\mathbb{R}}^N_+$ and there exists $c>0$ such that $$\|(-\Delta)^\alpha\xi(x)\|\leq c\mu(x),\quad \forall x\in{\mathbb{R}}^N_+;$$ \(iii) there exist $\varphi\in L^1({\mathbb{R}}^N_+,\rho^\alpha dx)$ and $\varepsilon_0>0$ such that $\|(-\Delta)_\varepsilon^\alpha\xi\|\le \varphi$ a.e. in ${\mathbb{R}}^N_+$, for all $\varepsilon\in(0,\varepsilon_0]$. Let $\mathbb{G}_{\alpha,{\mathbb{R}}^N_+}$ the Green’s function on ${\mathbb{R}}^N_+\times{\mathbb{R}}^N_+$ and $$\label{5.2} \Gamma_\alpha(x)=\lim_{t\to0}t^{-\alpha}\mathbb{G}_{\alpha,{\mathbb{R}}^N_+}(x,te_N).$$ \[lm 5.1\] Let $\Gamma_\alpha$ defined in (\[5.2\]), then $$\label{5.1.1} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha \Gamma_\alpha=\frac{\partial^\alpha \delta_0}{\partial e_N^\alpha}\qquad &{\rm in}\quad\bar {\mathbb{R}}^N_+,\\[2mm] \phantom{--- } \Gamma_\alpha=0 \quad & {\rm in} \quad {\mathbb{R}}^N_-. \end{array}$$ Moreover, $$\label{5.3} \Gamma_\alpha(x)=\|x\|^{-N+\alpha}\Gamma_\alpha\left(\frac{x}{\|x\|}\right),\qquad x\in{\mathbb{R}}^N,$$ and $$\Gamma_\alpha\left(\frac{x}{\|x\|}\right)\ \left\{\arraycolsep=1pt \begin{array}{lll} >0 \quad & {\rm if}\quad x\in {\mathbb{R}}^N_+,\\[2mm] =0 \quad & {\rm if}\quad x\not\in {\mathbb{R}}^N_+. \end{array} \right.$$ [Proof.]{} We observe that $$(-\Delta)^\alpha_x t^{-\alpha}\mathbb{G}_{\alpha,{\mathbb{R}}^N_+}(x,te_N)=t^{-\alpha}\delta_{te_N}$$ and $$\lim_{t\to0^+}\langle t^{-\alpha}\delta_{te_N},\xi\rangle=\frac{\partial^\alpha \xi(0)}{\partial e_N^\alpha},\qquad \forall \xi\in\mathbb{X}_{\alpha,{\mathbb{R}}^N_+}.$$ Then (\[5.1.1\]) holds in the weak sense. By the regularity results, $\Gamma_\alpha$ is a solution of $$\label{5.1.2} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha \Gamma_\alpha=0\qquad &{\rm in}\quad{\mathbb{R}}^N_+,\\[2mm] \phantom{--- } \Gamma_\alpha=0 \quad & {\rm in} \quad\overline{{\mathbb{R}}^N_-}\setminus\{0\}. \end{array}$$ Let $\Gamma_{\alpha,\lambda}(x)=\lambda^{N-\alpha}\Gamma_{\alpha}(\lambda x)$ and $\xi_\lambda(x)=\xi(x/\lambda)$ for $\xi\in\mathbb{X}_{\alpha,{\mathbb{R}}^N_+}$, then we have that $$\begin{aligned} \int_{{\mathbb{R}}^N_+}\Gamma_{\alpha,\lambda}(-\Delta)^\alpha \xi dx&=&\lambda^\alpha\int_{{\mathbb{R}}^N_+}\Gamma_{\alpha}(z)(-\Delta)^\alpha \xi_\lambda(x)dx, \\ &=& \lambda^\alpha \frac{\partial^\alpha \xi_\lambda(0)}{\partial e_N^\alpha},\end{aligned}$$ which implies that $$\int_{{\mathbb{R}}^N_+}\Gamma_{\alpha,\lambda}(-\Delta)^\alpha \xi dx=\frac{\partial^\alpha \xi(0)}{\partial e_N^\alpha}.$$ By the uniqueness, we derive that $$\lambda^{N-\alpha}\Gamma_{\alpha}(\lambda x)=\Gamma_{\alpha}(x),$$ which, choosing $\lambda=\frac1{\|x\|}$, implies (\[5.3\]). The last argument is obvious. $\Box$ \[teo 4.1\] Let $k>0$, then problem (\[14-08-0\]) has no any weak solution. [Proof.]{} If there exists a weak solution $u_k$ to (\[14-08-0\]), then we observe that $$u_k>0\qquad {\rm in}\quad {\mathbb{R}}^N_+.$$ By Maximum Principle, we have that $$\label{5.5} u_k\le k\Gamma_\alpha\qquad {\rm in}\quad {\mathbb{R}}^N.$$ Denoting $$u_\infty=\lim_{k\to\infty}u_k\qquad {\rm in}\quad {\mathbb{R}}^N.$$ We claim that $$\label{3.2.1} u_\infty(x)=\|x\|^{\alpha-N}u_\infty(\frac{x}{\|x\|}),\quad \forall x\in{\mathbb{R}}^N\setminus\{0\}.$$ Indeed, let $$\tilde u_\lambda(x)=\lambda^{N-\alpha}u_k(\lambda x),\quad \forall x\in {\mathbb{R}}^N\setminus\{0\}.$$ By direct computation, we have that for $x\in{\mathbb{R}}^N_+$, $$\begin{aligned} (-\Delta)^\alpha \tilde u_\lambda(x) +\tilde u_\lambda^{\frac{N+\alpha}{N-\alpha}}(x) &=&\lambda^{N+\alpha}[(-\Delta)^\alpha u_k(\lambda x) + u_k^{\frac{N+\alpha}{N-\alpha}}(\lambda x)] \nonumber \\ &=&0.\label{13-09-5}\end{aligned}$$ Moreover, for $f\in C_0^1({\mathbb{R}}^N_+)$, $$\begin{aligned} \langle(-\Delta)^\alpha \tilde u_\lambda +\tilde u_\lambda^{\frac{N+\alpha}{N-\alpha}}, f\rangle&=&\lambda^{N+\alpha}\int_{{\mathbb{R}}^N} [(-\Delta)^\alpha u_k(\lambda x) + u_k^{\frac{N+\alpha}{N-\alpha}}(\lambda x)]f(x)dx\nonumber \\&=&\lambda^{ \alpha }\int_{{\mathbb{R}}^N} [(-\Delta)^\alpha u_k(z) + u_k^{\frac{N+\alpha}{N-\alpha}}(z)]f\left(\frac{z}{\lambda}\right)dz\nonumber \\ &=&\lambda^{\alpha}k\frac{\partial^\alpha f(0)}{\partial e_N^\alpha}.\end{aligned}$$ Thus, $$\label{3.2.2} (-\Delta)^\alpha \tilde u_\lambda +\tilde u_\lambda^{\frac{N+\alpha}{N-\alpha}}=\lambda^{\alpha}k\frac{\partial^\alpha \delta_0}{\partial e_N^\alpha}\quad {\rm in}\ \ {\mathbb{R}}^N_+.$$ We observe that $\lim_{\|x\|\to\infty}\tilde u_\lambda(x)=0$ and $u_{k\lambda^{\alpha }}$ is the unique weak solution of (\[14-08-0\]) with $k$ replaced by $\lambda^\alpha k$, then for $x\in {\mathbb{R}}^N\setminus\{0\}$, $$\label{21-10-1} u_{k\lambda^{\alpha}}(x)=\tilde u_\lambda (x)=\lambda^{N-\alpha}u_k(\lambda x)$$ and letting $k\to\infty$ we have that $$u_{\infty}(x)=\lambda^{N-\alpha}u_\infty(\lambda x),\qquad \forall x\in {\mathbb{R}}^N\setminus\{0\},$$ which implies (\[3.2.1\]) by taking $\lambda=\|x\|^{-1}$. Combine (\[5.3\]), (\[5.5\]) and (\[21-10-1\]), then we have that $$\begin{aligned} u_{k\lambda^{\alpha}}(x) \le \lambda^{N-\alpha}k\Gamma_\alpha(\lambda x) = k\Gamma_\alpha(x),\quad \forall x\in {\mathbb{R}}^N. \end{aligned}$$ Thus, $$u_\infty(x)\le k\Gamma_\alpha(x),\quad \quad \forall x\in {\mathbb{R}}^N.$$ By arbitrary of $k$, it implies that $$u_\infty\equiv0,$$ then $u_1\equiv 0$ in ${\mathbb{R}}^N$, which is impossible. $\Box$ [Proof of Theorem \[teo 4\].]{} Without loss generality, we let $k=1$, $0\in\partial\Omega$ and $e_N$ is the unit normal vector pointing inside of $\Omega$ at $0$. If $$\arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^{\frac{N+\alpha}{N-\alpha}}=\frac{\partial^\alpha\delta_0}{\partial e_N^\alpha}\quad &{\rm in}\quad\ \ \bar\Omega,\\[3mm] \phantom{(-\Delta)^\alpha +u^{\frac{N+\alpha}{N-\alpha}}} u=0\quad &{\rm in}\quad\ \ \bar\Omega^c \end{array}$$ admits a solution weak $v_1$, we claim that there is a weak solution of (\[14-08-0\]), then the contradiction is obtained from Theorem \[teo 4.1\]. In fact, we may assume that $$\Omega=B_1(e_N)\quad{\rm and}\quad B_m=B_m(me_N).$$ Then $$\Omega\subset B_m\subset B_{m+1}\quad {\rm and}\quad \lim_{m\to\infty} B_m={\mathbb{R}}^N_+.$$ Let $$v_m(x)=m^{\alpha-N}v_1(\frac xm),\quad x\in{\mathbb{R}}^N.$$ By direct computation, $v_m$ is a weak solution of $$\label{15-08-0-0} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^{\frac{N+\alpha}{N-\alpha}}=\frac{\partial^\alpha\delta_0}{\partial e_N^\alpha}\quad &{\rm in}\quad\ \ \bar B_m,\\[3mm] \phantom{(-\Delta)^\alpha +u^{\frac{N+\alpha}{N-\alpha}}} u=0\quad &{\rm in}\quad\ \ \bar B_m^c, \end{array}$$ [We next show that $v_m\le v_{m+1}$ in ${\mathbb{R}}^N$. ]{} From Proposition \[pr 1\], $$\label{08-14-00} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^{\frac{N+\alpha}{N-\alpha}}=t^{-\alpha}\delta_{te_N}\quad &{\rm in}\quad\ \ B_m,\\[3mm] \phantom{(-\Delta)^\alpha +u^{\frac{N+\alpha}{N-\alpha}}} u=0\quad &{\rm in}\quad\ \ B_m^c \end{array}$$ admits a unique weak solution, denoting $v_{m,t}$. Choose a sequence nonnegative functions $\{f_{m,i}\}_{i\in{\mathbb{N}}}\subset C^1({\mathbb{R}}^N)$ with support $B_1(e_N)$ such that $f_{m,i}\rightharpoonup t^{-\alpha}\delta_{te_N}$ as $i\to\infty$ in the distribution sense. Let $v_{m,i,t}$ be the unique solution of $$\label{08-14-00-0} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^{\frac{N+\alpha}{N-\alpha}}=f_{m,i}\quad &{\rm in}\quad\ \ B_m,\\[3mm] \phantom{(-\Delta)^\alpha +u^{\frac{N+\alpha}{N-\alpha}}} u=0\quad &{\rm in}\quad\ \ B_m^c \end{array}$$ and by Maximum Principle, see [@CY Lemma 2.3], derive that $$v_{m,i,t}\le \tilde v_{m+1,i,t}\quad{\rm in}\quad {\mathbb{R}}^N.$$ Together with the facts that $v_{m,i,t}\to v_{m,t}$ a.e. in ${\mathbb{R}}^N$ and $v_{m+1,i,t}\to v_{m+1,t}$ a.e. in ${\mathbb{R}}^N$ as $i\to\infty$, we obtain that $$\label{3.2} v_{1,t}\le v_{m,t}\le v_{m+1,t}\quad{\rm a.e.\ in}\ \ {\mathbb{R}}^N$$ and $$\int_{B_m}v_{m,t}^{\frac{N+\alpha}{N-\alpha}}\rho^\alpha dx< {\\|\mathbb{G}_{\alpha, B_m}[f_{m,i}]\\|}_{L^1(\Omega,\ \rho^\alpha dx)},$$ which implies that $$\label{08-14-0-1} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u+u^{\frac{N+\alpha}{N-\alpha}}=\frac{\partial^\alpha \delta_0}{\partial e_N^\alpha}\quad &{\rm in}\quad\ \ \bar B_m,\\[3mm] \phantom{(-\Delta)^\alpha +u^{\frac{N+\alpha}{N-\alpha}}} u=0\quad &{\rm in}\quad\ \ \bar B_m^c \end{array}$$ admits a solution $v_m$ for any $m\in{\mathbb{N}}$ and $$\label{3.3} v_{m}\le v_{m+1}\quad {\rm a.e.\ in}\ \ {\mathbb{R}}^N.$$ We observe that $$\label{3.4} 0\le v_{m}\le \mathbb{G}_{\alpha, B_{m}}[\frac{\partial^\alpha \delta_0}{\partial e_N^\alpha}]\le \frac{c_5}{\|x\|^{N-\alpha}} \quad {\rm a.e.\ in}\ \ {\mathbb{R}}^N$$ and $$\int_{B_m}v_{m}^{\frac{N+\alpha}{N-\alpha}}\rho^\alpha dx< {\\|\mathbb{G}_{\alpha, B_m}[\frac{\partial^\alpha \delta_0}{\partial e_N^\alpha}]\\|}_{L^1(B_m,\ \rho^\alpha dx)}.$$ By (\[3.3\]) and (\[3.4\]), we see that the limit of $\{v_m\}$ exists, denoted it by $w_1$. Hence, $$\label{2.1.1} 0\le w_1\le \mathbb{G}_{\alpha,{\mathbb{R}}^N_+}[\frac{\partial^\alpha \delta_0}{\partial e_N^\alpha}]\quad {\rm a.e.\ in}\ \ {\mathbb{R}}^N$$ and $$\int_{{\mathbb{R}}^N_+}w_1^{\frac{N+\alpha}{N-\alpha}}\rho^\alpha dx< {\\|\mathbb{G}_{\alpha,{\mathbb{R}}^N_+}[\frac{\partial^\alpha \delta_0}{\partial e_N^\alpha}]\\|}_{L^1({\mathbb{R}}^N_+,\rho^\alpha\mu dx)},$$ which implies that $w_1\in L^1({\mathbb{R}}^N,\ \mu dx)$. Thus, $v_m\to w_1$ in $L^1({\mathbb{R}}^N,\ \rho^\alpha \mu dx)$ as $m\to\infty$. For $\xi\in \mathbb{X}_{\alpha,{\mathbb{R}}^N_+}$, there exists $N_0>0$ such that for any $m\ge N_0$, $${\rm supp}(\xi)\subset \bar B_m,$$ which implies that $\xi\in \mathbb{X}_{\alpha,B_m}$ and then $$\label{3.6} \int_{{\mathbb{R}}^N_+} [v_m(-\Delta)^\alpha\xi+v_m^{\frac{N+\alpha}{N-\alpha}}\xi]dx= \frac{\partial^\alpha \xi(0)}{\partial e_N^\alpha}.$$ By [@CY Lemma 3.1 ], $$\|(-\Delta)^\alpha\xi(x)\|\le \frac{c_{9}{\\|\xi\\|}_{L^\infty(\Omega)}}{1+\|x\|^{N+2\alpha}},\quad \forall x\in {\mathbb{R}}^N_+.$$ Thus, $$\label{3.7} \lim_{m\to\infty}\int_{{\mathbb{R}}^N_+} v_m(x)(-\Delta)^\alpha\xi(x) dx=\int_{{\mathbb{R}}^N_+} w_1(x)(-\Delta)^\alpha\xi(x) dx.$$ By (\[2.1.1\]) and increasing monotonicity of $v_m$, for any $n\ge N_0$, $$\label{3.8} \lim_{m\to\infty}\int_{{\mathbb{R}}^N_+} v_m^{\frac{N+\alpha}{N-\alpha}}\xi(x) dx=\int_{{\mathbb{R}}^N_+} w_1^{\frac{N+\alpha}{N-\alpha}}\xi(x) dx.$$ Combining (\[3.7\]), (\[3.8\]) and taking $m\to\infty$ in (\[3.6\]), we obtain that $$\label{3.10} \int_{{\mathbb{R}}^N_+} \left[w_1(-\Delta)^\alpha\xi+w_1^{\frac{N+\alpha}{N-\alpha}}\xi\right]dx=\frac{\partial^\alpha \xi(0)}{\partial e_N^\alpha}.$$ Since $\xi\in \mathbb{X}_{\alpha,{\mathbb{R}}^N_+}$ is arbitrary, $w_1$ is a weak solution of (\[14-08-0\]).$\Box$ Forcing nonlinearity ==================== This section is devoted to consider problem (\[eq 1.1\]) when $\epsilon=-1$, we call it as forcing case. In order to derive the existence of weak solution to (\[eq 1.1\]) with forcing nonlinearity, we first introduce the following propositions. \[general\] [@CFV Proposition 2.2] Let $\alpha\in(0,1]$, $\beta\in[0,\alpha]$ and $\nu\in\mathfrak{M}(\Omega,\rho^\beta_{\partial\Omega})$, then there exists $c_{44}>0$ such that $$\label{annex 00} \\|\mathbb{G}_\alpha[\nu]\\|_{M^{p_\beta^}(\Omega,\rho^\beta_{\partial\Omega} dx)}\le c_{44}\\|\nu\\|_{\mathfrak{M}(\Omega,\rho^\beta_{\partial\Omega})},$$ where $p_\beta^=\frac{N+\beta}{N-2\alpha+\beta}$. \[pr5\] [@CFV Proposition 2.3] Let $\alpha\in(0,1]$ and $\beta\in [0, \alpha]$, then the mapping $f\mapsto \mathbb G_\alpha[f]$ is compact from $L^{1}(\Omega,\rho^\beta_{\partial\Omega} dx)$ into $L^{q}(\Omega)$ for any $q\in [1,\frac{N}{N+\beta-2\alpha})$. Moreover, for $q\in [1,\frac{N}{N+\beta-2\alpha})$, there exists $c_{45}>0$ such that for any $f\in L^{1}(\Omega,\rho^{\beta}_{\partial\Omega}dx)$ $$\label{power1} {\\|\mathbb G_\alpha[f]\\|}_{L^q(\Omega)}\leq c_{45}{\\|f\\|}_{L^{1}(\Omega,\rho^{\beta}_{\partial\Omega}dx)}.$$ For $\nu\in\mathfrak{M}^b_{\partial\Omega}(\bar\Omega)$, $\nu_t$ is given in section 2.2 for $t\in(0,\sigma_0)$. Let $t_j=\frac1j\in(0,\sigma_0/4)$ if $j\ge j_0$ for some $j_0>0$. Choose $\{\tilde\nu_n\}_n\subset C_0^1(\Omega)$ a sequence of nonnegative functions such that supp$(\tilde\nu_n)\subset\Omega_{t_{j_0}-2^{-n}}\setminus\Omega_{t_{j_0}+2^{-n}}$ and $\tilde\nu_{n}\to\nu_{t_{j_0}} $ in the duality sense with $C(\bar \Omega)$. Denote $$\nu_{n,j}(x)=\left\{ \arraycolsep=1pt \begin{array}{lll} \tilde \nu_{n}(x+t_j\vec{n_x}),\quad& {\rm if}\quad x\in \Omega_{t_{j_0}-2^{-n}}\setminus\Omega_{t_{j_0}+2^{-n}}, \\[2mm] 0,&{\rm if\ not.} \end{array} \right.$$ \[lm 6.1\] Up to subsequence, we have that $\nu_{n,j_n}\to\nu $ in the duality sense with $C(\bar \Omega)$, that is, $$\label{06-08} \lim_{n\to\infty}\int_{\bar \Omega}\zeta \nu_{n,j_n }dx=\int_{\bar \Omega}\zeta d\nu,\qquad\forall \zeta\in C(\bar \Omega).$$ Moreover, $${\rm supp}(\nu_n)\subset \Omega_{\frac{t_n}2}\setminus \Omega_{2t_n}.$$ [Proof.]{} For any fixed $j$ and $\zeta\in C(\bar\Omega)$, we observe that $$\begin{aligned} \lim_{n\to\infty}\int_{\bar \Omega}\zeta \nu_{n,j}dx = \int_{\Omega}\zeta d\nu_{t_j}\end{aligned}$$ and pass $j\to\infty$, we derive that $$\begin{aligned} \lim_{j\to\infty}\lim_{n\to\infty}\int_{\bar \Omega}\zeta \nu_{n,j}dx = \int_{\Omega}\zeta d\nu.\end{aligned}$$ The second argument is obvious by the definition of $\nu_{n,j}$. $\Box$ Sub-linear ------------ In this subsection, we are devoted to prove the existence of weak solution to (\[eq 1.1\]) when the source nonlinearity is sub-linear. [Proof of Theorem \[teo 3\] $(i)$.]{} Let $\{\nu_n\}$ be a sequence of nonnegative functions such that $\nu_{n}\to\nu $ in sense of duality with $C(\bar\Omega)$, see Lemma \[lm 6.1\]. By the Banach-Steinhaus Theorem, we may assume that ${\\|\nu_n\\|}_{L^1(\Omega)}\le {\\|\nu\\|}_{\mathfrak M^b (\Omega)}=1$ for all $n$. We consider a sequence $\{g_n\}$ of $C^1$ nonnegative functions defined on ${\mathbb{R}}_+$ such that $g_n(0)=g(0)$, $$\label{06-08-1} g_n\le g_{n+1}\le g,\quad \sup_{s\in{\mathbb{R}}_+}g_n(s)=n\quad{\rm and}\quad \lim_{n\to\infty}{\\|g_n-g\\|}_{L^\infty_{loc}({\mathbb{R}}_+)}=0.$$ We set $$M(v)={\\|v\\|}_{L^{1}(\Omega)}.$$ [Step 1. To prove that for $n\geq 1$, $$\label{002.3} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u= g_{n}(u)+kt_n^{-\alpha}\nu_n\quad & {\rm in}\quad\Omega,\\[2mm] \phantom{ (-\Delta)^\alpha } u=0\quad & {\rm in}\quad \Omega^c \end{array}$$ admits a nonnegative solution $u_n$ such that $$M(u_n)\le \bar\lambda,$$ where $\bar\lambda>0$ independent of $n$. ]{} To this end, we define the operators $\{\mathcal{T}_n\}$ by $$\mathcal{T}_nu=\mathbb{G}_\alpha\left[g_n(u)+k t_n^{-\alpha}\nu_n\right],\qquad \forall u\in L^1_+(\Omega),$$ where $L^1_+(\Omega)$ is the positive cone of $L^1(\Omega)$. By (\[power1\]) and (\[06-08-2\]), we have that $$\label{23-05-0} \arraycolsep=1pt \begin{array}{lll} M(\mathcal{T}_nu)\le c_{45}{\\|g_n(u)+k t_n^{-\alpha}\nu_n\\|}_{L^1 (\Omega,\rho_{\partial\Omega}^{\alpha}dx)} \\[2mm] \phantom{---- } \le c_3c_{45} \int_{\Omega}u^{p_0}\rho^\alpha(x)dx+c_{46}(k+\epsilon) \\[2mm] \phantom{---- } \le c_3c_{47}\int_{\Omega}u^{p_0} dx+c_{46}(k+\epsilon) \\[2mm] \phantom{---- } \le c_3 c_{48}(\int_{\Omega}u dx)^{p_0}+c_{46}(k+\epsilon) \\[2mm] \phantom{---- } =c_3c_{48}M(u)^{p_0}+c_{46}(k+\epsilon), \end{array}$$ where $c_{47},c_{48}>0$ independent of $n$. Therefore, we derive that $$M(\mathcal{T}_nu)\le c_3c_{48} M(u)^{p_0}+c_{45}(k+\epsilon).$$ If we assume that $M(u)\le \lambda$ for some $\lambda>0$, it implies $$M(\mathcal{T}_nu)\le c_3c_{48} \lambda^{p_0}+c_{45}(k+\epsilon).$$ In the case of $p_0<1$, the equation $$c_3c_{48}\lambda^{p_0 }+c_{45}(k+\epsilon)=\lambda$$ admits a unique positive root $\bar\lambda$. In the case of $p_0=1$, for $c_3>0$ satisfying $c_3c_{48}<1$, the equation $$c_3c_{48}\lambda+c_{45}(k+\epsilon)=\lambda$$ admits a unique positive root $\bar\lambda$. For $M(u)\le \bar\lambda$, we obtain that $$\label{07-05-5jingxuan} M(\mathcal{T}_nu)\le c_3c_{48}\bar\lambda^{p_0}+c_{45}(k+\epsilon)= \bar\lambda.$$ Thus, $\mathcal{T}_n$ maps $L^1(\Omega)$ into itself. Clearly, if $u_m\to u$ in $L^1(\Omega)$ as $m\to\infty$, then $g_n(u_m)\to g_n(u)$ in $L^1(\Omega)$ as $m\to\infty$, thus $\mathcal{T}_n$ is continuous. For any fixed $n\in{\mathbb{N}}$, $\mathcal{T}_nu_m=\mathbb{G}_\alpha\left[g_n(u_m)+k \nu_n\right]$ and $\{g_n(u_m)+k \nu_n\}_m$ is uniformly bounded in $L^1(\Omega,\rho_{\partial\Omega}^\beta dx)$, then it follows by Proposition \[pr5\] that $\{\mathbb{G}_\alpha\left[g_n(u_m)+k t_n^{-\alpha}\nu_n\right]\}_m$ is pre-compact in $L^1(\Omega)$, which implies that $\mathcal{T}_n$ is a compact operator. Let $$\displaystyle\begin{array}{lll}\displaystyle \mathcal{G}=\{u\in L^1_+(\Omega): \ M(u)\le \bar\lambda \}, \end{array}$$ which is a closed and convex set of $L^1(\Omega)$. It infers by (\[07-05-5jingxuan\]) that $$\mathcal{T}_n(\mathcal{G})\subset \mathcal{G}.$$ It follows by Schauder’s fixed point theorem that there exists some $u_n\in L^1_+(\Omega)$ such that $\mathcal{T}_nu_n=u_n$ and $M(u_n)\le \bar\lambda,$ where $\bar\lambda>0$ independent of $n$. We observe that $u_n$ is a classical solution of (\[002.3\]). Let open set $O$ satisfy $ O\subset \bar O\subset \Omega$. By [@RS Proposition 2.3], for $\theta\in(0,2\alpha)$, there exists $c_{49}>0$ such that $${\\|u_n\\|}_{C^{\theta}(O)}\le c_{49}\{{\\|g(u_n)\\|}_{L^\infty(\Omega)}+kt_n^{-\alpha}{\\|\nu_n\\|}_{L^{\infty}(\Omega)}\},$$ then applied [@RS Corollary 2.4], $u_n$ is $C^{2\alpha+\epsilon_0}$ locally in $\Omega$ for some $\epsilon_0>0$. Then $u_n$ is a classical solution of (\[002.3\]). Moreover, from [@CV2 Lemma 2.2], we derive that $$\label{5.60000} \int_\Omega u_n(-\Delta)^\alpha\xi dx=\int_\Omega g(u_n)\xi dx+k\int_\Omega\xi t_n^{-\alpha}\nu_ndx,\quad \forall\xi\in \mathbb{X}_{\alpha}.$$ [Step 2. Convergence. ]{} We observe that $\{g_n( u_n)\}$ is uniformly bounded in $L^1(\Omega,\rho_{\partial\Omega}^\alpha dx)$, so is $\{\nu_n\}$. By Proposition \[pr5\], there exist a subsequence $\{u_{n_k}\}$ and $u$ such that $u_{n_k}\to u$ a.e. in $\Omega$ and in $L^1(\Omega)$, then by (\[06-08-2\]), we derive that $g_{n_k}(u_{n_k}) \to g( u)$ in $L^1(\Omega)$. Pass the limit of (\[5.60000\]) as $n_k\to \infty$ to derive that $$\int_\Omega u(-\Delta)^\alpha\xi=\int_\Omega g(u)\xi dx+k\int_\Omega\frac{\partial^\alpha\xi}{\partial \vec{n}^\alpha} d\nu,\quad \forall \xi\in\mathbb{X}_\alpha,$$ thus $u$ is a weak solution of (\[eq 1.1\]) and $u$ is nonnegative since $\{u_n\}$ are nonnegative. $\Box$ Integral subcritical --------------------- In this subsection, we prove the existence of weak solution to (\[eq 1.1\]) when the nonlinearity is integral subcritical. [Proof of Theorem \[teo 3\] $(ii)$.]{} Let $\{\nu_n\}\subset C^1(\bar \Omega)$ be a sequence of nonnegative functions given as the above and ${\\|\nu_n\\|}_{L^1(\Omega)}\le 2{\\|\nu\\|}_{\mathfrak M ^b(\bar\Omega)}=1$ for all $n$. We consider a sequence $\{g_n\}$ of $C^1$ nonnegative functions defined on ${\mathbb{R}}_+$ satisfying $g_n(0)=g(0)$ and (\[06-08-1\]). We set $$M_1(v)={\\|v\\|}_{M^{\frac{N+\alpha}{N-\alpha}}(\Omega,\rho^\alpha_{\partial\Omega} dx)}\quad{\rm and}\quad M_2(v)={\\|v\\|}_{L^{p_}(\Omega)},$$ where $p_$ is (\[1.4\]). We may assume that $p_\in(1, \frac{N}{N-\alpha})$. In fact, if $p_\ge \frac{N}{N-\alpha}$, then for any given $p\in(1, \frac{N}{N-\alpha})$, (\[1.4\]) implies that $$g(s)\le c_{4}s^p+\epsilon,\quad \forall s\in[0,1].$$ [Step 1. To prove that for $n\geq 1$, $$\label{2.3} \arraycolsep=1pt \begin{array}{lll} (-\Delta)^\alpha u= g_{n}(u)+kt_n^{-\alpha}\nu_n\quad & {\rm in}\quad\Omega,\\[2mm] \phantom{ (-\Delta)^\alpha_n } u=0\quad & {\rm in}\quad \Omega^c \end{array}$$ admits a nonnegative solution $u_n$ such that $$M_1(u_n)+M_2(u_n)\le \bar\lambda,$$ where $\bar\lambda>0$ independent of $n$. ]{} To this end, we define the operators $\{\mathcal{T}_n\}$ by $$\mathcal{T}_nu=\mathbb{G}_\alpha\left[g_n(u)+kt_n^{-\alpha} \nu_n\right],\qquad \forall u\in L^1_+(\Omega).$$ By Proposition \[general\], we have $$\begin{aligned} M_1(\mathcal{T}_nu) &\le& c_{44}{\\|g_n(u)+kt_n^{-\alpha} \nu_n\\|}_{L^1 (\Omega,\rho^{\alpha}_{\partial\Omega}dx)}\nonumber\\[2.5mm] &\le & c_{44} [{\\|g_n(u)\\|}_{L^1(\Omega,\rho^{\alpha}_{\partial\Omega}dx)}+k].\label{06-08-10}\end{aligned}$$ In order to deal with ${\\|g_n(u)\\|}_{L^1(\Omega,\rho^{\beta}_{\partial\Omega}dx)}$, for $\lambda >0$ we set $$S_\lambda=\{x\in\Omega:u(x)>\lambda\}\quad {\rm and}\quad \omega(\lambda)=\int_{S_\lambda}\rho^\alpha_{\partial\Omega} dx,$$ $$\label{chenyuhang1} \displaystyle\begin{array}{lll} \displaystyle{\\|g_n(u)\\|}_{L^1(\Omega,\rho^{\alpha}_{\partial\Omega}dx)}\le \int_{S^c_1}g(u)\rho^{\alpha}_{\partial\Omega}dx+\int_{S_1} g(u)\rho^{\alpha}_{\partial\Omega}dx. \end{array}$$ We first deal with $\int_{S_1} g(u)\rho^{\alpha}dx$. In fact, we observe that $$\int_{S_1} g(u)\rho^{\alpha}_{\partial\Omega}dx=\omega(1) g(1)+\int_1^\infty \omega(s)dg(s),$$ where $$\int_1^\infty g(s)d\omega(s)=\lim_{T\to\infty}\int_1^T g(s)d\omega(s).$$ It infers by Proposition \[pr 1\] and Proposition \[general\] that there exists $c_{50}>0$ such that $$\label{2.4} \omega(s)\leq c_{50}M_1(u)^{\frac{N+\alpha}{N-\alpha}}s^{-\frac{N+\alpha}{N-\alpha}}$$ and by (\[1.4\]) and Lemma \[lm 08-09\] with $p=\frac{N+\alpha}{N-\alpha}$, there exist a sequence of increasing numbers $\{T_j\}$ such that $T_1>1$ and $T_j^{-\frac{N+\alpha}{N-\alpha}} g(T_j)\to 0$ when $j\to\infty$, thus $$\displaystyle\begin{array}{lll} \displaystyle \omega(1) g(1)+ \int_1^{T_j} \omega(s)d g(s) \le c_{50}M_1(u)^{\frac{N+\alpha}{N-\alpha}} g(1)+c_{50}M(u)^{\frac{N+\alpha}{N-\alpha}}\int_1^{T_j} s^{-\frac{N+\alpha}{N-\alpha}}d g(s) \\[4mm]\phantom{-----}\displaystyle \leq c_{50}M_1(u)^{\frac{N+\alpha}{N-\alpha}}{T_j}^{-\frac{N+\alpha}{N-\alpha}} g(T_j)+\frac{c_{50}M_1(u)^{\frac{N+\alpha}{N-\alpha}}}{\frac{N+\alpha}{N-\alpha}+1}\int_1^{T_j} s^{-1-\frac{N+\alpha}{N-\alpha}} g(s)ds. \end{array}$$ Therefore, $$\label{06-08-11} \displaystyle\begin{array}{lll} \int_{S_1} g(u)\rho^{\alpha}dx=\omega(1)g(1)+ \int_1^\infty \omega(s)\ dg(s) \\[3mm]\phantom{------} \leq \frac{c_{50}M_1(u)^{\frac{N+\alpha}{N-\alpha}}}{\frac{N+\alpha}{N-\alpha}+1}\int_1^\infty s^{-1-\frac{N+\alpha}{N-\alpha}} g(s)ds \\[3mm]\phantom{------} \displaystyle \le c_{50}g_\infty M_1(u)^{\frac{N+\alpha}{N-\alpha}}, \end{array}$$ where $c_{50}>0$ independent of $n$. We next deal with $ \int_{S^c_1}g(u)\rho^{\alpha}_{\partial\Omega}dx$. For $p_\in(1, \frac{N}{N-2\alpha+\beta})$, we have that $$\label{4.1} \displaystyle\begin{array}{lll} \int_{S^c_1}g(u)\rho^{\alpha}_{\partial\Omega}dx\le c_{4}\int_{S_1^c}u^{p_}\rho^{\alpha}_{\partial\Omega}dx+\epsilon\int_{S_1^c}\rho^{\alpha}_{\partial\Omega}dx \\[3mm]\phantom{------} \le c_{4}c_{51}\int_{\Omega}u^{p_}dx+c_{51}\epsilon \\[3mm]\phantom{------} \leq c_{4}c_{51}M_2(u)^{p_} +c_{51}\epsilon, \end{array}$$ where $c_{51}>0$ independent of $n$. Along with (\[06-08-10\]), (\[chenyuhang1\]), (\[06-08-11\]) and (\[4.1\]), we derive $$\label{05-09-4} M_1(\mathcal{T}_nu)\le c_{44}c_{50}g_\infty M_1(u)^{\frac{N+\alpha}{N-\alpha}}+c_{44}c_{4}c_{51}M_2(u)^{p_}+c_{44}c_{51}\epsilon+c_{44}k.$$ By [@NPV Theorem 6.5] and (\[power1\]), we derive that $$\displaystyle\begin{array}{lll} M_2(\mathcal{T}_nu)\le c_{45}{\\|g_n(u)+k \nu_n\\|}_{L^1 (\Omega,\rho_{\partial\Omega}^{\alpha}dx)}, \end{array}$$ which along with (\[chenyuhang1\]), (\[06-08-11\]) and (\[4.1\]), implies that $$\label{4.3} M_2(\mathcal{T}_nu)\le c_{45}c_{50}g_\infty M_1(u)^{\frac{N+\alpha}{N-\alpha}}+c_{45}c_{4}c_{51}M_2(u)^{p_}+c_{45}c_{51}\epsilon+c_{45}k.$$ Therefore, inequality (\[05-09-4\]) and (\[4.3\]) imply that $$M_1(\mathcal{T}_nu)+M_2(\mathcal{T}_nu)\le c_{52}g_\infty M_1(u)^{\frac{N+\alpha}{N-\alpha}}+c_{53}c_4M_2(u)^{p_}+c_{54}\epsilon+c_{54}k,$$ where $c_{52}=(c_{44}+c_{45})c_{50}$, $c_{21}=(c_{44}+c_{45})c_{51}$ and $c_{54}=c_{44}+c_{45}$. If we assume that $M_1(u)+M_2(u)\le \lambda$, implies $$M_1(\mathcal{T}_nu)+M_2(\mathcal{T}_nu)\le c_{52}g_\infty\lambda^{\frac{N+\alpha}{N-\alpha}}+c_{21}\lambda^{p_}+c_{21}\epsilon+c_{54}k.$$ Since $\frac{N+\alpha}{N-\alpha},\ p_>1$, then there exist $k_0>0$ and $\epsilon_0>0$ such that for any $k\in(0,k_0]$ and $\epsilon\in(0,\epsilon_0]$, the equation $$c_{52}g_\infty\lambda^{\frac{N+\alpha}{N-\alpha}}+c_{21}\lambda^{p_}+c_{21}c_3\epsilon+c_{54}k=\lambda$$ admits the largest root $\bar\lambda>0$. We redefine $M(u)=M_1(u)+M_2(u)$, then for $M(u)\le \bar\lambda$, we obtain that $$\label{2.2} M(\mathcal{T}_nu)\le c_{52}g_\infty\bar\lambda^{\frac{N+\alpha}{N-\alpha}}+c_{21}\bar\lambda^{p_}+c_{21}\epsilon+c_{54}k= \bar\lambda.$$ Especially, we have that $${\\|\mathcal{T}_nu\\|}_{L^1(\Omega)}\le c_8M_1(\mathcal{T}_nu)\|\Omega\|^{\frac{2\alpha}{N+\alpha}}\le c_{23} \bar\lambda\quad{\rm if}\quad M(u)\le \bar\lambda.$$ Thus, $\mathcal{T}_n$ maps $L^1(\Omega)$ into itself. Clearly, if $u_m\to u$ in $L^1(\Omega)$ as $m\to\infty$, then $g_n(u_m)\to g_n(u)$ in $L^1(\Omega)$ as $m\to\infty$, thus $\mathcal{T}_n$ is continuous. For any fixed $n\in{\mathbb{N}}$, $\mathcal{T}_nu_m=\mathbb{G}_\alpha\left[g_n(u_m)+k \nu_n\right]$ and $\{g_n(u_m)+k \nu_n\}_m$ is uniformly bounded in $L^1(\Omega,\rho^\alpha dx)$, then it follows by Proposition \[pr5\] that $\{\mathbb{G}_\alpha\left[g_n(u_m)+k \nu_n\right]\}_m$ is pre-compact in $L^1(\Omega)$, which implies that $\mathcal{T}_n$ is a compact operator. Let $$\displaystyle\begin{array}{lll}\displaystyle \mathcal{G}=\{u\in L^1_+(\Omega): \ M(u)\le \bar\lambda \} \end{array}$$ which is a closed and convex set of $L^1(\Omega)$. It infers by (\[2.2\]) that $$\mathcal{T}_n(\mathcal{G})\subset \mathcal{G}.$$ It follows by Schauder’s fixed point theorem that there exists some $u_n\in L^1_+(\Omega)$ such that $\mathcal{T}_nu_n=u_n$ and $M(u_n)\le \bar\lambda,$ where $\bar\lambda>0$ independent of $n$. In fact, $u_n$ is a classical solution of (\[2.3\]). Let $O$ an open set satisfying $ O\subset \bar O\subset \Omega$. By [@RS Proposition 2.3], for $\theta\in(0,2\alpha)$, there exists $c_{55}>0$ such that $${\\|u_n\\|}_{C^{\theta}(O)}\le c_{55}\{{\\|g(u_n)\\|}_{L^\infty(\Omega)}+kt_n^{-\alpha}{\\|\nu_n\\|}_{L^{\infty}(\Omega)}\},$$ then applied [@RS Corollary 2.4], $u_n$ is $C^{2\alpha+\epsilon_0}$ locally in $\Omega$ for some $\epsilon_0>0$. Then $u_n$ is a classical solution of (\[2.3\]). Moreover, $$\label{5.6} \int_\Omega u_n(-\Delta)^\alpha\xi dx=\int_\Omega g(u_n)\xi dx+k\int_\Omega\xi \nu_ndx,\quad \forall\xi\in \mathbb{X}_{\alpha}.$$ [Step 2. Convergence.* ]{} Since $\{g_n( u_n)\}$ and $\{\nu_n\}$ are uniformly bounded in $L^1(\Omega,\rho^\beta_{\partial\Omega} dx)$, then by Propostion \[pr5\], there exist a subsequence $\{u_{n_k}\}$ and $u$ such that $u_{n_k}\to u$ a.e. in $\Omega$ and in $L^1(\Omega)$, and $g_{n_k}(u_{n_k}) \to g( u)$ a.e. in $\Omega$. Finally we prove that $g_{n_k}( u_{n_k})\to g( u)$ in $L^1(\Omega,\rho^\beta_{\partial\Omega} dx)$. For $\lambda >0$, we set $S_\lambda=\{x\in\Omega:\|u_{n_k}(x)\|>\lambda\}$ and $\omega(\lambda)=\int_{S_\lambda}\rho^{\alpha}_{\partial\Omega}dx$, then for any Borel set $E\subset\Omega$, we have that $$\label{chenyuhang1000} \displaystyle\begin{array}{lll} \displaystyle\int_{E}\|g_{n_k}(u_{n_k})\|\rho^\beta_{\partial\Omega} dx=\int_{E\cap S^c_{\lambda}}g(u_{n_k})\rho^\beta_{\partial\Omega} dx+\int_{E\cap S_{\lambda}} g(u_{n_k})\rho^\beta_{\partial\Omega} dx\\[4mm]\phantom{\int_{E}\|g(u_{n_k})\|\rho^{\alpha_{\partial\Omega}}dx} \displaystyle\leq \tilde g(\lambda)\int_E\rho^\beta_{\partial\Omega} dx+\int_{S_{\lambda}} g(u_{n_k})\rho^\beta_{\partial\Omega} dx\\[4mm]\phantom{\int_{E}g(u_{n_k})\rho^\beta_{\partial\Omega} dx} \displaystyle\leq \tilde g(\lambda)\int_E\rho^\beta_{\partial\Omega} dx+\omega(\lambda) g(\lambda)+\int_{\lambda}^\infty \omega(s)d g(s), \end{array}$$ where $\tilde g(\lambda)=\max_{s\in[0,\lambda]}g(s)$. On the other hand, $$\int_{\lambda}^\infty g(s)d\omega(s)=\lim_{T_m\to\infty}\int_{\lambda}^{T_m} g(s)d\omega(s).$$ where $\{T_m\}$ is a sequence increasing number such that $T_m^{-\frac{N+\alpha}{N-\alpha}} g(T_m)\to 0$ as $m\to\infty$, which could obtained by assumption (\[1.4\]) and Lemma \[lm 08-09\] with $p={\frac{N+\alpha}{N-\alpha}}$. It infers by (\[2.4\]) that $$\displaystyle\begin{array}{lll} \displaystyle \omega(\lambda) g(\lambda)+ \int_{\lambda}^{T_m} \omega(s)d g(s) \le c_{50} g(\lambda)\lambda^{-\frac{N+\alpha}{N-\alpha}}+c_{56}\int_{\lambda}^{T_m} s^{-\frac{N+\alpha}{N-\alpha}}d g(s) \\[4mm]\phantom{-----\ \int_{\lambda}^{T_m} g(s)d\omega(s)}\displaystyle \leq c_{56}T_m^{-\frac{N+\alpha}{N-\alpha}}g(T_m)+\frac{c_{56}}{\frac{N+\alpha}{N-\alpha}+1}\int_{\lambda}^{T_m} s^{-1-\frac{N+\alpha}{N-\alpha}} g(s)ds, \end{array}$$ where $c_{56}=c_{50}\frac{N+\alpha}{N-\alpha}$. Pass the limit of $m\to\infty$, we have that $$\omega(\lambda) g(\lambda)+ \int_{\lambda}^\infty \omega(s)\ d g(s)\leq \frac{c_{56}}{\frac{N+\alpha}{N-\alpha}+1}\int_{\lambda}^\infty s^{-1-\frac{N+\alpha}{N-\alpha}} g(s)ds.$$ Notice that the above quantity on the right-hand side tends to $0$ when $\lambda\to\infty$. The conclusion follows: for any $\epsilon>0$ there exists $\lambda>0$ such that $$\frac{c_{56}}{\frac{N+\alpha}{N-\alpha}+1}\int_{\lambda}^\infty s^{-1-\frac{N+\alpha}{N-\alpha}} g(s)ds\leq \frac{\epsilon}{2}.$$ Since $\lambda$ is fixed, together with (\[chenyuhang1\]), there exists $\delta>0$ such that $$\int_E\rho^{\alpha}_{\partial\Omega}dx\leq \delta\Longrightarrow g(\lambda)\int_E\rho^{\alpha}_{\partial\Omega}dx\leq\frac{\epsilon}{2}.$$ This proves that $\{g\circ u_{n_k}\}$ is uniformly integrable in $L^1(\Omega,\rho^\beta_{\partial\Omega} dx)$. Then $g\circ u_{n_k}\to g\circ u$ in $L^1(\Omega,\rho^\beta_{\partial\Omega} dx)$ by Vitali convergence theorem. Pass the limit of (\[5.6\]) as $n_k\to \infty$ to derive that $$\int_\Omega u(-\Delta)^\alpha\xi=\int_\Omega g(u)\xi dx+k\int_\Omega\frac{\partial^\alpha\xi}{\partial \vec{n}^\alpha} d\nu,\quad \forall \xi\in\mathbb{X}_\alpha,$$ thus $u$ is a weak solution of (\[eq 1.1\]) and $u$ is nonnegative since $\{u_n\}$ are nonnegative. $\Box$ [99]{} Ph. Bénilan and H. Brezis, Nonlinear problems related to the Thomas-Fermi equation, [J. Evolution Eq. 3]{}, 673-770 (2003). Ph. Bénilan, H. Brezis and M. Crandall, A semilinear elliptic equation in $L^1({\mathbb{R}}^N )$, [Ann. Sc. Norm. Sup. Pisa Cl. Sci. 2]{}, 523-555 (1975). H. Brezis and X. Cabré, Some simple PDE’s without solutions, [Boll. Unione Mat. Italiana 8]{}, 223-262 (1998). P. Baras and M. Pierre, Critéres d’existence de solutions positives pour des équations semi-linéaires non monotones, [Ann. Inst. H. Poincaré, Analyse Non Linéaire 2]{}, 185-212 (1985). M. F. Bidaut-Véron and L. Vivier, An elliptic semilinear equation with source term involving boundary measures: the subcritical case, [Rev. Mat. Iberoamericana 16]{}, 477-513 (2000). M. F. Bidaut-Véron and C. Yarur, Semilinear elliptic equations and systems with measure data: existence and a priori estimates. [Advances in Differential Equations 7(3)]{}, 257-296 (2002). Z. Chen and R. Song, Estimates on Green functions and poisson kernels for symmetric stable process, [Math. Ann. 312]{}, 465-501 (1998). H. Brezis, Some variational problems of the Thomas-Fermi type. Variational inequalities and complementarity problems, [Proc. Internat. School, Erice, Wiley, Chichester]{}, 53-73 (1980). R. Cignoli and M. Cottlar, An Introduction to Functional Analysis, [North-Holland, Amsterdam]{}, 1974. H. Chen, P. Felmer and A. Quaas, Large solution to elliptic equations involving fractional Laplacian, accepted by [Ann. Inst. H. Poincaré, Analyse Non Linéaire]{}, (arXiv:1311.6044). H. Chen, P. Felmer and L. Véron, Elliptic equations involving general subcritical source nonlinearity and measures, arXiv:1409.3067. H. Chen and L. Véron, Semilinear fractional elliptic equations involving measures, [J. Differential equations 257(5)]{}, 1457-1486 (2014). H. Chen and L. Véron, Semilinear fractional elliptic equations with gradient nonlinearity involving measures, [J. Funct. Anal. 266(8)]{}, 5467-5492 (2014). H. Chen and L. Véron, Weakly and strongly singular solutions of semilinear fractional elliptic equations, [Asymptotic Analysis 88]{}, 165-184 (2014). H. Chen and J. Yang, Semilinear fractional elliptic equations with measures in unbounded domain, arXiv: 1403.1530 (2014). W. Chen, Y. Fang and R. Yang, Semilinear equations involving the fractional Laplacian on domains, arXiv:1309.7499 (2013). P. Felmer and A. Quaas, Fundamental solutions and Liouville type theorems for nonlinear integral operators, [Advances in Mathematics 226,]{} 2712-2738 (2011). E. Di Nezza, G. Palatucci and E. Valdinoci, Hitchhiker’s guide to the fractional Sobolev spaces, [Bull. Sci. Math. 136]{}, 521-573 (2012). A.Gmira and L. Véron, Boundary singularities of solutions of some nonlinear elliptic equations, [Duke Math. J. 64]{}, 271-324 (1991). N. J. Kalton and I. E. Verbitsky, Nonlinear equations and weighted nor inequalities, [Trans. A. M. S. 351]{}, 3341-3397 (1999). M. Marcus and A. C. Ponce, Reduced limits for nonlinear equations with measures, [J. Funct. Anal. 258]{}, 2316-2372 (2010). M. Marcus and L. Véron, The boundary trace of positive solutions of semilinear elliptic equations: the subcritical case, [Arch. Rat. Mech. Anal. 144]{}, 201-231 (1998). M. Marcus and L. Véron, The boundary trace of positive solutions of semilinear elliptic equations: the supercritical case, [J. Math. Pures Appl. 77]{}, 481-524 (1998). M. Marcus and L. Véron, Removable singularities and boundary traces, [J. Math. Pures Appl. 80]{}, 879-900 (2001). M. Marcus and L. Véron, The boundary trace and generalized B.V.P. for semilinear elliptic equations with coercive absorption, [Comm. Pure Appl. Math. 56]{}, 689-731 (2003). A. C. Ponce, Selected problems on elliptic equations involving measures, arXiv:1204.0668 (2012). X. Ros-Oton and J. Serra, The Dirichlet problem for the fractional laplacian: regularity up to the boundary, [J. Math. Pures Appl. 101(3)]{}, 275-302 (2014). J. Vazquez, On a semilinear equation in ${\mathbb{R}}^2$ involving bounded measures, [Proc. Roy. Soc. Edinburgh 95A]{}, 181-202 (1983). L. Véron, Elliptic equations involving Measures, Stationary Partial Differential equations, [Vol. I, 593-712, Handb. Differ. Equ., North-Holland, Amsterdam]{} (2004). Huyuan Chen Department of Mathematics, Jiangxi Normal University, Nanchang, Jiangxi 330022, PR China and Institute of Mathematical Sciences, New York University Shanghai, Shanghai 200120, PR China Hichem Hajaiej Institute of Mathematical Sciences, New York University Shanghai, Shanghai 200120, PR China [^1]: [email protected] [^2]: [email protected]	{ "pile_set_name": "ArXiv" }
--- abstract: 'I discuss the calculation of QCD jet rates in $e^+e^-$ annihilation as a testing ground for parton shower simulations and jet finding algorithms.' --- Cavendish–HEP–10/16 [QCD Jets and Parton Showers]{}[^1] Bryan R. Webber$^$\ University of Cambridge, Cavendish Laboratory,\ J.J. Thomson Avenue, Cambridge CB3 0HE, UK\ $^$e-mail: [email protected] Introduction ============ The production of jets of hadrons in all kinds of high-energy collisions is dramatic evidence of the pointlike substructure of matter. QCD predictions of the rates of production of different numbers of jets are well confirmed and provide good measurements of the fundamental coupling $\as$. The latest triumph in this respect is the calculation of the 5-jet rate in $e^+e^-$ annihilation to next-to-leading order, i.e. ${\cal O}(\as^4)$ [@Frederix:2010ne]. Figure \[fig:y45NLO\] shows that calculation compared to data from the ALEPH experiment at LEP [@Heister:2003aj]. The observable shown is $L_{45}\equiv -\ln(y_{45})$, where $y_{45}$ is the value of the jet resolution parameter at which five jets are just resolved using the $k_t$-jet algorithm[@Catani:1991hj]. There is good agreement over the range shown, and the uncertainty in the prediction is remarkably small considering this quantity is ${\cal O}(\as^3)$ at leading order. The value of the strong coupling obtained from the NLO fit to the region $L_{45}<6$ is (M\_Z) = 0.1156\^[+0.0041]{}\_[-0.0034]{}, which is in good agreement with the world average value obtained from other observables. However, looking at a wider range of $y_{45}$ values, fig. \[fig:y45\], we see that the region used in the NLO fit represents only a small part of the full distribution. Most events have $L_{45}>6$, with a distribution that turns over at $L_{45}\sim 8$, whereas the fixed-order prediction continues to rise more and more rapidly with increasing $L_{45}$ (note the logarithmic vertical scale in fig. \[fig:y45NLO\]). What this means physically is that most events have a two-jet structure that can only be resolved into five jets by using a high-resolution jet algorithm. However, it is important to understand this internal structure of the jets as well as possible, for example to search for highly-boosted new particles whose decays might look like or be hiding inside QCD jets. To achieve better understanding we need progress on two fronts: 1. Calculations of jet substructure in the region beyond the reach of fixed-order perturbation theory; 2. Jet algorithms that probe jets in a way that reveals their substructure in informative ways. Although the era of LEP physics is past, $\ee$ annihilation can still serve as a good testing ground for ideas on both these topics, as I hope to illustrate in the following sections. Parton showers ============== The reason for the breakdown of fixed-order predictions at high $L_{45}$, where most of the data lie, is that QCD matrix elements have soft and collinear singularities that give rise to logarithmic enhancement of higher-order contributions. In fact there are up to two factors of $L_{45}$ for every extra power of $\as$, so if the coefficient were unity we would expect a breakdown at $L_{45}\sim 1/\sqrt{\as}\sim 3$. As we shall see, in fact the coefficient is more like $2/3\pi$, which does indeed imply a breakdown at $L_{45}\sim 6$. Ideally we would like to be able to sum these enhanced terms to all orders in a closed form that would exhibit the turnover in the distribution, as is the case for several other $\ee$ observables. [ccl]{} $R_{2 }$ &=& $ 1 + a(R_{21}L + R_{22}L^2) + a^2(R_{23}L^3 + R_{24}L^4) + a^3(R_{25}L^5 + R_{26}L^6) + \ldots $\ $R_{21} $ &=& $ 3C_F/2 $\ $R_{22} $ &=& $ -C_F/2 $\ $R_{23} $ &=& $ -3C_F^2/4 - 11C_F C_A/36 + C_F N_f /18 $\ $R_{24} $ &=& $ C_F^2/8 $\ $R_{25} $ &=& $ 3C_F^3/16 + 11C_F^2 C_A/72 - C_F^2 N_f /36 $\ $R_{26} $ &=& $ -C_F^3/48 $\ \ $R_{3 }$ &=& $ a(R_{31}L + R_{32}L^2) + a^2(R_{33}L^3 + R_{34}L^4) + a^3(R_{35}L^5 + R_{36}L^6) + \ldots $\ $R_{31} $ &=& $ -3C_F/2 $\ $R_{32} $ &=& $ C_F/2 $\ $R_{33} $ &=& $ 3C_F^2/2+7C_FC_A/12-C_FN_f/12 $\ $R_{34} $ &=& $ -C_F^2/4 - C_F C_A/48 $\ $R_{35} $ &=& $ -9C_F^3 /16 - 137C_F^2 C_A/288 - 7C_A^2 C_F /160 + 5C_F^2 N_f /72 + C_F C_AN_f /160 $\ $R_{36} $ &=& $ C_F^3 /16 + C_F^2 C_A/96 + C_F C_A^2 /960 $\ \ $R_{4 }$ &=& $ a^2(R_{43}L^3 + R_{44}L^4) + a^3(R_{45}L^5 + R_{46}L^6) + \ldots $\ $R_{43} $ &=& $ -3C_F^2/4-5C_FC_A/18+C_FN_f/36 $\ $R_{44} $ &=& $ C_F^2/8 + C_F C_A/48 $\ $R_{45} $ &=& $ 9C_F^3 /16 + 71C_F^2 C_A/144 + 217C_F C_A^2 /2880 - 41C_F^2 N_f /720 - C_F C_AN_f /120 $\ $R_{46} $ &=& $ -C_F^3/16-C_F^2C_A/48-7C_FC_A^2/2880 $\ \ $R_{5 }$ &=& $ a^3(R_{55}L^5 + R_{56}L^6) + \ldots $\ $R_{55} $ &=& $ -3C_F^3 /16 - 49C_F^2 C_A/288 - 91C_F C_A^2 /2880 + 11C_F^2 N_f /720 + C_F C_AN_f /480 $\ $R_{56} $ &=& $ C_F^3 /48 + C_F^2 C_A/96 + C_F C_A^2 /720 $ In ref. [@Catani:1991hj] we wrote down integral equations for generating functions that can be used to compute the leading and next-to-leading logarithms (NLL) in jet cross sections to any order. Table \[tab:Rn\] shows the results up to ${\cal O}(\as^3)$. These equations are for the jet fraction $R_n(\yc)$, which is the fraction of events that have precisely $n$ jets at resolution $\yc$. The differential jet rates, like the one in figs. \[fig:y45NLO\] and \[fig:y45\], are obtained from them by differentiating: \[eq:ykk\] 1[\_[tot]{}]{} = -\_[n=k]{}\^.\|\_[=y\_[k-1,k]{}]{}. Thus to NLL accuracy, in the notation of table \[tab:Rn\], \[eq:y45NLL\] 1[\_[tot]{}]{} = (6R\_[56]{} L\_[45]{}\^5+5R\_[55]{} L\_[45]{}\^4) + [O]{}(\^4). However, such fixed-order NLL predictions are not much use as they are invalid when $L_{45}$ is not large and need to be resummed when it is large. Indeed, since $6R_{56}=197/270=0.73$ while (for $n_f=5$ flavours) $5R_{55}=-7.77$, the prediction (\[eq:y45NLL\]) is actually negative for $L_{45}<10$. The leading double-logarithmic ‘abelian’ terms, i.e. those proportional to $(a C_F L^2)^{n-2}$, resum to an exponential form: \[eq:RnAb\] R\_[n+2]{}\^[(ab)]{} \~1[n!]{}(12 aC\_FL\^2)\^n (-12 aC\_FL\^2) This gives the correct qualitative features of the differential distribution (\[eq:ykk\]) at large $L$, but the numerical values are wrong, e.g. the turn-over occurs at $L_{45}\sim 10$. This is not surprising in view of the comparable non-abelian terms and large NLL corrections. The easiest way to resum the enhanced terms more completely is to encode them in a parton shower simulation. By this I mean a sequential $1\to 2$ parton branching process with branching probabilities of the form \[eq:dP\] dP(abc)= q P\_[ba]{}(z)dz where $q$ is an ordered evolution variable, $z$ measures the energy fraction in the branching, $P_{ba}$ is the corresponding DGLAP splitting function and the argument $q'$ of $\as$ is a function of $q$ and $z$ in general. The integral equations of ref. [@Catani:1991hj] are equivalent to such a process with the following simple properties: the evolution variable is the angle of branching and $q'$ is the relative transverse momentum. The [@Corcella:2000bw] event generator results shown in fig. \[fig:y45\] are based on a parton shower with precisely these properties. [PYTHIA]{}[@Sjostrand:2006za] also has a parton shower which, although organized in a different way, ought to be equivalent. [ARIADNE]{}[@Lonnblad:1992tz] is based on a different approach involving colour dipoles rather than partons.[^2] All the generators correctly reproduce the main features of the distribution, in particular the turn-over at $L_{45}\sim 8$. It should be said that the event generators include a lot of additional refinements, such as matching to fixed-order matrix elements at low $L_{45}$ and modelling of hadronization. In particular the latter has quite a strong effect at LEP energies and introduces free parameters which can be tuned to the data. Nevertheless a parton shower, or equivalent, with the correct features is an essential component for reliable extrapolation to the higher energies and different processes encountered at the LHC. Angular ordering is not the most convenient organization of the parton shower: physical quantities such as transverse momenta and jet masses have to be reconstructed from the shower variables. It would also be preferable to generate the hardest (highest transverse momentum) branchings first, which would make matching to fixed-order matrix elements[@Catani:2001cc] and NLO improvements[@Frixione:2002ik; @Frixione:2007vw] simpler. These considerations lead us to look at what happens if we order the shower in relative transverse momentum ($p_t$) rather than angle. Unfortunately with simple $p_t$-ordering things start to go wrong even at the leading-log level as soon as gluon branching is involved. Instead of the results in table \[tab:Rn\] for the LL coefficients in the 4-jet and 5-jet fractions, R\_[44]{} &=& C\_F\^2/8 + C\_F C\_A/48,\ R\_[56]{} &=& C\_F\^3 /48 + C\_F\^2 C\_A/96 + C\_F C\_A\^2 /720, we get[^3] \[eq:Rpt\] R\^[(p\_t)]{}\_[44]{} &=& C\_F\^2/8 + C\_FC\_A/24,\ R\^[(p\_t)]{}\_[56]{} &=& C\_F\^3/48 + C\_F\^2C\_A/48 + 13C\_FC\_A\^2/2880. We could try to fix things up by ordering in $p_t$ and rejecting branchings that are disordered in angle. For the 4-jet rate this cures the problem with gluon branching, fig. \[fig:fourjets\](c), but spoils the result for sequential quark branching, fig. \[fig:fourjets\](b), while for the 5-jet fraction everything is wrong: R\^[(p\_t,)]{}\_[44]{} &=& 5C\_F\^2/48 + C\_FC\_A/48,\ R\^[(p\_t,)]{}\_[56]{} &=& 7C\_F\^3/576 + 13C\_F\^2C\_A/1440 + C\_FC\_A\^2/960. To see what is going wrong, consider the $(z_2,\theta_2)$ integration regions for diagrams \[fig:fourjets\](b) and (c), depicted in fig. \[fig:ztheta\]. Here $z_1,z_2$ and $\theta_1,\theta_2$ are the (smaller) gluon energy fractions and opening angles in successive branchings, and $\eps=\sqrt{\yc}$. Thus in diagram \[fig:fourjets\](b), $p_t$-ordering corresponds to $\eps<z_2\theta_2<z_1\theta_1$, giving the integration region A+B. However, the correct region is the angular-ordered one A+C. If we impose angular ordering after $p_t$-ordering, we get only A, i.e. a deficit in the coefficient of $C_F^2$. Now it happens that for this diagram the inclusion of region B compensates for the loss of C as far as the logarithms are concerned, so in this case $p_t$-ordering alone gives the same result as angular ordering. I will come back to this point later. In diagram \[fig:fourjets\](c), $p_t$-ordering corresponds to $\eps<z_1z_2\theta_2<z_1\theta_1$, i.e. $\eps/z_1<z_2\theta_2<\theta_1$, as shown on the right in fig. \[fig:ztheta\]. The region C has disappeared and the $p_t$-ordered region A+B is just too large, giving an enhanced coefficient of $C_FC_A$. However, because region C is not there, imposing angular ordering after $p_t$-ordering is equivalent to simply angular ordering, giving the correct region A and hence the correct coefficient of $C_FC_A$. So perhaps the correct prescription for a $p_t$-ordered shower is to angular-order only the $g\to gg$ vertices? This corrects the 4-jet rate but in the 5-jet rate the coefficient of $C_FC_A^2$ is too small: R\^[(p\_t,gg)]{}\_[44]{} &=& C\_F\^2/8 + C\_F C\_A/48,\ R\^[(p\_t,gg)]{}\_[56]{} &=& C\_F\^3/48 + C\_F\^2C\_A/96 + C\_FC\_A\^2/960. However, the reason for this is the same as before: if the gluon that branches a second time in fig. \[fig:fivejets\](f) is the harder one coming from the first gluon branching, the situation is as on the left in fig. \[fig:ztheta\], and we should not angular-order the second gluon branching. In summary, the way to get the correct LL (and NLL) jet fractions, to all orders, from a $p_t$-ordered parton shower is to enforce angular ordering with respect to the branching at which each parton was “created”, where this means the branching at which it was the softer of the two produced [@Catani:2001cc]. More precisely, one should veto branchings that are disordered in angle with respect to their “creation”. Technically, a veto means not branching but resetting the $p_t$ scale as if the branching had occurred. This is a common kind of procedure in parton shower generators anyway, for example to correct for flavour thresholds or higher orders in the running coupling. This looks like a better way to do parton shower event generation. With $p_t$-ordering one can more easily correct the prediction to NLO, or indeed to any fixed order in $\as$ in principle. One only has to correct the first few steps in the shower. Unfortunately there is a catch. Everything works fine at the parton level as far as the distribution in phase space is concerned, but the colour structure of the partonic final state is not correct. Coming back to fig. \[fig:ztheta\] (left), we see that, compared to angular ordering, $p_t$-ordering includes a region of softer, wide-angle gluon emission, B, in place of a region of harder, more collinear emission, C. What this means is that gluon radiation is moved around within the shower, the amount and distribution remaining the same. This is depicted schematically in fig. \[fig:colstru\], where for simplicity we show the large-$N_c$ approximation, as used for hadronization in event generators. In fig. \[fig:colstru\](a), angular ordering assigns a soft, wide-angle gluon, actually emitted coherently by partons $b$ and $c$, to the parent parton $a$, which is reasonable because $a$ does have the coherent sum of the colour charges of $b$ and $c$. In contrast, $p_t$-ordering assigns this gluon to the harder of $b$ and $c$, in this case $c$, as in fig. \[fig:colstru\](b). That is reasonable as far as the momenta are concerned, but it spoils the colour structure by treating $c$ as the colour source and neglecting the coherent contribution of $b$. The colour structure matters when one wants to interface the parton shower to a non-perturbative hadronization model. In the cluster model used by , colour-singlet clusters are formed by splitting gluons at the end of the shower into $q\bar q$ pairs. Thus in the angular-ordered fig. \[fig:colstru\](a) the clusters connect $(gb)$ and $(bc)$, while in $p_t$-ordered fig. \[fig:colstru\](b) they connect $(bg)$ and $(gc)$. Similarly in the [PYTHIA]{} string hadronization model, the string connects $a-g-b-c$ in fig. \[fig:colstru\](a) but $a-b-g-c$ in fig. \[fig:colstru\](b). In conclusion, an angular-ordered parton shower sums the LL and NLL enhanced terms and provides partonic final states with colour structure consistent with QCD coherence. This is good for hadronization models but not so convenient for reconstruction of kinematics or for systematic improvement away from the soft and collinear regions. A $p_t$-ordered shower is better in those respects and, with the right angular veto procedure, can give the correct NLL jet fractions. However the colour structure then needs to be reconfigured according to angular ordering before the partonic final state can be hadronized. Jet algorithms ============== Recall that the $k_t$-algorithm for $\ee$ annihilation [@Catani:1991hj] is defined in terms of the resolution variable y\_[ij]{} = 2{E\_i\^2,E\_j\^2}(1-\_[ij]{})/Q\^2, where $E_{i,j}$ are the energies of final-state objects $i$ and $j$, $\theta_{ij}$ is the angle between their momenta and $Q$ is the centre-of-mass energy. The two objects with the smallest value of $y_{ij}$ are combined into one, this is repeated until all $y_{ij}>\yc$, and the remaining objects are called jets. For the purpose of counting large logarithms of $\yc$, we can write this in the small-angle approximation \_[ij]{} = {E\_i,E\_j}\_[ij]{}/Q >, where as before $\eps=\sqrt{\yc}$. As pointed out in ref. [@Cacciari:2008gp], this is just one of a continuum of possible jet algorithms with resolution variable \_[ij]{} = {E\^p\_i,E\^p\_j}\_[ij]{}/Q\^p, where $p$ can be any positive or negative number. In particular $p=-1$ defines the resolution for the $\ee$ analogue of the anti-$k_t$ algorithm[@Cacciari:2008gp], which has the advantage that objects are combined starting with those that have the highest energy rather than the lowest. When $p<0$ a supplementary condition is needed, otherwise infinitely soft emissions would be resolved. For anti-$k_t$ we define \_[ij]{} &=&{Q/E\_i,Q/E\_j}\_[ij]{},\ \_i &=& Q/E\_i. Then if the smallest of the set of $\{\eps_{ij},\eps_i\}$ is an $\eps_i$, we remove $i$ from the list of objects to be recombined, and if $\eps_i<1$ we call it a jet. Otherwise we just throw it away. Thus every jet has an energy greater than $\eps Q$ and is separated from other jets by an angle greater than $\eps$. The resulting LL coefficients in the 4- and 5-jet fractions are R\^[anti]{}\_[44]{} &=& C\_F\^2/2 + C\_F C\_A/8,\ R\^[anti]{}\_[56]{} &=& C\_F\^3/6 + C\_F\^2C\_A/8 + C\_FC\_A\^2/48, where as before the large logarithm is defined as $L=-2\ln\eps$. We could introduce an angular resolution $\delta$ different from the energy resolution $\eps$ by multiplying $\eps_{ij}$ by $\eps/\delta$. This would just replace $\ln^2\eps$ by $\ln\eps\ln\delta$. It is easy to see that leading double-logarithmic abelian terms in the anti-$k_t$ jet rates resum to an exponential form with twice the exponent of the $k_t$ rates (\[eq:RnAb\]): \[eq:RnAbanti\] R\_[n+2]{}\^[(anti, ab)]{} \~1[n!]{}(aC\_FL\^2)\^n (-aC\_FL\^2) It should also be possible to resum the non-abelian and NLL terms using techniques like those of ref. [@Catani:1991hj]. Conclusions =========== Although the era of high-energy $\ee$ collider experiments is past, at least for a while, it is helpful to study how our tools for analysing hadronic final states perform in the cleaner environment of the annihilation process. The $k_t$-jet algorithm has proven useful in all kinds of processes and the $\ee$ jet rates defined in this way are a good place to test alternative resummation methods, particular those involving parton showers ordered in different ways. We have seen that angular-ordered and $p_t$-ordered showers can both be arranged to resum the leading and next-to-leading logarithms of the $k_t$-jet resolution $\yc$. The $p_t$-ordering option is good for matching to fixed-order calculations but causes some difficulties in matching to hadronization models at low scales, owing to its disordered colour structure. The rather different anti-$k_t$ algorithm has been adopted as the preferred tool for jet finding at the LHC. An analogous $\ee$ algorithm can be defined and we saw that it has a simple pattern of leading logarithms, which should be amenable to resummation using techniques similar to those applied to the $k_t$ algorithm. Acknowledgments {#acknowledgments .unnumbered} =============== It is a pleasure to recall and acknowledge conversations with Volodya Gribov in many places during the all-too-brief times we spent together. I am also indebted to Stefano Catani, Gavin Salam and Mike Seymour for helpful comments and discussions. [10]{} R. Frederix, S. Frixione, K. Melnikov and G. Zanderighi, arXiv:1008.5313 (2010). A. Heister [et al.]{}, [Eur. Phys. J.]{} [C35]{}, 457 (2004). S. Catani, Y. L. Dokshitzer, M. Olsson, G. Turnock and B. R. Webber, [Phys. Lett.]{} [B269]{}, 432 (1991). G. Corcella [et al.]{}, [JHEP]{} [01]{}, 010 (2001). T. Sjostrand, S. Mrenna and P. Z. Skands, [JHEP]{} [05]{}, 026 (2006). L. Lonnblad, [Comput. Phys. Commun.]{} [71]{}, 15 (1992). S. Catani, F. Krauss, R. Kuhn and B. R. Webber, [JHEP]{} [11]{}, 063 (2001). S. Frixione and B. R. Webber, [JHEP]{} [06]{}, 029 (2002). S. Frixione, P. Nason and C. Oleari, [JHEP]{} [11]{}, 070 (2007). M. Cacciari, G. P. Salam and G. Soyez, [JHEP]{} [04]{}, 063 (2008). [^1]: Contribution to Proceedings of Gribov-80 Memorial Workshop on Quantum Chromodynamics and Beyond, ICTP, Trieste, Italy, 26-28 May, 2010. [^2]: I should emphasise that the discussion in this paper concerning alternative evolution variables and the colour structure of the shower refer only to parton showers as defined by eq. (\[eq:dP\]) and not to dipole showers. [^3]: Thanks to Mike Seymour for pointing out an error in my original calculation of the coefficient of $C_FC_A^2$.	{ "pile_set_name": "ArXiv" }
--- abstract: 'Pedagogical introduction into the problem of the mathematical description of the quantum correlation (entanglement) of composite quantum systems is represented. The notion is substantiated about the fact that the conventional algorithm of the reduction of von Neumann in the description of the dynamics of the observed subsystem is not universal and corresponds only to the case of maximum macroscopicity of the unobservable subsystem. Is clearly shown the sense of the algorithm of the correlated reduction proposed, which minimally changes the entropy of composite system.' --- N. K. Solovarov\ Zavoisky Physical-Technical Institute of Russian Academy of Sciences, Kazan, 420029 Russia\ Email: [email protected]\ PACS numbers: 03.65.Ta Introduction. Theoretical difficulties in the description of quantum correlation. ================================================================================= The property of quantum correlation (or entanglement) of composite quantum systems became last decade one of the most discussed questions of quantum physics. The number of reviews and articles, in which is used the idea about the quantum correlation, continuously and almost exponentially grows according to the Gisin’s estimation [@gisin] since 1995. The reasons for this avalanche-type interest several, and even classification reasons appears ambiguous. Here we first attempt ourselves to trace how idea about the quantum correlation is expressed mathematically and, to what experimental operations correspond the accepted mathematical models . Primary attention is paid to the difficulties and the contradictions in the description and the interpretation of the experimental manifestations of quantum correlation. In the introductory part of the report we refer mainly to the last reviews, where it is possible to find the comprehensive bibliography. The study of quantum correlation is conducted at present from different positions. Historically idea about the quantum correlation appeared with the examination of physical nature of quantum non-locality [@gisin]. Another approach is connected with the description of the decogerence phenomenon of quantum systems due to the measurement and it is called frequently the problem of quantum measurements [@dass; @zeh]. The third, most “published” (according to the number of articles) approach, is connected with the development of quantum information theory and the examination of the possibility of using the quantum correlation for transfer and processing of information [@batle; @plenio]. In spite of the use of different terminology and interpretation, in all enumerated approaches it is easy to trace the united mathematical and physical content. Our first purpose, to show mathematical generality and physical content of differences and difficulties, which are in the different approaches. Let us begin from the history, i.e., let us trace, what debating point, and what logical sequence of ideas and experiments led to the present situation. In the work [@gisin] this logical sequence of ideas is erected on the way of examining the axiomatically adopted (being seemed a priori obvious) property of the locality of nature - interaction occurs three-dimensional locally, the transfer of interaction between the three-dimensional different points is accomplished by certain material agent, the velocity of propagation of which is limited. Newton clearly formulated the being contained in its theory of universal gravitation (but by them categorically rejected) property of the non-locality (stone moved on the Moon instantly changes the weight of any object on the Earth). Newton’s non-locality mathematically is expressed in the absence in the law of universal gravitation of dependence on the time. To this, emphasized by Newton internal contradiction in the mathematical description of nature (mathematical model of gravity) the majority of physicists did not turn attention, since its experimental manifestations it was not observed. Non-locality (in the theory) was present up to 1915, when Einstein formulated general theory of relativity. Einstein by general theory of relativity returned to physics locality [@gisin]. Quantum mechanics was borne in 10 years and again in the description of nature non-locality returned. Einstein in principle did not agree with the statistical interpretation of quantum mechanics, expressing the sense of disagreement by the known phrase: > God do not plays dice! Paraphrasing the statement of Einstein Gisin [@gisin] formulates the fundamental results of the development of quantum physics of last decade by the assertion: > God does play dice, he even plays with nonlocal dice! Where in the quantum theory is contained quantum non-locality (quantum correlation, entanglement) and how it is expressed mathematically? The unavoidable preamble of answer to this question is reminding the axioms of quantum mechanics and interpretation accepted, (see, for example, the recent account of this preamble in the thesis [@batle]). Let us limit this introduction, after isolating only those positions, around which now goes the discussion about the quantum correlation. With the description of the quantum correlation: 1. One of the postulates of quantum theory is used - reduction postulate, i.e. the action of measurement consist in the collapse of wave function. 2. The tasks of quantum dynamics of composite quantum systems are examined. Let the state of quantum system be represented by the wave function $\vert\Psi_{l}\rangle=\sum_{\alpha}(c_{l\alpha}\vert\alpha\rangle)$, $\sum_{\alpha}\vert c_{l\alpha}\vert^{2} =1$, where $\vert\alpha\rangle$ is the complete orthonormal set of the eigenstates (basis) of the considered system, or by the density operator $\hat\rho=\sum_{l} p_{l}\vert\Psi_l\rangle\langle\Psi_l\vert$, $\sum_{l}p_l=1$ , where the pure states $\vert\Psi_{l}\rangle$ are not compulsorily orthogonal. The set of eigenstates of the Hamiltonian of physical system is selected in many instances as the basis. Let that observable $\hat A$ has eigenvalues $a$ with the appropriate eigenvectors $\vert ad\rangle$, where the additional index $d$ noted possible degeneracy $a=ad$. Then projector to the subspace of that observed $\hat A$ with the eigenvalue $a$ is an operator $\hat P_{a}=\sum_{ad=a}\vert ad\rangle\langle ad\vert$. Mathematically the reduction postulate is expressed by the following conversion of clean wave function as a result of macroscopic measurement of observable $a$: $$\label{eq1} \vert\Psi_{l}\rangle\to\vert ad\rangle=\frac{\hat P_{ad}\vert\Psi_{l}\rangle}{\sqrt{\langle\Psi_{l}\vert\hat P_{ad}\vert\Psi_{l}\rangle}}.$$ Normalizing coefficient in the radicand of denominator is equal to the probability of the realization of this event. We will note that in the right side of (1) the equality is carried out with an accuracy to the phase factor, which is considered usually unessential. If degeneration is absent, then this conversion is considered describing the ideal quantum von Neumann’s measurement. The conversion for the density operator corresponding to reduction is accepted to write in the form [@batle] $$\label{eq2} \hat\rho\to\hat\rho_{ad }=\frac{\hat P_{ad}\hat\rho\hat P_{ad}^{\dag}} {\textrm{Sp}\left(\hat P_{ad}\hat\rho\hat P_{ad}^{\dag}\right)},$$ with the assumptions of i) orthogonality $\textrm{Sp}\left(\hat P_{ad}\hat P_{ag}^{\dag}\right)=\delta_{dg}$ and ii) closure $\sum_{ad}\hat P_{ad}\hat P_{ad}^{\dag}=\hat I$, where $\hat I$ is the unit operator and subscript $\textrm{N}$ here and subsequently will be used for the designation of that reduction according to the axiomatic von Neumann’s algorithm. Normalizing factor in denominator of (2) it is possible to copy in the different form, since the trace does not depend on the non-diagonal matrix elements of operator. Conversions (1),(2) relate to the arbitrary closed quantum system, on which is conducted the measurement by external macroscopic measuring device. It is necessary to emphasize that very start of measurement is passage to the open system, but all the examination is conducted in the Hilbert space of the closed system. The property of quantum correlation is determined for the composite quantum systems. Generally speaking, in quantum physics the composite systems are always examined, since the reduction postulate, being the inherent part of the quantum theory, implies existence of two physically divided parts: the described quantum system and measuring (macroscopic) device. However, usually, keeping in mind composite quantum system, it is assumed that they are determined (i.e. experimentally they can be independently isolated and studied) several quantum subsystems $A,B,E,\ldots$(in the quantum informatics corresponding terms are accepted - Alice, Bob, Eve-Eavesdropper, Charlie...). The certainty of such subsystems means that are considered known their Hamiltonians $\hat H_{A},\hat H_{B},\hat H_{E},\ldots$, which correspond eigenstates $\vert\alpha\rangle,\vert\beta\rangle,\vert\varepsilon\rangle,\ldots$ (frequently selected as the bases), the operators of those observable $\hat A,\hat B,\hat E,\ldots$, their eigenvalues $a,b,e,\ldots$ and eigenstates $\vert a\rangle,\vert b\rangle,\vert e\rangle,\ldots$. We for the simplification not will here introduce special indices for the designation of possible degeneracy. Furthermore, are considered known the Hamiltonians of pairwise interactions between the subsystems $\hat H_{AB},\hat H_{AE},\hat H_{BE},\ldots$, moreover interactions are relied by the relatively weak $\left(\hat H_{AB},\hat H_{AB},\ldots\right)<\left(\hat H_{A},\hat H_{B},\ldots\right)$ (there are in the form the relative values of the eigenvalues of energy). The obvious terminology is accepted: bipartite system $\left(\hat H_{A+B}\right)$, tripartite system $\left(\hat H_{A+B+E}\right)$,…multipartite system $\left(\hat H_{A+B+E+\ldots}\right)$. The majorities of the fundamental properties of quantum correlation can be examined based on the example of the bipartite quantum system $\left(\hat H_{A+B}\right)$. The relative smallness of interactions between the subsystems makes possible to use the first mathematical assumption - to describe the state of composite system in the Hilbert space of dimensionality $N_{A}\times N_{B}$, which is been the direct product space of the space of the independent subsystems , where $N_{A}, N_{B}$ are the dimensionality (number of eigenstates) of subsystems respectively. I.e. the set of pared multiplications of eigenvectors of subsystems $\vert\alpha\rangle\times\vert\beta\rangle$ is selected as basis. Let the state of system in this basis be described by the density operator of $\hat\rho_{A+B}$ ($\hat\rho_{A+B}\in\hat H_{A}\otimes\hat H_{B}$). There is one additional taciturn adopted (implicit) limitation to many composite quantum systems, for which is determined the property of quantum correlation. It is assumed that there is a physical possibility to conduct local actions or measurements on the subsystems [@terhal]. Term “local” here does not bear in the general case of the content “three-dimensional localization”, although in many experimental cases this precisely thus. Mathematically this possibility is described by assumption about the validity of existence of the following (local) maps of the states of the quantum system: $$\label{eq3} \hat\rho'_{A+B}=\left(\hat U_{A}\otimes\hat U_{B}\right) \hat\rho_{A+B}\left(\hat U_{A}\otimes\hat U_{B}\right)^{\dagger},$$ where $\hat U_{A},\hat U_{B}$ are unitary operators, determined in the spaces of the subsystems, by which is described the action (most frequently external) on the subsystems. The equality $\hat U_{A(B)}=\hat I_{A(B)}$, where $\hat U_{A}$ are unit operators in the space of the corresponding subsystem, corresponds to the absence of local action on the subsystem. Existence of quantum correlation (entanglement) between the subsystems of such composite system is accepted to mathematically determine through the opposite property of the inseparability of quantum system, i.e., the impossibility to represent the density operator in the form of the convex linear combination of the direct (tensor) products of the pure density operators of subsystems [@werner]: $$\label{eq4} \hat\rho_{(A+B)s}=\sum_{l} p_{l}\left(\hat\rho_{Al}\otimes\hat\rho_{Bl}\right),$$ where $0<p_{l}\le 1$, $\sum_{l} p_{l}=1$, $\left(\hat\rho_{Al}\right)^2=\hat\rho_{Al}$, $\left(\hat\rho_{Bl}\right)^2=\hat\rho_{Bl}$, $\hat\rho_{Al}$, $\hat\rho_{Bl}$ are the density operators of subsystems. With the validity of equality (4) the system is considered separable that marked here by additional subscript $s$. In the separable system quantum correlation (entanglement, quantum non-locality) is absent. In this case statistical weights $p_{l}$ it is accepted to call the hidden parameters of quantum system. The states of the composite quantum system of form (4) can be obtained by local operations (3) from the originally separated subsystems. Answer to the question: is the state (the density operator) of composite quantum system separable, or entangled? - it is accepted to call the separability or disentanglement problem. At present the mathematical disentanglement problem seems unsolvable in the general case for two reasons [@batle]. From one side there is an infinite set of expansions of the separable density operator of form (4), since the number of clean and not compulsorily orthonormal states of subsystems is unconfined by anything. Physically to this the infinite set of the ways of creating this separable state with the aid of the local operations corresponds. From other side there is an intriguing mathematical property: the linear combination of the entangled states can be the separable state. But at the same time the linear combination of the separable states is always the separable state. In what the physical sense of quantum correlation and why must be known, are entangled subsystems or not? The traditional setting of physical experiments consists of the observation of the dynamics of physical subsystems as a result of their interaction. In many cases the state of composite system, described by the density operator, is obtained as a result of solving (most frequently the approximate) dynamic Schroedinger (Neumann) equation as the result of certain interaction between originally separated subsystems. The mixed character of this state (i.e. its description by the density matrix, but not by wave function) can appear, only if there are some uncontrollable degrees of freedom out of the subsystems in question. If state is described by the entangled density matrix, then between the subsystems there is (or there existed and it left its track) certain coherent interaction. But if the density matrix is separable, then was unknown interaction between the subsystems coherent, or not, and did remain there what or tracks of the mutual coherence of subsystems. Here term coherence is identical to term “quantum coherence”, i.e., the importance of the phase relationships between the subsystems in the process of interaction and up to the moment of measurement. Attempts at the mathematical solution of the disentanglement (separability) problem compose separate direction in the quantum informatics (in the linear algebra to it the unresolved problem of the characterization of positive maps corresponds) [@batle],[@terhal],[@plenio]. All approaches use as the criterion of separability disturbance of one of the physical limitations to the characteristics of the density operator. Let us group the mathematical criteria of the separability (entanglement) proposed in their physical content. 1. Positive partial transpose (PPT) criterion. This is necessary, but insufficient criterion of separability. In order to clearly present its content let us write down the density operator $\hat\rho_{A+B}$ in the selected basis: $$\label{eq5} \hat\rho_{(A+B)}=\sum_{\alpha,\alpha'}\sum_{\beta,\beta'} \langle\alpha\beta\vert\hat\rho_{(A+B)}\vert\alpha'\beta'\rangle \vert\alpha\rangle_{A}\langle\alpha'\vert\otimes \vert\beta\rangle_{B}\langle\beta'\vert.$$ Then the density operator partially transposed on the subsystem $A$ is determined by the expression: $$\label{eq6} \left(\hat\rho_{(A+B)}\right)_{T(A)}=\sum_{\alpha,\alpha'} \sum_{\beta,\beta'} \langle\alpha\beta\vert\hat\rho_{(A+B)}\vert\alpha'\beta'\rangle \vert\alpha'\rangle_{A}\langle\alpha\vert\otimes \vert\beta\rangle_{B}\langle\beta'\vert.$$ System is separable, if this operator $\left(\hat\rho_{(A+B)}\right)_{T(A)}$ has only positive eigenvalues. (It is equivalent for the transposition on the subsystem $B$: $\left([\hat\rho_{(A+B)}\right)_{T(B)}$, moreover $\left(\left(\hat\rho_{(A+B)}\right)_{T(A)}\right)_{T(B)}= \left(\left(\hat\rho_{(A+B)}\right)_{T(B)}\right)_{T(A)}= \left(\hat\rho_{(A+B)}\right)_{T}$). The condition of positive partial transposition makes simple physical sense. The separable system satisfies condition (4), which shows that the dynamics of subsystems occurs it independently and, therefore, satisfies the condition of local unitarity (3), i.e., reversibility in the time. The operation of transposition on the subsystem corresponds to time reversal in this subsystem. Consequently, the condition of positivity corresponds to the absence of nonphysical (negative) probabilities with the time reversal in one of the subsystems. 2. Reduction criterion. In the mathematical algorithm the reduction criterion is close to PPT criterion, since the conversion on one of the subsystems there also is done. It is possible to formulate this criterion as follows: system is in the separable state, if inequalities simultaneously are fulfilled [@batle]: $$\label{eq7} \frac{1}{N_{A}}\hat I_{A}\otimes\hat\rho_{B\textrm{N}}- \hat\rho_{(A+B)}\ge 0,\,\, \hat\rho_{A\textrm{N}}\otimes\frac{1}{N_{B}}\hat I_{B}- \hat\rho_{(A+B)}\ge 0,$$ where $\hat\rho_{A\textrm{N}}=\textrm{Sp}_{B}\hat\rho_{(A+B)}$, $\hat\rho_{B\textrm{N}}=\textrm{Sp}_{A}\hat\rho_{(A+B)}$ are reduced density operators of subsystems. Inequalities for the eigenvalues of operators here are implied, i.e. is assumed passage to the basis, in which the operators are diagonal.\ The enumerated criteria carry to the operational, i.e., to such, for which the algorithm of calculation and relationship for some values, which determine the conditions of separability, is indicated. The physical sense of last criterion is different from that indicated in the first point. It is possible to connect it with distinctions in kind in the entropy of the entangled and separeble composite quantum systems. Therefore let us transfer the first known properties of the entropy of composite quantum systems. Very determination of entropy is in this case debating point with the ambiguous and sometimes contradictory terminology.\ The quantity of information is evaluated , which we know about the quantum system globally (i.e. $\hat\rho_{(A+B)}$) and locally (most frequently there is in the form $\hat\rho_{A\textrm{N}}$, $\hat\rho_{B\textrm{N}}$). First remind of the von Neumann’s determination of the entropy of the quantum system, described by the density operator $\hat\rho$: $$\label{eq8} S\left(\hat\rho\right)=- \textrm{Sp}\left(\hat\rho\ln\hat\rho\right).$$ It is assumed that with the calculation of this value they pass to the basis, in which the density matrix is diagonal. Entropy describes the deflection of quantum system from the pure state. Its property: - Entropy is equal to zero (it is minimum!) for the pure state. Entropy is maximum and equal to $\ln N$ for a maximally mixed state, when $\hat\rho=(1/N)\hat I$. In the first case is known maximally possible, and in the second - minimally possible information about the quantum system [@fano]. Let us note the opposition of terminology from the accepted in the quantum informatics definition of the characteristics of the states of quantum system from the possibility of the content in them of minimum ($S(\hat\rho)=0$) and maximum ($S(\hat\rho)=\max$) quantity of information [@gorb]. Entropy is invariant with the unitary conversions of the basis: $S(\hat\rho)=S\left(\hat U\hat\rho\hat U^{\dagger}\right)$.\ - The entropy of the separable quantum systems is additive, i.e.\ $S\left(\hat\rho_{A}\otimes\hat\rho_{B}\right)= S(\hat\rho_{A})+S(\hat\rho_{B})$. However, for the composite entangled quantum system is characteristic the property of sub-additivity, determined by the inequality: $\left\vert S(\hat\rho_{A\textrm{N}})-S(\hat\rho_{B\textrm{N}})\right\vert\le S(\hat\rho_{A+B})\le S(\hat\rho_{A\textrm{N}})+S(\hat\rho_{B\textrm{N}})$. According to Shannon’s theory, the entropy of composite system never can be less than the entropy of any of its parts. For the entangled system with the local reduced (according to von Neumann’s algorithm) density operators - this is incorrect. It is considered that this property of entropy can serve as the indicator of the entanglement of state. However, there is a separate direction in quantum informatics, which argues the inapplicability of the classical determination of information to the composite quantum systems and is proposed a number of the alternative determinations of entropy for the composite quantum systems for the purpose to return by it the property of additivity [@batle]. It is considered that existence of entanglement experimentally is manifested in the inequality $\langle\hat A\hat B\rangle\ne\langle\hat A\rangle_{\textrm{N}}\langle\hat B\rangle_{\textrm{N}}$. To the left stands the experimentally specific average value of that nonlocal observable, specific by expression $\langle\hat A\hat B\rangle=\textrm{Sp}\hat\rho_{A+B}\left(\hat A\otimes\hat B\right)$. The product of the calculated local average values, determined by equations $\langle\hat A\rangle_{\textrm{N}}=\textrm{Sp}\hat\rho_{A+B}\hat A'$, $\langle\hat B\rangle_{\textrm{N}}=\textrm{Sp}\hat\rho_{A+B}\hat B'$ to the right stands. The upper prime noted operators, extended to the complete space: $\hat A'=\hat A\otimes\hat I_{B}$, $\hat P'_{B}(\beta)=\hat I_{A}\otimes\hat P_{B}(\beta)$. It is accepted to name the specially selected nonlocal observable $\hat W$ for the concrete systems the witness of entanglement. It is proven that the composite bipartite system is entangled if and only if there is witness of entanglement - Hermitian operator $\hat W\,(\hat W=\hat W^{\dagger})$, for whom are valid the inequalities: $\textrm{Sp}\hat\rho_{A+B}\hat W\le 0$ while $\textrm{Sp}\hat\rho_{(A+B)s}\hat W\ge 0$ for all separable states $\hat\rho_{(A+B)s}$. This is the sufficiently abstract non-operational criterion of quantum correlation, to which it is difficult to compare literal physical sense, however precisely this form of the inequality (them proposed much for the different experimental situations) is taken asd the experimental test of the quantum correlation of subsystems.\ Let us emphasize that it is always assumed that with the variety of approaches to the description of quantum correlation the local operators of the density of subsystems (and, correspondingly, the local dynamics of subsystems) are determined by the reduction algorithm of von Neumann ($\hat\rho_{A\textrm{N}}(t)=\textrm{Sp}_{B}\hat\rho_{A+B}(t)$, $\hat\rho_{B\textrm{N}}(t)=\textrm{Sp}_{A}\hat\rho_{A+B}(t)$). Our work is directed toward the assertion of idea about the approximate nature of von Neumann’s algorithm of the determination of the local density operators, its injustice in the general case and the needs for the calculation of the mutual correlation of subsystems with the description of their local dynamics.\ The idea of examination is based on what non-universality of von Neumann’s algorithm is already repeatedly demonstrated in different physical tasks. Apparently, historically as this first example can serve the construction of the quantum theory of relaxation [@fano],[@blum]. There one of the interacting subsystems (let $B$) from the physical considerations is relied stationary, that possesses by the properties of quasi-classical thermostat. Its state is approximately described by the Boltzmann’s density operator with the specific temperature $\textrm{T}$: $\hat\rho_{B}(t)\approx\hat\rho_{B}(\textrm{T})= \exp\left(-\left(1/k_{\textrm{B}}T\right)\hat H_{B}\right)$. For the definition of the state of subsystem $A$ the approximate disentanglement was postulated in the form of the relationship: $\hat\rho_{A+B}(t)\approx\hat\rho^{d}_{A+B(\textrm{T})}(t)= \hat\rho_{A}(t)\otimes\hat\rho_{B}(\textrm{T})$ , known as the “first assumption of the quantum theory of relaxation” or the “basic condition of irreversibility” [@fano],[@blum],[@prigozh] (here and further by superscript $d$ we note the approximate disentanglement).\ Another approximate algorithm of disentanglement, widely utilized in the quantum informatics, is based on the positions of the quantum theory of measurements [@preskill],[@zurek],[@men],[@koshino], when both subsystems are relied by quantum. It was initially postulated that the indirect determination of the state of subsystem $A$ is accomplished by means of the individual ideal projection quantum measurement on quasi-independent subsystem-pointer $B$ [@zurek],[@koshino]. As a result of measurement of the observable $\hat B$ by external macroscopic gauge the subsystem $B$ (according to the projection postulate of quantum mechanics) occurs in one of the eigenstates $\vert\beta\rangle$ of that observable. The density operator of subsystem-pointer after measurement is immediately considered equal: $\hat\rho_{B(\beta)}(t)=\hat P_{B}(\beta)=\vert\beta\rangle\langle\beta\vert$. The density operator of subsystem $A$ in this case is defined as the result of the quantum averaging of the density operator $\hat\rho_{A+B}(t)$ over the subsystem $B$ [@koshino]: $$\label{eq9} \hat\rho_{A(\beta)}(t)=\frac{\textrm{Sp}_{B} \left(\hat\rho_{A+B}(t)\hat P'_{B}(\beta)\right)} {\textrm{Sp}_{AB}\left(\hat\rho_{A+B}(t)\hat P'_{B}(\beta)\right)}.$$ Disentangled state: $$\label{eq10} \hat\rho_{A+B}(t)\to\hat\rho^{d}_{A+B(\beta)}= \hat\rho_{A(\beta)}(t)\otimes\hat P_{B}(\beta),$$ is considered as the initial state (relative to the moment of measurement) in the description of the subsequent dynamics of composite quantum system, which leads to the quantum Zeno effect [@koshino]. Such type indirect projection measurements (Zeno-like measurements) on the quasi-independent subsystem widely are discussed as one of the possible mechanisms of control of the state of subsystems in the quantum information theory.\ In the recent works [@vedral],[@grisha] was shown the importance of the calculation of the entanglement of quantum subsystem with the external (with respect to the considered composite system) gauge in the description of the result of quantum measurement. Is noted, that by the consequence of this calculation can be the incomplete loss of quantum coherence subsystem, i.e., a difference in its local density operator after measurement from the linear combination of projectors.\ In the enumerated non-Neumann’s algorithms of the approximate disentanglement the desired state of the observed subsystem $A$ is determined in the form of functional from the known density operator of the of composite system $\hat\rho_{A+B}(t)$ and density operator of subsystem $B$, postulated or obtained from the additional physical considerations. With the use of a traditional Neumann’s algorithm mathematically equivalent situation occurs - the state of composite system $\hat\rho_{A+B}(t)$ is known, and is postulated the algorithm of the calculation of the reduced density operator of subsystem $A$, $\hat\rho_{A\textrm{N}}(t)$. The inverse problem is obvious: to what state of subsystem $B$, $\hat\rho_{B?}(t)$ does correspond the Neumann’s determination of the density operator of the “observable” subsystem $A$? Or, to what disentanglement $\hat\rho_{A+B}(t)\to\hat\rho_{A\textrm{N}}\otimes\hat\rho_{B?}(t)$ it does correspond? Its examination was the starting point of ours work.\ Physical content of the operation of the reduction-disentanglement of von Neumann. ================================================================================== The necessary step of the approximate disentanglement is answer to the question: should be represented the density operator of composite system in the form of the tensor product of the local density operators of subsystems (tensor product structure [@zanardi]), or in the form (4) of the linear combination of such products? In the quantum information theory usually the first idea is postulated [@preskill]. However, in the recent work [@zanardi] the arguments were formulated, which show that precisely this selection is dictated by the experimental determination of observables and interactions between the subsystems. The idea of work [@zanardi] lies in the fact that the determination of composite system includes the possibility of conducting of local operations and measurements. Such composite system from an experimental point of view appears as two quasi-independent correlated subsystems, to what its mathematical idea in the form of the tensor product of the local density operators of subsystems corresponds. Assuming the positions of work [@zanardi], we will represent the entangled density operator in the form of the tensor product of some correlated density operators of subsystems: $$\label{eq11} \hat\rho_{A+B}(t)\approx\hat\rho^{d}_{A+B)}= \hat\rho_{Ac}(t)\otimes\hat\rho_{Bc}(t),$$ where subscript $c$ distinguishes the local correlated density operators from corresponding local reduced density operators.\ We transform (11) by the method, analogous to the method, used by von Neumann for the proof of the mutual correlation of the average values of observables (see chapter 6.2 [@von]). Let us multiply (11) to the right on $\hat\rho'_{Bc}(t)$ or $\hat\rho'_{Ac}(t)$ and let us take partial track on the subsystem $B$ or $A$. Using the orthonormality of the correlated density operators of subsystems, we will obtain the system of two connected equations [@solyear],[@soljetpl],[@solalg]: $$\label{eq12} \hat\rho_{Ac}(t)\approx\frac{\textrm{Sp}_{B}\hat\rho_{A+B}(t) \hat\rho'_{Bc}(t)} {\textrm{Sp}_{AB}\hat\rho_{A+B}(t) \hat\rho'_{Bc}(t)},\quad \hat\rho_{Bc}(t)\approx\frac{\textrm{Sp}_{A}\hat\rho_{A+B}(t) \hat\rho'_{Ac}(t)} {\textrm{Sp}_{AB}\hat\rho_{A+B}(t) \hat\rho'_{Ac}(t)}.$$ Right sides of (12) are the normalized quantum averaging of the density operator of the closed system over one of subsystems. Each equation defines the density operator of one subsystem as functional from the density operator of the closed system and density operator of another subsystem. Actually, this simply the more convenient record of expression (11). Such relationships are not unique. Repeating the procedure of right multiplication it is possible to obtain the set of the expressions of the form: $$\label{eq13} \hat\rho_{Ac}(t)\approx\frac{\textrm{Sp}_{B}\hat\rho_{A+B}(t) \left[\hat\rho'_{Bc}(t)\right]^{m}} {\textrm{Sp}_{AB}\hat\rho_{A+B}(t) \left[\hat\rho'_{Bc}(t)\right]^{m}},\quad \hat\rho_{Bc}(t)\approx\frac{\textrm{Sp}_{A}\hat\rho_{A+B}(t) \left[\hat\rho'_{Ac}(t)\right]^{m}} {\textrm{Sp}_{AB}\hat\rho_{A+B}(t) \left[\hat\rho'_{Ac}(t)\right]^{m}},$$ where $m$ - arbitrary integer. It is obvious that essential it is possible to consider only expressions with $m\le\left(N^{2}_{A(B)}-1\right)$ for the linearly independent degrees of the correlated density operators of subsystems.\ Now let us show, to what disentanglement does implicitly correspond von Neumann’s reduction? Let us examine the first of expressions (13), which determine the state of the “observed” subsystem $A$. Assume that its correlated density operator is determined by von Neumann’s reduction $\hat\rho_{Ac}(t)=\hat\rho_{A\textrm{N}}(t)$. Then precise equalities for all expressions (13) are carried out in two cases:\ 1. Or the density operator of subsystem $B$ corresponds to the pure state, when $\left[\hat\rho_{Bc}(t)\right]^{m}=\hat\rho_{Bc}(t)$ (that corresponds physically to the limiting case of the noninteracting subsystems).\ 2. Or it is proportional to the unit operator $\hat\rho_{Bc}(t)=\hat\rho_{B\max}(t)=(1/N_{B})\hat I_{B}$ (that corresponds to the steady state of subsystem $B$ with the maximum entropy, the minimum information about subsystem or the infinite temperature according to [@fano]). Since the trivial case of the noninteracting subsystems does not correspond to initial assumption about the entanglement of composite quantum system, remains the second version. Thus, the use of von Neumann’s algorithm of the reduction for determining the local density operator of the observed subsystem is physically equivalent to the approximate disentanglement, with which the unobservable subsystem is relied by being stationarily been in the state with the maximum entropy or infinite temperature, in which it with the equal probability is found in any from the eigenstates. In this case quantum coherence (non-diagonal matrix elements of density matrix) in the subsystem $B$ is equal to zero: $$\label{eq14} \hat\rho_{A+B}(t)\to\hat\rho^{d}_{A+B}(t)= \hat\rho_{A\textrm{N}}(t)\otimes\hat\rho_{B\max}.$$ To the same conclusion it is possible to come by another way, using positions of the quantum measurements theory [@zurek],[@men],[@koshino]. Let the subsystem $B$ be the quantum pointer, with the aid of which is accomplished the indirect measurement of the state of subsystem $A$. What must be the results of many projective measurements (1),(2) on the equally prepared system $\hat\rho_{A+B}(t)$, so that the result would be described by the density operator of $\hat\rho_{A\textrm{N}}(t)$?\ Let $p(\beta)$ is the probability of observing the value $\beta$, which is appeared the eigenvalue of that observable $\hat B$, with the individual ideal projection quantum measurements on the quasi-independent subsystem $B$ [@koshino]. With each measurement the density operator of subsystem $A$ is determined by the appropriate expression of form (2). Consequently, takeing into account the statistical nature of $\beta$ measurements, the density operator of subsystem $A$ it is necessary to define as the result of statistical averaging over the results of many individual projection measurements of $\hat B$: $$\label{eq15} \hat\rho_{AP(B)}(t)=\frac{\sum_{\beta}p(\beta)\textrm{Sp}_{B} \left(\hat\rho_{A+B}\hat P'_{B}(\beta)\right)} {\textrm{Sp}_{A}\sum_{\beta}p(\beta)\textrm{Sp}_{B} \left(\hat\rho_{A+B}\hat P'_{B}(\beta)\right)}.$$ Using permutability of the operations of summation over $\beta$ and the takings of the track on the subsystem $B$, let us write down the condition of the equality of this averaged density operator of subsystem $A$ to the operator $\hat\rho_{A\textrm{N}}(t)$: $$\label{eq16} \frac{\textrm{Sp}_{B}\left(\hat\rho_{A+B}(t)\left(\hat I_{A}\otimes\sum_{\beta}p(\beta)\hat P_{B}(\beta)\right) \right)} {\textrm{Sp}_{AB}\left(\hat\rho_{A+B}(t)\left(\hat I_{A}\otimes\sum_{\beta}p(\beta)\hat P_{B}(\beta)\right) \right)}=\textrm{Sp}\hat\rho_{A+B}(t).$$ Equality (16) is correct in the general case of arbitrary state only if $p(\beta)=\left(1/N_{\beta}\right)$. Thus, the description of the dynamics of the observed subsystem by the von Neumann’s reduced density operator corresponds in the quantum measurements theory to the case, when quantum subsystem-pointer with the projection measurements with the equal probability is revealed in any of the eigenstates. The use of a projection postulate in the quantum measurements theory determines the total loss of quantum coherence by subsystem-pointer with each individual measurement [@koshino],[@dass]. Comparing (16) and (9) we conclude that the state of subsystem-pointer in this case is physically equivalent to its presence in the state with the infinite temperature. By other words the use of von Neumann reduction with the disentanglement corresponds precisely to the “first assumption of the quantum relaxation theory” in the extreme case, when the temperature of thermostat is relied infinite [@fano],[@blum].\ If we use an idea about the correlation of subsystems (12)-(13), then von Neumann’s reduction [@von], the “first assumption of the quantum theory of relaxation” [@fano],[@blum] and definition (9) [@koshino] can be examined as special cases of approaching the assigned state of the “unobservable” subsystem in the disentanglement problem. In each case in the explicit or implicit form the state of one subsystem is postulated, and the state (density operator) of another subsystem correlated with it is determined by expressions (13). It is possible to conclude that in the conventional algorithm of the calculation of the average values of those observed by one of the subsystems the measuring projection postulate of quantum mechanics is implicitly used twice. For the first time with the determination of the density operator of subsystem from the known state of composite system (reduction-disentanglement of von Neumann), and for the second time with the calculation of the average values of the observables for the subsystem with the counted already known density operator. Our further consideration is based on the idea, that the first step (disentanglement) does not identify with the macroscopic projection measurement, and it must be based on the basis of the conditions of each specific objectives.\ Correlated disentanglement. =========================== The algorithms of the approximate disentanglement examined include in the explicit or implicit form assumption about state of one of the subsystems. By each concrete selection of the density operator of the “unobservable” subsystem is simulated the specific physical process of measurement and opposite effect of measurement on the state of subsystems. This idea for the projection measurements is contained in the quantum theory of measurements [@zurek],[@men],[@koshino] in the form of the mathematical expressions of form (9), (15). For the clarity let us show differences in the calculated dynamics of the observed subsystem, caused by the selection of the operator of the density of the unobservable subsystem, based on the example to model in the quantum informats of the composite system of two qubits [@preskill],[@dass]. Let us designate the eigenstates of the independent subsystems-qubits $A$ and $B$, $\vert 2\rangle_{A},\vert 1\rangle_{A}$ and $\vert 2\rangle_{B},\vert 1\rangle_{B}$, respectively. The operators of composite system are represented by the matrices of the fourth order, whose each element is designated by two pairs of the subscripts $(\alpha\beta,\alpha'\beta')$ (see, for example, [@blum] Appendix A). The first index of each pair $(\alpha,\alpha')$ corresponds to the eigenstate of subsystem $A$, and the second $\beta,\beta'$ - subsystem $B$. In the general case the unitary dynamics of the closed system in question is described by the density matrix of form [@blum]: $$\label{eq17} \left(\hat\rho_{A+B}(t)\right)=\left( \begin{array}{cccc} \rho_{22,22} & \rho_{22,21} & \rho_{22,12} & \rho_{22,11}\\ \rho_{21,22} & \rho_{21,21} & \rho_{21,12} & \rho_{21,11}\\ \rho_{12,22} & \rho_{12,21} & \rho_{12,12} & \rho_{12,11}\\ \rho_{11,22} & \rho_{11,21} & \rho_{11,12} & \rho_{11,11}\\ \end{array} \right),$$ where dependence on the time is omitted for simplicity in the expressions of matrix elements.\ According to (15) the dynamics of qubit $A$, determined from the results of many projective measurements of qubit $B$ with the measured probabilities $p_{B2},p_{B1}$, to reveal it in the states $\vert 2\rangle_{B},\vert 1\rangle_{B}$, correspondingly, is described by the density operator: $$\label{18} \hat\rho_{AP(B)}=\frac{p_{B2}\textrm{Sp}_{B} \left(\hat\rho_{A+B}\left(\hat I_{A}\otimes\hat P_{B2}\right)\right)+p_{B1}\textrm{Sp}_{B} \left(\hat\rho_{A+B}\left(\hat I_{A}\otimes\hat P_{B1}\right)\right)} {\textrm{Sp}_{A}\left( p_{B2}\textrm{Sp}_{B} \left(\hat\rho_{A+B}\left(\hat I_{A}\otimes\hat P_{B2}\right)\right)+p_{B1}\textrm{Sp}_{B} \left(\hat\rho_{A+B}\left(\hat I_{A}\otimes\hat P_{B1}\right)\right)\right)}.$$ Let $p_{B2}=p,\, p_{B1}=1-p$, i.e., the density matrix of the subsystem of qubit-pointer be relied by stationary and equal: $$\label{eq19} \left(\hat\rho_{B(P)}\right)= \left( \begin{array}{cc} p & 0\\ 0 & 1-p \end{array} \right).$$ Then the local, correlated with it density matrix, which describes the dynamics of the observed subsystem $A$, is equal: $$\begin{gathered} \label{eq20} \left(\hat\rho_{AP(B)}\right)= \left(\begin{array}{cc} p\rho_{22,22}+(1-p)\rho_{21,21} & p\rho_{22,12}+(1-p)\rho_{21,11}\\ p\rho_{12,22}+(1-p)\rho_{11,21} & p\rho_{12,12}+(1-p)\rho_{11,11} \end{array}\right)\times\\ \times\left[p\rho_{22,22}+(1-p)\rho_{21,21}+ p\rho_{12,12}+(1-p)\rho_{11,11}\right]^{-1}.\end{gathered}$$ This result immediately is obtained, if we substitute (19) into the first of (12). Thus, the approximate disentanglement is realized in this case by the following replacement: $$\begin{gathered} \label{eq21} \left(\hat\rho_{A+B}\right)\to\left(\hat\rho^{d}_{A+B}\right)= \left(\hat\rho_{AP(B)}\right)\otimes\left(\hat\rho_{B(P)}\right)= \\ \left(\begin{array}{cc} p^2\rho_{22,22}+p(1-p)\rho_{21,21} & 0 \\ 0 & p(1-p)\rho_{22,22}+(1-p)^{2}\rho_{21,21}\\ p^2\rho_{12,22}+p(1-p)\rho_{11,21} & 0 \\ 0 & p(1-p)\rho_{12,22}+(1-p)^{2}\rho_{21,21} \end{array}\right.\\ \left.\begin{array}{cc} p^2\rho_{22,12}+p(1-p)\rho_{21,11} & 0\\ 0 & p(1-p)\rho_{22,12}+(1-p)^{2}\rho_{21,11}\\ p^2\rho_{12,12}+p(1-p)\rho_{11,11} & 0\\ 0 & p(1-p)\rho_{12,12}+(1-p)^{2}\rho_{11,11} \end{array} \right)\times\\ \left[p\rho_{22,22}+(1-p)\rho_{21,21}+p\rho_{12,12}+(1-p)\rho_{11,11} \right]^{-1}\end{gathered}$$ This nontraditional presentation of the operation of reduction, recorded in the “disentangled” form of the direct product of the local density operators of subsystems, makes it possible to clearly present the sense of the made approximations, comparing matrix elements (21) and (17). The first difference - the non-diagonal on the indices of subsystem-pointer matrix elements of the density matrix of composite system are assumed equal to zero. By this step is mathematically reflected the postulated loss by the subsystem-pointer of quantum coherence (decoherence) with the external macroscopic projection measurement on the subsystem [@zurek],[@dass]. Let us note that in this case the quantum coherence in the subsystem $A$ does not disappear, i.e., complete dekogerence of both the composite system and the observed subsystem does not occur. The second difference - the diagonal on the indices of subsystem-pointer elements of complete density matrix are substituted with their linear combinations with the weights, determined according to the results of local projection measurements on the subsystem-pointer $B$. With the equal probability of detecting qubit $B$ in the eigenstates ($p=1/2$) the approximate disentanglement corresponds to the von Neumann’s reduction. With $p\ne 1/2$ joint use of both mathematical approximations is equivalent to physical assumption about the validity of the description of the state of the unobservable subsystem by the specific temperature $\textrm{T}$. It is evident that with $p\ne 1/2$ the matrix elements of the local density operator (20) quantitatively are differed from that obtained by traditional calculation and, correspondingly, the dynamics of the observed subsystem can differ from usually supposed.\ Qualitative difference in the dynamics of the observed subsystem $A$ will arise, if one assumes that the quantum coherence remains with the disentanglement in the subsystem $B$. To this assumption corresponds idea about that which the physical separation of composite system into the subsystems does not identify with conducting of projection macroscopic measurement. Either the physical separation of subsystems is accomplished before conducting of local projection measurement on qubit $B$, or the nondestructive quantum measurement is conducted . To the retention of quantum coherence in the subsystem $B$ mathematically corresponds adding to the nondiagonal elements of the local density matrix $\hat\rho_{B}$ of values $\vert b\vert\ne 0$ instead of their equality to zero, postulated in (19): $$\label{eq22} \left(\hat\rho_{Bc}\right)=\left(\begin{array}{cc} p & b\\ b^{} & 1-p \end{array}\right),\quad \left(\hat\rho'_{Bc}\right)=\left(\hat I_{A}\right)\otimes \left(\hat\rho_{Bc}\right)=\left(\begin{array}{cccc} p & b & 0 & 0\\ b^{} & 1-p & 0 & 0\\ 0 & 0 & p & b\\ 0 & 0 & b^{} & 1-p \end{array}\right).$$ Using first of (12) we will obtain correlated to (22) the local density matrix of the subsystem $A$: $$\begin{gathered} \label{eq23} \left(\hat\rho_{Ac}\right)=\\ =\left(\begin{array}{cc} \begin{array}{c} p\rho_{22,22}+b^{}\rho_{22,21}+\\ +b\rho_{21,22}+(1-p)\rho_{21,21}\end{array} & \begin{array}{c} p\rho_{22,12}+b^{}\rho_{22,11}+\\ +b\rho_{21,12}+(1-p)\rho_{21,11}\end{array}\\ \begin{array}{c} p\rho_{12,22}+b^{}\rho_{12,21}+\\ +b\rho_{11,22}+(1-p)\rho_{11,21}\end{array} & \begin{array}{c} p\rho_{12,12}+b^{}\rho_{12,11}+\\ +b\rho_{11,12}+(1-p)\rho_{11,11}\end{array} \end{array}\right)\times\\ \times\left[p\left(\rho_{22,22}+\rho_{12,12}\right)+ (1-p)(\left(\rho_{21,21}+\rho_{11,11}\right)+\right.\\ +\left.b\left(\rho_{21,22}+\rho_{11,12}\right)+ b^{}\left(\rho_{22,21}+\rho_{12,11}\right)\right]^{-1}\end{gathered}$$ From comparison (23) with (20) it is evident that the dynamics of populations (diagonal elements $\left(\hat\rho_{Ac}\right)$) and quantum coherence of qubit $A$, depends in this case not only on probabilities to reveal qubit $B$ in the eigenstates, but also from its quantum coherence. The analogous (21) expression of the approximately disentangled density matrix of composite system in this case is equal to: $$\label{eq24} \begin{array}{c} \left(\hat\rho_{A+B}\right)\to\left(\hat\rho^{d}_{(A+B)c}\right)= \left(\hat\rho_{A(Bc)}\right)\otimes\left(\hat\rho_{Bc)}\right)= \\ \left(\begin{array}{cc} \begin{array}{c} p^2\rho_{22,22}+p(1-p)\rho_{21,21}+\\ pb^{}\rho_{22,21}+ pb\rho_{21,22} \end{array} & \begin{array}{c} pb\rho_{22,22}+(1-p)b\rho_{21,21}+\\ \vert b\vert^{2}\rho_{22,21}+ b^{2}\rho_{21,22} \end{array} \\ \begin{array}{c} pb^{}\rho_{22,22}+(1-p)b^{}\rho_{21,21}+\\ \left(b^{}\right)^{2}\rho_{22,21}+ \vert b\vert^{2}\rho_{21,22} \end{array} & \begin{array}{c} p(1-p)\rho_{22,22}+(1-p)^{2}\rho_{21,21}+\\ (1-p)b^{}\rho_{22,21}+ (1-p)b\rho_{21,22} \end{array} \\ \begin{array}{c} p^2\rho_{12,22}+p(1-p)\rho_{11,21}+\\ pb^{}\rho_{12,21}+pb\rho_{11,22} \end{array} & \begin{array}{c} pb\rho_{12,22}+(1-p)b\rho_{11,21}+\\ \vert b\vert^{2}\rho_{12,21}+b^{2}\rho_{11,22} \end{array} \\ \begin{array}{c} pb^{}\rho_{12,22}+(1-p)b^{}\rho_{11,21}+\\ \left(b^{}\right)^{2}\rho_{12,21}+\vert b\vert^{2}\rho_{11,22} \end{array} & \begin{array}{c} p(1-p)\rho_{12,22}+(1-p)^{2}\rho_{11,21}+\\ (1-p)b{}\rho_{12,21}+(1-p)b\rho_{11,22} \end{array} \end{array}\right. \\ \left.\begin{array}{cc} \begin{array}{c} p^2\rho_{22,12}+p(1-p)\rho_{21,11}+\\ pb^{}\rho_{22,11}+ pb\rho_{21,12} \end{array} & \begin{array}{c} pb\rho_{22,12}+(1-p)b\rho_{21,11}+\\ \vert b\vert^{2}\rho_{22,11}+ b^{2}\rho_{21,12} \end{array}\\ \begin{array}{c} pb^{}\rho_{22,12}+(1-p)b^{}\rho_{21,11}+\\ \left(b^{}\right)^{2}\rho_{22,11}+ \vert b\vert^{2}\rho_{21,12} \end{array} & \begin{array}{c} p(1-p)\rho_{22,12}+(1-p)^{2}\rho_{21,11}+\\ (1-p)b^{}\rho_{22,11}+ (1-p)b\rho_{21,12} \end{array}\\ \begin{array}{c} p^2\rho_{12,12}+p(1-p)\rho_{11,11}+\\ pb^{}\rho_{12,11}+pb\rho_{11,12} \end{array} & \begin{array}{c} pb\rho_{12,12}+(1-p)b\rho_{11,11}+\\ \vert b\vert^{2}\rho_{12,11}+b^{2}\rho_{11,22} \end{array}\\ \begin{array}{c} pb^{}\rho_{12,12}+(1-p)b^{}\rho_{11,11}+\\ \left(b^{}\right)^{2}\rho_{12,11}+\vert b\vert^{2}\rho_{11,12} \end{array} & \begin{array}{c} p(1-p)\rho_{12,12}+(1-p)^{2}\rho_{11,11}+\\ (1-p)b{}\rho_{12,11}+(1-p)b\rho_{11,12} \end{array} \end{array} \right)\times \\ \left[p\left(\rho_{22,22}+\rho_{12,12}\right)+ (1-p)\left(\rho_{21,21}+\rho_{11,11}\right)+\right. \\ \left.b\left(\rho_{21,22}+\rho_{11,12}\right)+ b^{}\left(\rho_{22,21}+\rho_{12,11}\right)\right]^{-1}. \end{array}$$ From comparison (21),(24) and respectively (20),(23) it is evident that to idea about the retention of quantum coherence in the subsystem-pointer qualitatively corresponds the partial retention of the quantum coherence of composite system. In the general case all nondiagonal elements of density matrix (24) are not equal to zero. Simultaneously it is evident from this simplest example that the use of approximation of the assigned state of the subsystem-pointer (i.e. the selection of the model of measurement, including as a special case, the conventional algorithm of the reduction of von Neumann) leads always to redefining (change) of all matrix elements of the composite density matrix. Thus always with the realization of the mathematical operation of the approximate disentanglement occurs a change in the entropy or information about the composite quantum system [@vedral] and partial decogerence [@dass].\ Different dynamics of the observed subsystem is the consequence of different selection of the algorithm of disentanglement (model of measurement). The question about the criterion of the selection of the disentanglement algorithm arises and its correspondence to the experimental procedure of measurement accepted. Physically limiting cases are the von Neumann’s disentanglement, which corresponds to complete decogerence of the subsystem-pointer and the “nondestructive” disentanglement, to which corresponds the invariability (in the limit) of the density operator of composite system. If the approximation of the closed system and nondestructive quantum measurement is correct, it is justified to search for the algorithm of separation, which minimizes change of $\hat\rho_{A+B}$ and simultaneously the change of entropy. Such local mutually correlated density operators of subsystems it is possible to obtain by solving system of equations (12) by the method of sequential approximations [@solalg],[@soljetpl]: $$\begin{gathered} \label{eq25} \hat\rho_{Ac}=\lim_{n\to\infty}\hat\rho^{(n+1)}_{A}= \lim_{n\to\infty}\frac{\textrm{Sp}_{B}\hat\rho_{A+B}\hat\rho'^{(n)}_{B}} {\textrm{Sp}_{AB}\hat\rho_{A+B}\hat\rho'^{(n)}_{B}},\\ \hat\rho_{Bc}=\lim_{n\to\infty}\hat\rho^{(n+1)}_{B}= \lim_{n\to\infty}\frac{\textrm{Sp}_{A}\hat\rho_{A+B}\hat\rho'^{(n)}_{A}} {\textrm{Sp}_{AB}\hat\rho_{A+B}\hat\rho'^{(n)}_{B}},\end{gathered}$$ where, for example, the reduced density operators $\hat\rho_{A\textrm{N}},\hat\rho_{B\textrm{N}}$ it is possible to use as zero approximation (n=0). One should emphasize that this correlated reduction corresponds to idea about the equivalence of the subsystems of the closed quantum system. [100]{} N. Gisin. Can relativity be complete? From Newtonian non-locality to quantum non-locality and beyond. E-print: quant-ph/0512168.\ T. Dass. Measurement and Decoherence. E-print:quant-ph/0505070.\ H. D. Zeh. Roots and Fruits of Decoherence. Séminare Poincare, 2, 1-19 (2005); E-print: quant-ph/0512078.\ J. Batle-Vallespir. Characterization of Quantum Entangled States and Information Measures. PhD Thesis. Department de Fýsica. Universitat de les Balears, 2006; E-print: quant-ph/0603124.\ M. B. Plenio, Sh. Virmani. An introduction to entanglement measures. E-print quant-ph/0504163.\ B. M. Terhal. Detecting Quantum Entanglement. E-print: quant-ph/0101032.\ R. F. Werner. Quantum states with Einstein-Podolsky-Rosen correlations admitting a hidden-variable model. Phys. Rev. A40, 4277 (1989).\ U. Fano. Description of States in Quantum Mechanics by Density Matrix and Operator Techniques. Rev. Mod. Phys. 29, No. 1, 74-93 (1957). V. N. Gorbachev, A. I. Zhiliba. Physical bases of the contemporary information processes. Peterburgski institut pechati. Sankt-Peterburg, 2004 (in russian). K. Blum. Density Matrix. Theory and Applications, Plenum Press, NY and London, 1981. (K. Blum. Teoriya matritsy plotnosti i eyo prilozheniya, Mir, Moskwa, 1983). I. Prigogine. The End of Certainty The Free Press, NY-London-Toronto-Sidney-Singapore, 1997 (I. Prigozhin. Konets opredelennosti, Redaktsiya zhurnala “Regulyarnaya i chaoticheskaya dinamika”, Izhevsk, 1999). J. Preskill, Lecture notes for Physics 219: Quantum Computation, California Institute of Technology, Pasadena CA, 1998. URL: http://www.theory.caltech.edu/people/preskill/ph219/ W. H. Zurek . Decoherence, einselection, and the quantum origins of the classical. Rev. Mod. Phys. 75, 000715 (2003). M. B. Mensky. Quantum Measurements and Decoherence. Models and Phenomenology. Fundamental Theories of Physics. Vol 110. Kluwer Academic Publishers. 2000 (M. B. Mensky. Kvantovye izmereniya i dekogerentsiya, Fizmatlit, Moskwa, 2001). K. Koshino and A. Shimizu. Quantum Zeno effect by general measurements. E-print: quant-ph/0411145. V. Vedral. Classical Correlations and Entanglement in Quantum Measurements. Phys. Rev. Lett. 90, 050401 (2003). B. A. Grishanin and V. N. Zadkov. Entangling quantum measurements and their properties. Phys. Rev. A 68, 022309 (2003); Generalized quantum measurements. Part I: Information properties of soft quantum measurements. E-print: quant-ph/0506045; Generalized quantum measurements. Part II: Partially-destructive quantum measurements in finite-dimensional Hilbert spaces. E-print: quant-ph/0506046. P. Zanardi, D. A. Lidar, S. Lloyd. Quantum Tensor Product Structures are Observable Induced. Phys. Rev. Lett. 92, 060402 (2004). J. von Neumann, Mathematical Foundations of Quantum Mechanics. Princeton University Press, Princeton, 1955. N. K. Solovarov. Correlated Dynamics of Subsystems of the Closed Quantum System. In The 2001 Yearbook of Zavoisky Physical-Technical Institute, Kazan Scientific Center, Russian Academy of Sciences, Ed. by K.M. Salikhov (Fiztekhpress, Kazan, 2002, in russian) p. 104.\ N. K. Solovarov. Quantum Beats of Atomic Inversion in a Resonant Coherent Field. JETP Lett. 78, No. 5, 267 (2003).\ N. K. Solovarov. Algorithm of Reduction. E-print quant-ph/0304142*.\	{ "pile_set_name": "ArXiv" }
--- abstract: 'Numerical results for the transport coefficients of quark gluon plasma are calculated by lattice simulation of SU(3) pure gauge model. The bulk viscosity is consistent with zero. The shear viscosity is finite and increases with temperature $T$ roughly as $T^{3}$, and around the finite temperature transition points, it is slightly smaller than the typical hadron masses.' address: - 'Research Center for Information Science and Education, Hiroshima\' - 'Faculty of Education, Yamamagata University,Yamagata\' author: - 'A. Nakamura,, T.Saito, and S.Sakai$^{\rm b}$ [^1]' title: 'Numerical Results for Transport Coefficients of Quark Gluon Plasma with Iwasaki’s Improved Action' --- Introduction ============ In the phenomenological study of quark gluon plasma(QGP), when its bulk properties are concerned, the system of quarks and gluons is usually treated as gas or liquid. Then the fundamental parameters of QGP such as transport coefficients, are very important information. The aim of this work is to calculate them from the fundamental theory of QCD.\ The calculation of transport coefficients is formulated in the framework of linear response theory of Kubo[@Zubarev; @Hosoya; @Karsch; @Horsley]. They are expressed by the space time integral of retarded Green’s function of energy momentum tensors at finite temperature. The shear viscosity $\eta$ is expressed as follows. $$\begin{aligned} \eta = - \int d^{3}x' \int_{-\infty}^{t} dt_{1} e^{\epsilon(t_{1}-t)} \hskip 2cm \nonumber \\ \times \int_{-\infty}^{t_{1}} dt'<T_{12}(\vec{x},t)T_{12}(\vec{x'},t')>_{ret} $$ Similarly the bulk viscosity $\zeta$ and heat conductivity $\chi$ are expressed in terms of the retarded Green’s function of $T_{11}$ and $T_{41}$ components of energy momentum tensor. In the pure gauge model the energy momentum tensors are written by the the field strength tensor.\ The direct calculation of the retarded Green’s function at finite temperature is very difficult. Then the shortcut is to calculate Matsubara Green’s function($G_{\beta}$) and then by the analytic continuation, we obtain the retarded Green’s function at finite temperature. The analytic continuation is carried out by the use of the fact that the spectral function of Fourier transform of the both Green’s functions is the same. For the spectral function we use the following simplest ansatz[@Karsch]. $$\begin{aligned} \rho(\vec{p}=0,\omega) \hskip 4cm \nonumber \\ = \frac{A}{\pi} (\frac{\gamma}{(m-\omega)^2+\gamma^2}-\frac{\gamma}{(m+\omega)^2+\gamma^2}) $$ where $\gamma$ partially represents the effects of the interactions and is related to the imaginary part of self energy. Under this ansatz, the transport coefficients are calculated as, $$\alpha \times a^3 = 2A\frac{2\gamma m}{(\gamma^2+m^2)^2}$$ where $\alpha$ means $\eta$, $\frac{4}{3} \eta+\zeta$ and $\chi \cdot T$. We notice that if $m=0$ or $\gamma=0$, transport coefficient becomes zero. In order to determine these parameters, at least three independent data points in $G_{\beta}$ are required in the temperature direction. In the pioneering work of Karsch and Wyld[@Karsch], they performed lattice QCD calculations on $8^3 \times 4$ lattice and the resolution was not enough to determine these three parameters independently. Numerical Results ================= We carry out our simulation on $16^3 \times 8$ lattice. We have started the the simulation from U(1) gauge theory. It is found that the fluctuation of $G_{\beta}$ is very large that it need about $10^6$ data for the determination of $G_{\beta}$ in the deconfined phase. Further in the confined phase, the fluctuations become still larger and we could not obtain the Green’s functions even with the $\sim 1.5 \times 10^{6}$ data[@Sakai]. Similar results are obtained in the case of SU(2) gauge theory. Then in the calculation of SU(3) gauge theory, we carry out our simulation only in the deconfined phase.\ It is also found that the fluctuation of $G_{\beta}$ becomes larger as we proceed to $U(1) \rightarrow SU(2) \rightarrow SU(3)$. Then it is a very important problem to reduce the fluctuation of $G_{\beta}$. We find that by using the Iwasaki’s improved action, the fluctuation is much reduced as shown in Fig.1. Then in the following we apply Iwasaki’s Improved action for the simulation of SU(3) gauge theory.\ From roughly $0.5 \times 10^{6}$ data, we obtain $G_{\beta}$ for $T_{11}$ and $T_{12}$ . But they have still rather large errors. The fit of $G_{\beta}$ with parameters in the spectral function given by Eq.(2) is done with SALS. The shear and bulk viscosities are calculated by these parameters by Eqs.(3) and the error are estimated by the Jackknife method. The results for them are shown in Fig.2. It is found that the bulk viscosity is zero within errors. This result is consistent with the arguments of S. Gavin[@Gavin] that in the pure gauge theory, the bulk viscosity and heat conductivity should be zero. In our calculation, $G_{\beta}$ of $T_{14}$ from which the heat conductivity is calculated, has large background and it has been impossible to get signal from it.\ In order to know the shear viscosity in the physical unit, we should determine the lattice spacing $a$ in these $\beta$ values. For this purpose we have started to determine the finite temperature transition points $\beta_c$ at $N_{T}=8$. We made a histogram analysis on $16^3 \times 8$ lattice. The results for the Polyakov susceptibility are shown in Fig.3. The spacial volume is too small to make a precise determination of $\beta_c$ on our lattice, but we find that the transition region is $2.71 \leq \beta_c \leq 2.73$. By using the finite size scaling formula and the transition temperature $T_{c} \sim 276(3)(2)MeV$ determined by Tsukuba group [@Kaneko] and assuming asymptotic scaling for $\beta \geq 2.73$ region, the lattice spacing is estimated.\ The preliminary results for the shear viscosity in the physical unit are shown in Fig.4. It is found that it increases with temperature. The $T$ dependence is consistent with $T^{3}$, because $\eta \times a^3$ is almost independent of $\beta$ as shown in Fig.2 and $a \propto 1/T$ when $N_{T}$ is fixed. But clearly more accurate data are necessary. It is also seen that it seems finite around $T_{c}$, which is slightly smaller than the typical hadron massess. What is the physical effects on the phenomenology of quark gluon plasma, when it has shear viscosity with this magnitudes, is a very interesting problem. Conclusions and Further Problem =============================== We have obtaind the transport coefficients of quark gluon plasma from the lattice QCD calculation by using Iwasaki’s improved action, as shown in Figs.2 and 4. The shear viscosity increases with temperarure roughly as $T^3$ The bulk viscosity is consistent with zero. And the heat conductivity is very difficult to calculate because of large background for $G_{\beta}$ of $T_{14}$. Our results depend strongly on the ansatz of spectral function in Eq.(2). In order to study the functional form for the spectral function, we have started the simulation with anisotropic lattices. We could not calculate the $G_{\beta}$ in the confined phase[@Sakai]. We think that this is because the energy momentum tensor operators should be written by the hadron fields, rather than the gluon fields. In order to certify this hypothsis, we have started the calculation of gluon propagator at finite temperature with improved action. The preliminary results on the small lattice $8^3 \times 4$ shows that in the confined phase, gauge copies are found in lattice version of Lorentz gauge and the gluon propagators show the peculiar behavior, which Nakamura group[@Nakamura] has found on the large lattice with standard action.\ ACKNOWLEDGEMENT\ This work is supported by the Supercomputer Project (No.97-27) of High Energy Accelerator Research Organization (KEK). We would like to express our thanks to the members of KEK for their warm hospitality. [9]{} D.N.Zubarev, Nonequilibrium Statistical\ Thermodynamics Plenum, New York 1974. A.Hosoya, M.Sakagami and M Takao, Annals of Phys.154(1984) 229. F.Karsch and H.W.Wyld,Phys.Rev.\ D35(1987) 2518. R.Horsley and W.Schoenmaker, Nucl. Phys. B280\[FS18\](1987),716, ibid.,735. Y. Iwasaki preprint UTHEP-118,\ S.Itoh, Y.Iwasaki, Y.Oyanagi and T. Yoshié, Nucl. Phys. B274(1986), 33.\ Y.Iwasaki, K.Kanaya, S.Sakai and T. Yoshié, Nucl. Phys. B(Proc. Suppl.) 42(1995), 502. A.Nakamura, S.Sakai and K. Amemiya, Nucl. Phys. B(Proc. Suppl) 53(1996),432 S. Gavin, Nuclear Phys.A345(1985) 826. Y. Iwasaki, K. Kaneko, K. Kanaya and T. Yoshie Nucl. Phys.B(Proc. Suppl.) 53(1996) 429. A. Nakamura, H. Aiso, M. Fukuda, T. Iwatani T. Nakamura and M. Yoshida, Proceedings of Confinement95(1996) 90. [^1]: Presented by S.Sakai	{ "pile_set_name": "ArXiv" }
--- abstract: 'The Global Historical Climatology Network-Daily database contains, among other variables, daily maximum and minimum temperatures from weather stations around the globe. It is long known that climatological summary statistics based on daily temperature minima and maxima will not be accurate, if the bias due to the time at which the observations were collected is not accounted for. Despite some previous work, to our knowledge, there does not exist a satisfactory solution to this important problem. In this paper, we carefully detail the problem and develop a novel approach to address it. Our idea is to impute the hourly temperatures at the location of the measurements by borrowing information from the nearby stations that record hourly temperatures, which then can be used to create accurate summaries of temperature extremes. The key difficulty is that these imputations of the temperature curves must satisfy the constraint of falling between the observed daily minima and maxima, and attaining those values at least once in a twenty-four hour period. We develop a spatiotemporal Gaussian process model for imputing the hourly measurements from the nearby stations, and then develop a novel and easy to implement Markov Chain Monte Carlo technique to sample from the posterior distribution satisfying the above constraints. We validate our imputation model using hourly temperature data from four meteorological stations in Iowa, of which one is hidden and the data replaced with daily minima and maxima, and show that the imputed temperatures recover the hidden temperatures well. We also demonstrate that our model can exploit information contained in the data to infer the time of daily measurements.' author: - 'Maxime Rischard[^1]' - 'Karen A. McKinnon' - Natesh Pillai bibliography: - 'temper.bib' title: ' [ Bias correction in daily maximum and minimum temperature measurements through Gaussian process modeling ]{}' --- 0 [0]{} 1 [0]{} [[ Bias correction in daily maximum and minimum temperature measurements through Gaussian process modeling ]{}]{} Introduction {#sec:introduction} ============ Long, high-quality records of temperature provide an important basis for our understanding of climate variability and change. Historically, there has been a focus on monthly-average temperature records that are sufficient for certain analyses, such as quantifying long-term changes in temperature. As our knowledge of climate change expands, however, there is increasing interest in understanding changes in temperature on shorter timescales, with a particular focus on extreme events. To do so, it is necessary to utilize temperature data with higher temporal resolution. Recent work has led to the development of the Global Historical Climatology Network-Daily (GHCND) database [@menne2012overview], which contains, among other variables, daily maximum and minimum temperatures from weather stations around the globe. The database draws from a range of different sources, and the data within it undergoes basic quality control to remove erroneous values. The current quality control methodology, however, does not account for so-called “inhomogeneities.” Inhomogeneities result from changes in measurement practices that impact the recorded temperatures. For temperature, known inhomogeneities include changes in the time of observation, changes in the thermometer technology, station relocation, and changes in land use around a station [@menne2009us]. While these inhomogeneities have a small effect on, for example, the estimation of global mean temperature, they can have a large effect on estimation of temperature variability and change at a more local scale. There is a large body of work focused on homogenizing monthly-average temperatures [e.g., @karl1986model; @easterling1996development; @peterson1998homogeneity; @ducre2003comparison; @menne2009homogenization; @vincent2012second], resulting in widely available, large-scale homogenized monthly temperature datasets. Homogenization typically proceeds through identifying non-climatic ‘breakpoints’ in a given time series through comparison with neighboring stations. Once a breakpoint is identified, the measurements recorded after the breakpoint are adjusted in some way to reduce or remove the inhomogeneity. Most applications of these methods, however, focus on adjusting the mean state of the data rather than the shape of the distribution [see @della2006method and references therein]. While this may be sufficient for monthly data, it is known that changes in measurement practices may affect different quantiles of the daily temperature distribution unequally. To address this issue, some homogenization methods have also employed frequency distribution matching techniques, so that each temperature recorded after a breakpoint is adjusted according to its percentile within the time series [@della2006method; @trewin2013daily]. The Problem {#sec:theproblem} ----------- \[sec:illustrate\_bias\] ![ \[fig:waterloo\_triangles\] An extract of the temperature measurements from KALO. The blue and red triangles respectively indicate the coldest and warmest temperature of each diurnal cycle. The blue and red lines respectively show the observed maximum and temperature recorded each day at 17:00 (top) or 5:00 (bottom) for the 24-hour period preceding the measurement. Discrepancies between the 24-hour extrema, and the peaks and troughs of the diurnal cycle, are indicated with dotted lines. ](figures/waterloo_triangles.pdf){height="0.4\textheight" width="99.00000%"} Many historical measurements of daily temperatures are provided as daily maximum and minimum temperatures (${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ respectively), which ideally measure the peak and trough of each diurnal temperature cycle. ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ are often recorded by an observer who every 24 hours visits a weather station equipped with a maximum-minimum thermometer, and notes the maximum and minimum registered by the instrument in the last 24 hours. In this section we explain how this measurement practice can cause the ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ measurements to fail to capture the peaks and troughs of some diurnal cycles. This has long been recognized in the scientific literature; see for example @baker1975effect and references therein. ![ \[fig:Iowa\_map\] Map of the four airport weather stations in Iowa providing hourly temperature records. Each airport is identified by its ICAO code. ](figures/Iowa_map.pdf){width="60.00000%"} illustrates the problem with ten days of hourly temperature measurements from the Waterloo Municipal Airport (KALO) weather station in Iowa. gives a map of the four Iowa weather stations used as examples throughout this paper. We emulate daily ${{\mathrm{T}}_{x}}$/${{\mathrm{T}}_{n}}$ measurements by dividing the data into 24 hour measurement windows, and reporting the minimum and maximum temperature that was recorded in this window. On most days, the measurements successfully capture the peak and trough of the diurnal cycle. But there are also several discrepancies (indicated with vertical dotted lines), typically in ${{\mathrm{T}}_{x}}$ when the measurements are made near the warmest hour of the day, and in ${{\mathrm{T}}_{n}}$ when the measurements are made near the coldest hour. A blatant example occurs on April 3rd, where the peak of the diurnal cycle is 7.2${{}^{\circ}\mathrm{C}}$ and occurs at 15:00 (all times are in the UTC-6 time zone, and tick marks are at midnight at the start of each day), but with measurements made at 17:00, the day’s ${{\mathrm{T}}_{x}}$ record of 16.1${{}^{\circ}\mathrm{C}}$ is reached immediately after the previous day’s measurement: a 8.9${{}^{\circ}\mathrm{C}}$ overestimate. Ideally, measurements of the diurnal cycle peak and trough would be obtained by recording ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ at the coldest and warmest time of day respectively. This would minimize the possibility of the previous or next diurnal cycle setting the measured ${{\mathrm{T}}_{x}}$ or ${{\mathrm{T}}_{n}}$. For convenience, however, most observers instead record data at a single daytime hour. Our goal is to address the bias that results from this measurement practice. ![ \[fig:waterloo\_avgTnTx\] Mean daily ${{\mathrm{T}}_{x}}$ (top left) and ${{\mathrm{T}}_{n}}$ (top right), and mean absolute daily change in ${{\mathrm{T}}_{x}}$ (bottom left) and ${{\mathrm{T}}_{n}}$ (bottom right), extracted from hourly temperature records at KALO in 2015, under varying measurement hours of ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$.](figures/waterloo_bias_2x2.pdf){height="0.3\textheight" width="99.00000%"} The bias in the daily records can in turn induce bias in the long-term summary statistics that are of climatological interest. A statistic as simple as the average daily maximum temperature for an entire year (2015) increases by over 1${{}^{\circ}\mathrm{C}}$ if the measurements are made at 15:00 compared to 9:00, as seen in . Conversely, the average ${{\mathrm{T}}_{n}}$ is colder by over 1${{}^{\circ}\mathrm{C}}$ if ${{\mathrm{T}}_{n}}$ is measured at 5:00 rather than 15:00. If the time of observation remained constant over time, this systematic bias would still exist, but it would not be linked to spurious trends in the data. However, there have been known (and likely unknown) changes in the time of observation. In the United States, for example, observers were instructed to switch from recording data in the afternoon to recording data in the morning beginning in the 1950s. This change has led to an apparent decrease in both ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ over time [@menne2009us]. Such spurious trends also compromise the study of weather variability, through summary statistics such as the average absolute change in daily temperature maxima and minima from one day to the next, as seen in . Our Approach {#sec:approach} ------------ One of our goals is to be able to infer the “true” ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ peaks and troughs of the diurnal cycle throughout the data records, so as to correct both the variance biases and the spurious trends. This stands in constrast to previous work, which has focused directly on addressing spurious trends. We approach the problem as a missing data problem: if we had access to the full temperature time series at the station rather than just ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ measured at an arbitrary time, we would be able to retrospectively choose the hour of measurements, to avoid the issues described in . Our idea therefore is to impute the hourly time series of temperatures at the location of the ${{\mathrm{T}}_{x}}$/${{\mathrm{T}}_{n}}$ measurements. In turn, the imputed time series can be used to create accurate summaries of temperature extremes. Our imputation strategy is to borrow information from the nearby weather stations, usually located at airports, that record the current temperature about once an hour. Although it should be noted that the sampling times are not always equally spaced, we refer to these measurements as “hourly” throughout this paper. They cannot be used directly for climatology, as the weather stations that provide them are not always as carefully documented, calibrated, and situated as the research stations included in the GHCND. For instance, weather stations at locations experiencing a lot of human activity, like airports, may record higher temperatures on average. However, even if mis-calibrated or systematically biased, the time series data from these nearby stations do contain valuable information about the hourly changes in temperatures on any given day. In this paper, we develop a spatiotemporal Gaussian process model pooling the information from nearby stations with hourly data and simulate multiple realizations of hourly temperature time series at each station of interest. The key technical difficulty is that these imputations of the temperature curves must satisfy the constraint of falling between the observed daily minima and maxima, and attaining those values at least once in a twenty-four hour period. We develop SmoothHMC, a novel and easy to implement Markov Chain Monte Carlo (MCMC) algorithm to sample from the posterior distribution satisfying the above constraints. Our constrained imputations are implemented in the Stan programming language [@stancite]; our code is publicly available on the first author’s GitHub account. Compared to a custom implementation, the Stan model code is short and Stan’s MCMC samplers are well-optimized, which makes our imputation strategy efficient and easy to reproduce. A First Spatiotemporal Model {#a-spatiotemporal-model} ============================ In order to pool the information from temperatures measured at various locations and times, we develop a spatio-temporal Gaussian process model. In its simplest form, we posit that temperatures from stations that are near each other are more correlated than distant stations, and that those correlations also decay in time. We model the simultaneous temperatures throughout a region as a Gaussian process, with covariance between two locations ${{\bm{x}}}$ and ${{\bm{x}}}'$ given by the squared exponential (SE) covariance with characteristic lengthscale $\ell_{{\mathrm{space}}}$ and variance $\sigma_{{\mathrm{space}}}^2$: $$\operatorname{{cov}}{\parenthesis{T({{\bm{x}}}), T({{\bm{x}}}') \mid t}} = k_{{\mathrm{space}}}({{\bm{x}}}, {{\bm{x}}}') = \sigma_{{\mathrm{space}}}^2 \exp{\parenthesis{-\frac{{\parenthesis{{{\bm{x}}}-{{\bm{x}}}'}}{^{\intercal}}{\parenthesis{{{\bm{x}}}-{{\bm{x}}}'}}}{2\ell_{{\mathrm{space}}}^2}}}\,. \label{eq:kspace}$$ Similarly, the time series of temperatures at a single location can be modeled as a Gaussian process with characteristic timescale $\ell_{{\mathrm{time}}}$ and variance $\sigma_{{\mathrm{time}}}^2$: $$\operatorname{{cov}}{\parenthesis{T(t), T(t') \mid {{\bm{x}}}}} = k_{{\mathrm{time}}}(t, t') = \sigma_{{\mathrm{time}}}^2 \exp{\parenthesis{-\frac{{\parenthesis{t-t'}}^2}{2\ell_{{\mathrm{time}}}^2}}}\,.$$ We combine the spatial and temporal model by multiplying the covariances functions: $$k({{\bm{x}}},{{\bm{x}}}',t,t') = k_{{\mathrm{time}}}(t,t') \cdot k_{{\mathrm{space}}}({{\bm{x}}}, {{\bm{x}}}')\,.$$ This yields the covariance of the Gaussian process underlying the spatio-temporal model of temperatures. The variances $\sigma_{{\mathrm{space}}}^2$ and $\sigma_{{\mathrm{time}}}^2$ are not separately identifiable, so we arbitrarily fix $\sigma_{{\mathrm{space}}}^2=1$. To allow for systematic differences between stations, we add a mean temperature parameter $\mu_{{\mathrm{station}{\squarebracket{i}}}}$ for each station, where ${\mathrm{station}{\squarebracket{i}}}$ is the index of the station at which observation $i$ was recorded. This parameter captures both systematic differences in temperature between locations, for example due to differences in altitude, vegetation, or built environment around the station, and also calibration errors in the measurement apparatus. The observation model depends on the type of measurement obtained at a given location. At stations $j$ that provide a full temperature time series, we model the $i^{\mathrm{th}}$ temperature record as a noisy measurement from the true time series, with iid normal noise: $$\begin{split} & {\mathrm{T}}_{ij} = \mu_{j} + f({{\bm{x}}}_j, t_{ij}) + \epsilon_{ij}\,,\quad \epsilon_{ij} \overset{{iid}}{\sim} \operatorname{\mathcal{N}}{\parenthesis{0,{\sigma_{\epsilon}}^2}}\,\\ & f({{\bm{x}}}_j, t_{ij}) \sim \operatorname{\mathcal{GP}}{\parenthesis{0, k_{}({{\bm{x}}},{{\bm{x}}}',t,t')}}\,.\\ \end{split} \label{eq:gpmodel}$$ The noise term captures measurement error and micro-fluctuations occuring on time scales much shorter than $\ell_{{\mathrm{time}}}$. At stations $j$ that only provide daily ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ records, we denote the time of the ${d}^{\mathrm{th}}$ daily measurement by ${t^{\mathrm{meas}}}_{{d}}$, and approximate the ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ observation respectively as the maximum or minimum temperatures at a discretized set of times $t_{ij}$ inside of ${({t^{\mathrm{meas}}}_{{d}-1},\,{t^{\mathrm{meas}}}_{{d}}]}$: $$\begin{split} {\parenthesis{{{\mathrm{T}}_{x}}}}_{{d}j} &= \max{\curlybracket{{\mathrm{T}}_{ij}\text{, for all $i$ such that } t_{ij} \in {({t^{\mathrm{meas}}}_{{d}-1},\,{t^{\mathrm{meas}}}_{{d}}]}}} \,,\\ {\parenthesis{{{\mathrm{T}}_{n}}}}_{{d}j} &= \min{\curlybracket{{\mathrm{T}}_{ij}\text{, for all $i$ such that } t_{ij} \in {({t^{\mathrm{meas}}}_{{d}-1},\,{t^{\mathrm{meas}}}_{{d}}]}}} \,, \end{split} \label{eq:obs_Tn_Tx}$$ with ${\mathrm{T}}_{ij}$ modeled as in . Fitting the Spatiotemporal Model -------------------------------- Software is readily available in many programming languages for fitting Gaussian process models, including inference on the covariance parameters. We chose to use the julia `GaussianProcesses.jl` package to fit the above spatiotemporal model to the hourly temperatures at four Iowa weather stations. The Iowa data set includes 47,864 measurements, which is computationally challenging to fit directly with a single Gaussian process. There are many methods to handle large data sets with Gaussian processes: for example @quinonero2007approximation review sparse approximations to Gaussian processes from a machine learning perspective, while @banerjee2008gaussian develop a method specifically for large spatial data sets. For simplicity, we chose instead to divide the data into 10-day chunks, modeled as independent Gaussian processes with shared hyperparameters. We put weak normal priors on $\mu_{{\mathrm{station}{\squarebracket{i}}}}$ with large standard deviation $\sigma_{\mu}=10\,{{}^{\circ}\mathrm{C}}$, which can be incorporated into the Gaussian process with an additional term $$k_{\mu}({{\bm{x}}}, {{\bm{x}}}') = \begin{cases} \sigma_\mu^2 &\text{if } {{\bm{x}}}= {{\bm{x}}}'\,, \\ 0 &\text{otherwise}\,. \end{cases}$$ added to the covariance function. The spatio-temporal covariance function becomes $$\operatorname{k_{\mathtt{SExSE}}}({{\bm{x}}},{{\bm{x}}}',t,t') = k_{{\mathrm{time}}}(t,t') \cdot k_{{\mathrm{space}}}({{\bm{x}}}, {{\bm{x}}}') + k_\mu({{\bm{x}}}, {{\bm{x}}}') \,, \label{eq:ksese}$$ which we denote $\operatorname{k_{\mathtt{SExSE}}}$ to distinguish it from the covariance functions developed later in . Our model thus has four free parameters, $\sigma_{{\mathrm{time}}}$, $\ell_{{\mathrm{time}}}$, $\ell_{{\mathrm{space}}}$ and ${\sigma_{\epsilon}}$, which we fit by maximizing the marginal likelihood of ${\mathrm{T}}$, the complete 2015 temperature time series provided at the four Iowa weather stations: $$\label{eq:optimization} \widehat\sigma_{{\mathrm{time}}},\widehat\ell_{{\mathrm{time}}},\widehat\ell_{{\mathrm{space}}},\widehat{\sigma_{\epsilon}}= \operatorname{arg\,max}_{\sigma_{{\mathrm{time}}},\ell_{{\mathrm{time}}},\ell_{{\mathrm{space}}},{\sigma_{\epsilon}}} {\curlybracket{ \operatorname{\mathbb{P}}{\parenthesis{ {\mathrm{T}}\mid {\sigma_{{\mathrm{time}}},\ell_{{\mathrm{time}}},\ell_{{\mathrm{space}}},{\sigma_{\epsilon}}} }} }}\,.$$ The fitted covariance values are found in . ------------------------------- ---------------------------- ----- ------ ------------- ${\sigma_{\epsilon}}$ (${{}^{\circ}\mathrm{C}}$) 0.4 0.4 0.2 $\sigma_{\mathrm{time}}$ (${{}^{\circ}\mathrm{C}}$) 3.7 3.1 0.5,0.9,4.4 $\ell_{\mathrm{time}}$ ($\mathrm{hr}$) 2.7 2.8 0.3,1.9,8.9 $\ell_{\mathrm{space}}$ ($\mathrm{km}$) 176 154 10,59,370 $\alpha_{{\mathrm{time}}}$ 0.3,1.1,0.3 $\sigma_{24}$ (${{}^{\circ}\mathrm{C}}$) 2.4 2.7 $\ell_{24}$ ($\mathrm{hr}$) 0.7 0.8 $\ell_{{\mathrm{space}}{}24}$ ($\mathrm{km}$) 1414 785 ------------------------------- ---------------------------- ----- ------ ------------- : Fitted parameters for each specification of the Gaussian process covariance function. For $\operatorname{k_{\mathtt{sumprod}}}$ , the parameters of the short-term, medium-term, and long-term components are separated by commas. Notice how shorter timescales $\ell_{\mathrm{time}}$ are associated with shorter lengthscales $\ell_{\mathrm{space}}$ by the fitted covariance function. \[table:fitted\_params\] Predictions Using Nearby Data ============================= \[sec:predict\_nearby\] ![\[fig:imputations\_2x2\]Imputations of the temperature time series at Waterloo Municipal Airport (KALO) between May 28, 2015 and June 1, 2015 (a) using only nearby data and the product of squared exponentials model; (b) using only nearby data and the sum of products model; (c) incorporating ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$ measurements under the product of squared exponentials model; and (d) incorporating ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$ measurements under the sum of products model. The mean is subtracted from each time series in (a) and (b) as the models leave the average temperature at the imputation site as a free parameter. For each imputation distribution, the mean is shown as a thick line, surrounded by an 80% credible envelope in lighter color, and example imputations as thinner lines.](figures/imputations_2x2.pdf){width="99.00000%"} After optimizing the parameters of the spatio-temporal covariance , we use the model —fitted to the data from nearby stations with full time series—to provide time series predictions at the station that only collects ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ data. Gaussian processes give closed-form expressions for the posterior distribution of the predicted temperatures. We denote the temperatures we wish to impute as ${\mathrm{T}}_{\mathrm{miss}}{}$ at times $t_{\mathrm{miss}}$ and location ${{\bm{x}}}_{\mathrm{miss}}$ and those observed at nearby stations as ${\mathrm{T}}_{\mathrm{nearby}}{}$, at times $t_{\mathrm{nearby}}$ and locations $X_{\mathrm{nearby}}$. Under the spatio-temporal model , ${\mathrm{T}}_{\mathrm{miss}}$ and ${\mathrm{T}}_{\mathrm{nearby}}$ are jointly multivariate normal, with mean zero and covariance given by $\operatorname{k_{\mathtt{SExSE}}}({{\bm{x}}},{{\bm{x}}}',t,t')$. Standard results for conditioning within multivariate normals then yield: $$\begin{split} {\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}&\sim \operatorname{\mathcal{N}}{\parenthesis{\mu_{{\mathrm{miss}}\mid {\mathrm{nearby}}}, \Sigma_{{\mathrm{miss}}\mid {\mathrm{nearby}}}}}\,\text{, with} \\ \mu_{{\mathrm{miss}}\mid {\mathrm{nearby}}} &= \operatorname{\mathbb{E}}{\parenthesis{{\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}}} = {\mathbf{K}_{{\mathrm{miss}},{\mathrm{nearby}}}}{{\bm{\Sigma}}_{{\mathrm{nearby}},{\mathrm{nearby}}}}^{-1} {\mathrm{T}}_{\mathrm{nearby}}\,\text{, and} \\ \Sigma_{{\mathrm{miss}}\mid {\mathrm{nearby}}} &= \operatorname{{var}}{\parenthesis{{\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}}} = {\mathbf{\Sigma}_{{\mathrm{miss}},{\mathrm{miss}}}}- {\mathbf{K}_{{\mathrm{miss}},{\mathrm{nearby}}}}{{\bm{\Sigma}}_{{\mathrm{nearby}},{\mathrm{nearby}}}}^{-1} {\mathbf{K}_{{\mathrm{miss}},{\mathrm{nearby}}}}{^{\intercal}}\,. \end{split} \label{eq:unconstrained_post}$$ All covariance matrices can be derived from the model. For example, the $ij^{\text{th}}$ entry of ${\mathbf{K}_{{\mathrm{miss}},{\mathrm{nearby}}}}= \operatorname{{cov}}{\parenthesis{{\mathrm{T}}_{\mathrm{miss}}, {\mathrm{T}}_{\mathrm{nearby}}}}$ is given by $\operatorname{k_{\mathtt{SExSE}}}({{\bm{x}}}_{\mathrm{miss}},X_{\mathrm{nearby}}{\squarebracket{j}},t_{\mathrm{miss}}{\squarebracket{i}},t_{\mathrm{nearby}}{\squarebracket{j}})$, where $X_{\mathrm{nearby}}{\squarebracket{j}}$ gives the location of the $j$th observation, and $t_{\mathrm{nearby}}{\squarebracket{j}}$ its time. The two ${\bm{\Sigma}}$ matrices have an additional ${\sigma_{\epsilon}}^2$ diagonal component for measurement noise. In (a), we show an example of predictions obtained from this spatio-temporal model. We withheld temperature measurements from KALO (shown in black), and then used data from the three remaining stations (KCID, KDSM and KMCW, shown in orange) to predict the 2015 temperature time series At KALO. To speed up computations, we process 73 days of data at a time, with 48 days overlapping between adjacent prediction windows so that predictions can always be made away from the edge of the prediction window (except at the start and end of the year). The predictions can be seen to combine information from the three other stations, giving less weight to KDSM, which is further away from KALO. We will discuss the quality of these predictions in more detail in , after completing the exposition of our imputation strategy. Imputing by Conditioning on Extrema {#imputations} =================================== Our aim is not simply to predict temperatures at a location with no measurements, but rather to impute hourly temperatures at a location with accurate measurements of the daily temperature extrema. This is an instance of a more general statistical problem: if a random $p$-vector ${\curlybracket{X_i:~i=1,\dotsc,p}}$ has a known distribution $F_X$, and its maximum ${X_{\max}}= \max_i{\curlybracket{X_i}}$ and minimum ${X_{\min}}= \min_i{\curlybracket{X_i}}$ are measured, how does one draw samples from ${F_{X \mid {X_{\max}},{X_{\min}}}}$, the distribution of $X$ conditional on ${X_{\max}}$ and ${X_{\min}}$? Conditional draws from ${F_{X \mid {X_{\max}},{X_{\min}}}}$ need to respect three constraints: one component of $X$ must be equal to ${X_{\min}}$, another to ${X_{\max}}$, and all other components must lie between ${X_{\min}}$ and ${X_{\max}}$. Conceptually, we could implement a valid imputation algorithm by drawing random samples $F_X$, and accepting only those samples that satisfy the three constraints. Unfortunately, if $F_X$ is a continuous distribution, the probability of a random draw from $F_X$ satisfying such sharp constraints is zero. One could envision adding some tolerance, so that samples with minimum and maximum within a small margin of ${X_{\max}}$ and ${X_{\min}}$ are retained, but as the dimensionality $p$ grows, the rejection probability will rapidly go to 1, thus requiring huge sample sizes. Ultimately, this rejection sampling strategy is therefore bound to fail. Markov Chain Monte Carlo (MCMC) techniques can also be used to draw samples from arbitrary distributions with densities known up to a constant. The density of ${F_{X \mid {X_{\max}},{X_{\min}}}}$ is obtained up to a constant multiplier through a simple application of Bayes’ theorem. It is proportional to the prior density of $F_X$ multiplied by indicators ensuring that the extrema are respected: $$\operatorname{\mathbb{P}}{\parenthesis{X \mid {X_{\max}},{X_{\min}}}} \propto \operatorname{\mathbb{P}}{\parenthesis{X}} \operatorname{\mathbb{I}}{\curlybracket{ \max_i{\curlybracket{X_i}} = {X_{\max}}}}\operatorname{\mathbb{I}}{\curlybracket{ \min_i{\curlybracket{X_i}} = {X_{\min}}}} \,. \label{eq:bayes_exact}$$ However, once again, this distribution is zero everywhere in $\mathbb{R}^p$, except in a (p-2) dimensional subspace where the $\min$ and $\max$ constraints are met. Consequently, out-of-the-box generic MCMC algorithms targeting will not successfully converge to ${F_{X \mid {X_{\max}},{X_{\min}}}}$. We therefore loosen the constraint by replacing the likelihood term $\operatorname{\mathbb{P}}{\parenthesis{{X_{\max}},{X_{\min}}\mid X}}$ with two narrow independent normal distributions around the minimum and maximum of $X$: $$\begin{split} \operatorname{\mathbb{P}}{\parenthesis{X \mid {X_{\max}},{X_{\min}}}} &\propto \operatorname{\mathbb{P}}{\parenthesis{X}} \operatorname{\mathcal{N}}{\parenthesis{{X_{\max}}\mid \max_i{\curlybracket{X_i}}, \epsilon^2}} \operatorname{\mathcal{N}}{\parenthesis{{X_{\min}}\mid \min_i{\curlybracket{X_i}}, \epsilon^2}}\,, \end{split} \label{eq:normal_lik}$$ where $\operatorname{\mathcal{N}}{\parenthesis{x \mid \mu, \sigma^2}}$ denotes the density of a normal distribution with mean $\mu$ and variance $\sigma^2$ evaluated at $x$. For small $\epsilon$, this is a reasonable approximation enabling the use of MCMC techniques. ![\[fig:constraints3d\] With three variables $X_1$, and $X_2$ and $X_3$, ${F_{X \mid {X_{\max}},{X_{\min}}}}$ resides in the one-dimensional six-sided loop shown with thicker green lines. This is a 1D manifold embedded in 3D space, and possessing sharp corners, making it difficult for most MCMC algorithms to explore.](figures/constraints3d.pdf){height="0.3\textheight" width="99.00000%"} This approximation to ${F_{X \mid {X_{\max}},{X_{\min}}}}$ remains a difficult distribution to sample from. We illustrate the constraint in a 3-dimensional setting in . The MCMC must travel efficiently along the six edges of the allowed subspace, and navigate corners when the index of the extremum components change. Hamiltonian Monte Carlo (HMC) has shown a remarkable ability to navigate complicated distributions, including distributions where the typical set has “pinch points” of strong curvature [@betancourt2017conceptual], similar to the “corners“ in ${F_{X \mid {X_{\max}},{X_{\min}}}}$. HMC’s efficient sampling relies on gradient information in order to move towards regions of high probability. The normal likelihood softened the extrema constraints, but the maximum and minimum functions also remove information from the gradient. The partial derivative of the log-likelihood of the maximum term with respect to $X_i$ is proportional to: $$\frac{\partial \log \operatorname{\mathcal{N}}{\parenthesis{{X_{\max}}\mid \max_i{\curlybracket{X_i}}, \epsilon^2}}}{\partial X_i} \propto {\parenthesis{{X_{\max}}- X_i}} \operatorname{\mathbb{I}}{\curlybracket{\operatorname{arg\,max}_j{\parenthesis{X_j}} = i}} \,.$$ The gradient pulls the maximum of the current MCMC state towards ${X_{\max}}$, and ignores all other components. This makes it difficult for HMC to efficiently explore scenarios where other components set the maximum. In order to assist the HMC algorithm, we make another approximation. We replace the $\max$ and $\min$ functions in with the $\operatorname{smoothmax}$ and $\operatorname{smoothmin}$ functions, defined on real inputs $x_1, \dotsc, x_p$ as: $$\begin{split} \operatorname{smoothmax}{\parenthesis{x_1, \dotsc, x_p ; k}} &= \frac{1}{k} \log{\parenthesis{\sum_{i=1}^p e^{kx_i}}}\,, \\ \operatorname{smoothmin}{\parenthesis{x_1, \dotsc, x_p ; k}} &= -\operatorname{smoothmax}{\parenthesis{-x_1, \dotsc, -x_p; k}}\,. \end{split} \label{eq:softmax}$$ As the sharpness parameter $k$ goes to infinity, $\operatorname{smoothmax}$ approaches the maximum, and $\operatorname{smoothmin}$ approaches the minimum. This substitution costs a small price in accuracy due to the approximation, but there is an important benefit: the gradient is now informative for all components of $X$: $$\frac{\partial \log \operatorname{\mathcal{N}}{\parenthesis{{X_{\max}}\mid \operatorname{smoothmax}{\parenthesis{X_{1:p} ; k}}, \epsilon^2}}}{\partial X_i} \propto {\parenthesis{{X_{\max}}- \operatorname{smoothmax}{\parenthesis{X_{1:p} ; k}}}} \frac{e^{k X_i}} {\sum_{j=1}^p e^{k X_j}} \,.$$ These modifications make HMC a viable algorithm to efficiently draw samples from the constrained posterior. Setting $k$ and $\epsilon$ is a compromise between exactness and efficiency; we found $k=10$ and $\epsilon=0.1\,{{}^{\circ}\mathrm{C}}$ to perform well for our application. Henceforth, we refer to this use of HMC and a smoothmax approximation to the target distribution as SmoothHMC. SmoothHMC provides a generally applicable algorithm to draw from a multivariate distribution conditionally on the observed minimum and maximum of its components. Demonstration of SmoothHMC {#sec:toy_example} -------------------------- ![\[fig:toy\_quantiles\](a) Prior distribution of $X_i$ displayed as mean $\mu_i$ (shown in every subplot to ease comparison) with 2 SD envelope; (b) Quantiles of the analytically derived posterior ${F_{X \mid {X_{\max}},{X_{\min}}}}$ conditioned on ${X_{\min}}$ (dark blue line) and ${X_{\max}}$ (red line); (c) Quantiles of the samples drawn from ${F_{X \mid {X_{\max}},{X_{\min}}}}$ using HMC (without the $\operatorname{smoothmax}$ approximation); (d) Quantiles of the samples drawn from ${F_{X \mid {X_{\max}},{X_{\min}}}}$ using SmoothHMC.](figures/toy_quantiles.pdf){width="99.00000%" height="0.55\textheight"} We demonstrate SmoothHMC’s ability to obtain draws from ${F_{X \mid {X_{\max}},{X_{\min}}}}$ in a simplified setting where the distribution function of ${F_{X \mid {X_{\max}},{X_{\min}}}}$ can be derived analytically and computed easily. In our application, $F_X$ is the posterior predictive multivariate normal distribution ${\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}$ obtained from nearby measurements, with mean and marginal variance evolving smoothly from one prediction to the next. To parallel this, we specify a random vector $X$ with each component $X_i$ normally distributed, and with sinusoidal means and variances, but without any correlations between them, so as to avoid a combinatorial explosion when obtaining ${F_{X \mid {X_{\max}},{X_{\min}}}}$ analytically: $$\begin{split} & X_i \overset{{\perp}}{\sim} \operatorname{\mathcal{N}}{\parenthesis{\mu_i, \sigma_i}} \,,\quad i = 1, \dotsc, 100 \,, \\ & \mu_i = 10 + \sin{\parenthesis{2\pi i / 50}} \,,\quad \sigma_i = 0.1+\cos^2{\parenthesis{2\pi i / 50}} \,.\\ & {X_{\max}}= \max_i{\curlybracket{X_i}} \quad\text{and}\quad {X_{\min}}= \min_i{\curlybracket{X_i}} \,. \end{split} \label{eq:toyspec}$$ The unconstrained distribution of $X_i$ is shown in (a). In this example, we aim to sample from the distribution of $X_i$ subject to the observation that ${X_{\max}}=12.5$ and ${X_{\min}}=8.8$. An analytical derivation of the marginals of ${F_{X \mid {X_{\max}},{X_{\min}}}}$ is provided in , and its quantiles shown in (b). To obtain samples from ${F_{X \mid {X_{\max}},{X_{\min}}}}$, we use the implementation of HMC provided by the Stan probabilistic programming language [@stancite]. In Stan, the user specifies a probabilistic data-generating process for the observed data, based on parameters and latent variables with accompanying priors. Stan then compiles this model into a custom `C++` program that efficiently implements posterior sampling using HMC. We implement two Stan models to draw from ${F_{X \mid {X_{\max}},{X_{\min}}}}$; code for both is available from the GitHub account of the first author. The first model implements , with a narrow normal likelihood term around the maximum and minimum, while the second model also uses the $\operatorname{smoothmax}$ approximation . For each Stan model, we obtain 4 HMC chains each with 10,000 warm-up samples followed by 10,000 samples. The quantiles of the samples obtained without the $\operatorname{smoothmax}$ approximation are shown in (c). By default, Stan initizializes each $X_i$ uniformly at random between -2 and 2, and for most variables, the algorithm remains stuck near the initial values. Most samples do not conform to the constraints imposed by the observed ${X_{\min}}$ and ${X_{\max}}$ values, which invalidates these imputations. However, once we replace the maximum function with the $\operatorname{smoothmax}$ function, with quantiles shown in (d), SmoothHMC is able to draw samples that respect the observed extrema. Furthermore, a visual comparison of the analytical quantiles in (a) and the SmoothHMC sample quantiles in (d) confirms that this sampling algorithm delivers a close approximation of the marginal distribution of each variable $X_i$ in ${F_{X \mid {X_{\max}},{X_{\min}}}}$. ![\[fig:toy\_joint\] Comparison of the joint joint PDF of $X_{23}$ and $X_{52}$ obtained analytically and from SmoothHMC samples. The central scatterplot shows the 40,000 SmoothHMC samples. Superimposed thereon are a contour plot (dash-dotted) of the joint marginal PDF of ${F_{X \mid {X_{\max}},{X_{\min}}}}$ for $X_{23}$ and $X_{52}$, and a contour plot (solid lines) of kernel density estimates for the subset of SmoothHMC samples where neither $X_{23}$ or $X_{52}$ is the min or max, obtained with a normal kernel with bandwidth $0.2$ (estimates are divided by the integrated mass of the kernel that is inside of the ${X_{\min}}$/${X_{\max}}$ boundaries). The dotted lines are one bandwidth away from the ${X_{\min}}$/${X_{\max}}$ boundaries, beyond which kernel density estimates are less reliable. The four histograms around the scatter plot are of the SmoothHMC samples adjacent to their x-axis, when one of the variables is an extremum. For example, the top histogram is of $X_{23}$ for samples where $X_{52}$ is the max, while the super-imposed pink line is the (truncated normal) marginal PDF of $X_{23}$ if it is neither the max nor the min, times the probability that $X_{52}$ is the max. Blue and red lines indicate ${X_{\min}}$ and ${X_{\max}}$ respectively. ](figures/toy_joint.pdf){width="97.00000%" height="0.7\textheight"} We also visually verify that SmoothHMC samples correctly from the joint distribution of any combination of variables. We do this for a pair of variables, $X_{23}$ and $X_{52}$, with results shown in . There is a close match between the contours of the analytical joint distribution function (dash-dotted contour lines) and of the kernel density estimate (solid contour lines) of the SmoothHMC samples. Each of the four histogram of samples where $X_{23}$ or $X_{52}$ occupies the minimum or maximum position matches the corresponding analytical distribution function well. This visual comparison of the sample and analytical distributions shows that SmoothHMC is yielding a good approximation of a sample drawn from the true ${F_{X \mid {X_{\max}},{X_{\min}}}}$ in this example. We did not examine the behavior of the sampling algorithm for the joint distribution of more than two variables due to the difficulty of visualizing such a distribution, but we see no reason to suspect that the algorithm suffers from pathological behaviors that do not appear in these univariate and bivariate inspections. Smoothmax Temperature Model --------------------------- Armed with the SmoothHMC algorithm implemented in Stan, we now return to the problem of imputing hourly temperature measurements. To impute the missing temperatures, we need to draw from the posterior distribution ${\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}, {{\mathrm{T}}_{n}}, {{\mathrm{T}}_{x}}$. Bayes’ theorem conditional on ${\mathrm{T}}_{\mathrm{nearby}}$ gives $$\operatorname{\mathbb{P}}{\parenthesis{{\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}, {{\mathrm{T}}_{n}}, {{\mathrm{T}}_{x}}}} = \frac{ \operatorname{\mathbb{P}}{\parenthesis{{{\mathrm{T}}_{n}}, {{\mathrm{T}}_{x}}\mid {\mathrm{T}}_{\mathrm{nearby}}, {\mathrm{T}}_{\mathrm{miss}}}} \operatorname{\mathbb{P}}{\parenthesis{{\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}}} }{ \operatorname{\mathbb{P}}{\parenthesis{{{\mathrm{T}}_{n}}, {{\mathrm{T}}_{x}}\mid {\mathrm{T}}_{\mathrm{nearby}}}} }\,.$$ The second term in the numerator is the posterior obtained in now acting as a prior. The denominator is a normalizing constant. The first term in the numerator is either zero or one, indicating whether ${\mathrm{T}}_{\mathrm{miss}}$ satisfies the constraint imposed by the observed ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$. Therefore, the posterior distribution takes a similar form to , which motivates the use of SmoothHMC. A small leap of faith is needed to accept that SmoothHMC’s success in a toy example in will extend to this application. There are three important differences between the toy example and the temperature time series model. Firstly, $F_X$ is now a multivariate normal distribution with strong correlations obtained as the posterior distribution of a Gaussian process in . Secondly, instead of a single minimum and maximum, we observe extrema for every 24 hour period. Thirdly, we allow for the mean temperature to be different at different locations, and so the imputed temperatures are shifted by an additional parameter $\mu_{{\mathrm{miss}}}$, to which we attach a vague prior. To summarize, the probabilistic model that we wish to draw posterior imputations of ${\mathrm{T}}_{\mathrm{miss}}$ from is given by: $$\begin{split} {\mathrm{T}}_{\mathrm{miss}}&= \mu_{\mathrm{miss}}+ {\mathrm{T}}_{{\mathrm{miss}}\mid {\mathrm{nearby}}} \quad\text{with}\quad \mu_{{\mathrm{miss}}} \sim \operatorname{\mathcal{N}}{\parenthesis{0,10^2}} \,\text{, and} \\ {\mathrm{T}}_{{\mathrm{miss}}\mid {\mathrm{nearby}}} &= {\mathrm{T}}_{{\mathrm{miss}}} \mid {\mathrm{T}}_{\mathrm{nearby}}\sim \operatorname{\mathcal{N}}{\parenthesis{\mu_{{\mathrm{miss}}\mid {\mathrm{nearby}}}, \Sigma_{{\mathrm{miss}}\mid {\mathrm{nearby}}}}}\, \\ {\parenthesis{{{\mathrm{T}}_{x}}}}_{{d}} &= \max{\curlybracket{{\mathrm{T}}_{{\mathrm{miss}},\,i}\text{, for all $i$ such that } t_{{\mathrm{miss}}\,,i} \in {({t^{\mathrm{meas}}}_{{d}-1},\,{t^{\mathrm{meas}}}_{{d}}]}}} \,,\\ {\parenthesis{{{\mathrm{T}}_{n}}}}_{{d}} &= \min{\curlybracket{{\mathrm{T}}_{{\mathrm{miss}},\,i}\text{, for all $i$ such that } t_{{\mathrm{miss}}\,,i} \in {({t^{\mathrm{meas}}}_{{d}-1},\,{t^{\mathrm{meas}}}_{{d}}]}}} \,. \end{split} \label{eq:idealmodel}$$ To sample from this model with SmoothHMC, we modify it with the $\operatorname{smoothmax}$ approximation to the maximum, and a normal likelihood: $$\begin{split} {\parenthesis{{{\mathrm{T}}_{x}}}}_{{d}} &\sim \operatorname{\mathcal{N}}{\parenthesis{\operatorname{smoothmax}_{i \in {({t^{\mathrm{meas}}}_{{d}-1},\,{t^{\mathrm{meas}}}_{{d}}]}} {\curlybracket{ T_{{\mathrm{miss}},i}; k=10}}, 0.1^2}}\,, \\ {\parenthesis{{{\mathrm{T}}_{n}}}}_{{d}} &\sim \operatorname{\mathcal{N}}{\parenthesis{\operatorname{smoothmin}_{i \in {({t^{\mathrm{meas}}}_{{d}-1},\,{t^{\mathrm{meas}}}_{{d}}]}} {\curlybracket{ T_{{\mathrm{miss}},i}; k=10}}, 0.1^2}}\,. \end{split} \label{eq:smoothed_model}$$ A few samples from this imputation procedure are shown in (c). From May 28, 2015 to June 1, 2015, hourly temperatures are imputed at KALO, using the hourly temperature measurements from nearby stations to inform the course of the temperatures, and constraining the imputations within the ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ extracted from the withheld time series at 11:00 each day. Imputations are obtained in nine day windows for computational reasons, with three days of overlap between adjacent windows so each imputation can be made at least three days away from the window’s edges. One can verify visually that the imputations respect the ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$ constraints, reaching but not exceeding each extreme on each day. Since we actually have hourly data for KALO, yet fed our algorithm only the daily extremes, we can also plot the hidden temperatures (in black), and see how faithfully the imputations reproduce them. We see that the imputations indeed track the true measurements closely. This success demonstrates that SmoothHMC is capable of imputing temperature time series from the constrained posterior distribution ${\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}, {{\mathrm{T}}_{n}}, {{\mathrm{T}}_{x}}$. Model Diagnostics {#sec:diagnostics} ================= Variogram --------- Model fit can be visually inspected by plotting temporal and spatial semi-variograms. The semi-variogram of a stationary spatio-temporal function $Y({{\bm{x}}},t)$ is a function of the spatial lag ${{\bm{h}}}$ and the temporal lag $r$ [see for example @sherman2011spatial chapter 6]: $$\gamma{\parenthesis{{{\bm{h}}},r}} = \frac{1}{2} \operatorname{\mathbb{E}}{\squarebracket{{\parenthesis{Y{\parenthesis{{{\bm{x}}},t}}-Y{\parenthesis{{{\bm{x}}}+{{\bm{h}}},t+r}}}}^2}} $$ For a Gaussian Process model, with a stationary covariance function $k({{\bm{h}}},r)=k({{\bm{x}}},{{\bm{x}}}+{{\bm{h}}},t,t+r)$, this can be expressed as: $$\label{eq:gp_variogram} \gamma{\parenthesis{{{\bm{h}}},r}} = {\sigma_{\epsilon}}^2 + k{\parenthesis{0,0}} - k({{\bm{h}}},r)\,.$$ From the data, the variogram can also be estimated empirically, by averaging the square differences of any two observations that are separated by ${{\bm{h}}}$ in space, and $r$ in time (in practice, time lags are binned). By comparing the empirical variogram to the variogram of the fitted covariance, we obtain a visual diagnosis of the model. ![\[fig:spatial\_variogram\] Semi-variograms of the temperature temperature time series at four Iowa weather stations, each labeled by its ICAO code. The empirical semi-variograms are shown in black, and the fitted variograms for the three covariance models proposed in this paper are shown in color. The temporal semi-variograms are shown on the diagonal, while the off-diagonal plots show the semi-variograms as a function of time lag for a fixed distance ${{\bm{h}}}$ equal to the distance between the two stations. ](figures/spatial_variogram.pdf){width="99.00000%"} In our Iowa example, there are only four possible locations. For each location, we plot the empirical temporal variogram $\widehat\gamma{\parenthesis{0,r}}$. Then, for each pair of stations separated by ${{\bm{h}}}$ (fixed), we can also plot the estimate $\widehat\gamma{\parenthesis{{{\bm{h}}},r}}$. We overlay the model’s semi-variogram from equation , resulting in . For each variogram, we have removed the effect of the $k_\mu$ covariance, which would shift the variogram between two stations by a large arbitrary constant. Correspondingly, we subtract the mean of each observed time series before obtaining the empirical variogram. We notice that the variogram of the model with product covariance tracks the empirical variogram well at short lags, but fails to capture the periodicity in the empirical variogram, and the fit degrades at long lag. We improve the model in . Error and Expected Error ------------------------ The variogram gives us a visual diagnostic of the overall model fit. To quantify the model’s predictive ability in the Iowa example, we compare the posterior mean temperature to the withheld truth, and obtain the empirical mean squared error (MSE) for $N$ predictions as: $$\label{eq:mse} \frac{1}{N} \sum_{i=1}^N {\squarebracket{ {\mathrm{T}}_{{\mathrm{miss}},i} - \operatorname{\mathbb{E}}{\parenthesis{{\mathrm{T}}_{{\mathrm{miss}},i} \mid {\mathrm{T}}_{\mathrm{nearby}},{{\mathrm{T}}_{x}},{{\mathrm{T}}_{n}}}} }}^2\,.$$ Equation is for the final predictions obtained using nearby hourly temperatures and local daily maxima and minima. A similar diagnostic can be computed for the intermediary predictions, which exclude the local ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$ information. At that stage, we are not concerned with any overall bias in the predicted temperatures, so we instead compute the sample variance of the errors as $$\operatorname{{var}}{\parenthesis{{\mathrm{err}}}} = \operatorname{{var}}_i {\curlybracket{ {\mathrm{T}}_{{\mathrm{miss}},i} - \operatorname{\mathbb{E}}{\parenthesis{{\mathrm{T}}_{{\mathrm{miss}},i} \mid {\mathrm{T}}_{\mathrm{nearby}}}} }}\,. \label{eq:varerr}$$ ----------------------------------------- --------- ---------------------------------------- -------------------------------------------------- ------------------------ ---------------------------------- Model $\operatorname{{var}}({\mathrm{err}})$ $\widehat{\operatorname{{var}}}({\mathrm{err}})$ $\operatorname{{MSE}}$ $\widehat{\operatorname{{MSE}}}$ $\operatorname{k_{\mathtt{SExSE}}}$ -55,614 1.59 0.88 1.12 0.44 $\operatorname{k_{\mathtt{SESE_{24}}}}$ -54,472 1.63 0.97 1.12 0.69 $\operatorname{k_{\mathtt{sumprod}}}$ -45,944 1.32 1.19 1.04 0.81 ----------------------------------------- --------- ---------------------------------------- -------------------------------------------------- ------------------------ ---------------------------------- : Model diagnostics for three Gaussian process covariance functions. \[table:diagnostics\] For our purposes, it isn’t sufficient for the spatio-temporal model to yield good predictions; we also require a good estimate of its own accuracy. We estimate the error variance expected by the model by sampling random draws ${\mathrm{T}}^{(k)}_{\mathrm{miss}}$, $k=1,\dotsc,K$ from the multivariate normal posterior distribution ${\mathrm{T}}_{{\mathrm{miss}},i} \mid {\mathrm{T}}_{\mathrm{nearby}}$, and computing the variance between the samples and the posterior expectation: $$\widehat{\operatorname{{var}}}{\parenthesis{{\mathrm{err}}}} = \frac{1}{K} \sum_{k=1}^K \operatorname{{var}}_i {\curlybracket{{\mathrm{T}}^{(k)}_{{\mathrm{miss}},i} - \operatorname{\mathbb{E}}{\parenthesis{{\mathrm{T}}_{{\mathrm{miss}},i} \mid {\mathrm{T}}_{\mathrm{nearby}}}}}} \label{eq:expected_varerr}$$ Similarly, to estimate the MSE expected by the model we use the MCMC draws $\tilde{\mathrm{T}}^{(k)}_{\mathrm{miss}}$, $k=1,\dotsc,K$ from SmoothHMC, and compute the MSE between the samples and the posterior expectation: $$\widehat{\operatorname{{MSE}}} = \frac{1}{K} \sum_{k=1}^K \frac{1}{N} \sum_{i=1}^N {\squarebracket{\tilde{\mathrm{T}}^{(k)}_{{\mathrm{miss}},i} - \operatorname{\mathbb{E}}{\parenthesis{{\mathrm{T}}_{{\mathrm{miss}},i} \mid {\mathrm{T}}_{\mathrm{nearby}},{{\mathrm{T}}_{x}},{{\mathrm{T}}_{n}}}}}}^2 \label{eq:expected_mse}$$ When evaluating models, we want the errors to be small, and so the error variance and MSE to be low. A well-calibrated model should also have the estimated error variance and MSE close to their empirical values and respectively. These diagnostics for our first spatio-temporal model, the product of squared exponentials, are found in the first row of . The error variance using only nearby measurements is already fairly low, with typical errors of order $\sqrt{1.59}=1.26\,{{}^{\circ}\mathrm{C}}$. Incorporating ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$ using SmoothHMC reduces it further to $\sqrt{1.12}=1.06\,{{}^{\circ}\mathrm{C}}$. However, the model is overly optimistic, and the expected errors underestimate the empirical errors. Improving the Basic Model {#sec:improving_model} ========================= In this section, we develop more sophisticated Gaussian process covariances than the simple product of squared exponential kernels $\operatorname{k_{\mathtt{SExSE}}}$ . We then assess whether these models improve the variogram and the predictive diagnostics that we presented in . The most salient feature of the empirical variogram that is not captured by the $\operatorname{k_{\mathtt{SExSE}}}$ covariance is the oscillation with a 24-hour period. It is intuitively clear that the diurnal cycle induces this periodic covariance, and that our model should be improved by incorporating this feature. Gaussian processes allow for periodic components of the covariance, for example the periodic squared exponential covariance function, which we use with a 24-hour period $$k_{24}(t,t') = \sigma_{24}^2 \exp{\squarebracket{ - \frac{2}{\ell_{24}^2} \sin^2{\parenthesis{ \pi \frac{t-t'}{\text{24 hrs}} }}}}\,.$$ We modify the spatiotemporal model by adding this diurnal component to it, with its own spatial decay component $k_{{\mathrm{space}}{}24}$ (with the same form as $k_{{\mathrm{space}}}$ in , and again with variance parameter fixed to 1): $$\operatorname{k_{\mathtt{SESE_{24}}}}({{\bm{x}}},{{\bm{x}}}',t,t') = k_{{\mathrm{time}}}(t,t') \cdot k_{{\mathrm{space}}}({{\bm{x}}}, {{\bm{x}}}') + k_{24}(t,t') \cdot k_{{\mathrm{space}}{}24}({{\bm{x}}}, {{\bm{x}}}') + k_\mu({{\bm{x}}}, {{\bm{x}}}') \,.$$ We also propose a more complex model, which breaks up $k_{{\mathrm{time}}}$ into short-term, medium-term and long-term correlation components: $$\begin{aligned} \operatorname{k_{\mathtt{sumprod}}}({{\bm{x}}},{{\bm{x}}}',t,t') &= k_{{\mathrm{time}}{}1}(t,t') \cdot k_{{\mathrm{space}}{}1}({{\bm{x}}}, {{\bm{x}}}') &\text{(short-term variation)} \\ &+ k_{{\mathrm{time}}{}2}(t,t') \cdot k_{{\mathrm{space}}{}2}({{\bm{x}}}, {{\bm{x}}}') &\text{(medium-term variation)} \\ &+ k_{{\mathrm{time}}{}3}(t,t') \cdot k_{{\mathrm{space}}{}3}({{\bm{x}}}, {{\bm{x}}}') &\text{(long-term variation)} \\ &+ k_{24}(t,t') \cdot k_{{\mathrm{space}}{}24}({{\bm{x}}}, {{\bm{x}}}') &\text{(diurnal cycle)} \\ &+ k_\mu({{\bm{x}}}, {{\bm{x}}}') &\text{(station mean)} \end{aligned} \label{eq:sumprod_kernel}$$ Each of $k_{{\mathrm{time}}{}1}$, $k_{{\mathrm{time}}{}2}$, and $k_{{\mathrm{time}}{}3}$, is a rational quadratic kernel: $$k_{RQ}(t,t') = \sigma^2 {\parenthesis{1 + \frac{{\parenthesis{t-t'}}^2}{2\alpha\ell^2} }}^{-\alpha}$$ and is multiplied by a spatial decay component, specified as a squared exponential with variance fixed at 1. Fitted covariance parameters for $\operatorname{k_{\mathtt{SESE_{24}}}}$ and $\operatorname{k_{\mathtt{sumprod}}}$ are found in . We now have three competing Gaussian process models, with covariance functions $\operatorname{k_{\mathtt{SExSE}}}$, $\operatorname{k_{\mathtt{SESE_{24}}}}$, and $\operatorname{k_{\mathtt{sumprod}}}$ respectively. We can compare them in four ways. Firstly, the variogram fit in is visibly improved by the introduction of the the diurnal component in $\operatorname{k_{\mathtt{SESE_{24}}}}$, and by the additional spatio-temporal correlation decay components in $\operatorname{k_{\mathtt{sumprod}}}$. Secondly, the marginal log-likelihood is the quantity maximized by the parameter fitting procedure in , with maximized values found in the second column of . The more complex models indeed yield a higher log-likelihood, promising a better model fit which should yield better predictions. Thirdly, we compare the variance of the error in the predicted temperatures specified in when withholding all the data from a test station. Averaged over all of 2015, this is given in the third column of , and shows more mixed results. The diurnal model $\operatorname{k_{\mathtt{SESE_{24}}}}$ performs slightly worse than the simple $\operatorname{k_{\mathtt{SExSE}}}$ model, and $\operatorname{k_{\mathtt{sumprod}}}$ only yields a small improvement. Fourthly, we compare the mean squared error specified in for imputations at the test station incorporating ${{\mathrm{T}}_{n}}$/${{\mathrm{T}}_{x}}$. Results in the fifth column also show more modest improvements for the more complex models. That said, with an expected MSE closer to its true value, $\operatorname{k_{\mathtt{sumprod}}}$ does give better estimates of its own inaccuracy. We interpret these results as a reminder that prediction accuracy using Gaussian process is sensitive to model specification when extrapolating, but fairly insensitive to the model when interpolating [@stein2012interpolation]. Our imputations interpolate the temperatures from nearby stations, further aided by the constraints imposed by the daily ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$ measurements, which could explain why the choice of model does not seem to have a large impact on the performance of our imputation procedure. This insensitivity can be seen as reassuring, as it shows robustness against model misspecification. Imputed Summary Statistics ========================== \(d) shows the imputations produced under the $\operatorname{k_{\mathtt{sumprod}}}$ covariance . This is the primary output of our imputation method, and the results are promising. Firstly, just like in the toy example presented in , the individual imputations meet the three constraints imposed by the measured minimum and maximum. Each day, the imputations stay between ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$, and the temperatures always drop to ${{\mathrm{T}}_{n}}$ and rise to ${{\mathrm{T}}_{x}}$ at some time of the day. The imputations reflect the uncertainty in the time at which the extrema are reached. Notably, on some days, the posterior distribution of the warmest (or coldest) time is bimodal: during the May 31 measurement window (from May 30 at 11:00 to May 31 at 11:00) for example, 72.5% of SmoothHMC imputations reach their peak before 20:00 on May 30, 27.5% after 8:00 on May 31, and none in between. We view as a particular strength of our approach that the imputations are able to capture this ambiguity, rather than being restricted to a single mode of the posterior distribution. ![\[fig:imputed\_summary\_stats\] Average minimum (left) and maximum (right) daily temperature obtained under varying hour of measurement from KALO data (shown in blue), and from imputations of the withheld data (shown in orange with 2 SD envelope) obtained under the $\operatorname{k_{\mathtt{SExSE}}}$ covariance function (top) and the $\operatorname{k_{\mathtt{sumprod}}}$ covariance function (bottom).](figures/imputed_summary_stats.pdf){height="0.35\textheight" width="99.00000%"} These imputations however are not the final aim of our analysis. Rather, our stated goal is to undo, or at least account for, the sensitivity of summary statistics to measurement time, for example the average ${{\mathrm{T}}_{x}}$ in . Equipped with these imputations, is it possible to infer what the value of the summary statistic would have been for different measurement hours? This possibility is demonstrated in , which shows the same summary statistic as in applied to the imputations as well as the (withheld) hourly data at KALO. It can be seen that the imputed summary statistics track within about 0.1 ${{}^{\circ}\mathrm{C}}$ of the true values. The product covariance $\operatorname{k_{\mathtt{SExSE}}}$ and the sum of products covariance $\operatorname{k_{\mathtt{sumprod}}}$ seem to perform equally well imputing the summary statistics for different times, but the $\operatorname{k_{\mathtt{sumprod}}}$ gives more honest, wider credible intervals. Inference on Measurement Hour ============================= ![\[fig:measure\_hour\_example\] Constrained and unconstrained imputations in an eight-day window, assuming (top) the correct measurement hour (11:00 UTC-6), and (bottom) a wrong measurement hour (23:00 UTC-6). Assuming the wrong measurement time drives the constrained mean imputation away from the unconstrained mean imputation.](figures/measure_hour_example.pdf){width="90.00000%" height="0.4\textheight"} Our analysis thus far has focused on the case where the hour of measurement ${\mathtt{hr}}$ is known in advance. This is a sometimes unrealistic assumption, and so inference on ${\mathtt{hr}}$ is desirable. It is conceptually straightforward to modify the measurement model with a uniform prior on ${\mathtt{hr}}$. However, ${\mathtt{hr}}$ affects which observations are attributed to each day’s measurements, which has a discontinuous (observations suddenly jump from one day to the next) and non-differentiable effect on the posterior, and so Hamiltonian Monte Carlo becomes unviable. We therefore do not consider the introduction of a uniform prior on ${\mathtt{hr}}$ in Stan to be feasible. ![ \[fig:hr\_inference\] Concordance ${\delta}_{\mathtt{hr}}$ for imputations of temperatures at KALO assuming measurement hours ${\mathtt{hr}}=1,\dotsc,24$. The true hour of measurement is 11:00, and obtains the highest $\delta_{\mathtt{hr}}$. Observations are associated with the date on which the observation occurs in the UTC timezone, which causes the discontinuity at 18:00 UTC-6. ](figures/hr_inference.pdf){height="0.4\textheight" width="90.00000%"} Our procedure allows us to obtain imputation samples of ${\mathrm{T}}_{\mathrm{miss}}$ conditional on ${\mathrm{T}}_{\mathrm{nearby}}$, ${{\mathrm{T}}_{n}}$, ${{\mathrm{T}}_{x}}$ and ${\mathtt{hr}}$. If we do so for ${\mathtt{hr}}=0,\dotsc,23$, is there information available in these samples to infer ${\mathtt{hr}}$? We examine sample imputations in to gain intuition. Rather unsurprisingly, assuming an incorrect measurement time leads to wildly inaccurate imputations, for example on March 2nd. But notice also that assuming the wrong time causes the mean constrained imputation to depart further from the unconstrained imputation (that is, the green and orange lines are further apart). This can be interpreted as an indication of an incompatibility between ${\mathrm{T}}_{\mathrm{nearby}}$ and ${{\mathrm{T}}_{n}}$/${{\mathrm{T}}_{x}}$, caused by assuming the wrong ${\mathtt{hr}}$. We therefore propose to calculate the probability ${\delta}_{\mathtt{hr}}$ of the mean constrained imputation under the unconstrained posterior given by $\autoref{eq:unconstrained_post}$, which we interpret as a measure of concordance between ${\mathrm{T}}_{\mathrm{nearby}}$ and ${{\mathrm{T}}_{n}}$/${{\mathrm{T}}_{x}}$: $$ {\delta}_{\mathtt{hr}}= \log \operatorname{\mathbb{P}}{\parenthesis{ {\mathrm{T}}_{\mathrm{miss}}\! = \! \mu{\parenthesis{{\mathtt{hr}}}} \mid {\mathrm{T}}_{\mathrm{nearby}}}} \,\text{, where}\ \, \mu{\parenthesis{{\mathtt{hr}}}} = \operatorname{\mathbb{E}}{\parenthesis{ {\mathrm{T}}_{\mathrm{miss}}\mid {\mathrm{T}}_{\mathrm{nearby}}, {{\mathrm{T}}_{n}}, {{\mathrm{T}}_{x}}, {\mathtt{hr}}}}\,, \label{eq:concordance}$$ Intuitively, ${\delta}_{\mathtt{hr}}$ will drop when the wrong ${\mathtt{hr}}$ is assumed, and we may be able to infer the true ${\mathtt{hr}}$ by maximizing ${\delta}_{\mathtt{hr}}$. In , we demonstrate this method on the withheld KALO time series, which has been replaced by ${{\mathrm{T}}_{x}}$/${{\mathrm{T}}_{n}}$ observations made at ${\mathtt{hr}}=11$. We use SmoothHMC to impute the withheld data for all of 2015 under each possible measurement hour ${\mathtt{hr}}=0,1,\cdots,23$. For each set of imputations, we compute the posterior mean $\mu{\parenthesis{{\mathtt{hr}}}}$ from the SmoothHMC samples, and the concordance ${\delta}_{\mathtt{hr}}$ (by necessity, modified to treat the center of each 73-day prediction window as an independent prediction). Pleasantly, the concordance is highest when the true hour of measurement is used so that, in this example at least, the correct hour of measurement would be infered. Conclusion {#sec:conclusion} ========== Climatological research relies on the ability to track small changes over long periods. For this reason, the bias induced by the measurement time that we demonstrate in could lead to wrong estimates and conclusions regarding long-term trends in temperature records. We reformulated the source of this bias as a missing data problem, and imputed the missing hourly temperatures at the weather station using posterior samples from a spatiotemporal Gaussian process model. The model allows the combination of information from the measured daily minimum (${{\mathrm{T}}_{n}}$) and maximum (${{\mathrm{T}}_{x}}$) temperatures, and from measurements of hourly temperatures at nearby meteorological stations. While ours is not a physical model, it is very flexible, and it performs well for the task of interpolating temperatures between nearby locations and times. Indeed, more complex covariance functions (with a diurnal component and a sum of short-range and long-range components) showed only modest improvements in the mean squared error of the imputations compared to a withheld hourly temperature record. Our model accounts for miscalibration and bias in the hourly temperature measurements by assigning a mean parameter to each location, which is given a weak independent prior with no spatial correlation. Therefore, our model only makes predictions at new locations up to a constant shift, and it only extracts information about the trajectory of the temperature time series from each weather station. However, our strategy rests on the assumption that the trajectory is not affected by biases and miscalibration. This assumption is violated for example if the presence of an airport has a very different effect on measured temperatures during the day and during the night, which would introduce bias in the imputations. Our model could be improved in the future with a more complete characterization of how daily temperatures differ systematically between locations. In order to condition the imputations on the daily ${{\mathrm{T}}_{x}}$ and ${{\mathrm{T}}_{n}}$, we developed SmoothHMC, a general algorithm based on Hamiltonian Monte Carlo with a smoothed approximation of the target distribution that can sample from a multivariate distribution conditionally on its observed minimum and maximum. It showed an excellent ability to sample from the conditional distribution in an example where the distribution function can also be obtained analytically. SmoothHMC is the main technical contribution of this paper, and we believe the method could find applications beyond the present setting. We used this method to obtained imputations of the temperature time series that satisfied the constraints imposed by the measured ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$. The imputation of withheld temperatures at KALO track the true temperatures, within a root mean square error of $1.02\,{{}^{\circ}\mathrm{C}}$. We view as particularly encouraging that the imputations successfully capture the uncertainty and sometimes bimodality in the time of the maximum or minimum temperature on days where this time is difficult to infer from the available information. Future improvements to the imputation strategy would include the inclusion of rounding errors in the measurement model, explicit treatment of non-stationarity due to coastlines or other geographical features, and of altitude differences. Gaussian process modeling allows for much flexibility in the choice of covariance kernels, and improved modeling should lead to more accurate imputations. The imputed time series are the primary output of this work, but they are intended as a starting point for further analyses motivated by different scientific goals. In particular, summary statistics can be applied to the imputations, such as the average ${{\mathrm{T}}_{x}}$, under different choices of daily measurement hours. Using imputations obtained for the withheld time series at KALO, we have demonstrated a good ability to recover this information (). The average ${{\mathrm{T}}_{x}}$ or ${{\mathrm{T}}_{n}}$ is an example of a possible follow-up analysis, chosen mostly as an illustrative proof of concept. We plan to use this method to compare the average temperature to the average of the measured ${{\mathrm{T}}_{n}}$ and ${{\mathrm{T}}_{x}}$ for a given location and year, with the former estimated using imputed time series. Lastly, we discussed the possibility of inferring the hour of measurement ${\mathtt{hr}}$. We gave some intuition for maximizing the concordance in order to infer ${\mathtt{hr}}$, and a single example where this strategy is successful. While promising, we lack a theoretical justification for this approach. It remains to be seen whether our approach is generalizable and successful in other examples, and whether it can be placed on sound theoretical bases. Ideally we would wish to estimate the posterior probability $\operatorname{\mathbb{P}}{\parenthesis{{\mathtt{hr}}\mid {\mathrm{T}}_{\mathrm{nearby}},{{\mathrm{T}}_{n}},{{\mathrm{T}}_{x}}}}$, for example by sampling from the joint posterior of ${\mathrm{T}}_{\mathrm{miss}}, {\mathtt{hr}}\mid {\mathrm{T}}_{\mathrm{nearby}},$${{\mathrm{T}}_{n}}$,${{\mathrm{T}}_{x}}$, but this is computationally difficult. Furthermore, it would be desirable not merely to infer the hour of measurement for an entire year, but to detect changepoints: days on which the measurement practice changed from one hour of measurement to another. We leave these improvements to inference of the measurement hour to future work. Derivation of the analytic posterior for toy example {#sec:analytical_posterior} ==================================================== In this appendix we derive and compute the conditional distribution ${F_{X \mid {X_{\max}},{X_{\min}}}}$ for the toy example of . We denote by $f_i(\cdot)$ and $F_i(\cdot)$ the prior probability distribution function and cumulative distribution function of $X_i$, i.e. the normal PDF and CDF with means and variances given by . Let ${{\operatorname{\mathbb{P}}{}_{ij}}}$ be the probability that $X_i$ is the minimum of $X$ and $X_j$ is its maximum. We also define ${{\operatorname{\mathbb{P}}{}_{i\bullet}}}= \sum_{j=1}^{100} {{\operatorname{\mathbb{P}}{}_{ij}}}$, the probability that $X_i$ is the minimum, and ${{\operatorname{\mathbb{P}}{}_{\bulletj}}}= \sum_{i=1}^{100} {{\operatorname{\mathbb{P}}{}_{ij}}}$, the probability that $X_j$ is the maximum. The cumulative distribution function of $X_i$ is then given by: $$\operatorname{\mathbb{P}}{\parenthesis{X_i \leq x \mid {X_{\max}}, {X_{\min}}}} = \begin{cases} 0 &\text{if } x < {X_{\min}}\,, \\ 1 &\text{if } x \geq {X_{\max}}\,, \\ {\operatorname{\mathbb{P}}{}_{i\bullet}} + {\parenthesis{1 \!-\! {\operatorname{\mathbb{P}}{}_{i\bullet}} \!-\! {\operatorname{\mathbb{P}}{}_{\bulleti}}}} {\squarebracket{\frac{F_i(x) - F_i({X_{\min}}) } {F_i({X_{\max}}) - F_i({X_{\min}}) } }} &\text{otherwise.}\\ \end{cases}$$ Meanwhile, ${{\operatorname{\mathbb{P}}{}_{ij}}}$ is proportional to: $$f_i({X_{\min}}) f_j({X_{\max}}) \prod_{k \neq i,j}^{100} {\parenthesis{F_k({X_{\max}}) - F_k({X_{\min}})}} \,,$$ which we compute for all $i,j$ and normalize to obtain the $100 \times 100$ matrix $\operatorname{\mathbb{P}}$ of probabilities of each pair of element occupying the extremes. We sum over its rows and columns to obtain ${{\operatorname{\mathbb{P}}{}_{\bulletj}}}$ and ${{\operatorname{\mathbb{P}}{}_{i\bullet}}}$. While this algorithm has cubic complexity in the dimensionality $p$ of $X$, for $p=100$, it only take seconds to compute the entries of $\operatorname{\mathbb{P}}$ and evaluate $\operatorname{\mathbb{P}}{\parenthesis{X_i \leq x \mid {X_{\max}}, {X_{\min}}}}$ over a range of $x$. (b) shows the analytical quantiles of ${F_{X \mid {X_{\max}},{X_{\min}}}}$ marginally for each $X_i$. [^1]: We thank Peter Huybers, Debdeep Pati and Martin Lysy for their ideas, questions and suggestions.	{ "pile_set_name": "ArXiv" }
--- abstract: 'Advanced LIGO’s sensitivity will be limited by coating noise. Though this noise depends on beam shape, and though nongaussian beams are being seriously considered for advanced LIGO, no published analysis exists to compare the quantitative thermal noise improvement alternate beams offer. In this paper, we derive and discuss a simple integral which completely characterizes the dependence of coating thermal noise on shape. The derivation used applies equally well, with minor modifications, to all other forms of thermal noise in the low-frequency limit.' author: - 'R. O’Shaughnessy' bibliography: - 'apj-jour.bib' - 'LIGO-design-2-mirrors-coatings.bib' - 'LIGO-design-2-mirrors-control.bib' - 'popsyn.bib' - 'short-grb-data-analysis.bib' - 'supernovae-theory.bib' - 'gw-astronomy.bib' - 'gw-astronomy-pulsars.bib' - 'gw-astronomy-mergers.bib' date: 'Received ?? Month 2006, printed ' title: 'Coating thermal noise for arbitrary-shaped beams' --- Introduction ============ Though gravitational wave detectors such as LIGO are presently taking data, the best estimates from the astrophysical community for gravitational waves from compact object merger rates [@PSconstraints; @StarTrack; @Chunglee-nsns-1] (though some disagree [@Nakar]), cosmic strings [@2006PhRvD..73j5001S], rotating neutron stars [@2006PhRvD..73h4001N; @2006CQGra..23S...1O; @2006PhRvD..73b4021L], and supernovae [@astro-ph..0605493; @2006ApJ...640..878B] suggest that discoveries are most likely to begin with next-generation ground based interferometers like advanced LIGO. The present consensus advanced LIGO design has astrophysical reach (e.g., as measured by the distance to which a pair of inspiralling neutron stars could be detected) limited by coating thermal noise [@2006ApOpt..45.1569H]. In this context, thermal noise denotes the phase noise in the IFO produced by elastic oscillations of the mirror excited by the thermal bath of the remaining degrees of freedom [@1999PhLA..264....1B]; coating thermal noise denotes strong contributions to the noise arising when couplings between elastic modes and the thermal bath (i.e., losses) are predominantly located in the thin mirror coating off which the test beam reflects. Thermal noise depends strongly on beam shape: as one can show by applying the fluctuation-dissipation theorem to a low-frequency limit @2000PhRvD..62l2002L, for gaussian beams thermal noise power goes as $\propto r_o^{-2}$. A flatter beam which more equitably averages over fluctuations, such as “mesa” beams [@DOSTV-2004; @OSV-2004] or hyperboloidal generalization [@2004.gr-qc..0409083; @2004.gr-qc..0409084; @2006PhRvD..73l7101G], should reduce coating thermal noise. In this paper, we provide a simple two-dimensional integral to allow comparison of the coating thermal noise for different mode shapes. Specifically, the fluctuation dissipation theorem, plus symmetry arguments about half-infinite mirrors, plus some scaling arguments imply the power spectrum $S_x(\omega)$ of coating thermal noise must be proportional to the coating thickness $d$ and to two-dimensional integral over the fourier transform $\tilde{P}(K)$ normalized beam shape $P(r)$ $$S \propto d \int d^2 K \|\tilde{P}(K)\|^2$$ where $\tilde{P}(K)$, its two-dimensional fourier transform. However, the symmetry arguments presented can be applied to nearly any system with (approximate) two-dimensional translation symmetry (i.e., with a small beam on a large mirror): $S\propto \int d^2 K K^p \|\tilde{P}(K)\|^2$ for some constant index $p$. This index can be uniquely determined by comparison to other calculations for gaussian beams; thus the correct noise dependence on beam shape can be easily determined and understood for bulk thermoelastic noise ($S\propto r_o^{-3}$ implies $p=1$) [@OSV-2004; @1999PhLA..264....1B], for bulk thermal noise ($S\propto r_o^{-1}$ implies $p=-1$) [@1998PhRvD..57..659L; @1999PhLA..264....1B; @1998PhRvD..57..659L], and for coating thermoelastic noise ($p=0$) [@2004PhRvD..70h2003F]. Scaling argument ================ According to the fluctuation-dissipation theorem, coating thermal noise is proportional to the power dissipation rate $W_\text{diss}$ associated with a fluctuating pressure of shape $P(r)$ on the mirror surface. Manifestly (for half-infinite mirrors), $W_\text{diss}$ must be proportional to a translation-invariant inner product on $P$, of form $$\begin{aligned} W_\text{diss}\propto \int d^2 R \int d^2 R' V(R-R') P(R) P(R') \nonumber\\ \propto \int d^2 K \tilde{G}(K) \|\tilde{P}(K)\|^2\end{aligned}$$ By definition, the coating thermal noise is the contribution of the coating to the total thermal noise; thus, expanding in powers of coating thickness, $$\tilde{G}(K,d) \approx \tilde{G}_o(K) + d \tilde{G}_1(K) + \ldots$$ Since no other transverse scale exists in the half-infinite mirror, the kernel $\tilde{G}_1$ must be scale invariant, and therefore satisfy $\tilde{G}_1(\lambda K)=\lambda^p \tilde{G}(K)$, and thereforem be of form $$\tilde{G}_1(K) = K^p c_1$$ for some constant $c_1$. Finally, to recover the usual result for gaussian beams (i.e. $S\propto d/r_o^2$, as has been extensively calculated [@2006ApOpt..45.1569H; @2002PhRvD..65j2001N; @1998PhRvD..57..659L]), we must have $p=0$. Detailed calculation ==================== To check this simple scaling argument, we can perform the full fluctuation-dissipation calculation of coating thermal noise in a special case where the exact solution is known: when the elastic properties of the medium and coating are identical. The relevant elastic green’s functions for a half-infinite mirror are provided in an appendix of @2002PhRvD..65j2001N. From Nakagawa et al’s Eq. (1) , we know $$S \propto \int d^2 R \int d^2 R' P(R) P(R') \text{Im}\chi_{zz}(R-R')$$ where $\text{Im}\chi_{zz}(R,R')$ is given by their Eqs. (4-5): $$\begin{aligned} \text{Im}\chi_{zz}(r) &=& \phi \frac{1-\sigma^2}{\pi E} \left[ F(r,0)-F(r,d) \right] \\ F(r,z)&=& \frac{1}{\sqrt{r^2+4z^2}} \nonumber \\ & & \times \left( 1+ \frac{z^2/(1-\sigma)}{r^2+4z^2} + 12 \frac{z^4/(1-\sigma)}{(r^2+4z^2)^2} \right)\end{aligned}$$ We can equivalently represent this integral in the fourier domain, as $$S \propto \int d^2 K \|\tilde P(K)\|^2 \left[ \tilde{F}(K,0) -\tilde{F}(K,d) \right]$$ where \[Nakagawa et al Eq. A1\] $$\tilde{F}(K,d) = 2 \pi \frac{e^{-2 K d}}{K} \left[ 1+ \frac{K d}{1-\sigma} + \frac{(K d)^2}{1-\sigma} \right]$$ In other words $$\begin{aligned} S&\propto& \int_0^\infty d^2K \|\tilde P(K)\|^2 \\ & & \times \left[ \frac{-1+\exp[-2 Kd]}{K} + \frac{d \exp[-2 K d] (1+Kd)}{1-\sigma} \right] \nonumber\end{aligned}$$ Naturally, $\tilde{P}(K)$ drops to zero well before $K\approx 1/d$; therefore, we may take a small-$d$ limit. We therefore conclude $$S \propto \int d^2 K \|\tilde{P}(K)\|^2 \; .$$ Conclusions =========== Complementing similar earlier studies by @OSV-2004 on thermoelastic noise. in this paper we describe how to calculate how coating thermal noise varies with beam shape. An independent derivation, as well as detailed discussion of alternative beam-shape applications, will be forthcoming by other authors (Lovelace et al, in preparation). More generally, this paper describes a simple way to unify several disparate calculations for the beam shape dependence of thermally-driven noise (e.g., coating thermal noise; bulk thermoelastic noise; bulk thermal noise) produced when a beam reflects off a large mirror. Geoffrey Lovelace deserves considerable thanks for pointing out an error (an incorrect power index for coating thermoelastic noise). This work is partially supported by NSF Gravitational Physics PHYS grant-0353111.	{ "pile_set_name": "ArXiv" }
--- abstract: 'A key goal of the Stage IV dark energy experiments Euclid, LSST and WFIRST is to measure the growth of structure with cosmic time from weak lensing analysis over large regions of the sky. Weak lensing cosmology will be challenging: in addition to highly accurate galaxy shape measurements, statistically robust and accurate photometric redshift (photo-z) estimates for billions of faint galaxies will be needed in order to reconstruct the three-dimensional matter distribution. Here we present an overview of and initial results from the Complete Calibration of the Color-Redshift Relation (C3R2) survey, designed specifically to calibrate the empirical galaxy color-redshift relation to the Euclid depth. These redshifts will also be important for the calibrations of LSST and WFIRST. The C3R2 survey is obtaining multiplexed observations with Keck (DEIMOS, LRIS, and MOSFIRE), the Gran Telescopio Canarias (GTC; OSIRIS), and the Very Large Telescope (VLT; FORS2 and KMOS) of a targeted sample of galaxies most important for the redshift calibration. We focus spectroscopic efforts on under-sampled regions of galaxy color space identified in previous work in order to minimize the number of spectroscopic redshifts needed to map the color-redshift relation to the required accuracy. Here we present the C3R2 survey strategy and initial results, including the 1283 high confidence redshifts obtained in the 2016A semester and released as Data Release 1.' author: - \| Daniel C. Masters, Daniel K. Stern, Judith G. Cohen, Peter L. Capak,\ Jason D. Rhodes, Francisco J. Castander, Stéphane Paltani bibliography: - 'biblio.bib' title: \| The Complete Calibration of the Color-Redshift Relation (C3R2) Survey:\ Survey Overview and Data Release 1 --- Introduction ============ The upcoming large-scale cosmology experiments Euclid [@Laureijs11], LSST [@Ivezic08] and WFIRST [@Spergel15] will depend on robust photometric redshift (photo-z) estimates for billions of faint galaxies in order to obtain a three-dimensional picture of the growth of cosmic structure. Small ($\gtrsim$0.2%) redshift biases can easily dominate the overall error budget in the cosmological parameters measured by these surveys (e.g., [@Huterer06]). Spectroscopic calibration efforts for these missions must therefore measure the color-redshift relation of galaxies with sufficient fidelity to reconstruct the redshift distributions of shear samples with negligible systematic bias. While photometric redshift estimation techniques have grown in sophistication and precision over the past few decades (e.g., [@Benitez00; @Brammer08; @Ilbert09; @Carrasco13; @Speagle16]), existing methods have not met the photo-z accuracy requirements set by weak lensing cosmology. The relation of seven or eight galaxy broadband colors (referred to henceforth by the vector $C$) to redshift is ultimately an empirical question. In @Masters15 (hereafter M15) we demonstrated a method, based on the self-organizing map (SOM; [@Kohonen90]) algorithm, to constrain the empirical multidimensional color distribution of galaxies present in a survey. This technique allowed us to project the multicolor distribution of galaxies in a topologically ordered way onto a two-dimensional grid. By applying this technique to a well-studied deep field with uniform ugrizYJH photometry, we were able to demonstrate that spectroscopic surveys to date do not sample the full color-space of galaxies in a Euclid-like survey, and thus the color-redshift relation is not fully constrained with existing spectroscopy. This issue is of particular concern for machine learning-based photo-z estimation, which requires color-complete training samples, but also affects the calibration of template-based techniques. The analysis in M15 motivated a survey designed to systematically map the color-redshift relation over the currently undersampled regions of galaxy color space relevant to Euclid. M15 estimated that $\sim$5000 new redshifts, carefully distributed in color space, would be sufficient to meet the stringent requirements for weak lensing cosmology. This “direct" approach to photo-z calibration is complementary to approaches based on spatial cross-correlation of photometric samples with spectroscopic samples (e.g., @Newman08 [@Rahman15]. At least two independent methods to measure $N(z)$ for the tomographic shear samples will be required to ensure no systematic photo-z biases exist; these methods can therefore serve as useful checks on each other. Here we describe the initial stage of what we are calling the Complete Calibration of the Color-Redshift Relation (C3R2) survey, designed to fill out the color space of galaxies with secure redshifts to the Euclid weak lensing depth. By doing so, the empirical relation obeyed by galaxies can be constrained with sufficient accuracy to meet the cosmological requirements of Euclid. The spectra will also be of significant value for the LSST and WFIRST calibrations, which will be more difficult than for Euclid due to the greater photometric depth of those surveys (Hemmati et al, 2017, in prep). We estimate that $\sim$40 Keck nights in total (or their equivalent) could achieve the fidelity required to meet the cosmological requirements for Euclid, when combined with extensive existing spectroscopy. -- -- -- -- -- -- \[figure:colors\] This paper gives an overview of the C3R2 survey and presents results from the 2016A semester, which constituted the first five nights of observing. All 2016A observations were done with Keck. The structure of this paper is as follows. In §2 we give an overview of C3R2 survey strategy. In §3 we discuss the observations and data reduction for the first five nights of observing. In §4 we describe redshift determination and the identification of serendipitous sources. In §5 we present initial results from the survey. In §6 we conclude with a discussion. High confidence redshifts from DR1 are provided in a machine readable table. C3R2 Survey Overview ==================== The Keck portion of the C3R2 survey is a joint effort between Caltech (PI: J. Cohen), NASA (PI: D. Stern), the University of California (PI: B. Mobasher), the University of Hawaii (PI: D. Sanders). European participation in C3R2 with the GTC (PI: C. Gutierrez) and VLT (PI: F. Castander), as well as Harvard participation with MMT (PI: D. Eisenstein), will commence in 2017. The first five nights of observing with Keck were allocated by Caltech in the 2016A semester. Here we provide a brief overview of the C3R2 strategy for these observations. The Self-Organized Map of Galaxy Colors --------------------------------------- In M15 we used COSMOS [@Capak07; @Scoville07; @Laigle16] ugrizYJH photometry of $\sim$130k galaxies, closely resembling what will be obtained by the Euclid survey, to map the color distribution of galaxies to the Euclid depth ($i$$\sim$24.5 AB). We used the SOM algorithm (a manifold learning technique for nonlinear dimensionality reduction) to generate a topologically ordered 2D representation of the high-dimensional color distribution[^1]. Galaxies from COSMOS were then matched back to the self-organized map according to their best-matching color cell in the SOM. This sorting of galaxies enables a variety of analyses, including the density of galaxies in different parts of color space, the median 30-band photometric redshifts from COSMOS as a function of position in color space, and the distribution of spectroscopic redshifts on the map (Figure \[figure:colors\]). Importantly, by placing all existing spectroscopy from the COSMOS field on the map, we reveal regions of color space for which no galaxies have existing high-confidence redshifts. Of greatest importance for the C3R2 survey are: (1) the current spectroscopic sampling across color space, and (2) the source density as a function of position in color space, as more common galaxies will contribute more to the cosmic shear signal. Existing Spectroscopy Across Galaxy Color Space ----------------------------------------------- For C3R2 we need to identify the regions of galaxy color space for which spectroscopic redshifts already exist and where they are systematically missing. We collected existing spectroscopy in COSMOS to do this, as described in M15. These redshifts include (but are not limited to) those from VLT-VIMOS [@Lilly07; @LeFevre15], Keck-MOSFIRE [@Kriek15], Keck-DEIMOS [@Kartaltepe10], and Magellan-IMACS [@Trump07]. For the 2016A run we used only the spectroscopy taken in the COSMOS survey to identify undersampled regions of color space. The reason we could not incorporate spectroscopy from other fields for these observations is that the photometry between fields has to be highly consistent in multiple bands to reliably place galaxies on the same color map; at the time this problem had not been solved. Significant subsequent work has been done to solve this problem for upcoming runs, to be described in a forthcoming paper. The fields that have subsequently been put on a highly consistent color frame in ugrizYJH to the Euclid depth are VVDS, SXDS, and EGS (in addition to COSMOS). Target Prioritization --------------------- For the 2016A observations we used the SOM derived in M15 to prioritize regions of galaxy multicolor space that are currently undersampled by existing spectroscopic surveys. For observed fields in 2016A other than COSMOS (SXDS and EGS), we attempted to bring the photometry on to the COSMOS color system in order to select the targets in a consistent way. We used the CANDELS [@Grogin11] photometry in a subset of the COSMOS field together with the CANDELS photometry in SXDS and EGS to derive a rough color conversion between the fields. Target prioritization for C3R2 is based on two main factors: (1) the usefulness of a galaxy for calibrating the relation, and (2) the likelihood of obtaining a secure redshift given the instrument, exposure time, and expected galaxy properties. The usefulness of a particular galaxy to the redshift calibration effort depends both on how common its colors are in the data and whether high-confidence redshifts for galaxies with similar colors already exist. Based on these considerations, we developed a prioritization scheme for galaxies that weights sources in unsampled cells of the SOM more heavily, and also gives preference to more common galaxy colors. The priorities for C3R2 are adaptive as new data is obtained and more of the color space is filled in. For the 2016A run our priority scheme was as follows: 1. [We assign a initial priority value of 10 to objects occupying cells with no spectroscopic redshifts of even moderate quality (the gray regions of the SOM in the middle panel of Figure \[figure:colors\]), a starting priority of 3 to objects in cells with a spectroscopic redshift(s) of only moderate confidence, and a starting priority of 1 to galaxies in color cells that already have one or more high-confidence spectroscopic redshifts. Galaxies with existing redshifts of at least moderate confidence were not targeted.]{} 2. [We multiply each galaxy’s priority by the number of objects in its color cell, effectively upweighting sources with common SEDs.[^2]]{} 3. [We penalize objects that are color outliers within their color cell in order to avoid using them for calibration. A small fraction of objects in the sample are not represented well in the SOM, either because they have abnormal colors from photometric errors or the superposition of two or more sources, or are truly rare objects (e.g., X-ray sources). We want to avoid calibrating with these.]{} As will be described in future data releases, this prioritization scheme has been refined for the 2016B and later observations to more efficiently map the color-redshift relation. In M15 we pointed out that spectroscopic effort could also be intentionally directed at regions of color space with intrinsically higher redshift uncertainty (e.g., with double-peaked redshift PDFs). For now we have not prioritized based on redshift uncertainty; however, as the survey progresses and the color map is filled in we may incorporate this quantity. Estimating Required Instruments & Exposure Times ------------------------------------------------ A crucial element of the C3R2 survey is the use of best-fit spectral templates to the galaxies to predict the exposure times with different instruments needed to obtain a secure redshift. If we then fail to obtain a redshift under nominal observing conditions we can prioritize the target further for follow-up. This potential re-targeting is important to avoid systematic biases in the redshifts obtained in different parts of color space. We use a technique developed for the proposed SPHEREx mission [@Dore14; @Stickley16] to predict the spectrum of galaxies based on their broadband photometry. In brief, this method fits a set of templates based on the libraries of @Brown14 (for galaxies) and @Salvato09 (for AGN) to deep multiband photometry. Based on the analysis in @Stickley16, we can estimate the continuum to within 20% and the emission line strengths to within a factor of two. We then use the instrumental response curves for each telescope and instrument to estimate the required integration time to obtain a redshift to that galaxy, given its estimated photometric redshift. Primary objects for a mask are those expected to yield a redshift within a factor of two of the intended mask integration time. The time estimates were compared with previous observations to verify their accuracy. As described in §4.2, we use a flagging scheme to keep track of objects for which a redshift was expected but not obtained. These sources can then be prioritized for additional observations. [lllcl]{} 2015 Dec 15 & N01-D & DEIMOS & 4 & clear, 065 seeing\ 2016 Feb 28 & N02-M & MOSFIRE & 6 & clear, 05-065 seeing\ 2016 Feb 29 & N03-D & DEIMOS & 4 & clear, 065 seeing; moon\ 2016 Mar 01 & N04-D & DEIMOS & 7 & clear, 10 seeing; moon\ 2016 Apr 09 & N05-L & LRIS & 4 & thin cirrus, 097 seeing \[table:obsevations\] Observations and Data Reductions ================================ Five nights were allocated by Caltech in the 2016A semester: three nights on DEIMOS [@Faber03], and one night each with LRIS [@Oke95] and MOSFIRE [@McLean12]. Tables 1 and 2 summarize the nights and observed slitmasks. All five nights had excellent observing conditions. Here we describe the observations and data reduction. DEIMOS ------ DEIMOS observations were conducted using the 600 groove mm$^{-1}$ grating blazed at 7200 Å and the GG400 blocking filter, with dithering performed to improve sky subtraction. In the initial observing run, we experimented with minimum slit lengths of both 6$''$ and 10$''$, with no significant difference in redshift success rate. In the subsequent DEIMOS observations we settled on a minimum slit width of 8$''$ as a balance between getting the most targets possible on the mask and getting good sky measurements. Data were reduced using a modified version of the DEEP2 pipeline designed to deal with dithered data. LRIS ---- We used the 400 groove mm$^{-1}$ blue grism blazed at 3400 Å and the 400 groove mm$^{-1}$ red grating blazed at 8500 Å, with the D560 dichroic. Our choice of blue grism gives high sensitivity at bluer wavelengths where identifying features are likely to be found for objects with photometric redshifts of $z\sim1.5-3$, while the red coverage allows for the detection of for some sources out to $z\sim1.6$. The LRIS spectra were reduced using the IRAF-based BOGUS software developed by D. Stern, S. A. Stanford, and A. Bunker, and flux calibrated using observations of standard stars from @Massey90 observed on the same night using the same instrument configuration. MOSFIRE ------- MOSFIRE was used in its default configuration. For instrumental details we refer the reader to @Steidel14. We observed four masks in $Y$ band and two in $K$ band, using integration times of 180s with ABAB dithering to improve sky subtraction. Reductions were performed with the MOSFIRE Data Reduction Pipeline (DRP) made available by the instrument team[^3]. Redshift Determination ====================== Each observed source was assessed independently by two co-authors to determine the redshift and associated quality flag. These results were then compared for conflicts in either redshift or quality flag. Conflicts were reconciled through a joint review of the spectra, usually with the help of a third, independent reviewer. As a final step in the process, we investigated all Q=4 (highest quality, see §4.1) sources for which the spectroscopic redshift ($z_s$) was highly discrepant from the expected photometric redshift ($z_p$, defined as the median photometric redshift of sources in the relevant SOM cell). Specifically, we investigated all sources with $\lvert z_{p} - z_{s} \rvert / (1 + z_{s}) \geq 0.15$. For most of these outliers, the spectroscopic redshift was deemed solid and we discuss the nature of the discrepancy in more detail in § 5.2. However, for two cases, this step caused us to modify the final redshift assessment. One of these final modifications was due to confusing a target and very close ($\sim 1\arcsec$ separation) serendipitous source, while the other modification was due to a genuine error in line identification aggravated by incomplete sky-line subtraction mimicking a corroborating emission line. Quality Flags -------------- The redshift flagging scheme we use is similar to that adopted by the zCOSMOS [@Lilly07], DEEP2 [@Newman13], and VUDS surveys [@LeFevre15]. The quality flags range from 0-4 with 4 indicating the highest confidence redshift and 0 indicating that no redshift could be found. The interpretation of the flags is roughly as follows: [lcccccc]{} 16A-D01 / UDS-m1n1 & N01-D & 2:17:27.0 & $-$5:15:07 & 90.0 & 2$\times$1800 & 86 / 51 / 5\ 16A-D02 / UDS-m3n1 & N01-D & 2:17:27.0 & $-$5:14:07 & 90.0 & 2$\times$1800 & 100 / 66 / 5\ 16A-D03 / COSMOS-m3n1 & N01-D & 10:00:22.0 & $+$2:20:00 & 90.0 & 4$\times$1800 & 104 / 72 / 7\ 16A-D04 / COSMOS-m4n1 & N01-D & 10:00:22.0 & $+$2:35:00 & 90.0 & 4$\times$1800 & 70 / 60 / 19\ 16A-M05 / COSMOS-m1-Y & N02-M & 10:00:57.2 & $+$1:48:40 & 85.0 & 20$\times$180 & 24 / 8 / 1\ 16A-M06 / COSMOS-m2-Y & N02-M & 10:00:54.4 & $+$2:01:47 & 55.0 & 20$\times$180 & 29 / 3 / 0\ 16A-M07 / COSMOS-m3-Y & N02-M & 10:00:57.7 & $+$2:14:38 & 40.0 & 20$\times$180 & 24 / 5 / 1\ 16A-M08 / COSMOS-m1-K & N02-M & 10:00:10.5 & $+$2:14:20 & 30.0 & 20$\times$180 & 12 / 5 / 0\ 16A-M09 / COSMOS-m4-Y & N02-M & 10:00:14.2 & $+$2:03:34 & 30.0 & 16$\times$180 & 25 / 6 / 1\ 16A-M10 / EGS-m1-K & N02-M & 14:17:57.4 & $+$52:35:51 & 25.0 & 22$\times$180 & 23 / 3 / 0\ 16A-D11 / COSMOS-m1n2 & N03-D & 9:58:43.2 & $+$1:42:00 & 90.0 & 3$\times$1200 & 93 / 62 / 2\ 16A-D12 / COSMOS-m8n2 & N03-D & 9:58:43.3 & $+$2:12:47 & 90.0 & 6$\times$1200 & 92 / 56 / 17\ 16A-D13 / COSMOS-m2n2 & N03-D & 9:58:43.2 & $+$1:46:15 & 90.0 & 3$\times$1200 & 91 / 77 / 7\ 16A-D14 / COSMOS-m9n2 & N03-D & 9:58:43.2 & $+$2:17:00 & 90.0 & 6$\times$1200 & 99 / 64 / 3\ 16A-D16 / COSMOS-m3n2 & N04-D & 9:58:43.2 & $+$1:50:24 & 90.0 & 3$\times$1200 & 89 / 43 / 4\ 16A-D17 / COSMOS-m4n2 & N04-D & 9:58:43.2 & $+$1:54:36 & 90.0 & 3$\times$1200 & 95 / 70 / 4\ 16A-D18 / COSMOS-m7n2 & N04-D & 9:58:43.2 & $+$2:08:15 & 90.0 & 6$\times$1200 & 91 / 52 / 13\ 16A-D19 / COSMOS-m6n2 & N04-D & 9:58:43.2 & $+$2:04:16 & 90.0 & 4$\times$1200 & 94 / 72 / 9\ 16A-D20 / COSMOS-m5n2 & N04-D & 9:58:43.2 & $+$1:58:48 & 90.0 & 3$\times$1200 & 98 / 80 / 3\ 16A-D21 / EGS-m1n2 & N04-D & 14:18:00.0 & $+$52:33:00 & 90.0 & 3$\times$1200 & 100 / 62 / 10\ 16A-D22 / EGS-m2n2 & N04-D & 14:18:00.0 & $+$52:41:24 & 90.0 & 3$\times$1200 & 104 / 72 / 3\ 16A-L23 / COSMOS-m1n5 & N05-L & 9:59:44.1 & $+$2:36:12 & $-$60.0 & 4$\times$1200 & 25 / 3 / 1\ 16A-L24 / COSMOS-m3n5 & N05-L & 9:58:58.7 & $+$2:45:56 & $-$110.0 & 2$\times$1200 & 18 / 8 / 0\ 16A-L25 / EGS-m1n5 & N05-L & 14:19:08.6 & $+$52:28:48 & 0.0 & 5$\times$1200 & 28 / 11 / 0\ 16A-L26 / EGS-m2n5 & N05-L & 14:18:04.8 & $+$52:42:01 & 0.0 & 5$\times$1200 & 26 / 4 / 1 \[table:slitmasks\] - [Q=4: A quality flag of 4 indicates an unambiguous redshift identified with multiple features or the presence of the split doublet.]{} - [Q=3.5: A quality flag of 3.5 indicates a high-confidence redshift based on a single line, with a very remote possibility of an incorrect identification. An example might be a strong, isolated emission line identified as H$\alpha$, where other identifications of the line are highly improbable due to the lack of associated lines or continuum breaks. This flag is typically only adopted for LRIS and MOSFIRE spectra.]{} - [Q=3: A quality flag of 3 indicates a high-confidence redshift with a low probability of an incorrect identification. An example might be the low signal-to-noise ratio detection of an emission line, possibly corrupted by telluric emission or absorption, identified as , but where the data quality is insufficient to clearly resolve the doublet.]{} - [Q=2/1: A quality flag of 2 indicates a reasonable guess, while a quality flag of 1 indicates a highly uncertain guess. Sources with these low confidence redshifts are not included in the data release.]{} - [Q=0: A quality flag of 0 indicates that no redshift could be identified. As described next, a code indicating the cause of the redshift failure is assigned in place of the redshift.]{} -- -- -- -- Figure \[figure:example\_spectra\] shows C3R2 spectra from 2016A as examples of Q=3, Q=3.5, and Q=4 redshift assignments. Failure Codes ------------- It is important for C3R2 to track redshift failures, as well as the reasons for the failures, in order to avoid systematic biases in the sources selected for calibration. Failed targets that were expected to yield a redshift given the instrument and exposure time can be prioritized for additional follow-up. On the other hand, if no spectroscopic redshift was obtained because of a problem with the observing conditions or data (i.e., bad rows, or the target ended up in a region between two detector arrays), no additional prioritization of that source may be needed. With these considerations in mind, we developed a system to flag different “failure modes” for objects not yielding a redshift. Four categories of failures are used, with the corresponding code assigned in place of a redshift in our catalog. The failure modes we identify are: 1. [Code = $-$91: Object too faint to identify the redshift. Indicates that a deeper exposure and/or different instrument and/or different wavelength coverage is required to obtain a secure redshift. An example of such a source might be a galaxy expected to have a strong emission line at 9800 Å, but where the slit placement caused the wavelength coverage to end at 9500 Å, yielding a continuum detection without any strong spectroscopic features. These sources will be further prioritized in future observations.]{} 2. [Code = $-$92: Object well-detected, but no redshift could be determined. May require a different instrument for secure redshift determination due to an incorrect photometric redshift or the wavelength coverage obtained for a given observation. We emphasize that the dividing line between $-$91 and $-$92 failure codes is imprecise, and no strong effort was made to homogenize the classification. Fundamentally, both codes can be considered as two aspects of the same issue. Again, these sources will increase in priority going forward.]{} 3. [Code = $-$93: Corrupted slit, typically due to bad rows/columns in the data or the source falling on or near detector chip gaps. Does not affect object priority in future observations.]{} 4. [Code = $-$94: Missing slit, as an extreme case of code $-$93. Does not affect object priority in future observations.]{} [Failure codes $-$91 and $-$92 essentially correspond to spectral quality issues (i.e., signal-to-noise ratio, wavelength range), while codes $-$93 and $-$94 correspond to data quality issues (i.e., slitmask design issues, detector issues). While in DR1 we distinguished between these four failure modes, for many analysis purposes, considering just the two general categories will be sufficient.]{} Failure code $-$91, the most common failure code, generally indicates that the signal-to-noise ratio of the data was insufficient for redshift determination. Indeed, considering the 131 DEIMOS-observed sources in COSMOS with this failure code, 122 (93%) were anticipated to fail based on our estimated exposure time needed to get a redshift. As with low-confidence redshifts, sources for which we failed to find a redshift are not included in this data release. [@c@]{}\ \[figure:specz\_somz\] Serendipitous Sources --------------------- We measured the redshifts of 134 serendipitously detected sources that happened to fall in slits with primary C3R2 targets. The coordinates of these sources were identified and they were matched back to the survey catalogs. The redshifts for these sources are included in our published catalog. Literature sources ------------------ Some (unintentional) overlap with literature redshifts allows a check on our results. In COSMOS and EGS we observed 38 sources that have previously existing high-quality redshifts. Most (24) were serendipitous detections. We find an RMS discrepancy between our redshifts and the literature values of $4\times10^{-4}$. C3R2 redshifts are often higher precision than the literature values, which likely explains this small difference. There is no systematic difference between the C3R2 and literature redshifts. Redshift results and Calibration progress ========================================= A total of 1825 sources were targeted in the 2016A observations. We identified 1131 Q=4 redshifts, 27 Q=3.5 redshifts, and 125 Q=3 redshifts. In principle, only the highest confidence redshifts would be used for calibration for cosmology; whether this restricts usable sources to those with Q=4 is worth investigating. Another 99 spectra yielded redshifts of low confidence (Q=1/2), while there were 443 failures. Of these, 409 were failure code $-91$ or $-92$, indicating that the source was too faint or lacking in identifying features, while 34 were code $-93$ or $-94$, indicating a corrupted or missing slit. In terms of the SOM presented in M15, and using only the C3R2 sources observed in COSMOS with Q $\geq$ 3 for this analysis (911 redshifts), we have increased color space coverage by 5.4%. Figure \[figure:mag\_and\_z\_dist\] shows the i-band magnitude distribution and redshift distribution of the 2016A “gold" sample of Q=4 sources as well as the Q=3/3.5 sources. The distributions are very similar to the overall distribution of the unsampled cells of galaxy color space identified in M15, indicating that we are targeting the correct sources. SOM-based photo-z performance ----------------------------- The self-organizing map colored by the median photo-z of sources per cell (the left panel of Figure \[figure:colors\]) effectively defines a photometric redshift estimate for each galaxy based on its position in the ugrizYJH color space of Euclid/WFIRST. Figure \[figure:specz\_somz\] compares our Q=4 spectroscopic redshifts with the redshift that would be inferred based on the SOM, with encouraging results. The normalized median absolute deviation (a dispersion measure which is not sensitive to catastrophic outliers [@Ilbert09; @Dahlen13]) defined as $$\sigma_{\mathrm{NMAD}} = 1.48\times \mathrm{median}\Big(\frac{\lvert z_{p}-z_{s} \rvert}{1+z_{s}}\Big)$$ is 0.027 (2.7%) for the sample, which is quite low. Using the standard definition of catastrophic photo-z outliers as those with $\lvert z_{p} - z_{s} \rvert/(1+z_{s}) \geq 0.15$, we measure a low outlier fraction of 3.8%. The measured bias, defined as $$\mathrm{mean}\Big(\frac{ z_{p}-z_{s}}{1+z_{s}}\Big)$$ is $\lesssim$0.1% after removing the catastrophic outliers. Further improvements to these results will result from folding in all spectroscopic information from C3R2 and other surveys to the relation encoded by the SOM. Notably, these results are already competitive with or better than the photo-z results of codes tested in @Dahlen13, where the photometry used comprised 14 bands including full depth CANDELS and Spitzer data. -- -- -- -- -- -- \[figure:som\_update\] While the performance we find is quite good, and may be representative of what can be achieved with a survey such as Euclid or WFIRST, the results depend on the depth and stability of the photometry. The photometry used to place objects on the SOM in order to estimate a photo-z in the above analysis is quite deep ($i$-band depth $\sim$25.4 AB). The results will degrade as the photometry gets shallower or bands are lost in a manner that can be directly characterized via the SOM. A detailed study of the expected performance from the SOM-based photo-z approach will be the subject of a future paper. Outliers -------- Out of 1079 sources with Q=4 redshifts and reliable SOM-based photo-z estimates, only 41 (3.8%) are outliers according to the standard definition, $\lvert z_{p} - z_{s} \rvert/(1+z_{s}) \geq 0.15$. If, instead of the SOM-based photo-z, we use the photo-z for each object based on deep multiband data (e.g., the 30-band COSMOS data), we find an outlier fraction for the same sources of $\sim$3.1%. Thus the SOM photo-z (effectively based only on the seven color Euclid-like SEDs) performs nearly as well in terms of outlier fraction. We have analyzed all of the outliers on a case-by-case basis. The majority (24/41; 59%) have individual (rather than SOM-based) photo-z estimates more in line with the measured redshift, indicating that the color cells they belong to have real redshift scatter. For nearly all of these sources, the measured dispersion in the 30-band photo-z’s within the relevant color cell is significantly larger than the median redshift dispersion per cell; in other words, these are sources that fall in more degenerate regions of the color space. The SOM can be used to identify these regions in a consistent way in order to either reject them in weak lensing analysis or direct extra spectroscopy at them to characterize the redshift distribution in those cells. In addition, there are several other examples easily understood as Galactic stars (3) or obvious quasars/active galaxies (2), which are known not to have typical galaxy colors (total = 5/41; 12%). This process caught one mistaken line identification where our initial assessment of a MOSFIRE spectrum identified an isolated, narrow, strong line as \[\] $\lambda 5007$ with corroborating \[\] $\lambda 4959$ emission. Subsequent analysis reveals the latter emission line to be due to poorly subtracted telluric emission, and we now identify the strong emission line as H$\alpha$ (Q=3.5). The remaining cases seem to be genuine mismatches between the spectroscopic redshift and the photometric redshifts, both the individual photometric redshift of the galaxy and the SOM-based photometric redshift. Consideration of [Hubble]{} imaging reveals at least some of these likely due to two close-separation galaxies, where the ground-based imaging used for the photometry was unable to separate the sources. Increased color space coverage ------------------------------ The five nights of observing in 2016A filled in $\sim$6% of the map, in addition to existing spectroscopy which already filled $\sim$50% (see Figure \[figure:som\_update\]). Thus we completed $\gtrsim$10% of the required calibration. However, some of the remaining observations may prove more challenging. Given the recent progress on bringing multiple deep fields (and their spectroscopy) onto a consistent color system, the requirements may also change to some extent, in the sense that somewhat fewer spectra are required due to the inclusion of other spectroscopic surveys. It should also be noted that a certain fraction of the remaining cells represent faint, red sources for which spectroscopic redshifts are prohibitively difficult to obtain with current instruments. These constitute a small ($\sim$3%) fraction of the unsampled cells. If needed, the SOM provides a consistent method of identifying such objects and removing them from the weak lensing sample. [lccccccc]{} UDS-3583 & 02:17:30.65 & -05:15:24.4 & UDS-m1n1 & 001 & 23.4 & 0.7877 & 4\ UDS-10246 & 02:17:17.55 & -05:13:06.9 & UDS-m1n1 & 002 & 23.9 & 0.8028 & 4\ UDS-767 & 02:17:59.05 & -05:16:21.2 & UDS-m1n1 & 003 & 25.0 & 0.5558 & 4\ UDS-7109 & 02:17:00.35 & -05:14:15.4 & UDS-m1n1 & 004 & 23.6 & 1.0314 & 3\ UDS-2276 & 02:17:52.83 & -05:15:55.2 & UDS-m1n1 & 005 & 22.9 & 0.9388 & 4\ UDS-8536 & 02:17:53.37 & -05:13:40.3 & UDS-m1n1 & 006 & 24.7 & 0.8619 & 4\ UDS-9784 & 02:17:56.80 & -05:13:15.9 & UDS-m1n1 & 009 & 23.6 & 0.8533 & 4\ UDS-10739 & 02:17:44.88 & -05:12:58.7 & UDS-m1n1 & 010 & 23.2 & 1.0594 & 4\ UDS-9730 & 02:17:32.64 & -05:13:17.4 & UDS-m1n1 & 012 & 23.8 & 1.0949 & 4\ UDS-12725 & 02:17:14.84 & -05:12:19.8 & UDS-m1n1 & 013 & 23.7 & 1.0351 & 4\ $\cdots$ \[table:spectroscopy\] Conclusion ========== We have presented initial results of the C3R2 survey based on five nights of Keck spectroscopy in the 2016A semester. C3R2 is designed to supplement extensive existing spectroscopy in order to provide a spectroscopic sample spanning the observed colors of galaxies to the Euclid weak lensing photometric depth. The ultimate aim of the survey is to calibrate the color-redshift relation sufficiently to meet the requirements set by weak lensing cosmology. We estimate that the survey would require $\sim$40 Keck nights (or their equivalent) in total to meet the requirements set by Euclid. In future papers we will present the updated survey strategy based on bringing multiple Euclid calibration fields onto a consistent color system, as well as realistic tests of the performance of the method. Initial tests show that the empirical color mapping technique performs very well in reproducing $N(z)$ distributions with low bias. Additional data, including results from 16.5 nights allocated in 2016B as well as time allocated in 2017A and 2017B will be presented in follow-on papers. Combined with data from VLT, GTC, and MMT, we expect the calibration samples will be sufficient to meet the needs of Euclid. Work is ongoing to understand the needs for WFIRST calibration, but these spectra will form part of the foundation of that survey as well. Further tests and refinements of the calibration method, as well as studies to determine the optimal way to incorporate all existing spectroscopic and photometric information from deep fields into photo-z estimation using a limited set of broad band observations, are avenues of continuing research. ©2017. All rights reserved. [^1]: The SOM algorithm was used mainly for its relative simplicity and visualization power; however, any technique that manages to quantify the density of galaxies in multicolor space would be equally appropriate. A number of other techniques for nonlinear dimensionality reduction (e.g., the generative topographic map, growing neural gas, and local linear embedding) may, in principle, offer some advantages over the SOM. [^2]: We have since substantially lessened the extent to which we weight by cell occupation, because it is effectively accomplished already by the source density on the sky - i.e., more common sources will find their way onto masks more frequently. [^3]: https://keck-datareductionpipelines.github.io/MosfireDRP/	{ "pile_set_name": "ArXiv" }
--- abstract: \| Strong asymptotics on the whole complex plane of a sequence of monic Jacobi polynomials $P_n^{(\alpha _n, \beta _n)}$ is studied, assuming that $$\lim_{n\to\infty} \frac{\alpha_n}{n}=A\,, \qquad \lim_{n\to\infty} \frac{\beta _n}{n}=B\,,$$ with $A$ and $B$ satisfying $ A > -1$, $ B>-1$, $A+B < -1$. The asymptotic analysis is based on the non-Hermitian orthogonality of these polynomials, and uses the Deift/Zhou steepest descent analysis for matrix Riemann-Hilbert problems. As a corollary, asymptotic zero behavior is derived. We show that in a generic case the zeros distribute on the set of critical trajectories $\Gamma$ of a certain quadratic differential according to the equilibrium measure on $\Gamma$ in an external field. However, when either $\alpha _n$, $\beta _n$ or $\alpha _n+\beta _n$ are geometrically close to ${{\mathbb{Z}}}$, part of the zeros accumulate along a different trajectory of the same quadratic differential. author: - 'A.B.J. Kuijlaars[^1]' - 'A.Mart[í]{}nez-Finkelshtein[^2]' date: 'January 5, 2004' title: Strong asymptotics for Jacobi polynomials with varying nonstandard parameters --- Introduction ============ We consider Jacobi polynomials $P_n^{(An, Bn)}$ with varying negative parameters $An$ and $Bn$ such that $$\label{ABinequalities} -1 < A < 0, \qquad -1 < B < 0, \qquad -2 < A + B < -1.$$ We will obtain strong asymptotics as $n \to \infty$ of $P_n^{(An, Bn)}(z)$ uniformly for $z$ in any region of the complex plane and uniformly for $A$ and $B$ in compact subsets of the set of parameter values satisfying (\[ABinequalities\]). Since the asymptotics is uniform in $A$ and $B$, we also find the asymptotics for general sequences of Jacobi polynomials $P_n^{(\alpha_n, \beta_n)}$ such that the limits $$\label{ABlimits} A = \lim_{n \to \infty} \frac{\alpha_n}{n} \qquad \mbox{and} \qquad B = \lim_{n \to \infty} \frac{\beta_n}{n}$$ exist, and satisfy (\[ABinequalities\]). From the asymptotics of the polynomials we will also be able to describe the limiting behavior of the zeros. From the point of view of behavior of zeros, the Jacobi polynomials with varying parameters $\alpha_n$, $\beta_n$ such that (\[ABlimits\]) and (\[ABinequalities\]) hold are the most interesting general case. Indeed, Mart[í]{}nez-Finkelshtein et al.[@MR2002d:33017] distinguish five cases depending on the values of the limits (\[ABlimits\]) (cf. Fig.\[fig:Jacobi\_cases\]). The first case is the case where $A, B > 0$, which corresponds to classical Jacobi polynomials with varying positive parameters. These polynomials are orthogonal on the interval $[-1,1]$, and as a result all their zeros are simple and belong to the interval $(-1,1)$. The asymptotic behavior of Jacobi polynomials with varying positive parameters is discussed in [@Bosbach99; @ChenIsmail; @DetteStudden:95; @Gawronski:91; @KuijlaarsVanAssche:99; @MoakSaffVarga]. We also consider the parameter combinations $B > 0$, $A + B < -2$ and $A > 0$, $A + B < -2$ as classical. Indeed, the transformation formula $$\label{transformation1} P_n^{(\alpha, \beta)}(x) = \left(\frac{1-x}{2}\right)^n P_n^{(-2n-\alpha-\beta-1, \beta)} \left(\frac{x+3}{x-1}\right)$$ see [@szego:1975 §4 .22], expresses a Jacobi polynomial with parameters $\alpha$ and $\beta$ satisfying $\alpha + \beta < -2n$ and $\beta > -1$ directly in terms of a Jacobi polynomial with positive parameters. It follows that (\[transformation1\]) reduces the study of Jacobi polynomials with varying parameters $\alpha_n$ and $\beta_n$ such that the limits (\[ABlimits\]) hold with $B > 0$ and $A + B < -2$ to the study of Jacobi polynomials with varying positive parameters. The analogous formula $$\label{transformation2} P_n^{(\alpha, \beta)}(x) = \left(\frac{1+x}{2}\right)^n P_n^{(\alpha, -2n-\alpha-\beta-1)} \left(\frac{3-x}{x+1}\right)$$ shows similarly how to reduce the case $A > 0$ and $A+B<-2$ to the classical case. \[fig:Jacobi\_cases\] ![The five different cases in the classification of Jacobi polynomials with varying parameters according to [@MR2002d:33017].](Jacobi_cases "fig:") The second case in the classification of [@MR2002d:33017] corresponds to limits $A$ and $B$ in (\[ABlimits\]) satisfying one of the three combinations $A < -1$, $A + B > -1$, or $B < -1$, $A + B > -1$, or $A < -1$, $B < -1$. In this case the zeros accumulate along an open arc in the complex plane. Their asymptotic distribution was found in [@MR2002d:33017] in terms of the equilibrium measure in an external field (cf.[@Saff:97]). The approach followed there was based on the non-hermitian orthogonality of the Jacobi polynomials with these parameters. See [@Kuijlaars03] for an overview of non-hermitian orthogonality properties of Jacobi polynomials with general parameters. The remaining cases correspond to combinations of $A$ and $B$ values such that one or more of the inequalities $-1 < A < 0$, $-1 < B< 0$, and $-2 < A+B < -1$ are satisfied. In these cases, the zero behavior is more involved due to the possible occurrence of multiple zeros (at $\pm 1$ only) or a zero at $\infty$ (which means a degree reduction). To be precise, if $\alpha = -k$ is a negative integer with $k \in \{1, \ldots, n\}$, then we have (see [@szego:1975 formula (4.22.2)]), $$P_{n}^{\left( -k,\beta \right) }(z) = \frac{\Gamma(n+\beta+1) }{\Gamma(n+\beta +1-k)}\, \frac{(n-k)!}{n!} \left( \frac{z-1}{2}\right) ^{k}P_{n-k}^{\left( k,\beta \right) }\left( z\right), \label{integer 1}$$ so that $P_n^{(-k,\beta)}$ has a zero at $1$ of multiplicity $k$. Similarly, if $\beta = -l$ with $l \in \{1, \ldots, n\}$ then $P_n^{(\alpha, -l)}$ has a zero at $-1$ of multiplicity $l$. A degree reduction may occur when $\alpha + \beta$ is a negative integer, namely if $\alpha +\beta =-n -k-1$ with $k \in \{0, \ldots,n-1\}$, then $$P_{n}^{\left( \alpha ,\beta \right) }\left( z\right) =\frac{\Gamma(n+\alpha+1) }{\Gamma(k+\alpha+1)} \, \frac{k!}{n!}\, P_{k}^{\left( \alpha ,\beta \right) }\left( z\right), \label{integer 3}$$ see [@szego:1975 Eq. (4.22.3)]; see §4.22 of [@szego:1975] for a more detailed discussion. Now assume we have varying parameters $\alpha_n$, $\beta_n$ such that the limits (\[ABlimits\]) exist. If $-1 < A < 0$, and if the $\alpha_n$ are integers, then we have for each $n$ large enough, that $P_n^{(\alpha_n,\beta_n)}$ has a multiple zero at $1$. In the weak limit of the zero counting measures this corresponds to a point mass $\|A\|$ at $1$. Similarly, if $-1 < B < 0$, and if the $\beta_n$ are integers, then we have in the limit a point mass $\|B\|$ at $-1$. Finally, if $-2 < A+B< -1$ and $\alpha_n + \beta_n$ are integers, then we have in the limit a point mass $2+A+B$ at infinity. The classification of the remaining cases in [@MR2002d:33017] depends on the number of inequalities $-1 < A<0$, $-1 < B< 0$, $-2 < A+B < -1$ that are satisfied. The third, fourth and fifth case correspond to combination of parameters $A$ and $B$ such that exactly one, exactly two, or exactly three, respectively, of the inequalities are satisfied (cf. Fig. \[fig:Jacobi\_cases\]). In these three cases the limiting behavior of zeros will be very sensitive to the proximity of $\alpha_n$ (if $-1 < A < 0$), $\beta_n$ (if $-1 < B < 0$) or $\alpha_n + \beta_n$ (if $-2 < A +B < -1$) to integer values. For Laguerre polynomials the same phenomenon was analyzed recently in [@Kuijlaars/Mclaughlin:01b]. Since all three kinds of singular behavior can occur in the fifth case, this is the most interesting case and that is the reason why we consider it here. The other cases can also be treated with our methods. Fig. \[fig:Jacobi\_cases\] shows the behavior of zeros which is typical for case 5. From the figure it appears that the zeros accumulate on a contour consisting of three analytic arcs. From our analysis below it follows that this is indeed the case, provided that the parameters are not too close to integers. We identify the curves as trajectories of a quadratic differential as well as the limiting density of the zeros on the curves, see Theorems \[theoremweak1\] and \[theoremweak2\] for the exact statement. To be able to explain the remarkable zero behavior was the main motivation for the present work. ![Zeros of $P_{100}^{(\alpha ,\beta )}$ for $\alpha = -70 + 10^{-5}$, $\beta = -80 + 10^{-5}$, which corresponds to Case 5 in Fig.\[fig:Jacobi\_cases\]. []{data-label="fig_zeros_intro"}](zeros_intro) We remark that the different possibilities within the cases 1, 2, 3, and 4 can be transformed to one another using the transformation formulas (\[transformation1\]), (\[transformation2\]) for Jacobi polynomials. It is interesting to note that case 5 is invariant under these transformations, see [@MR2002d:33017]. We also remark that the transitions between the five cases (i.e., $A=0$, $A=-1$, or $B=0$, $B=-1$, $A+B = -1$ or $A+B=-2$) will present additional difficulties. These are the non-general cases, in contrast to what we call the general cases 1–5. The zero distribution in some of these cases has been studied by Driver, Duren and collaborators (see also a recent survey [@temme03] on the large parameter cases of the hypergeometric function). In [@Driver/Duren99] the case $P_n^{(k n+1 , -n-1 )}$, $k \in {{\mathbb{N}}}$, has been analyzed, corresponding to $A=k\in {{\mathbb{N}}}$ and $B=-1$; this result was generalized in [@Duren01] using a saddle-point method to allow $k$ to be any positive real number. Case $P_n^{(n+b , -n-b )}$ has been studied in [@Driver/Moler01]. In general, these works establish the accumulation set of the zeros but not the limiting distribution. Trajectories of the zeros of the Gegenbauer polynomials $P_n^{(-n-b, -n-b)}$ with fixed $n$ as $b$ varies from $-1/2$ to $-\infty$ have been described in [@Driver/Duren01a]. The rest of the paper is organized as follows. The main results are stated in Section 2. We start defining the basic configuration on the plane used in the description of the zero (Subsection 2.2) and strong (Subsections 2.3–2.4) asymptotics of the polynomials. In Section 3 we prove two technical lemmas. The cornerstone of our approach is the matrix Riemann-Hilbert problem formulated in Section 4; the transformations of this problem in the framework of the Deift-Zhou steepest descent analysis (Section 5) are used in Section 6 to prove the main results of the paper. The Deift-Zhou steepest descent method for asymptotics of Riemann-Hilbert problems was introduced in [@deift/zhou] and applied first to orthogonal polynomials in [@MR2001f:42037; @MR2001g:42050], see also [@MR2000g:47048]. We use an adaptation of the method to orthogonality on curves in the complex plane. The optimal curves are trajectories of a quadratic differential and they were used for steepest descent analysis of Riemann-Hilbert problems first in [@baik01] and later in [@aptekarev02; @Kamvissis03; @Kuijlaars/Mclaughlin:01b; @Kuijlaars/Mclaughlin:01a; @Kuijlaars03b]. Statement of results ==================== Geometry of the problem ----------------------- We assume $A$ and $B$ satisfy the inequalities (\[ABinequalities\]) and define $$\label{zetapm} \zeta_{\pm} = \frac{B^2 - A^2 \pm 4i \sqrt{(A+1)(B+1)(-A-B-1)}} {(A+B+2)^2}.$$ Because of the inequalities (\[ABinequalities\]) we have that all factors in the square root in (\[zetapm\]) are positive, so that $\zeta_+ \in {{\mathbb{C}}}^+=\{z\in {{\mathbb{C}}}:\, {\mathop{\rm Im}}z > 0\}$ and $\zeta_-$ is the complex conjugate of $\zeta_+$. Regardless of the branch of the square root and of the path of integration we choose, the set $$\label{defGamma} \Gamma =\Gamma^{(A,B)}:= \left\{ z \in \mathbb C :\, {\mathop{\rm Re}}\int_{\zeta_-}^z \frac{((t-\zeta_+)(t-\zeta_-))^{1/2}}{t^2-1} \, dt = 0 \right\}$$ is well defined, and consists of the union of the critical trajectories of the quadratic differential (cf. [@Strebel84]) $$\label{quaddiff} -\frac{(z-\zeta_-)(z-\zeta_+)}{(z^2-1)^2} \, dz^2.$$ \[lemma1\] We have that $\Gamma$ is the union of three analytic arcs, which we denote by $\Gamma_L$, $\Gamma_C$, and $\Gamma_R$. All three arcs connect the two points $\zeta_{\pm}$ and intersect the real line in exactly one point, in such a way that each of the intervals $(-\infty,-1)$, $(-1,1)$, $(1,\infty)$ is cut by one of the arcs. The contour $\Gamma$ is oriented as indicated in Fig. \[fig\_Gamma1\]. That is, $\Gamma_L$ and $\Gamma_C$ are oriented from $\zeta_+$ to $\zeta_-$, and $\Gamma_R$ is oriented from $\zeta_-$ to $\zeta_+$. The orientation of $\Gamma$ induces a $+$ and $-$ side in a neighborhood of the contour, where the $+$ side is on the left while traversing $\Gamma$ according to its orientation and the $-$ side is on the right. We say that a function $f$ on ${{\mathbb{C}}}\setminus \Gamma$ has a boundary value $f_+(t)$ for $t \in \Gamma \setminus \{ \zeta_+, \zeta_- \}$ if the non-tangential limit of $f(z)$ as $z \to t$ with $z$ on the $+$ side of $\Gamma $ exists; similarly for $f_-(t)$. ![Contour $\Gamma = \Gamma_L \cup \Gamma_C \cup \Gamma_R$ with the orientation chosen. []{data-label="fig_Gamma1"}](Jac_fig33_setN_orient) Also, we denote by $\Omega_{-1}$ and $\Omega_{1}$ the bounded components of ${{\mathbb{C}}}\setminus \Gamma$ containing $-1$ and $1$, respectively, and by $\Omega_\infty$ the unbounded component of ${{\mathbb{C}}}\setminus \Gamma$ (Fig. \[fig\_Gamma1\]). In what follows we write $$\label{defRz} R(z) = ((z-\zeta_+)(z-\zeta_-))^{1/2}, \qquad z \in \mathbb C \setminus \Gamma_C,$$ which is defined and analytic in the cut plane $\mathbb C \setminus \Gamma_C$, such that $R(z) \sim z$ as $z \to \infty$. We also need the critical orthogonal trajectories of the quadratic differential (\[quaddiff\]). These are defined by $$\Gamma^{\perp} = \Gamma^{\perp+} \cup \Gamma^{\perp-}$$ where $$\label{gammaperp1} \Gamma^{\perp-} = \left\{ z \in \mathbb C^- :\, {\mathop{\rm Im}}\int_{\zeta_-}^z \frac{R(t)}{t^2-1} dt = 0 \right\}$$ where the integration is along a path from $\zeta_-$ to $z$ in $\mathbb C^- \setminus \Gamma_C$, and $$\label{gammaperp2} \Gamma^{\perp+} = \left\{ z \in \mathbb C^+ :\, {\mathop{\rm Im}}\int_{\zeta_+}^z \frac{R(t)}{t^2-1} dt = 0 \right\}$$ where the integration is along a path from $\zeta_+$ to $z$ in $\mathbb C^+ \setminus \Gamma_C$. ![Typical structure of the sets $\Gamma$ (solid lines) and $\Gamma^{\perp}$ (dotted lines). []{data-label="fig_trajectories1"}](Jac_fig31_setN) The typical structure of the orthogonal trajectories $\Gamma^{\perp}$ is shown in Fig. \[fig\_trajectories1\]. Three orthogonal trajectories emanate from both $\zeta_+$ and $\zeta_-$, ending at $1$, $-1$ and $\infty$, respectively (see the dotted lines in Fig. \[fig\_trajectories1\]). We denote by $\gamma_1^+$, $\gamma_{-1}^+$, $\gamma_{\infty}^+$ the arcs of $\Gamma^{\perp}$ that connect $\zeta_+$ with the points $1$, $-1$, and $\infty$, respectively; this is also the part of $\Gamma^{\perp}$ in the upper half plane. The corresponding arcs in the lower half plane are denoted by $\gamma_1^-$, $\gamma_{-1}^-$ and $\gamma_{\infty}^-$, so that $\gamma_s^-$ is the mirror image of $\gamma_s^{1}$ in the real axis, for $s \in \{1,-1,\infty\}$. Weak convergence of zeros ------------------------- Then we define the absolutely continuous (a priori, complex) measure $\mu$ on $\Gamma$ by $$\label{defmu} d\mu(z) = \frac{A + B+2}{2 \pi i} \frac{R_+(z)}{z^2-1} \,dz, \qquad z\in \Gamma,$$ where $R_+$ denotes the boundary value of $R$ on the $+$-side of $\Gamma$. (Only on $\Gamma_C$ there is a difference between the $+$ and $-$ boundary values.) The line element $dz$ is taken according to the orientation of $\Gamma$. \[lemma2\] The measure [(\[defmu\])]{} is positive and $$\label{measures_arcs1} \mu(\Gamma_L)=1+A>0, \qquad \mu(\Gamma_C)=-1-A-B>0, \qquad \mu(\Gamma_R)=1+B>0.$$ In particular we have that $\mu$ is a probability measure on $\Gamma$. The importance of $\mu$ is shown in the following result. \[theoremweak1\] Let $(\alpha_n)$ and $(\beta_n)$ be two sequences such that $\alpha_n/n \to A$ and $\beta_n/n \to B$ where $A$ and $B$ satisfy [(\[ABinequalities\])]{}. Suppose that $$\label{notclosetoZ} \lim_{n \to \infty} \left[ {\mathop{\rm dist}}(\alpha_n, \mathbb Z)\right]^{1/n} = \lim_{n \to \infty} \left[ {\mathop{\rm dist}}(\beta_n, \mathbb Z)\right]^{1/n} = \lim_{n \to \infty} \left[ {\mathop{\rm dist}}(\alpha_n+\beta_n, \mathbb Z)\right]^{1/n} = 1.$$ Then, as $n \to \infty$, the zeros of the Jacobi polynomial $P_n^{(\alpha_n,\beta_n)}$ accumulate on $\Gamma$ and $\mu$ is the weak$^$ limit of the sequence of normalized zero counting measures. The conditions (\[notclosetoZ\]) imply that $\alpha_n$, $\beta_n$, and $\alpha_n+\beta_n$ are not too close to the integers. That such a condition is necessary is easily seen from the case when these numbers are in fact integers (cf.(\[integer 1\])). To describe the general case, we need the function $$\label{defphi} \phi(z) = \frac{A+B+2}{2} \int_{\zeta_-}^z \frac{R(t)}{t^2-1} dt,$$ which is a multi-valued function. However, its real part is well-defined, and we see from the definition (\[defGamma\]) that $\Gamma = \{ z : {\mathop{\rm Re}}\phi (z) = 0 \}$. For every $r$ we introduce the level set $$\label{defGammar} \Gamma_r = \{ z \in \mathbb C :\, {\mathop{\rm Re}}\phi (z) = r \}.$$ We note that by the selection of the branch in (\[defRz\]), ${\mathop{\rm Re}}\phi > 0$ in the unbounded region $\Omega_{\infty}$ and ${\mathop{\rm Re}}\phi < 0$ in the two bounded regions $\Omega_{\pm 1}$. For $r > 0$, we have that $\Gamma_r$ is a simple closed contour in $\Omega_{\infty}$, while for $r < 0$, we have that $\Gamma_r$ consists of two simple closed contours, one contained in $\Omega_1$ and the other in $\Omega_{-1}$. We define for $r < 0$, $$\label{defGammarmp} \Gamma_{r,-1} = \Gamma_r \cap \Omega_{-1}, \qquad \Gamma_{r,+1} = \Gamma_r \cap \Omega_1.$$ We choose the positive (=counterclockwise) orientation on each of the closed contours. All these contours are trajectories of the quadratic differential (\[quaddiff\]). See Fig. \[trajectories\] for the trajectories. ![Some trajectories of the quadratic differential (\[quaddiff\]), or equivalently, some level sets $\Gamma_r$, for the values $A = -0.7$ and $B=-0.8$. []{data-label="trajectories"}](trajectories.eps) Finally, we introduce three numbers $r_{\alpha}$, $r_{\beta}$, and $r_{\alpha+\beta}$ and we assume that $$\begin{aligned} \label{closetoZ1} \lim_{n \to \infty} \left[ {\mathop{\rm dist}}(\alpha_n, \mathbb Z)\right]^{1/n} & = & e^{-r_{\alpha}}, \\ \label{closetoZ2} \lim_{n \to \infty} \left[ {\mathop{\rm dist}}(\beta_n, \mathbb Z)\right]^{1/n} & = & e^{-r_{\beta}}, \\ \label{closetoZ3} \lim_{n \to \infty} \left[ {\mathop{\rm dist}}(\alpha_n+\beta_n, \mathbb Z)\right]^{1/n} & = & e^{-r_{\alpha+\beta}}.\end{aligned}$$ It is easily seen that these numbers are non-negative and that the case $r_{\alpha} = r_{\beta} = r_{\alpha+\beta} = 0$ corresponds to Theorem \[theoremweak1\]. It is also easily seen that at least two of the numbers $r_{\alpha}$, $r_{\beta}$ and $r_{\alpha+\beta}$ should be equal, and if the third one is different, it will be greater than the other two. So we distinguish four cases in the next theorem. \[theoremweak2\] Let $(\alpha_n)$ and $(\beta_n)$ be two sequences such that $\alpha_n/n \to A$ and $\beta_n/n \to B$ where $A$ and $B$ satisfy [(\[ABinequalities\])]{}. Suppose that there exist three numbers $r_{\alpha}$, $r_{\beta}$, and $r_{\alpha+\beta}$ such that the limits [(\[closetoZ1\])]{}, [(\[closetoZ2\])]{}, and [(\[closetoZ3\])]{} exist. Then the following hold. 1. If $r_{\alpha} = r_{\beta} = r_{\alpha + \beta}$, then the zeros of $P_n^{(\alpha_n,\beta_n)}$ accumulate on $\Gamma$ as $n \to \infty$, and $\mu$ is the weak$^$ limit of the normalized zero counting measures. 2. If $r_{\alpha} = r_{\beta} < r_{\alpha+\beta}$, then the zeros of $P_n^{(\alpha_n,\beta_n)}$ accumulate on $\Gamma_C \cup \Gamma_r$ where $r = (r_{\alpha+\beta}-r_{\alpha})/2 >0$ and $$\frac{A+B+2}{2\pi } \frac{R_+(z)}{z^2-1} dz, \qquad z \in \Gamma_C \cup \Gamma_r$$ is the weak$^$ limit of the normalized zero counting measures. 3. If $r_{\alpha} = r_{\alpha+\beta} < r_{\beta}$, then the zeros of $P_n^{(\alpha_n,\beta_n)}$ accumulate on $\Gamma_R \cup \Gamma_{r,-1}$ where $r = (r_{\alpha}-r_{\beta})/2 < 0$, and $$\frac{A+B+2}{2\pi i} \frac{R(z)}{z^2-1} dz, \qquad z \in \Gamma_R \cup \Gamma_{r,-1}$$ is the weak$^$ limit of the normalized zero counting measures. 4. If $r_{\beta} = r_{\alpha+\beta} < r_{\alpha}$, then the zeros of $P_n^{(\alpha_n,\beta_n)}$ accumulate on $\Gamma_L \cup \Gamma_{r,+1}$ where $r = (r_{\beta}-r_{\alpha})/2 < 0$, and $$\frac{A+B+2}{2\pi i} \frac{R(z)}{z^2-1} dz, \qquad z \in \Gamma_L \cup \Gamma_{r,+1}$$ is the weak$^$ limit of the normalized zero counting measures. Of course the statement of Theorem \[theoremweak1\] is a special case of part (a) of Theorem \[theoremweak2\]. We choose to mention Theorem \[theoremweak1\] separately, since it represents the generic case. The statements of Theorem \[theoremweak2\] are also valid along subsequences of $\mathbb N$, if we assume existence of the limits (\[closetoZ1\])–(\[closetoZ3\]) as $n\to \infty$ for $n$ in a subsequence $\Lambda$ of $\mathbb N$. To illustrate the different phenomena that can happen we show some figures (Fig. \[fig\_zeros\_case1a\] and Fig. \[fig:zerosCases23a\]). ![Zeros of $P_{100}^{(\alpha ,\beta )}$ for $\alpha = -70 + 10^{-5}$, $\beta = -80 + 10^{-5}$, together with the set $\Gamma$ corresponding to $A=-0.7$, $B=-0.8$. []{data-label="fig_zeros_case1a"}](zeros_case1) -- -- -- -- A general approach to the limiting zero behavior of polynomials satisfying a non-hermitian orthogonality property has been established in the works of Stahl [@Stahl:88] and Gonchar-Rakhmanov [@Gonchar:87]. These authors describe the limit distribution in terms of the equilibrium measure in an external field on a contour satisfying a symmetry property in ${{\mathbb{C}}}$. Our contour $\Gamma$ possesses this property, but the theorems of [@Stahl:88] and [@Gonchar:87] are not applicable: an essential assumption in these papers is the connectedness of the complement to the contour. Nevertheless, the measure $\mu$ from (\[defmu\]) is the above mentioned equilibrium measure on $\Gamma$ in a certain external field. Also the contours $\Gamma_r$ have the symmetry property and the measures given in parts (b)–(d) of Theorem \[theoremweak2\] are the equilibrium measures in the external fields on these contours. So Theorem \[theoremweak2\] shows that in a certain sense the results of Gonchar-Rakhmanov-Stahl are also valid for Jacobi polynomials with varying negative parameters. It seems likely that similar results hold in more general situations. Strong asymptotics away from $\zeta_{\pm}$ ------------------------------------------ The weak convergence results of Theorems \[theoremweak1\] and \[theoremweak2\] follow from the strong asymptotic results that we obtain for the Jacobi polynomials. We state the result here for the sequence $P_n^{(An, Bn)}$. We use $\widehat P_n^{(An,Bn)}$ to denote the corresponding monic Jacobi polynomial. Note that $\Gamma$ and $\Gamma^{\perp}$ divide the complex plane into six domains, which we number from left to right by I, II, III, IV, V, and VI, as shown in Fig. \[fig\_regions1\]. ![Domains defined by trajectories $\Gamma \cup \Gamma^\perp$. []{data-label="fig_regions1"}](Jac_fig34_regions) To state the asymptotic results we need to be specific about the branches of the functions that are involved. We already defined $\phi$ in (\[defphi\]) as a multi-valued function. Now we specify that $$\label{defphi2} \phi(z) = \frac{A+B+2}{2} \int_{\zeta_-}^z \frac{R(t)}{t^2-1} dt, \qquad z \in \mathbb C \setminus (\Gamma_C \cup \gamma_1^+ \cup \gamma_{-1}^+ \cup \gamma_{\infty}^+)$$ where integration from $\zeta_-$ to $z$ is along a curve in $\mathbb C \setminus (\Gamma_C \cup \gamma_1^+ \cup \gamma_{-1}^+ \cup \gamma_{\infty}^+)$. Note that this definition prevents the curve from going around the cut $\Gamma_C$ and also from going around one of the poles $\pm 1$. Near infinity, $\phi$ behaves like $$\label{defc} \phi(z) = \frac{A+B+2}{2} \log z + c + O\left(\frac{1}{z}\right)$$ for some constant $c$. This constant $c$ will also appear in the asymptotic formulas below. In our formulas we will also see fractional powers $(z-1)^{-An/2}$ and $(z+1)^{-Bn/2}$. We will choose these to be defined and analytic in $\mathbb C \setminus (\gamma_1^+ \cup \gamma_{\infty}^+)$ and $\mathbb C \setminus (\gamma_{-1}^+ \cup \gamma_{\infty}^+)$, respectively, and to be positive for real $z > 1$. Finally, we define $$\label{defN11} N_{11}(z) = \frac{1}{2} \left(\left(\frac{z-\zeta_-}{z-\zeta_+}\right)^{1/4} + \left(\frac{z-\zeta_+}{z-\zeta_-}\right)^{1/4} \right)$$ and $$\label{defN12} N_{12}(z) = \frac{1}{2i} \left(\left(\frac{z-\zeta_-}{z-\zeta_+}\right)^{1/4} - \left(\frac{z-\zeta_+}{z-\zeta_-}\right)^{1/4} \right)$$ which are defined and analytic in $\mathbb C \setminus \Gamma_C$. The fourth-roots are chosen so that they approach $1$ as $z \to \infty$. We call these functions $N_{11}$ and $N_{12}$ since they will appear later as the corresponding entries of a matrix $N$. Now we have all the ingredients to state our main theorem. \[maintheorem\] Let $A$ and $B$ satisfy [(\[ABinequalities\])]{}. Then the monic Jacobi polynomials $\widehat P_n^{(An,Bn)}$ have the following asymptotic behavior as $n \to \infty$. 1. For $z$ in domains I and II, $$\begin{aligned} \nonumber \lefteqn{\widehat P_n^{(An,Bn)}(z) = e^{-nc} (z-1)^{-An/2}(z+1)^{-Bn/2} } \\ & & \label{asformleft} \left( e^{n\phi(z)} N_{11}(z) \left(1 + O\left(\frac{1}{n}\right)\right) \right. \left. - e^{-An\pi i} \frac{\sin (Bn\pi)}{\sin((A+B)n \pi)} e^{-n\phi(z)} N_{12}(z) \left(1 + O\left(\frac{1}{n}\right)\right) \right)\end{aligned}$$ 2. For $z$ in domain III, $$\begin{split} \widehat P_n^{(An,Bn)}(z) = & \ e^{-nc} (z-1)^{-An/2}(z+1)^{-Bn/2} \\ & \left( e^{Bn \pi i} \frac{\sin(An\pi)}{\sin((A+B)n \pi)} e^{n\phi(z)} N_{11}(z) \left(1 + O\left(\frac{1}{n}\right)\right) \right. \\ & \label{asformmiddle1} \qquad \left. - e^{-An\pi i} \frac{\sin (Bn\pi)}{\sin((A+B)n \pi)} e^{-n\phi(z)} N_{12}(z) \left(1 + O\left(\frac{1}{n}\right)\right) \right) \end{split}$$ 3. For $z$ in domain IV, $$\begin{split} \widehat P_n^{(An,Bn)}(z) = & \ e^{-nc} (z-1)^{-An/2}(z+1)^{-Bn/2} \\ & \left(e^{-An\pi i} \frac{\sin (Bn\pi)}{\sin((A+B)n \pi)} e^{n\phi(z)} N_{11}(z) \left(1 + O\left(\frac{1}{n}\right)\right) \right. \\ & \label{asformmiddle2} \qquad \left. + e^{Bn \pi i} \frac{\sin(An\pi)}{\sin((A+B)n \pi)} e^{-n\phi(z)} N_{12}(z) \left(1 + O\left(\frac{1}{n}\right)\right) \right) \end{split}$$ 4. For $z$ in domains V and VI, $$\begin{aligned} \nonumber \lefteqn{\widehat P_n^{(An,Bn)}(z) = e^{-nc} (z-1)^{-An/2}(z+1)^{-Bn/2} } \\ & & \label{asformright} \left( e^{n\phi(z)} N_{11}(z) \left(1 + O\left(\frac{1}{n}\right)\right) \right. \left. + e^{Bn \pi i} \frac{\sin(An\pi)}{\sin((A+B)n \pi)} e^{-n\phi(z)} N_{12}(z) \left(1 + O\left(\frac{1}{n}\right)\right) \right)\end{aligned}$$ These asymptotic formulas hold uniformly for $z$ in the indicated domains away from the branch points, uniformly for $A$ and $B$ in compact subsets of the region $-1 < A < 0$, $-1 < B < 0$, $-2 < A+B < -1$, and for values of $n$ such that $(A+B)n$ is not an integer. One can verify that the asymptotic formulas (\[asformleft\])–(\[asformright\]) agree on the boundaries of the respective domains. The fact that the formulas (\[asformleft\])–(\[asformright\]) hold uniformly for $A$ and $B$ in compact subsets of the region given by (\[ABinequalities\]) implies that we can allow varying values of $A$ and $B$. In particular, we can consider two sequences $(\alpha_n)$ and $(\beta_n)$ such that the limits (\[ABlimits\]) exist and satisfy (\[ABinequalities\]). We then have asymptotic formulas for the Jacobi polynomials $\widehat P_n^{(\alpha_n,\beta_n)}$ as in (\[asformleft\])–(\[asformright\]) with $A$ replaced by $A_n = \alpha_n/n$ and $B$ replaced by $\beta_n/n$. Then we also have to realize that $\phi$, $c$, $N_{11}$ and $N_{12}$ are going to be $n$-dependent. Indeed, these quantities are defined using $A$ and $B$, which here we have to replace by $A_n$ and $B_n$; we chose to state the theorem for $\alpha _n=An$ and $\beta _n=Bn$ for the sake of brevity of notation. The expressions between brackets in the right hand-sides of (\[asformleft\])–(\[asformright\]) contain two terms that correspond to the Liouville-Green approximation of two linearly independent solutions of the differential equation satisfied by the corresponding Jacobi polynomials (cf. [@Olver74 Ch.VI]). In different regions of the plane and depending on the parameters, these two terms are of comparable sizes (and then zeros of the polynomials arise), or one of them is dominating the other. If we assume that $An$, $Bn$, and $(A+B)n$ are not close to integers, the expressions $\sin(An\pi)/\sin((A+B)n\pi)$ and $\sin(Bn\pi)/\sin((A+B)n\pi)$ have moderate sizes (not too small, not too big). In that case the dominant term is determined by ${\mathop{\rm Re}}\phi$. For $z \in \Omega_{\infty}$, we have ${\mathop{\rm Re}}\phi(z) > 0$, and then (\[asformleft\]) and (\[asformright\]) both reduce to $$\label{asformoutside} \widehat P_n^{(An,Bn)}(z) = e^{-nc} (z-1)^{-An/2}(z+1)^{-Bn/2} e^{n\phi(z)} N_{11}(z) \left(1 + O\left(\frac{1}{n}\right)\right)$$ for $z$ in domains I and VI. For $z \in \Omega_1 \cup \Omega_{-1}$ we have ${\mathop{\rm Re}}\phi(z) < 0$, so that $e^{-n\phi(z)}$ dominates $e^{n\phi(z)}$ for large $n$. Then (\[asformleft\])–(\[asformright\]) reduce to $$\label{asforminside-1} \widehat P_n^{(An,Bn)}(z) = - e^{-An\pi i-nc} \frac{\sin (Bn\pi)}{\sin((A+B)n \pi)} (z-1)^{-An/2}(z+1)^{-Bn/2} e^{-n\phi(z)} N_{12}(z) \left(1 + O\left(\frac{1}{n}\right)\right)$$ for $z$ in domains II and III (that is, for $z \in \Omega_{-1}$), and to $$\label{asforminside1} \widehat P_n^{(An,Bn)}(z) = e^{Bn \pi i-nc} \frac{\sin(An\pi)}{\sin((A+B)n \pi)} (z-1)^{-An/2}(z+1)^{-Bn/2} e^{-n\phi(z)} N_{12}(z) \left(1 + O\left(\frac{1}{n}\right)\right)$$ for $z$ in domains IV and V (that is, for $z \in \Omega_1$). We emphasize that (\[asformoutside\]), (\[asforminside-1\]), and (\[asforminside1\]) only hold if $An$, $Bn$, and $(A+B)n$ are not close to integers. In general one has to use the compound asymptotic formulas (\[asformleft\])–(\[asformright\]). If $An$ is an integer, then (\[asformmiddle2\]) and (\[asformright\]) reduce to (\[asformoutside\]) for $z$ in domains IV, V, and VI. Then we see the multiple zero at $z=1$, not only because of the factor $(z-1)^{-An/2}$, but also because $$\phi(z) = -\frac{A}{2} \log(z-1) + O(1) \qquad \mbox{ as } z \to 1$$ so that $$\label{ephinear1} e^{n\phi(z)} = (z-1)^{-An/2}(1+ O(z-1)) \qquad \mbox{ as } z \to 1.$$ So we have a zero at $z=1$ of multiplicity $-An$, as it should be. Similar remarks apply if $Bn$ is an integer. In that case we have a zero at $z=-1$ of multiplicity $-Bn$. If $An$ is not an integer, then $\widehat P_n^{(An,Bn)}$ does not have a zero at $z=1$. This is in agreement with formulas (\[asformmiddle2\]) and (\[asformright\]) since the zero at $z=1$ due to the factor $(z-1)^{-An/2}$ is compensated exactly by the singularity in $e^{-n\phi(z)}$ at $z=1$, see (\[ephinear1\]). Strong asymptotics near $\zeta_-$. ---------------------------------- The asymptotic formulas (\[asformleft\]) and (\[asformright\]) are not valid near the branch points $\zeta_-$ and $\zeta_+$. Near those points, there is an asymptotic formula involving Airy functions. We need the following particular combination of Airy functions, depending on $A$, $B$, and $n$, $$\begin{aligned} \label{Airycombination1} {{\mathcal A}}(s;A,B,n) &= & - e^{Bn \pi i} \frac{\sin(An\pi)}{\sin((A+B)n\pi)}\, \omega {\mathop{\rm Ai}}(\omega s) +e^{-An\pi i} \frac{\sin(Bn\pi)}{\sin((A+B)n\pi)}\, \omega^2 {\mathop{\rm Ai}}(\omega^2 s) \\ & = & \frac{1}{2i} \frac{\cos((A+B)n\pi)- \exp((B-A)n\pi i)}{\sin((A+B)n\pi)} {\mathop{\rm Ai}}(s) + \frac{1}{2i} {\mathop{\rm Bi}}(s), \label{Airycombination}\end{aligned}$$ where $\omega = e^{2\pi i/3}$ and ${\mathop{\rm Ai}}$ and ${\mathop{\rm Bi}}$ are the usual Airy functions [@Abramowitz]. Note that ${{\mathcal A}}(s;A,B,n)$ is defined for combinations of $A$, $B$, and $n$ that are such that $(A+B)n$ is not an integer. \[asymnearbranch\] Let $A$ and $B$ satisfy [(\[ABinequalities\])]{}. Then there is a $\delta > 0$ such that for every $z$ with $\|z-\zeta_-\| < \delta$, the monic Jacobi polynomials $\widehat P_n^{(An,Bn)}$ have the following asymptotic behavior as $n \to \infty$: $$\begin{split} \widehat P_n^{(An, Bn)}(z) = & e^{-nc} (z-1)^{-An/2} (z+1)^{-Bn/2} \sqrt{\pi} i \\ & \times \left[ n^{1/6} \left( \frac{z-\zeta_+}{z-\zeta_-} f(z)\right)^{1/4} {{\mathcal A}}(n^{2/3} f(z);A,B,n) \left(1+O\left(\frac{1}{n}\right)\right) \right. \\ & \label{asformnearzeta-} \left. \qquad + n^{-1/6} \left(\frac{z-\zeta_+}{z-\zeta_-} f(z)\right)^{-1/4} {{\mathcal A}}'(n^{2/3} f(z);A,B,n) \left(1+O\left(\frac{1}{n}\right)\right) \right] \end{split}$$ with $$\label{deffz} f(z) = \left[ \frac{3}{2}\, \phi(z) \right]^{2/3}$$ where the $2/3$rd root chosen is real and positive on $\gamma_{\infty}^-$. The $O$-terms in [(\[asformnearzeta-\])]{} hold uniformly for $\|z-\zeta_-\| < \delta$ and for $A$ and $B$ in compact subsets of the region $-1 < A < 0$, $-1 < B < 0$, $-2 < A+B < -1$, and for values of $n$ such that $(A+B)n$ is not an integer. There is a similar asymptotic formula for the behavior near $\zeta_+$. From the uniform asymptotics in Theorem \[asymnearbranch\] it is possible to establish a more precise behavior of the zeros of $P_n^{(An, Bn)}$ close to the branch points $\zeta_{\pm}$. In fact, the zeros of the function ${{\mathcal A}}$ defined in (\[Airycombination1\])–(\[Airycombination\]) model the behavior of these zeros of $P_n^{(An, Bn)}$. For instance, in the generic case (\[notclosetoZ\]) both terms in (\[Airycombination1\]) or (\[Airycombination\]) have approximately the same size, and ${{\mathcal A}}$ has its zeros aligned along three curves emanating from $0$ and forming the same angle. The situation is different in cases (b)–(d) of Theorem \[theoremweak2\]. For instance, if $r_{\alpha+\beta}> r_{\alpha}= r_{\alpha}$, then the first term in (\[Airycombination\]) dominates the second term. But if $r_{\alpha} \neq r_{\alpha}$ then one of the two terms in (\[Airycombination1\]) dominates the other. In these cases the zeros of ${{\mathcal A}}$ behave like zeros of the dominating Airy function and are aligned along a single curve emanating from $0$. Proof of Lemmas \[lemma1\] and \[lemma2\] ========================================= of Lemma \[lemma1\].* The quadratic differential (\[quaddiff\]) has a simple zero at $\zeta_{\pm}$ and a double pole at $\pm 1$ and at $\infty$. This determines the local structure of the trajectories as follows, see also [@baik01], [@Pommerenke Chapter 8] or [@Strebel84 Chapter III], 1. Three trajectories emanate from $\zeta_{\pm}$ at equal angles. These are the critical trajectories. 2. Near $\pm 1$ the trajectories are simple closed contours. Here we use the fact that $$- \frac{(z-\zeta_+)(z-\zeta_-)}{(z^2-1)^2} = \frac{c_{\pm 1}}{(z \mp 1)^2} + O\left(\frac{1}{z \mp 1}\right) \qquad \textrm{as } z \to \pm 1,$$ with $c_{\pm 1} < 0$. 3. The trajectories near $\infty$ are also simple closed contours. This follows from the fact that in the expansion $$- \frac{(z-\zeta_+)(z-\zeta_-)}{(z^2-1)^2} = \frac{c_{\infty}}{z^2} + O\left(\frac{1}{z^3}\right) \qquad \textrm{as } z \to \infty,$$ we have $c_{\infty} = -1 < 0$. In the lower half-plane $\mathbb C^-$ there is only the simple zero at $\zeta_-$. The other points are regular points. This means that the three critical trajectories that emanate from $\zeta_-$ extend to the boundary of $\mathbb C^-$, cf. [@Pommerenke Lemma 8.4]. Because of (2) and (3) and the fact that trajectories do not intersect, the critical trajectories do not tend to infinity, or come to $\pm 1$. So each critical trajectories exits the lower half-plane in a point from $\mathbb R \setminus \{-1,1\}$ and these points are mutually distinct, say $\xi_L$, $\xi_C$, and $\xi_R$, with $\xi_L < \xi_C < \xi_R$. Because of the symmetry with respect to the real axis, $${\mathop{\rm Re}}\int_{\zeta _-}^{\zeta _+} \frac{R(t)}{t^2-1}\, dt =0.$$ Hence, the three trajectories extend into the upper half-plane as their mirror images in $\mathbb R$, and so they continue to $\zeta_+$. This proves the existence of three arcs $\Gamma_L$, $\Gamma_C$, and $\Gamma_R$ contained in $\Gamma$ and connecting $\zeta_{\pm}$, where $\xi_s \in \Gamma_s$ for $s\in \{L,C,R\}$. Next, we note that $\Gamma_L \cup \Gamma_C$ is a closed contour consisting of trajectories. It follows from [@Pommerenke Lemma 8.3] that it has to surround a pole. Similarly $\Gamma_C \cup \Gamma_R$ has to surround a pole. This can only happen if $\xi_L < -1 < \xi_C < 1 < \xi_R$. To complete the proof of the lemma, we need to establish that $\Gamma$ consists only of $\Gamma_L$, $\Gamma_C$, and $\Gamma_R$ and nothing more. We use that the function $$\label{defhz} h(z) = {\mathop{\rm Re}}\int_{\zeta_-}^z \frac{R(t)}{t^2-1} dt, \qquad z \in \mathbb C \setminus (\Gamma_C \cup \{-1,1\}),$$ is single-valued and harmonic in $\mathbb C \setminus (\Gamma_C \cup \{-1,1\})$. The path of integration in (\[defhz\]) is in $\mathbb C \setminus \Gamma_C$. It is easy to see that $$\lim_{z \to \infty} h(z) = +\infty.$$ Since $h = 0$ on $\Gamma_L \cup \Gamma_R = \partial \Omega_{\infty}$, it follows by the maximum principle for harmonic functions that $h(z) > 0$ for $z \in \Omega_{\infty}$. Similarly, since $$\lim_{z \to \infty} h(z) = - \infty,$$ and $h = 0$ on $\Gamma_L \cup \Gamma_C = \partial \Omega_{-1}$, and on $\Gamma_R \cup \Gamma_C = \partial \Omega_1$, we have that $h(z) < 0$ for $z \in \Omega_{\pm}$. Since $\Gamma = \{h = 0\}$, we get that $\Gamma$ consists exactly of $\Gamma_L$, $\Gamma_C$, and $\Gamma_R$. This completes the proof of Lemma \[lemma1\]. of Lemma \[lemma2\]. Recall that $ R(z):=\sqrt{(z-\zeta_+)(z-\zeta_-)}$ denotes the single-valued branch in ${{\mathbb{C}}}\setminus \Gamma_C$ such that $R(z) \sim z$ as $z \to \infty$. With this convention and taking into account (\[zetapm\]) it is straightforward to check that $$\label{R_1} R(-1)= \frac{2B}{A+B+2} <0, \qquad R(1)=\frac{-2A}{A+B+2} > 0.$$ From the definition of $\Gamma$ it follows that $d \mu(z)$ is real-valued on $\Gamma$ and does not change sign on each component of $\Gamma \setminus \{\zeta_-,\zeta_+ \}$. Using the residue theorem, we have that $$\mu(\Gamma_C \cup \Gamma_R)=\int_{\Gamma_C \cup \Gamma_R} d\mu (t)= (A + B+2) {\mathop{\rm res}}_{z=1} \left( \frac{R(z)}{z^2-1} \right) = -A$$ where we have used (\[R\_1\]). Analogously, $$\mu(\Gamma_L \cup \Gamma_C)=\int_{\Gamma_L \cup \Gamma_C} d\mu (t)= (A + B+2) {\mathop{\rm res}}_{z=-1} \left( \frac{R(z)}{z^2-1} \right) =-B.$$ Finally, $$\mu(\Gamma_L \cup \Gamma_R)=\int_{\Gamma_L \cup \Gamma_R} d\mu (t)= (A + B+2) {\mathop{\rm res}}_{z=\infty} \left( \frac{R(z)}{z^2-1} \right) =A+B+2.$$ Hence, $$\mu(\Gamma)=\frac{1}{2}\, \left( \mu(\Gamma_C \cup \Gamma_R)+\mu(\Gamma_L \cup \Gamma_C) + \mu(\Gamma_L \cup \Gamma_R)\right)=1,$$ and $$\begin{aligned} \mu(\Gamma_L) & = & 1-\mu(\Gamma_C \cup \Gamma_R)\ =\ 1+A, \\ \mu(\Gamma_R) & = & 1-\mu(\Gamma_L \cup \Gamma_C)\ = \ 1+B, \\ \mu(\Gamma_C) & = & 1 - \mu(\Gamma_L \cup \Gamma_R)\ =\ -1-A-B,\end{aligned}$$ which proves (\[measures\_arcs1\]). Since each part has positive total $\mu$-mass and $\mu$ does not change sign on each of the parts, we find that $\mu$ is a positive measure. This completes the proof. A Riemann-Hilbert problem for Jacobi polynomials ================================================ Consider a closed path $\Gamma_u$ encircling the points $+1$ and $-1$ first in the positive direction and then in the negative direction, as shown in Fig. \[fig:pathGamma\]. The point $\xi \in (-1,1)$ is the begin and endpoint of $\Gamma_u$. ![ The universal curve $\Gamma_u$.[]{data-label="fig:pathGamma"}](JAC_fig21_curve) For $\alpha , \beta \in {{\mathbb{C}}}$, denote $$\label{weight} w(z; \alpha , \beta ):=(1-z)^\alpha (1+z)^\beta=\exp [\alpha \log (1-z) + \beta \log (1+z)].$$ This is a multi-valued function with branch points at $\infty$ and $\pm 1$. However, if we start with a value of $w(z;\alpha, \beta)$ at a particular point of $\Gamma_u$, and extend the definition of $w(z; \alpha, \beta)$ continuously along $\Gamma_u$, then we obtain a single-valued function $w(z; \alpha, \beta)$ on $\Gamma_u$ if we view $\Gamma_u$ as a contour on the Riemann surface for the function $w(z;\alpha, \beta)$. For definiteness, we assume that the “starting point” is $\xi \in (-1,1)$, and that the branch of $w$ is such that $w(\xi; \alpha , \beta )>0$. We prefer to view $\Gamma_u$ as a subset of the complex plane. Then $\Gamma_u$ has points of self-intersection, as shown in Fig. \[fig:pathGamma\]. At points of self-intersection the value of $w(z;\alpha,\beta)$ is not well-defined. In [@Kuijlaars03] it was shown that for $k \in \{0, 1, \ldots, n\}$, we have $$\label{orthMain} \int_{\Gamma_u} t^k \, P_n^{(\alpha , \beta )}(t) w(t; \alpha, \beta)\, dt = \frac{- \pi^2 2^{n+\alpha +\beta +3} e^{\pi i (\alpha +\beta )}}{\Gamma(2n+\alpha +\beta +2) \Gamma(-n-\alpha ) \Gamma(-n-\beta)}\, \delta_{kn}.$$ This shows that the Jacobi polynomials $P_n^{(\alpha,\beta)}$ are orthogonal on the universal curve $\Gamma_u$. The right-hand side of (\[orthMain\]) vanishes for $k=n$ if and only if either $-2n-\alpha-\beta-2$, or $n+\alpha$ or $n+\beta$ is a non-negative integer. In some of these cases the zero comes from integrating a single-valued and analytic function along a curve in the region of analyticity; other values of $\alpha$ and $\beta$ correspond to the special cases mentioned before when there is a zero at $\pm 1$. It is shown in [@Kuijlaars03] that the orthogonality conditions (\[orthMain\]) characterize the Jacobi polynomial $P_n^{(\alpha, \beta)}$ provided the parameters satisfy $$\label{condition} -n-\alpha-\beta \notin {{\mathbb{N}}}, \quad \text{and} \quad n+\alpha \notin {{\mathbb{N}}}, \quad \text{and} \quad n+\beta \notin {{\mathbb{N}}}.$$ Then $P_n^{(\alpha , \beta )}$ is of degree exactly $n$, and we will denote by $\widehat P_n^{(\alpha , \beta )}$ the corresponding monic Jacobi polynomial. Based on the orthogonality (\[orthMain\]) a Riemann-Hilbert problem is constructed in [@Kuijlaars03], whose solution is given in terms of $\widehat P_n^{(\alpha,\beta)}$ with parameters satisfying (\[condition\]). Let $\Gamma_u$ be a curve in ${{\mathbb{C}}}$ as described above with three points of self-intersection as in Fig. \[fig:pathGamma\]. We let $\Gamma_u^o$ be the curve without the points of self-intersection. Recall that the orientation of $\Gamma_u$ (see also Fig. \[fig:pathGamma\]) induces a $+$ and $-$ side in a neighborhood of $\Gamma_u$, where the $+$ side is on the left while traversing $\Gamma_u$ according to its orientation and the $-$ side is on the right. Again, we say that a function $Y$ on ${{\mathbb{C}}}\setminus \Gamma_u$ has a boundary value $Y_+(t)$ for $t \in \Gamma_u^o$ if the limit of $Y(z)$ as $z \to t$ with $z$ on the $+$ side of $\Gamma_u$ exists; similarly for $Y_-(t)$. The Riemann-Hilbert problem for Jacobi polynomials is then as follows. We look for a $2\times 2$ matrix valued function $Y =Y^{(\alpha , \beta )}:\, {{\mathbb{C}}}\setminus \Gamma_u \to {{\mathbb{C}}}^{2\times 2}$ such that the following four conditions are satisfied: 1. $Y$ is analytic on ${{\mathbb{C}}}\setminus \Gamma_u$. 2. $Y$ has continuous boundary values on $\Gamma_u^o$, denoted by $Y_{+}$ and $Y_{-}$, such that $$Y_{+}(t) = Y_{-}(t) \begin{pmatrix} 1 & w(t; \alpha, \beta) \\ 0 & 1 \end{pmatrix} \qquad \mbox{for } t\in \Gamma_u^o.$$ 3. As $z\to\infty$, $$Y(z) = \left(I + O\left(\frac{1}{z}\right)\right) \begin{pmatrix} z^{n} & 0 \\ 0 & z^{-n} \end{pmatrix}.$$ 4. $Y(z)$ remains bounded as $z \to t \in \Gamma_u \setminus \Gamma_u^o$. This Riemann-Hilbert problem is similar to the Riemann-Hilbert problem for orthogonal polynomials due to Fokas, Its, and Kitaev [@Fokas92], see also [@MR2000g:47048]. Also the solution is similar. \[solRH\] Assume that the parameters $\alpha, \beta$ satisfy [(\[condition\])]{}. Then the above Riemann-Hilbert problem for $Y$ has a unique solution, which is given by $$\label{formulaY} Y(z)= \begin{pmatrix} \widehat P_{n}^{(\alpha,\beta)}(z) & \frac{1}{2\pi i} \int\limits_{\Gamma_u} \frac{\widehat P_{n}^{(\alpha,\beta)}(t) w(t; \alpha, \beta)}{t-z} \, dt \\[10pt] c_{n-1} P_{n-1}^{(\alpha,\beta)}(z) & \frac{c_{n-1}}{2\pi i} \int\limits_{\Gamma_u} \frac{P_{n-1}^{(\alpha,\beta)}(t) w(t; \alpha, \beta)}{t-z}\, dt \end{pmatrix}, \qquad z \in \mathbb C \setminus \Gamma_u,$$ for some non-zero constant $c_{n-1}$. The Riemann-Hilbert problem holds for any combination of parameters $\alpha$ and $\beta$ such that (\[condition\]) is satisfied. Also the contour $\Gamma_u$ is rather arbitrary. It could be modified to any curve that is homotopic to it in ${{\mathbb{C}}}\setminus \{-1,1\}$. Transformations of the Riemann-Hilbert problem ============================================== In this section we consider parameters $A$ and $B$ satisfying the inequalities (\[ABinequalities\]). We also assume that $n \in \mathbb N$ is such that $An$, $Bn$ and $(A+B)n$ are non-integers. Throughout this section $A$, $B$, and $n$ remain fixed. From Proposition \[solRH\] we know that the Jacobi polynomial $\widehat P_n^{(An,Bn)}$ is characterized as the $(1,1)$ entry of the solution of the Riemann-Hilbert problem for $Y$ given in the previous section with $\alpha = An$ and $\beta = Bn$. In this section we apply the steepest descent method of Deift and Zhou to this Riemann-Hilbert problem in order to reduce it to a Riemann-Hilbert problem that is normalized at infinity and whose jump matrices are close to the identity. In the next section we derive the asymptotic results from this analysis. The Deift/Zhou steepest descent method proceeds through a number of transformations of the original Riemann-Hilbert problem. Choice of contour ----------------- In the first step of the analysis we have to pick the right contour. For $A$ and $B$ satisfying (\[ABinequalities\]) we have the contour $\Gamma = \Gamma^{(A,B)}$ defined in (\[defGamma\]), which according to Lemma \[lemma1\] consists of three analytic arcs $\Gamma = \Gamma_L \cup \Gamma_C \cup \Gamma_R$. We modify $\Gamma_u$ to a contour that follows $\Gamma$ in such a way that every part of $\Gamma$ is covered twice as shown in Fig. \[fig\_tautening\]. ![Tautening $\Gamma_u$ on the set $\Gamma$.[]{data-label="fig_tautening"}](Jac_fig32_tautening) Passing from the Riemann-Hilbert problem on $\Gamma_u$ to the Riemann-Hilbert problem on $\Gamma$, we have that on each part of $\Gamma$ two of the jumps are combined. The new jump matrices take the form $$\begin{pmatrix} 1 & w(t_1; An, Bn) \\ 0 & 1 \end{pmatrix} \begin{pmatrix} 1 & -w(t_2; An, Bn) \\ 0 & 1 \end{pmatrix} = \begin{pmatrix} 1 & w(t_1; An,Bn) - w(t_2; An, Bn) \\ 0 & 1 \end{pmatrix},$$ where $t_1$ and $t_2$ are points on the Riemann surface, both lying above $t$. The values of $w(t_j; An, Bn)$, $j=1, 2$, differ from each other by a phase factor. To make this precise we specify a single-valued branch for the weight $$w(z;An, Bn) = (1-z)^{An} (1+z)^{Bn}$$ on $\Gamma$. Since $\Gamma \setminus \{ \zeta_+\}$ is simply connected, we can define a single valued branch on $\Gamma \setminus \{ \zeta_+\}$, and we will do it in such a way that $w(\xi; An, Bn) > 0$, where $\xi = \xi_C$ is the intersection point of $\Gamma_C$ with the interval $(-1,1)$. Then the jump on each of the contours $\Gamma_L$, $\Gamma_C$, and $\Gamma_R$ can be calculated. The result is the following Riemann-Hilbert problem on $\Gamma$ for a matrix valued function which we continue to call $Y$. The contour $\Gamma$ has the orientation shown in Fig. \[fig\_Gamma1\]. 1. $Y$ is analytic on ${{\mathbb{C}}}\setminus \Gamma$. 2. $Y$ has continuous boundary values on $\Gamma \setminus \{\zeta_-, \zeta_+\}$, denoted by $Y_+$ and $Y_-$, such that $$\label{RHmatrix1} Y_+(t) = Y_-(t) \begin{pmatrix} 1 & d_s w(t; An, Bn) \\ 0 & 1 \end{pmatrix} \qquad \mbox{for } t \in \Gamma_s \setminus \{ \zeta_{\pm}\}, \quad s \in \{L,C,R\},$$ with constants $$\label{constants} d_L = e^{2\pi Bn i} \left(e^{2\pi An i} -1\right), \qquad d_C = 1-e^{2\pi (A+B)ni}, \qquad d_R = 1-e^{2\pi B n i},$$ and we follow the convention about the branch of $w(t; An, Bn)$ on $\Gamma$ mentioned above. 3. As $z\to\infty$, $$Y(z) = \left(I + O\left(\frac{1}{z}\right)\right) \begin{pmatrix} z^{n} & 0 \\ 0 & z^{-n} \end{pmatrix}.$$ 4. $Y(z)$ remains bounded as $z \to \zeta_\pm $. Of course the solution to the above Riemann-Hilbert problem is similar to the solution (\[formulaY\]) to the earlier Riemann-Hilbert problem. In particular we still have $$\label{Y11} Y_{11}(z) = \widehat P_n^{(An,Bn)}(z)$$ The constants $d_L$, $d_C$ and $d_R$ from (\[constants\]) will play an important role in what follows. These numbers are non-zero, exactly because of our assumption that $An$, $Bn$, and $(A+B)n$ are non-integers. Observe that $$\label{constants2} d_L + d_C = d_R,$$ which is a relation that will be used a number of times. Auxiliary functions ------------------- In order to make the first transformation of the Riemann-Hilbert problem we need some auxiliary functions. We already know from Lemma \[lemma2\] that $\mu$ defined in (\[defmu\]) is a positive measure on $\Gamma$ such that $$\label{measures_arcs} \mu(\Gamma_L)=1+A>0, \qquad \mu(\Gamma_C)=-1-A-B>0, \qquad \mu(\Gamma_R)=1+B>0.$$ Let $g$ be the complex logarithmic potential of the measure $\mu$, $$g(z) = \int \log(z-t) d\mu(t).$$ This is a multivalued function; however its derivative is single valued: $$\label{g_prime} g'(z)=\int \frac{d\mu(t)}{z-t}=\begin{cases} \dfrac{\strut A+B+2}{2}\, \dfrac{R(z)}{z^2-1}- \dfrac{A/2}{z-1}-\dfrac{B/2}{z+1}, & \text{ for } z \in \Omega_\infty,\\[10pt] -\dfrac{\strut A+B+2}{2}\, \dfrac{R(z)}{z^2-1}- \dfrac{A/2}{z-1}-\dfrac{B/2}{z+1}, & \text{ for } z \in \Omega_{-1} \cup \Omega_{1}. \end{cases}$$ We define $$\label{def_G} G(z)=\exp \left(\int_{\zeta_-}^z g'(t)\, dt\right), \qquad z \in \mathbb C \setminus \Gamma,$$ where the path of integration lies entirely in $\mathbb C \setminus \Gamma$ except for the initial point $\zeta_-$. From the fact that $\mu$ is a positive unit measure on $\Gamma$ it follows that $G$ is single-valued in each component of $\mathbb C \setminus \Gamma$. Furthermore, $G$ is analytic, $G(\zeta_-) = 1$, and the following limit exists $$\label{kappa} \kappa := \lim_{z \to \infty} \frac{G(z)}{z} = \zeta_- \exp\left( \int_{\zeta_-}^{\infty} (g'(t) - 1/t)\, dt \right).$$ We calculate the jumps of $G$. We have $$\label{G+timesG-} G_+(z) G_-(z) = \frac{w(\zeta_-;A,B)}{w(z;A,B)}, \qquad \textrm{ for } z \in \Gamma,$$ and $$\label{G+divideG-} \frac{G_+(z)}{G_-(z)} = \exp(-2 \phi_+(z)), \qquad \textrm{ for } z \in \Gamma,$$ where $\phi$ is defined by (\[defphi2\]). It will be useful to introduce also the related function $$\label{defphitilde} \tilde{\phi}(z) = \frac{A+B+2}{2} \int_{\zeta_+}^z \frac{R(t)}{t^2-1} dt = \overline{\phi(\bar{z})}, \qquad \text{ for } z \in \mathbb C \setminus (\Gamma_C \cup \gamma_{-1}^{-} \cup \gamma_1^- \cup \gamma_{\infty}^-).$$ To relate $\tilde{\phi}$ with $\phi$ it is necessary to compute $\frac{A+B+2}{2} \int_{\zeta_-}^{\zeta_+} \frac{R(t)}{t^2-1} dt$. This integral depends on the path from $\zeta_-$ to $\zeta_+$. We can distinguish four paths, namely $\Gamma_R$, $-\Gamma_{C,+}$, $-\Gamma_{C,-}$ and $-\Gamma_L$. (Recall that $\Gamma_C$ and $\Gamma_L$ are oriented from $\zeta_+$ to $\zeta_-$. So we put a minus sign to indicate that the path is from $\zeta_-$ to $\zeta_+$.) We obtain $$\frac{A+B+2}{2} \int_{\zeta_-}^{\zeta_+} \frac{R(t)}{t^2-1} dt = \left\{ \begin{array}{lcll} \pi i \mu(\Gamma_R) & = & \pi i (1+B) & \text{ integral over $\Gamma_R$} \\ -\pi i \mu(\Gamma_C) & = & \pi i (1+A+B) & \text{ integral over $-\Gamma_{C,+}$} \\ \pi i \mu(\Gamma_C) & = & -\pi i (1+A+B) & \text{ integral over $-\Gamma_{C,-}$} \\ -\pi i \mu(\Gamma_L) & = & -\pi i (1+A) & \text{ integral over $-\Gamma_L$} \end{array} \right.$$ where we have used (\[measures\_arcs\]). It follows that $$\begin{aligned} \label{phirelation1} \left. \begin{array}{rcl} \phi_+(z) & = & \tilde{\phi}(z) + \pi i(1+B) \qquad \\ \phi_-(z) & = & \tilde{\phi}(z) - \pi i(1+A) \end{array} \right\} && \text{ for $z$ on } \gamma_{\infty}^+, \\ \label{phirelation2} \left. \begin{array}{rcl} \phi_+(z) & = & \tilde{\phi}(z) - \pi i(1+A+B) \\ \phi_-(z) & = & \tilde{\phi}(z) - \pi i(1+A) \end{array} \right\} && \text{ for $z$ on } \gamma_{-1}^+, \\ \label{phirelation3} \left. \begin{array}{rcl} \phi_+(z) & = & \tilde{\phi}(z) + \pi i(1+B) \\ \phi_-(z) & = & \tilde{\phi}(z) + \pi i(1+A+B) \end{array} \right\} && \text{ for $z$ on } \gamma_{1}^+.\end{aligned}$$ Observe also that by construction both $\phi$ and $\tilde{\phi}$ have negative real parts in the bounded components $\Omega_{-1}$ and $\Omega_{1}$ of ${{\mathbb{C}}}\setminus \Gamma$ (where defined) and positive real part in $\Omega_\infty$ (with the appropriate cuts). First transformations $Y \mapsto U$ ----------------------------------- Now we introduce a new matrix valued function $U$ by $$\label{defUz} U(z) = \kappa^{n\sigma_3}\, w(\zeta_-; An ,Bn )^{-\sigma_3/2}\, Y(z) \, G(z)^{-n\sigma_3}\, w(\zeta_-; An ,Bn )^{\sigma_3/2},$$ where $\sigma_3 =\begin{pmatrix} 1 & 0 \\ 0& -1 \end{pmatrix}$ is the Pauli matrix, and for any non-zero $x$, $ x^{\sigma_3}=\begin{pmatrix} x & 0 \\ 0& 1/x \end{pmatrix} $. Here $G$ is the function introduced in (\[def\_G\]), and $\kappa$ is the limit defined in (\[kappa\]). Then $U$ satisfies the Riemann-Hilbert problem 1. $U$ is analytic on ${{\mathbb{C}}}\setminus \Gamma$. 2. $U$ has continuous boundary values on $\Gamma \setminus \{\zeta_\pm\}$ such that $$\label{jumpU} U_{+}(z) = U_{-}(z) \begin{pmatrix} \exp(2n \phi_+(z)) & d_s \\ 0 & \exp(-2n \phi_+(z)) \end{pmatrix} \qquad \mbox{for } z\in \Gamma_s^o, \quad s \in \{L,C,R\}.$$ 3. $ U(z) = I + O\left(\frac{1}{z}\right)$ as $z \to \infty$. 4. $U(z)$ remains bounded as $z \to \zeta_{\pm}$. To obtain the jumps in (\[jumpU\]) we used the relations (\[G+timesG-\]) and (\[G+divideG-\]). For the asymptotic behavior in (c) we used the limit (\[kappa\]). We use the following factorizations of the jump matrices in (\[jumpU\]) $$\label{factorjump1} \begin{pmatrix} e^{2n\phi_+} & d_C \\ 0 & e^{-2n\phi_+} \end{pmatrix} = \begin{pmatrix} 1 & 0 \\ \frac{1}{d_C} e^{2n\phi_-} & 1 \end{pmatrix} \begin{pmatrix} 0 & d_C \\ - \frac{1}{d_C} & 0 \end{pmatrix} \begin{pmatrix} 1 & 0 \\ \frac{1}{d_C} e^{2n\phi_+} & 1 \end{pmatrix}$$ and $$\label{factorjump2} \begin{pmatrix} e^{2n\phi} & d_s \\ 0 & e^{-2n\phi} \end{pmatrix} = \begin{pmatrix} 1 & 0 \\ \frac{1}{d_s} e^{-2n\phi} & 1 \end{pmatrix} \begin{pmatrix} e^{2n\phi} & d_s \\ -\frac{1}{d_s} & 0 \end{pmatrix}, \quad \text{ for } s=L,R,$$ in order to define the next transformation. Second transformation $U \mapsto T$ ----------------------------------- The trajectories $\Gamma$ and the orthogonal trajectories $\Gamma^{\perp}$ divide the complex plane into six domains, which we number from left to right as domains I, II, III, IV, V and VI, see Fig. \[fig\_regions1\]. We define $\tilde{T}$ in each of these six domains separately. We put $$\begin{aligned} \tilde{T} &= U \label{tildeT1} \begin{pmatrix} 1 & 0 \\ \frac{1}{d_L} e^{-2n\phi} & 1 \end{pmatrix} \quad \text{in domain I}, \\ \tilde{T} &= U \label{tildeT2} \begin{pmatrix} 1 & 0 \\ \frac{1}{d_R} e^{-2n\phi} & 1 \end{pmatrix} \quad \text{in domain VI}, \\ \tilde{T} &=U \label{tildeT3} \begin{pmatrix} 1 & 0 \\ -\frac{1}{d_L} e^{2n\phi} & 1 \end{pmatrix} \begin{pmatrix} 0 & -d_L \\ \frac{1}{d_L} & 0 \end{pmatrix}= U \begin{pmatrix} 0 & -d_L \\ \frac{1}{d_L} & e^{2n\phi} \end{pmatrix} \quad \mbox{in domain II,} \\ \tilde{T} &=U \label{tildeT4} \begin{pmatrix} 1 & 0 \\ -\frac{1}{d_R} e^{2n\phi} & 1 \end{pmatrix} \begin{pmatrix} 0 & -d_R \\ \frac{1}{d_R} & 0 \end{pmatrix}= U \begin{pmatrix} 0 & -d_R \\ \frac{1}{d_R} & e^{2n\phi} \end{pmatrix} \quad \text{in domain V,} \\ \tilde{T} &= U \label{tildeT5} \begin{pmatrix} 1 & 0 \\ \frac{1}{d_C} e^{2n\phi} & 1 \end{pmatrix} \begin{pmatrix} 0 & -d_L \\ \frac{1}{d_L} & 0 \end{pmatrix} \quad \mbox{in domain III,} \\ \tilde{T} & = U \label{tildeT6} \begin{pmatrix} 1 & 0 \\ -\frac{1}{d_C} e^{2n\phi} & 1 \end{pmatrix} \begin{pmatrix} 0 & -d_R \\ \frac{1}{d_R} & 0 \end{pmatrix} \quad \mbox{in domain IV}.\end{aligned}$$ Since $\phi(z)$ behaves like $\frac{A+B+2}{2} \log z$ as $z \to \infty$, we have $\|e^{-2n \phi(z)}\| \sim \|z\|^{-(A +B +2 )n}$, so that $$\begin{pmatrix} 1 & 0 \\ \frac{1}{d_s}\, e^{-2n\phi(z)} & 1 \end{pmatrix} = I + O(1/z) \quad \mbox{ as } z \to \infty.$$ Thus $\tilde{T}(z) = I + O(1/z)$ as $z \to \infty$. By definition, $\tilde{T}$ is analytic in $\mathbb C \setminus (\Gamma \cup \Gamma^{\perp})$. However, we have arranged our transformation in a way that the jumps on $\Gamma_L$ and $\Gamma_R$ disappear (due to the factorization (\[factorjump2\]) and the definition of $\tilde{T}$) so $\tilde{T}$ is analytic in $\mathbb C \setminus (\Gamma_C \cup \Gamma^{\perp})$. We compute the jumps on $\Gamma_C \cup \Gamma^{\perp}$ with the convention that these curves are oriented as shown in Fig. \[fig\_regions1\]. Straightforward computations then show that $$\begin{aligned} \tilde{T}_+ &= \tilde{T}_- \begin{pmatrix} 0 & \frac{d_Ld_R}{d_C} \\ - \frac{d_C}{d_Ld_R} & 0 \end{pmatrix} \quad \mbox{ on } \Gamma_C,\\ \tilde{T}_+ &= \tilde{T}_- \begin{pmatrix} 1 & 0 \\ \frac{1}{d_R} e^{-2n\phi_+} - \frac{1}{d_L} e^{-2n\phi_-} & 1 \end{pmatrix} \quad \mbox{ on } \gamma_{\infty}^+ \cup \gamma_{\infty}^-, \\ \tilde{T}_+ &= \tilde{T}_- \begin{pmatrix} 1 & - d_L e^{2n\phi_-} - \frac{d_L^2}{d_C} e^{2n\phi_+} \\ 0 & 1 \end{pmatrix} \quad \mbox{ on } \gamma_{-1}^+ \cup \gamma_{-1}^-,\\ \tilde{T}_+ &= \tilde{T}_- \begin{pmatrix} 1 & -\frac{d_R^2}{d_C} e^{2n\phi_-} + d_R e^{2n\phi_+} \\ 0 & 1 \end{pmatrix} \quad \mbox{ on } \gamma_{1}^+ \cup \gamma_{1}^-.\end{aligned}$$ Since $\phi$ is analytic across the curves $\gamma_{j}^-$, the jumps on these curves simplify to (we also use (\[constants2\])) $$\begin{aligned} \tilde{T}_+ &= \tilde{T}_- \begin{pmatrix} 1 & 0 \\ - \frac{d_C}{d_Ld_R} e^{-2n\phi} & 1 \end{pmatrix} \quad \mbox{ on } \gamma_{\infty}^- , \\ \tilde{T}_+ &= \tilde{T}_- \begin{pmatrix} 1 & -\frac{d_L d_R}{d_C} e^{2n\phi} \\ 0 & 1 \end{pmatrix} \quad \mbox{ on } \gamma_{-1}^- \cup \gamma_{1}^-.\end{aligned}$$ If we now express the jumps on the contours $\gamma_{j}^+$ in terms of $\tilde{\phi}$, see (\[phirelation1\])–(\[phirelation3\]), they look as those on the lower half plane, but with $\phi$ replaced by $\tilde \phi$: $$\begin{aligned} \tilde{T}_+ &= \tilde{T}_- \begin{pmatrix} 1 & 0 \\ - \frac{d_C}{d_Rd_L} e^{-2n\tilde \phi} & 1 \end{pmatrix} \quad \mbox{ on } \gamma_{\infty}^+ , \\ \tilde{T}_+ &= \tilde{T}_- \begin{pmatrix} 1 & -\frac{d_L d_R}{d_C} e^{2n\tilde \phi} \\ 0 & 1 \end{pmatrix} \quad \mbox{ on } \gamma_{-1}^+ \cup \gamma_{1}^+.\end{aligned}$$ Now with $\tau$ such that $$\label{def tau} \tau^2 = \frac{d_L d_R}{d_C}$$ we define $T$ by $$\label{defT} T = \begin{pmatrix} \tau^{-1} & 0 \\ 0 & \tau \end{pmatrix} \tilde{T} \begin{pmatrix} \tau & 0 \\ 0 & \tau^{-1} \end{pmatrix}.$$ The effect on the jump matrices is that the $(1,2)$ entries are multiplied by $\tau^{-2}$ and the $(2,1)$ entries are multiplied by $\tau^2$. So $T$ satisfies the following Riemann-Hilbert problem: 1. $T$ is analytic on ${{\mathbb{C}}}\setminus ( \Gamma_C \cup \Gamma^\perp)$. 2. $T$ has continuous boundary values on $(\Gamma_C \cup \Gamma^\perp) \setminus \{\zeta_\pm\}$ such that $$\begin{aligned} T_+ &= T_- \label{jumpT1} \begin{pmatrix} 0 & 1 \\ -1 & 0 \end{pmatrix} \quad \text{ on } \Gamma_C, \\ T_+ & = T_- \label{jumpT2} \begin{pmatrix} 1 & 0 \\ - e^{-2n\phi} & 1 \end{pmatrix} \quad \text{ on } \gamma_{\infty}^- ,\\ T_+ & = T_- \label{jumpT3} \begin{pmatrix} 1 & - e^{2n\phi} \\ 0 & 1 \end{pmatrix} \quad \text{ on } \gamma_{-1}^- \cup \gamma_{1}^- ,\\ T_+ & = T_- \label{jumpT4} \begin{pmatrix} 1 & 0 \\ - e^{-2n\tilde{\phi}} & 1 \end{pmatrix} \quad \text{ on } \gamma_{\infty}^+ ,\\ T_+ & = T_- \label{jumpT5} \begin{pmatrix} 1 & - e^{2n\tilde{\phi}} \\ 0 & 1 \end{pmatrix} \quad \text{ on } \gamma_{-1}^+ \cup \gamma_{1}^+. \end{aligned}$$ 3. $T(z) = I + O\left(\frac{1}{z}\right)$ as $z \to \infty$. 4. $T(z)$ remains bounded as $z \to \zeta_{\pm}$. The problem for $T$ is by now relatively standard. However, compared with earlier works, the triangularity of the jump matrices on the curves $\gamma_j^{\pm}$ is reversed. The inverse transposed matrix $T^{-t}$ satisfies the jumps $$\begin{aligned} T_+^{-t} &= T_-^{-t} \begin{pmatrix} 0 & 1 \\ -1 & 0 \end{pmatrix} \quad \text{ on } \Gamma_C, \\ T_+^{-t} & = T_-^{-t} \begin{pmatrix} 1 & e^{-2n\phi}\\ 0 & 1 \end{pmatrix} \quad \text{ on } \gamma_{\infty}^- ,\\ T_+^{-t} & = T_-^{-t} \begin{pmatrix} 1 & 0\\ e^{2n\phi} & 1 \end{pmatrix} \quad \text{ on } \gamma_{-1}^- \cup \gamma_{1}^- ,\\ T_+^{-t} & = T_-^{-t} \begin{pmatrix} 1 & e^{-2n\tilde{\phi}}\\ 0 & 1 \end{pmatrix} \quad \text{ on } \gamma_{\infty}^+ ,\\ T_+ ^{-t}& = T_-^{-t} \begin{pmatrix} 1 & 0 \\ e^{2n\tilde{\phi}} & 1 \end{pmatrix} \quad \text{ on } \gamma_{-1}^+ \cup \gamma_{1}^+, \end{aligned}$$ which are exactly of the form considered for example in [@MR2001g:42050; @Kuijlaars/Mclaughlin:01a]. Outside parametrix ------------------ The jump matrices in (\[jumpT2\])–(\[jumpT5\]) are close to the identity matrix if $n$ is large. Therefore we expect that the main term in the asymptotic behavior of $T$ is given by the solution $N$ to the following model Riemann-Hilbert problem: 1. $N$ is analytic in $\mathbb C \setminus \Gamma_C$, 2. $N_+ = N_- \begin{pmatrix} 0 & 1 \\ -1 & 0 \end{pmatrix}$ on $\Gamma_C \setminus \{\zeta_\pm\}$, 3. $N(z) = I + O(1/z)$ as $z \to \infty$. In analogy with the condition (d) in the Riemann-Hilbert problem for $T$ we would like to ask that $N(z)$ remains bounded as $z \to \zeta_{\pm}$. However, this would lead to a Riemann-Hilbert problem with no solution. Instead we allow for moderate singularities of $N$ at $\zeta_{\pm}$: 1. $N(z)=O(\|z-\zeta_{\pm}\|^{-1/4})$ as $z \to \zeta_\pm$. The solution to this problem is given by $$\label{defN} N(z) = \begin{pmatrix} \frac{a(z)+a(z)^{-1}}{2} & \frac{a(z) - a(z)^{-1}}{2i} \\ -\frac{a(z)-a(z)^{-1}}{2i} & \frac{a(z) + a(z)^{-1}}{2} \end{pmatrix}$$ with $$a(z) = \frac{(z-\zeta_-)^{1/4}}{(z-\zeta_+)^{1/4}}, \qquad z \in \mathbb C \setminus \Gamma_C,$$ being analytic and single-valued in ${{\mathbb{C}}}\setminus \Gamma_C$, such that $a(z) \to 1$ as $z \to \infty$, see [@MR2000g:47048; @MR2001f:42037; @MR2001g:42050; @Kuijlaars/Mclaughlin:01a]. In [@Kuijlaars/Mclaughlin:01a] also an alternative expression for $N$ has been established in terms of $R(z):=\sqrt{(z-\zeta_+)(z-\zeta_-)}$: $$\label{alternativeForN} N(z) = \begin{pmatrix} \left(\frac{1+R'(z)}{2}\right)^{1/2} & -\left(\frac{1-R'(z)}{2}\right)^{1/2} \\ \left(\frac{1-R'(z)}{2}\right)^{1/2} & \left(\frac{1+R'(z)}{2}\right)^{1/2}\end{pmatrix}.$$ Local parametrices ------------------ Near the branch points $\zeta_{\pm}$ we construct local parametrices in the same way as done by Deift et al [@MR2001f:42037; @MR2001g:42050; @MR2000g:47048], see also [@Kuijlaars/Mclaughlin:01b; @Kuijlaars/Mclaughlin:01a]. In a neighborhood $U_{\delta} =\{ z \in \mathbb C :\, \|z-\zeta_-\| < \delta\}$ of $\zeta_-$ we construct a $2\times 2$ matrix valued $P$ that is analytic in $U_{\delta} \setminus (\Gamma_C \cup \gamma_{-1}^- \cup \gamma_1^- \cup \gamma_{\infty}^-)$, satisfies the same jump conditions as $T$ does on $U_{\delta} \cap (\Gamma_C \cup \gamma_{-1}^- \cup \gamma_1^- \cup \gamma_{\infty}^-)$ and that matches with $N$ on the boundary $C_{\delta}$ of $U_{\delta}$ up to order $1/n$. ![Conformal mapping $f$. []{data-label="fig_Jacobiconfmap"}](Jacobiconfmap) We need the function $$\label{deff} f(z) = \left[ \frac{3}{2} \phi(z) \right]^{2/3}$$ where the $2/3$rd root is chosen which is real and positive on $\gamma_{\infty}^-$. This is a conformal map from $U_{\delta}$ onto a neighborhood of $0$ provided $\delta > 0$ is small enough. We note that $\gamma_{\infty}^-$ is mapped to the positive real axis, $\Gamma_C$ to (a part of) the negative real axis. Recall that $\phi$ is real and negative on $\gamma_1^-$ and $\gamma_{-1}^-$ and we see that $\gamma_1^-$ is mapped to $\arg w = 2\pi/3$ and $\gamma_{-1}^-$ to $\arg w = - 2\pi/3$ (Fig. \[fig\_Jacobiconfmap\]). Then the Riemann-Hilbert problem for $P$ is solved by (cf.[@Kuijlaars/Mclaughlin:01a]) $$\label{defP} P(z) = \left[ E(z) \Psi(n^{2/3} f(z)) e^{n\phi(z)\sigma_3} \right]^{-t},$$ where $$\label{defE} E(z) = \sqrt{\pi} e^{\frac{\pi i}{6}} \left(\begin{array}{cc} 1 & -1 \\ -i & -i \end{array} \right) \left(\frac{n^{1/6} f(z)^{1/4}}{a(z)} \right)^{\sigma_3},$$ and $\Psi$ is built out of the Airy function ${\mathop{\rm Ai}}$ and its derivative ${\mathop{\rm Ai}}'$ as follows $$\label{defPsi1} \Psi(s) = \left\{ \begin{array}{ll} \left(\begin{array}{cc} {\mathop{\rm Ai}}(s) & {\mathop{\rm Ai}}(\omega^2 s) \\ {\mathop{\rm Ai}}'(s) & \omega^2 {\mathop{\rm Ai}}'(\omega^2 s) \end{array} \right) e^{- \frac{\pi i}{6} \sigma_3} & \mbox{for } 0 < \arg s < 2 \pi/3, \\[10pt] \left(\begin{array}{cc} {\mathop{\rm Ai}}(s) & {\mathop{\rm Ai}}(\omega^2 s) \\ {\mathop{\rm Ai}}'(s) & \omega^2 {\mathop{\rm Ai}}'(\omega^2 s) \end{array} \right) e^{- \frac{\pi i}{6} \sigma_3} \left(\begin{array}{cc} 1 & 0 \\ -1 & 1 \end{array} \right) & \mbox{for } 2\pi/3 < \arg s < \pi, \\[10pt] \left(\begin{array}{cc} {\mathop{\rm Ai}}(s) & -\omega^2 {\mathop{\rm Ai}}(\omega s) \\ {\mathop{\rm Ai}}'(s) & -{\mathop{\rm Ai}}'(\omega s) \end{array} \right) e^{-\frac{\pi i}{6} \sigma_3} \left(\begin{array}{cc} 1 & 0 \\ 1 & 1 \end{array} \right) & \mbox{for } -\pi < \arg s < -2\pi/3, \\[10pt] \left(\begin{array}{cc} {\mathop{\rm Ai}}(s) & -\omega^2 {\mathop{\rm Ai}}(\omega s) \\ {\mathop{\rm Ai}}'(s) & -{\mathop{\rm Ai}}'(\omega s) \end{array} \right) e^{- \frac{\pi i}{6} \sigma_3} & \mbox{for } -2\pi/3 < \arg s < 0, \end{array} \right.$$ with $\omega = e^{2\pi i/3}$. Note that we take the inverse transpose in (\[defP\]), which is absent in the construction in [@Kuijlaars/Mclaughlin:01a]. This is of course due to the fact that the Riemann-Hilbert problem for $T^{-t}$ is comparable to the Riemann-Hilbert problem found in [@Kuijlaars/Mclaughlin:01a], see the remark at the end of subsection 5.4. A similar construction yields a parametrix $\tilde{P}$ in a neighborhood $\tilde{U}_{\delta} = \{ z : \|z-\zeta_+\| < \delta\}$. Final transformation $T \mapsto S$ ---------------------------------- ![Contour $\Gamma_S$ for the Riemann-Hilbert problem of $S$.[]{data-label="fig_Jac_fig36_finaljumps"}](Jac_fig36_finaljumps) The final transformation $T \mapsto S$ is $$\begin{aligned} \label{defS1} S = TN^{-1} && \text{ outside the disks $U_{\delta}$ and $\tilde{U}_{\delta}$}, \\ \label{defS2} S = TP^{-1} && \text{ inside the disk $U_{\delta}$}, \\ \label{defS3} S = T \tilde{P}^{-1} && \text{ inside the disk $\tilde{U}_{\delta}$}.\end{aligned}$$ Then by construction, $S$ has jumps on the circles $C_{\delta}=\partial U_{\delta}$ and $\tilde{C}_{\delta}=\partial \tilde{U}_{\delta}$ as well as on $\Gamma_C \cup \Gamma^\perp$. Since the jumps of $T$ and $N$ on $\Gamma_C$ agree, we have that $S$ is analytic across the part of $\Gamma_C$ outside the disks $U_{\delta}$ and $\tilde{U}_{\delta}$. Similarly, the jumps of $T$ and $P$ agree inside the disk $U_{\delta}$, and the jumps of $T$ and $\tilde{P}$ agree inside the disk $\tilde{U}_{\delta}$, so that $S$ is analytic in $U_{\delta}$ and $\tilde{U}_{\delta}$ with possible isolated singularities at $\zeta_{\pm}$. However it follows from the behavior of $T$ and $N$ near $\zeta_{\pm}$ that the singularities are removable. Thus $S$ solves the following Riemann-Hilbert problem. 1. $S$ is analytic on $\mathbb C \setminus \Gamma_S$, where $\Gamma_S$ consists of the circles $C_{\delta}$ and $\tilde{C}_{\delta}$, and of the parts of $\gamma_{-1}$, $\gamma_1$ and $\gamma_{\infty}$ outside the disks, see Fig. \[fig\_Jac\_fig36\_finaljumps\]. 2. $S$ has continuous boundary values on $\Gamma_S$ such that $$\begin{aligned} S_+ & = S_- \label{jumpS1} N \begin{pmatrix} 1 & 0 \\ - e^{-2n\phi} & 1 \end{pmatrix} N^{-1} \quad \text{ on } \gamma_{\infty}^- \setminus \tilde{U}_{\delta},\\ S_+ & = S_- \label{jumpS2} N \begin{pmatrix} 1 & - e^{2n\phi} \\ 0 & 1 \end{pmatrix} N^{-1} \quad \text{ on } \left(\gamma_{-1}^- \cup \gamma_{1}^- \right) \setminus \tilde{U}_{\delta},\\ S_+ & = S_- \label{jumpS3} N \begin{pmatrix} 1 & 0 \\ - e^{-2n\tilde{\phi}} & 1 \end{pmatrix} N^{-1} \quad \text{ on } \gamma_{\infty}^+\setminus U_{\delta},\\ S_+ & = S_- \label{jumpS4} N \begin{pmatrix} 1 & - e^{2n\tilde{\phi}} \\ 0 & 1 \end{pmatrix} N^{-1} \quad \text{ on } \left( \gamma_{-1}^+ \cup \gamma_{1}^+\right) \setminus U_{\delta}, \\ S_+ & = S_- \label{jumpS5} P N^{-1} \qquad \qquad \text{ on } C_{\delta}, \\ S_+ & = S_- \label{jumpS6} \tilde{P} N^{-1} \qquad \qquad \text{ on } \tilde{C}_{\delta}.\end{aligned}$$ 3. $S(z) = I + O\left(\frac{1}{z}\right) $ as $z \to \infty$. Asymptotics: Proofs of the theorems =================================== Asymptotics of $S$ ------------------ The analysis in the last section is done for fixed values of $A$, $B$, and $n$. All the transformations are exact for finite $n$. It is now our aim to let $n \to \infty$ and control the jump matrices in the Riemann-Hilbert problem for $S$. We want to do it in a way which is valid locally uniformly for parameters $A$ and $B$ satisfying (\[ABinequalities\]). Then first of all we should study the dependence of the contour $\Gamma_S$ on the parameters $A$ and $B$. Note that $\Gamma_S$ does not depend on $n$, but it does depend on $A$ and $B$. In fact, we have that $\Gamma^\perp$ is completely determined by $A$ and $B$, while the radius $\delta$ of the circles around $\zeta_{\pm}$ is only restricted by the requirement that the mapping $f$ from (\[deff\]) is a conformal mapping on $U_{\delta}$. With that in mind, it is clear that we may assume that the curve $\Gamma_S$ depends on $A$ and $B$ in a continuous way. Now we can see what happens with the jump matrices in (\[jumpS1\])–(\[jumpS6\]) as $n \to \infty$. On $\gamma_{\infty}^- \setminus \tilde{U}_{\delta}$ we have that ${\mathop{\rm Re}}\phi$ is strictly positive. Hence the jump matrix in (\[jumpS1\]) is $I + O(e^{-cn})$ as $n \to \infty$, uniformly on $\gamma_{\infty}^- \setminus \tilde{U}_{\delta}$. This estimate is also valid uniformly for $A$ and $B$ in compact subsets of the set $$\label{ABset} \{ (A,B) \mid -1 < A < 0,\ -1 < B < 0,\ -2 < A+B< -1 \}.$$ Similarly, the jump matrices in (\[jumpS2\])–(\[jumpS4\]) are $I + O(e^{-cn})$ as $n \to \infty$, uniformly on the respective contours and uniformly for $A$ and $B$ in compact subsets of (\[ABset\]). For (\[jumpS5\]) and (\[jumpS6\]) we make use of the matching conditions $$\label{matching1} P(z) = \left(I + O\left(\frac{1}{n}\right)\right) N(z) \qquad \mbox{uniformly for } z \in C_{\delta}.$$ and $$\label{matching2} \tilde{P}(z) = \left(I + O\left(\frac{1}{n}\right)\right) N(z) \qquad \mbox{uniformly for } z \in \tilde{C}_{\delta}.$$ So that the jump matrices in (\[jumpS5\]) and (\[jumpS6\]) are $I + O(1/n)$ as $n \to \infty$, uniformly on the two circles. A closer analysis also reveals that the $O$-terms in (\[matching1\]) and (\[matching2\]) are valid uniformly for $A$ and $B$ in compact subsets of (\[ABset\]). So the conclusion is that all jumps in (\[jumpS1\])–(\[jumpS6\]) are $I + O(1/n)$ uniformly for $z$ on $\Gamma_S$, and uniformly for $A$ and $B$ in compact subsets of (\[ABset\]). Then arguments such as in [@MR2000g:47048; @MR2001f:42037; @MR2001g:42050] lead to the following conclusion. We have that $$\label{asympS} S(z) = I + O\left(\frac{1}{n}\right)$$ uniformly for $z \in \mathbb C \setminus \Gamma_S$ and uniformly for $A$ and $B$ in compact subsets of the set [(\[ABset\])]{}. The estimate (\[asympS\]) is the basic asymptotic result. Unravelling the sequence of transformations $Y \mapsto U \mapsto \tilde{T} \mapsto T \mapsto S$, we obtain asymptotic formulas for $Y$ in any region of the complex plane. In this way we obtain the asymptotic formulas for $\widehat{P}_n = Y_{11}$. Proof of Theorem \[maintheorem\] -------------------------------- of Theorem \[maintheorem\]. Suppose $A$ and $B$ satisfy (\[ABinequalities\]) and let $n \in \mathbb N$ such that $An$, $Bn$, and $(A+B)n$ are non-integers. Then we have $\widehat{P}_n^{(An,Bn)}(z) = Y_{11}(z)$ by (\[formulaY\]). For $z$ in domain I away from the branch points, we get by using (\[defUz\]), (\[tildeT1\]), (\[defT\]) and (\[defS1\]), $$\begin{aligned} \nonumber Y_{11}(z) &= & \left( \frac{G(z)}{\kappa} \right)^n U_{11}(z) \\ & = & \nonumber \left( \frac{G(z)}{\kappa} \right)^n \left( \tilde{T}_{11}(z) - \frac{1}{d_L} e^{-2n\phi(z)} \tilde{T}_{12}(z) \right) \\ & = & \nonumber \left( \frac{G(z)e^{-\phi(z)}}{\kappa} \right)^n \left( e^{n \phi(z)}T_{11}(z) - \frac{d_R}{d_C} e^{-n\phi(z)} T_{12}(z) \right) \\ & = & \label{Y11formula1} \left( \frac{G(z)e^{-\phi(z)}}{\kappa} \right)^n \left( e^{n\phi(z)}(SN)_{11}(z) - \frac{d_R}{d_C} e^{-n\phi(z)} (SN)_{12}(z) \right).\end{aligned}$$ Since $S = I + O(\frac{1}{n})$ and since the entries of $N$ are bounded and bounded away from zero away from the branch points, we have that $$\label{SN11andSN12} (SN)_{11} = N_{11}(I + O(1/n)) \quad \text{ and } (SN)_{12} = N_{12}(I + O(1/n)).$$ Next we recall that for $z$ in domain I, $$\frac{G'(z)}{G(z)} = g'(z) = \frac{A+B+2}{2} \frac{R(z)}{z^2-1} - \frac{A/2}{z-1} - \frac{B/2}{z+1},$$ so that $$\log G(z) = \phi(z) - \frac{A}{2} \log(z-1) - \frac{B}{2} \log(z+1) + const.$$ Thus there is a constant $c$ such that $$\label{relationc} \frac{G(z)e^{-\phi(z)}}{\kappa} = e^{-c} (z-1)^{-A/2} (z+1)^{-B/2} \qquad \text{for $z$ in domain I}.$$ Since $Y_{11}(z)$ is a monic polynomial of degree $n$, we find by letting $z \to \infty$ in (\[Y11formula1\]) and using (\[defc\]) and (\[relationc\]), that $c$ should be as defined in (\[defc\]). Plugging (\[SN11andSN12\]), (\[relationc\]) and the formulas (\[constants\]) for $d_R$ and $d_C$ into formula (\[Y11formula1\]) we obtain (\[asformleft\]) for $z$ in domain I. For $z$ in domain II away from the branch points, we find in the same way $$\begin{aligned} Y_{11}(z) \label{Y11formula2} & = & \left( \frac{G(z)e^{\phi(z)}}{\kappa} \right)^n \left( e^{n\phi(z)} (SN)_{11}(z) - \frac{d_R}{d_C} e^{-n\phi(z)}(SN)_{12}(z) \right)\end{aligned}$$ Since $G_+ = G_- e^{-2\phi}$ on $\Gamma_L$, see (\[G+divideG-\]), we have that $G e^\phi$ is the analytic continuation of $G e^{-\phi}$ into domain II. So we have by (\[relationc\]) $$\label{relationc2} \frac{G(z) e^{\phi}(z)}{\kappa} = e^{-c} (z-1)^{-A/2} (z+1)^{B/2} \qquad \text{for $z$ in domain II}.$$ Then using (\[constants\]), (\[SN11andSN12\]), and (\[relationc2\]) in (\[Y11formula2\]), we obtain (\[asformleft\]) for $z$ in domain II. For $z$ in domain III away from the branch points, we obtain $$\begin{aligned} \label{Y11formula3} Y_{11}(z) &= & \left( \frac{G(z)e^{\phi(z)}}{\kappa} \right)^n \left( - \frac{d_L}{d_C} e^{n\phi(z)} (SN)_{11}(z) - \frac{d_R}{d_C} e^{-n\phi(z)} (SN)_{12}(z) \right).\end{aligned}$$ Again we use (\[constants\]), (\[SN11andSN12\]), and (\[relationc2\]) to obtain (\[asformmiddle1\]) from (\[Y11formula3\]) for $z$ in domain III. The proofs of the formulas (\[asformmiddle2\]) and (\[asformright\]) for $z$ in the other domains IV, V, and VI are the same. We have derived the formulas (\[asformleft\])–(\[asformright\]) under the assumption that $n$ is such that $An$, $Bn$, and $(A+B)n$ are non-integers. Since the formulas hold uniformly in $A$ and $B$ in compact subsets of (\[ABset\]) and $\widehat{P}_n^{(An,Bn)}$ depends continuously on $A$ and $B$, they continue to hold if $An$ or $Bn$ is an integer. However, we cannot allow $(A+B)n$ to be an integer, since then there is a reduction in the degree of $P_n^{(An,Bn)}$ and we cannot normalize the Jacobi polynomial to be monic. This completes the proof of Theorem \[maintheorem\]. Proof of Theorem \[asymnearbranch\] ----------------------------------- of Theorem \[asymnearbranch\]. Suppose $A$ and $B$ satisfy (\[ABinequalities\]) and let $n \in \mathbb N$ such that $An$, $Bn$, and $(A+B)n$ are non-integers. Then we have $\widehat{P}_n^{(An,Bn)}(z) = Y_{11}(z)$ by (\[formulaY\]). Let $z \in U_{\delta}$ be in domain VI. Following the transformations (\[defUz\]), (\[tildeT6\]), (\[defT\]), we see that $$\begin{pmatrix} Y_{11}(z) \\ * \end{pmatrix} = \left(\frac{G(z) e^{-\phi(z)}}{\kappa} \right)^{n} T(z) \begin{pmatrix} e^{n\phi(z)} \\ - \frac{d_L}{d_C} e^{-n\phi(z)} \end{pmatrix},$$ where $$ denotes an unimportant unspecified entry. For $z$ in domain VI, we have (\[relationc\]) so that $$\begin{pmatrix} Y_{11}(z) \\ \end{pmatrix} = e^{-nc} (z-1)^{-An/2} (z+1)^{-Bn/2} T(z) \begin{pmatrix} e^{n\phi(z)} \\ - \frac{d_L}{d_C} e^{-n\phi(z)} \end{pmatrix}.$$ Since $z$ belongs to $U_{\delta}$, we have $T(z) = S(z)P(z)$ by (\[defS2\]). By (\[defP\]) we have $P(z) = E^{-t}(z) \Psi^{-t}(s) e^{-n\phi(z) \sigma_3}$ where $s = n^{2/3} f(z)$. Thus $$\label{Y11nearbranch} \begin{pmatrix} Y_{11}(z) \\ * \end{pmatrix} = e^{-nc} (z-1)^{-An/2} (z+1)^{-Bn/2} S(z)E^{-t}(z) \Psi^{-t}(s)\begin{pmatrix} 1 \\ - \frac{d_L}{d_C} \end{pmatrix}.$$ From (\[defE\]) we see that $$\label{defE2} E^{-t}(z)=\frac{1}{2\sqrt{\pi}}\, e^{-\pi i/6} \begin{pmatrix} 1 & -1 \\ i & i \end{pmatrix}\, \left( \frac{a(z)}{s^{1/4}}\right)^{\sigma_3}.$$ Furthermore, we have $0 < \arg s < \pi/3$ for $s = n^{2/3} f(z)$ since $z$ is in domain VI, so that we use the formula (\[defPsi1\]) to evaluate $\Psi^{-t}(s)$. Taking into account that $$\det \left(\begin{array}{cc} {\mathop{\rm Ai}}(s) & {\mathop{\rm Ai}}(\omega^2 s) \\ {\mathop{\rm Ai}}'(s) & \omega^2 {\mathop{\rm Ai}}'(\omega^2 s) \end{array} \right) =\frac{1}{2\pi }\, e^{\pi i/6},$$ we have from (\[defE2\]) and (\[defPsi1\]), $$E^{-t}(z) \Psi^{-t}(s) = \sqrt{\pi} e^{\pi i/6} \begin{pmatrix} -i & i \\ 1 & 1 \end{pmatrix} \left( \frac{a(z)}{s^{1/4}}\right)^{\sigma_3} \begin{pmatrix} \omega^2 {\mathop{\rm Ai}}'(\omega^2 s) & - {\mathop{\rm Ai}}'(s) \\ - {\mathop{\rm Ai}}(\omega^2 s) & {\mathop{\rm Ai}}(s) \end{pmatrix} e^{\pi i/6 \sigma_3}.$$ Plugging this into (\[Y11nearbranch\]) we get $$\begin{pmatrix} Y_{11}(z) \\ * \end{pmatrix} = e^{-nc} (z-1)^{-An/2}(z+1)^{-Bn/2} \sqrt{\pi} S(z) \begin{pmatrix} -i & i \\ 1 & 1 \end{pmatrix} \left( \frac{a(z)}{s^{1/4}} \right)^{\sigma_3} \begin{pmatrix} - \left(-\frac{d_L}{d_C} {\mathop{\rm Ai}}(s) + \omega^2 {\mathop{\rm Ai}}(\omega^2 s)\right)' \\ -\frac{d_L}{d_C} {\mathop{\rm Ai}}(s) + \omega^2 {\mathop{\rm Ai}}(\omega^2 s) \end{pmatrix}$$ where the prime denotes the derivative with respect to $s$. Comparing with (\[constants\]) and (\[Airycombination\]) we see that $${{\mathcal A}}(s) = {{\mathcal A}}(s;A,B,n) = - \frac{d_L}{d_C} {\mathop{\rm Ai}}(s) + \omega^2 {\mathop{\rm Ai}}(\omega^2s).$$ Thus $$\label{asympLocal} \begin{pmatrix} Y_{11}(z) \\ * \end{pmatrix} = e^{-nc} (z-1)^{-An/2} (z+1)^{-Bn/2} \sqrt{\pi} S(z) \begin{pmatrix} i \frac{s^{1/4}}{a(z)} {{\mathcal A}}(s) + i \frac{a(z)}{s^{1/4}} {{\mathcal A}}'(s) \\[10pt] \frac{s^{1/4}}{a(z)} {{\mathcal A}}(s) - \frac{a(z)}{s^{1/4}} {{\mathcal A}}'(s) \end{pmatrix}$$ We derived the formula (\[asympLocal\]) for $z \in U_{\delta}$ in the domain VI. Similar calculations for $z \in U_{\delta}$ in the other domains give the same result, so (\[asympLocal\]) is valid in the full neighborhood $U_{\delta}$ of $\zeta_-$. Now it remains to recall that $Y_{11} = \widetilde{P}_n^{(An,Bn)}$ and that $S = I + O(1/n)$ as $n \to \infty$ in order to obtain (\[asformnearzeta-\]). We have derived (\[asformnearzeta-\]) under the assumption that $An$ and $Bn$ are non-integers. Since the formula holds uniformly for $A$ and $B$ in compact subsets of (\[ABset\]) and $\widetilde{P}_n^{(An,Bn)}$ depends continuously on $A$ and $B$, they continue to hold if $An$ or $Bn$ is an integer. This completes the proof of Theorem \[asymnearbranch\]. Proof of Theorem \[theoremweak2\] --------------------------------- of Theorem \[theoremweak2\]. Conclusions of Theorem \[theoremweak2\] are based upon the strong asymptotics obtained in Theorem \[maintheorem\]. We let $(\alpha_n)$ and $(\beta_n)$ be two sequences such that $\alpha/n \to A$ and $\beta_n/n \to B$ where $A$ and $B$ satisfy the inequalities (\[ABinequalities\]) and we assume that the limits (\[closetoZ1\])–(\[closetoZ3\]) exist. Taking into account that formula (\[asformleft\]) is uniform in $A$, $B$, for $z$ in domains I and II we have $$\begin{split} \widehat P_n^{(\alpha _n, \beta _n)}(z) = & \ e^{-nc} (z-1)^{-\alpha_n/2} (z+1)^{-\beta_n/2} \\ & \left(e^{n\phi(z)} N_{11}(z) \left(1+O\left(\frac{1}{n}\right)\right) - e^{-\alpha_n\pi i} \frac{\sin (\beta_n \pi)}{\sin((\alpha_n+\beta_n) \pi)}\, e^{-n\phi(z)} N_{12}(z)\left(1+O\left(\frac{1}{n}\right)\right) \right). \end{split}$$ Since the first factors in the right hand side have no zeros in domains I–II, $z$ is a zero of $P_n^{(\alpha _n, \beta _n)}$ only if $$\label{asymFor Zeros} e^{2n\phi(z)}= e^{-\alpha_n \pi i} \frac{\sin (\beta_n\pi)}{\sin((\alpha_n+\beta_n) \pi)} \frac{N_{12}(z)}{N_{11}(z)} \left(1+O\left(\frac{1}{n}\right)\right).$$ But $N_{12}/N_{11}$ is uniformly bounded and uniformly bounded away from zero, if we stay away from the branch points $\zeta_\pm$. Thus, taking the absolute values in (\[asymFor Zeros\]), we see that the zeros in domains I–II away from the branch points must satisfy $$2 {\mathop{\rm Re}}\phi(z) = \frac{1}{n} \log \left\| \frac{\sin (\beta_n \pi)}{\sin((\alpha_n+\beta_n)\pi)}\right\| + O\left(\frac{1}{n}\right).$$ Analogously, the zeros of $P_n^{(\alpha_n, \beta_n)}$ in the other domains III–VI away from the branch points satisfy $$2 {\mathop{\rm Re}}\phi(z) = \frac{1}{n} \log \left\| \frac{ \sin(\beta_n \pi)}{ \sin(\alpha_n \pi)}\right\| + O\left(\frac{1}{n} \right), \qquad \textrm{for $z$ in domain III},$$ $$2 {\mathop{\rm Re}}\phi(z) = \frac{1}{n} \log \left\| \frac{\sin(\alpha_n \pi)}{\sin(\beta_n \pi)} \right\| + O\left(\frac{1}{n} \right), \qquad \textrm{for $z$ in domain IV},$$ $$2 {\mathop{\rm Re}}\phi(z)=\frac{1}{n} \log \left\|\frac{\sin(\alpha_n \pi)}{\sin((\alpha_n+\beta_n)\pi)}\right\| + O\left(\frac{1}{n}\right), \qquad \textrm{for $z$ in domains V--VI}.$$ Furthermore, for any sequence of real numbers $(\kappa_n)$, $$\lim_n \frac{1}{n}\, \log \left\|\sin (\pi \kappa _n) \right\|= \lim_n \frac{1}{n}\, \log \left\|{\mathop{\rm dist}}( \kappa _n, {{\mathbb{Z}}}) \right\|\,,$$ whenever either one of these limits exists. Thus, under the assumptions of the theorem, the zeros of $P_n^{(\alpha_n, \beta_n)}$ away from the branch points must satisfy $$\label{casesZeros} 2 {\mathop{\rm Re}}\phi(z) = r+ O\left(\frac{1}{n}\right), \quad r=\begin{cases} r_{\alpha +\beta }-r_{\beta }, & \text{for $z$ in domains I and II,}\\ r_{\alpha }-r_{\beta }, & \text{for $z$ in domain III,}\\ r_{\beta }-r_{\alpha }, & \text{for $z$ in domain IV,}\\ r_{\alpha +\beta }-r_{ \alpha }, & \text{for $z$ in domains V and VI.} \end{cases}$$ From the definition of the constants (\[closetoZ1\])–(\[closetoZ3\]) it follows that the “generic” case is $$\label{genericCase} r_{\alpha }=r_{\beta }=r_{\alpha +\beta }\,.$$ Recall that ${\mathop{\rm Re}}\phi(z)>0$ in domains I and VI, and ${\mathop{\rm Re}}\phi(z)<0$ in domains II–V. Hence by (\[casesZeros\]), if $z\in {{\mathbb{C}}}\setminus \Gamma$, then $P_n^{(\alpha_n, \beta_n)}(z)\neq 0$ for sufficiently large $n$, which proves that the zeros can accumulate only on $\Gamma$. Next assume we are in case (b) of the Theorem \[theoremweak2\], that is, $$r_{\alpha +\beta }>r_\alpha =r_\beta\,.$$ By (\[casesZeros\]), the zeros cannot accumulate in domains II, III, IV and V, nor on $\Gamma_L\cup \Gamma_R$. In domains I and VI they must satisfy $${\mathop{\rm Re}}\phi(z) = r+ O\left(\frac{1}{n}\right), \quad r= \frac{ r_{\alpha +\beta }-r_{\beta }}{2}=\frac{ r_{\alpha +\beta }-r_{ \alpha }}{2}\,,$$ showing that they must accumulate on the curve $\Gamma_r$ defined in (\[defGammar\]). Hence, in this case the accumulation set belongs to $\Gamma_C \cup \Gamma_r$. The rest of the cases is analyzed in the same fashion. Once we have established that the zeros accumulate along curves in the complex plane, it remains to find the asymptotic zero distribution. We can use the differential equation (see e.g.[@szego:1975 §4.22]) $$(1-z^2)\, y_n''(z) +\left[ \beta _n-\alpha _n-(\alpha _n+\beta _n+2) z \right]\, y_n'(z) + n (n+\alpha _n+\beta _n+1)\, y_n(z)=0$$ satisfied by $y_n=P_n^{(\alpha_n, \beta_n)}$. Rewriting this equation in terms of $h_n=y'_n/(n y_n)$ we reduce it to the Riccati form $$\label{riccati} (1-z^2)\, \left( \frac{1}{n}\, h_n'(z)+h_n^2(z) \right) + \frac{\beta _n-\alpha _n-(\alpha _n+\beta _n+2) z }{n} \, h_n'(z) + \frac{ n+\alpha _n+\beta _n+1 }{n}=0\,.$$ Let $\nu_n$ denote the normalized zero counting measures of $P_n^{(\alpha _n, \beta _n)}$. Using the week compactness of the sequence $( \nu_n )$ we may take a subsequence $\Lambda\subset {{\mathbb{N}}}$ such that $\nu_n$ converge along $\Lambda$ to a certain unit measure $\nu$ in the weak\-topology. By the discussion above, $\nu$ is supported on a finite union of analytic arcs or curves (level sets $\Gamma_r$), and every compact subset of ${{\mathbb{C}}}\setminus {\mathop{\rm supp}}(\nu)$ contains no zeros of $P_n^{(\alpha _n, \beta _n)}$ for $n$ sufficiently large. Hence, $$h_n(z)=\int_\Gamma \frac{d\nu_n(t)}{z-t} \longrightarrow h(z)=\int_\Gamma \frac{d\nu(t)}{z-t}\,, \quad n\in \Lambda\,,$$ locally uniformly in ${{\mathbb{C}}}\setminus {\mathop{\rm supp}}(\nu)$. Taking limits in (\[riccati\]) we obtain that $h$ satisfies the quadratic equation $$(1-z^2)\, h^2(z) + \left[ B-A-(A+B) z \right]\, h(z) + A +B+1 =0\,,$$ so that $$\int_\Gamma \frac{d\nu(t)}{z-t}=\frac{A+B+2}{2}\, \frac{R(z)}{z^2-1}-\frac{1}{2}\, \left( \frac{A}{z-1}+ \frac{B}{z+1}\right)\,, \quad z \in {{\mathbb{C}}}\setminus {\mathop{\rm supp}}(\nu)\,,$$ with $R$ defined in (\[defRz\]) and $\zeta_\pm$ given in (\[zetapm\]). By Sokhotsky-Plemelj’s formulas, on every arc of ${\mathop{\rm supp}}(\nu)$, $$\label{nu} d\nu(z) = \frac{A + B+2}{2 \pi i} \frac{R_+(z)}{z^2-1} \,dz\,.$$ Assume that (\[genericCase\]) holds, so that ${\mathop{\rm supp}}(\nu)\subset \Gamma$. By (\[nu\]), $\nu'=\mu'$ almost everywhere on ${\mathop{\rm supp}}(\nu)$. Taking into account Lemma \[lemma2\] and that $\nu$ is a unit measure it follows that $\nu=\mu$. If $r_{\alpha +\beta }>r_\alpha =r_\beta$ then ${\mathop{\rm supp}}(\nu)\subset \Gamma_C \cup \Gamma_r$, $r=(r_{\alpha +\beta }-r_{ \alpha })/2>0$. Again taking into account Lemma \[lemma2\] and the normalization of $\nu$ it follows that ${\mathop{\rm supp}}(\nu)= \Gamma_C \cup \Gamma_r$. The remaining cases are analyzed analogously. This completes the proof of Theorem \[theoremweak2\]. Acknowledgements {#acknowledgements .unnumbered} ================ This research was partially supported by the Ministry of Science and Technology of Spain through the grant BFM2001-3878-C02-02. A.B.J.K. was also supported by FWO research projects G.0176.02 and G.0455.04. Additionally, A.M.F. acknowledges the support of Junta de Andaluc[í]{}a, Grupo de Investigaci[ó]{}n FQM 0229, and of the Ministry of Education, Culture and Sports of Spain through the grant PR2003–0104. [10]{} , [Handbook of Mathematical Functions]{}, Dover Publications, New York, 1968. , [Sharp constants for rational approximations of analytic functions]{}, Sb. Math. 193 (2002), no. 1-2, 1–72. , [Optimal tail estimates for directed last passage site percolation with geometric random variables]{}, Adv. Theor. Math. Phys. 5 (2001), 1207-1250. , [Strong asymptotics for Jacobi polynomials with varying weights]{}, Methods Appl. Anal. 6 (1999), 39–54. , [On asymptotics of Jacobi polynomials]{}, SIAM J. Math. Anal. 22 (1991), no. 5, 1442–1449. , [Orthogonal Polynomials and Random Matrices: a [R]{}iemann-[H]{}ilbert approach]{}, New York University Courant Institute of Mathematical Sciences, New York, 1999. , [Strong asymptotics of orthogonal polynomials with respect to exponential weights]{}, Comm. Pure Appl. Math. 52 (1999), pp. 1491–1552. , [Uniform asymptotics for polynomials orthogonal with respect to varying exponential weights and applications to universality questions in random matrix theory]{}, Comm. Pure Appl. Math. 52 (1999), pp. 1335–1425. , [A steepest descent method for oscillatory Riemann-Hilbert problems: asymptotics for the MKdV equation]{}, Ann. Math. 137 (1993), 295-368. , [Some New Asymptotic Properties for the Zeros of [J]{}acobi, [L]{}aguerre, and [H]{}ermite Polynomials]{}, Constr. Approx. 11 (1995), pp. 227–238. , [Asymptotic zero distribution of hypergeometric polynomials]{}, Numer. Algorithms 21 (1999), pp. 147–156. height 2pt depth -1.6pt width 23pt, [Trajectories of the zeros of hypergeometric polynomials [$F(-n, b; 2b; z)$]{} for $b< -\frac{1}{2}$]{}, Constr. Approx. 17 (2001), pp. 169–179. , [Zeros of hypergeometric polynomials $F(-n,b;-2n;z)$]{}, J. Approx. Theory 110 (2001), pp. 74–87. , [Zeros of hypergeometric functions]{}, Comput. Methods Func. Theory 1 (2001), pp. 275–287. , [Asymptotic properties of zeros of hypergeometric polynomial]{}, J. Approx. Theory 111 (2001), pp. 329–343. , [The isomonodromy approach to matrix models in [2D]{} quantum gravity]{}, Comm. Math. Phys. 147 (1992), pp. 395–430. , [Strong asymptotics and the limit distribution of the zeros of [J]{}acobi polynomials ${P}_n^{(an+\alpha,bn+\beta)}$]{}, in Progress in Approximation Theory, P. Nevai and A. Pinkus, eds., New York, 1991, Academic Press, pp. 379–404. , [Equilibrium distributions and rate of the rational approximation of analytic functions]{}, Mat. USSR Sbornik 62 (1989), pp. 305–348. Translation from Mat. Sb., Nov. Ser. 134(176), No. 3 (11), pp. 306–352 (1987). , [Semiclassical soliton ensembles for the focusing nonlinear Schr[ö]{}dinger equation]{}, Annals of Mathematics Studies 154, Princeton University Press, Princeton, 2003. , [ Orthogonality of [J]{}acobi polynomials with general parameters]{}, to appear in Electr. Trans. Numer. Anal., preprint math.CA/0301037. , [Asymptotic zero behavior of [L]{}aguerre polynomials with negative parameter]{}, to appear in Constr. Approx., preprint math.CA/0205175. height 2pt depth -1.6pt width 23pt, [[R]{}iemann-[H]{}ilbert analysis for [L]{}aguerre polynomials with large negative parameter]{}, Comput. Meth. Funct. Theory 1 (2001), pp. 205–233. , [The asymptotic zero distribution of orthogonal polynomials with varying recurrence coefficients]{}, J. Approx. Theory 99 (1999), 167–197. , [Quadratic Hermite-Pad[é]{} approximation to the exponential function: a Riemann-Hilbert approach]{}, preprint math.CA/0302357. , [Zeros of [J]{}acobi polynomials with varying non-classical parameters]{}, in Special functions (Hong Kong, 1999), World Sci. Publishing, River Edge, NJ, 2000, pp. 98–113. , [On the zeros of Jacobi polynomials $P_n^{(\alpha_n,\beta_n)}(x)$]{}, Trans. Amer. Math. Soc. 249 (1979), pp. 159–162. , [Asymptotics and Special Functions]{}, Acdemic Press, Boston, 1974. , [Univalent Functions]{}, Vandenhoeck & Ruprecht, Göttingen, 1975. , [Logarithmic Potentials with External Fields]{}, vol. 316 of Grundlehren der Mathematischen Wissenschaften, Springer-Verlag, Berlin, 1997. , [Orthogonal polynomials with complex-valued weight function. [I]{}, [II]{}]{}, Constr. Approx. 2 (1986), pp. 225–240, 241–251. , [Quadratic Differentials]{}, Springer Verlag, Berlin, 1984. , [Orthogonal Polynomials]{}, vol. 23 of Amer. Math. Soc. Colloq. Publ., Amer. Math. Soc., Providence, [RI]{}, fourth ed., 1975. , [Large parameter cases of the [G]{}auss hypergeometric function*]{}, J. Comput. Appl. Math. 153 (2003), 441–462. [^1]: Department of Mathematics, Katholieke Universiteit Leuven, Celestijnenlaan 200 B, 3001 Leuven, Belgium, email: [[email protected]]{} [^2]: University of Almer[í]{}a and Instituto Carlos I de F[í]{}sica Te[ó]{}rica y Computacional, Granada University (SPAIN). Corresponding author, email: [[email protected]]{}	{ "pile_set_name": "ArXiv" }
--- abstract: 'One of the most prominent tools for abstract argumentation is the Dung’s framework, AF for short. It is accompanied by a variety of semantics including grounded, complete, preferred and stable. Although powerful, AFs have their shortcomings, which led to development of numerous enrichments. Among the most general ones are the abstract dialectical frameworks, also known as the ADFs. They make use of the so–called acceptance conditions to represent arbitrary relations. This level of abstraction brings not only new challenges, but also requires addressing existing problems in the field. One of the most controversial issues, recognized not only in argumentation, concerns the support cycles. In this paper we introduce a new method to ensure acyclicity of the chosen arguments and present a family of extension–based semantics built on it. We also continue our research on the semantics that permit cycles and fill in the gaps from the previous works. Moreover, we provide ADF versions of the properties known from the Dung setting. Finally, we also introduce a classification of the developed sub–semantics and relate them to the existing labeling–based approaches.' author: - \| Sylwia Polberg\ Vienna University of Technology\ Institute of Information Systems\ Favoritenstraße 9-11, 1040 Vienna, Austria [^1] bibliography: - 'references.bib' title: 'Extension–based Semantics of Abstract Dialectical Frameworks' --- Introduction ============ Over the last years, argumentation has become an influential subfield of artificial intelligence [@book:argai]. One of its subareas is the abstract argumentation, which became especially popular thanks to the research of Phan Minh Dung [@article:dung]. Although the framework he has developed was relatively limited, as it took into account only the conflict relation between the arguments, it inspired a search for more general models (see [@general] for an overview). Among the most abstract enrichments are the abstract dialectical frameworks, ADFs for short [@inproc:adf]. They make use of the so–called acceptance conditions to express arbitrary interactions between the arguments. However, a framework cannot be considered a suitable argumentation tool without properly developed semantics. The semantics of a framework are meant to represent what is considered rational. Given many of the advanced semantics, such as grounded or complete, we can observe that they return same results when faced with simple, tree–like frameworks. The differences between them become more visible when we work with more complicated cases. On various occasions examples were found for which none of the available semantics returned satisfactory answers. This gave rise to new concepts: for example, for handling indirect attacks and defenses we have prudent and careful semantics [@inproc:careful; @inproc:prudent]. For the problem of even and odd attack cycles we can resort to some of the SCC–recursive semantics [@BaroniGG05a], while for treatment of self attackers, sustainable and tolerant semantics were developed [@Bodanza2009]. Introducing a new type of relation, such as support, creates additional problems. The most controversial issue in the bipolar setting concerns the support cycles and is handled differently from formalism to formalism. Among the best known structures are the Bipolar Argumentation Frameworks (BAFs for short) [@incoll:bipolar; @article:newbaf], Argumentation Frameworks with Necessities (AFNs) [@incoll:newafn] and Evidential Argumentation Systems (EASs) [@inproc:eas]. While AFNs and EASs discard support cycles, BAFs do not make such restrictions. In ADFs cycles are permitted unless the intuition of a given semantics is clearly against it, for example in stable and grounded cases. This variety is not an error in any of the structures; it is caused by the fact that, in a setting that allows more types of relations, a standard Dung semantics can be extended in several ways. Moreover, since one can find arguments both for and against any of the cycle treatments, lack of consensus as to what approach is the best should not be surprising. Many properties of the available semantics can be seen as “inside” ones, i.e. “what can I consider rational?”. On the other hand, some can be understood as on the “outside”, e.g. “what can be considered a valid attacker, what should I defend from?”. Various examples of such behavior exist even in the Dung setting. An admissible extension is conflict–free and defends against attacks carried out by any other argument in the framework. We can then add new restrictions by saying that self–attackers are not rational. Consequently, we limit the set of arguments we have to protect our choice from. In a bipolar setting, we can again define admissibility in the basic manner. However, one often demands that the extension is free from support cycles and that we only defend from acyclic arguments, thus again trimming the set of attackers. From this perspective semantics can be seen as a two–person discussion, describing what “I can claim” and “what my opponent can claim”. This is also the point of view that we follow in this paper. Please note that this sort of dialogue perspective can already be found in argumentation [@inproc:dungdialect; @inproc:dialect], although it is used in a slightly different context. Although various extension–based semantics for ADFs have already been proposed in the original paper [@inproc:adf], many of them were defined only for a particular ADF subclass called the bipolar and were not suitable for all types of situations. As a result, only three of them – conflict–free, model and grounded – remain. Moreover, the original formulations did not solve the problem of positive dependency cycles. Unfortunately, neither did the more recent work into labeling–based semantics [@tofix:newadf], even though they solve most of the problems of their predecessors. The aim of this paper is to address the issue of cycles and the lack of properly developed extension–based semantics. We introduce a family of such semantics and specialize them to handle the problem of support cycles, as their treatment seems to be the biggest difference among the available frameworks. Furthermore, a classification of our sub–semantics in the inside–outside fashion that we have described before is introduced. We also recall our previous research on admissibility in [@inproc:adm] and show how it fits into the new system. Our results also include which known properties, such as Fundamental Lemma, carry over from the Dung framework. Finally, we provide an analysis of similarities and differences between the extension and labeling–based semantics in the context of produced extensions. The paper is structured as follows. In Sections \[sec:dungintro\] to \[sec:adf\] we provide a background on argumentation frameworks. Then we introduce the new extension–based semantics and analyze their behavior in Section \[sec:sem\]. We close the paper with a comparison between the new concepts and the existing labeling–based approach. Dung’s Argumentation Frameworks {#sec:dungintro} =============================== Let us recall the abstract argumentation framework by Dung [@article:dung] and its semantics. For more details we refer the reader to [@article:semintro]. A Dung’s abstract argumentation framework (AF for short) is a pair $(A, R)$, where $A$ is a set of arguments and $R \subseteq A \times A$ represents an attack relation. Let $AF = (A, R)$ be a Dung’s framework. We say that an argument $a \in A$ is defended[^2] by a set $E$ in $AF$, if for each $b \in A$ s.t. $(b, a)\in R$, there exists $c \in E$ s.t. $(c, b)\in R$. A set $E \subseteq A$ is: - conflict–free in $AF$ iff for each $a, b \in E,\, (a,b) \notin R$. - admissible iff conflict–free and defends all of its members. - preferred iff it is maximal w.r.t set inclusion admissible. - complete iff it is admissible and all arguments defended by $E$ are in $E$. - stable iff it is conflict–free and for each $a\in A \setminus E$ there exists an argument $b \in E$ s.t. $(b,a) \in R$. The characteristic function $F_{AF} : 2^A \rightarrow 2^A$ is defined as: $F_{AF}(E) = \{a \mid a \text{ is defended by } E \text{ in } AF\}$. The grounded extension is the least fixed point of $F_{AF}$. In the context of this paper, we would also like to recall the notion of range: Let $E^+$ be the set of arguments attacked by $E$ and $E^-$ the set of arguments that attack $E$ . $E^+ \cup E$ is the range of $E$. Please note the concepts $E^+$ and the $E^-$ sets can be used to redefine defense. This idea will be partially used in creating the semantics of ADFs. Moreover, there is also an alternative way of computing the grounded extension: The unique grounded extension of $AF$ is defined as the outcome $E$ of the following “algorithm”. Let us start with $E=\emptyset$: 1. put each argument $a \in A$ which is not attacked in $AF$ into $E$; if no such argument exists, return $E$. 2. remove from $AF$ all (new) arguments in $E$ and all arguments attacked by them (together with all adjacent attacks) and continue with Step 1. \[prop:dung-grd\] What we have described above forms a family of the extension–based semantics. However, there exist also labeling–based ones [@CaminadaG09; @article:semintro]. Instead of computing sets of accepted arguments, they generate labelings, i.e. total functions $Lab: A \rightarrow \{in, out, undec\}$. Although we will not recall them here, we would like to draw the attention to the fact that for every extension we can obtain an appropriate labeling and vice versa. This property is particularly important as it does not fully carry over to the ADF setting. Finally, we would like to recall several important lemmas and theorems from the original paper on AFs [@article:dung]. [Dung’s Fundamental Lemma]{} Let $E$ be an admissible extension, $a$ and $b$ two arguments defended by $E$. Then $E' = E \cup \{a\}$ is admissible and $b$ is defended by $E'$. Every stable extension is a preferred extension, but not vice versa. Every preferred extension is a complete extension, but not vice versa. The grounded extension is the least w.r.t. set inclusion complete extension. The complete extensions form a complete semilattice w.r.t. set inclusion. [^3] Argumentation Frameworks with Support {#sec:bip} ===================================== Currently the most recognized frameworks with support are the Bipolar Argumentation Framework BAF [@article:newbaf], Argumentation Framework with Necessities AFN [@incoll:newafn] and Evidential Argumentation System EAS [@inproc:eas]. We will now briefly recall them in order to further motivate the directions of the semantics we have taken in ADFs. The original bipolar argumentation framework BAF [@incoll:bipolar] studied a relation we will refer to as abstract support: A bipolar argumentation framework is a tuple $(A, R, S)$, where $A$ is a set of arguments, $R \subseteq A \times A$ represents the attack relation and $S \subseteq A \times A$ the support. The biggest difference between this abstract relation and any other interpretation of support is the fact that it did not affect the acceptability of an argument, i.e. even a supported argument could be accepted “alone”. The positive interaction was used to derive additional indirect forms of attack and based on them, stronger versions of conflict–freeness were developed. We say that an argument $a$ support attacks argument $b$, if there exists some argument $c$ s.t. there is a sequence of supports from $a$ to $c$ (i.e. $a S...S c$) and $c R b$. We say that $a$ secondary attacks $b$ if there is some argument $c$ s.t. $c S...S b$ and $a R c$. We say that $B\subseteq A$ is: - +conflict–free iff $\nexists a, b \in B$ s.t. $a$ (directly or indirectly) attacks $b$. - safe iff $\nexists b\in A$ s.t. $b$ is at the same time (directly or indirectly) attacked by $B$ and either there is a sequence of supports from an element of $B$ to $b$, or $b \in B$. - closed under $S$ iff $\forall b\in B, a \in A$, if $bSa$ then $a\in B$. The definition of defense remains the same and any Dung semantics is specialized by choosing an given notion of conflict–freeness or safety. Apart from the stable semantics, no assumptions as to cycles occurring in the support relation are made. The later developed deductive support [@inproc:support] remains in the BAF setting and is also modeled by new indirect attacks [@article:newbaf]. Consequently, acyclicity is not required. The most recent formulation of the framework with necessary support is as follows [@incoll:newafn]: An argumentation framework with necessities is a tuple $(A, R, N)$, where $A$ is the set of arguments, $R \subseteq A \times A$ represents (binary) attacks, and $N \subseteq (2^A \setminus \emptyset) \times A$ is the necessity relation. Given a set $B\subseteq A$ and an argument $a$, $B N a$ should be read as “at least one element of $B$ needs to be present in order to accept $a$”. The AFN semantics are built around the notions of coherence: We say that a set of arguments $B$ is coherent iff every $b\in B$ is powerful, i.e. there exists a sequence $a_0,..,a_n$ of some elements of $B$ s.t $a_n = b$, $\nexists C\subseteq A$ s.t. $CN a_0$, and for $1\leq i \leq n$ it holds that for every set $C\subseteq A$ if $CN a_i$, then $C \cap \{a_0,...,a_{i-1}\} \neq \emptyset$. A coherent set $B$ is strongly coherent iff it is conflict–free. Although it may look a bit complicated at first, the definition of coherence grasps the intuition that we need to provide sufficient acyclic support for the arguments we want to accept. Defense in AFNs is understood as the ability to provide support and to counter the attacks from any coherent set. We say that a set $B\subseteq A$ defends $a$, if $B\cup \{a\}$ is coherent and for every $c\in A$, if $c R a$ then for every coherent set $C\subseteq A$ containing $c$, $B R C$. Using the notion of strong coherence and defense, the AFN semantics are built in a way corresponding to Dung semantics. It is easy to see that, through the notion of coherency, AFNs discard cyclic arguments both on the “inside” and the “outside”. This means we cannot accept them in an extension and they are not considered as valid attackers. The last type of support we will consider here is the the evidential support [@inproc:eas]. It distinguishes between standard and prima facie arguments. The latter are the only ones that are valid without any support. Every other argument that we want to accept needs to be supported by at least one prima facie argument, be it directly or not. An evidential argumentation system (EAS) is a tuple $(A, R, E)$ where $A$ is a set of arguments, $R \subseteq (2^A \setminus \emptyset) \times A$ is the attack relation, and $E \subseteq (2^A \setminus \emptyset) \times A$ is the support relation. We distinguish a special argument $\eta \in A$ s.t. $\nexists (x,y) \in R$ where $\eta \in x$; and $\nexists x$ where $(x, \eta) \in R$ or $(x, \eta) \in E$. $\eta$ represents the prima facie arguments and is referred to as evidence or environment. The idea that the valid arguments (and attackers) need to trace back to it is captured with the notions of e–support and e–supported attack[^4]. \[def:esup\] An argument $a\in A$ has evidential support (e–support) from a set $S\subseteq A$ iff $a=\eta$ or there is a non-empty $S' \subseteq S$ s.t. $S' E a$ and $\forall x \in S'$, $x$ has e–support from $S \setminus \{a\}$. \[def:eatt\] A set $S\subseteq A$ carries out an evidence supported attack (e–supported attack) on $a$ iff $(S',a) \in R$ where $S' \subseteq S$, and for all $s \in S'$, $s$ has e–support from $S$. An e–supported attack by $S$ on $a$ is minimal iff there is no $S' \subset S$ that carries out an e–supported attack on $a$. The EASs semantics are built around the notion of acceptability in a manner similar to those of Dung’s. However, in AFs only the attack relation was considered. In EASs, also sufficient support is required: An argument $a$ is acceptable w.r.t. a set $S\subseteq A$ iff $a$ is e–supported by $S$ and given a minimal e–supported attack by a set $T\subseteq A$ against $a$, it is the case that $S$ carries out an e–supported attack against a member of $T$. The notion of conflict–freeness is easily adapted to take set, not just binary conflict into account. With this and the notion of acceptability, the EASs semantics are built just like AF semantics. From the fact that every valid argument needs to be grounded in the environment it clearly results that EAS semantics are acyclic both on the inside and outside. Abstract Dialectical Frameworks {#sec:adf} =============================== Abstract dialectical frameworks have been defined in [@inproc:adf] and further studied in [@tofix:newadf; @inproc:adm; @report:strass; @strass13instantiating; @strass-wallner14complexity]. The main goal of ADFs is to be able to express arbitrary relations and avoid the need of extending AFs by new relation sets each time they are needed. This is achieved by the means of the acceptance conditions, which define what arguments should be present in order to accept or reject a given argument. An abstract dialectical framework (ADF) as a tuple $(S, L, C)$, where $S$ is a set of abstract arguments (nodes, statements), $L \subseteq S \times S$ is a set of links (edges) and $C = \{C_ s \}_{s\in S}$ is a set of acceptance conditions, one condition per each argument. An acceptance condition is a total function $C_s : 2^{par(s)} \rightarrow \{in, out\}$, where $par(s) = \{ p \in S \mid (p,s) \in L\}$ is the set of parents of an argument $s$. One can also represent the acceptance conditions by propositional formulas [@thesis:stefan] rather than functions. By this we mean that given an argument $s \in S$, $C_s = \varphi_s$, where $\varphi_s$ is a propositional formula over arguments $S$. As we will be making use of both extension and labeling–based semantics, we need to provide necessary information on interpretations first (more details can be found in [@tofix:newadf; @inproc:adm]). Please note that the links in ADFs only represent connections between arguments, while the burden of deciding the nature of these connections falls to the acceptance conditions. Moreover, parents of an argument can be easily extracted from the conditions. Thus, we will use of shortened notation $D = (S,C)$ through the rest of this paper. Interpretations and decisiveness {#sec:premdec} -------------------------------- A two (or three–valued) interpretation is simply a mapping that assigns the truth values $\{{\mathbf{t}}, {\mathbf{f}}\}$ (respectively $\{{\mathbf{t}}, {\mathbf{f}}, {\mathbf{u}}\}$) to arguments. We will be making use both of partial (i.e. defined only for a subset of $S$) and the full ones. In the three–valued setting we will adopt the precision (information) ordering of the values: ${\mathbf{u}}\leq_i {\mathbf{t}}$ and ${\mathbf{u}}\leq_i {\mathbf{f}}$ The pair $(\{\textbf{t}, \textbf{f}, \textbf{u}\}, \leq_i)$ forms a complete meet–semilattice with the meet operation $\sqcap$ assigning values in the following way: $\textbf{t} \, \sqcap \, \textbf{t} = \textbf{t}$, $\textbf{f} \, \sqcap \, \textbf{f} = \textbf{f}$ and $\textbf{u}$ in all other cases. It can naturally be extended to interpretations: given two interpretations $v$ and $v'$ on $S$, we say that $v'$ contains more information, denoted $v \leq_i v'$, iff $\forall_{s\in S}\, v(s) \leq_i v'(s)$. Similar follows for the meet operation. In case $v$ is three and $v'$ two–valued, we say that $v'$ extends $v$. This means that elements mapped originally to ${\mathbf{u}}$ are now assigned either ${\mathbf{t}}$ or ${\mathbf{f}}$. The set of all two–valued interpretations extending $v$ is denoted $\lbrack v \rbrack_2$. Let $v=\{a:{\mathbf{t}}, b:{\mathbf{t}}, c:{\mathbf{f}}, d:{\mathbf{u}})$ be a three–valued interpretation. We have two extending interpretations, $v'=\{a:{\mathbf{t}}, b:{\mathbf{t}}, c:{\mathbf{f}}, d:{\mathbf{t}})$ and $v''=\{a:{\mathbf{t}}, b:{\mathbf{t}}, c:{\mathbf{f}}, d:{\mathbf{f}})$. Clearly, it holds that $v \leq_i v'$ and $v\leq_i v''$. However, $v'$ and $v''$ are incomparable w.r.t. $\leq_i$. Let now $w=\{a:{\mathbf{f}}, b:{\mathbf{f}}, c:{\mathbf{f}}, d:{\mathbf{t}})$ be another three–valued interpretation. $v \sqcap w$ gives us a new interpretation $w' = \{a:{\mathbf{u}}, b:{\mathbf{u}}, c:{\mathbf{f}}, d:{\mathbf{u}})$: as the assignments of $a,b$ and $d$ differ between $v$ and $w$, the resulting value is ${\mathbf{u}}$. On the other hand, $c$ is in both cases ${\mathbf{f}}$ and thus retains its value. We will use $v^x$ to denote a set of arguments mapped to $x$ by $v$, where $x$ is some truth–value. Given an acceptance condition $C_s$ for some argument $s \in S$ and an interpretation $v$, we define a shorthand $v(C_s)$ as $C_s(v^{\mathbf{t}}\cap par(s))$. For a given propositional formula $\varphi$ and an interpretation $v$ defined over all of the atoms of the formula, $v(\varphi)$ will just stand for the value of the formula under $v$. However, apart from knowing the “current” value of a given acceptance condition for some interpretation, we would also like to know if this interpretation is “final”. By this we understand that no new information will cause the value to change. This is expressed by the notion of decisive interpretations, which are at the core of the extension–based ADF semantics. Given an interpretation $v$ defined over a set $A$, completion of $v$ to a set $Z$ where $A\subseteq Z$ is an interpretation $v'$ defined on $Z$ in a way that $\forall a \in A \; v(a) = v'(a)$. By a ${\mathbf{t}}/{\mathbf{f}}$ completion we will understand $v'$ that maps all arguments in $Z\setminus A$ respectively to ${\mathbf{t}}/{\mathbf{f}}$. The similarity between the concepts of completion and extending interpretation should not be overlooked. Basically, given a three–valued interpretation $v$ defined over $S$, the set $\lbrack v \rbrack_2$ precisely corresponds to the set of completions to $S$ of the two–valued part of $v$. However, the extension notion from the three–valued setting can be very misleading when used in the extension–based semantics. Therefore, we would like to keep the notion of completion. We say that a two–valued interpretation $v$ is decisive for an argument $s \in S$ iff for any two completions $v_{par(s)}$ and $v'_{par(s)}$ of $v$ to $A \cup par(s)$, it holds that $v_{par(s)}(C_s) =v'_{par(s)}(C_s)$. We say that $s$ is decisively out/in wrt $v$ if $v$ is decisive and all of its completions evaluate $C_s$ to respectively $out,in$. \[->,>=stealth,shorten >=1pt,auto,node distance=1.5cm, thick,main node/.style=[circle,fill=none,draw,minimum size = 0.7cm,font=**]{}, condition/.style=[circle,fill=none,draw=none,minimum size = 0.3cm,font=]{}\] \(a) [$a$]{}; (b) \[right of=a\] [$b$]{}; (c) \[right of=b\] [$c$]{}; (d) \[left of =a\] [$d$]{}; (ca) \[above of= a, yshift=-0.8cm\] [$b \rightarrow d$]{}; (cb) \[above of= b, yshift=-0.8cm\] [$a \land c$]{}; (cc) \[above of= c, yshift=-0.8cm\] [$\bot$]{}; (cd) \[above of= d, yshift=-0.8cm\] [$d$]{}; (a.east)–(b.west); (c.west)–(b.east); (d.east)–(a.west); (d) edge \[loop left\] node (d); Let $(\{a,b,c,d\}, \{ \varphi_a: b\rightarrow d, \varphi_b: a\land c, \varphi_c: \bot, \varphi_d:d\})$ be an ADF depicted in Figure \[fig:dec\]. Example of a decisively in interpretation for $a$ is $v=\{b: {\mathbf{f}}\}$. It simply means that knowing that $b$ is false, not matter the value of $d$, the implication is always true and thus the acceptance condition satisfied. From the more technical side, it is the same as checking that both completions to $\{b,d\}$, namely $\{b:{\mathbf{f}}, d:{\mathbf{t}}\}$ and $\{b:{\mathbf{f}}, d:{\mathbf{f}}\}$ satisfy the condition. Example of a decisively out interpretation for $b$ is $v'=\{c: {\mathbf{f}}\}$. Again, it suffices to falsify one element of a conjunction to know that the whole formula will evaluate to false. Acyclicity ---------- Let us now focus on the issue of positive dependency cycles. Please note we refrain from calling them support cycles in the ADF setting in order not to confuse them with specific definitions of support available in the literature [@article:newbaf]. Informally speaking, an argument takes part in a cycle if its acceptance depends on itself. An intuitive way of verifying the acyclicity of an argument would be to “track” its evaluation, e.g. in order to accept $a$ we need to accept $b$, to accept $b$ we need to accept $c$ and so on. This basic case becomes more complicated when disjunction is introduced. We then receive a number of such “paths”, with only some of them proving to be acyclic. Moreover, they might be conflicting one with each other, and we can have a situation in which all acyclic evaluations are blocked and a cycle is forced. Our approach to acyclicity is based on the idea of such “paths” that are accompanied by sets of arguments used to detect possible conflicts. Let us now introduce the formal definitions. Given an argument $s\in S$ and $x\in\{in, out\}$, by $min\_dec(x,s)$ we will denote the set of minimal two–valued interpretations that are decisively $x$ for $s$. By minimal we understand that both $v^{{\mathbf{t}}}$ and $v^{{\mathbf{f}}}$ are minimal w.r.t. set inclusion. Let $A \subseteq S$ be a nonempty set of arguments. A positive dependency function on $A$ is a function $pd$ assigning every argument $a \in A$ an interpretation $v \in min\_dec(in, a)$ s.t. $v^t \subseteq A$ or $\mathcal{N}$ (null) iff no such interpretation can be found. An acyclic positive dependency evaluation $ace^a$ for $a \in A$ based on a given pd–function $pd$ is a pair $((a_0,...,a_n), B)$, [^5] where $B = \bigcup_{i=0}^n \, pd(a_i)^{{\mathbf{f}}}$ and $(a_0,...,a_n)$ is a sequence of distinct elements of $A$ s.t.: $\forall_{i=0}^n \, pd(a_i) \neq \mathcal{N}$, $a_n = a$, $pd(a_0)^{\mathbf{t}}= \emptyset$, and $\forall_{i=1}^n, \, pd(a_i)^{{\mathbf{t}}} \subseteq \{a_0,...,a_{i-1}\}$. We will refer to the sequence part of the evaluation as pd–sequence and to the $B$ as the blocking set. We will say that an argument $a$ is pd–acyclic in $A$ iff there exist a pd–function on $A$ and a corresponding acyclic pd–evaluation for $a$. We will write that an argument has an acyclic pd–evaluation on $A$ if there is some pd–function on $A$ from which we can produce the evaluation. There are two ways we can “attack” an acyclic evaluation. We can either discard an argument required by the evaluation or accept one that is capable of preventing it. This corresponds to rejecting a member of a pd–sequence or accepting an argument from the blocking set. We can now formulate this “conflict” by the means of an interpretation: Let $A\subseteq S$ be a set of arguments and $a \in A$ s.t. $a$ has an acyclic pd–evaluation $ace^a = ((a_0,...,a_n), B) $ in $A$. We say that a two–valued interpretation $v$ blocks $ace^a$ iff $\exists b\in B$ s.t. $v(b) = {\mathbf{t}}$ or $\exists a_i \in \{a_0,...,a_n\}$ s.t. $v(a_i) = {\mathbf{f}}$. Let us now show on an example why we require minimality on the chosen interpretations and why do we store the blocking set: Let us assume an ADF $(\{a,b,c\}, \{C_a:\neg c \lor b, C_b:a, C_c:c\})$ depicted in Figure \[fig:cf\]. For argument $a$ there exist the following decisively in interpretations: $v_1 = \{c:{\mathbf{f}}\},v_2 = \{b:{\mathbf{t}}\},v_3 = \{b:{\mathbf{t}}, c:{\mathbf{f}}\},v_4= \{b:{\mathbf{t}}, c:{\mathbf{t}}\}, v_5 = \{b:{\mathbf{f}}, c:{\mathbf{f}}\}$. Only the first two are minimal. Considering $v_4$ would give us a wrong view that $a$ requires $c$ for acceptance, which is not a desirable reading. The interpretations for $b$ and $c$ are respectively $w_1 = \{a:{\mathbf{t}}\}$ and $z_1 = \{c:{\mathbf{t}}\}$. Consequently, we have two pd–functions on $\{a,b,c\}$, namely $pd_1 = \{a:v_1, b:w_1, c:z_1\}$ and $pd_2 = \{a:v_2, b:w_1, c:z_1\}$. From them we obtain one acyclic pd–evaluation for $a$: $((a),\{c\})$, one for $b$: $((a,b),\{c\})$ and none for $c$. Let us look closer at a set $E=\{a,b,c\}$. We can see that $c$ is not pd–acyclic in $E$. However, the presence of $c$ also “forces” a cycle between $a$ and $b$. The acceptance conditions of all arguments are satisfied, thus this simple check is not good enough to verify if a cycle occurs. Only looking at the whole evaluations shows us that $a$ and $b$ are both blocked by $c$. Although $a$ and $b$ are pd–acyclic in $E$, we see that their evaluations are in fact blocked and this second level of conflict needs to be taken into account by the semantics. \[->,>=stealth,shorten >=1pt,auto,node distance=1.5cm, thick,main node/.style=[circle,fill=none,draw,minimum size = 0.7cm,font=]{}, condition/.style=[circle,fill=none,draw=none,minimum size = 0.3cm,font=*]{}\] \(a) [$a$]{}; (b) \[left of=a\] [$b$]{}; (c) \[right of=a\] [$c$]{}; (ca) \[above of= a, yshift=-0.8cm\] [$\neg c \lor b$]{}; (cb) \[above of= b, yshift=-0.8cm\] [$a$]{}; (cc) \[above of= c, yshift=-0.8cm\] [$c$]{}; \(a) edge \[bend left\] node (b) (b) edge \[bend left\] node (a) (c) edge node (a) (c) edge \[loop right\] node (c); \[ex:cf\] As a final remark, please note that it can be the case that an evaluation is self–blocking. We can now proceed to recall existing and introduce new semantics of the abstract dialectical frameworks. Extension–Based Semantics of ADFs {#sec:sem} ================================= Although various semantics for ADFs have already been defined in the original paper [@inproc:adf], only three of them – conflict–free, model and grounded (initially referred to as well–founded) – are still used (issues with the other formulations can be found in [@tofix:newadf; @inproc:adm; @report:strass]). Moreover, the treatment of cycles and their handling by the semantics was not sufficiently developed. In this section we will address all of those issues. Before we continue, let us first motivate our choice on how to treat cycles. The opinions on support cycles differ between the available frameworks, as we have shown in Section \[sec:bip\]. Therefore, we would like to explore the possible approaches in the context of ADFs by developing appropriate semantics. The classification of the sub–semantics that we will adopt in this paper is based on the inside–outside intuition we presented in the introduction. Appropriate semantics will receive a two–element prefix $xy-$, where $x$ will denote whether cycles are permitted or not on the “inside” and $y$ on the “outside”. We will use $x, y \in \{a,c\}$, where $a$ will stand for acyclic* and $c$ for cyclic constraints. As the conflict–free (and naive) semantics focus only on what we can accept, we will drop the prefixing in this case. Although the model, stable and grounded fit into our classification (more details can be found in this section and in [@report:semantics]), they have a sufficiently unique naming and further annotations are not necessary. We are thus left with admissible, preferred and complete. The BAF approach follows the idea that we can accept arguments that are not acyclic in our opinion and we allow our opponent to do the same. The ADF semantics we have developed in [@inproc:adm] also shares this view. Therefore, they will receive the $cc-$ prefix. On the other hand, AFN and EAS semantics do not permit cycles both in extensions and as attackers. Consequently, the semantics following this line of reasoning will be prefixed with $aa-$. Please note we believe that a non–uniform approach can also be suitable in certain situations. By non–uniform we mean not accepting cyclic arguments, but still treating them as valid attackers and so on (i.e. $ca-$ and $ac-$). However, in this paper we would like to focus only on the two perspectives mentioned before. Conflict–free and naive semantics --------------------------------- In the Dung setting, conflict–freeness meant that the elements of an extension could not attack one another. Providing an argument with the required support is then a separate condition in frameworks such as AFNs and EASs. In ADFs, where we lose the set representation of relations in favor of abstraction, not including “attackers” and accepting “supporters” is combined into one notion. This represents the intuition of arguments that can stand together presented in [@article:semintro]. Let us now assume an ADF $D= (S,C)$. A set of arguments $E \subseteq S$ is conflict–free in $D$ iff for all $s \in E$ we have $C_s (E \cap par(s )) = in$. In the acyclic version of conflict–freeness we also need to deal with the conflicts arising on the level of evaluations. To meet the formal requirements, we first have to show how the notions of range and the $E^+$ set are moved to ADFs. Let $E\subseteq S$ a conflict–free extension of $D$ and $v_E$ a partial two–valued interpretation built as follows: 1. Let $M=E$ and for every $a\in M$ set $v_E(a)={\mathbf{t}}$; 2. For every argument $b \in S\setminus M$ that is decisively out in $v_E$, set $v_E(b)={\mathbf{f}}$ and add $b$ to $M$; 3. Repeat the previous step until there are no new elements added to $M$. By $E^+$ we understand the set of arguments $v_E^{\mathbf{f}}$ and we will refer to it as the discarded set. $v_E$ now forms the range interpretation of $E$. However, the notions of the discarded set and the range are quite strict in the sense that they require an explicit “attack” on arguments that take part in dependency cycles. This is not always a desirable property. Depending on the approach we might not treat cyclic arguments as valid and hence want them “out of the way”. Let $E\subseteq S$ a conflict–free extension of $D$ and $v_E^a$ a partial two–valued interpretation built as follows: 1. Let $M=E$. For every $a\in M$ set $v_E^a(a)={\mathbf{t}}$. 2. For every argument $b \in S\setminus M$ s.t. every acyclic pd–evaluation of $b$ in $S$ is blocked by $v_E^a$, set $v_E^a(b)={\mathbf{f}}$ and add $b$ to $M$. 3. Repeat the previous step until there are no new elements added to $M$. By $E^{a+}$ we understand the set of arguments mapped to ${\mathbf{f}}$ by $v_E^a$ and refer to it as acyclic discarded set. We refer to $v_E^a$ as acyclic range interpretation of $E$. We can now define an acyclic version of conflict–freeness: A conflict–free extension $E$ is a pd–acyclic conflict–free extension of $D$ iff every argument $a \in E$ has an unblocked acyclic pd–evaluation on $E$ w.r.t. $v^E$. As we are dealing with a conflict– free extension, all the arguments of a given pd–sequence are naturally ${\mathbf{t}}$ both in $v_E$ and $v_E^a$. Therefore, in order to ensure that an evaluation $((a_0,...,a_n), B)$ is unblocked it suffices to check whether $E \cap B = \emptyset$. Consequently, in this case it does not matter w.r.t. to which version of range we are verifying the evaluations. The naive and pd–acyclic naive extensions are respectively maximal w.r.t. set inclusion conflict–free and pd–acyclic conflict–free extensions. Recall the ADF $(\{a,b,c\}, \{C_a:\neg c \lor b, C_b:a, C_c:c\})$. The conflict–free extensions are $\emptyset, \{a\}, \{c\}, \{a,b\}$ and $\{a,b,c\}$. Their standard discarded set in all cases is just $\emptyset$ – none of the sets has the power to decisively out the non–members. The acyclic discarded set of $\emptyset$, $\{a\}$ and $\{a,b\}$ is now $\{c\}$, since it has no acyclic evaluation to start with. In the case of $\{c\}$, it is $\{a,b\}$, which is to be expected since $c$ had the power to block their evaluations. Finally, $\{a,b,c\}^{a+}$ is $\emptyset$. In the end, only $\emptyset, \{a\}$ and $\{a,b\}$ qualify for acyclic type. The naive and pd–acyclic naive extensions are respectively $\{a,b,c\}$ and $\{a,b\}$. Model and stable semantics -------------------------- The concept of a model basically follows the intuition that if something can be accepted, it should be accepted: A conflict–free extension $E$ is a model of $D$ if $\forall \;s \in S, \; \;C_s (E \cap par (s )) = in$ implies $s \in E$. Although the semantics is simple, several of its properties should be explained. First of all, given a model candidate $E$, checking whether a condition of some argument $s$ is satisfied does not verify if an argument depends on itself or if it “outs” a previously included member of $E$. This means that an argument we should include may break conflict–freeness of the set. On the other hand, an argument can be $out$ due to positive dependency cycles, i.e. its supporter is not present. And since model makes no acyclicity assumptions on the inside, arguments outed this way can later appear in a model $E\subset E'$. Consequently, it is clear to see that model semantics is not universally defined and the produced extensions might not be maximal w.r.t. subset inclusion. The model semantics was used as a mean to obtain the stable models. The main idea was to make sure that the model is acyclic. Unfortunately, the used reduction method was not adequate, as shown in [@tofix:newadf]. However, the initial idea still holds and we use it to define stability. Although the produced extensions are now incomparable w.r.t. set inclusion, the semantics is still not universally defined. A model $E$ is a stable extension iff it is pd–acyclic conflict–free. Let us again come back to the ADF $(\{a,b,c\}, \{C_a:\neg c \lor b, C_b:a, C_c:c\})$. The conflict–free extensions were $\emptyset, \{a\}, \{c\}, \{a,b\}$ and $\{a,b,c\}$. The first two are not models, as in the first case $a$ and in the latter $b$ can be accepted. Recall that $\emptyset, \{a\}$ and $\{a,b\}$ were the pd–acyclic conflict–free extensions. The only one that is also a model is $\{a,b\}$ and thus we obtain our single stable extension. \[ex:mod\] Grounded semantics ------------------ Next comes the grounded semantics [@inproc:adf]. Just like in the Dung setting, it preserves the unique–status property, i.e. produces only a single extension. Moreover, it is defined in the terms of a special operator: Let $\Gamma'_D(A,R) = (acc(A, R ),reb (A, R ))$, where $acc(A, R)=\{ r\in S \mid A \subseteq S' \subseteq (S\backslash R) \Rightarrow C_r(S' \cap par(r))=in\}$ and $reb(A, R )=\{ r\in S \mid A \subseteq S' \subseteq (S\backslash R) \Rightarrow C_r(S' \cap par(r))=out\}$. Then $E$ is the grounded model of $D$ iff for some $E' \subseteq S, (E,E')$ is the least fix–point of $\Gamma'_D$. Although it might look complicated at first, this is nothing more than analyzing decisiveness using a set, not interpretation form (please see [@report:semantics] for more details). Thus, one can also obtain the grounded extension by an ADF version of Proposition \[prop:dung-grd\]: Let $v$ be an empty interpretation. For every argument $a\in S$ that is decisively in w.r.t. $v$, set $v(a) = {\mathbf{t}}$ and for every argument $b\in S$ that is decisively w.r.t. $v$, set $v(b) = {\mathbf{f}}$. Repeat the procedure until no further assignments can be done. The grounded extension of $D$ is then $v^{\mathbf{t}}$. \[prop:adf-grd\] Recall our ADF $(\{a,b,c\}, \{C_a:\neg c \lor b, C_b:a, C_c:c\})$. Let $v$ be an empty interpretation. It is easy to see that no argument is decisively in/out w.r.t. $v$. If we analyze $a$, it is easy to see that if we accept $c$, the condition is out, but if we accept both $b$ and $c$ it is in again. Although both $b$ and $c$ are out in $v$, the condition of $b$ can be met if we accept $a$, and condition of $c$ if we accept $c$. Hence, we obtain no decisiveness again. Thus, $\emptyset$ is the grounded extension. Admissible and preferred semantics ---------------------------------- In [@inproc:adm] we have presented our first definition of admissibility, before the sub–semantics classification was developed. The new, simplified version of our previous formulation, is now as follows: A conflict–extension $E \subseteq S$ is cc–admissible in $D$ iff every element of $E$ is decisively in w.r.t to its range interpretation $v_E$. It is important to understand how decisiveness encapsulates the defense known from the Dung setting. If an argument is decisively in, then any set of arguments that would have the power to out the acceptance condition is “prevented” by the interpretation. Hence, the statements required for the acceptance of $a$ are mapped to ${\mathbf{t}}$ and those that would make us reject $a$ are mapped to ${\mathbf{f}}$. The former encapsulates the required support, while the latter contains the “attackers” known from the Dung setting. When working with the semantics that have to be acyclic on the “inside”, we not only have to defend the members, but also their acyclic evaluations: A pd–acyclic conflict–free extension $E$ is aa–admissible iff every argument in $E$ is decisively in w.r.t. acyclic range interpretation $v_E^a$, and has an unblocked acyclic pd–evaluation on $E$ s.t. all members of its blocking set $B$ are mapped to ${\mathbf{f}}$ by $v_E^a$. \[def:adm1\] A set of arguments is xy–preferred iff it is maximal w.r.t. set inclusion xy–admissible. The following example shows that decisiveness encapsulates defense of an argument, but not necessarily of its evaluation: Let us modify the ADF depicted in Figure \[fig:cf\] by changing the condition of $c$: $(\{a,b,c\}, \{C_a:\neg c \lor b, C_b:a, C_c:\top\})$. The new pd–evaluations are $((a),\{c\})$ for $a$, $((a,b),\{c\})$ for $b$ and $((c), \emptyset)$ for $c$. The conflict–free extensions are now $\emptyset, \{a\}, \{c\}, \{a, b\}$ and $\{a, b, c\}$. Apart from the last, all are pd–acyclic conflict–free. $\emptyset$ and $\{c\}$ are trivially both aa and cc–admissible and $\{a,b,c\}$ cc–admissible. The standard and acyclic discarded sets of $\{a\}$ are both empty, thus $a$ is not decisively in (we can always utter $c$) and the set is neither aa nor cc–admissible. The discarded sets of $\{a,b\}$ are also empty; however, it is easy to see that both $a$ and $b$ are decisively in. Although uttering $c$ would not change the values of acceptance conditions, it blocks the pd–evaluations of $a$ and $b$. Thus, $\{a,b\}$ is cc, but not aa–admissible. The cc and aa–preferred extensions are respectively $\{a,b,c\}$ and $\{c\}$. Let us come back to the original ADF $(\{a,b,c\}, \{C_a:\neg c \lor b, C_b:a, C_c:c\})$. $\emptyset, \{a\}, \{c\}, \{a,b\}$ and $\{a,b,c\}$ were the standard and $\emptyset, \{a\}, \{a,b\}$ pd–acyclic conflict–free extensions. $\emptyset$ is trivially both aa and cc, while $\{c\}$ and $\{a,b,c\}$ cc–admissible. The standard discarded sets of $\{a\}$ and $\{a,b\}$ are both empty, while the acyclic ones are $\{c\}$. Consequently, $\{a\}$ is aa, but not cc–admissible. $\{a,b\}$ is both, but for different reasons; in the cc–case, all arguments are decisively in (due to cyclic defense). In aa–approach, they are again decisively in, but the evaluations are “safe” only because $c$ is not considered a valid attacker. \[ex:adm\] Complete semantics ------------------ Completeness represents an approach in which we have to accept everything we can safely conclude from our opinions. In the Dung setting, “safely” means defense, while in the bipolar setting it is strengthened by providing sufficient support. In a sense, it follows the model intuition that what we can accept, we should accept. However, now we not only use an admissible base in place of a conflict–free one, but also defend the arguments in question. Therefore, instead of checking if an argument is in, we want it to be decisively in. A cc–admissible extension $E$ is cc–complete in $D$ iff every argument in $S$ that is decisively in w.r.t. to range interpretation $v_E$ is in $E$. An aa–admissible extension $E$ is aa–complete in $D$ iff every argument in $S$ that is decisively in w.r.t. to acyclic range interpretation $v_E^a$ is in $E$. Please note that in the case of aa–complete semantics, no further “defense” of the evaluation is needed, as visible in AA Fundamental Lemma (i.e. Lemma \[fund4\]). This comes from the fact that if we already have a properly “protected” evaluation, then appending a decisively in argument to it is sufficient for creating an evaluation for this argument. Let us now finish with the ADF $(\{a,b,c\}, \{C_a:\neg c \lor b, C_b:a, C_c:c\})$. It is easy to see that all cc–admissible extensions are also cc–complete. However, only $\{a,b\}$ is aa–complete. Due to the fact that $c$ is trivially included in any discarded set, $a$ can always be accepted (thus, $\emptyset$ is disqualified). Then, from acceptance of $a$, acceptance of $b$ follows easily and $\{a\}$ is disqualified. Properties and examples ----------------------- Although the study provided here will by not be exhaustive, we would like to show how the lemmas and theorems from the original paper on AFs [@article:dung] are shifted into this new setting. The proofs can be found in [@report:semantics]. Even though every pd–acyclic conflict–free extension is also conflict–free, it does not mean that every aa–admissible is cc–admissible. These approaches differ significantly. The first one makes additional restrictions on the “inside”, but due to acyclicity requirements on the “outside” there are less arguments a given extension has to defend from. The latter allows more freedom as to what we can accept, but also gives this freedom to the opponent, thus there are more possible attackers. Moreover, it should not come as a surprise that these differences pass over to the preferred and complete semantics, as visible in Example \[ex1\]. Our results show that admissible sub–semantics satisfy the Fundamental Lemma. [CC Fundamental Lemma:]{} Let $E$ be a cc–admissible extension, $v_E$ its range interpretation and $a, b\in S$ two arguments decisively in w.r.t. $v_E$. Then $E' = E \cup \{a\}$ is cc–admissible and $b$ is decisively in w.r.t. $v_E'$. \[fund1\] [AA Fundamental Lemma:]{} Let $E$ be an aa-admissible extension, $v_E^a$ its acyclic range interpretation and $a, b\in S$ two arguments decisively in w.r.t. $v_E^a$. Then $E' = E \cup \{a\}$ is aa–admissible and $b$ is decisively in w.r.t. $v_E'$. \[fund4\] The relations between the semantics presented in [@article:dung] are preserved by some of the specializations: Every stable extension is an aa–preferred extension, but not vice versa. Every xy–preferred extension is an xy–complete extension for $x,y \in \{a,c\}$, but not vice versa. The grounded extension might not be an aa–complete extension. The grounded extension is the least w.r.t. set inclusion cc–complete extension. Let $(\{a,b,c,d\},\{C_a: \neg b, C_b: \neg a, C_c: b\land \neg d, C_d:d \})$ be the ADF depicted in Figure \[fig:adf1\]. The obtained extensions are visible in Table \[tab:ext\]. The conflict–free, model, stable, grounded, admissible, complete and preferred semantics will be abbreviated to CF, MOD, STB, GRD, ADM, COMP and PREF. The prefixing is visible in second column. In case of conflict–freeness, $C$ will denote the standard, and $A$ the pd–acyclic one. \[->,>=stealth,shorten >=1pt,auto,node distance=1.5cm, thick,main node/.style=[circle,fill=none,draw,minimum size = 0.7cm,font=**]{}, condition/.style=[circle,fill=none,draw=none,minimum size = 0.3cm,font=]{}\] \(a) [$a$]{}; (b) \[right of=a\] [$b$]{}; (c) \[right of=b\] [$c$]{}; (d) \[right of=c\] [$d$]{}; (ca) \[above of= a, yshift=-0.8cm\] [$\neg b$]{}; (cb) \[above of= b, yshift=-0.8cm\] [$\neg a$]{}; (cc) \[above of= c, yshift=-0.8cm\] [$b \land \neg d$]{}; (cd) \[above of= d, yshift=-0.8cm\] [$d$]{}; \(b) edge node (c) (b) edge \[bend left\] node (a) (a) edge \[bend left\] node (b) (d) edge node (c) (d) edge \[loop right\] node (d); ----- ------ ------------------------------------------------------------------------- $C$ $\emptyset$, $\{a\}$, $\{b\}$, $\{d\}$, $\{b,c\}$, $\{a,d\}$, $\{b,d\}$ $A$ $\emptyset$, $\{a\}$, $\{b\}$, , $\{b,c\}$, MOD , $\{a\}$, , $\{b,c\}$, $\{a,d\}$, $\{b,d\}$ STB , $\{a\}$, , $\{b,c\}$, GRD $\emptyset$, $CC$ $\emptyset$, $\{a\}$, $\{b\}$, $\{d\}$, , $\{a,d\}$, $\{b,d\}$ $AA$ $\emptyset$, $\{a\}$, $\{b\}$, , $\{b,c\}$, $CC$ $\emptyset$, $\{a\}$, $\{b\}$, $\{d\}$, , $\{a,d\}$, $\{b,d\}$ $AA$ $\emptyset$, $\{a\}$, , $\{b,c\}$, $CC$ , $\{a,d\}$, $\{b,d\}$ $AA$ , $\{a\}$, , $\{b,c\}$, ----- ------ ------------------------------------------------------------------------- : Extensions of the ADF from Figure \[fig:adf1\]. \[tab:ext\] Labeling–Based Semantics of ADFs {#sec:comparison} ================================ The two approaches towards labeling–based semantics of ADFs were developed in [@report:strass; @tofix:newadf]. We will focus on the latter one, based on the notion of a three–valued characteristic operator: Let $V_S$ be the set of all three–valued interpretations defined on $S$, $s$ and argument in $S$ and $v$ an interpretation in $V_S$. The three–valued characteristic operator of $D$ is a function $\Gamma_D: V_S \rightarrow V_S$ s.t. $\Gamma_D(v) = v'$ with $v'(s) = \bigsqcap_{w \in \lbrack v \rbrack_2} C_s(par(s) \cap w^{\mathbf{t}})$. Verifying the value of an acceptance condition under a set of extensions$\lbrack v \rbrack_2$ of a three–valued interpretation $v$ is exactly checking its value in the completions of the two–valued part of $v$. Thus, an argument that is ${\mathbf{t}}$/${\mathbf{f}}$ in $\Gamma_D(v)$ is decisively in/out w.r.t. to the two–valued part of $v$. It is easy to see that in a certain sense this operator allows self–justification and self–falsification, i.e. that status of an argument depends on itself. Take, for example, a self–supporter; if we generate an interpretation in which it is false then, obviously, it will remain false. Same follows if we assume it to be true. This results from the fact that the operator functions on interpretations defined on all arguments, thus allowing a self–dependent argument to affect its status. The labeling–based semantics are now as follows: Let $v$ be a three–valued interpretation for $D$ and $\Gamma_D$ its characteristic operator. We say that $v$ is: - three–valued model iff for all $s \in S$ we have that $v(s) \neq {\mathbf{u}}$ implies that $v(s) = v(\varphi_s)$; - admissible iff $v \leq_i \Gamma_D(v)$; - complete iff $v = \Gamma_D(v)$; - preferred iff it is $\leq_i$–maximal admissible; - grounded iff it is the least fixpoint of $\Gamma_D$. Although in the case of stable semantics we formally receive a set, not an interpretation, this difference is not significant. As nothing is left undecided, we can safely map all remaining arguments to ${\mathbf{f}}$. The current state of the art definition [@report:strass; @tofix:newadf] is as follows: Let $M$ be a model of $D$. A reduct of $D$ w.r.t. $M$ is $D^M = (M, L^M, C^M)$, where $L^M = L \cap (M \times M)$ and for $m \in M$ we set $C_m^M = \varphi_m\lbrack b/{\mathbf{f}}: b \notin M\rbrack$. Let $gv$ be the grounded model of $D^M$. Model $M$ is stable iff $M = gv^{{\mathbf{t}}}$. Let us now compute the possible labelings of our ADF. As there are over twenty possible three–valued models, we will not list them. We have in total 15 admissible interpretations: $v_1 = \{a: {\mathbf{f}},b: {\mathbf{t}},c: {\mathbf{u}},d: {\mathbf{t}}\}, v_2 = \{a: {\mathbf{t}},b: {\mathbf{f}},c: {\mathbf{u}},d: {\mathbf{u}}\}, v_3 = \{a: {\mathbf{u}},b: {\mathbf{u}},c: {\mathbf{u}},d: {\mathbf{t}}\}, v_4 = \{a: {\mathbf{t}},b: {\mathbf{f}},c: {\mathbf{u}},d: {\mathbf{t}}\}, v_5 = \{a: {\mathbf{f}},b: {\mathbf{t}},c: {\mathbf{u}},d: {\mathbf{f}}\}, v_6 = \{a: {\mathbf{t}},b: {\mathbf{f}},c: {\mathbf{u}},d: {\mathbf{f}}\}, v_7 = \{a: {\mathbf{u}},b: {\mathbf{u}},c: {\mathbf{u}},d: {\mathbf{u}}\}, v_8 = \{a: {\mathbf{u}},b: {\mathbf{u}},c: {\mathbf{f}},d: {\mathbf{t}}\}, v_9 = \{a: {\mathbf{t}},b: {\mathbf{f}},c: {\mathbf{f}},d: {\mathbf{t}}\}, v_{10} = \{a: {\mathbf{f}},b: {\mathbf{t}},c: {\mathbf{t}},d: {\mathbf{f}}\}, v_{11}= \{a: {\mathbf{u}},b: {\mathbf{u}},c: {\mathbf{u}},d: {\mathbf{f}}\}, v_{12} = \{a: {\mathbf{t}},b: {\mathbf{f}},c: {\mathbf{f}},d: {\mathbf{u}}\}, v_{13} = \{a: {\mathbf{f}},b: {\mathbf{t}},c: {\mathbf{u}},d: {\mathbf{u}}\}, v_{14} = \{a: {\mathbf{f}},b: {\mathbf{t}},c: {\mathbf{f}},d: {\mathbf{t}}\}$ and $v_{15} = \{a: {\mathbf{t}},b: {\mathbf{f}},c: {\mathbf{f}},d: {\mathbf{f}}\}$. Out of them $v_7$ to $v_{15}$ are complete. The ones that maximize the information content in this case are the ones without any ${\mathbf{u}}$ mappings: $v_9$, $v_{10}$, $v_{14}$ and $v_9$. $v_{10}$ and $v_{15}$ are stable and finally, $v_7$ is grounded. \[ex1lab\] Comparison with the extension–based approach -------------------------------------------- We will start the comparison of extensions and labelings by relating conflict–freeness and three–valued models. Please note that the intuitions of two–valued and three–valued models are completely different and should not be confused. We will say that an extension $E$ and a labeling $v$ correspond iff $v^{\mathbf{t}}= E$. Let $E$ be a conflict–free and $A$ a pd–acyclic conflict–free extension. The ${\mathbf{u}}$–completions of $v_E$, $v_A$ and $v_A^a$ are three–valued models. Let us continue with the admissible semantics. First, we will tie the notion of decisiveness to admissibility, following the comparison of completions and extending interpretations that we have presented in Section \[sec:premdec\]. Let $v$ be a three–valued interpretation and $v'$ its (maximal) two–valued sub–interpretation. $v$ is admissible iff all arguments mapped to ${\mathbf{t}}$ are decisively in w.r.t. $v'$ and all arguments mapped to ${\mathbf{f}}$ are decisively out w.r.t. $v'$. \[thm:declab\] Please note that this result does not imply that admissible extensions and labelings “perfectly” coincide. In labelings, we guess an interpretation, and thus assign initial values to arguments that we want to verify later. If they are self–dependent, it of course affects the outcome. In the extension based approaches, we distinguish whether this dependency is permitted. Therefore, the aa– and cc– approaches will have a corresponding labeling, but not vice versa. Let $E$ be a cc–admissible and $A$ an aa–admissible extension. The ${\mathbf{u}}$–completions of $v_E$ and $v_A^a$ are admissible labelings. \[thm:extlabadm\] Let us now consider the preferred semantics. Information maximality is not the same as maximizing the set of accepted arguments and due to the behavior of $\Gamma_D$ we can obtain a preferred interpretation that can map to ${\mathbf{t}}$ a subset of arguments of another interpretation. Consequently, we fail to receive an exact correspondence between the semantics. By this we mean that given a framework there can exist an (arbitrary) preferred extension without a labeling counterpart and a labeling without an appropriate extension of a given type. For any xy–preferred extension there might not exist a corresponding preferred labeling and vice versa. Let us look at $ADF_1 = (\{a,b,c\}, \{C_a:\neg a, C_b: a, C_c: \neg b \lor c\}) $, as depicted in Figure \[figa\]. $a$ and $b$ cannot form a conflict–free extension to start with, so we are only left with $c$. However, the attack from $b$ on $c$ can be only overpowered by self–support, thus it cannot be part of an aa–admissible extension. Therefore, we obtain only one aa–preferred extension, namely the empty set. The single preferred labeling solution would be $v=\{a: {\mathbf{u}}, b:{\mathbf{u}}, c:{\mathbf{t}}\}$ and we can see there is no correspondence between the results. On the other hand, there is one with the cc–preferred extension $\{c\}$. Finally, we have $ADF_2 =(\{a,b,c\}, \{C_a:\neg a \land b, C_b: a,C_c: \neg b\}) $ depicted in Figure \[figb\]. The preferred labeling is $\{a:{\mathbf{f}}, b:{\mathbf{f}}, c:{\mathbf{t}}\}$. The single cc–preferred extension is $\emptyset$ and again, we receive no correspondence. However, it is compliance with the aa–preferred extension $\{c\}$. [0.20]{} \[->,>=stealth,shorten >=1pt,auto,node distance=1.5cm, thick,main node/.style=[circle,fill=none,draw,minimum size = 0.7cm,font=]{}, condition/.style=[circle,fill=none,draw=none,minimum size = 0.3cm,font=]{}\] \(a) [$a$]{}; (b) \[right of=a\] [$b$]{}; (c) \[right of=b\] [$c$]{}; (ca) \[above of= a, yshift=-0.8cm\] [$\neg b$]{}; (cb) \[above of= b, yshift=-0.8cm\] [$a$]{}; (cc) \[above of=c, yshift=-0.8cm\] [$\neg b \lor c$]{}; (a) edge \[bend left\] node (b) (b) edge \[bend left\] node (a) (c) edge \[loop below\] node (c) (b) edge node (c); [0.20]{} \[->,>=stealth,shorten >=1pt,auto,node distance=1.5cm, thick,main node/.style=[circle,fill=none,draw,minimum size = 0.7cm,font=]{}, condition/.style=[circle,fill=none,draw=none,minimum size = 0.3cm,font=**]{}\] \(a) [$a$]{}; (b) \[right of=a\] [$b$]{}; (c) \[right of=b\] [$c$]{}; (ca) \[above of= a, yshift=-0.8cm\] [$\neg a \land b$]{}; (cb) \[above of= b, yshift=-0.8cm\] [$a$]{}; (cc) \[above of=c, yshift=-0.8cm\] [$\neg b$]{}; (a) edge \[loop below\] node (a) (a) edge \[bend left\] node (b) (b) edge \[bend left\] node (a) (b) edge node (c); The labeling–based complete semantics can also be defined in terms of decisiveness: Let $v$ be a three–valued interpretation and $v'$ its (maximal) two–valued sub–interpretation. $v$ is complete iff all arguments decisively out w.r.t. $v'$ are mapped to ${\mathbf{f}}$ by $v$ and all arguments decisively in w.r.t. $v'$ are mapped to ${\mathbf{t}}$ by $v$. \[thm:complab\] Fortunately, just like in the case of admissible semantics, complete extensions and labelings partially correspond: Let $E$ be a cc–complete and $A$ an aa–complete extension. The ${\mathbf{u}}$–completions of $v_E$ and $v_A^a$ are complete labelings. Please recall that in the Dung setting, extensions and labelings agreed on the sets of accepted arguments. In ADFs, this relation is often only one way – like in the case of admissible and complete cc– and aa– sub–semantics – or simply nonexistent, like in preferred approach. In this context, the labeling–based admissibility (and completeness) can be seen as the most general one. This does not mean that specializations, especially handling cycles, are not needed. Even more so, as to the best of our knowledge no methods for ensuring acyclicity in a three–valued setting are yet available. Due to the fact that the grounded semantics has a very clear meaning, it is no wonder that both available approaches coincide, as already noted in [@tofix:newadf]. We conclude this section by relating both available notions of stability. The relevant proofs can be found in [@report:semantics]. The two–valued grounded extension and the grounded labeling correspond. A set $M \subseteq S$ of arguments is labeling stable iff it is extension–based stable. Concluding Remarks ================== In this paper we have introduced a family of extension–based semantics as well as their classification w.r.t. positive dependency cycles. Our results also show that they satisfy ADF versions of Dung’s Fundamental Lemma and that appropriate sub–semantics preserve the relations between stable, preferred and complete semantics. We have also explained how our formulations relate to the labeling–based approach. Our results show that the precise correspondence between the extension–based and labeling–based semantics, that holds in the Dung setting, does not fully carry over. It is easy to see that in a certain sense, labelings provide more information than extensions due to distinguishing false and undecided states. Therefore, one of the aims of our future work is to present the sub–semantics described here also in a labeling form. However, since our focus is primarily on accepting arguments, a comparison w.r.t. information content would not be fully adequate for our purposes and the current characteristic operator could not be fully reused. We hope that further research will produce satisfactory formulations. [^1]: The author is funded by the Vienna PhD School of Informatics. This research is a part of the project I1102 supported by the Austrian Science Fund FWF. [^2]: Please note defense is often also termed acceptability, i.e. if a set defends an argument, the argument is acceptable w.r.t. this set. [^3]: A partial order $(A,\leq)$ is a complete semilattice iff each nonempty subset of $A$ has a glb and each increasing sequence of $S$ has a lub. [^4]: The presented definition is slightly different from the one available in [@inproc:eas]. The new version was obtained through personal communication with the author. [^5]: Please note that it is not required that $B\subseteq A$	{ "pile_set_name": "ArXiv" }
--- address: \| Physics Department, Old Dominion University,\ Norfolk, VA 23529, USA\ and\ Thomas Jefferson National Accelerator Facility\ Newport News, VA 23606, USA author: - 'A.V. RADYUSHKIN[^1]' title: 'QCD SUM RULES AND VIRTUAL COMPTON SCATTERING[^2]' --- JLAB-THY-96-06\ September 1996 Soft $vs.$ hard dynamics in QCD ================================ [QCD and virtual Compton scattering.]{} The kinematics of the amplitude of the process $\gamma^* p \to \gamma p'$ can be specified by the initial nucleon momentum $p$, the momentum transfer $r=p-p'$ and the momentum $q$ of the initial virtual photon, $q^2 \equiv -Q^2$. The final photon momentum $q'$ is then given by $q'= q +r$ with $q'^2=0$. Other important momentum invariants are $t \equiv (p'-p)^2=(q-q')^2$ and $s \equiv (q+p)^2$. Taking $Q^2$ large, $i.e.,$ at least above $1 \, GeV^2$, one can hope to enter the region where the amplitude is dominated by short distances between the two photon vertices and pQCD may be applicable. In this situation, it is tempting to speak about the “virtual Compton scattering on a single quark” implying that the large-$Q^2$ behaviour is given just by the quark propagator (see Fig.$1a$), while the long-distance information is accumulated in a distribution function $F(X,t)$ described by the matrix element of $\langle p' \| \bar q \ldots q \| p \rangle$ type. However, the factorization formula $$M(Q^2,s,t) \to \int m(s/Q^2,X) F(X,t) dX \label{fact}$$ only makes sense if $\|t\| \ll Q^2$. Otherwise, if $\|t\| \sim Q^2$, large momentum enters into the hadron wave function and one deals with the scattering process on the hadron as a whole. For $\|t\| \ll Q^2$, the function $F(X,t)$ in eq.(\[fact\]) looks like a parton distribution function $f(x)$ with an additional form-factor-type dependence on $t$. To make analogy with deep inelastic scattering, it is instructive to recall that the imaginary part of the virtual forward Compton amplitude (for which $q'=q$ and $p'=p$) in the limit of large $Q^2$ and fixed Bjorken variable $\zeta \equiv Q^2/2(pq)$ can be written as $$\begin{aligned} \int_0^1 f(x) \delta((q+xp)^2) \cdot 2 (qp) dx = \int_0^1 f(x) \delta(x-Q^2/ 2 (qp) ) dx = f(\zeta)\, , \end{aligned}$$ where $xp$ is the fraction of the initial hadron momentum carried by the interacting quark. The usual parton distribution functions $f(\zeta)$ correspond to exactly forward matrix elements, with $r \equiv p' -p =0$, while the kinematics of VCS requires that $r \neq 0$ and $t \equiv r^2 \neq 0$. Hence, we need a new type of parton functions $F(X,t)$ [@ji]. In the limit $t \to 0$, they reduce to the “asymmetric distribution functions” $F(X)$ [@compton; @gluon]. Hence, the studies of deeply virtual Compton scattering (DVCS) are related to a new field of pQCD applications. As shown in refs.[@compton; @gluon], the asymmetric distributions $F(X)$ have features of both the distribution functions and distribution amplitudes (wave functions). A more detailed discussion of DVCS at small $t$ will be given in Section 3. Another situation in which pQCD is applicable is when both $Q^2$ and $\|t\|$ are asymptotically large. Then the virtual Compton scattering amplitude factorizes into a convolution of the short-distance amplitude $m(\{x_i\},\{y_j\},Q^2,s,t)$ and two distribution amplitudes $\varphi(x_1,x_2,x_3)$, $\varphi(y_1,y_2,y_3)$ describing the proton in the initial and final state, respectively (see Fig.$1b$). They are related to matrix elements of $\langle 0\|q\ldots q \ldots q \|p \rangle$ type and give the probability amplitude, that $e.g.,$ the initial proton can be treated as three collinear quarks with the momentum $p$ divided into fractions $x_1 p,x_2 p, x_3 p$ with $x_1+x_2+x_3 =1$. The short-distance amplitude $m(\{x_i\},\{y_j\},Q^2,s,t)$ is given by Feynman diagrams involving two hard gluon exchanges, suppressed by a factor $(\alpha_s/\pi)^2 \sim 1/100$ compared to the “soft contribution” produced by a simple overlap of soft wave functions, without any gluon exchanges. The soft term, however, has an extra power of $1/Q^2$ for large $Q^2$. As a result, the hard term asymptotically dominates, though the soft term may be much larger than the hard one for accessible $Q^2$. [Quark-hadron duality and $\gamma^ p \to \Delta^+$ transition.]{} In particular, a purely soft contribution to $G_M^p(Q^2)$ calculated within the local quark-hadron duality approach [@nr83] is in good agreement with experimental data up to $Q^2 \sim \, 20 \, GeV^2$. The same approach was used recently [@del] to get the estimates of the soft term for the $\gamma^ p \to \Delta^+$ transition. For the magnetic form factor $G_M^(Q^2)$ these estimates are rather close to the results of the analysis of inclusive SLAC data [@stoler; @keppel]. A small value for the ratio $G_E^(Q^2)/G_M^(Q^2)$ obtained in ref.[@del] is also in agreement with available data [@burkert], in contrast to the pQCD prediction [@carlson] which gives $G_E^(Q^2)/G_M^(Q^2) \to \, -1$ for the ratio of hard contributions. Within different nonperturbative approaches [@nr82; @nr83; @chizhit; @kroll96], it was observed that soft terms are sufficiently large to describe the data or that the hard terms are too small compared to the data. Hence, there is a growing evidence that soft terms dominate the exclusive amplitudes at accessible energies. Of course, the magnitude of the hard contribution depends on the shape of distribution amplitudes (DA’s). The latter are usually integrated with the weights like $1/x_1 x_2$, so the humpy DA’s of Chernyak-Zhitnitsky (CZ) type [@cz82; @cz84] produce contributions which are much larger than those obtained with smooth DA’s close to the “asymptotic” forms. The CZ wave functions were originally motivated by QCD sum rule analysis [@cz82]. However, the results of the QCD sum rule calculations of the DA’s are extremely model-dependent and unreliable. Furthermore, for the theoretically most clean case of the $\gamma^ \gamma \to \pi^0$ form factor, both a direct QCD sum rule calculation of this form factor and available experimental data show no enhancement compared to the pQCD result obtained with the asymptotic DA for the pion. $\gamma^* \gamma \to \pi^0$ form factor ======================================= [pQCD analysis.]{} The transition $\gamma^* \gamma^* \to \pi^0$ of two virtual photons $\gamma^$ into a neutral pion provides an exceptional opportunity to test QCD predictions for exclusive processes. In the lowest order of perturbative QCD, its asymptotic behaviour is due to the subprocess $\gamma^(q) + \gamma^(q^{\prime}) \to \bar q(\bar xp) + q (xp) $ with $x$ ($\bar x$) being the fraction of the pion momentum $p$ carried by the quark produced at the $q$ ($q')$ photon vertex (see Fig.1$c$). The relevant diagram is similar to the handbag diagram for deep inelastic scattering, with the main difference that one should use the pion distribution amplitude $\varphi_{\pi}(x)$ instead of parton densities. For large $Q^2$, the perturbative QCD prediction is given by [@bl80]: $$F_{\gamma^ \gamma^* \pi^0 }^{as}(Q^2, q^{\prime 2}) = \frac{4\pi}{3} \int_0^1 {{\varphi_{\pi}(x)}\over{xQ^2 - \bar x q^{\prime 2} }} \, dx \stackrel{q^{\prime 2}=0}{\longrightarrow} \frac{4\pi}{3} \int_0^1 {{\varphi_{\pi}(x)}\over{xQ^2}} \, dx \equiv \frac{4\pi}{3Q^2} I \, . \label{eq:ggpipqcd}$$ Experimentally, the most important situation is when the lower virtuality photon is (almost) real $q^{\prime 2} \approx 0$. In this case, necessary nonperturbative information is accumulated in the same integral $I$ (see eq.(\[eq:gg\pipqcd\])) that appears in the one-gluon-exchange diagram for the pion electromagnetic form factor [@cz84; @pl80; @blpi79]. The value of $I$ depends on the shape of the pion distribution amplitude $\varphi_{\pi}(x)$. In particular, using the asymptotic form $ \varphi_{\pi}^{as}(x) = 6 f_{\pi} x \bar x $ [@pl80; @blpi79] gives $F_{\gamma \gamma^* \pi^0 }^{as}(Q^2) = 4 \pi f_{\pi}/Q^2 $ for the asymptotic behaviour [@bl80]. If one takes the Chernyak-Zhitnitsky form [@cz82] $\varphi_{\pi}^{CZ}(x) = 30 f_{\pi} x(1-x)(1-2x)^2$, the integral $I$ increases by a sizable factor of 5/3, and this difference can be used for experimental discrimination between the two forms. Note, that the pQCD hard scattering term for $\gamma \gamma^* \to \pi^0$ has the zeroth order in the QCD coupling constant $\alpha_s$, just like in deep inelastic scattering. Hence, there are good reasons to expect that pQCD for $F_{\gamma \gamma^* \pi^0 }(Q^2)$ may work at rather low $Q^2$. The $Q^2=0$ limit of $F_{\gamma \gamma^* \pi^0 }(Q^2)$ is known from $\pi^0 \to \gamma \gamma$ decay rate. Using PCAC and ABJ anomaly [@ABJ], one can calculate $F_{\gamma \gamma^* \pi^0 }(0)$ theoretically: $ F_{\gamma \gamma^* \pi^0 }(0) =1/ \pi f_{\pi} . $ It is natural to expect that a complete QCD result does not strongly deviate from a simple interpolation [@blin] $ \pi f_{\pi} F_{\gamma \gamma^* \pi^0 }(Q^2) = 1/(1+ Q^2/4 \pi^2 f_{\pi}^2) $ between the $Q^2=0$ value and the large-$Q^2$ asymptotics [^3]. This interpolation implies the asymptotic form of the distribution amplitude for the large-$Q^2$ limit and agrees with CELLO experimental data [@CELLO]. It was also claimed [@CLEO] that the new CLEO data available up to $8 \, GeV^2$ also agree with the interpolation formula. Comparing the data with theoretical predictions, one should take into account the one-loop pQCD radiative corrections to the hard scattering amplitude calculated in ref.[@braaten]. Effectively, the correction decreases the leading-order result by about $20 \%$, still leaving a sizable gap between the prediction based on the CZ amplitude and the phenomenologically successful Brodsky-Lepage interpolation formula[@blin]. Hence, the new preliminary data[@CLEO] seem to indicate that the magnitude of $I$ is close to that corresponding to the asymptotic form of the pion distribution amplitude. Because of the far-reaching consequences of this conclusion, it is desirable to have a direct calculation of the $\gamma\gamma^* \to \pi^0$ form factor in the intermediate region of moderately large momentum transfers $Q^2 {\raisebox{-.2ex}{$\stackrel{\textstyle>} {\raisebox{-.6ex}[0ex][0ex]{$\sim$}}$}}1 \, GeV^2$. As we will see below, the QCD sum rules allow one to calculate $F_{\gamma \gamma^* \pi^0 }(Q^2)$ for large $Q^2$ without any assumptions about the shape of the pion distribution amplitude, and the result can be used to get information about $\varphi_{\pi}(x)$. [QCD sum rules and pion distribution amplitude.]{} The CZ sum rule written directly for the pion distribution amplitude $\varphi_\pi(x)$ is $$\begin{aligned} f_\pi\varphi_\pi(x)&=&\frac{3M^2}{2\pi^2}(1-e^{-s_0/M^2})x(1-x) +\frac{\alpha_s\langle GG\rangle}{24\pi M^2}[\delta(x)+\delta(1-x)] \nonumber \\ &+ & \frac{8}{81}\frac{\pi\alpha_s\langle\bar qq\rangle^2}{M^4} \{11[\delta(x)+\delta(1-x)]+2[\delta^\prime(x)+\delta^\prime(1-x)]\}. \label{eq:wfsr}\end{aligned}$$ Here, $M$ is the auxiliary Borel parameter which must be taken in the region where the r.h.s. is least sensitive to its variations, and $s_0$ is the effective onset of the continuum fitted to maximize the $M^2$-stability region. From the QCD sum rule for $f_{\pi}$, [@svz] $s_0 \approx 0.7 \, GeV^2$. As emphasized in ref.[@MR], the lowest condensates $\langle GG\rangle$ and $\langle \bar qq\rangle^2$ taken into account in eq.(\[eq:wfsr\]) do not provide all the information necessary for a reliable determination of $\varphi_\pi(x)$. The humpy CZ shape is, in fact, a compromise between the $\delta(x)$, $\delta(1-x)$ condensate peaks and the smooth $x(1-x)$ behaviour of the perturbative term. Adding higher condensates, $e.g.,$ $\langle \bar q D^2 q\rangle$, one would get even higher derivatives of $\delta(x)$ and $\delta(1-x)$. The sum of such singular terms can be treated as an expansion of some finite-width function $\delta \varphi (x)$: $$\delta \varphi (x) = a_0 \delta(x) + a_1 \delta^{\prime} (x) + a_2 \delta^{\prime \prime}(x) + \ldots + \{x \to 1-x \}.$$ Of course, the knowledge of $a_0$ alone is not sufficient for a reliable reconstruction of $\delta \varphi (x)$. On the other hand, the higher coefficients $a_1, a_2, etc.$ are given by a sum of several higher condensates whose magnitudes are completely unknown. Hence, no strict conclusions can be made. The CZ procedure is equivalent to assuming that $a_1, a_2, \ldots \, \sim 0$, though other choices ($e.g.,$ nonlocal condensate model [@MR]) may look more realistic. [QCD sum rule for doubly virtual form factor $F_{\gamma^ \gamma^* \pi^0} (q^2, q^{\prime 2})$.]{} Instead of following the steps dictated by the old logic: $1)$ pQCD factorization for $F_{\gamma^ \gamma \pi^0} (Q^2)$; $2)$ QCD sum rules for the moments of $\varphi_{\pi} (x)$ (which are unreliable); $3)$ calculation of $I= \int_0^1 \varphi_{\pi} (x)/x \, dx$, we developed in ref. [@rr] the approach which $1)$ starts with the QCD sum rule for $ F_{\gamma^* \gamma^* \pi^0} (q^2, q^{\prime 2})$ in the $q^{\prime 2} \to 0$ limit; $2)$ information about $I$ is extracted from this sum rule and $3)$ then used to make conclusions about the shape of $\varphi_{\pi} (x)$. When both virtualities of the photons are large, we have the following QCD sum rule: $$\begin{aligned} \pi f_{\pi} \mbox{$F_{\gamma^\gamma^\pi^\circ}$}(Q^2, q^{\prime 2})= 2\int_0^{s_o} ds \, e^{-s/{M^2}} \int_0^1 \frac{x\bar{x}(xQ^2 - \bar x q^{\prime 2})^2} {[s{x}\bar{x}+xQ^2 - \bar x q^{\prime 2}]^3} \,dx \, \nonumber \\ +\frac{\pi^2}{9} {\langle \frac{\alpha_s}{\pi}GG \rangle} \left(\frac{1}{2M^2 Q^2} - \frac{1}{2M^2 q^{\prime 2}} + \frac1{Q^2 q^{\prime 2}}\right) \nonumber\\ + \frac{64}{243}\pi^3\alpha_s{\langle \bar{q}q\rangle}^2 \left( \frac1{M^4} \left [ \frac{Q^2}{q^{\prime 4}} - \frac9{2q^{\prime 2}}+\frac9{2Q^2}-\frac{q^{\prime 2}}{Q^4} \right ] + \frac9{Q^2 q^{\prime 4}} -\frac9{Q^4q^{\prime 2} } \right ) . \label{eq:SR1}\end{aligned}$$ In this situation, the pQCD approach is also expected to work. Indeed, neglecting the $s{x}\bar{x}$-term compared to $xQ^2 - \bar x q^{\prime 2}$ and keeping only the leading $O(1/Q^2)$ and $O(1/q^{\prime 2})$ terms in the condensates, we can write eq.(\[eq:SR1\]) as $$\begin{aligned} \mbox{$F_{\gamma^\gamma^\pi^\circ}$}(,Q^2) = \frac{4\pi}{3f_{\pi}} \int_0^1 \frac{dx}{ xQ^2 - \bar x q^{\prime 2}} \, \left \{ \frac{3M^2}{2\pi^2}(1-e^{-s_0/M^2}) x\bar{x} \right. \nonumber \\ \left. + \frac{1}{24M^2} \langle \frac{\alpha_s}{\pi}GG\rangle [\delta(x) + \delta (\bar{x})] \right. \nonumber \\ + \left. \frac{8}{81M^4}\pi\alpha_s{\langle \bar{q}q\rangle}^2 \biggl ( 11[\delta(x) + \delta (\bar{x})] + 2[\delta^{\prime}(x) + \delta ^{\prime}(\bar{x})] \biggr ) \right \} \label{eq:SRlargeQ2wf}. \end{aligned}$$ The expression in curly brackets coincides with the QCD sum rule (\[eq:wfsr\]) for the pion distribution amplitude $f_{\pi} \varphi_{\pi}(x)$. Hence, when both $Q^2$ and $q^{\prime 2}$ are large, the QCD sum rule (\[eq:SR1\]) exactly reproduces the pQCD result (\[eq:gg\pipqcd\]). One may be tempted to get a QCD sum rule for the integral $I$ by taking $ q^{\prime 2}=0$ in eq.(\[eq:SR1\]). Such an attempt, however, fails immediately because of the power singularities $1/q^{\prime 2}$, $1/ q^{\prime 4}$, $etc.$ in the condensate terms. It is easy to see that these singularities are produced by the $\delta(x)$ and $\delta'(x)$ terms in eq.(\[eq:SRlargeQ2wf\]). In fact, it is precisely these terms that generate the two-hump form for $\varphi_{\pi}(x)$ in the CZ-approach [@cz82]. The advantage of having a direct sum rule for $F_{\gamma \gamma^ \pi^0}(Q^2,q^{\prime 2})$ is that the small-$q^{\prime 2}$ behavior of $F_{\gamma \gamma^* \pi^0}(Q^2,q^{\prime 2})$ is determined by the position of the closest resonances in the $q^{\prime }$ channel, which is known. Eventually, $1/q^{\prime 2}$ is substituted for small $q^{\prime 2}$ by something like $1/m_{\rho}^2$ and the QCD sum rule in the $q^{\prime 2}$ limit is $$\begin{aligned} & \,& \pi f_{\pi} F_{\gamma \gamma^* \pi^0}(Q^2) = \int_0^{s_0} \left \{ 1 - 2 \frac{Q^2-2s}{(s+Q^2)^2} \left (s_{\rho} - \frac{s_{\rho}^2}{2 m_{\rho}^2} \right ) \right. \nonumber \\ &+& \left. 2\frac{Q^2-6s+3s^2/Q^2}{(s+Q^2)^4} \left (\frac{s_{\rho}^2}{2} - \frac{s_{\rho}^3}{3 m_{\rho}^2} \right ) \right \} e^{-s/M^2} \frac{Q^2 ds }{(s+Q^2)^2} \nonumber \\ &+&\frac{\pi^2}{9} {\langle \frac{\alpha_s}{\pi}GG \rangle} \left \{ \frac{1}{2 Q^2 M^2} + \frac{1}{Q^4} - 2 \int_0^{s_0} e^{-s/M^2} \frac{ds }{(s+Q^2)^3} \right \} \nonumber \\ &+&\frac{64}{27}\pi^3\alpha_s{\langle \bar{q}q\rangle}^2 \lim_{\lambda^2 \to 0} \left \{ \frac1{2Q^2 M^4} + \frac{12}{Q^4 m_{\rho}^2 } \left [ \log \frac{Q^2}{\lambda ^2} -2 \right. \right. \nonumber \\ &+& \left. \left. \int_0^{s_0} e^{-s/M^2} \left ( \frac{s^2+3sQ^2+4Q^4} {(s+Q^2)^3} - \frac1{s+\lambda ^2} \right) ds \right] \right. \label{eq:finsr} \\ &-& \left. \frac4{Q^6} \left [ \log \frac{Q^2}{\lambda^2} -3+ \int_0^{s_0} e^{-s/M^2} \left ( \frac{s^2+3sQ^2+6Q^4} {(s+Q^2)^3} - \frac1{s+\lambda ^2} \right) ds \right] \right \} .\nonumber\end{aligned}$$ In Fig.\[fig:3\], we present a curve for $Q^2F_{\gamma \gamma^* \pi^0}(Q^2)/4\pi f_{\pi}$ calculated from eq.(\[eq:finsr\]) for standard values of the condensates, $\rho$- and $\pi$-meson duality intervals $s_{\rho} = 1.5 \, GeV^2$, [@svz], $s_0 = 0.7 \, GeV^2$ and $M^2 = 0.8\, GeV^2$. It is rather close to the curve corresponding to the Brodsky-Lepage interpolation formula $\pi f_{\pi} F_{\gamma \gamma^* \pi^0}(Q^2) = 1/(1+Q^2/4\pi^2 f_{\pi}^2)$ and to that based on the $\rho$-pole approximation $\pi f_{\pi} F(Q^2) = 1/(1+Q^2/m_{\rho}^2)$. Hence, our result favors a pion distribution amplitude which is close to the asymptotic form. It should be noted, that the $\rho$-pole behaviour in the $Q^2$-channel has nothing to do with the explicit use of the $\rho$-contributions in our models for the correlators in the $q^{\prime 2}$-channel: the $Q^2$-dependence of the $\rho$-pole type emerges due to the fact that the pion duality interval $s_0 \approx 0.7 \, GeV^2$ is numerically close to $m_{\rho}^2\approx 0.6\,GeV^2$. Taking the lowest-order perturbative spectral density $\rho^{quark}(s,q^{\prime 2}=0 ,Q^2) = {{Q^2}/{(s+Q^2)^2}}$ and assuming the local quark-hadron duality, we obtain the result $$f_{\pi} F_{\gamma \gamma^* \pi^0}^{LD}(Q^2) = \frac1{\pi } \int_0^{s_0} \rho^{quark}(s,0,Q^2) \, ds = \frac1{\pi (1+Q^2/s_0)} \label{eq:FLDgg}$$ coinciding, for $s_0=4 \pi^2 f_{\pi}^2 \approx 0.67 \, GeV^2$ with the BL-interpolation formula. [Lessons.]{} $1)$ CZ sum rule is an unreliable source of information about the pion DA; $2)$ Pion DA is narrow; $3)$ Since the diagrams for the nucleon DA’s have the structure as in the pion case, we should expect that the nucleon DA’s are also close to asymptotic and , hence, the two-gluon-exchange hard terms are very small for accessible $Q^2$ and $t$. Thus, it is very important to get the estimates for the soft contributions to the virtual Compton scattering amplitude (see, $e.g.,$ refs. [@crs], where the high-$t$ real Compton scattering on the pion was considered). Small-$t$, large-$Q^2$ limit of VCS: a new pQCD area ==================================================== Recently, X. Ji [@ji] suggested to use the deeply virtual Compton scattering (DVCS) to get information about some parton distribution functions inaccessible in standard inclusive measurements. He also emphasized that the DVCS amplitude has a scaling behavior in the region of small $t$ and fixed $x_{Bj}$ which makes it a very interesting object on its own ground. [Double distributions.]{} In the scaling limit, the square of the proton mass $m_p^2=p^2$ can be neglected compared to the virtuality $Q^2 \equiv -q^2$ of the initial photon and the energy invariant $p \cdot q \equiv m_p \nu$. Thus, we set $p^2=0$ and, for small $t$, we also have $r^2 = 0$. Then the requirement $p'^2 \equiv (p-r)^2=p^2$ reduces to the condition $p\cdot r = 0$ which can be satisfied only if $r$ is proportional to $p$: $r= \zeta p$, where $\zeta$ coincides with the Bjorken variable $x_{Bj} \equiv Q^2/2(p \cdot q)$, $0 \leq x_{Bj} \leq 1$. Naturally, the light-like limit of 4-momenta $p$, $r$ is more convenient to visualize in a frame where the initial proton is moving fast, rather than in its rest frame. Though the momenta $p$ and $r$ are proportional to each other, one should make a clear distinction between them since $p$ and $r$ specify the momentum flow in two different channels. Since the initial quark momentum originates both from $p$ and $r$, we write it as $xp +y r$. In more formal terms, the relevant light-cone matrix elements are parameterized as $$\begin{aligned} && \hspace{-6mm} \langle p-r\, \| \, \bar \psi_a(0) \hat z E(0,z;A) \psi_a(z) \, \| \, p \rangle \|_{z^2=0} = \bar u(p-r) \hat z u(p) \label{eq:vec} \\ && \hspace{-6mm} \int_0^1 \int_0^1 \, \left ( e^{-ix(pz)-iy(r z)}F_a(x,y) \right.- \left. e^{ix(pz)-i\bar y(r z)}F_{\bar a}(x,y) \right ) \, \theta( x+y \leq 1) dy \, dx , \nonumber \end{aligned}$$ $etc.,$ where $\hat z \equiv \gamma_{\mu} z^{\mu}$ and $\bar u(p-r), u(p)$ are the Dirac spinors for the nucleon. Taking the limit $r =0$ gives the matrix element defining the parton distribution functions $f_a(x)$, $f_{\bar a}(x)$. This leads to the reduction formula: $$\int_0^{1-x} \, F_a(x,y)\, dy= f_a(x) . \label{eq:redf}$$ [Asymmetric distribution functions.]{} Since $r = \zeta p$, the variable $y$ appears in eq.(\[eq:vec\]) only in $x+y\zeta \equiv X$ and $x- \bar y\zeta \equiv X - \zeta$ combinations, where $X$ and $(X - \zeta)$ are the total fractions of the initial hadron momentum $p$ carried by the quarks. Integrating the double distribution $F(X-y \zeta,y)$ over $y$ we get the asymmetric distribution function $${\cal F}_{\zeta}^a (X) = \int_0^{{\rm min} \{ X/\zeta, \bar X / \bar \zeta \}} F_a(X-y \zeta,y) \, dy, \label{eq:asdf}$$ where $\bar \zeta \equiv 1- \zeta$. Since $\zeta \leq 1$ and $x+y \leq 1$, the variable $X$ satisfies a natural constraint $0\leq X \leq 1$. In the region $X > \zeta$ (Fig.4$a$), the initial quark momentum $Xp$ is larger than the momentum transfer $r = \zeta p$, and we can treat ${\cal F}_{\zeta}^a (X)$ as a generalization of the usual distribution function $f_a(x)$. In this case, the quark goes out of the hadron with a positive fraction $Xp$ of the original hadron momentum and then comes back into the hadron with a changed (but still positive) fraction $(X - \zeta)p$. The Bjorken ratio $\zeta$ specifies the momentum asymmetry of the matrix element. Hence, one deals now with a family of asymmetric distribution functions ${\cal F}_{\zeta}^a (X)$ whose shape changes when $\zeta$ is changed. The basic distinction between the double distributions $F(x,y)$ and the asymmetric distribution functions ${\cal F}_{\zeta} (X)$ is that the former do not depend on the momentum asymmetry parameter $\zeta$, while the latter are explicitly labelled by it. When $\zeta \to 0$, the limiting curve for ${\cal F}_{ \zeta}(X)$ reproduces the usual distribution function: $${\cal F}^a_{\zeta=0} \, (X) = f_a(X) \ . \label{eq:Fzeta0}$$ Another region is $X < \zeta$ (Fig.4$b$), in which the “returning” quark has a negative fraction $(X- \zeta)$ of the light-cone momentum $p$. Hence, it is more appropriate to treat it as an antiquark going out of the hadron and propagating together with the original quark. Writing $X$ as $X = Y \zeta$, we see that the quarks carry now positive fractions $Y \zeta p \equiv Y r$ and $\bar Y r \equiv (1-Y)r $ of the momentum transfer $r$, and the asymmetric distribution function in the region $X= Y \zeta < \zeta$ looks like a distribution amplitude $\Psi_{\zeta}(Y)$ for a $\bar q q $ state with the total momentum $r= \zeta p$: $$\Psi_{\zeta}(Y) = \int_0^Y F((Y-y) \zeta , y ) \, dy . \label{eq:Psi}$$ [Leading-order contribution.]{} Using the parameterization for the matrix elements given above, we get a parton-type representation for the handbag contribution at $t=0$: $$T^{\mu \nu}_{symm} (p,q,r) = \left (g^{\mu \nu} -\frac1{p \cdot q } (p^{\mu}q^{\nu} +p^{\nu}q^{\mu}) \right ) \, \sum_a e_a^2\, \sqrt{1- \zeta} \ ( T_V^a(\zeta ) + T_V^{\bar a}(\zeta ) ) ,$$ where only the $\{\mu \leftrightarrow \nu \}$-symmetric part is shown explicitly and $T_V^a(\zeta )$ is the invariant amplitude depending on the scaling variable $\zeta $: $$T_V^a(\zeta ) = \int_0^{1} \left ( \frac1{X-\zeta +i\epsilon} + \frac1{X} \right ) {\cal F}_{\zeta}^a (X) \, dX \, . \label{eq:tv}$$ The term containing $1/(X-\zeta +i\epsilon)$ generates the imaginary part: $$- \frac1{\pi}\, {\rm Im} \, T_V^a(\zeta ) = {\cal F}^a_{\zeta} \, (\zeta)\, . \label{eq:imtv}$$ Though ${\cal F}_{\zeta = 0}^a(X) = f_a(X)$, in the general case when $\zeta \neq 0$, these two functions differ. Furthermore, the imaginary part appears for $X= \zeta$, $i.e.,$ in a highly asymmetric configuration in which the second quark carries a vanishing fraction of the original hadron momentum, in contrast to the usual distribution $f_a(\zeta)$ which corresponds to a symmetric configuration with the final quark having the momentum equal to that of the initial one. A characteristic feature of the asymmetric distribution functions ${\cal F}_{\zeta}^a(X)$ is that they rapidly vary in the region $X {\raisebox{-.2ex}{$\stackrel{\textstyle<} {\raisebox{-.6ex}[0ex][0ex]{$\sim$}}$}}\, \zeta$ and vanish for $X=0$. However, the limiting curve ${\cal F}_{\zeta=0}(X)$ does not necessarily vanish for $X=0$, $i.e.,$ the limits $\zeta \to 0$ and $X \to 0$ do not commute. For this reason, if $\zeta$ is small, the substitution of ${\cal F}_{\zeta}^a(X)$ by $f_a(X)$ may be a good approximation for all $X$-values except for the region $X {\raisebox{-.2ex}{$\stackrel{\textstyle<} {\raisebox{-.6ex}[0ex][0ex]{$\sim$}}$}}\, \zeta$, and it is not clear [a priori]{} how close are the functions ${\cal F}_{\zeta}^a(\zeta)$ and $ f_a(\zeta)$. [ Evolution of the double distributions.]{} The purely scaling behavior of the DVCS amplitude is violated by the logarithmic $Q^2$-dependence of $F_{NS}(x,y;Q^2)$ governed by the evolution equation $$Q \frac{d}{d Q} F_{NS}(x,y;Q^2) = \int_0^1 d \xi \int_0^1 R_{NS} (x,y; \xi, \eta;g) F_{NS}( \xi, \eta;Q^2) d \eta \label{eq:nfwdev}$$ (the flavor-nonsinglet (NS) component is taken for simplicity). Since integration over $y$ converts $F_{NS} (x,y)$ into the parton distribution function $f_{NS} (x)$, whose evolution is governed by the GLAPD equation [@gl; @ap; @d], our kernel has the property $$\int_0^ {1-x} R_{NS} (x,y; \xi, \eta;g) d y = \frac1{\xi} P_{NS} (x/\xi). \label{eq:rtop}$$ For a similar reason, integrating $R_{NS}(x,y; \xi, \eta;g)$ over $x$ one should get the evolution kernel $V(y,\eta;g)$ [@blpi79; @pl80] for the pion distribution amplitude $$\int_0^{1-y} R_{NS}(x,y; \xi, \eta;g) d x = V(y,\eta;g). \label{eq:rtov}$$ In the formal $Q^2 \to \infty$ limit, $F(x,y; Q^2\to \infty) \sim \delta(x) y \bar y,$ $i.e.,$ in each of its variables $x,y$, the double distribution tends to the characteristic asymptotic form: $\delta(x)$ is specific for the distribution functions, while the $y \bar y$-form is the asymptotic shape for the lowest-twist two-body distribution amplitudes [@pl80; @blpi79]. [ Evolution of asymmetric distribution functions.]{} As a result, the evolution of the asymmetric distribution functions ${\cal F}_{\zeta}^a(X)$ proceeds in the following way. Due to the GLAP-type evolution, the momenta of the partons decrease and distributions become peaked in the regions of smaller and smaller $X$. However, when the parton momentum degrades to values smaller than the momentum transfer $r = \zeta p$, the further evolution is like that for a distribution amplitude: it tends to make the distribution symmetric with respect to the central point $X= \zeta/2$ of the $(0, \zeta)$ segment. [Conclusions.]{} DVCS opens a new class of scaling phenomena characterized by absolutely new nonperturbative functions describing the structure of the proton. The continuous electron beam accelerators like TJNAF and ELFE may be an ideal place to study DVCS [@afanas]. The asymmetric distributions can also be studied in the processes of large-$Q^2$ meson electroproduction [@gluon], $etc.$ [Acknowledgement.]{} This work was supported by the US Department of Energy under contract DE-AC05-84ER40150. [99]{} X. Ji, preprint MIT-CTP-2517, Cambridge (1996); hep-ph/9603249. A.V. Radyushkin, Phys.Lett. [B380]{} (1996) 417. A.V. Radyushkin, CEBAF-TH-96-06, May 1996; hep-ph/9605431. V.A. Nesterenko and A.V. Radyushkin, Phys. Lett. [128B]{} (1983) 439. V.M. Belyaev and A.V. Radyushkin, Phys.Rev. [D53]{} (1996) 6509. P. Stoler, Phys. Rev. Lett. [66]{} (1991) 1003; Phys. Rev. [D44]{} (1991) 73. C. Keppel, Ph.D. Thesis, The American University (1995). V.D. Burkert and L. Elouadrhiri, Phys. Rev. Lett. [75]{} (1995) 3614. C. E. Carlson, [Phys.Rev.]{} [D34]{} (1986) 2704. V.A. Nesterenko and A.V. Radyushkin, Phys. Lett. [B115]{} (1982) 410. B. Chibisov and A. Zhitnitsky, [Phys.Rev. ]{} [D52]{} (1995) 5273. J. Bolz and P.Kroll, preprint WUB-95-35, hep-ph/9603289. V.L.Chernyak and A.R.Zhitnitsky, [Nucl. Phys.]{} [201]{} (1982) 492. V.L.Chernyak and A.R.Zhitnitsky, [Phys.Reports]{} [112]{} (1984) 173. G.P.Lepage and S.J.Brodsky, [Phys.Rev. ]{} [D22]{} (1980) 2157. A.V.Efremov and A.V.Radyushkin, [Phys.Lett. ]{} [94B ]{} (1980) 245. S.J.Brodsky and G.P.Lepage, [Phys.Lett.]{} [87B]{} (1979) 359. S.L.Adler, [Phys.Rev.]{} [177]{}, 2426 (1969);\ J.S.Bell, R.Jackiw, [Nuovo Cim.]{} [A60]{}, 47 (1967). S. J. Brodsky and G.P. Lepage [Phys.Rev. ]{} [D24]{} (1981) 1808. H.Ito, W.W.Buck and F.Gross, [Phys.Lett.]{} [B287]{} (1992) 23. CELLO collaboration, H.-J.Behrend et al., [Z. Phys. ]{} [C 49]{} (1991)401. CLEO collaboration, V.Savinov, hep-ex/9507005 (1995). E. Braaten, [Phys. Rev.]{} [D28 ]{} (1983) 524. M.A.Shifman,A.I.Vainshtein and V.I.Zakharov, [Nucl.Phys.]{} [B147]{} (1979) 385,448. S.V.Mikhailov and A.V.Radyushkin, [Phys.Rev.]{} [D45]{} (1992) 1754. A.V. Radyushkin and R. Ruskov, [Phys.Lett.]{} [B374]{} (1996) 173;\ CEBAF-TH-95-18-REV, March 1996, hep-ph/9603408. C. Coriano, A.V. Radyushkin and G. Sterman, [Nucl. Phys.]{} [405]{} (1993) 481; C. Coriano and H.-N. Li, [Nucl. Phys.]{} [434]{} (1995) 535. V.N. Gribov and L.N. Lipatov, [Sov. J. Nucl. Phys. ]{} [15]{} (1972) 78;\ L.N. Lipatov, [Sov. J. Nucl. Phys. ]{} [20]{} (1975) 94. G. Altarelli and G. Parisi, [Nucl. Phys. ]{} [B126]{} (1977) 298. Yu. L. Dokshitser, [JETP ]{} [46 ]{} (1977) 641. A. Afanasev, JLAB-THY-96-01; hep-ph/9608305. [^1]: Also Laboratory of Theoretical Physics, JINR, Dubna, Russian Federation [^2]: Talk at the Workshop on Virtual Compton Scattering, Clermont-Ferrand, France, June 26-29,1996. [^3]: In particular, such an interpolation agrees with the results of a constituent quark model calculation [@hiroshi]	{ "pile_set_name": "ArXiv" }
--- author: - Richard Drociuk title: On the Root Ambiguity in the Complete Solution to the Most General Fifth Degree Polynomial --- Abstract ======== Starting from the solution to Bring’s equation the root ambiguity is removed from the solution to the quintic equation. This gives the five complex roots of the quintic equation as indicated by Gauss’s Fundamental Theorem of Algebra. Introduction ============ In the previous paper\[Drociuk,1\], the solution to the quintic was given, but the root ambiguity was not correctly removed. This problem arises because Ferrari’s method for solving the quartic equation does not introduce all possible roots. Instead if one uses a Tshirnhaussen transformation as demonstrated\[Drociuk,1\], to the quartic equation, the five correct roots of the quintic equation,\ $$\label{1} x^{5}+mx^{4}+nx^{3}+px^{2}+qx+r=0$$ are selcted for arbitrary coefficients $m$, $n$, $p$, $q$, and $r$. The roots are selected from all possible roots using conditional loops to an arbitrary precission, $\epsilon$. Ambiguity in the Quartic Solution ================================= The quartic equation, $$\label{2} x^{4}+a_{3}x^{3}+a_{2}x^{2}+a_{1}x+a_{0}=0$$ is transformed with, $$\label{3} Tsh2 = x^{3}+b_{2}x^{2}+b_{1}x+b_{0}+y_{n}$$ to the quadratic in $y_{n}^{2}$, $$\label{4} y_{n}^{4}+B_{2}y_{n}^{2}+B_{0}=0$$ whose four roots are, $$\label{5} y_{1}= {\frac{1}{2}}(-2B_{2}+2(B_{2}^{2}-4B_{0})^{\frac{1}{2}})^{\frac{1}{2}}$$ $$\label{6} y_{2}= -{\frac{1}{2}}(-2B_{2}+2(B_{2}^{2}-4B_{0})^{\frac{1}{2}})^{\frac{1}{2}}$$ $$\label{7} y_{3}= {\frac{1}{2}}(-2B_{2}-2(B_{2}^{2}-4B_{0})^{\frac{1}{2}})^{\frac{1}{2}}$$ $$\label{8} y_{4}= -{\frac{1}{2}}(-2B_{2}-2(B_{2}^{2}-4B_{0})^{\frac{1}{2}})^{\frac{1}{2}}$$ where,\ \ $B0 = 3b_{0}^{2}a_{3}^{2}a_{0}+b_{0}^{3}b_{2}a_{3}^{2}+b_{0}^{2}b_{2}^{2}a_{2}^{2}-2b_{0}^{3}b_{2}a_{2}+3a_{3}a_{2}b_{0}^{3}+b_{0}b_{2}^{3}a_{1}^{2}-b_{0}^{3}b_{1}a_{3} +4b_{0}b_{2}a_{0}^{2}+4b_{0}^{2}b_{1}a_{0}+b_{0}^{2}b_{1}^{2}a_{2}-2b_{0}^{2}b_{1}a_{2}^{2}-b_{2}^{3}a_{0}^{2}a_{3}-2a_{0}b_{1}^{3}a_{2}+a_{3}^{2}a_{0}b_{1}^{3} -b_{2}a_{0}^{2}a_{1}-2b_{1}a_{0}^{2}a_{2}+b_{1}a_{0}a_{1}^{2}+a_{0}b_{1}^{2}a_{2}^{2}+a_{1}b_{0}^{2}b_{2}a_{2}-3b_{0}a_{3}a_{0}^{2}+2a_{1}b_{0}^{2}b_{2}a_{3}^{2} -3a_{1}a_{3}a_{2}b_{0}^{2}+b_{0}^{2}a_{3}^{2}a_{2}b_{1}-5b_{0}^{2}b_{2}a_{3}a_{0}-b_{0}^{2}a_{3}a_{2}^{2}b_{2}+3b_{0}^{2}b_{1}b_{2}a_{1} -2b_{0}^{2}b_{2}^{2}a_{3}a_{1}-2b_{0}b_{2}^{3}a_{0}a_{2}-4b_{0}b_{2}a_{0}b_{1}^{2}-2b_{0}b_{2}a_{0}a_{2}^{2}+2b_{0}a_{1}b_{1}^{2}a_{2}+b_{0}b_{2}^{2}a_{0}a_{1} -b_{0}a_{3}^{2}a_{1}b_{1}^{2}+2b_{0}a_{3}a_{1}^{2}b_{1}-b_{0}b_{2}^{2}a_{3}a_{1}^{2}+b_{0}b_{2}a_{2}a_{1}^{2}-b_{0}a_{1}b_{1}a_{2}^{2}+3b_{0}a_{1}a_{2}a_{0} -3b_{0}b_{1}b_{2}a_{1}^{2}-5b_{0}a_{1}b_{1}a_{0}+3b_{1}b_{2}a_{3}a_{0}^{2}+a_{0}b_{0}b_{1}^{2}a_{3}+a_{0}b_{1}a_{3}a_{2}b_{0}-a_{0}b_{1}b_{2}a_{2}a_{1} -a_{0}b_{2}a_{2}b_{1}^2a_{3}+2b_{1}^{2}a_{0}^{2}+b_{2}^{4}a_{0}^{2}+b_{1}^{4}a_{0}-b_{0}b_{1}^{3}a_{1}+a_{0}^{2}b_{2}^{2}a_{2}-4b_{2}^{2}a_{0}^{2}b_{1} -3a_{1}b_{0}^{3}+3b_{0}^{2}a_{1}^{2}-b_{0}a_{1}^{3}-3b_{0}^{2}a_{2}a_{0}+b_{0}^{4}+2b_{2}^{2}a_{0}b_{0}^{2}-b_{0}^{3}a_{3}^{3}+b_{0}^{2}a_{2}^{3}-a_{1}b_{0}^{2}b_{1}a_{3} +a_{0}^{3}-b_{0}^{2}b_{2}a_{2}b_{1}a_{3}-b_{0}a_{1}b_{2}a_{3}a_{0}+b_{0}b_{2}a_{1}b_{1}^{2}a_{3}+2b_{0}a_{0}b_{2}^{2}a_{3}a_{2}-3b_{0}a_{0}b_{1}b_{2}a_{3}^{2} +4b_{0}b_{2}a_{0}b_{1}a_{2}+3b_{0}b_{2}^{2}a_{0}b_{1}a_{3}-b_{0}b_{1}a_{1}b_{2}^{2}a_{2}+b_{0}a_{1}b_{2}a_{2}b_{1}a_{3}+3b_{1}^{2}a_{0}b_{2}a_{1} -2b_{1}^{2}a_{0}a_{1}a_{3}-b_{1}a_{0}b_{2}^{3}a_{1}+b_{1}^{2}a_{0}b_{2}^{2}a_{2}+b_{2}^{2}a_{0}b_{1}a_{3}a_{1}-b_{2}a_{0}b_{1}^{3}a_{3}$ $$\label{9}$$ and\ \ $B2 = a_{3}^{2}a_{2}b_{1}+2a_{1}b_{2}a_{3}^{2}-6b_{0}b_{2}a_{2}-5b_{2}a_{3}a_{0}+4b_{1}a_{0}-3a_{2}a_{0}+3a_{3}^{2}a_{0}-3a_{3}a_{2}a_{1}-3b_{0}a_{3}^{3}+6b_{0}^{2}-9b_{0}a_{1}+b_{2}^{2}a_{2}^{2}+3a_{1}^{2}-2b_{1}a_{2}^{2}+b_{1}^{2}a_{2}-b_{2}a_{2}b_{1}a_{3}-a_{3}a_{2}^{2}b_{2}+a_{2}^{3}+3b_{1}b_{2}a_{1}-a_{1}b_{1}a_{3}-3b_{0}b_{1}a_{3}+3b_{0}b_{2}a_{3}^{2}+2a_{0}b_{2}^{2}+9a_{3}a_{2}b_{0}+b_{2}a_{2}a_{1}-2b_{2}^{2}a_{3}a_{1}$ $$\label{10}$$ The coefficients of the Tshirnhaussen transformation (\[3\]) are,\ $b_{0} = \frac{(36b_{01}b_{02}b_{03}-108b_{00}b_{03}^{2}-8b_{02}^{3}+12(3)^{\frac{1}{2}}(4b_{01}^{3}b_{03}-b_{01}^{2}b_{02}^{2}-18b_{01}b_{02}b_{03}b_{00}+27b_{00}^{2}b_{03}^{2}+4b_{00}b_{02}^{3})^{\frac{1}{2}}b_{03})^{\frac{1}{3}}}{6b_{03}}$\ $-{\frac{2}{3}}{\frac{(3b_{01}b_{03}-b_{02}^{2})}{(b_{03}(36b_{01}b_{02}b_{03}-108b_{00}b_{03}^{2}-8b_{02}^{3}+12(3)^{\frac{1}{2}}(4b_{01}^{3}b_{03}-b_{01}^{2}b_{02}^{2}-18b_{01}b_{02}b_{03}b_{00}+27b_{00}^{2}b_{03}^{2}+4b_{00}b_{02}^{3})^{\frac{1}{2}}b_{03})^{\frac{1}{3}}-{\frac{1}{3}}{\frac{b_{02}}{b_{03}}})}} $ $$\label{11}$$ choose,\ $$\label{12} b_{1} = 0$$ and\ $$\label{13} b_{2} = {\frac{(a_{3}^{3}+3a_{1}-4b_{0}+b_{1}a_{3}-3a_{3}a_{2})}{(a_{3}^{2}-2a_{2})}}$$ the coefficients $b_{00}$, $b_{01}$, $b_{02}$ and $b_{03}$ are given by\ \ $b_{00} = (a_{0}^{2}a_{3}^{7}+20a_{2}^{3}a_{1}^{3}+2a_{3}^{6}a_{1}^{3}+18a_{3}^{3}a_{1}^{4}-36a_{1}a_{2}^{3}a_{0}a_{3}^{2}+150a_{3}a_{1}^{2}a_{2}^{2}a_{0}+29a_{1}a_{2}^{2}a_{3}^{4}a_{0}-54a_{3}^{3}a_{1}^{2}a_{0}a_{2}-4a_{3}^{6}a_{0}a_{1}a_{2}-48a_{3}^{2}a_{0}^{2}a_{1}a_{2}+27a_{1}^{5}-a_{3}^{5}a_{2}^{2}a_{1}^{2}+28a_{3}^{3}a_{0}^{2}a_{2}^{2}-24a_{2}^{3}a_{0}^{2}a_{3}+48a_{2}^{2}a_{0}^{2}a_{1}+24a_{2}^{5}a_{0}a_{3}+12a_{3}^{4}a_{0}^{2}a_{1}-10a_{3}^{5}a_{0}^{2}a_{2}-14a_{3}^{3}a_{0}a_{2}^{4}+2a_{3}^{5}a_{0}a_{2}^{3}-12a_{3}a_{2}^{4}a_{1}^{2}+7a_{3}^{3}a_{2}^{3}a_{1}^{2}-48a_{1}a_{2}^{4}a_{0}+3a_{3}^{5}a_{1}^{2}a_{0}+21a_{2}^{2}a_{1}^{3}a_{3}^{2}-15a_{2}a_{1}^{3}a_{3}^{4}-72a_{1}^{3}a_{2}a_{0}+9a_{1}^{3}a_{3}^{2}a_{0}-63a_{3}a_{1}^{4}a_{2})/(a_{3}^{2}-2a_{2})^{3}$ $$\label{14}$$ $b_{01} = (-10a_{3}^{6}a_{1}^{2}-84a_{3}^{3}a_{1}^{3}+80a_{0}a_{2}^{4}-76a_{2}^{3}a_{1}^{2}-64a_{0}^{2}a_{2}^{2}-2a_{3}^{4}a_{2}^{4}+12a_{3}^{2}a_{2}^{5}-16a_{3}^{4}a_{0}^{2}-108a_{1}^{4}-16a_{2}^{6}+120a_{3}^{3}a_{1}a_{2}a_{0}-344a_{3}a_{1}a_{2}^{2}a_{0}+16a_{2}^{3}a_{3}^{2}a_{0}-10a_{3}^{5}a_{1}a_{0}-24a_{2}^{2}a_{3}^{4}a_{0}-74a_{2}^{2}a_{1}^{2}a_{3}^{2}+70a_{2}a_{1}^{2}a_{3}^{4}+72a_{3}a_{2}^{4}a_{1}-52a_{3}^{3}a_{2}^{3}a_{1}+8a_{3}^{5}a_{2}^{2}a_{1}+4a_{3}^{6}a_{0}a_{2}+64a_{3}^{2}a_{0}^{2}a_{2}240a_{3}a_{1}^{3}a_{2}+12a_{3}^{2}a_{0}a_{1}^{2}+192a_{0}a_{2}a_{1}^{2})/(a_{3}^{2}-2a_{2})^{3}$ $$\label{15}$$ $b_{02} = (-64a_{3}^{3}a_{2}a_{0}+8a_{3}^{5}a_{0}+16a_{1}a_{3}^{6}-104a_{1}a_{3}^{4}a_{2}+112a_{3}^{2}a_{1}a_{2}^{2}-64a_{3}a_{2}^{4}+128a_{1}^{2}a_{3}^{3}-8a_{3}^{5}a_{2}^{2}-304a_{3}a_{2}a_{1}^{2}+144a_{1}^{3}+64a_{1}a_{2}^{3}+48a_{3}^{3}a_{2}^{3}+192a_{0}a_{3}a_{2}^{2}-80a_{1}a_{3}^{2}a_{0}-128a_{1}a_{2}a_{0})/(a_{3}^{2}-2a_{2})^{3}$ $$\label{16}$$ $b_{03} = (-64a_{3}^{2}a_{2}^{2}+48a_{3}^{4}a_{2}-8a_{3}^{6}+64a_{3}^{2}a_{0}-64a_{1}^{2}+128a_{3}a_{2}a_{1}-64a_{1}a_{3}^{3})/(a_{3}^{2}-2a_{2})^{3}$ $$\label{17}$$ Now the cubic(\[3\])is solved for it’s three roots,\ \ $x_{1n} = \frac{\Delta_{n}-(2b_{1}-\frac{2}{3}b_{2}^{2})}{6\Delta_{n}-\frac{1}{3}b_{2}}$ $$\label{18}$$ $x_{2n} = -\frac{1}{12}\Delta_{n}+\frac{b_{1}-\frac{1}{3}b_{2}^{2}}{\Delta_{n}}-\frac{1}{3}b_{2}+i\sqrt{3}(\frac{1}{12}\Delta_{n}+\frac{b_{1}-\frac{1}{3}b_{2}^{2}}{\Delta_{n}})$ $$\label{19}$$ $x_{3n} = -\frac{1}{12}\Delta_{n}+\frac{b_{1}-\frac{1}{3}b_{2}^{2}}{\Delta_{n}}-\frac{1}{3}b_{2}-i\sqrt{3}(\frac{1}{12}\Delta_{n}+\frac{b_{1}-\frac{1}{3}b_{2}^{2}}{\Delta_{n}})$ $$\label{20}$$ $\Delta_{n} =(36b_{1}b_{2}-108y_{1}-108b_{0}-8b_{2}^{3}+12(12b_{1}^{3}-3b_{1}^{2}b_{2}^{2}-54b_{1}b_{2}y_{n}-54b_{1}b_{2}b_{0}+81y_{n}^{2}+162y_{n}b_{0}+12y_{n}b_{2}^{3}+81b_{0}^{2}+12b_{0}b_{2}^{3})^{\frac{1}{2}})^{\frac{1}{3}}$ $$\label{21}$$ By substituting the values of $y_{n}$, from equations (\[5\]) to (\[8\]), into equations (\[18\]) to (\[20\]), one can generate twelve possible roots, $x_{mn}$ with $m \in (1,2,3)$ and $n \in (1,2,3,4)$. Only four roots satisfy equation (\[2\]), they are found by direct substitution. So for all complex coeffients $a_{0}$, $a_{1}$, $a_{2}$ and $a_{3}$ in equation (\[2\]), the four roots of the quartic are contained in $x_{mn}$. Ambiguity in the Quintic Solution ================================= Recall in \[Drociuk,1\], the coefficients of the quartic Tshirnhausen Transformation, $$\label{22} Tsh1 = x^{4}+dx^{3}+cx^{2}+bx+a+y$$ and let the coefficients of equation (\[2\]) be, $$\label{23} a_{0} = a+y$$ $$\label{24} a_{1} = b$$ $$\label{25} a_{2} = c$$ $$\label{26} a_{3} = d$$ and calculate the twelve $x_{mn}$ using the equations of the previous section. Then using nested “for loops” and “if-then” statements on Maple, $x_{mn}$ is substituted into both the quintic (\[1\]) and the quartic (\[22\]). When both these equations are made less than $\epsilon$, we have obtained the first root of the quintc equation (\[1\]), let it be $r_{1}$. $\epsilon$ can be made arbitrarily close to zero. Let $\epsilon = 10^{-100}$, you may chose it to be zero if you want to wait or have a faster computer. Now $r_{1}$ is then factored out of the quintic (\[1\]), leaving a quartic equation to be solved with the following coefficients, $$\label{27} a_{0} = q+r_{1}p+r_{1}^{2}n+mr_{3}+r_{1}^{4}$$ $$\label{28} a_{1} = p+r_{1}n+r_{1}^{2}m+r_{1}^{3}$$ $$\label{29} a_{2} = n+mr_{1}+r_{1}^{2}$$ $$\label{30} a_{3} = m+r_{1}$$ Equations (\[27\]) to (\[30\]) are then substituted into the solution of the general quartic of the previous section, which gives another twelve roots, $x_{mn}$. This time four of these roots satisfy both the quartic (\[2\]) with coefficients (\[27\]) to (\[30\]) and the quintic (\[1\]), let them be $r_{2}$,$ r_{3}$, $r_{4}$ and $r_{5}$. They are determined in the same way as $r_{1}$, using nested “for loops” and “if-then” statements on Maple. The the final array has five non-zero elements, they are the five complex roots of the quintic equation,(\[1\]). Conclusion ========== The root ambiguity is removed from the solution to the general fifth and fourthdegree polynomials. Reference ========= 1\) Drociuk, Richard, “On the Complete Solution to the Most General Fifth Degree Polynomial”, GM/0005026v1 3 May 2000.	{ "pile_set_name": "ArXiv" }
--- abstract: 'The duality symmetries of WZW and coset models are discussed. The exact underlying symmetry responsible for semiclassical duality is identified with the symmetry under affine Weyl transformations. This identification unifies the treatement of duality symmetries and shows that in the compact and unitary case they are exact symmetries of string theory to all orders in $\alpha''$ and in the string coupling constant. Non-compact WZW models and cosets are also discussed. A toy model is analyzed suggesting that duality will not generically be a symmetry.' --- ‘=11 \#1 =by60 = \#1[[bsphack@filesw [ gtempa[auxout[ ]{}]{}]{}gtempa @nobreak esphack]{} eqnlabel[\#1]{}]{} eqnlabel vacuum \#1 \#1[@underline\#1 $\@@underline{\hbox{#1}}$]{} ‘@=12 \#1\#2\#3[ [Phys. Rev.]{} [\#1]{}, \#2(\#3)]{} \#1\#2\#3[ [Phys. Rev. Lett.]{} [\#1]{}, \#2(\#3)]{} \#1\#2\#3[ [Nucl. Phys.]{} [\#1]{}, \#2(\#3)]{} \#1\#2\#3[ [Comm. Math. Phys.]{} [\#1]{}, \#2(\#3)]{} \#1\#2\#3[ [Phys. Lett.]{} [\#1]{}, \#2(\#3)]{} \#1\#2\#3[ [Ann. Phys.]{} [\#1]{}, \#2(\#3)]{} publist\[\#1\] @publist plus 1000pt minus 1000pt \#1 \#1[= to]{} \#1[$^{#1)}$]{} u LPTENS-92-29\ November 1992\ hep-th/9211081\ [Exact Duality Symmetries in CFT and String Theory]{}[^1] .8in Ĕlias Kiritsis[^2][^3]\ .1in [Laboratoire de Physique Théorique\ de l’Ecole Normale Supérieure\ 24 rue Lhomond\ Paris, CEDEX 05, F-75231, FRANCE]{} .8in Introduction, Results and Conclusions ===================================== Strings, being extended objects, sense the target space, into which they are embended, in a different way than point particles. The difference comes because, strings, embended in a compact space, except from their local excitations, that mimic point particle behaviour (“momentum" modes), have “winding" excitations where the string wraps around non-contractible cycles of the manifold. The masses of momentum modes are inversely proportional to the volume of the manifold, whereas those of the winding modes are proportional to the volume, since it costs energy in order to stretch the string. In the simplest possible example, that of a string moving on a circle, it was observed that the spectrum of the theory with radius $R$ and that with radius $1/R$ are identical, [@dual]. This duality symmetry is the same as the electric-magnetic duality symmetry of the underlying 2-d gausian model. Such duality symetries persist in all flat compact backgrounds, [@grv] and imply the existence of discrete symmetries for the effective theory of string theory around such backgrounds. These discrete symmetries are local, in the sense that they can be considerent as remnants of broken gauge symmetries, present at special points in the space of such flat backgrounds, [@gau]. The existence of such symmetries poses important questions about the background interpretation of such string ground states (CFTs). Obviously, the string senses the geometry of the target space in a rather “confusing" way. For example, when the string moves on a circle of radius R, just looking at the scattering data, we cannot tell if the radius is $R$ or $1/R$. When $R$ is large or small, then the distinction of the momentum and winding modes makes sense (although which is which depends on whether $R$ is large or small). For $R\sim {\cal O}(1)$ however, such a distinction does not make sense any more. To be more specific, we will discuss here the $R\rightarrow 1/R$ duality of a free scalar field in order to set the notation and to derive the formula that will be of use for all semiclassical $\s$-model duality symmetries, [@k]. Consider a scalar field $\phi$ taking values on a circle of radius $R$. We will use the convention that the $R$ dependence is explicit and $\phi\in [0,2\pi)$. Let’s consider the partition function in the presence of an external current $J_{\mu}$ $$Z_{R}(J)=\int_{0}^{2\pi}[RD\phi]exp\left[-{R^{2}\over 4\pi}\int\p_{\mu} \phi\p^{\mu}\phi+\int\p_{\mu}\phi J^{\mu}\right].\eqno(1.1)$$ In order to perform the duality transformation, we will use an infinite dimensional version of the gaussian integration formula, $$e^{-ab^{2}}={1\over 2\sqrt{\pi a}}\int_{-\infty}^{+\infty}dxe^{-{x^{2}\over 4a}+ibx}\eqno(1.2)$$ in order to make the exponent in (1.1) linear in $\phi$. Thus, we obtain $$Z_{R}(J)=\int_{0}^{2\pi}[RD\phi]\int\left[{DB_{\mu}\over R^{2}}\right] exp\left[-{\pi\over R^{2}}\int B_{\mu}B^{\mu}+i\int B_{\mu}\left(\p^{\mu}\phi-{2\pi\over R^{2}}J^{\mu}\right)+{\pi\over R^{2}}\int J_{\mu}J^{\mu}\right].\eqno(1.3)$$ The crucial step is to go from the (dummy) vector field $B_{\mu}$ to its dual, $B_{\mu}=\varepsilon_{\mu\nu}A^{\nu}$. By integrating out $\phi$ we obtain $$Z_{R}(J)=\int\left[{2\pi\over R^{2}}DA_{\mu}\right]\delta(F(A))exp\left[-{\pi\over R^{2}}\int A_{\mu}A^{\mu}-{2\pi i\over R^{2}}\int\varepsilon^{\mu\nu}J_{\mu}A_{\nu}+{\pi\over R^{2}}\int J_{\mu}J^{\mu}\right],\eqno(1.4)$$ where $F(A)=\varepsilon^{\mu\nu}\p_{\mu}A_{\nu}$. The original theory was invariant under translations of $\phi$ by a constant. This implies that $\int F(A)=0$. We will subsequently solve the $\delta$-function constraint by $A_{\mu}=\p_{\mu}\phi/2\pi$ (the jacobian for this is 1) to finally obtain $$Z_{R}(J)=\int^{2\pi}_{0}\left[{D\phi\over R}\right]exp\left[-{1 \over 4\pi R^{2}}\int\p_{\mu}\phi\p^{\mu}\phi-{i\over R^{2}} \int\varepsilon^{\mu\nu}J_{\mu}\p_{\nu}\phi+{\pi\over R^{2}}\int J_{\mu}J^{\mu}\right].\eqno(1.5)$$ Eq. (1.5) will be enough to derive all $\s$-model duality transformations. In particular, setting $J=0$, we obtain the usual duality symmetry $Z_{R}=Z_{1/R}$.[^4] The discussion above generalizes to strings propagating on an d-dimensional torus, where there are d generating duality transformations, each for every coordinate. In order to apply (1.5) to a general $\s$-model, the presence of a Killing symmetry is needed. In the appropriate coordinates, one can write the action of such a $\s$-model as $$S={1\over 4\pi}\int \left[G_{ij}\p_{\mu}x^{i}\p^{\mu}x^{j}+ iB_{ij}\varepsilon^{\mu\nu}\p_{\mu}x^{i}\p_{\nu}x^{j}\right] ,\eqno(1.6)$$ where we assume that $G_{ij},B_{ij}$ do not depend on the coordinate $x^{0}$. In terms of $x^{0}$ the action (1.6) has the same form as in (1.1) (we will assume here that $G_{00}$ is a constant although this is not necessary[^5]). The identifications are $R^{2}\rightarrow G_{00}$ and $$J_{\mu}\rightarrow -{1\over 2\pi}\left(G_{0i}\p_{\mu}x^{i}+B_{0i} \varepsilon_{\mu\nu}\p^{\nu}x^{i}\right).\eqno(1.7)$$ Then application of (1.5) gives a dual action with $${\tilde G_{00}}={1\over G_{00}}\;\;,\;\;{\tilde G_{0i}}={B_{0i} \over G_{00}}\;\;,\;\;{\tilde B_{0i}}={G_{0i}\over G_{00}}\eqno(1.8a)$$ $${\tilde G_{ij}}=G_{ij}+{B_{0i}B_{0j}-G_{0i}G_{0j}\over G_{00}} \;\;,\;\;{\tilde B_{ij}}=B_{ij}+{B_{0i}G_{0j}-B_{0j}G_{0i}\over G_{00}}.\eqno(1.8b)$$ There is a change also in the measure, as in (1.5), which can be interpeted as a shift of the dilaton, (see for example [@bu; @k; @sch]). In a $\s$-model with $d$ Killing symmetries, the structure of the group of duality transformations is as follows. There are $d$ generating duality transformations $D_{i}$, corresponding to doing the transformation (1.8) in the i-th Killing direction. These transformations are commutative $$D_{i}D_{j}=D_{j}D_{i}\eqno(1.9a)$$ and each one generates a $Z_{2}$ group, $$D_{i}D_{i}=1\;\;,\;\; \forall i=1,2,\cdots,d\eqno(1.9b)$$ The transformation $\prod_{i=1}^{d}D_{i}$ inverts the generalized metric $G+B\rightarrow (G+B)^{-1}$. When the target space is a $d$-torus, the $\s$-model is described by (1.6) with $G,B$ constants. The partition function can be calculated directly via instanton sums $$Z=\left({\sqrt{Im\tau}\over \|\eta\|^{2}}\right)^{d}\sum_{{\vec m},{\vec n}}e^{-{\pi\over Im\tau}(\tau{\vec m}+{\vec n})_{i}(G+B)_{ij}({\bar \tau}{\vec m}+{\vec n})_{j}}\;\;.\eqno(1.10)$$ Modular invariance is obvious in (1.10). Let us introduce the $2d\times 2d$ matrices $$M=\left(\matrix{G^{-1}&-G^{-1}B\cr BG^{-1}&G-BG^{-1}B\cr}\right)\;\;,\;\; H=\left(\matrix{0&1\cr 1&0\cr}\right)\;.\eqno(1.11)$$ They satisfy $$M=M^{T}\;\;,\;\;M^{-1}=HMH\;\;,\;\; M^{T}HM=H\;,\eqno(1.12)$$ which imply that $M\in O(d,d,R)$. $O(d,d,R)$ transformations act implicitly on $G,B$ via $M\rightarrow \Omega M \Omega^{T}$. Upon Poisson resumming (1.10), it can be cast in character form $$Z={det(G+B)^{-d/2}\over \|\eta\|^{2d}}\sum_{{\vec m},{\vec n}}q^{\Delta_{ {\vec m},{\vec n}}}{\bar q}^{{\bar \Delta}_{{\vec m},{\vec n}}}\;.\eqno(1.13)$$ Introducing a $2d$ vector ${\vec N}\sim \left(\matrix{{\vec n}\cr {\vec m}\cr}\right)$, we can write the conformal weights as $$\Delta_{{\vec m},{\vec n}}=Q_{i}(G^{-1})_{ij}Q_{j}=N_{i}(M+H)_{ij}N_{j}\;,\eqno(1.14a)$$ $${\bar \Delta}_{{\vec m},{\vec n}}={\bar Q}_{i}(G^{-1})_{ij}{\bar Q}_{j}= N_{i}(M-H)_{ij}N_{j}\;,\eqno(1.14b)$$ $$Q_{i}=n_{i}+(G-B)_{ij}m_{j}\;\;,\;\;{\bar Q}_{i}=n_{i}-(G+B)_{ij}m_{j}\;.\eqno(1.14c)$$ The generating duality transformations can be represented as $O(d,d)$ transformations $$D_{i}\rightarrow \left(\matrix{1-e_{i}&e_{i}\cr e_{i}&1-e_{i}\cr}\right)\;,\eqno(1.15)$$ where $e_{i}$ is a $d\times d$ matrix with all elements zero except the diagonal $ii$ element being $1$. It is obvious from (1.14) that $D_{i}$ interchanges $m_{i}\leftrightarrow n_{i}$ and thus the invariance of the partition function is obvious. In the basis for the currents in which the metric is unity, this amounts to the transformation ${\bar J}^{i}\rightarrow -{\bar J}^{i}$. Provided that $J_{\mu}$ is a classical source, eq. (1.5) is exact. However, subtleties can (and do) arise when $J_{\mu}$ depends on other quantum fields, as is the case in (1.7). Of course our semiclassical considerations will still be valid, but, in general, we expect discrepancies in higher loops. One of our tasks in this paper is to investigate when semiclassical duality is exact (in the sense that it can be corrected beyond 1-loop to yield a genuine symmetry). There are two points of view relevant here. One is the $\s$-model point of view, which has the advantage that the background interpretation is manifest. The other is the CFT point of view, where, although the background interpretation is not always obvious, it has the advantage that one can get exact results easier. In this paper we will discuss all the duality symmetries of well-understood CFTs, that is WZW models [@wit] and their cosets [@c1; @c2].[^6] This class is quite large and contains (modulo a mild assumption) all the CFTs which describe string propagation in a target space with $d$ Killing symmetries. It was argued in [@gr] that any such $\s$-model can be obtained by gauging $d$ abelian currents in a WZW model. The first step is to understand duality in the WZW model. From the $\s$-model point of view, there are many semiclassical duality transformations, of the type (1.8). By analyzing their effect on the affine primaries, we will be able to identify them with Weyl transformations acting on the current algebra representations. The question of exact duality invariance then translates into invariance under the affine Weyl group. In the case of compact current algebra and unitary (integrable) representations the affine Weyl group is a genuine symmetry. This is not the case in general (where it relates inequivalent representations). We will also see explicitly that the action of the exact duality transformation on the fields is, in general, more complicated than the semiclassical duality transformation. Once we understand how duality works in the WZW model, we can proceed to the coset models. When we gauge a semisimple subgroup, then the duality symmetry of the coset theory is inherited from that of the original WZW model, and the different dual actions are obtained by gauging the different dual actions of the WZW model. The non-trivial duality transformations are those that leave the subgroup structure invariant. The generic coset $G/H$ model with H semisimple, has extra Killing symmetries, which can be used to generate duality transformations. However these transformations are included in the ones mentioned above. When $H$ is maximal, then the $\s$-model describing the $G/H$ coset has no Killing symmetries. However, according to our previous discussion it still posseses duality symmetries. More interesting things happen when H is abelian. In this case, we have the option to gauge an axial or a vector abelian current. Semiclassically, it can be shown that, these two theories are dual to each other, [@k; @dvv]. We will see that the original affine Weyl symmetry of the WZW model guarantees that this extra axial-vector duality is an exact symmetry (although in the $\s$-model language it needs corrections beyond one-loop ). This type of duality is a generalization of the order-disorder (Kramers-Wannier) duality of the critical Ising model. Using axial-vector duality, one can generate new conformal $\s$-models using $O(d,d,R)$ transformations, [@ven; @sen; @js]. The $O(d,d,R)$ transformations need to be corrected beyong one-loop, however, in the compact case, this can always be done. One implication of this result is that there are marginal $J{\bar J}$ perturbations in $\s$ models with Killing symmetries. If the currents are abelian and chiral this already known. However marginality persists for some combinations of non-chiral abelian currents. The presence of duality symmetries in compact targets complicates the background interpretation of the $\s$ model. When $\s$-model couplings are strong, it is difficult to have a geometric notion of a target manifold (even the notion of dimensionality can break down, and many such instances are known, for example $SU(2)_{k=1}\sim U(1)_{R=1}$ etc.). The only case where one has an (almost) unabiguous notion of a manifold is when all couplings are weak, ($\a '\rightarrow 0$). In curved backgrounds, the dual versions obtained for example by (1.8) are not trustworthy guides of geometry since the dual background describes strong coupling regions. When one considers string propagation in non-compact backgrounds, the situation is quite different. For Euclidean non-compact cosets, generically, duality is not expected to be a symmetry, since the underlying affine Weyl group relates, in general, inequivalent representations. If one considers a model where the spectrum can be classified into complete orbits of the affine Weyl group, then duality will be restored.[^7] There are two potential problems with this procedure. The first is that the required orbits contain representations that are not positive. However, this might not be lethal for the associated string model, but positivity of the string Hilbert space needs to be addressed. [^8] The second is that the background interpretation of such theories is obscure. In order to investigate whether the semiclassical duality is exact in the non-compact case we will analyse the simplest possible model, where axial-vector duality relates the 2-d Euclidean plane (free field theory), to a certain singular manifold. Although we cannot compute the latter partition function exactly, we will compute it in the “minisuperspace” approximation where it will turn out to be different than that of the plane. We will show, however, that at weak coupling the two coincide. Although, this computation does not settle the issue of exactness of non-compact duality symmetries it does give some useful indications. For non-compact cosets with Minkowskian signature, the meaning of the duality transformation is different. Instead of relating two different manifolds, it interchanges various regions of spacetime, [@giv; @dvv]. The same remarks apply here as in the Euclidean case. Duality here, although it might not be a symmetry, provides a map that can give meaning to regions of spacetime that one otherwise would traditionally neglect. The structure of the rest of this paper is as follows. In section 2 we will analyze semiclassically and exactly the duality symmetries of compact WZW models. The same will be done for compact cosets in section 3. The extension of duality symmetries to $O(d,d)$ symmetries will be discussed in section 4. Section 5 contains some remarks on marginal current-current perturbations implied by $O(d,d)$ covariance. Finally, in section 6 we will discuss non-compact cosets. Duality in the WZW model. ========================= In this section we will analyze in detail the duality symmetries of WZW model, both from the $\s$-model and the CFT (affine current algebra) point of view. We will consider for simplicity a compact group $G$ which is simple and simply laced. It will turn out that understanding the simplest such group, SU(2), will suffice. In the case of non-simply laced simple groups there are some minor changes due to the short roots that will be dealt with latter on. The case of non-simple groups has further complications that we will not consider here. The action of the WZW model is $$I(g)={k\over 4\pi}I_{NS}(g)+{ik\over 6\pi}\Gamma_{WZ}(g)\eqno(2.1)$$ $$I_{NS}(g)=\int d^{2}x Tr[U_{\mu}U^{\mu}]\;\;,\;\;\Gamma_{WZ}(g)=\int \limits_{B\atop \p B=S^{2}} d^{3}y\varepsilon^{\mu\nu\rho}Tr[U_{\mu}U_{\nu}U_{\rho}]\eqno(2.2)$$ where $$U_{\mu}=g^{-1}\p_{\mu}g\;\;,\;\;V_{\mu}=\p_{\mu}g g^{-1}\eqno(2.3)$$ $g$ is a matrix in the fundamental representation of $G$ and $Tr$ is a properly normalized trace such that $${1\over 12\pi^{2}}\int_{S^{3}}Tr[U\wedge U\wedge U]\in Z\;.\eqno(2.4)$$ The action $I(g)$ is invariant under the group $G_{R}\otimes G_{L}$, generated by left and right group transformations, $g\rightarrow h_{1}gh_{2}$, with associated conserved currents $$J^{\mu}_{R}={k\over 2\pi}P_{-}^{\mu\nu}U_{\nu}\;\;,\;\;J^{\mu}_{L}= {k\over 2\pi}P_{+}^{\mu\nu}V_{\nu}\eqno(2.5)$$ with $P_{\pm}^{\mu\nu}\equiv\delta^{\mu\nu}\pm i\varepsilon^{\mu\nu}$. These currents are conserved and chirally conserved and they generate two copies of the affine $\hat G$ current algebra. An important property of the WZW action is that it satisfies the Polyakov-Wiegman formula $$I(gh)=I(g)+I(h)-{k\over 2\pi}\int d^{2}xP^{\mu\nu}_{+}Tr[U_{\mu}(g)V_{\nu}(h)]\eqno(2.6)$$ To generate duality transformations in the WZW model, we pick a generator of the Lie algebra of $G$, $T^{0}$, normalized as $Tr[(T^{0})^{2}]=1$. We can then parametrize $g=e^{i\phi T^{0}}h$. Using (1.6), the action $I(g)$ takes the form $$I(g)=I(h)+{k\over 4\pi}\int \p_{\mu}\phi\p^{\mu}\phi -{ik\over 2\pi} \int P^{\mu\nu}_{+}\p_{\mu}\phi V^{0}_{\nu}(h)\eqno(2.7)$$ where $V^{0}_{\mu}(h)=Tr[T^{0}V_{\mu}(h)]$. We can now apply the duality map (1.5) $\rightarrow$ (2.7) to obtain[^9] $$I^{\rm dual}(g)=I(h)+{1\over 4\pi k}\int \p_{\mu}\phi\p^{\mu}\phi- {i\over 2\pi}\int P_{+}^{\mu\nu}\p_{\mu}V^{0}_{\nu}(h)\;.\eqno(2.8)$$ The angle $\phi$ was originally normalized to take values in $[0,2\pi]$. It is obvious from (2.8) that the effect of the duality transformation is to change the range of values to $[0,2\pi/k]$. To see how many independent duality transformations exist, we have to explicitly parametrize the Cartan torus dependence of the WZW model. Pick a basis in the Cartan algebra, $T^{i}$, $i=1,2,\cdots,r$, $[T^{i},T^{j}]=0$, $Tr[T^{i}T^{j}]=\delta^{ij}$ and parametrize, $$g=e^{i\sum_{i=1}^{r}\a^{i}T^{i}}\,h\,e^{i\sum_{i=1}^{r}\g^{i}T^{i}} \;\;.\eqno(2.9)$$ Then using (2.6) the WZW action becomes $$I(g)=I(h)+{k\over 4\pi}\int(\p_{\mu}\a^{i}\p^{\mu}\a^{i}+\p_{\mu}\g^{i} \p^{\mu}\g^{i})-{ik\over 2\pi}\int(P_{+}^{\mu\nu}\p_{\mu}\a^{i}V^{i}_{\nu} (h)+P_{-}^{\mu\nu}\p_{\mu}\g^{i}U_{\nu}^{i}(h))+$$ $$+{k\over 2\pi}\int P_{+}^{\mu\nu}\p_{\mu}\a^{i}\p_{\nu}\g^{j}M^{ij}(h)\eqno(2.10)$$ where $$U_{\mu}^{i}(h)=Tr[T^{i}U_{\mu}(h)]\;,\;V_{\mu}^{i}(h)=Tr[T^{i}V_{\mu}(h)] \;,\; M^{ij}(h)=Tr[T^{i}hT^{j}h^{-1}]\;.\eqno(2.11)$$ It is obvious from (2.10) that we can apply the duality transformation using any of the $\a^{i}$, $\g^{i}$. Thus, there are $2^{2r}-1$ non-trivial duality transformations. A duality transformation on $\a^{i}$ effectively makes the substitution $\a^{i}\rightarrow \a^{i}/k$ in the action whereas a duality transformation on $\g^{i}$ makes the substitution $\g^{i}\rightarrow -\g^{i}/k$. In order to identify the underlying property of the WZW model, responsible for the invariance under these duality transformations, we have delve a bit into such elements of the the representation theory of the affine Lie algebras as the affine Weyl group and external automorphisms. Here, I will just state some properties that we need. More information can be obtained in [@gw] and references therein. The affine Weyl group ${\hat W}$ is a semidirect product of the Lie algebra Weyl group $W$ times a translation group, ${\hat W}=W\triangleright T$. Appart from the action of finite Weyl group elements, there are Weyl transformations associated to roots which have a component in the direction of the imaginary simple root. The action of such an element ${\hat W}_{\vec\a}$ on a finite Lie algebra weight $\vec\lambda$ and on the grade $n$ is $${\hat W}_{\vec\a}({\vec\lambda})=W_{\vec\a}({\vec\lambda})-k{\vec\beta} \eqno(2.12a)$$ $${\hat W}_{\vec\a}(n)=n-{\vec\lambda}\cdot{\vec\beta}-{k\over 2}{\vec\beta}\cdot{\vec\beta}\eqno(2.12b)$$ where ${\vec\beta}=2{\vec\a}/{\vec\a}\cdot{\vec\a}$ is the coroot associated to the finite Lie algebra root $\vec\a$, the grade $n$ is basically the mode number[^10] and $W_{\vec\a} (\vec\lambda)={\vec\lambda}-{\vec\a}({\vec\lambda}\cdot{\vec\beta})$ is a finite Weyl transformation. It is important to note that affine Weyl transformations, in general, map states inside a representation at different levels. There are also external automorphisms of the affine algebra which are essentially associated to symmetries of the affine Dynkin diagram. For the $SU(n)$ case, the affine Dynkin diagram consists of $n$ nodes connected around a circle. The external automorphisms are generated by a basic rotation, and a reflection which corresponds to the finite Lie algebra external automorphism (that maps a representation to its complex conjugate). When we write a highest weight ${\vec\Lambda}=\sum_{i=1}^{n-1} m_{i}{\vec\Lambda}_{i}$ in terms of the fundamental weights ${\vec\Lambda}_{i}$, ($m_{i}$ are non-negative integers), the action of the generating rotation of the affine Dynkin diagram is as follows $$\sigma({\vec\Lambda})=(k-\sum_{i=1}^{n-1}m_{i}){\vec\Lambda}_{1}+ m_{1}{\vec\Lambda}_{2}+\cdots+m_{n-2}{\vec\Lambda}_{n-1}\;.\eqno(2.13)$$ $\sigma$ generates a $Z_{n}$ group[^11] where $\sigma^{n}=1$ on the heighest weights, but acts as an affine Weyl transformation in the representation. Specializing to SU(2), let $m\in Z/2$ be the weight, and $j\in Z/2$ the highest weight (spin of a representation). Then the finite Weyl group acts as $m\rightarrow -m$, and combined with the affine translation $m\rightarrow m+k$ they generate the affine Weyl group. The only nontrivial outer automorphism $\sigma$ acts as $j\rightarrow k-j$ and $\sigma^{2}$ is a Weyl translation. The non-trivial statement now is: For compact groups, integer level and integrable heighest weight representations, both the affine Weyl group and the external automorphisms are symmetries. In particular, in a WZW model the Hilbert space is constructed by tying together (in a modular invariant way) two copies of representations of the affine algebra. Thus, we have invariance under independent affine Weyl transformations acting on left or right representations. Moreover, since the modular transformation properties of the affine characters reflect the external automorphism symmetries, the theory is invariant under external automorphisms that act at the same time on left and right representations. These invariance properties can be verified for correlation functions on the sphere and the torus. This then implies that they hold on an arbitrary Riemann surface since the sphere and torus data are sufficient in order to construct the correlators at higher genus. As an example, we will present the SU(2) case and focus on the spectrum. We introduce the (affine) SU(2)$_{k}$ characters $$\chi_{l}(q=e^{2\pi i\tau}, w)=Tr_{l}\left[q^{L_{0}}e^{2\pi iwJ^{3}_{0}} \right]=\sum_{m=-k+1}^{k} c^{l}_{m}(q)\vartheta_{m,k}(q,w)\eqno(2.14)$$ where $l$ is twice the spin (a non-negative integer) and $m$ is twice the $ J^{3}_{0}$ eigenvalue. The trace is in the affine hw representation of spin $l$, $$\vartheta_{m,k}(q,w)=\sum_{n\in Z} q^{k(n+{m\over 2k})^{2}}e^{2\pi iw(kn+{m\over 2})}\eqno(2.15)$$ and $c^{l}_{m}$ are the standard string functions [@kp] which satisfy $c^{l}_{m}=0$ when $l-m=1(mod$ $2)$ (which means that the spin is increased or decreased in units of 1) . For integrable representations ($k$ is a positive integer and $0\leq l\leq k$), invariance under the affine Weyl group is equivalent to $$c^{l}_{m}=c^{l}_{-m}\;\;,\;\; c^{l}_{m}=c^{l}_{m+2k}\eqno(2.16)$$ The first relation is due to the Weyl group of $SU(2)$ while the second is the generating translation in the affine Weyl group. There is another important relation $$c^{l}_{m}=c^{k-l}_{k-m}\eqno(2.17)$$ which is a consequence of the external affine automorphism, [@kp]. The duality tranformation on $\a^{i}$ amounts to replacing ${\bar J}^{i} \rightarrow -{\bar J}^{i}$, where ${\bar J}^{i}$ is the right Cartan current in the $T^{i}$ basis of the Cartan subalgebra. Similarly the duality transformation on $\g^{i}$ amounts to the replacement $J^{i}\rightarrow -J^{i}$ at the level of the Cartan subalgebra. This is not the whole story however. With a bit more effort one can see that they act as Weyl transformations on the left or right SU(2) currents. This identification can be seen clearly by coupling the WZW action to external gauge fields and monitoring the effect of the duality transformation on the currents. It can also be recovered from the twisted partition function via the action of the duality transformation on the gauge field moduli (for the Cartan). Let us now check that the duality transformations $$D_{i}\;\;:\;\; J^{i}\rightarrow -J^{i}\eqno(2.18a)$$ $${\bar D}_{i}\;\;:\;\;{\bar J}^{i}\rightarrow -{\bar J}^{i}\eqno(2.18b)$$ are exact symmetries of the model. We will consider again for simplicity SU(2) and then generalize to an arbitrary group. The partition function is $$Z(q,{\bar q})=\sum_{l,{\bar l}=0}^{k}N^{l,{\bar l}}\chi_{l}(q,w=0){\bar \chi}_{\bar l}({\bar q},{\bar w}=0)\eqno(2.19)$$ where $N^{l,{\bar l}}$ is one of the CIZ modular invariants. The diagonal one $N^{l,{\bar l}}=\delta_{l,{\bar l}}$ corresponds to the usual WZW model. Putting everything together we obtain $$Z(q,{\bar q})=\sum_{l,{\bar l}=0}^{k}\sum_{m=-k+1}^{k} \sum_{\mb =-k+1}^{k}N^{l,{\bar l}}c_{m}^{l}(q){\bar c}^{\bar l}_{\mb}(\qb) \cdot$$ $$\cdot \sum_{n,\nb \in Z}exp\left[2\pi i\left(\tau k(n+{m\over 2k})^{2}-{\bar \tau}k(\nb +{\mb \over 2k})^{2}\right)\right]\eqno(2.20)$$ The two generating duality transformations here correspond to $n\rightarrow -n$, $m\rightarrow -m$, and ${\bar n}\rightarrow -{\bar n}$, ${\bar m}\rightarrow -{\bar m}$. They are symmetries of (2.20) if we use the invariance of the string functions under the affine Weyl group, (2.16). This invariance is similar, but qualitatively different than that present in flat backgrounds. There, one has a family of theories parametrized by $G,B$ and duality is the statement that two theories are equivalent for different values of the parameters. Here, there is no parameter present and, in this sense, this is what we could call self-duality. This becomes more transparent if we consider the one parameter family of theories, parametrized by the radius of the cartan torus of SU(2). The partition function is known, [@sky] $$Z(R)=\sum_{l,{\bar l}=0}^{k}\sum_{m=-k+1}^{k}\sum_{r=0}^{k-1}N^{l,{\bar l}}c^{l}_{m}(q){\bar c}^{\bar l}_{m-2r}(\qb)\sum_{M,N\in Z}q^{\Delta_{M,N}}{\bar q}^{{\bar \Delta}_{M,N}}\eqno(2.21)$$ with $$\Delta_{M,N}={1\over 4k}\left({kM+m-r\over R}+R(kN+r)\right)^{2}\;\; ,\;\,{\bar \Delta}_{M,N}={1\over 4k}\left({kM+m-r\over R}-R(kN+r)\right)^{2}\eqno(2.22)$$ In (2.21) there is a duality symmetry $R\rightarrow 1/R$, which becomes self-duality at the point $R=1$, that corresponds to the WZW model. Now we are in a position to discuss the general WZW model for a simple group $G$. Let $M$ be the root lattice, $M_{L}$ the long root lattice and $M^{}$ the weight lattice. The character of a hw representation of ${\hat G}$ with hw $\vec \Lambda$ is defined as $$\chi_{\vec \Lambda}(q,{\vec w})=Tr[q^{L_{0}}e^{2\pi i{\vec w}\cdot{\vec J}_{0}}]\eqno(2.23)$$ where ${\vec J}_{0}$ generates the cartan subalgebra of $G$. The character admits the string function decomposition, [@kp] $$\chi_{\vec \Lambda}=\sum_{{\vec \lambda}\in M^{}/kM_{L}} c^{\vec \Lambda}_{\vec \lambda}(q)\Theta_{\vec \lambda}({\vec w},q)\eqno(2.24)$$ with $\Theta_{\vec\lambda}$ being the classical $\vartheta$-function of level $k$ of the Lie algebra of $G$ $$\Theta_{\vec\lambda}({\vec w},q)=\sum_{{\vec\gamma}\in M_{L}} q^{{k\over 2}\left({\vec\gamma}+{{\vec\lambda}\over k}\right)^{2}} e^{2\pi i{\vec w}\cdot(k{\vec\gamma}+{\vec\lambda})}\;.\eqno(2.25)$$ The string functions are invariant under the Weyl group and Weyl translations $$c^{\vec\Lambda}_{w({\vec\lambda})}=c^{\vec\Lambda}_{\vec\lambda}\;\;,\;\; c^{\vec\Lambda}_{{\vec\lambda}+k{\vec\beta}}=c^{\vec\Lambda}_{\vec\lambda} \eqno(2.26)$$ where $w$ is a Weyl transformation and ${\vec\beta}\in M_{L}$. The (left) generating duality transformations $D_{i}$ correspond to Weyl reflections generated by the simple roots ${\vec\a}_{i}$ which implement the transformations (2.18a). The invariance of the spectrum (and partition function) is encoded in the fact, obvious from (2.25,26), that $\chi_{\vec\Lambda}$ is invariant under $w_{i}\rightarrow -w_{i}$. Although $w_{{\vec\a}_{i}}$ do not commute, they do so when applied to the character, thus at the level of the partition function they generate a group isomorphic to (1.9). However, at the level of correlation functions the (left) duality group is larger and in fact isomorphic to the finite Weyl group of $G$, $W_{G}$. Thus the full duality group of the WZW model is $W_{G}\times W_{G}$ the first acting on the left current modules while the second acting on the right current modules.[^12] The structure of the (self)-duality group is different than the one present in flat backgrounds. Compact Cosets ============== A host of CFTs can be obtained from the coset costruction [@c1]. In Langrangian form it amounts to gauging a subgroup $H$ of $G$ in a conformally invariant way, [@c2]. We will assume $G$ to be simple, and $H$ regularly embedded[^13]. Let us first consider $H$ to be semi-simple. Then, there is one possible gauging , the vectorial one, [@k]. The gauged WZW action is $$S_{V}(g,A)=I(g)+{k\over 2\pi}\int d^{2}xTr[(J^{\mu}_{R}-J^{\mu}_{L})A_{\mu}+P_{+}^{\mu\nu}A_{\mu}gA_{\nu} g^{-1}-A_{\mu}A^{\mu}]\eqno(3.1)$$ where $A_{\mu}$ belongs to the Lie algebra of $H$. The action (3.1) is invariant under $$g\rightarrow hgh^{-1}\;\;\;,\;\;\;A_{\mu}\rightarrow h^{-1}A_{\mu}h+h^{-1}\p_{\mu}h\;.\eqno(3.2)$$ Since the gauge field is quadratic in the action (3.1) one can integrate it out, and fix a physical gauge in order to obtain a $\s$-model desription of the coset theory. There are two possible ways to generate duality transformations for the non-abelian coset theory. The first is to gauge different dual versions of the original WZW model. The group of the left duality transformations obtained this way is equivalent to the original Weyl group $W_{G}$ with the restriction that its subgroup $W_{H}$ acts trivially. The action of $W_{H}$ can be absorbed in a redefinition of the gauge fields, and in the $\s$-model form (where the gauge fields have be integrated out) is trivial. The other possibility is that the action (3.1) has Killing symmetries which can be exploited in order to generate duality transformations. The constant vector gauge transformations are symmetries of (3.1) but will not survive the passage (gauge fixing) to the $\s$-model. We can directly hunt for such Killing symmetries. The result is that whenever there exists a subgroup $H'$ of $G$, such that $[H,H']=0$, then, there are extra conserved currents which can be calculated from (3.1), $$J^{\mu}_{R}=P_{+}^{\mu\nu}g^{-1}(\p_{\nu}g+\{g,A_{\nu}\})\|_{H'}\eqno(3.3a)$$ $$J^{\mu}_{L}=P_{-}^{\mu\nu}(\p_{\nu}g-\{g,A_{\nu}\})g^{-1}\|_{H'}\eqno(3.3b)$$ where $\{,\}$ stands for anti-commutator, and $\|_{H'}$ implies a projection onto the Lie algebra of $H'$. The currents (3.3) transform covariantly under $H$ gauge transformations and they are conserved: $\p_{\mu}J^{\mu}_{R,L}=0$. A less obvious, but verifiable statement is that these currents are also chirally conserved: $\varepsilon_{\mu\nu}\p^{\mu}J^{\nu}_{L,R}=0$. Thus, they generate a $H'$ current algebra, and this implies that, locally the $G/H$ model can be factorized into a $G/(H\times H')$ model times a $H'$ WZW model. This guarantees the presence of the Killing symmetries associated to the Cartan of $H'$ and the duality transformations they imply have been discussed in the previous section. A more interesting case is when $H$ is abelian. We will assume without much loss of generality that $H=U(1)$. The situation with more $U(1)$’s will become obvious. In this case, there are two possible ways to gauge. The vector as in the non-abelian case with action given in (3.1) and the axial with action $$S_{A}(g,A)=I(g)+{k\over 2\pi}\int d^{2}xTr[(J^{\mu}_{R}+J^{\mu}_{L})A_{\mu}-P_{+}^{\mu\nu}A_{\mu}gA_{\nu} g^{-1}-A_{\mu}A^{\mu}]\;.\eqno(3.4)$$ The axial and vector actions are related by a duality transformation, [@dvv; @k]. In order to show this, we have to parametrize the group element $g$ as in (2.7) where $T^{0}$ is the generator of the U(1) subgroup. In order to write the gauged action (3.1), we need the expressions for the left and right U(1) currents $$J^{\mu}_{R}={k\over 2\pi}P_{-}^{\mu\nu}(iX(h)\p_{\nu}\phi+U^{0}_{\nu} (h))\;\;,\;\;J^{\mu}_{L}={k\over 2\pi}P_{+}^{\mu\nu}(i\p_{\nu}\phi+ V^{0}_{\nu}(h))\eqno(3.5)$$ where $X(h)=Tr[T^{0}hT^{0}h^{-1}]$, as well as (2.7) for the WZW action. Then, $$S_{V}(g,A)=I(h)+{k\over 4\pi}\int \p_{\mu}\phi\p^{\mu}\phi -{ik\over 2\pi} \int P^{\mu\nu}_{+}\p_{\mu}\phi V^{0}_{\nu}(h)+$$ $$+{ik\over 2\pi}\int A_{\mu}\left(P_{-}^{\mu\nu}(iX(h)\p_{\nu}\phi+U^{0}_{\nu}(h)) -P_{+}^{\mu\nu}(i\p_{\nu}\phi+V^{0}_{\nu}(h))\right)-{k\over 2\pi}\int (1-X(h))A_{\mu}A^{\mu}.\eqno(3.6)$$ Applying the duality transformation (1.5), we obtain $$S_{V}\rightarrow S_{V}^{\em dual}=I^{\em dual}(g)+{ik\over 2\pi}\int A_{\mu}({\tilde J}^{\mu}_{R}+{\tilde J}^{\mu}_{L})-{k\over 2\pi}\int (1+M)A_{\mu}A^{\mu}\eqno(3.7)$$ where ${\tilde J}^{\mu}_{R,L}$ are the respective currents of the dual theory $${\tilde J}^{\mu}_{R}={k\over 2\pi}P_{-}^{\mu\nu}\left({i\over k}X(h) \p_{\nu}\phi+U^{0}_{\nu}(h)\right)\;\;,\;\;{\tilde J}^{\mu}_{L}={k\over 2\pi}P_{+}^{\mu\nu}\left({i\over k}\p_{\nu}\phi+V^{0}_{\nu}(h)\right) \eqno(3.8)$$ Inspection of (3.7) shows that it is the axially gauged dual WZW model action. Of course, this is not unexpected, since, at the naive level, the vector coset is the WZW model with the constraint $J_{L}-J_{R}=0$. As we have seen, a duality transformation of the WZW model changes the sign of one of the currents, and this gives the axial constraint $J_{L}+J_{R}=0$. In order to investigate to what extend this semiclassical axial-vector duality is exact, we will analyze the partition function. In particular we will need a method to compute exactly the partition function for both the axial and the vector gauge theory. The easiest way is the operator method.[^14] We will start by considering the $SU(2)_{k}/U(1)$ coset which captures the relevant effects. Once we understand it, the generalization will be simple. Since we are concerned with the partition function, we will be working on the torus, in the standard flat metric. It is well known, [@c2] that, in the gauge $\p_{\mu}A^{\mu}=0$, the gauged WZW action factorizes (up to gauge field moduli) to that of the original WZW plus the quadratic action for the gauge field (and the FP determinant, det$'$). The effect of the gauge field moduli is to introduce twisted boundary conditions for the field $g$ of the WZW model around the two non-contactible cycles of the torus. The strategy will be to compute the WZW partition function in the presence of the gauge-field moduli (twists), then integrate over them as specified by the gauge field measure, and then add the contribution of the (decoupled) local part of the gauge field. Consider first the twist in the “space" direction, (vector gauging is considered here). Its effect is to impose a boundary condition on $g$ which is a global $U(1)_{V}$ transformation , $$g(\sigma +1)=e^{i\pi\alpha\s_{3}}g(\s)e^{-i\pi\alpha\s_{3}}\;.\eqno(3.9)$$ The left and right currents are defined in the standard fashion $$J={k\over 2\pi}g^{-1}\p g\,\,\,,\,\,\,{\bar J}={k\over 2\pi}{\bar \p}g g^{-1}\;.\eqno(3.10)$$ If we introduce cylinder coordinates $$z=e^{2\pi(t+i\s)}\;\;,\;\;{\bar z}=e^{2\pi(t-i\s)}\eqno(3.11)$$ (3.9) amounts to $$J^{\pm}(ze^{2\pi i})=e^{\pm 2\pi i\a}J^{\pm}(z)\;\Rightarrow \; J^{\pm}(z)=\sum _{m\in Z\mp \a} {J^{\pm}_{m}\over z^{m+1}}\eqno(3.12a)$$ $${\bar J^{\pm}}({\bar z}e^{-2\pi i})=e^{\pm 2\pi i\a}{\bar J}^{\pm}({\bar z}) \;\Rightarrow \; {\bar J}^{\pm}({\bar z})=\sum _{m\in Z\mp \a} {{\bar J}^{\pm}_{m} \over {\bar z}^{m+1}}\eqno(3.12b)$$ while it leaves $J^{3},{\bar J}^{3}$ almost invariant. In fact, due to the central term in the current algebra, both $J^{3},{\bar J}^{3}$ are shifted by the same constant, to be determined below. The vector current $J^{3}-{\bar J}^{3}$ is invariant, as it should be. The twisted currents satisfy an algebra that is isomorphic to the untwisted SU(2) current algebra. In particular, the Cartan currents are shifted, $$J_{m}^{3}(\a)=J^{3}_{m}+{k\a\over 2}\delta_{m,0}\eqno(3.13)$$ and similarly for ${\bar J}^{3}$. Then, $$[J^{+}_{m-\a},J^{-}_{n+a}]={k\over 2}m\delta_{m+n,0}+J^{3}_{m+n}(\a)\eqno(3.14a)$$ $$[J^{3}_{m}(\a),J^{\pm}_{n\mp\a}]=\pm J^{\pm}_{m+n\mp\a}\eqno(3.14b)$$ $$[J^{3}_{m}(\a),J^{3}_{n}(\a)]={k\over 2}m\delta_{m+n,0}\eqno(3.14c)$$ and similarly for the left sector. The Virasoro operator also get shifted. This is standard, we can see it by either doing the Sugawara construction using the twisted currents or checking that the following expression has the proper commutation relations $$L_{m}(\a)=L_{m}+\a J^{3}_{m}+{\a^{2}k\over 4}\delta_{m,0}\;.\eqno(3.15)$$ Now we need to twist in the “time" direction. This is achieved in the standard way by inserting a factor $$e^{2\pi i\b(J^{3}_{0}(\a)-{\bar J}^{3}_{0}(\a))}\eqno(3.16)$$ which will eventually project onto invariant states (this is similar to the orbifold case). Thus, collecting everything together, we obtain the (vectorially) twisted WZW partition function $$Z_{\a}^{\b}(q,{\bar q};V)=Tr_{H}\left[ q^{L_{0}(\a)}{\bar q}^{{\bar L}_{0} (\a)}e^{2\pi i\b(J^{3}_{0}(\a)-{\bar J}^{3}_{0}(\a))}\right]\eqno(3.17)$$ where $H$ is the Hilbert space of the WZW theory. Using (2.14,15,19) and (3.13,15) we can explicitly evaluate (3.17), $$Z_{\a}^{\b}(q,{\bar q})=\sum_{l,{\bar l}=0}^{k}\sum_{m=-k+1}^{k} \sum_{\mb =-k+1}^{k}N^{l,{\bar l}}c_{m}^{l}(q){\bar c}^{\bar l}_{\mb}(\qb) \cdot$$ $$\cdot \sum_{n,\nb \in Z}exp\left[2\pi i\left(\tau k(n+{m\over 2k}+ {\a\over 2})^{2}-{\bar \tau}k(\nb +{\mb \over 2k}+{\a\over 2})^{2}+ \b(k(n-\nb)+{m-\mb\over 2})\right)\right].\eqno(3.18)$$ The twisted partition function satisfies $$Z_{\alpha}^{\beta}=Z^{\beta}_{\alpha+1}=Z^{\beta+1}_{\alpha}= Z^{-\beta}_{-\alpha}\eqno(3.19)$$ which can be shown, using the invariance under the affine Weyl group, (2.16) $and$ the invariance under the proper outer automorphism, (2.17). Eq. (3.19) specifies the fundamental domain for the gauge field moduli, and agrees with the periodicity implied by the $U(1)$ transformations (3.9,16). Under modular transformations it transforms as $$Z^{\beta}_{\alpha}(\tau+1,{\bar \tau}+1)=Z^{\alpha+\beta}_{\alpha}(\tau,{\bar \tau})\eqno(3.20a)$$ $$Z^{\beta}_{\alpha}(-{1\over \tau},-{1\over {\bar \tau}}) =Z^{-\alpha}_{\beta}(\tau,{\bar \tau})\eqno(3.20b)$$ Eqs. (3.15) imply that, under a modular transformation $$\tau\rightarrow {a\tau+b\over c\tau+d}\;\;,\;\; \left(\matrix{a&b\cr c&d\cr}\right)\,\in\, SL(2,Z)\eqno(3.21)$$ the gauge field moduli transform linearly $$\left(\matrix{\a\cr \b\cr}\right)\rightarrow \left(\matrix{a&b\cr c&d\cr} \right)\;\left(\matrix{\a\cr \b\cr}\right)\eqno(3.22)$$ The meaning of (3.20-22) becomes more transparent if we introduce complex coordinates in the gauge field moduli space, $u=\alpha\tau+\beta$. Then, using (3.19) we can see that the twisted partition function $Z(u,{\bar u}, \tau,{\bar \tau})$ is invariant under the mapping class group of a torus with coordinate $u$ and modulus $\tau$, $$u\rightarrow u+1\,\,\,,\,\,\,u\rightarrow u+\tau\eqno(3.23a)$$ $$\tau\rightarrow\tau+1\,\,\,,\,\,\, u\rightarrow u\eqno(3.23b)$$ $$\tau\rightarrow -{1\over \tau}\,\,\,,\,\,\, u\rightarrow {u\over \tau}\;.\eqno(3.23c)$$ It remains to calculate the integral over the fundamental region of the moduli $$\int_{0}^{1}d\a\int_{0}^{1}d\b Z_{\a}^{\b}(q,\qb;V)\;.$$ We can do the integral over $\b$ first. The only terms in the sum (3.18) that contribute are those that satisfy $k(n-\nb)+{m-\mb\over 2}=0$ Taking into account the ranges of $m,\mb$ and the fact that they are both even or both odd, the only solution is $n=\nb$ and $m=\mb$. Using $$\int_{0}^{1}d\a \sum_{n\in Z} F(n+\a)=\int_{-\infty}^{\infty}d\a F(\a)\eqno(3.24)$$ we finally obtain $$\int_{0}^{1}d\a\int_{0}^{1}d\b Z_{\a}^{\b}(q,\qb;V)={\pi\over \sqrt{kIm \tau}} \sum_{l,{\bar l}=0}^{k}\sum_{m=-k+1}^{k}N^{l,{\bar l}}c_{m}^{l}(q) {\bar c}^{\bar l}_{m}(\qb)\eqno(3.25)$$ To obtain the full partition function for the coset we have to multiply (3.25) with the contribution from the local part of the gauge field , $(det'$$)^{-1/2}$ and the FP determinant, $(det'$$)$ giving a net contribution $(det'$$)^{1/2} =\|\eta(q)\|^{2}$ and an extra factor of $\sqrt{Im\tau}$ coming from the measure of the twists (This factor is standard and can be read from the norm of the gauge field $\|\delta A\|^{2}=\int \sqrt{g}g^{\mu\nu}\delta A_{\mu}\delta A_{\nu}$ using the proper flat metric for a torus parametrized by $\tau$). Putting everything together we obtain (up to constants) $$Z^{V}_{SU(2)/U(1)}=\|\eta(q)\|^{2}\sum_{l,{\bar l}=0}^{k}\sum_{m=-k+1}^{k} N^{l,{\bar l}}c_{m}^{l}(q){\bar c}_{m}^{\bar l}({\bar q})\eqno(3.26)$$ which is the correct parafermionic partition function, [@gq]. Let us now consider the axial case. The boundary condition (3.9) is replaced by $$g(\sigma +1)=e^{i\pi\a\sigma_{3}}g(\sigma)e^{i\pi\a\sigma_{3}}\eqno(3.27)$$ From (3.10) we can verify that ${\bar J}^{\pm}$ is twisted with the oposite sign of $\a$ compared to $J^{\pm}$. Thus, the axial partition function is proportional to $$\int_{0}^{1}d\a\int_{0}^{1}d\b Tr_{H}\left[q^{L_{0}(\a)} \qb^{{\bar L}_{0}(-\a)}e^{2\pi i\b(J^{3}_{0}(\a)+{\bar J}_{0}^{3} (-\a))}\right]\;.\eqno(3.28)$$ Doing the integrals over the moduli, we obtain in this case $$Z^{A}={1\over 2}\|\eta(q)\|^2\sum^{k}_{l,{\bar l}=0}\sum_{m=-k+1}^{k}N^{l,{\bar l}}c^{l}_{m}(q)({\bar c}^{\bar l}_{-m}({\bar q})+{\bar c}^{\bar l}_{-m-2k}({\bar q}))\eqno(3.29)$$ Using again the symmetry under the affine Weyl group (2.26) we obtain that $$Z^{A}=Z^{V}\eqno(3.30)$$ One final comment is in order here, concerning the SU(2)/U(1) case: We can also compute the parafermionic partition function $Z_{SU(2)/U(1)}(r,s)$ twisted around the two cycles of the torus by two elements of its parafermionic symmetry $Z_{k}\times{\tilde Z}_{k}$, ($e^{2\pi i r/k}$, $e^{2\pi i s/k}$). The way to do this is to allow a general twist $$g(\sigma+1)=e^{i\pi\a\sigma_{3}}g(\sigma)e^{i\pi{\bar\a}\sigma_{3}} \eqno(3.31)$$ with $k(\a-{\bar\a})/2=r$ mod $k$. We must also project in the time direction on $J-{\bar J}=s$ mod $k$. Since the twist is now neither axial nor vector there is the standard modular anomaly that can be cancelled by multiplying the twisted partition function by $exp(\pi k \|\tau\|^2(\a-{\bar\a})^2/4Im\tau)$. The procedure described above for the $SU(2)/U(1)$ coset easily generalizes. Consider a simple group $G$. We will gauge the maximal abelian subgroup, namely the Cartan subalgebra. Thus we will be looking at the theory of generalized parafermions, [@g]. A convenient basis to work with is the Chevaley basis. Let $J^{i}$ be a basis of the Cartan, and $\va,\va_{i}\in M$ denote the roots and simple roots respectively. The zero modes of the currents in this basis satisfy $$[J^{i},J^{j}]=0\;\;,\;\;[J^{i},J^{\va}]={2\va\cdot\va_{i}\over \va^{2}}J^{\va} \;\;,\;\;[J^{\va},J^{-\va}]=\sum_{i}m_{i}J^{i}\eqno(3.32a)$$ $${2\va\over \va^{2}}=\sum_{i}m_{i}{2\va_{i}\over \va_{i}^{2}}\;\;,\;\;[J^{\va},J^{\vb}]=\varepsilon_{\va,\vb}r_{\va,\vb} J^{\va+\vb}\;\;{\em for}\;\va+\vb\in M\eqno(3.32b)$$ $$[J^{\va},J^{\vb}]=0\;\;{\em for}\;\;\va+\vb\;\notin M\eqno(3.32c)$$ where $r_{\va,\vb}$ is the smallest integer such that $\vb-r\va\notin M$ and $\varepsilon_{\va,\vb}=\pm 1$. The non-zero components of the Killing form in this basis are given by $$\k(J^{i},J^{j})\equiv \k_{ij} ={4\va_{i}\cdot\va_{j}\over \va_{i}^{2}\va_{j}^{2}}\;\;,\;\;\k(J^{\va},J^{-\va})={2\over \va^{2}}\eqno(3.33)$$ Finally, the central term in the current algebra is given by $k$ times the Killing form. We can now impose the twisted boundary conditions similar to (3.9) $$g(\s+1)=e^{2\pi iz_{i}T^{i}}g(\s)e^{-2\pi iz_{i}T^{i}}\eqno(3.34)$$ Their effect is to twist the algebra in the following way $$J^{\va}_{m}\rightarrow J^{\va}_{m-s(\va)}\;\;,\;\;s(\va)=\sum_{i}z_{i} {2\va\cdot\va_{i}\over \va^{2}}\eqno(3.35a)$$ $$J^{i}_{m}\rightarrow J^{i}_{m}+k(\sum_{j}\k_{ij}z_{j})\delta_{m,0} \eqno(3.35b)$$ $$L_{m}\rightarrow L_{m}+\sum_{i}z_{i}J^{i}_{m}+{k\over 2}\left(\sum_{i,j} \k_{ij}z_{i}z_{j}\right)\delta_{m,0}\;.\eqno(3.35c)$$ The projection factor now becomes $$e^{2\pi i\sum_{i,j}w_{i}\k_{ij}(J^{j}_{0}-{\bar J}^{j}_{0})}\;.\eqno(3.36)$$ Using the string decomposition formulae, (2.24,25) (and properly accounting for the change in the metric) we obtain $$Z_{G}=\sum_{\Lambda,{\bar \Lambda}}\sum_{{\vec \lambda}_{1,2}\in {M^{}\over kM_{L}}}N^{\Lambda,{\bar \Lambda}} c^{\Lambda}_{\vec\lambda_{1}}(q){\bar c}^{\bar \Lambda}_{\vec\lambda_{2}} (\qb)\sum_{\va,\vb\in M_{L}} q^{{k\over 2}(\va+{\vec z}+{\vec\lambda_{1}\over k})^{2}}\qb^{{k\over 2} (\vb+{\vec z}+{\vec\lambda_{2}\over k})^{2}}e^{2\pi i{\vec w}\cdot (k(\va-\vb)+{\vec\lambda}_{1}-{\vec\lambda}_{2})}.\eqno(3.37)$$ At this stage, inspection of (3.37) reveals that all we have found in the SU(2) case goes through here. In particular, a Weyl reflection generated by the simple root $\va_{i}$ is generating from (3.37) the partition function with the $U(1)$ subgroup in that direction axially gauged. An interesting point is that the (axial-vector) duality transformation in the abelian coset can be effected also via an orbifold construction , (this has been observed for $SU(2)_{k}/U(1)$ in [@gq]). Let us consider the parafermionic partition function on the torus with boundary conditions around the two cycles twisted by elements $e^{2\pi ir/k}$, $e^{2\pi is/k}$ of the $Z_{k}$ parafermionic symmetry. This can be evaluated to be $$Z(r,s)={1\over 2}\|\eta\|^{2}e^{-2\pi i{s\over k}}\sum_{l=0}^{k}\sum_{m=-k+1}^{k}c^{l}_{m}{\bar c}^{l}_{m-2r}\eqno(3.38)$$ and the usual vector partition function is $Z(0,0)$. If we construct the orbifold of the original theory with repect to the $Z_{k}$ symmetry, (which amounts to summing over $r,s$), we obtain the axial partition function. In this respect the orbifold projection throws out the order operators and adds as twisted sectors the disorder operators. This is precisely the generalization of what is known to happen in the Ising model. This automatically generalizes to arbitrary abelian cosets where the parafermionic symmetry group is isomorphic to $M^{}/kM_{L}$. Thus we have the following sequence. We gauge the vector U(1), and thus obtain a model with a Killing symmetry associated with axial U(1). This U(1) symmetry is broken to a discrete group (the parafermionic symmetry). Doing an orbifold on that symmetry (which amounts to a flat gauging) we obtain the axially gauged theory. $O(d,d)$ symmetries =================== Combining duality transformations with antisymmetric tensor shifts and linear transformations on the Cartan angles, we can generate a bigger “duality” group which at the level of the partition function acts as $O(d,d,Z)$, [@gr]. In the case of toroidal backgrounds it is easy to see how this works. Invariance under $B_{ij}\rightarrow B_{ij}+N_{ij}$ is obvious from (1.14c), where $N_{ij}$ is an antisymmetric matrix with integer entries. Also, invariance under $G+B\rightarrow U(G+B)U^{T}$ is obvious from (1.10), where $U$ is an arbitrary matrix with integer entries. The duality group (1.9) and the transformations above generate the $O(d,d,Z)$ group. In the non-flat case, similar arguments apply, [@gr], with one difference: the duality group acting on the full operator content of the theory is more complicated than its reduction on the partition function. This is the reflection of our observation that the full duality group in the non-abelian case is isomorphic to the finite Weyl group (or its reductions by subgroups). Keeping this in mind, we can reproduce easily the argument for the partition function. The most general $\s$-model action with $d$ chiral currents is of the form (2.10). We will rewrite it in chiral form, and we will explicitly parametrize the $I(h)$. We will be a bit more general than [@gr] by allowing arbitrary radii for the Cartan angles. The general action takes then the form (up to total derivatives) $$S={1\over 4\pi}\int \left[\kappa_{ij}(\p\a^{i}\pb\a^{j}+\p\g^{i}\pb\g^{j})+ 2\Sigma_{ij}(x)\p\a^{i}\pb\g^{j}+\Gamma^{1}_{ai}(x)\p x^{a}\pb\g^{i}+ \Gamma^{2}_{ia}(x)\pb x^{a}\p\a^{i}+\right.$$ $$\left. +\Gamma_{ab}(x)\p x^{a}\pb x^{b} \right]-{1\over 8\pi}\int R^{(2)}\Phi(x)\eqno(4.1)$$ where $\a^{i},\g^{i}$ take values in $[0,2\pi]$. The action (4.1) is invariant under $\a^{i}\rightarrow \a^{i}+\varepsilon^{i}( {\bar z})$ and $\g^{i}\rightarrow \g^{i}+\zeta^{i}(z)$ with associated chiral (abelian) currents, $$J^{i}=\p\g^{j}\kappa_{ji}+\p\a^{j}\Sigma_{ji}+{1\over 2}\p x^{a}\Gamma^{1}_{ai}\eqno(4.2a)$$ $${\bar J}^{i}=\kappa_{ij}\pb\a^{j}+\Sigma_{ij}\pb\g^{j}+{1\over 2}\Gamma^{2}_{ia}\pb x^{a}\;.\eqno(4.2b)$$ This automatically implies (assuming conformal invariance) that $S$ describes a (not direct, in general) tensor product of a WZW model and some arbitrary decoupled CFT. The currents (4.2) generate the Cartan subalgebra of the full current algebra of the WZW model. We can gauge vectorially the Cartan subalgebra, $$S_{V}=S+{1\over 4\pi}\int\left[A^{i}{\bar J}^{i}-{\bar A}^{i}J^{i}+ {1\over 2}A^{i}(\kappa-\Sigma)_{ij}{\bar A}^{j}\right]\;.\eqno(4.3)$$ Integrating out the gauge fields and gauge fixing $\a^{i}=\g^{i}$ we obtain $$S_{c}={1\over 4\pi}\int\left[E_{ij}(x)\p\a^{i}\pb\a^{j}+F^{1}_{ai}(x)\p x^{a} \pb\a^{i}+F^{2}_{ia}(x)\p\a^{i}\pb x^{a}+F_{ab}(x)\p x^{a}\pb x^{b}\right]-{1\over 8\pi}\int R^{(2)}\phi(x)\eqno(4.4)$$ where $$E_{ij}(x)=4\kappa (1+\kappa^{-1}\Sigma)(1-\kappa^{-1}\Sigma)^{-1}\eqno(4.5a)$$ $$F^{2}(x)=2\kappa(\kappa-\Sigma)^{-1}\Gamma^{2}\;\;,\;\;F^{1}(x)=2\Gamma^{1} (\kappa-\Sigma)^{-1}\kappa\;\;,\;\;F=\Gamma-{1\over 2}\Gamma^{1} (\kappa-\Sigma)^{-1}\Gamma^{2}\eqno(4.5b)$$ $$\phi(x)=\Phi+{\rm log}({\rm det}(\kappa-\Sigma))\;.\eqno(4.5c)$$ The interesting observation is that, given a $\s$-model (4.4) with $d$ Killing symmetries we can always construct it as an abelian coset of a WZW model (4.1). The reason is that relations (4.5) are generically invertible. There is an underlying assumption in this, that should be kept in mind, namely that conformal invariance of (4.4) implies conformal invariance of (4.1). The duality generators $D_{i}$ correspond to switching from vector to axial in the $i$-th component of the gauging. There are also the following obvious symmetries, integer shifts of the antisymmetric tensor $E_{ij}(x)\rightarrow E_{ij}(x)+N_{ij}$ with $N$ an antisymmetric integer matrix, and integer linear transformations of the angles $\a^{i}$ which act as $E\rightarrow UEU^{T}$, $F^{2}\rightarrow UF^{2}$, $F^{1}\rightarrow F^{1}U^{T}$. The full group of invariance of the partition function is the $O(d,d,Z)$ group acting as $$\left(\matrix{E& F^{2}\cr F^{1}&F\cr}\right)\rightarrow \left(\matrix{(aE+b)(cE+d)^{-1}&(ac^{-1}d-b)(cE+d)^{-1}cF^{2}\cr F^{1}(cE+d)^{-1}&F-F^{1}(cE+d)^{-1}cF^{2}\cr}\right)\eqno(4.6)$$ where $$\left(\matrix{a&b\cr c&d\cr}\right)\;\in\; O(d,d,Z)\eqno(4.7)$$ The exact underlying picture of the symmetries above is as follows. The antisymmetric tensor shift in the action corresponds to combined affine Weyl translations on the left and right parts of the theory. The duality transformations, as we argued in the previous section are isomorphic to Weyl transformations. Finally the $GL(d)$ group acts a linear integer transformations of the weight lattice. Again, these are exact symmetries at least when the coset is compact. The reasoning above can be extended to derive the $O(d,d,R)$ action on conformal backgrounds. This was first observed as an invariance of the one-loop string effective action, with backgrounds having $d$ Killing symmetries, [@ven]. The observation is the following. Starting from a background (CFT) with $d$ Killing symmetries, an arbitrary constant shift of the antisymetric tensor, as well as an arbitrary linear combination of the coordinates corresponding to the Killing directions provide another theory which is also conformally invariant. If these transformations are intertwined with the duality transformations $D_{i}$ it can be shown that the full group of transformations is isomorphic to $O(d,d,R)$ which acts as in (4.6). The $O(d,d,Z)$ subgroup generates the same string theory. It is obvious that linear transformations and antisymmetric tensor shifts are exact to all orders in $a'$ and the string loop expansion. In the compact case we have shown in the previous sections that, although the action of the duality transformations $D_{i}$ has to be modified beyond one-loop, there is such a modification, that is exact again non-perturbatively in $\a '$ and perturbatively in the string loop expansion.[^15] One further comment applicable to the compact case: $O(d,d,R)$ transformations do not in general preserve the positivity (unitarity in Minkowski space) of the appropriate conformal field theory. This is obvious for antisymmetric tensor shifts, since they correspond to arbitrary shifts of the weight lattice and thus map integrable to non-integrable reps. It remains to be seen if the string theory constructed from such CFTs remains unitary. On marginal current-current perturbations ========================================= In a CFT with chiral abelian currents like (4.1) it is well known [@cs] that the perturbation $$S_{I}=\lambda\int g_{ij}J^{i}{\bar J}^{j}\eqno(5.1)$$ is marginal. This corresponds to the deformation of the Cartan torus, and generalizes the SU(2) case which we explicitly discussed in section 2. It is also well known from CFT that such perturbations break the non-abelian symmetry while leaving the (abelian) Cartan symmetries intact. However a simple calculation shows that the action $S+S_{I}$ has (deformed) chiral symmetries only to order $\cal{O}(\lambda)$. The way to improve this situation is via (in this case) a special $O(2d,2d)$ transformation, or equivalently by considering the tensor product of this theory with $d$ free scalar fields and gauging an arbitrary linear combination of the two $U(1)^{d}$ symmetries.[^16] To see that we can get the current-current perturbation from an $O(2d,2d)$ transformation we can study first infinitesimal perturbations. Let us consider the infinitesimal form of the transformations in (4.6). $$a\sim 1+\lambda A+\cal{O}(\lambda^{2})\;\;,\;\;d\sim 1-\lambda A^{T}+ \cal{O}(\lambda^{2})\eqno(5.2a)$$ $$b\sim \lambda B+\cal{O}(\lambda^{2})\;\;,\;\; c\sim \lambda C+\cal{O}(\lambda ^{2})\;\;,\;\;B^{T}=-B\;\;,\;\;C^{T}=-C\eqno(5.2b)$$ It is not difficult to see that the following infinitesimal $O(2d,2d)$ transformation $$\left(\matrix{1&0&0&-\lambda g^{T}\cr 0&1&\lambda g &0\cr 0&-\lambda g^{T}&1&0\cr \lambda g&0&0&1\cr}\right)+\cal{O}(\lambda^{2}) \eqno(5.3)$$ generates the perturbation (5.1) Of course there will be also a non-trivial dilaton that can be calculated from (4.5c), which will ensure conformal invariance at the one-loop level. For the $SU(2)$ case there is only one marginal perturbation and (5.3) can be integrated automatically to obtain the finite transformation. The advantage of the finite transformation is that the theory with the fully transformed action will have chiral abelian currents for all values of the parameters. However conformal invariance will still have to corrected beyong one loop. What we remarked so far is hardly surprising. When we look however at the action of $O(d,d)$ transformations on the abelian coset theory, which generically has no chiral currents, we can observe that it still implies that certain current-current perturbations are marginal. Let us first compute the conserved currents in (4,4) associated to the Killing symmetries $\a^{i}\rightarrow \a^{i}+\varepsilon^{i}$ $$J^{i}=\p\a^{j}E_{ji}(x)+\p x^{a}F^{1}_{ai}(x)\eqno(5.4a)$$ $${\bar J}^{i}=E_{ij}(x)\pb\a^{j}+F^{2}_{ia}(x)\pb x^{a}\;.\eqno(5.4b)$$ These currents are conserved $$\pb J^{i}+\p{\bar J}^{i}=0\eqno(5.5)$$ but not chirally conserved. The infinitesimal transformations corresponding to (4.6) become $$\delta E=\lambda (AE+EA^{T}+B-ECE)\;\;,\;\;\delta F=-\lambda F^{1}CF^{2}\eqno(5.6a)$$ $$\delta F^{1}=\lambda F^{1}(A^{T}-CE)\;\;,\;\;\delta F^{2}=\lambda (A-EC)F^{2} \;.\eqno(5.6b)$$ We can now observe that the infinitesimal change in the action (4.4) under the special transformation $A=B=0$ has the form $$\delta S =-\lambda \int J^{i}C_{ij}{\bar J}^{j}\eqno(5.7)$$ which is a specific current-current perturbation (because the matrix $C$ is forced by (5.2b) to be antisymmetric). Of course, there are corrections again to the dilaton via (4.5c). In CFT, whenever there are conserved but not chirally conserved currents, they are bad conformal fields. This can be proven in general in 2-d by showing that normal conservation of a current and conformal invariance (which fixes the form of the two-point function) implies chiral conservation. Moving a bit off criticality we can see that conserved but not chirally conserved currents have severe IR divergences and decouple from the spectrum as one approaches the critical point. It is thus surprising that a perturbation of the form (5.7) is a marginal perturbation of such models. Non-compact cosets ================== Non compact cosets attracted attention recently, [@wit2; @giv; @dvv; @k] as CFTs that provide curved backgrounds for consistent string propagation. They also generically exhibit (semi-classically) spacetime singularities. The prototype theory (and apparently the simplest) is the $SL(2,R)/U(1)$ model describing a two dimensional target manifold. We will consider Euclidean targets, which means that the U(1) we are going to gauge will be compact. If we parametrize the $SL(2,R)$ matrix using Euler angles as $$g=e^{i{\phi\over 2}\s_{2}}e^{{r\over 2}\s_{1}}e^{i{\psi\over 2}\s_{2}}\eqno(6.1)$$ then, upon integrating out the gauge fields, we arrive at the following partition functions for the axial and vector theory $$Z_{A}=\int_{0}^{\infty}{{\rm sinh}rdr\over 1+{\rm cosh}r}\int_{0}^{2\pi}d\phi e^{-{k\over 4\pi}\int[\p_{\mu}r\p^{\mu}r+4{\rm tanh}^{2}{r\over 2}\p_{\mu}\phi \p^{\mu}\phi]}\eqno(6.2)$$ $$Z_{V}=\int_{0}^{\infty}{{\rm sinh}rdr\over 1-{\rm cosh}r}\int_{0}^{2\pi}d\phi e^{-{k\over 4\pi}\int[\p_{\mu}r\p^{\mu}r+4{\rm coth}^{2}{r\over 2}\p_{\mu}\phi \p^{\mu}\phi]}\eqno(6.3)$$ where we have incorporated the dilaton into the measure. In the axial theory, the string propagates on a manifold with the shape of a cigar, which becomes a cylinder asymptotically ($r\rightarrow\infty$). However the manifold of the vector theory, although similar when $r\rightarrow \infty$, has a different, and in fact singular behaviour as $r\rightarrow 0$. The line element and scalar curvature behave as follows, in this region $$ds^{2}\sim dr^{2}+{1\over r^{2}}d\phi^{2}\;\;\;;\;\;\;R\sim {1\over r^{2}} \;.\eqno(6.4)$$ In the Minkowskian (2-d black hole) model the analytic continuation of the axial Euclidean model (6.2) generates region I of spacetime (the asymptotically flat region till the horizon). Region III (from the horizon to the sigularity) corresponds to the self-dual SU(2)/U(1) model. Finally region V (behide the singularity) corresponds to the vector Euclidean model (6.3). The pertinent question here is: are the two models (6.2,3) equivalent, like in the compact case? Our semiclassical derivation of the axial to vector duality is still valid here. However there are reasons to make us distrustful of such a semiclassical reasoning in the non-compact case. One is that the two targets are radically different , unlike the compact case where the target manifold of the axial theory is a reparametrization of that of the vector theory (even when higher loop corrections are included, [@dvv]). The other reason is that the semiclassical spectrum of the two theories in the non-compact case is quite different. In the axial theory, only the continuous series of the SL(2,R) representations contribute, while in the vector theory there are extra contributions from the discrete series. The partition function of the axial model has been computed by Gawedski [@gaw], however that of the vector model remains a mystery. Trying to understand the situation, we will analyse a simpler (but not trivial) case of (potential) axial-vector duality in a non-compact model. Let us consider the conformal field theory on a 2-d Euclidean plane $$Z_{E}=\int d^{2}x e^{-{1\over 4\pi}\int[(\p x^{1})^{2}+(\p x^{2})^{2}]}\eqno(6.5)$$ This is a free field theory that we know everything about, in particular, its exact torus partition function is (up to constants) $$Z^{t}_{E}={1\over Im\tau}{1\over \|\eta(\tau)\|^{4}}\eqno(6.6)$$ where the $1/Im\tau$ factor comes from the integration of the zero modes. We can write this theory in polar coordinates $(r,\t)$, $$Z_{E}=\int_{0}^{\infty}rdr\int_{0}^{2\pi}d\t e^{-{1\over 4\pi}\int[(\p r)^{2}+ r^{2}(\p \t)^{2}]}. \eqno(6.7)$$ We can now apply the O(1,1) duality transformation corresponding to the Killing symmetry associated with translations of $\t$ to obtain the “dual” theory $$Z_{\e}=\int_{0}^{\infty}{dr\over r}\int_{0}^{2\pi}d\t e^{-{1\over 4\pi}\int[ (\p r)^{2}+r^{-2}(\p \t)^{2}]}\eqno(6.8)$$ where the effects of the dilaton transformation were also taken into account by the change in the measure. A naive extrapolation of our results from the compact case would imply that these theories are equivalent, and in particular that the theory (6.8) is a free field theory. However, as in the SL(2,R)/U(1) example, the manifold corresponding to (6.8) is radically different from the flat Euclidean plane of (6.6); it is a curved manifold with a curvature singularity at the origin. It concides with a region close to the origin, of the vector SL(2,R)/U(1) model as can be seen from (6.4). It is also interesting to note that the two theories (6.6,8) can be viewed as axial and vector gauged models of the following $\s$-model, with 3-d target $$Z_{3}=\int_{0}^{\infty}g(r)dr\int_{0}^{2\pi}d\t d\varphi \;\; e^{-{1\over 4\pi}\int[(\p \t)^{2}+(\p \varphi)^{2}+(\p r)^{2}+2f(r)\p\t\pb\varphi]}\eqno(6.9)$$ Choosing $$g(r)={r\over 1+r^{2}}\;\;\;,\;\;\;f(r)={1-r^{2}\over 1+r^{2}}\eqno(6.10)$$ we can verify with a simple computation that by gauging the axial symmetry $\t\rightarrow\t+\varepsilon$, $\varphi\rightarrow\varphi+\varepsilon$ we obtain the free model (6.7) while gauging the vector symmetry $\t\rightarrow\t+\varepsilon$, $\varphi\rightarrow\varphi-\varepsilon$ we obtain model (6.8). It is an interesting question whether the the model (6.9,10) is conformally invariant. There are abelian chiral currents in this model associated with the symmetries $\t\rightarrow\t+\varepsilon({\bar z})$ and $\varphi \rightarrow \varphi+\zeta(z)$ $$J=\p\varphi+f(r)\p\t\;\;,\;\;{\bar J}=\pb\t+f(r)\pb\varphi\;\;;\;\;\p{\bar J}= \pb J=0\eqno(6.11)$$ We can easily check that unless $f(r)$ is the one which corresponds to the SL(2,R) model there are no other chiral currents in the theory (this might seem trivial, but it is possible in principle that the model can be mapped to that of SL(2,R) through a complicated reparametrization). Thus if (6.9) is conformally invariant it describes the product (certainely not direct) of a U(1) theory and some other CFT. We will now proceed to tackle the question posed above: is the free model (6.6) and (6.8) equivalent? Unfortunately it seems extremely difficult to compute exactly the torus partition function of the $\e$ model. Thus we will resort to the so called “minisuperspace” approximation. This amounts essentially to a dimensional reduction to 1-d, that is, neglecting the $\s$ dependence. Thus, we will have to deal with a quantum mechanical model with Langrangian given by $$L_{\e}={1\over 4\pi}[{\dot r}^{2}+r^{-2}{\dot \t}^{2}]\eqno(6.12)$$ The minisuperspace approximation of a CFT is not an approximation in the usual sense of the word. However, it is well understood that, since it describes the quantum mechanics of zero modes, it does not “see” the oscilator part of the spectrum, finite renormalizations of couplings and unitary truncations of the Hilbert space. For example in the “minisuperspace” approximation to the $SU(2)_{k}$ WZW model all representations of $SU(2)$ contribute whereas in the 2-d theory their range is restricted to $0\leq j\leq k/2$. However, if two CFTs are different in this approximation they are certainely different as 2-d theories, whereas the converse is not necessarily true. The Euclidean time quantum mechanical partition function of the $E$ theory is given by dropping the $\eta$-function contributions from (6.6) $$\Omega_{E}\sim \int d^{2}x \int d^{2}pe^{-\tau {\vec p}^{2}}\sim {1\over \tau}\int d^{2}x\eqno(6.13)$$ In general, the quantum mechanical partition function will be given by the trace of $exp[-\tau {\hat H}]$ $$\Omega_{H}=\int \langle x\|e^{-\tau {\hat H}}\|x\rangle\eqno(6.14)$$ and is proportional to the volume of the manifold. The Hamiltonian of the $\e$ theory is minus the Laplacian on the manifold $$H_{\e}=-{\p^2\over \p r^{2}}+{1\over r}{\p \over \p r}-r^{2}{\p^{2}\over \p \t^{2}}\eqno(6.15)$$ $\Omega_{\e}$ is then the integrated trace of the heat kernel and in order to compute it we need to know the spectrum of this Laplacian. This is done in a straightforward manner. The wave functions are labeled by the energy E and the eigenvalues $m$ of angular momentum (${\p \over \p \t}$), which are integers. When $m\not= 0$, the energy spectrum is discrete, $E_{m,n}=4\|m\|(n+1)$, $n=0,1,2, \cdots$, and the energy eigenvalues are doubly degenerate. Their respective eigenfunctions are $$\Psi_{m,n}(r,\t)={mr^{2}\over \sqrt{\pi (n+1)}}e^{-{\|m\|\over 2}r^{2}}L_{n}^{1}(\|m\|r^{2})e^{im\t}\eqno(6.16)$$ properly normalized $$\int_{0}^{\infty}{dr\over r}\int_{0}^{2\pi}d\t \;\Psi_{m,n}^{}(r,\t)\Psi_{m',n' }(r,\t)=\delta_{m,m'}\delta_{n,n'}\eqno(6.17)$$ where $L^{1}_{n}$ is a Laguerre polynomial. When $m=0$, the energy is continuous and non-negative, the corresponding eigenfunctions being $$\Psi_{E}(r,\t)=rJ_{1}(\sqrt{E}r)\;\;\;,\;\;\; E\geq 0\eqno(6.18)$$ normalized as $$\int_{0}^{\infty}{dr\over r}\int_{0}^{2\pi}d\t \;\Psi_{E}^{}(r,\t)\Psi_{E'} (r,\t)=4\pi \delta(E-E')\eqno(6.19)$$ and $J_{1}$ is the standard Bessel function. Already, at the level of the spectrum, the two theories $E$ and $\e$ look quite different. The $E$ theory has positive continuous spectrum of infinite multiplicity. The $\e$ theory has both continuous and discrete spectrum, both of finite multiplicity. These features are also common in the two versions of the SL(2,R)/U(1) model, presented above, namely (6.2,3). The completeness condition can be verified explicitly using standard formulae of special functions $${1\over 4\pi}\int_{0}^{\infty}dE\Psi^{}_{E}(r,\t)\Psi_{E}(r',\t')+ \sum_{m\not=0}\sum_{n=0}^{\infty}\Psi^{}_{m,n}(r,\t)\Psi_{m,n}(r',\t')= r\delta (r-r')\delta(\t-\t')\eqno(6.20)$$ We can now write the trace of the heat kernel $$\langle r,\t\|e^{-\tau {\hat H}_{\e}}\|r,\t\rangle={r^{2}\over 4\pi\tau} e^{r^{2}/2\tau}I_{1}({r^{2}\over 2\tau})+$$ $$+{r^{2}\over 2\pi}\sum_{m=1}^{\infty}{m\over sinh(2m\tau)}e^{-mr^{2} coth(2m\tau)}I_{1}\left({mr^{2}\over sinh(2m\tau)}\right)\eqno(6.21)$$ where the first term comes from the continuous part of the spectrum whereas the second from the discrete. $I_{1}$ is the standard Bessel function. The partition function is given by $$\Omega_{\e}(\tau)=\int_{0}^{\infty}{dr\over r}\int_{0}^{2\pi}\; \langle r,\t\|e^{-\tau {\hat H}_{\e}}\|r,\t\rangle\eqno(6.22)$$ Performing this integral we find that the continuous spectrum contributes a $\tau$-independent divergent piece (a linear divergence) which moreover does not even scale with the volume of the manifold that is logarithmically divergent. The discrete part of the spectrum gives a finite contribution $$\Omega_{\e}^{finite}(\tau)=\sum_{m=1}^{\infty} {1\over e^{4m\tau}-1}\eqno(6.23)$$ Thus, not only the two quantum mechanical partition functions differ but model (6.8) has the pathological behaviour that the free energy per unit volume is infinite. In order to rederive our semiclassical expectations it is instructive to go back to model (6.9) (where $\a'$ was set to one) and re-introduce it explicitly. The free model is of course insensitive to this, since the $\a'$ dependence can be scaled away. However in the “dual” model (6.8) $\a'$ can be scaled away from the action at the expense of changing the range of $\t$ from $[0,2\pi]$ to $[0,2\pi\a']$. Thus we see that at weak coupling, we have to go to the universal cover of the manifold, the eigenvalues of the angular momentum become continuous and the partition function is given by $$\Omega_{\e}^{c}\sim \int_{0}^{infty}dx {1\over e^{4x\tau}-1}\sim {1\over \tau}\times log-divergence\eqno(6.24)$$ which maps properly to the free model. What we have seen so far is that semiclassically we have recovered the duality of the compact case. However the example above raises serious doubts about its validity beyond weak coupling. One could of course contemplate modifications that could bypass the discussion above (like alternative quantization of the dual theory)[^17]. Our point here is that, unlike the compact case, duality if present is certainly not manifest. Similar remarks apply to the more “realistic” SL(2,R)/U(1) model, (6.2,3). As we mentioned earlier the two versions (6.2) and (6.3) differ substantially only in a neighbourhood of $r=0$, and there they are approximated (possibly crudely) by our toy models (6.7) and (6.8). An analysis similar to the above is underway for (6.2,3) in order to settle this question. At the full 2-d level, duality might require, as in the compact case, invariance of the theory under affine translations. This was manifestly true in compact unitary cosets but it is not difficult to see that n the non-compact case, affine Weyl translations map in general a representation to a different one. This can be seen at the level of the non-compact string functions, [@bk]. One way to proceed is to consider orbits under the translation group, but in that case one has always to cope with non-positive representations and the spacetime interpretation is not manifest. In many issues associated with black-holes one usually invokes some analytic continuation from Minkowski to Euclidean space. As we have seen it plausible that there are two inequivalent such continuations depending on the region of spacetime. This might imply a different behaviour (and maybe interpretation) for such issues as Hawking radiation etc. [99]{} K. Kikkawa, M Yamazaki, Phys. Lett. B149 (1984) 357; N. Sakai, I. Senda, Prog. theor. Phys. 75 (1986) 692; V. P. Nair, A. Shapere, A. Strominger, F. Wilczek, Nucl. Phys. B287 (1987) 402. A. Giveon, E. Rabinovici, G. Veneziano, Nucl. Phys. B322 (1989) 167; A. Giveon, N. Malkin, E. Rabinovici, Phys. Lett. B238 (1990) 57. E. Kiritsis, Mod. Phys. Lett. A6 (1991) 2871. E. Verlinde, M. Rocek, Nucl. Phys. B373 (1992) 630. A. Giveon, M. Rocek, “[Generalized Duality in Curved String Backgrounds]{}", Princeton preprint, hep-th/9112070. T. H. Buscher, Phys. Lett. B201 (1988) 466. A. Schwarz, A. Tseytlin, “[Dilaton Shift under Duality and Torsion of the Elliptic Complex]{}", Imperial preprint, hep-th/9210015. E. Witten, Comm. Math. Phys. 92 (1984) 455. K. Bardakci, M. Halpern, Phys. rev. D3 (1971) 2493; M. Halpern, Phys. Rev. D4 (1971) 2398; P. Goddard, A. Kent, D. Olive, Phys. Lett. B152 (1985) 88; M. Douglas, PhD Thesis, Caltech 1988, unpublished. K. Bardakci, E. Rabinovici, B. Säring, Nucl. Phys. B299 (1988) 157; D. Altschuler, K. Bardakci, E. Rabinovici, Comm. math. Phys. 118 (1988) 241; K. Gawedski, A. Kupiainen, Phys. Lett. B215 (1988) 119; Nucl. Phys. B320 (1989) 625; D. Karabali, Q.-H. Park, H. Schnitzer, Z. Yang, Phys. Lett. B216 (1989) 307; D. Karabali, H. Schnitzer, Nucl. Phys. B329 (1990) 649. M. Halpern, E. Kiritsis, Mod. Phys. Lett. A4 (1989) 1373; ibid. A4 1797 (E). R. Dijkgraaf, E. Verlinde, H. Verlinde, Nucl. Phys. B371 (1992) 269. K. Meissner, G. Veneziano, Phys. Lett. B267 (1991) 33; Mod. Phys. Lett. A6 (1991) 3397; M. Gasperini, G. Veneziano, Phys. Lett. B277 (1992) 256; M. Gasperini, J. Maharana, G. Veneziano, “[Boosting away Singularities from Conformal String Backgrounds]{}”, CERN preprint, hep-th/9209052. A. Sen, Phys. Lett. B271 (1991) 295; ibid. B274 (1992) 34; Phys. Rev. Lett. 69 (1992) 1006; S. Hassan, A. Sen, Nucl. Phys. B375 (1992) 103. J. Maharana, J. Schwarz, “[Non-compact Symmetries in String Theory]{}", Caltech preprint, CALT-68-1790. M. Henninngson, S. Hwang, P. Roberts, B. Sundborg, Phys. Lett. B267 (1991) 350. A. Giveon, Mod. Phys. Lett. A6 (1991) 2843. D. Gepner, E. Witten, Nucl. Phys. B278 (1986) 493. V. Kac, D. Peterson, Adv. Math. 53 (1984) 125. S. K. Yang, Phys. Lett. B209 (1988) 242. X. De La Ossa, F. Quevedo, “[Duality Symmetries from Non-Abelian Isometries in String Theory]{}”, Neuchâtel preprint, hepth/9210021. D. Gepner, Z. Qiu, Nucl. Phys. B285 \[FS19\] (1987) 423. D. Gepner, Nucl. Phys. B290 \[FS20\] (1987) 10. S. Chaudhuri, J. A. Schwarz, Phys. Lett. B219 (1989) 291. S. Hassan, A. Sen, “[Marginal Deformations of WZNW and Coset Models from O(d,d) Transformations]{}”, Tata preprint, hepth/9210121. E. Witten, Phys. Rev. D44 (1991) 314; E. Martinec, S. Shatasvili, Nucl. Phys. B368 (1992) 338; D. Kutasov, A. Bershadsky, Phys. Lett. B266 (1991) 345; J. Horne, G. Horowitz, A. Steif, Phys. Rev. Lett. 68 (1992) 568; N. Ishibashi, M. Li, A. Steif, Phys. Rev. Lett. 67 (1992) 3336; P. Horava, Phys. Lett. B278 (1992) 101; C. Nappi, E. Witten, Phys. Lett. B293 (1992) 309; C. Kounnas, D. Lüst, Phys. Lett. B289 (1992) 56; D. Gershon, “[Exact Solutions of 4-D Black Holes in String Theory"]{}, Tel Aviv preprint, TAUP-1937-91; P. Ginsparg, F. Quevedo, ”[Strings in Curved Spacetimes, Black Holes, Torsion and Duality]{}“, Los Alamos preprint, hepth/9202092; S. Chaudhuri, J. Lykken, ”[String Theory, Black Holes and SL(2,R) Current Algebra]{}", Fermilab preprint, hepth/9206107; J. L. Gervais, M. Saveliev, Phys. Lett. B286 (1992) 271; J. Ellis, N. Mavromatos, D. Nanopoulos, Phys. Lett. B289 (1992) 25, and references therein; I. Bars, “[Superstrings in Curved Spacetimes]{}", USC preprint, hepth/9210079 and references therein; K. Gawedski, Lectures given at the Cargese Summer Inst. “New Symmetry Principles in QFT”, July 1991. I. Bakas, E. Kiritsis, Proceedings of the RIMS “Infinite Analysis” project, Kyoto, 1991, published in Int. J. Mod. Phys. A7 \[Sup. 1A\] (1992) 55. [^1]: Invited talk presented at the International Workshop on String Theory, Quantum Gravity and the Unification of Fundamental Interactions, Roma, 21-26 Sept. 1992. [^2]: Present address: Theory Division, CERN, CH-1211, Geneva 23, Switzerland [^3]: email: [email protected] [^4]: Similar results can be obtained for correlation functions. In the path integral framework, the general conformal operator (affine U(1) primary) with $(\Delta, {\bar \Delta})=((mR+nR^{-1})^{2}/4, (mR-nR^{-1})^{2}/4)$ is represented by the insertion of the field $e^{in\phi(z_{0},{\bar z}_{0})}$ and the instruction to do the integration over maps $\phi(z,{\bar z})$ which wind $m$ times around the point $z_{0}$. [^5]: Non-constant, but $x^{0}$-independent $G_{00}$ can be handled by the quotient method, [@vr; @gr]. [^6]: There is a more general class of CFTs whose structure is much less understood, namely the affine-Virasoro constructions, [@hk]. [^7]: This has been effectively done for the SL(2,R)/U(1) coset in [@hw]. [^8]: In the case of $SL(2,R)$ this orbit method works for the discrete series but it is not at all obvious how it could be implemented in the continuous series. [^9]: The measure also changes by a finite computable piece, see [@k] [^10]: In a highest weight representation where the affine primaries have $L_{0}$ eigenvalue $\Delta$, the grade $n$ of a state is the eigenvalue of $L_{0}-\Delta$ on that state. [^11]: In general this group is isomorphic to the center of the finite Lie group [^12]: In [@dq] a non-abelian form of duality transformations was introduced. It is not clear if these are related to the extended duality group introduced above. [^13]: The analysis can be extended to non-simple $G$ and/or irregularly embedded $H$, but it is certainely more involved. [^14]: More precisely it is a hybrid of operator methods and the path integral approach of Gawedski, [@c2]. [^15]: Arguments, to the extend that $O(d,d,R)$ transformations can be made exact symmetries to all orders in $\a'$ where also given in [@sen] from a string field theory point of view. [^16]: Similar observations were made independently in [@hs]. [^17]: A little analysis shows that the freedom in quantizing the theory is so wide that can reproduce any possible Hamiltonian, that commutes with the angular momentum.	{ "pile_set_name": "ArXiv" }
--- abstract: 'Observations of exoplanetary systems provide clues about the intrinsic distribution of planetary systems, their architectures, and how they formed. We develop a forward modelling framework for generating populations of planetary systems and “observed” catalogues by simulating the detection pipeline (`SysSim`). We compare our simulated catalogues to the DR25 catalogue of planet candidates, updated to include revised stellar radii from Gaia DR2. We constrain our models based on the observed 1D marginal distributions of orbital periods, period ratios, transit depths, transit depth ratios, transit durations, transit duration ratios, and transit multiplicities. Models assuming planets with independent periods and sizes do not adequately account for the properties of the multiplanet systems. Instead, a clustered point process model for exoplanet periods and sizes provides a significantly better description of the population, particularly the observed multiplicity and period ratio distributions. We find that $0.56^{+0.18}_{-0.15}$ of FGK stars have at least one planet larger than $0.5 R_\oplus$ between 3 and 300 d. Most of these planetary systems ($\sim 98\%$) consist of one or two clusters with a median of three planets per cluster. We find that the dichotomy is evidence for a population of highly inclined planetary systems and is unlikely to be solely due to a population of intrinsically single planet systems. We provide a large ensemble of simulated physical and observed catalogues of planetary systems from our models, as well as publicly available code for generating similar catalogues given user-defined parameters.' author: - \| Matthias Y. He$^{1,2,3,4}$[^1], Eric B. Ford$^{1,2,3,4}$, and Darin Ragozzine$^{5}$\ $^{1}$Department of Astronomy and Astrophysics, 525 Davey Laboratory, The Pennsylvania State University, University Park, PA 16802, USA\ $^{2}$Center for Exoplanets and Habitable Worlds, 525 Davey Laboratory, The Pennsylvania State University, University Park, PA 16802, USA\ $^{3}$Center for Astrostatistics, 525 Davey Laboratory, The Pennsylvania State University, University Park, PA 16802, USA\ $^{4}$Institute for CyberScience, The Pennsylvania State University, University Park, PA 16802, USA\ $^{5}$Department of Physics and Astronomy, N283 ESC, Brigham Young University, Provo, UT 84602, USA\ date: 'Accepted ?. Received ?; in original form ?' title: 'Architectures of Exoplanetary Systems. I: A Clustered Forward Model for Exoplanetary Systems around ’s FGK Stars' --- \[firstpage\] methods: statistical – planetary systems – planets and satellites: detection, fundamental parameters, terrestrial planets – stars: statistics Introduction {#Introduction} ============ Within the past decade, NASA’s mission [@B2010; @B2011a; @B2011b; @B2013] has discovered thousands of exoplanets and hundreds of multiplanet systems around Sun-like stars. Within these systems, there is an abundance of short period planets (i.e. with orbits much smaller than that of Earth) and tightly packed multiple planets [@La2011; @Li2011b; @Li2014; @R2014]. The wealth of transit detections generated by and the relative homogeneity of the sample allows for the study of exoplanet systems as a whole, enabling statistical exploration of the population of exoplanets detectable by . Multitransiting planetary systems are especially valuable because they serve as crucial tests for our models of planetary formation, their resulting architectures, and their subsequent evolution and stability [@RH2010; @Fo2014; @WF2015]. The abundance of systems with many transiting planets indicates that systems with small mutual inclinations are common, as small mutual inclination angles between planets in the same system are required to explain how a single observer can see so many planets in a transiting configuration. These observations contribute to our theories of planet formation, supporting the picture that planets likely formed in relatively flat, gaseous discs. Previous studies have explored the degree of coplanarity required to explain the multitude of multitransiting systems, focusing almost exclusively on the observed multiplicity and the transit duration ratio distributions [@La2011; @Li2011b; @FM2012; @F2012; @J2012; @TD2012; @WSS2012; @F2014]. However, one emergent puzzle from the studies of observed transiting multiplicities is the apparent excess of single transiting systems, which has led some authors to speculate the existence of a second population of intrinsic singles or highly inclined multiplanet systems, a so-called “ dichotomy” [@Li2011b; @J2012; @HM2013; @BJ2016]. It is unclear how the planet multiplicities and their orbital inclinations mesh with other observed properties of the transiting population, such as the distributions of their orbital periods and period ratios. The period ratios of adjacent planet pairs in multiplanet systems, which is a direct measure of their physical separations and thus stability, have been studied [@F2014; @SH2015]. The period ratios range from close to unity to over several dozen, with 1.172 being the smallest observed period ratio in the KOI-1665 system [@Li2011a]. Smaller period ratios of 1.038 (in KOI-284) and 1.065 (in KOI-2248) are now interpreted as due to transiting planets with very similar periods around different stars in a binary system [@Li2011a; @Li2014]. These studies also find relative excesses of (apparently) adjacent planet pairs with period ratios near (slightly larger than) first-order mean-motion resonances (MMRs), namely near the 3:2 and 2:1 ratios [@F2014; @SH2015]. Inferring the true rate of near resonant systems is difficult, due to the interaction of the geometric joint transit probability, detection efficiency, and the unknown distribution of mutual inclinations and eccentricities. A few studies have also attempted to understand the size distribution of planets in multiplanet systems, by probing their relative radii using transit depth ratios in order to minimize the effects of our uncertainties in the stellar radii. For example, @C2013 found that for planet pairs with planets larger than $\sim 3 R_\oplus$, the outer planet tends to be larger than the inner one, although they did not observe this trend for planets $\lesssim 3 R_\oplus$. In addition, planet sizes appear to be highly correlated, as evidenced by the peaked nature of the adjacent radii ratio distribution [@W2018a], and this clustering extends to planet masses [@M2017]. While the mechanisms of photoevaporation [@OW2013; @F2017; @OW2017; @vE2017; @C2018] or core-powered mass-loss from formation [@GSS2016; @GSS2018; @GS2018] have been proposed to explain these features in the distributions of radii and radii ratios, complex observational biases limit our ability to distinguish models or understand the relative contribution of physical and observational effects. To further illustrate this point, most recently @Z2019 has challenged the statistical significance of the correlations reported by [@C2013] and @W2018a. @Z2019 found that testing the significance of correlations by bootstrap re-sampling with cuts on the signal-to-noise ratios (as opposed to planet size) resulted in smaller and less significant correlations for sizes of planets in one system, uniformity of spacing, and preference for the outer planet to be larger than the inner planet. They suggest that a better way to study such correlations would be via forward modelling the detection and selection processes using a model for the intrinsic distribution of planetary properties. This study takes that approach and provides a fresh perspective on the significance of correlations in orbital period, planet size, and uniformity of spacing. There is an overwhelming wealth of information in the observed population of exoplanets, especially encoded in the multiplicity, period ratio, and transit depth and duration ratio distributions. Exploratory analyses are difficult to interpret due to complex geometric and sensitivity biases in the detection pipeline. Earlier works such as those discussed above attempted to study these features but had fewer detections to work with, a more limited understanding of the pipeline’s detection efficiency, greater uncertainties in the stellar properties, and intractable biases in the planet catalogue due to human-influenced vetting of planet candidates. In this work, we are able to address these concerns by making use of several recent analyses and updates to the stellar and planetary properties. We use the final DR25 catalogue [@T2018], which was generated using a completely automated procedure for the vetting of planet candidates using the Kepler Robovetter. We rely and build on the Exoplanets Systems Simulator (“SysSim”), which makes use of several additional DR25 data products describing ’s detection pipeline [@BC2017a; @BC2017b; @BC2017c; @C2017; @Co2017], and is described in @H2018 [@H2019]. Finally, we adopt the improved stellar properties from the European Space Agency’s Gaia mission [@Gaia2018], and planet candidates around a clean sample of FGK dwarfs as defined by @H2019. In this paper, we outline a new framework for simulating catalogues of observed transiting planets in §\[Methods\], which we use to explore several statistical, yet physically motivated, models for the intrinsic distribution of planetary systems. In particular, we explore a marked, clustered point process for generating planetary systems. For each planetary system, we attempt to generate the planets by first drawing a period and radius scale for each cluster and then drawing orbital periods and planet sizes for each planet in a cluster conditioned on the cluster’s period and radius scales. These models are described in §\[Models\]-§\[Obs\_pipeline\]. We define a set of summary statistics, a distance function to compare models to data, and a subset of the DR25 catalogue we use to constrain our models, in §\[Obs\_compare\]. We describe the optimization procedure used to explore the multidimensional parameter space of each model in §\[Optimization\]. In §\[GP\], we provide a brief review of the statistical machinery of Gaussian processes, and describe how we adapt it to serve as an “emulator” for our models and compute the credible regions for each model parameter using Approximate Bayesian Computation. The results for the credible regions of the parameters for each model are presented and discussed in §\[Model\_params\]. The results of each of our models, for both the observed and physical (intrinsic) distributions of planetary systems, and a direct comparison between them are described in §\[Discussion\]. We compute some estimates for the fraction of stars with planets and the number of planets per star or planetary system in §\[Planet\_rates\]. A discussion of the dichotomy is presented in §\[Dichotomy\]. We discuss how our models can be improved in §\[secFuture\], and summarize our main results and encourage the use of our code and model catalogues in §\[Conclusions\]. Methods {#Methods} ======= The models described in this paper are developed in the framework of the Exoplanets Systems Simulator, which we refer to as “SysSim”. The SysSim codebase is written in the Julia language [@B2014] and can be installed as the ExoplanetsSysSim.jl package [@F2018b]. Our models can be accessed at <https://github.com/ExoJulia/SysSimExClusters>. Our general framework for performing an (approximate) Bayesian analysis can be summarized with the following steps: Step 0: Define a statistical description for the intrinsic distribution of exoplanetary systems\ We present three separate statistical models in this paper (§\[Models\]). Each model is described by a set of physically motivated model parameters. *Step 1: Generate an underlying population of exoplanetary systems (physical catalogue) from a given model\ Each model involves randomly drawing stars from the stellar catalogue and populating them with planets. Planet radii and periods are drawn. Then, masses are assigned from the radii and a rejection-sampling algorithm is applied to only keep planetary systems that are physical. Step 2: Generate an observed population of exoplanetary systems (observed catalogue) from the physical catalogue*\ The systems are oriented by assigning orbits to each planet and the SysSim forward model simulating the detection pipeline is applied to generate a catalogue of observed, transiting planets along with their “measured” properties. Step 3: Compare the simulated observed catalogue with the data\ A set of summary statistics are computed for both the simulated observed and data, and a distance function is computed to compare the two catalogues. Step 4: Optimize the distance function to find the best-fitting model parameters\ Steps 0–3 are repeated by passing the distance function into an optimization algorithm in order to explore the parameter space and attempt to find the best-fitting model parameters. Step 5: Explore the posterior distribution of model parameters using a Gaussian Process (GP) emulator\ For each model, a GP model is trained on a set of points (model parameters with computed distances) in order to form an emulator for the full forward model. The emulator is used to quickly characterize the parameter space. Step 6: Compute credible intervals for model parameters and simulated catalogues using Approximate Bayesian Computing\ Model parameters are drawn from prior distributions, and used to draw an emulated distance from the GP emulator. If the emulated distance is less than a specified distance threshold, we accept the set of model parameters. The credible regions are reported based on the accepted sample of model parameters, and we present and analyze the simulated physical and observed catalogues that resulted in the distance from the full forward model being less than the distance threshold. We describe each of these steps in full detail in the following subsections. Models for generating planetary systems {#Models} --------------------------------------- We describe three separate models, starting with a baseline non-clustered model, extending to a clustered periods model, and then a clustered periods and sizes model. First, we give a brief overview of all three models. Then we provide details about the method for drawing each of the properties under the different models. Results from the non-clustered model shown in §\[Discussion\], will provide empirical support for motivating use of the clustered models. In the non-clustered model, we first draw a number of planets for each star and then draw orbital periods and planet sizes independently for each planet in a system, using a simple power law for period and a broken power law for radius. In both of the clustered models, we first draw a number of “clusters” of planets for each star. For each cluster, we draw a number of planets and a period scale. In the clustered periods and sizes model, we also draw a radius scale for each cluster. Then, the periods and sizes of the planets are drawn from distributions centred on the period scale and the radius scale, respectively (i.e. conditioned on the properties of their parent cluster). Thus, the properties of planets are explicitly correlated with those of other planets from the same cluster. This leads to closely spaced planets often having strong correlations, but more widely spaced planets having weaker correlations. We show a cartoon illustration of our three models in Figure \[fig:models\_cartoon\]. ### Numbers of planets We will constrain our models based on the observed multiplicity distribution (as described in §\[Summary\_stats\]) and use the results to address both the overall rate of planets per star and the fraction of stars with at least one planet in this study. Therefore, it is important to consider the process for assigning the number of planets to each star. For the non-clustered model, we draw the number of planets in each system from a Poisson distribution, $N_p \sim {\rm Poisson}(\lambda_p)$. In the clustered models we first draw a number of clusters from a Poisson distribution, $N_c \sim {\rm Poisson}(\lambda_c)$. Then, we draw a number of planets for each cluster from a zero-truncated Poisson (ZTP) distribution, $N_p \sim {\rm ZTP}(\lambda_p)$, where the probability mass function is given by: $$\begin{aligned} g(k \| \lambda) = \frac{\lambda^k}{(e^\lambda - 1)k!}. \label{eq_ZTP}\end{aligned}$$ The choice of the ZTP is simply to avoid drawing empty clusters with no planets. [^2] The way in which zero-planet systems are drawn is an important consideration for our models. In our non-clustered model, assigning a star no planets occurs when $N_p = 0$ is drawn from the Poisson distribution for the number of planets per system, while in our clustered models, this occurs only when $N_c = 0$ is drawn from the Poisson distribution for the number of clusters per system. Both the number of clusters and planets per cluster are truncated to not exceed maximum values, $N_c \leq N_{c,\rm max}$ and $N_p \leq N_{p,\rm max}$, respectively. We set $N_{c,\rm max} = 10$ and $N_{p,\rm max} = 15$ in our clustered models, and $N_{p,\rm max} = 20$ in our non-clustered model, unless stated otherwise in order to avoid generating systems with extremely large numbers of planets, as these are computationally expensive. In addition to truncation effects, the true mean rates of planets per system (non-clustered model), clusters per system and planets per cluster (clustered models) are somewhat less than the values suggested by the parameters $\lambda_p$ and $\lambda_c$ due to our stability criteria and rejection sampling algorithm (described in §\[Stability\] and §\[Procedure\]). Thus, while the values of $\lambda_p$ (non-clustered) and $\lambda_c \lambda_p$ (clustered) serve as approximations for the mean number of planets per star, they should not be overinterpreted. We use the true rates of planets per star for more detailed calculations in §\[Discussion\]. ### Orbital Periods {#sec:OribtalPeriods} Non-clustered model:\ Orbital periods are drawn independently from a simple power-law distribution: $$\begin{aligned} f(P) &\propto {P}^{\alpha_P}, \quad P_{\rm min} \leq P \leq P_{\rm max}, \label{eq_P} \end{aligned}$$ where $f(P)$ is the probability density function (PDF) and $\alpha_P$ is the power-law index for the period distribution. We choose $P_{\rm min} = 3$ d and $P_{\rm max} = 300$ d. Clustered periods and clustered periods and sizes models:\ The orbital period of each planet is drawn conditionally on the period scale for its parent cluster, $P_c$. For each cluster we draw a trial $P_c$ from a simple power law, and the planets in each cluster have trial periods drawn from a log-normal distribution conditioned on $P_c$ with a cluster width that is scaled to the number of planets: $$\begin{aligned} f(P_c) &\propto {P_c}^{\alpha_P} \label{eq_Pc} \\ P'_i &\sim {\rm Lognormal}(0, N_p\sigma_P) \label{eq_P_clusP} \\ P_i &= P_c P'_i,\quad P_{\rm min} \leq P_c \leq P_{\rm max} \end{aligned}$$ where $P'_i$ are the unscaled periods, $P_i$ are the true periods, $N_p$ is the number of planets in the cluster, and $\sigma_P$ is the cluster width scale factor, per planet in the cluster. The trial periods are accepted or rejected based on a heuristic for orbital instability, as described in §\[Procedure\].[^3] ### Planet sizes Non-clustered model and clustered periods model:\ Planet radii are drawn independently from a broken power law: $$\begin{aligned} f(R_p) &\propto \begin{cases} {R_p}^{\alpha_{R1}}, & R_{p,\rm min} \leq R_p \leq R_{p,\rm break} \label{eq_brokenR} \\ {R_p}^{\alpha_{R2}}, & R_{p,\rm break} < R_p \leq R_{p,\rm max} \end{cases}, \end{aligned}$$ where $f(R_p)$ is the PDF, and $\alpha_{R1}$ and $\alpha_{R2}$ are the broken power-law indices for the planet radius distribution below and above $R_{p,\rm break}$, the break radius, respectively. Our choice of a broken power law for the radius distribution is motivated by previous studies suggesting that there is an observed break at around 2–3$R_\oplus$, with a rise in occurrence down to $\sim 2 R_\oplus$ and a plateau below that [@Y2011; @H2012; @PMH2013b]. Clustered periods and sizes models:*\ The radius of each planet is drawn conditionally on the radius scale for its parent cluster, $R_{p,c}$. The cluster radius scales $R_{p,c}$ are drawn from a broken power law, while the planets in each cluster have their radii drawn from a log-normal distribution conditioned on $R_{p,c}$ with a fixed scale for the cluster width $\sigma_R$: $$\begin{aligned} f(R_{p,c}) &\propto \begin{cases} {R_{p,c}}^{\alpha_{R1}}, & R_{p,\rm min} \leq R_{p,c} \leq R_{p,\rm break} \label{eq_brokenRc} \\ {R_{p,c}}^{\alpha_{R2}}, & R_{p,\rm break} < R_{p,c} \leq R_{p,\rm max} \end{cases}. \\ R_{p,i} &\sim {\rm Lognormal}(R_{p,c}, \sigma_R). \label{eq_R_clusR} \end{aligned}$$ For all three models, we limit our analyses to $R_{p,\rm min} = 0.5 R_\oplus$ and $R_{p,\rm max} = 10 R_\oplus$, as the distribution of larger or smaller planets is not well constrained by observations. ### Planet masses Planets vary significantly in structure and composition [@WL2012; @LF2014; @CK2016; @WRF2016]. We adopt a non-parametric mass–radius relation from @NWG2018 in order to assign planet masses probabilistically given their drawn planet radii. This mass-radius relation involves a series of Bernstein polynomials used to model the joint distribution of mass and radius based on a sample of 127 exoplanets with RV or TTV masses, and is more flexible than simpler, power law mass–radius relations. To accelerate computations, we pre-compute and use a $1001\times1001$ look-up table for the mass–radius relation. ### Eccentricities We draw the orbital eccentricities from a Rayleigh distribution, $e \sim {\rm Rayleigh}(\sigma_e)$, where the Rayleigh parameter ($\sigma_e$) defines the scale for eccentricities, following previous works (e.g. @M2011 [@F2014]). ### Mutual inclinations {#Incl} Finally, we allow for two populations of planetary systems with separate distributions of mutual inclinations between planets. We use Rayleigh distributions for both populations, as the Rayleigh distribution has been used to describe mutual inclinations in previous works (e.g. @FW2009 [@Li2011b; @FM2012; @DLC2016]). Thus, for a fraction $f_{\sigma_{i,\rm high}}$ of systems we draw from a broad distribution of mutual inclinations while for the remaining $1 - f_{\sigma_{i,\rm high}}$ fraction of systems we draw from a narrower distribution: $$\begin{aligned} i_m \sim \begin{cases} {\rm Rayleigh}(\sigma_{i,\rm high}), & u < f_{\sigma_{i,\rm high}} \label{eq_incl} \\ {\rm Rayleigh}(\sigma_{i,\rm low}), & u \geq f_{\sigma_{i,\rm high}} \end{cases},\end{aligned}$$ where $u \sim {\rm Unif}(0,1)$ and $\sigma_{i,\rm low} \leq \sigma_{i,\rm high}$. In addition, we check whether each planet is near a 2:1, 3:2, 4:3, or 5:4 MMR with another planet, and draw $i_m \sim {\rm Rayleigh}(\sigma_{i,\rm low})$ for planets that are. For our purposes, we regard adjacent planet pairs as “near an MMR” if their period ratio is within 5% exterior to the MMR. This amounts to period ratios between 1.5 and 1.575 as near the 3:2 MMR and period ratios between 2 and 2.1 as near the 2:1 MMR, for example. Our motivation for forcing planets near MMRs to have more coplanar orbits than other planets is to explore whether this alone can produce the apparent relative excesses of planets with period ratios just wide of MMRs [@F2014; @SH2015], or whether it is essential to generate more planets near MMRs than what is drawn naturally from our model for orbital periods (see §\[sec:OribtalPeriods\]). Since planets with more coplanar orbits are more likely to transit together than highly mutually inclined planets, our model has the effect of producing slightly more observed planet pairs near MMRs than planet pairs with arbitrary period ratios. To test the robustness of our conclusions to this special treatment of the near-MMR planets, we also explore a model without the lowering of mutual inclinations for such planets, using our clustered periods and sizes model. We refer to this model as the “alternative MMR inclinations” model for the remainder of the paper. The results of this model will be primarily discussed in §\[Model\_params\] and §\[Dichotomy\]. We leave $f_{\sigma_{i,\rm high}}$, $\sigma_{i,\rm high}$, and $\sigma_{i,\rm low}$ as free parameters of the models. ### Stability criteria {#Stability} We test whether planetary systems are likely to be long-term Hill stable after drawing their periods. When a planetary system is identified as likely unstable, then it is discarded and we attempt to redraw orbital periods. Our instability criterion is based on the spacing between adjacent planets ($\Delta$) normalized by the mutual Hill radius ($R_H$), which is given by: $$\begin{aligned} R_H = \bigg(\frac{a_{\rm in} + a_{\rm out}}{2}\bigg)\bigg[\frac{m_{\rm in} + m_{\rm out}}{3 M_\star}\bigg]^{1/3}, \label{eq_mhill}\end{aligned}$$ where $a_{\rm in}$, $a_{\rm out}$ are the semimajor axes and $m_{\rm in}$, $m_{\rm out}$ are the masses of the inner and outer planets, respectively, and $M_\star$ is the mass of the stellar host. We define an instability criterion that is parametrized in terms of $R_H$: $$\begin{aligned} \Delta = \frac{a_{\rm out}(1-e_{\rm out}) - a_{\rm in}(1+e_{\rm in})}{R_H}. \label{eq_Nhill}\end{aligned}$$ To avoid generating planetary systems likely to be unstable, we require $\Delta \geq \Delta_c$ where $\Delta_c$ is the minimum separation (held fixed). For two circular and coplanar orbits, the minimum separation required to be Hill stable (i.e., no close encounters) is given by $\Delta > 2\sqrt{3} \simeq 3.46$ [@G1993]. @CWB1996 used numerical simulations to conclude that systems with equal-mass planets separated by $\Delta < 10$ are virtually always unstable. @PW2015 find that $\Delta \gtrsim12$ is required for long-term stability of planets in multiplanet systems given certain assumptions about the masses and eccentricity distribution for the exoplanets. In our early exploratory analyses where $\Delta_c$ was treated as a free parameter, we found that our models preferred smaller values, $\Delta \lesssim 10$. Therefore, we set $\Delta_c = 8$ for the remainder of this paper unless otherwise stated. ### Summary of the free parameters of our models Our non-clustered model has a total of 10 free parameters: $\lambda_p$, $\alpha_P$, $\alpha_{R1}$, $\alpha_{R2}$, $R_{p,\rm break}$, $\sigma_e$, $\Delta_c$, $f_{\sigma_{i,\rm high}}$, $\sigma_{i,\rm high}$, and $\sigma_{i,\rm low}$. The clustered periods model has an additional two parameters $\lambda_c$ and $\sigma_P$ (the cluster width in log-period per planet in the cluster) to give 12 free parameters, while the clustered periods and sizes model adds yet another parameter $\sigma_R$ (the cluster width in log-radius) for a total of 13 free parameters. In our early exploratory analysis we found that the break radius $R_{p,\rm break}$ and the minimum separation $\Delta_c$ were not well constrained by observations (see §\[Optimization\] for more details). In particular, $R_{p,\rm break}$ can take on a wide range of values and the models seem to prefer smaller values of $\Delta_c \lesssim 10$. Thus, in order to improve the efficiency of the fitting procedure, we reduce the number of free parameters and thus the size of the parameter space by setting these parameters to $R_{p,\rm break} = 3 R_\oplus$ and $\Delta_c = 8$ based on previous studies (e.g., @PMH2013b). This reduces the number of free parameters for each model by two. A list of all the parameters of our models used in this paper is provided in Table \[tab:params\]. Parameter Definition of parameter Equation Relevant quantity/distribution --------------------------- ------------------------------------------------------------------------------------------------------------------- --------------------------------------- ----------------------------------------------------------------------------- $N_{\rm stars,sim}$ Number of simulated stars - - $\lambda_c$ Mean number of clusters per star (before rejection sampling) - $N_c \sim {\rm Poisson}(\lambda_c)$ $N_{c,\rm max}$ Maximum number of clusters per star - $N_c \leq N_{c,\rm max}$ $\lambda_p$ Mean number of planets per system (non-clustered model) or cluster (clustered models) (before rejection-sampling) - $N_p \sim {\rm Poisson}(\lambda_p)$ or ${\rm ZTP}(\lambda_p)$ $N_{p,\rm max}$ Maximum number of planets per cluster - $N_p \leq N_{p,\rm max}$ $\alpha_P$ Power law index for distribution of periods (non-clustered model) or period scales (clustered models) (\[eq\_P\]), (\[eq\_Pc\]) $P_c \sim f(P_c) \propto {P_c}^{\alpha_P}$ $P_{\rm min}$ Minimum period (days) (\[eq\_P\]), (\[eq\_Pc\]) $P = P'P_c \geq P_{\rm min}$ $P_{\rm max}$ Maximum period (days) (\[eq\_P\]), (\[eq\_Pc\]) $P = P'P_c \leq P_{\rm max}$ $\alpha_{R1}$ Power law index for planetary radius distribution for $R_p \leq R_{p,\rm break}$ (\[eq\_brokenR\]), (\[eq\_brokenRc\]) $R_{p,c} \sim f(R_p) \propto {R_p}^{\alpha_{R1}}, R_p \leq R_{p,\rm break}$ $\alpha_{R2}$ Power law index for planetary radius distribution for $R_p > R_{p,\rm break}$ (\[eq\_brokenR\]), (\[eq\_brokenRc\]) $R_{p,c} \sim f(R_p) \propto {R_p}^{\alpha_{R2}}, R_p > R_{p,\rm break}$ $R_{p,\rm break}$ Break radius for planetary radii ($R_\oplus$) (\[eq\_brokenR\]), (\[eq\_brokenRc\]) $R_{p,\rm min} < R_{p,\rm break} < R_{p,\rm max}$ $R_{p,\rm min}$ Minimum planetary radius ($R_\oplus$) (\[eq\_brokenR\]), (\[eq\_brokenRc\]) $R_p \geq R_{p,\rm min}$ $R_{p,\rm max}$ Maximum planetary radius ($R_\oplus$) (\[eq\_brokenR\]), (\[eq\_brokenRc\]) $R_p \leq R_{p,\rm max}$ $\sigma_P$ Scale factor, per planet in the cluster, for the cluster (unscaled) period distribution (\[eq\_P\_clusP\]) $P' \sim {\rm Lognormal}(0, {N_p}\sigma_P)$ $\sigma_R$ Scale factor for the cluster radius distribution (\[eq\_R\_clusR\]) $R_p \sim {\rm Lognormal}(R_{p,c}, \sigma_R)$ $\sigma_e$ Scale for orbital eccentricities - $e \sim {\rm Rayleigh}(\sigma_e)$ $\Delta_c$ Minimum allowed separation between adjacent planets in mutual Hill radii - $\Delta \geq \Delta_c$ $f_{\sigma_{i,\rm high}}$ Fraction of systems with relatively high mutual inclinations, $\sigma_{i,\rm high}$ (\[eq\_incl\]) - $\sigma_{i,\rm high}$ Scale for mutual inclinations for systems with high mutual inclinations (deg) (\[eq\_incl\]) $i_m \sim {\rm Rayleigh}(\sigma_{i,\rm high})$ $\sigma_{i,\rm low}$ Scale for mutual inclinations for systems with low mutual inclinations (deg) (\[eq\_incl\]) $i_m \sim {\rm Rayleigh}(\sigma_{i,\rm low})$ $N_{\rm attempts}$ Maximum number of attempts for re-sampling periods and period scales for each cluster - - \[tab:params\] Procedure for generating a physical catalogue* {#Procedure} ----------------------------------------------- We outline the procedure for generating an underlying population of planetary systems (a physical catalogue) from the clustered periods and sizes model as follows. The procedure for the other two models are analogous (we explain how to modify the procedure after these steps). 1. Set a number of target stars $N_{\rm stars,sim}$ for each simulated catalogue, typically equal to the number of targets being used as observational constraints. 2. Set a value for each of the model parameters. 3. For each target, assign stellar properties drawn from the stellar catalogue (updated based on Gaia DR2, as described in @H2019, and allowing for the uncertainties in the stellar properties). Stellar properties include radius and mass which affect the observed transit properties, as well as the one-sigma depth function, window function, and contamination that are necessary for the planet detection model [@H2019]. 4. Draw a number of clusters in the system, $N_c \sim {\rm Poisson}(\lambda_c)$. Re-sample until $N_c \leq N_{c,\rm max}$. 5. For each cluster: 1. Draw a number of planets in the cluster from a zero-truncated Poisson (ZTP) distribution, $N_p \sim {\rm ZTP}(\lambda_p)$. Re-sample until $N_p \leq N_{p,\rm max}$. 2. Draw a radius for each planet in the cluster: first, draw a characteristic radius $R_{p,c}$ for the cluster according to Equation (\[eq\_brokenRc\]). If $N_p = 1$, the radius of the cluster’s one planet is simply $R_p = R_{p,c}$. If $N_p > 1$, draw a radius for each of the cluster’s planets, $R_{p,i} \sim {\rm Lognormal}(R_{p,c}, \sigma_R)$, where $i = 1,...,N_p$ (the log is base-$e$). Draw their masses using the mass–radius relation from the non-parametric model in @NWG2018. 3. Draw an orbital eccentricity $e \sim {\rm Rayleigh}(\sigma_e)$ and argument of periastron $\omega \sim {\rm Unif}(0,2\pi)$ for each planet in the cluster. 4. Draw orbital periods for each planet in the cluster: If $N_p = 1$, assign an unscaled period of $P' = 1$. If $N_p > 1$, draw their unscaled periods $P'_i \sim {\rm Lognormal}(0, N_p\sigma_P)$, where $i = 1,...,N_p$ (the log is base-$e$), and sort them in increasing order. Check if $\Delta = [a_{i+1}(1-e_{i+1}) - a_i(1+e_i)]/R_H(i,i+1) \geq \Delta_c$ for all $i$ in the cluster. Re-sample the unscaled periods $P'_i$ until this condition is satisfied or the maximum number of attempts $N_{\rm attempts}$ is reached.[^4] If the case is the latter, discard the cluster. Draw a period scale factor $P_c$ (d) according to Equation (\[eq\_P\]) and multiply each planet’s unscaled periods by the period scale for its parent cluster: $P_i = {P'_i}P_c$. 5. Test for stability: Check if $\Delta \geq \Delta_c$ for all adjacent planet pairs in the entire system, including planets from previously drawn clusters. If the system is identified as likely unstable, redraw $P_c$ for the current cluster until this condition is satisfied or until the maximum number of attempts $N_{\rm attempts}$ is reached. If the latter case occurs, discard the cluster. 6. For each system, draw an inclination angle (relative to the plane of the sky) for the reference plane of the system isotropically, $\cos{i_{\rm ref}} \sim {\rm Unif}(0,1)$. For each system, draw a number $u \sim {\rm Unif}(0,1)$. If $u < f_{\sigma_{i,\rm high}}$, set $\sigma_i = \sigma_{i,\rm high}$; otherwise, set $\sigma_i = \sigma_{i,\rm low}$. Assign an orbit to each planet in the system: 1. Compute the period ratios $\mathcal{P} = P_{i+1}/P_i$ of all adjacent planet pairs in the system. For each planet, check if it is near any MMRs with any adjacent planet by checking if $\mathcal{P}_{\rm mmr} \leq \mathcal{P} \leq 1.05\mathcal{P}_{\rm mmr}$ for any $\mathcal{P}_{\rm mmr}$ in $\{2, 1.5, 4/3, 1.25\}$ (aside from the inner- and outer-most planet, each planet is part of two adjacent planet pairs). If the planet is not near any MMRs with another adjacent planet, draw a mutual inclination angle (relative to the reference plane) $i_m \sim {\rm Rayleigh}(\sigma_i)$; otherwise, draw a mutual inclination angle $i_m \sim {\rm Rayleigh}(\sigma_{i,\rm low})$ regardless of whether $\sigma_{i,\rm high}$ or $\sigma_{i,\rm low}$ was assigned to the non-resonant planets of the system. 2. For each planet’s orbit, draw an angle of ascending node $\Omega \sim {\rm Unif}(0,2\pi)$ in the reference plane. 3. Compute the inclination angle $i$ (relative to the plane of the sky) for each planet’s orbit using the spherical law of cosines, $\cos{(i)} = \cos{(i_{\rm ref})}\cos{(i_m)} + \sin{(i_{\rm ref})}\sin{(i_m)}\cos{(\Omega)}$. The procedure is very similar for the other models. For the clustered periods model, instead of drawing a characteristic radius $R_{p,c}$ for each cluster as in Step 5(b), we simply assign planet radii by drawing them directly from Equation (\[eq\_brokenR\]). In the case of the non-clustered model, Steps 4–5 are simplified to drawing a number of planets $N_p \sim {\rm Poisson}(\lambda_p)$ and directly drawing their periods and radii by Equations (\[eq\_P\]) and (\[eq\_brokenR\]).[^5] Procedure for generating an observed catalogue {#Obs_pipeline} ------------------------------------------------ The underlying population of planetary systems is then used to generate an observed catalogue of exoplanets using a procedure which simulates the observational pipeline employed by . This procedure is the same regardless of the choice of model used for generating the physical catalogue. The full pipeline that is implemented in SysSim as described in @H2018 [@H2019] incorporates many DR25 data products, including tabulated window functions, one sigma depth functions, and a detection efficiency derived from analysing pixel-level transit injections. We briefly summarize the main considerations here. First, we reduce the number of planets by only keeping (in the observed catalogue) planets that transit.[^6] This amounts to requiring that the impact parameter $b$ is less than $1+R_p/R_\star$, $$\begin{aligned} b = \frac{a\cos{i}}{R_\star}\frac{1 - e^2}{1 + e\sin{\omega}} < 1+R_p/R_\star. \label{eq_b}\end{aligned}$$ Next, we select a subset weighted by their detection probabilities. We require at least three transits to be observed by the mission based on window functions provided as part of DR25, as described in @H2019. A planet is labelled as detected if a random number $u \sim {\rm Unif}(0,1)$ is less than the planet’s detection probability (conditioned on it transiting), as calculated by assuming the joint detection and vetting efficiency model described in @H2019. For each of the detected transiting planets, we compute the true transit depths accounting for limb darkening and the true transit durations according to @K2010, equation (15) therein: $$\begin{aligned} t_{\rm dur} &= \frac{P}{\pi}\frac{\rho_c^2}{\sqrt{1-e^2}}\sin^{-1}{\bigg(\frac{\sqrt{1-b^2}}{a_R\rho_c\sin{i}}\bigg)} \label{eq_tdur} \\ &\simeq{} \frac{P}{\pi{a_R}}\frac{\sqrt{1-e^2}}{1+e\sin{\omega}}\sqrt{1-b^2}, \label{eq_tdur_approx} \\ \rho_c &\equiv \frac{1-e^2}{1+e\sin{\omega}},\end{aligned}$$ where $t_{\rm dur}$ is the full-width half-maximum transit duration and $a_R \equiv a/R_\star$ is the semimajor axis in units of the stellar radius. The one-term analytic expression in Equation (\[eq\_tdur\]) only assumes that the planet–star separation does not change during the entire transit event and neglects the planet size [@K2010]. Using this definition grazing transits (i.e., $1\le b<1+R_p/R_\star$) have zero duration. Finally, we add measurement noise to the true period, transit depth, and transit durations and compile the “observed” properties of the detected transiting planets into a simulated observed catalogue. Some of the simulations used for exploring parameter space and training the emulator (see §\[Optimization\]-§\[GP\]) use a simplified noise model where the fractional uncertainties in orbital periods, transit depths, and transit durations are held fixed. However, the final simulations for constraining model parameters and inferring the distributions of observed and physical properties use a diagonal version of the transit noise model of @PR2014, which is based on a trapezoidal transit model and accounts for the finite integration time. Observational comparisons {#Obs_compare} ------------------------- Next, we compare the simulated observed catalogues of exoplanets derived from our models to the actual observed population of exoplanets by the mission. In principle, one could attempt to generate simulated catalogues that precisely match all planet properties to those of the actual catalogue. Of course, the odds of generating such a catalogue are minuscule, even if one could use the perfect model for the underling exoplanet population due to the stochastic nature of the model. Instead, we identify a set of summary statistics that encode the most physically important properties of the distributions of planetary systems observed by in §\[Summary\_stats\]. Next, we define a distance function that quantifies the degree of similarity between the summary statistics for the two catalogues in §\[Distance\]. We describe a procedure for identifying sets of parameters for our physical models that approximately minimize the distance between simulated and observed catalogues (allowing for inevitable shot noise due to the stochastic nature of the model) in §\[Optimization\]. Finally, we constrain the model parameters using approximate Bayesian Computing (ABC). Given the cost of the full model, we train a Gaussian process emulator to approximate the distribution of the distances computed from our forward model in §\[GP\]. We obtain samples from the ABC posterior by drawing trial model parameters based on the GP emulator, computing the distance with the full SysSim forward model, and accepting parameter values that result in a small distance between the simulated observed catalogue and the DR25 catalogue. ### Summary statistics {#Summary_stats} Our procedure for generating observed catalogues yields an observed catalogue with a “measured” orbital period $P$, transit duration $t_{\rm dur}$, and transit depth $\delta$ for each observed planet. A good forward model must result in a similar number of detected planets, as well as a similar number of systems with $m$ detected planets. Additionally, we want our models to reproduce the observed distributions of orbital periods, transit depths and transit durations. Finally, we want our forward models to produce planetary systems that have realistic correlations between the orbital periods and sizes of planets within the same system. Therefore, we compute the following summary statistics for each observed catalogue: 1. the overall rate of observed planets relative to the number of target stars, $f = N_{p,\rm tot}/N_{\rm stars}$, where $N_{p,\rm tot}$ and $N_{\rm stars}$ are the total numbers of observed planets and stars in the catalogue, respectively, 2. the observed multiplicity distribution, $\{N_m\}$, where $N_m$ is the number of systems with $m$ observed planets and $m = 1,2,3,...$, 3. the observed orbital period distribution, $\{P\}$, 4. the observed distribution of period ratios of apparently adjacent planets, $\{\mathcal{P}\}$, where $\mathcal{P} = P_{i+1}/P_i$, 5. the observed transit depth distribution, $\{\delta\}$, 6. the observed distribution of the transit depth ratios of apparently adjacent planets, $\{\delta_{i+1}/\delta_i\}$, 7. the observed transit duration distribution, $\{t_{\rm dur}\}$, 8. the observed distribution of (period-normalized) transit duration ratios of apparently adjacent planets near mean motion resonances, $\{\xi_{\rm res}\}$, and 9. the observed distribution of (period-normalized) transit duration ratios of apparently adjacent planets not near mean motion resonances, $\{\xi_{\rm non-res}\}$. Previously published studies have attempted to match a subset of these summary statistics, but not all at once. Each summary statistic is most sensitive to one or two model parameters, but typically have weaker dependencies on other model parameters. For example, the transit duration distribution is useful for characterizing the orbital eccentricities [@FQV2008]. Moreover, the period-normalized transit duration ratio $\xi$ (also known as the orbital-velocity normalized transit duration ratio; @F2014) is a useful probe of the orbital properties of exoplanets in multitransiting systems. It can be shown that the ratio of the period to the transit duration cubed is proportional to the stellar density, $P/t_{\rm dur}^3 \propto \rho_\star$ [@SM2003]. Thus, for two planets transiting the same star, it is useful to consider the ratio of their period-normalized transit durations. This quantity has been used to test whether two planets really transit the same star; if they do, $\xi$ should be near unity [@Li2012; @F2014]. Following @S2010 and @F2014, we define $\xi$ as: $$\begin{aligned} \xi = \bigg(\frac{t_{\rm dur, in}}{t_{\rm dur, out}}\bigg)\bigg(\frac{P_{\rm out}}{P_{\rm in}}\bigg)^{1/3}. \label{eq_xi}\end{aligned}$$ More importantly for our purposes, the distribution of $\xi$ encodes information about the orbital eccentricities and impact parameters (i.e., inclination angles) of the transiting planets [@F2014; @M2016phd]. It is useful to transform this quantity by taking the logarithm, $\log{\xi}$, so that negative values imply $\xi < 1$ and positive values denote $\xi > 1$. For planets in circular, coplanar orbits around the same star, $\log{\xi}$ must be non-negative (if they both transit at the equator, $b = 0$ and $\xi = 1$; otherwise, the inner planet must have a smaller impact parameter $b$ than the outer planet, and thus a longer period-normalized transit duration, $\xi > 1$). Deviating from coplanarity by increasing the mutual inclination angles results in a lower skewness (i.e., greater symmetry around 0) for the distribution of $\log{\xi}$, because the impact parameters become more randomized as the outer transiting planets need not necessarily have larger values of $b$ than the inner planets. For eccentric orbits, the distribution of $\log{\xi}$ becomes more spread out (i.e., more extreme values of $\xi$ are more common) with increasing eccentricities because the velocities of the planets during transit become more randomized, depending on whether the transits occur near periastron or apastron. Our forward model involves assigning systems to one of two separate mutual inclination scales and assigning planets near MMRs to follow the smaller mutual inclination distribution ($i_m \sim {\rm Rayleigh}(\sigma_{i,\rm low})$; see Step 6(a) in §\[Procedure\]). In order to constrain the inclinations of both populations, we include two summary statistics based on the $\xi$ distribution: $\{\xi_{\rm res}\}$ calculated using only observed planet pairs apparently near a MMR (i.e. based on the observed periods) and $\{\xi_{\rm non-res}\}$ calculated using the remaining planet pairs (i.e., planets pairs not apparently near any MMRs). ### Distance function {#Distance} ABC is a powerful method for performing inference on models where it is impractical to write an explicit likelihood, such as the case for studying multiplanet systems observed by [@H2018]. In ABC, one must define a distance function to quantify how different two catalogues are. This distance is a function of only the summary statistics for each catalogue and goes to zero when the two summary statistics are identical. For simple analytic distributions, one can identify sufficient summary statistics for which one can rigorously prove that the ABC posterior approaches the true posterior as a distance threshold goes to zero. In practice, ABC is most useful for complex problems like ours, for which it is impractical to identify sufficient statistics. In these cases, the ABC posterior will be broader than the true posterior, since some draws may reasonably reproduce the summary statistics, but differ in some way that did not contribute to the distance function. Having identified summary statistics that are both observable and physically significant in §\[Summary\_stats\], we proceed to define a distance function for each summary statistic. For most of our summary statistics, we use either the two-sample Kolmogorov–Smirnov (KS; @K1933 [@S1948]) distance or a rescaled variant of the two-sample Anderson–Darling (AD; @AD1952 [@P1976]) distance. A component distance of zero would mean that the marginal distributions for each of these summary statistics are identical. The use of KS distances rewards distributions that agree best near the median of the marginal distribution, while the use of AD distances is more sensitive to differences in the tails of the marginal distributions. We perform calculations using both, primarily as a sensitivity test. By comparing results using two different distance functions, we can check whether any of our conclusions are sensitive to the choice of distance function. For our total distance, we take a linear combination of individual distance terms for each summary statistic: $$\begin{aligned} \mathcal{D}_W &= \left[ \sum_{i'}{w_{i'} \left\\|\mathcal{D}_{i'}\right\\|}^{\alpha_D} \right]^{1/\alpha_D}, \label{eq_dist_general} $$ where $\mathcal{D}_W$ is the total weighted distance, we set $\alpha_D =1$, and $w_{i'} = 1/\hat{\sigma}(\mathcal{D}_{i'})$ is the weight (defined below) of the $i'^{\rm th}$ distance term, $\mathcal{D}_{i'}$. In principle, one could have chosen another value of $\alpha_D$, corresponding to the Euclidean normal or maximum norm. Either of these would result in a total distance that is more sensitive to the most discrepant summary statistic than our choice of $\alpha_D=1$, which was informed by our preliminary exploratory analyses. We want a total distance function that takes into account each of the observed marginal distributions of the population described in §\[Summary\_stats\], as well as the overall rate of planets per observed star, $f_{\rm Kepler}$. We aim to assign weights to the individual distances that reflect the precision with which they were measured by . Therefore, we specialize Equation \[eq\_dist\_general\] to: $$\begin{aligned} \mathcal{D}_W &= \frac{D_f}{\hat{\sigma}(D_f)} + \frac{D_{\rm mult}}{\hat{\sigma}(D_{\rm mult})} + \sum_{i=1}^{7} \frac{\mathcal{D}_i}{\hat{\sigma}(\mathcal{D}_i)}, \label{eq_dist}\end{aligned}$$ where $D_f$ is the distance between the two rates of observed planets, $D_{\rm mult}$ is the distance of the multiplicity distributions, $\mathcal{D}_i$ is the distance between the distributions of the $i^{\rm th}$ observable, and $\hat{\sigma}(\mathcal{D})$ is the estimated root-mean-square (RMS) of the distance $\mathcal{D}$ given a “perfect” model (i.e., comparisons of simulated catalogues to a reference simulated catalogue with the same model parameters). The indices for the observables run from $i = 1$ to $i = 7$, denoting the following distributions of measured properties in order: period $\{P\}$, period ratio $\{\mathcal{P}\}$, transit depth $\{\delta\}$, transit depth ratio $\{\delta_{i+1}/\delta_i\}$, transit duration $\{t_{\rm dur}\}$, and period-normalized transit duration ratio for near-resonance $\{\xi_{\rm res}\}$ and not-near-resonance $\{\xi_{\rm non-res}\}$ planet pairs. [Rate of Observed Planets:]{} Since the rate of observed planets is a scalar, the distance function is simply, $D_f = \| f_{\rm sim} - f_{\rm Kepler} \|$. [Multiplicity:]{} The observed multiplicities $\left\{N_m\right\}$ is a vector of integers. If each system had the same probability to be observed as an $m$-planet system, then it would be a multinomial distribution. Formally, $\left\{N_m\right\}$ is not drawn from a multinomial since each system has a different set of probabilities for being observed as an $m$-planet system. Nevertheless, the theory of multinomial distribution is useful for identifying an appropriate distance function for the observed multiplicities. The Cressie–Read power divergence (CRPD) statistic [@CR1984] is commonly used in comparing multinomial distributions like the multiplicity distribution and is known to be more robust than other choices like $\chi^2$ for cases like ours where there is a large dynamic range between the values in each category and one of the categories ($O_5$) often has values less than 5.[^7] Therefore, for $D_{\rm mult}$, we adopt the CRPD statistic which is given by: $$\begin{aligned} \rho_{\rm CRPD} = \frac{9}{5}\sum_j O_j \bigg[{\bigg(\frac{O_j}{E_j}\bigg)}^{2/3} - 1\bigg], \label{eq_CRPD}\end{aligned}$$ where $O_j$ is the number of “observed” systems according to one of our models and $E_j$ is the number of expected systems in the $j^{th}$ bin based on the rate of such systems in the data set. The indices $j$ label the bins corresponding to 1, 2, 3, 4, and $5+$ observed planets. Note that by definition, this statistic only sums over bins $j$ for which $E_j \neq 0$ (as it is formally impossible to have a match between the observed and expected distributions when $E_j = 0$ and $O_j > 0$). These caveats are not significant for our analysis since our distance function $D_{\rm mult}$ meets the primary goal of clearly favouring better matches to the Kepler data. [^8] [Continuous Distributions:]{} For remaining $\mathcal{D}_i$ terms, we perform the full analysis with two different distance functions for comparing samples of continuous random variables: the two-sample Kolmogorov–Smirnov (KS) distance, and a rescaled variant of the two-sample Anderson–Darling (AD) distance. For two finite samples of sizes $n$ and $m$, described by empirical distribution functions $F_n(x)$ and $G_m(x)$, the two-sample KS and AD distances are defined as: $$\begin{aligned} \mathcal{D}_{\rm KS} &= {\rm max}\| F_n(x) - G_m(x) \|, \label{eq_KS} \\ \mathcal{D}_{\rm AD} &\equiv \frac{nm}{N} \int_{-\infty}^{\infty} \frac{[F_n(x) - G_m(x)]^2}{H_N(x)[1 - H_N(x)]} dH_N(x), \label{eq_AD_def} \\ &= \frac{1}{nm} \sum_{i=1}^{N-1} \frac{(M_i N - ni)^2}{i(N - i)}, \label{eq_AD}\end{aligned}$$ where $N = n+m$ is the combined sample size, $H_N(x) = [nF_n(x) + mG_m(x)]/N$ is the empirical distribution function for the combined sample, and $M_i$ is the number of observations less than or equal to the $i^{\rm th}$ smallest in the combined sample [@P1976]. The KS distance is simply the maximum absolute difference in the cumulative distribution functions (CDFs) and is thus most sensitive to the difference in the bulk locations of the two distributions. The AD distance on the other hand, more heavily weights the tails of the distributions and is thus more sensitive to differences at the extremes of the distributions. In practice, we find that the standard AD distance given by Equation (\[eq\_AD\]) does not sufficiently account for the differences in the sample sizes $n$ and $m$; in other words, two samples can give a very small AD distance even when $n$ and $m$ are vastly different. In the context of our models, this has the consequence that very small simulated observed catalogues (i.e. with only a handful of observed planets) can still result in small AD distances, even enough to counteract larger distances of $D_f$ for the overall rate of planets per star.[^9] This is clearly undesirable as such models are terrible fits to the observations. In order to counteract this unintended consequence, we modify the AD distance to penalize such models (i.e. those that result in almost no observed planets) by dividing out the $nm/N$ constant in front of the integral, giving: $$\begin{aligned} \mathcal{D}_{\rm AD'} &= \int_{-\infty}^{\infty} \frac{[F_n(x) - G_m(x)]^2}{H_N(x)[1 - H_N(x)]} dH_N(x) \\ &= \frac{N}{(nm)^2} \sum_{i=1}^{N-1} \frac{(M_i N - ni)^2}{i(N - i)}. \label{eq_AD_mod}\end{aligned}$$ For the remainder of the paper, we will refer to our modified AD distance given by Equation (\[eq\_AD\_mod\]) as simply “the AD distance” unless otherwise noted. We refer to the total weighted distance functions involving the KS or rescaled AD distance terms as $\mathcal{D}_{W,\rm KS}$ and $\mathcal{D}_{W,\rm AD'}$, respectively. In order to compute the weights $w_{i'}$ for the individual distance terms, we first generate a simulated observed catalogue using the clustered periods and sizes model which serves as a reference catalogue, and then repeatedly simulate catalogues using the same model (i.e. with identical model parameters) which would be a “perfect” model. The RMS ($\hat{\sigma}(\mathcal{D})$) of each individual distance term and the weights (the inverse of the RMS, $w = 1/\hat{\sigma}$) computed this way are listed in Table \[tab:weights\], while the model parameters used to generate this reference catalogue are given in Table \[tab:param\_ranges\]. The distances for each term are not zero because the simulations involve Monte Carlo noise and only a finite number of planets are drawn per iteration. Indeed, the true catalogue of observed planets is finite in size. Thus, we simulate a reference catalogue that contains a similar number of observed planets given the same number of target stars as the mission. However, for the purposes of computing the weights and optimizing the distance function to find the best-fitting model parameters (see §\[Optimization\]), we wish to reduce stochastic noise in order to improve our power to distinguish between different models. Therefore, we simulate catalogues from the “perfect” model with five times as many stars to give observed catalogues that are five times as large as that of our sample, balancing the desire to reduce stochastic noise and computational time. We compute the weights with 1000 repeated simulations from the “perfect” model. By summing the distances weighted by their variations for a “perfect” model, the total weighted distance emphasizes the distances of observables that are well characterized and prevents any single distance term with high stochastic noise from dominating the total distance function. Also, this implies that each individual distance term included in Equation (\[eq\_dist\]) typically contributes a weighted value of roughly 1 to the total distance in the case of using the perfect model. Distance term --------------------------------------------------- ----------------------------- ---------------------- ----------------------------- ---------------------- $D_f$ $D_{\rm mult}$ $\hat{\sigma}(\mathcal{D})$ $w = 1/\hat{\sigma}$ $\hat{\sigma}(\mathcal{D})$ $w = 1/\hat{\sigma}$ $\mathcal{D}_1$ (for $\{P\}$) 0.0173 58 0.000448 2230 $\mathcal{D}_2$ (for $\{\mathcal{P}\}$) 0.0288 35 0.00113 882 $\mathcal{D}_3$ (for $\{\delta\}$) 0.0179 56 0.000480 2085 $\mathcal{D}_4$ (for $\{\delta_{i+1}/\delta_i\}$) 0.0299 33 0.000911 1098 $\mathcal{D}_5$ (for $\{t_{\rm dur}\}$) 0.0213 47 0.000524 1907 $\mathcal{D}_6$ (for $\{\xi_{\rm res}\}$) 0.0678 15 0.00652 153 $\mathcal{D}_7$ (for $\{\xi_{\rm non-res}\}$) 0.0328 30 0.00121 827 : Table of weights for the individual distance terms as computed from a reference clustered periods and sizes model. The weights $w$ are computed from the root mean squares of the distances, $\hat{\sigma}(\mathcal{D})$, and are shown here as rounded whole numbers simply for guidance purposes. \[tab:weights\] ### The catalogue: stellar and planetary properties {#Data} [c]{}\ \ \ \ \ \ \ [Stellar Catalogue:]{} We couple the DR25 stellar properties catalogue with the results of the second Gaia data release (DR2) [@A2018; @Gaia2018] in order to take advantage of its significantly improved stellar parameters for a large fraction of the target stars. These improvements (of primary interest, the stellar radii) result from the refined parallax and thus distance measurements of the Gaia mission. Furthermore, the Gaia DR2 parallaxes allow for a cleaner sample of main–sequence target stars thanks to a more precise positioning of the targets in color-luminosity space. Additionally, the astrometric information allows for identification of targets likely consisting of multiple stars with comparable luminosity. We use the same target list as in @H2019, who identified a clean sample of FGK main-sequence (M-S) stars for occurrence rate studies. This involved performing a series of cuts on the DR25 stellar table based on measurements reported in the Gaia DR2 and updating the stellar radii with values reported in the Gaia DR2. For full details, see §3.1 of @H2019. In order to minimize sensitivity to uncertainties in stellar radii, impact parameters, and the limb darkening model, we have chosen a distance function based on the distribution of transit depths instead of the distribution of planet radii or planet–star radius ratios, since the measured transit depth does not depend on our knowledge of the stellar radius (like the planet radius) and is better modelled as a Gaussian distribution than the planet–star radius ratio (due to effects of limb darkening and covariance with impact parameter). Nevertheless, the uncertainties in stellar radii still affect our simulations via the transit depths of the simulated planets. [Planet Catalogue:]{} We start from the Data Release 25 (DR25) [@T2018] Objects of Interest (KOI) catalogue as the basis for our study. This table is most suitable for a population study of the exoplanets because it involves uniform vetting of the Q1-Q17 light curves obtained by , and thus does not involve human biases across individual systems. It is derived from processing using the SOC pipeline release 9.3, and involves fully automated dispositioning of the Threshold Crossing Events (TCEs) using the Kepler Robovetter [@T2015; @T2016]. Specifically, the automated procedure involves determining whether the TCEs are transit-like, and if so, tests whether there is any evidence of an eclipsing binary, shift in the in-transit centroid position, or contamination from another source [@M2015; @C2016]. TCEs that pass all the tests are dispositioned as planetary candidates, and their planetary and orbital parameters are computed using a Markov Chain Monte Carlo (MCMC) fitting algorithm [@R2015]. Starting from the DR25 KOI catalogue, we remove planet candidates around targets that were excluded from the stellar catalogue, as described above. Next, we keep only KOIs designated as planet candidates by the Kepler Robovetter. Then, we replace the transit depths and durations in the DR25 catalogue (which were maximum likelihood estimators) with the median values from the MCMC-based posterior samples described in @R2015. We update the planet radii based on the observed transit depths, the updated stellar radii from Gaia DR2, and the limb darkening model from the DR25 stellar catalogue. Finally, we limit the planetary catalogue to include planet candidates with periods between $P_{\rm min}=3$ d and $P_{\rm max}=300$ d (see §\[Methods\]) and with updated planet radii between $R_{p,\rm min} = 0.5 R_\oplus$ and $R_{p,\rm max} = 10 R_\oplus$. These cuts result in a total of 79 935 targets (hereafter denoted by $N_{\rm stars,Kep}$), with 2137 total planet candidates in 1561 systems (with periods between 3 and 300 d and planet radii between 0.5 and 10 $R_\oplus$). Of these, a total of 390 multiplanet systems with 966 planets are included, with the remaining 1171 planets being in single systems. The resulting population of KOIs is shown in Figure \[fig:Kepler\_DR25\], where we plot histograms of the number of planets per system ($N_m$, planet multiplicity), periods ($P$), period ratios of apparently adjacent planet pairs ($P_{i+1}/P_i$), transit durations ($t_{\rm dur}$), period-normalized transit duration ratios ($\xi$), transit depths ($\delta$), and transit depth ratios ($\delta_{i+1}/\delta_i$). These distributions serve as the target observed distributions for our models. In particular, the distributions of the period ratios, transit duration ratios, and transit depth ratios are especially insightful to model because they probe the architectures of planetary systems, yet are insensitive to the stellar parameters. Optimization of the distance function {#Optimization} ------------------------------------- In order to compare our forward models to the observations, we need to find model parameters that result in simulated catalogues that are similar to the DR25 catalogue. Given the complexity and computational expense of the model, we take a multistage approach. First, we use an optimizer to identify good regions of parameter space. Results from the optimizer are then used to train a Gaussian Process emulator (described in §\[GP\]). Finally, we draw samples from prior distributions for model parameters and use rejection sampling to construct the ABC posterior for inference. Here we describe the optimization stage that seeks the set of model parameters that minimize the distance function (for given a model, observed data set, and distance function). This is a challenging problem for several reasons. First, evaluating the distance function is computationally expensive, primarily due to the catalogue simulation. Secondly, the parameter space is large. Our models involve several free parameters: 8, 10, and 11 for the non-clustered, clustered periods, and clustered periods and sizes models, respectively (even after fixing the break radius $R_{p,\rm break}$ and minimum separation for stability $\Delta_c$). There can be correlations or potentially complicated interplay between model parameters. The third and most challenging factor is that the forward model is stochastic due to sampling variance and the finite number of targets. Even for fixed model parameters, the computed distance varies from one realization to the next due to Monte Carlo randomness in drawing target properties, physical properties of planetary systems, and simulated measurement noise. This means that traditional optimization algorithms that assume a deterministic function are not appropriate for our problem. In theory, one could reduce the variance in the distances drawn from our forward model by simulating significantly more targets than observed. While this could be a useful (but expensive) way to find the “best-fitting” model parameters, it is not appropriate for accurately characterizing the uncertainties in the model parameters. The summary statistics for the catalogue include features that might be real or merely the result of small number statistics. In ABC, our forward model should also have variance due to the finite number of targets observed, in order for the ABC posterior to properly weight model parameters accounting for the extent of variations due to the finite number of targets. Therefore, we use the same number of targets as in the catalogue during both the optimization stage (this section) and the emulator stage (as described in the next section). We use an adaptive Differential Evolution optimizer function with radius limited sampling, implemented by “BlackBoxOptim” (<https://github.com/robertfeldt/BlackBoxOptim.jl>). This package includes a general purpose optimization function “bboptimize()” which provides various algorithms, some of which are designed to deal with stochastic noise. For our purposes, we use the optimizer “adaptive\_de\_rand\_1\_bin\_radiuslimited”. This optimizer uses a population-based algorithm that iteratively searches a specified region in parameter space in order to try and minimize a target fitness function without assuming that the function is differentiable. Thus it is well suited to high dimensional problems with stochastic noise. We use the total weighted distance given by Equation (\[eq\_dist\]) as the target fitness function, and leave most of the key model parameters of each model as free parameters with an allowed search range (minimum, maximum) for each parameter. Table \[tab:param\_ranges\] lists the search ranges we used for each free parameter, in each model. We repeat 50 runs of the optimizer on each of our models, each with a different set of initial values for the free parameters drawn randomly (uniformly, while for $\lambda_c$ and $\lambda_p$, uniformly in log) within each of the search ranges for each parameter. For each run, we set the “population size” parameter equal to $4n_{\rm params}$, where $n_{\rm params}$ is the number of free parameters in the model. We simulate $N_{\rm stars,sim} = 79 935$ targets for each generation of the model. In order to enforce the criteria $\sigma_{i,\rm low} \leq \sigma_{i,\rm high}$, we transform these two parameters into two dummy variables $r_1$, $r_2$ using a mapping from the unit square to a triangle [@OFCD2002].[^10] We also set a condition to avoid simulating the model if $\ln{(\lambda_c)} + \ln{(\lambda_p)} > 2.5$, in order to avoid wasting time on models with too many planets. For computational expediency, we stop the optimization process after 5000 model iterations have been computed for each of the 50 runs. We observe from preliminary runs that the best models found during this stage do not improve appreciably with significantly more iterations (e.g. 10,000 iterations). We verify that the local minima found by each of the runs are in a similar region of parameter space. ---------------------------------- ---------------------------- ---------------------------- ---------------------------- ----------- -- Parameter Non- Clustered Clustered periods Ref. clustered periods and sizes catalogue $f_{\sigma_{i,\rm high}}$ $(0, 1)$ $(0, 1)$ $(0, 1)$ 0.5 $\lambda_c$ - $(0.2, 5)$ $(0.2, 5)$ 1.6 $\lambda_p$ $(1, 8)$ $(0.5, 10)$ $(0.5, 10)$ 1.6 $\alpha_P$ $(-2, 2)$ $(-2, 2)$ $(-2, 2)$ $0$ $\alpha_{R1}$ $(-4, 2)$ $(-4, 2)$ $(-4, 2)$ $-1$ $\alpha_{R2}$ $(-6, 0)$ $(-6, 0)$ $(-6, 0)$ $-4$ $\sigma_e$ $(0, 0.1)$ $(0, 0.1)$ $(0, 0.1)$ 0.01 $\sigma_{i,\rm high}$ ($^\circ$) $(0, 90)$ $(0, 90)$ $(0, 90)$ 10 $\sigma_{i,\rm low}$ ($^\circ$) $(0, \sigma_{i,\rm high})$ $(0, \sigma_{i,\rm high})$ $(0, \sigma_{i,\rm high})$ 1 $\sigma_R$ - - $(0, 0.5)$ 0.25 $\sigma_P$ - $(0, 0.3)$ $(0, 0.3)$ 0.15 ---------------------------------- ---------------------------- ---------------------------- ---------------------------- ----------- -- : Table of the free parameters and their search ranges (min, max) explored, of each model. The last column lists the parameter values of the clustered periods and sizes model used to generate the reference catalogue as described in §\[Distance\]. We set $R_{p,\rm break} = 3 R_\oplus$ and $\Delta_c = 8$. \[tab:param\_ranges\] Exploring the parameter space with a Gaussian process emulator {#GP} -------------------------------------------------------------- Computing the full forward model as detailed in §\[Procedure\]-§\[Obs\_pipeline\] is computationally expensive. Generating a physical catalogue with $N_{\rm stars,sim} = N_{\rm stars,Kep} = 79 935$ typically takes $\sim$10–20 s for reasonable model parameters, although this is highly dependent on the mean rates of clusters and planets per cluster $\lambda_c$, $\lambda_p$, and the cluster width in log-period per planet $\sigma_P$ due to repeated draws for stability. (The procedure to simulate an observed catalogue from a pre-existing physical catalogue is faster, taking just a few seconds.) Thus, it would be prohibitively time-consuming to simulate the full model for millions of iterations. Fortunately, for the purpose of finding the region(s) of parameter space that result in observed catalogues similar to the catalogue, we only need to store the total weighted distance for each proposed set of model parameters and do not necessarily need to save all the information in the catalogues. Based on our initial exploratory analyses, the total weighted distance has a single dominant mode for each physical model, which is identified by the optimizer from §\[Optimization\]. In the vicinity of that mode, the mean of distance function varies smoothly, but with considerable variance due to Monte Carlo noise from the finite number of stars. Thus, we can dramatically accelerate the exploration of parameter space by approximating the total weighted distance of our full model using a fast emulator. We adopt the machinery of Gaussian processes (GPs) to train an emulator for our distance function and use the GP to explore the model parameter space in a computationally feasible manner (see the textbook by @RW2006 for an extensive guide to and discussion of GPs, and @O2004 for an introduction to the GP emulator approach). A Gaussian process emulator is a statistical model that aims to mimic a more complicated and expensive function by “emulating” the outputs of the expensive function given the same inputs. A covariance function specifies the correlation between draws from the GP for any pair of input values. For any set of inputs (i.e., model parameters), the GP emulator returns a Gaussian distribution for its prediction of the model output (i.e., distance) that depends on the observed values of the function at a set of training points. While the detailed outputs of our physical (clustered and non-clustered) models are complex, we only require the GP emulator to provide a good approximation to the total weighted distance in a region of parameter space that results in simulated catalogues that are a good match to the data. For each set of model parameters, the GP emulator returns a distribution, which approximates the mean and variance of the distribution of distances that would be returned by the full model (including the effects of the finite number of targets). This allows us to explore the parameter space very quickly in order to efficiently estimate how often realizations with a given set of model parameters would result in a weighted distance less than the maximum acceptable distance for our ABC posterior sample. Here we provide an overview of the GP emulator before providing more specific details below. For the remainder of this paper, we let $\bm{\theta}$ denote the free parameters of our physical models (e.g. $\lambda_p$, $\alpha_P$, etc. as listed in Table \[tab:param\_ranges\], of our non-clustered and clustered models) and $\bm{\phi}$ denote the (hyper) parameters of the GP emulator. We have a distance function $\mathcal{D}_W(\bm{\theta})$ (this is either $\mathcal{D}_{W,\rm KS}(\bm{\theta})$ or $\mathcal{D}_{W,\rm AD'}(\bm{\theta})$) that we evaluated at a large number of points $\bm\theta$ during the optimization stage. As discussed below, we use a subset of points that yielded low distances during the optimization stage as training points for the GP emulator (see §\[secTrainingPoints\]). For a given model and set of training points, we find the “best-fitting” values of the hyperparameters $\bm\phi_{bf}$ which maximize the log likelihood for the GP. Then, we use the emulator with $\bm\phi_{bf}$ to predict the distance $\mathcal{D}_W(\bm{\theta})$ for a much larger number of points in the model parameter space. We draw trial values of $\bm{\theta}$ from a prior and reject draws that result in a predicted distance greater than a distance threshold. The distribution of the accepted points provides a sample from the ABC posterior (see §\[CredibleRegions\]). We compute credible regions for each model parameter $\theta$ from the quantiles of the accepted points. ### Choice of mean and covariance function {#secCovarFunction} GPs are especially useful in our context due to their flexibility in modelling stochastic processes with intractable functional forms. This is exactly the case for our distance function. Mathematically, a GP is described by a prior mean function $m(\bm{x})$ and a covariance (i.e. kernel) function $k(\bm{x},\bm{x'};\bm{\phi})$: $$\begin{aligned} f(\bm{x}) \sim \mathcal{GP}\big(m(\bm{x}), k(\bm{x},\bm{x'};\bm{\phi})\big), \label{eq_GP}\end{aligned}$$ where $f(\bm{x})$ is the function we wish to model ($f(\bm{x}) = \mathcal{D}_W(\bm{\theta})$ for our purposes), and $\bm{\phi}$ are the hyperparameters of the kernel. The prior mean function can be used to model an underlying deterministic process if one is known (such as the periodic motion of a star in a series of radial velocity measurements, e.g. in @Rj2015). For our problem, an underlying functional dependence (i.e. $\mathcal{D}_W$ as a function of the model parameters $\bm{\theta}$) is not known; thus we use a constant prior mean function. When evaluating the emulator near training points, the predicted values will be strongly affected by the training points and only minimally affected by the prior mean. The choice of our mean function will be important when evaluating the emulator far from training points. In an ideal world, we would supply enough training points to adequately characterize the model behavior over the full parameter space, so the emulator would return accurate predictions at any given point. However, this is not practical for the entirety of our 8–11 dimensional parameter space, as the number of training points is limited to just a few thousand, due to the computational cost of training the GP emulator. Therefore, we select training points to be in the vicinity of the best-fitting region as found during optimization. We verified that the distribution of distances returned by the GP emulator is accurate in the region of interest. In order to ensure that the GP emulator consistently returns values well above the distance threshold in other regions, we set the prior GP mean to a large value, i.e., near the high end of the distance for the training points (e.g., $m(\bm{x}) = 75$ for emulating the distance function involving KS distance terms, $\mathcal{D}_{W,\rm KS}$, and $m(\bm{x}) = 250$ for emulating the distance function involving AD distance terms, $\mathcal{D}_{W,\rm AD'}$). Since the prior GP mean is significantly larger than the distance threshold to be used by ABC, the emulator will almost certainly return emulated distances well above distance threshold, when it is given parameter values far from the training points. For the emulator to have a reasonable chance of returning a distance that would be accepted, the model parameters must be near a sizeable number of training points that cause the mean of the prediction to drop well below the prior mean. For the covariance function, we choose the squared exponential kernel with a separate length scale, $\lambda_i$, for each dimension (i.e. model parameter $\theta_i$, with $i = 1,2,...,d$ where we let $d$ denote the number of dimensions/model parameters), $$\begin{aligned} k(\bm{x},\bm{x'};\bm{\phi}) = \sigma_f^2 {\rm exp} \Bigg[-\frac{1}{2} \sum_i \frac{(x_i - {x_i}')^2}{\lambda_i^2} \Bigg], \label{eq_kernel}\end{aligned}$$ where $\bm{\phi} = (\sigma_f, \lambda_1, \lambda_2, ..., \lambda_d)$ are the hyperparameters of the kernel. The hyperparameter $\sigma_f$ controls the strength of correlation between points, and also serves as the standard deviation of the Gaussian prior (i.e. far away from any training data). Intuitively, the length scales $\lambda_i$ govern how far points can be from one another, in each dimension, before they become uncorrelated with each other. We find the best values for the hyperparameters $\bm{\phi}$ by attempting to maximizing the log marginal likelihood: $$\begin{aligned} \log p(\bm{y}\|\bm{X},\bm{\phi}) = -\frac{1}{2}\bm{y}^T k_y^{-1}\bm{y} - \frac{1}{2}\log{\| k_y \|} - \frac{n}{2}\log{2\pi}, \label{eq_log_likelihood}\end{aligned}$$ where $\bm{y} = \bm{\mathcal{D}_W}(\bm{\theta})$ are the function values at the training points, $k_y = k(\bm{\theta},\bm{\theta'};\bm{\phi}) + \sigma_n^2 I$ is the covariance matrix for $\bm{y}$ (and $\sigma_n = \hat{\sigma}(\mathcal{D}_W)$ are the estimated uncertainties at the training points), and $n$ is the total number of training points ($\bm{X} = \bm{\theta}$). The choice of training points is described in §\[secTrainingPoints\]. In practice, optimizing a 8–11D function is computationally expensive. Rather than simultaneously optimizing all of the hyperparameters, we set $\sigma_f = 1$ and fixed values for the relative length scales ${\lambda_i}'$ (listed in Table \[tab:GP\_hparams\]) informed by inspection of the distribution of training points. Then, we maximize the log-marginal likelihood by varying an overall length scale factor $\lambda_{\rm global}$, where $\lambda_i = \lambda_{\rm global}{\lambda_i}'$. Note that for the purposes of training the emulator for our clustered models, we also transform the parameters $\lambda_c$ and $\lambda_p$ into a sum and difference of their log-values, $\ln{(\lambda_c \lambda_p)}$ and $\ln{(\frac{\lambda_p}{\lambda_c})}$, as the region of best values for these transformed parameters are more Gaussian than that of the rates of clusters and planets per cluster parameters themselves. Note. \This is $\ln({\lambda_p})$ for the non-clustered model. \[tab:GP\_hparams\] ### Training points {#secTrainingPoints} For each model, we choose a set of training points $\{\bm{\theta}\}$ for the GP emulator by taking a subset of the model evaluations from the optimization runs as described in §\[Optimization\]. Since we ran 50 individual optimizers with 5000 total model evaluations per run, we have a pool of $2.5\times10^5$ model evaluations scattered around the $d$-dimensional parameter space. We rank order these points by $\mathcal{D}_W$ and choose the top $10^5$ points, keeping every $10^{\rm th}$ point for a total of $10^4$ points. The choice of keeping every $10^{\rm th}$ point instead of all points is due to (1) the computational limits in calculating and inverting the kernel matrix $k(\bm{\theta},\bm{\theta'};\bm{\phi})$, which scales as $n^3$ and thus prevents us from using more than a few thousand training points, and (2) the desire to include some points far enough away from the minimum so that the GP emulator makes reasonable predictions for a wider region of parameter space. The combination of stochastic noise in $\mathcal{D}_W$ due to the Monte Carlo noise of our simulations and keeping the best rank-ordered points would introduce a bias towards smaller than average distances at these points. In order to avoid this bias, we recompute the values $\mathcal{D}_W$ by simulating a new catalogue with the full SysSim model at each of these points. We train the GP emulator using updated values of $\mathcal{D}_W$ for a random subset of 2000 points of the $10^4$ available points. ### Computing credible regions for model parameters {#CredibleRegions} The prior mean function, kernel function, and training points fully define a GP model. For each model and distance function ($\mathcal{D}_{W,\rm KS}$ or $\mathcal{D}_{W,\rm AD'}$) combination[^11] we train a GP emulator and use it to predict the distance function at a large number of points $\bm\theta$ in the $d$-dimensional model parameter space. In order to improve computational efficiency, we draw samples from a reduced range for each parameter (based on inspecting the results of the optimization stage). Effectively, we assume a uniform prior for each model parameter by drawing points uniformly in the $d$-dimensional box, which the bounds for each parameter are specified in Table \[tab:GP\_hparams\]. For each draw of model parameters from the prior distribution, we draw an emulated distance from the GP emulator. Finally, we only accept draws if the emulated distance is less than a distance threshold. We repeat this procedure until we have accumulated $10^4$ draws from the ABC posterior. We adopt distance thresholds of $\mathcal{D}_{W,\rm KS} = 35$ and $\mathcal{D}_{W,\rm AD'} = 140$ for both the clustered models, and a distance threshold of $\mathcal{D}_{W,\rm KS} = 55$ for the non-clustered model. The distance thresholds are less than the medians of the training points, so that the mean and variance of the GP emulators are constrained both in and around the perimeter of the regions of parameter space that are plausibly good fits. Given the stochastic nature of our forward model, even a perfect model would typically return a distance of $\sim 9 \times \sqrt{5}\simeq 20$, i.e., the number of component distance functions in our total weighted distance times the square root of the ratio of the number of target stars used when generating weights to the number of target stars used during the inference step). The best distances found during the optimization stage ($\sim 20$ for the clustered models and $\sim 45$ for the non-clustered model, when using the KS distance) set the smallest distance thresholds we could have chosen. The use of larger distance thresholds, means that credible regions based on our ABC posterior samples will be somewhat larger than the ideal posterior if we had been able to use a smaller distance threshold. In practice, the size of the ABC posterior credible intervals appear to be shrinking only slowly as we decrease the distance threshold further. Since the credible regions can only shrink with a smaller distance threshold, the credible regions that we report are conservative, i.e., larger than if we would have obtained with infinite computational resources. In Figure \[fig:clustered\_P\_R\_corner\_KS\], we display a “corner” plot (plotted using `corner.py`; @Fm2016) showing the ABC posterior based on a sample of $10^4$ for the clustered periods and sizes model with the KS distance function. Similar figures for the non-clustered, clustered periods, and alternative MMR inclinations models are available in supplementary online material. Similarly, supplemental Figure \[fig:clustered\_P\_R\_corner\_AD\] shows the analogous plot using the AD distance function. We interpret and discuss the meaning of these results in detail in the next section. Results {#Model_params} ======= Parameter ------------------------------------------ ---------------------------------------------------------------- --------------------------- -------------------------------------------------------------- --------------------------- --------------------------- ----------------------------------------------------------------- --------------------------- --------------------------- --------------------------- --------------------------- Fig. \[fig:non\_clustered\_model\], \[fig:models\_underlying\] Best-fit KS Fig. \[fig:clustered\_P\_model\], \[fig:models\_underlying\] Best-fit KS Best-fit AD Fig. \[fig:clustered\_P\_R\_model\], \[fig:models\_underlying\] Best-fit KS Best-fit AD Best-fit KS Best-fit AD \[5pt\] $f_{\sigma_{i,\rm high}}$ 0.03 $0.03_{-0.02}^{+0.02}$ 0.4 $0.42_{-0.10}^{+0.10}$ $0.41_{-0.13}^{+0.11}$ 0.4 $0.42_{-0.07}^{+0.08}$ $0.40_{-0.12}^{+0.11}$ $0.32_{-0.06}^{+0.06}$ $0.31_{-0.10}^{+0.10}$ \[5pt\] $\ln{(\lambda_c)}$ - - $-0.51$ $-0.39_{-0.31}^{+0.32}$ $-0.28_{-0.27}^{+0.32}$ $-0.22$ $-0.10_{-0.35}^{+0.44}$ $0.24_{-0.40}^{+0.33}$ $0.04_{-0.45}^{+0.46}$ $0.35_{-0.41}^{+0.32}$ \[5pt\] $\lambda_c$ - - $0.6$ $0.68_{-0.18}^{+0.25}$ $0.76_{-0.18}^{+0.28}$ $0.8$ $0.90_{-0.26}^{+0.50}$ $1.27_{-0.42}^{+0.50}$ $1.04_{-0.38}^{+0.61}$ $1.42_{-0.48}^{+0.54}$ \[5pt\] $\ln{(\lambda_p)}$ $1.25$ $1.26_{-0.15}^{+0.16}$ $1.44 $ $1.50_{-0.50}^{+0.30}$ $1.27_{-0.50}^{+0.38}$ $1.31$ $1.35_{-0.44}^{+0.36}$ $0.73_{-0.56}^{+0.60}$ $1.05_{-0.34}^{+0.39}$ $0.56_{-0.43}^{+0.50}$ \[5pt\] $\lambda_p$ $3.5$ $3.53_{-0.50}^{+0.61}$ $4.2$ $4.48_{-1.76}^{+1.57}$ $3.56_{-1.40}^{+1.65}$ $3.7$ $3.86_{-1.38}^{+1.67}$ $2.08_{-0.89}^{+1.70}$ $2.86_{-0.82}^{+1.36}$ $1.75_{-0.61}^{+1.14}$ \[5pt\] $\alpha_P$ $-0.1$ $-0.13_{-0.07}^{+0.10}$ 0.1 $0.13_{-0.48}^{+0.73}$ $0.01_{-0.29}^{+0.71}$ 0.4 $0.40_{-0.56}^{+0.64}$ $0.07_{-0.45}^{+0.66}$ $0.19_{-0.57}^{+0.69}$ $-0.05_{-0.33}^{+0.37}$ \[5pt\] $\alpha_{R1}$ $-1.0$ $-1.08_{-0.20}^{+0.31}$ $-0.8$ $-0.75_{-0.46}^{+0.46}$ $-0.85_{-0.26}^{+0.26}$ $-1.0$ $-1.02_{-0.70}^{+0.64}$ $-1.27_{-0.25}^{+0.26}$ $-1.28_{-0.57}^{+0.72}$ $-1.32_{-0.29}^{+0.30}$ \[5pt\] $\alpha_{R2}$ $-5.3$ $-5.30_{-0.40}^{+0.53}$ $-4.9$ $-4.91_{-0.62}^{+0.96}$ $-5.49_{-0.33}^{+0.42}$ $-4.4$ $-4.41_{-0.79}^{+1.36}$ $-5.08_{-0.54}^{+0.71}$ $-4.50_{-0.70}^{+1.03}$ $-4.81_{-0.71}^{+0.79}$ \[5pt\] $\sigma_e$ 0.003 $0.003_{-0.002}^{+0.003}$ 0.01 $0.014_{-0.007}^{+0.009}$ $0.010_{-0.006}^{+0.008}$ 0.02 $0.020_{-0.010}^{+0.014}$ $0.014_{-0.008}^{+0.010}$ $0.019_{-0.009}^{+0.011}$ $0.014_{-0.008}^{+0.010}$ \[5pt\] $\sigma_{i,\rm high}$ ($^\circ$) 60 $62_{-50}^{+16}$ 50 $49_{-21}^{+22}$ $49_{-25}^{+25}$ 50 $48_{-17}^{+17}$ $49_{-25}^{+23}$ $49_{-18}^{+19}$ $42_{-20}^{+26}$ \[5pt\] $\sigma_{i,\rm low}$ ($^\circ$) 0.3 $0.31_{-0.10}^{+0.10}$ 1.1 $1.13_{-0.33}^{+0.38}$ $1.16_{-0.26}^{+0.31}$ 1.4 $1.40_{-0.39}^{+0.54}$ $1.29_{-0.32}^{+0.35}$ $1.38_{-0.38}^{+0.43}$ $1.29_{-0.31}^{+0.34}$ \[5pt\] $\sigma_R$ - - - - - 0.3 $0.31_{-0.07}^{+0.07}$ $0.32_{-0.07}^{+0.07}$ $0.30_{-0.07}^{+0.07}$ $0.30_{-0.06}^{+0.06}$ \[5pt\] $\sigma_P$ - - 0.2 $0.20_{-0.04}^{+0.03}$ $0.17_{-0.03}^{+0.04}$ 0.2 $0.21_{-0.04}^{+0.04}$ $0.20_{-0.04}^{+0.04}$ $0.21_{-0.04}^{+0.04}$ $0.18_{-0.03}^{+0.04}$ \[tab:param\_fits\] In Table \[tab:param\_fits\], we report the best-fitting values of the free model parameters for each of the models. We discuss the results and the effect of each of the free model parameters on the simulated observed population in this section. Comparison of clustered and non-clustered models {#secCompClusteredNonClustered} ------------------------------------------------ First, we briefly summarize how our models compare with each other, before reporting the results for each parameter of each model in detail. Our clustered models are clearly preferred over the non-clustered model, as evidenced by the significantly smaller best-fitting weighted distances as described in §\[GP\]. The improvement in the clustered models can be traced to improvements in the component distances for the multiplicities, period ratios, transit duration ratios for planet pairs not near resonance, and to a lesser extent periods and transit durations. Going from clustered periods to clustered periods and radii results in some component distance improving (based on period ratios and radius ratios), but the change in total distance is more modest. All three models are able to match the overall observed rate of planets (via our analysis involving KS distances; the analysis involving AD distances fails for the non-clustered model). The period and radius (power/broken-power law) distributions are broadly consistent across all the models and distances, suggesting a shallowly increasing number of planets in log-period, a roughly flat distribution of planet sizes below the break radius, and a sharply decreasing occurrence of planets above the break radius. Both clustered models prefer: (1) a sizable fraction ($\sim 40\%$) of highly mutually inclined systems and (2) the remaining majority of systems with small mutual inclinations of $\sim 1^\circ$. These findings are consistent across analyses using either the KS and AD distances, corroborating previous studies (e.g., @Mu2018). All three models imply that most planets have small orbital eccentricities. Finally, the clustered models suggest that both periods and planet sizes are highly clustered, with consistent results for the cluster widths in log-period and radius using both KS and AD distances. Our results are also relatively unchanged when considering the alternative MMR inclinations model, i.e. our fully clustered model without the reduction of mutual inclinations for the near-MMR planets. All the parameters with the exception of $f_{\sigma_{i,\rm high}}$ remain consistent within statistical uncertainties. While $f_{\sigma_{i,\rm high}}$ is somewhat reduced ($\sim 0.32$), the true fraction of highly mutually inclined planets is similar, since in our standard clustered models the $f_{\sigma_{i,\rm high}} \sim 40\%$ of highly mutually inclined systems includes planets near an MMR that have low mutual inclinations. Overall, the rate of planets per star, the period and radius distributions, the eccentricity and mutual inclination scales, and the period and radius cluster widths all remain the same. Rates of clusters per system and planets per cluster\ ($\lambda_c$, $\lambda_p$) {#Params_rates} ----------------------------------------------------- In our two clustered models, the parameters $\lambda_c$ and $\lambda_p$ parameterize the mean numbers of clusters and planets per cluster, respectively, before rejection-sampling and any truncation. The true mean rate of clusters in our simulations is somewhat less than $\lambda_c$, while the true mean rate of planets per cluster is typically greater than $\lambda_p$ for small values, and always greater than one, due to the draws from the zero-truncated Poisson distribution. These parameters largely control the overall rates of planets and how they are distributed within and between clusters. It is clear that with increasing $\lambda_c$ and $\lambda_p$, the overall number of observed planets increases, due to there being intrinsically more planets around each star. This effect is perhaps slightly more sensitive to $\lambda_c$, likely because increasing $\lambda_p$ is more likely to result in rejected systems which slows the increase in the number of planets with increasing $\lambda_p$ (since it is harder to fit more planets into the same cluster than to increase the number of clusters, due to stability). The product of these two parameters is most directly constrained by the total rate of planets per star (i.e. the distance term $D_f$ in equation \[eq\_dist\]). In our clustered periods and sizes model, we find that $\lambda_c \simeq 0.90_{-0.26}^{+0.50}$ indicating that $\sim 80\%$ of systems with planets have one cluster. The number of planets per cluster peaks at four with $\lambda_p \simeq 3.86_{-1.38}^{+1.67}$. The combination of these gives a typical $\lambda_c \lambda_p \sim 3.5$ planets per star. As seen in Table \[tab:param\_fits\], using the AD distances leads to more clusters per star (1.27) and fewer planets per cluster (2.08), though the 1$\sigma$ credible regions overlap. The overall rate of planets per star ($\sim 2.6$) is similar. For the clustered periods model, we find similar values for the rates of clusters and planets per cluster using both KS and AD distances: $\lambda_c \simeq 0.68_{-0.18}^{+0.25}$ and $\lambda_p \simeq 4.48_{-1.76}^{+1.57}$ using KS distances. Again, while the KS analysis results in slightly more planets per cluster and fewer clusters than the AD analysis, the difference is not significant given the credible regions. These results are very similar to those of the clustered periods and sizes model using KS distances. For our non-clustered model, the number of planets per system $N_p$ is described by a single parameter, $N_p \sim {\rm Poisson}(\lambda_p)$. We find that using KS distances, $\lambda_p \simeq 3.53_{-0.50}^{+0.61}$, implying a median value of $\sim 3.5$ for the mean number of planets per system which is comparable to the value in our best-fitting clustered models. As discussed before, we do not recover meaningful results for this model using AD distances. Returning to our fully clustered model, the values of $\lambda_c$ and $\lambda_p$ are somewhat shifted in our alternative MMR inclinations model. There are slightly more clusters per system and fewer planets within each cluster, using both KS and AD distances, although these differences are within the statistical uncertainties. The overall rate of planets per system effectively remains the same. We conclude that the observations robustly constrain the mean number of planets per system and the results are insensitive to the choice of forward model or distance function. While there is more variance across models and distance functions for the inferred ratio of planets per cluster to clusters per system, the differences are less than the statistical uncertainties for three of the four cases considered (i.e., clustered periods and fully clustered models, each with KS and AD distances). Period distribution ($\alpha_P$) {#Params_periods} -------------------------------- For all our models, the overall period distribution is described by a single power law between 3 and 300 d with index $\alpha_P$. We allowed $\alpha_P$ to vary between $-2$ and 2 during the optimization process (note that $\alpha_P = -1$ corresponds to a flat distribution in log-period). All three of our models result in an increasing occurrence of planets with log-period, strongly disfavouring a flat distribution ($\alpha_P = -1$). The estimates of $\alpha_P$ are consistent within statistical uncertainties across all of our clustered models. However, assuming a non-clustered model results in both a shallower slope and significantly smaller uncertainties on $\alpha_P$. The clustered periods and sizes model results in $\alpha_P \simeq 0.40_{-0.56}^{+0.64}$ using KS distances, and favours a shallower slope when using AD distances, $\alpha_P \simeq 0.07_{-0.45}^{+0.66}$. The results for the clustered periods models are very similar for the KS and AD analyses, giving $\alpha_P \simeq 0.13_{-0.48}^{+0.73}$ and $\alpha_P \simeq 0.01_{-0.29}^{+0.71}$, respectively. The constraint on $\alpha_P$ appears to be much tighter for the non-clustered model, with $\alpha_P \simeq -0.13_{-0.07}^{+0.10}$. We conclude that one should be cautious about overinterpreting measurements of $\alpha_P$ that have assumed orbital periods are drawn independently for planets within the same system. Radius distribution ($\alpha_{R1}$, $\alpha_{R2}$) {#Params_radii} -------------------------------------------------- We allowed $\alpha_{R1}$ and $\alpha_{R2}$, the power law indices for the broken power-law as given in Equation (\[eq\_brokenR\]), to vary from $-4$ to 2 and $-6$ to 0, respectively, for all our models (Table \[tab:param\_ranges\]). These two parameters along with $R_{p,\rm break}$ control the overall distribution of planetary radii; we fix $R_{p,\rm break} = 3 R_\oplus$ here, so $\alpha_{R1}$ controls the distribution of Earth-sized planets between 0.5 and 3$R_\oplus$ and $\alpha_{R2}$ controls the distribution of larger planets between 3 and 10$R_\oplus$. The continuity of $p(R_p)$ induces a correlation between $\alpha_{R1}$ and $\alpha_{R2}$. In our distance function, the most direct constraint on the radius distribution comes from fitting the transit depth and transit depth ratio distributions. The sizes of planets also affect the observed period ratio distribution, since small planets can be packed closer to each other but may be harder to detect. Formally, the radius distribution must be inferred simultaneously with the period distribution and other model parameters [@Y2011]. Nevertheless, it appears that one could infer the overall radius distribution reasonably well even without simultaneously modelling the planetary architectures. There is a clear need for a broken power law for the radius distribution, given the joint posterior for $\alpha_{R1}$ and $\alpha_{R2}$ strongly excludes equal values. Moreover, the radius distribution falls much more quickly above $R_{p,\rm break}$ than below it; $\alpha_{R2} \simeq -4.41_{-0.79}^{+1.36}$ while $\alpha_{R1} \simeq -1.02_{-0.70}^{+0.64}$ for our clustered periods and sizes model with KS distances (and slightly steeper values using AD distances). Similarly, the clustered periods and non-clustered models also prefer $\alpha_{R2} \lesssim -5$ and $\alpha_{R1} \simeq -1$. This is expected given that the population does not have a high occurrence of larger, Jupiter-sized planets. The differences in the constraints on $\alpha_{R1}$ and $\alpha_{R2}$ across models and distances functions are smaller than the statistical uncertainties. The 68.3% credible interval for $\alpha_{R1}$ includes a flat distribution of $\alpha_{R1} = -1$ for almost all our models and distance functions, consistent with previous results [e.g., @H2012; @PMH2013b; @Mu2018]. While each of our analyses yields similar constraints on $\alpha_{R1}$, we caution that our radius model cannot include a local minima, i.e., the radius valley in the underlying distribution [@F2017; @vE2017; @H2019]. Therefore, we leave the generalization of our model to allow for a more flexible radius distribution for future studies. The eccentricity distribution ($\sigma_e$) {#Params_ecc} ------------------------------------------ The orbital eccentricity distribution primarily affects the transit duration and duration ratio distributions, with a slight influence on the period ratio distribution due to the stability criteria. In all our models, the eccentricities are very low; for the clustered models, both KS and AD analyses result in $\sigma_e \simeq 0.01$–0.02 for the Rayleigh distribution. The eccentricity scale preferred by our non-clustered model is even smaller, centred around $\sigma_e \sim 0.003$. We speculate that this may be caused by the poor fit to the period ratio distribution (see §\[Non\_clustered\_model\]): this model is unable to produce enough observed planets with small period ratios, and thus desires near circular orbits in order to space planets as close as possible while remaining stable. As such, the fits to the transit duration and $\xi$ distributions are also worse in the non-clustered model (see supplemental Figure \[fig:distances\]). Our results are consistent with those of several previous studies based on properties of systems with multiple transiting planets. @F2014 used an early sample of 899 planets in multitransiting systems and fit to the $\xi$ distribution to constrain both eccentricity and mutual inclination dispersions. They found a best fit of $\sigma_e \simeq 0.032$ (also assuming a Rayleigh distribution of $e$), but caution that a wide range of $\sigma_e$ were plausible since the $\xi$ distribution is not as sensitive to $e$ as it is to mutual inclinations. @WL2013, who found $e \sim 0.01$, and @HL2014, who reported an rms of $\sigma_e \simeq 0.018_{-0.004}^{+0.005}$, both performed analyses of TTVs to arrive at these results. @vEA2015 assumed a Rayleigh distribution and arrived at a slightly larger value of $\sigma_e = 0.049 \pm 0.013$. More recently, @W2019 examined the dynamical stability of multiplanet systems and the period ratio distribution, and found an upper limit of $\sigma_e = 0.03$ (also using the Rayleigh distribution). Previous studies that included systems with only a single detected transiting planet have led to more varied conclusions. For example, @M2011 assumed a Rayleigh distribution of eccentricities and performed an analysis comparing normalized transit durations (durations assuming eccentric orbits divided by durations for circular orbits) with a small subset (104 planets) of the initial discoveries. They found a larger mean eccentricity in the range of 0.1–0.25 for their population, but cautioned that the measurement uncertainty and potential systematic effects in the host stellar densities could bias the derived eccentricities. This concern has now been mitigated thanks to improved stellar properties and the ability to select a cleaner sample of host stars thanks to Gaia DR2. @SDC2015 investigated the eccentricity distribution of a small sample of mostly single, large, short-period planets using both transit and occultation ($h = e\cos{\omega}$ and $k = e\sin{\omega}$) measurements. When they assumed a one-component Gaussian distribution for $h$ and $k$ (equivalent to a Rayleigh distribution for $e$), they found a value of $\sigma_e = 0.081_{-0.003}^{+0.014}$. However, they also show that adopting a two-component Gaussian (very similar to a mixture of two Rayleigh distributions) is favoured. That model results in about 90% of planets with $\sigma_e = 0.01_{-0.002}^{+0.014}$ and 10% with $\sigma_e = 0.22_{-0.026}^{+0.1}$. Some studies have found evidence suggesting that the eccentricity distribution may differ between systems with a single transiting planet and multiple transiting planets. @M2011 found a difference in the transit duration distribution (normalized by the duration estimated for a transit over the diameter of the star with a circular orbit of the same period) that was statistically significant, but cautioned about the potential impact of uncertainties in the stellar parameters. @X2016 also found a difference in the eccentricity distribution for single transiting planets ($\bar{e} \approx 0.3$) and systems with multiple detected transiting planets ($\bar{e} = 0.04_{-0.04}^{+0.03}$), based on a larger catalogue and stellar properties updated by LAMOST. A similar finding is reported by @vE2019, who find $\sigma_e = 0.32 \pm 0.06$ for singles and $\sigma_e = 0.083_{-0.020}^{+0.015}$ for multis, using a smaller sample of planets but with much more precisely measured stellar properties, thanks to asteroseismology. @M2019 updated stellar properties based on Gaia* and spectra from the California–Kepler Survey for a large sample of planets. They found that singles have higher mean eccentricities, with $\bar{e} = 0.167_{-0.008}^{+0.013}$ for singles and $\bar{e} = 0.036 \pm 0.012$ for multis. @M2019 proceeded to fit a mixture model for the single transiting planet systems, that included $\simeq 69\%$ of singles being drawn from a low-eccentricity population with $\sigma_{e,\mathrm{low}}<0.05$ and the remainder coming from a high-eccentricity population with $\sigma_{e,\mathrm{high}}>0.3$. Formally, our low-inclination population contributes a small number of systems with a single transiting planet detected. However, we find that this fraction is so small that the transit duration distribution of singles will remain a robust probe of the properties of the eccentricity distribution of the highly excited population. Several of these studies also find correlations of eccentricities with planet radii (e.g., @HL2014 [@SDC2015; @DLC2016]) and stellar metallicity (e.g., @SDC2015 [@M2019]). In this study, we adopted a single Rayleigh distribution of $e$ for simplicity and to avoid increasing the number of model parameters. Thus, at some level, our results are likely averaging over a small population with larger eccentricities and/or any correlations between eccentricity and planetary or stellar properties. However, we expect this to have a relatively small effect on our results for the eccentricity distribution, since the $\xi$ distribution is only observable for systems with multiple transiting planets. While the transit duration distribution is affected by planets regardless of multiplicity, this provides a weaker constraint than the $\xi$ distribution as almost half of the known transiting planets are in systems with multiple transiting planets. We encourage future work to explore the impact of allowing the high-inclination population to have a broader distribution of eccentricities and potentially correlations with stellar properties. The mutual inclination distribution ($f_{\sigma_{i,\rm high}}$, $\sigma_{i,\rm high}$, $\sigma_{i,\rm low}$) {#Params_incl} ------------------------------------------------------------------------------------------------------------ The mutual inclination angles between planets have pronounced effects on the multiplicity distribution and the period-normalized transit duration ratio, as discussed in §\[Summary\_stats\]. As listed in Table \[tab:param\_ranges\], we allow the fraction of systems with broad mutual inclinations $f_{\sigma_{i,\rm high}}$ to vary in the entire range between 0 and 1, and the mutual inclination scales $\sigma_{i,\rm high}$ and $\sigma_{i,\rm low}$ for the Rayleigh distributions to vary between $0^\circ$ (coplanar orbits) and $90^\circ$ (effectively isotropic), with the constraint that $\sigma_{i,\rm low} \leq \sigma_{i,\rm high}$. As discussed in §\[Incl\], $\sigma_{i,\rm low}$ also governs the mutual inclinations for planets near MMRs. For both of our clustered models, the fraction of systems with $\sigma_{i,\rm high}$ is close to 40%; $f_{\sigma_{i,\rm high}} \simeq 0.42_{-0.07}^{+0.08}$ ($0.40_{-0.12}^{+0.11}$) using KS (AD) distances for the clustered periods and sizes model. Results with the clustered periods model are very similar. The broad mutual inclination scale $\sigma_{i,\rm high}$ is clearly greater than $\sim 10^\circ$, but poorly unconstrained at significantly larger inclinations, since these systems contribute almost solely to the number of observed single-transiting systems. Our result of $f_{\sigma_{i,\rm high}} \sim 0.4$ is very similar to the value of $f_{\rm iso} \simeq 0.38$ reported in @Mu2018 and consistent with previous studies on the dichotomy (e.g. @BJ2016 for M dwarfs). For the rest of the systems, the mutual inclinations between planets are only a few degrees, $\sigma_{i,\rm low} \simeq 1.40_{-0.39}^{+0.54}$ ($1.29_{-0.32}^{+0.35}$) degrees using KS (AD) distances. Additionally, $\simeq 30\%$ of the planets from the $\simeq 40\%$ of systems labelled high mutual inclination, were actually assigned a low inclination to the system’s reference plane, due to being near a first-order MMR with an adjacent planet. While $\sigma_{i,\rm low}$ is small, strictly coplanar orbits are strongly prohibited primarily due to the $\log{\xi}$ distribution. We note that in the clustered periods and sizes model, if we do not include the lowering of mutual inclinations for planets near an MMR, $f_{\sigma_{i,\rm high}}$ decreases to $\sim 0.32$. This is expected because more systems need to host multiple planets with relatively small mutual inclinations in order to make up for the lost contribution to the multitransiting (2+ observed planets) systems from the near-MMR planets. Thus, this value is comparable to the true fraction of systems with highly mutually inclined planets in our usual clustered models. Notably, the values of $\sigma_{i,\rm high}$ and $\sigma_{i,\rm low}$ do not change in this model, highlighting their robustness to our treatment of the near-MMR planets. The non-clustered model is unable to adequately match the observed multiplicity distribution as shown in Figure \[fig:non\_clustered\_model\] and discussed later in §\[Non\_clustered\_model\]. This leads the non-clustered model to favour very small values of $f_{\sigma_{i,\rm high}}$ and $\sigma_{i,\rm low}$. Since only a few percent of systems have high inclinations, the value of $\sigma_{i,\rm high}$ is only weakly constrained. This demonstrates the importance of adopting a clustered model, not just for inferring the distribution of orbital periods, but also for inference about the distribution of mutual inclinations. One reason for this is that the correlation between the true multiplicity and inclination distribution noticed by earlier authors [e.g., @Li2011b; @TD2012] is partially broken when including the period ratio distribution, which sets the typical spacing of planets and affects the observed multiplicity in a non-clustered model. Our results for the mutual inclination distribution are in strong agreement with many previous studies. The first study on the mutual inclinations of the exoplanets was done by @Li2011b, who also simulated planetary systems in order to match the observed multiplicities of the first few months of data. They found an excess of single transiting systems, and by fitting the multis alone with a mixture of 3- and 4-planet systems found a best fit of $\sigma_i = 2^\circ$ using a Rayleigh distribution of mutual inclinations. @TD2012 combined both and RV survey results to estimate mean inclinations in the range 0–$5^\circ$; a similar result of $\leq 5^\circ$ was reported by @J2012, who tried to match the number of single, double, and triple transiting systems by assuming all systems are intrinsic triples. @F2012 also used and RV (HARPS) data to constrain $\sigma_i \leq 1^\circ$ assuming a Rayleigh distribution, although they included only planets larger than $2 R_\oplus$ and noted that larger inclinations of $\sim 5^\circ$ are possible if the mass–radius relationship is more extreme. @FM2012 explored combinations of bounded-uniform and zero-truncated Poisson distributions of multiplicities, with Rayleigh and Rayleigh of Rayleigh distributions of mutual inclinations, in order to fit the observed multiplicity and $\xi$ distributions. They found the best fits with the bounded-uniform and either Rayleigh or Rayleigh of Rayleigh distributions with $\sigma = 1^\circ$ (or $\sigma_\sigma = 1^\circ$). @F2014 also used the observed multiplicity and $\xi$ distributions to constrain mutual inclinations in the range 1–$2.2^\circ$. A similar range of 0.3–$2.2^\circ$ (for mean mutual inclinations) was reported by @X2016. Thus, our results from the clustered models corroborate these previous studies and also serve as the tightest constraints on the mutual inclination distribution to date, with consistent results via two independent analyses involving KS and AD distances. @Z2018 took a different approach to modelling the mutual inclination distribution compared to these previous works, by parametrizing the mutual inclination dispersion as a function of planet multiplicity $k$, with $\sigma_{i,k} \propto k^\alpha$. By matching to both the observed transiting multiplicity and TTV-inferred multiplicity distributions, they found a steep inverse relation, $-4 \lesssim \alpha < -2$, and a normalization of $\sigma_{i,5} = 0.8^\circ$. The results of our clustered models, considering only the low mutual inclination population, are generally consistent with their mutual inclination dispersion for high-multiplicity systems. The period and radius clustering of planets ($\sigma_P$, $\sigma_R$) {#Params_clustering} -------------------------------------------------------------------- The parameters $\sigma_P$ and $\sigma_R$ refer to the scale factors for the cluster period and radius distributions, respectively. While $\sigma_P$ is effectively the width of the cluster in log-period, per planet in the cluster, $\sigma_R$ is the fixed width of the cluster in log-radius regardless of the number of planets (recall Equations \[eq\_P\_clusP\] and \[eq\_R\_clusR\]). These parameters are largely constrained by the period ratio and transit depth (i.e. radius) ratio distributions, respectively. We find that $\sigma_P \simeq 0.20_{-0.04}^{+0.03}$ ($0.17_{-0.03}^{+0.04}$) using KS (AD) distances for our model with just clustered periods and very similar values for the model that includes clustering in radii, where $\sigma_R \simeq 0.31_{-0.07}^{+0.07}$ using KS ($0.32_{-0.07}^{+0.07}$ using AD) distances. We find that extremely small values of $\sigma_P \lesssim 0.05$ are highly excluded, likely due to the effects of our stability criteria which causes some clusters to be discarded after many attempts to fit planets in and thus significantly decreasing the overall rate of observed planets. These values suggest that the periods and radii of planets are highly clustered, as smaller values imply a greater degree of clustering (i.e. smaller cluster widths). For example, if we assume $\sigma_P \simeq 0.2$ and consider a four-planet cluster centred around $P_c = 20$ d, this implies that $\simeq 68\%$ of draws ($\sim 3$ planets) to populate this cluster will have periods between 9 and 45 d. Similarly, $\simeq 68\%$ of planets drawn for a cluster of super-Earth sized planets around $2 R_\oplus$ will have radii between $1.48$ and $2.7 R_\oplus$, assuming $\sigma_R \simeq 0.3$. Discussion {#Discussion} ========== Comparison of models for reproducing the observed data {#Model_comparison_observed} ------------------------------------------------------ As outlined in §\[Methods\], our forward modelling procedure produces a simulated observed catalogue of transiting exoplanets, including measured planet multiplicities per system, observed properties for each planet (period, transit duration, and transit depth), and the ratios of period, transit depth, and transit duration for pairs of apparently adjacent planets. In this section, we display simulated observed catalogues from each of our three models, non-clustered, clustered periods, and clustered periods and sizes, in Figures \[fig:non\_clustered\_model\]–\[fig:clustered\_P\_R\_model\] respectively. For each model, we generate one catalogue with $N_{\rm stars,sim} = 5N_{\rm stars,Kep} = 399 675$ in order to reduce stochastic noise and provide smoother marginal distributions for comparison. The model parameters $\bm{\theta}$ used to generate these catalogues are listed in Table \[tab:param\_fits\]. In order to compute the 16 and 84 percentiles for each bin, we also generate 100 additional catalogues with model parameters drawn from our ABC posterior and that pass our (KS) distance thresholds after generating an observed catalogue using the full forward model. Appendix Figure \[fig:distances\] shows how each of these catalogues compare to the DR25 catalogue in terms of the total and individual (weighted and unweighted) distance terms. We discuss how each of our models compare to the data below. -- -- -- -- ### Non-clustered model {#Non_clustered_model} In Figure \[fig:non\_clustered\_model\], we show the results of our non-clustered model (the model parameters for the population plotted in black are listed in Table \[tab:param\_fits\]). The panels from top to bottom display the marginal distribution of each summary statistic while the left-hand and right-hand sides show the empirical PDFs and CDFs, respectively. The DR25 population (as shown in Figure \[fig:Kepler\_DR25\]) is overplotted as grey histograms for comparison. The lower right corners of each CDF panel display the relevant distance terms between the model and the population. The non-clustered model is able to produce a population of simulated observed exoplanets that resembles the DR25 population in some regards. The overall rate of observed planets per target can be matched almost exactly (the $D_f$ term is almost zero). The bulk of the period distribution appears well modelled by a simple power law with $\alpha_P \sim -0.1$ (i.e. shallowly increasing in log-period, see §\[Model\_comparison\_underlying\] for the underlying distribution), although there appears to be a slight excess of planets at long periods. The transit depth distribution is reasonably well modelled with planet radii drawn independently from a broken power law, although the observed distribution of (log) $\delta$ is slightly left-skewed (i.e., mode is at larger value than the median) whereas our model produces a symmetric distribution. The transit duration distribution is also reasonably well modelled, although we produce an excess of grazing transits with zero duration. The distribution is also somewhat shifted to longer durations due to the extremely small eccentricities. The $\xi$ distribution of our simulated catalogue appears to be slightly more skewed towards values $\log{\xi} > 0$, suggesting that the mutual inclinations of the planets are larger than $\sigma_{i,\rm low} \sim 0.3^\circ$. However, there are several major, clear differences between the non-clustered model and the planet catalogues. First, the planet multiplicity distribution is poorly modelled and produces far too few multiplanet systems. This shortcoming of the model is persistent for many realizations near the best-fitting parameters. Secondly, the period ratio distribution is clearly not well captured by this model, motivating our use of clustered models. While most of the adjacent planet pairs have small period ratios and the distribution falls rapidly above $\mathcal{P} \sim 3$, the non-clustered model produces a much more gradual decline in the tail of the distribution. A stability criteria of $\Delta_c = 8$ appears to sculpt the inner edge of the period ratio distribution in a similar manner as seen in the data. Finally, the transit depth ratio distribution is inadequately described by our non-clustered model; the model produces a wide, symmetric distribution around unity whereas the observed radii ratios are highly peaked around unity. In summary, the non-clustered model described in the previous section performed well in some respects, but there are at least three obvious differences between the simulated observed and DR25 populations: the observed multiplicity, period ratio, and transit depth ratio distributions. These shortcomings motivate our clustered models. -- -- -- -- ### Clustered periods model {#Clustered_P} Before we consider our full clustered periods and sizes model, we first explore the effect of adding just a clustering of orbital periods, i.e., our clustered periods model (the planetary radii are drawn in the same manner as in our non-clustered model). In Figure \[fig:clustered\_P\_model\], we show the results of this model with the DR25 population (again, the model parameters used are listed in Table \[tab:param\_fits\]). Our clustered periods model is a significantly better description of the data than our non-clustered model. Notably, the observed multiplicity and period ratio distributions bear a much closer resemblance to the data, and have significantly reduced distances (CRPD distance for the observed multiplicity counts; KS and AD distances for the period ratio distribution). This is perhaps not surprising given that the main difference between this model and the previous model is the introduction of period clusters. The clustering in periods allows for some systems to contain planets that are closely packed more often than in the case of independently drawn periods, while also allowing for a more gradually falling tail to large orbital period ratios. The clustering in periods also provide a more flexible model for the numbers of planets per system since there are two parameters ($\lambda_c$ and $\lambda_p$) instead of just one ($\lambda_p$ of the non-clustered model). The fits to the period and transit duration distributions are also slightly improved, while the transit duration ratio distribution (for planets not near resonances; $\{\xi_{\rm non-res}\}$) is significantly improved, in both KS and AD distances. This suggests that the distribution of orbital eccentricities is not as low as what is implied by the non-clustered model, and is instead well described by a Rayleigh distribution with $\sigma_e \simeq 0.01$. These results also imply that the mutual inclination distribution is well described by $\sigma_{i,\rm low} \simeq 1.1^\circ$ for $\sim 60\%$ of systems and planets near resonances. This also appears to produce noticeable peaks near the first-order MMRs in the period ratio distribution (most apparent near $\mathcal{P} \simeq 1.5$ and to a smaller extent at $\mathcal{P} \simeq 2$), although we only observe a marginal improvement to the KS and AD distances for the $\xi_{\rm res}$ distribution. There is no significant change in the fit to the transit depth distribution. The transit depth ratio distribution, on the other hand, is modelled just as poorly as and perhaps even worse than before. These results show that clustered periods alone cannot reproduce the highly peaked nature of the planet radii ratios observed in the multiplanet systems. The clustering in orbital periods is able to substantially improve the modelling of the period ratio distribution, but the transit depth ratio distribution remains poorly modelled by both the non-clustered and clustered periods models. The transit depth ratios are more highly peaked around one in the actual population than in the simulated catalogues of these models (see Figures \[fig:non\_clustered\_model\] and \[fig:clustered\_P\_model\]), motivating our next investigation of the clustered periods and sizes model. -- -- -- -- ### Clustered periods and sizes model {#Clustered_P_R} Several previous studies have found that the radii of exoplanets observed by around a single target star are correlated, causing planets to be more similar in size within each system compared to between systems [@C2013; @M2017; @W2018a]. However, most recently these results have been called into question by @Z2019, who show that the clustered radii can be reproduced by a re-sampling of signal-to-noise ratios and conclude that these correlations are largely due to detection biases. Thus, we extend the clustering point process to include clustering of the planetary radii (and implicitly planet masses). This is the full implementation of our clustered periods and sizes model as detailed in §\[Procedure\]. We plot the results of this model in Figure \[fig:clustered\_P\_R\_model\] and list the model parameters in Table \[tab:param\_fits\]. As is the case in our previous clustered periods model, the distances for the observed multiplicity, period, and period ratio distributions are small, and are significantly improved compared to the non-clustered model. There are no significant differences in the distances (KS and AD) for these observables between the two clustered models. A similar conclusion can be drawn for the transit durations and transit duration ratios; while both clustering in periods and clustering in both periods and sizes fit these marginal distributions equally well, both clustered models provide substantially better descriptions of $\{t_{\rm dur}\}$ and $\{\xi_{\rm non-res}\}$ and perhaps slightly better fits to $\{\xi_{\rm res}\}$ as compared to the non-clustered model. However, the transit depth distribution appears to be modelled slightly worse than in the previous models. Also, while the transit depth ratios qualitatively appear to be better fit with this model, there is almost no improvement in the KS or AD distance. A closer examination reveals that there is a slight offset in the CDFs, due to the observed distribution of transit depth ratios of adjacent planet pairs being asymmetric (in log). This suggests that while models with non-clustered planet sizes fail to predict the highly peaked nature of the transit depth ratio distribution, a model with clustered sizes still does not adequately reproduce this property; there are additional features shaping the distribution of adjacent-planet radii ratios that require a more complex model to explain. In particular, modelling the depth and depth ratio distributions would likely be improved by allowing for a valley in the planet radius distribution (perhaps due to photoevaporation, core heating, or some other process, see §\[Introduction\]) that is not included in our model. We discuss our speculations for how future models can be generalized to better match these features in §\[secFuture\]. Differences in the underlying populations (physical catalogues) between the models {#Model_comparison_underlying} ------------------------------------------------------------------------------------ -- -- -- -- -- -- -- -- A major benefit of forward modelling the mission is that we can directly analyze the predictions of our models for the true, underlying exoplanetary systems. While the model parameters describe the underlying system properties including the rate of planets and clusters per system, the period and radius distributions, and the orbital architectures, the physical catalogues generated by our models can be directly examined to see Monte Carlo realizations of populations of planetary systems. In Figures \[fig:models\_underlying\] and \[fig:clustered\_P\_R\_underlying\], we plot the underlying distributions of planetary systems as generated by one instance of each model and 100 realizations of the clustered periods and sizes model, respectively. The physical catalogues shown here are the same simulated populations generating the observed catalogues as shown in Figures \[fig:non\_clustered\_model\]–\[fig:clustered\_P\_R\_model\]. We make the following observations: - Planets and clusters per system: There is a huge difference in the way planets are distributed across different systems between our clustered and non-clustered models. The non-clustered model (dotted red) produces very few ($\lesssim 5\%$) systems with no planets due to the single Poisson distribution describing the number of planets, $N_p \sim {\rm Poisson}(\lambda_p)$. In this model, the most common planetary system consists of three planets (between 3 and 300 d and 0.5–10$R_\oplus$). The clustered models (dashed green and solid blue) produce many more systems with no planets due to draws of zero-cluster systems from $N_c \sim {\rm Poisson}(\lambda_c)$ (since the number of planets per cluster is a zero-truncated Poisson distribution, $N_p \sim {\rm ZTP}(\lambda_p)$ in the clustered models, and $\lambda_c$ tends to be small). For systems with planets, the multiplicity tends to be higher; the most common planetary system is a four-planet system. While there is a clear preference for clustered planetary systems, the number of actual clusters tends to be small. For planet-harbouring stars, $\sim$80% of systems have just a single cluster consisting of three or four planets (over the range of periods considered, 3–300 d). Thus, assuming that planetary systems are highly clustered, our model predicts a large fraction of stars with no planets larger than 0.5 $R_\oplus$ with periods between 3 and 300 d, while the remaining systems to have many planets in this range. We discuss the occurrence rates of planets in more detail in §\[Planet\_rates\]. - Orbital period distribution: All three models exclude flat or falling power laws in log-period (i.e, $\alpha_P \leq -1$). Instead, our assumed power law for the period distribution leads to a shallowly increasing planet occurrence as a function of log-period. There is a pile-up of planets near $P_{\rm max} = 300$ d, likely due to edge effects of our rejection-sampling algorithm; this pile-up is most severe in our non-clustered model. However, this has a minimal effect on our observed catalogues, as the probability of transits decreases with longer periods. Additionally, the period distributions produced by our clustered models appear to be “rounded” and deviate from a power law, due to the draws of period clusters. - Spacings of adjacent planets: Given the same stability criteria of $\Delta \geq \Delta_c = 8$, the clustered models produce slightly narrower separation ($\Delta$) and underlying period ratio distributions, suggesting that planetary systems are more tightly spaced than one would infer from a model with periods drawn independently. All three models produce distributions of separations in mutual Hill radii that are sharply truncated at $\Delta_c$, suggesting that many period or period scale draws are attempted more than once. The distribution of $\Delta$ begins to fall out around $20$, which is similar to the findings of @W2018a although they assumed a much simpler mass–radius relationship [@WM2014] and only analysed the spacings of observed planets. The underlying period ratio distributions are highly peaked around $\sim 1.5$. - Planet radius distribution: As listed in Table \[tab:param\_fits\], we chose $\alpha_{R1} = -1, -0.8$, and $-1$ for these realizations of the non-clustered, clustered periods, and clustered periods and sizes models, respectively. As expected, the radius distribution is flat for small sizes up to the break radius ($R_{p,\rm break} = 3 R_\oplus$) and sharply falls above it for this non-clustered model. However, the catalogue generated from the clustered periods and sizes model exhibits a rounded distribution for small radii and the break radius is not as clear (despite also having set $\alpha_{R1} = -1$), due to the clustering in radii which tends to smooth out the distribution. The catalogue shown here from the clustered periods model has a radius distribution that slightly increases up towards the break radius, although the credible regions for $\alpha_{R1}$ are consistent with a flat distribution for all three models. - Planet radius ratios: The clustered periods and sizes (solid blue) model produces a radius ratio distribution strongly peaked around unity. The effect of this peak increases for decreasing values of $\sigma_R$ due to more highly clustered planet radii. The radius ratio distributions of the other two models reflect what the distribution looks like if planet pairs with radii drawn independently from a broken power law are randomly paired. Implications for the fraction of stars with planets {#Planet_rates} --------------------------------------------------- Since we also fit our simulated results to match the overall rate of observed planets per surveyed (FGK) star in the mission, we can directly estimate the true fraction of stars with planets and the mean rate of planets per star, given various ranges of period and planetary radius. We use the same 100 simulated catalogues generated for Figure \[fig:clustered\_P\_R\_underlying\] to calculate the following planet occurrence rates (and their 16% and 84% quantiles) from our clustered periods and sizes model. [Planets between 0.5 and 10$R_\oplus$:]{} The fraction of stars with planets (FSWP) with 3 d $<P<$ 300 d and $0.5R_\oplus < R_p < 10 R_\oplus$ is $0.56_{-0.15}^{+0.18}$. Roughly half of all FGK stars have at least one planet in this range based on our clustered periods and sizes model. In contrast, our non-clustered model gives a remarkably different result, producing just a few percent ($\sim 3\%$) of stars with no planets (see the dotted red curve in Figure \[fig:models\_underlying\]). This emphasizes the importance of adopting a clustered model; while both our clustered and non-clustered models can fit the overall rate of planets equally well, the distribution of these planets across systems is very different. A similar result was drawn by @WSS2012, who studied the intrinsic multiplicity and mutual inclination distributions and concluded that either planet occurrence is correlated between planets in the same system and/or some stars are significantly more planet-rich than others. Returning to our clustered periods and sizes model, the mean number of planets per star is $2.28_{-0.53}^{+0.94}$. This rate is somewhat less than the credible region values of $\lambda_c \lambda_p$, due to the rejection-sampling procedure. Excluding the stars with no planets (in this range of periods and sizes), the mean number of such planets per system rises to $4.26_{-0.37}^{+0.52}$. @Z2018 find that the fraction of Sun-like stars with at least one “-like” planet is $30\pm3\%$ based on multiplicity and rate of TTVs, although they define “-like” to only include planets larger than $1R_\oplus$ (and periods less than 400 d). In contrast, @ZCH2019 find that $0.72^{+0.04}_{-0.03}$ of stars have at least one planet (with orbital periods between 0.5 and 500 d and radii between 0.5 and $16 R_\oplus$), while modelling just the observed periods, radii, and multiplicity. @ZCH2019 report $8.4\pm0.31$ planets per star hosting a planetary system (but excluding stars that host a single transiting planet with $R_p\ge 6.7R_\oplus$). This implies that most stars have at least seven planets over the range they consider and only 5.5% of planets have one to four such planets. Our finding of $0.56_{-0.15}^{+0.18}$ FGK dwarf stars hosting at least one planet is greater than that suggested by @Z2018 and less than that suggested by @ZCH2019. While each study considers a slightly different range of orbital periods and sizes, we attribute the bulk of the differences relative to our model as due to our use of a clustered model for the distribution of orbital periods within a planetary system. When we applied a non-clustered model, we find $\sim 97\%$ of stars host at least one planet. Switching to a model which accounts for clustering in orbital period and planet size (or just orbital periods) dramatically increases the fraction of stars with no planets in our model. Similarly the mean number of planets per star with at least one planet (within the range of periods and sizes considered) increases from $3.41_{-0.39}^{+0.31}$ to $4.26_{-0.37}^{+0.52}$ when switching from a non-clustered to a clustered model (see Figure \[fig:clustered\_P\_R\_underlying\]). This demonstrates the importance of allowing for clustering when performing inference on the fraction of stars with planets or mean number of planet per stars with at least one planet. Another important way in which our model differs from those of @Z2018 and @ZCH2019 is that we allow for the mutual inclination distribution to have both a low and high inclination component. Since we treat the fraction of systems from the high inclination component as a free parameter, our simulations could have resulted in $f_{\sigma_{i,\mathrm{high}}}$ consistent with zero. Instead, values of zero were strongly excluded in both clustered models, regardless of which distance function was chosen. Interestingly, the simulations using a non-clustered model did result in small values of $f_{\sigma_{i,\mathrm{high}}}$ and did not exclude $f_{\sigma_{i,\mathrm{high}}}=0$. This finding also underscores the importance of allowing for clustering when performing inference on the mutual inclination distribution of planetary systems. [Earth-sized planets:]{} We provide results for the rate of Earth-sized planets (here defined to be 0.75–1.25$R_\oplus$) around FGK stars, using our clustered periods and sizes model. The fraction of stars with at least one Earth-sized planet (between $3-300$ days) is $0.42_{-0.16}^{+0.24}$. The mean number of these planets per star is $0.85_{-0.30}^{+0.51}$. Considering systems with at least one planet (with sizes between 0.5 and 10$R_\oplus$ and periods between 3 and 300 d), the probability that a planetary system contains an Earth-sized planet (in the same period range) is $0.75 \pm 0.14$. If we focus on planetary systems with at least one planet (not necessarily Earth-size), then the mean number of Earth-sized planets per system is $1.62_{-0.38}^{+0.23}$ (both with periods 3–300 d). Broadly, about half of stars have an Earth-sized planet and most inner planetary systems have Earth-sized planets. Due to the increasing frequency of planets at longer periods, most of these Earth-like planets are in 100–300 d orbits. These conclusions follow directly from combining the inferred distributions for orbital period and planet radius (that are consistent with previous results) with the inferred fraction of stars with planetary systems. dichotomy {#Dichotomy} ---------- Previous studies have shown that many simple parametrized models with a single population significantly underpredict the number of systems with only one transiting planet. Since this implies two populations, this result is known as the dichotomy. This paper focuses on models that allow for two populations for the mutual inclinations, each with similar distributions for the remaining properties of the system (i.e., no explicit dichotomy in the intrinsic distributions of multiplicity, orbital period, planet size, or eccentricity). A model with a single inclination population is nested inside our model. Therefore, we can examine whether our posterior distribution for the fraction of stars with a high mutual inclination planetary has significant weight near zero. For each of the clustered models and distance functions we considered, we find that the predicted fraction of high-inclination systems ($40 \pm 10$%) is inconsistent with zero, supporting the theorized dichotomy. Another proposed solution to the dichotomy is an alternative model consisting of one population of high-multiplicity systems (i.e., very similar to our low-inclination population) and a second population of systems with a single planet (both within 3 d $<P<$ 300 d and $0.5 R_\oplus < R_p < 10 R_\oplus$)[^12]. As shown in §\[Planet\_rates\], the average number of planets per planetary system (i.e., star with at least one planet) is $\left<{\rm NPPS}\right> \sim 4.3$ and the fraction of stars with at least one planet is ${\rm FSWP} \sim 0.56$ in our fully clustered model. Therefore, the average star coming from the high-inclination population in our model actually hosts multiple detectable transiting planets, greatly increasing the fraction of viewing angles for which the star would be perceived as hosting a single transiting planet. Using values from our fully clustered model, simply replacing all the stars with at least one planet ($\simeq 0.56$) and assigned to the high mutual inclination population ($f_{\sigma_{i,\rm high}} = 0.42$) with stars hosting a single planet, would require ${\rm FSWP} \times f_{\sigma_{i,\rm high}} \times \left<{\rm NPPS}\right> \simeq 101\%$ of stars to host a single planet in order to provide the same number of planets (or the same number of systems with a single transiting planet), leaving no stars available to host multiple planet systems. We can reject this model for the dichotomy, since the only way to fit enough systems with one transiting planet around stars that are not already spoken for is to have more than one (highly inclined) planet per star. Before completely discarding an abundance of single-planet systems as an explanation for the dichotomy, it is worth considering this argument in more detail. In the fully clustered model, $\simeq 29\%$ of planets around stars nominally in the high-inclination population were actually assigned a low inclination (relative to the system mid-plane), due to the planet being near a first-order MMR with another planet in the same system. This results in a subset of the planet pairs from the high-inclination population contributing to the number of systems observed to have two transiting planets and detracting from the fraction of viewing geometries that the system would be observed as a single planet system. We consider the possibility of replacing the planets with truly high-inclinations with a population of intrinsic single-planet systems by considering the fraction of stars hosting each of three types of planetary systems: (1) the fraction of stars with planetary systems (i.e. with at least one planet) initially assigned to the low-inclination population is ${\rm FSWP} \times (1 - f_{\sigma_{i,\rm high}}) \simeq 32\%$; (2) the fraction of stars hosting planets in a system initially assigned to the high-inclination population, but containing planets assigned a small mutual inclination due to being near a first-order MMR, is ${\rm FSWP} \times f_{\sigma_{i,\rm high}} \times 0.29 \simeq 7\%$; and (3) the fraction of stars required to host exactly a single planet. The final category would need to be $101\% \times (1 - 0.29) \simeq 71\%$ of stars in order to fully replace the planets that remain assigned to high mutual inclinations. Thus, in the fully clustered model, the sum of the three populations comes to $\simeq 110\%$ of target stars. The net effect of accounting for the reassignment of high inclination planets to the low inclinations is to reduce, but not eliminate the tension in having enough stars to accommodate the three populations. Repeating this calculation over 100 realizations of our fully clustered model results in the total number of stars required to be 110% $\pm$ 23%, exceeding unity 63 times. Using the clustered periods model, the sum exceeds unity for 12 of 100 realizations. A similar conclusion is drawn by applying the same argument to our fully clustered model without the special treatment of planets near an MMR. In this model, the fraction of high mutual-inclination systems is somewhat lower ($f_{\sigma_{i,\rm high}} \sim 0.31$, as discussed in §\[Params\_incl\]), but none of the planets in these systems are reassigned to a lower mutual inclination. The fraction of stars with planets is higher, ${\rm FSWP} \sim 0.67$. The mean number of planets per planetary system is similar, $\left<{\rm NPPS}\right> \sim 4.2$, meaning that we would require ${\rm FSWP} \times f_{\sigma_{i,\rm high}} \times \left<{\rm NPPS}\right> \simeq 87\%$ of stars to host a single planet. Again, this does not leave enough stars to host the planetary systems from the low mutual-inclination population, for which we require ${\rm FSWP} \times (1 - f_{\sigma_{i,\rm high}}) \simeq 46\%$ of stars. Repeating this calculation results in the same conclusion for 73 of 100 realizations. Therefore, while explaining the dichotomy with a population of single planet systems cannot be strictly excluded by our work, it is disfavoured and strongly constrained. Even in the case where we cannot formally exclude it, the fraction of stars with planets is quite high, requiring extremely efficient planet formation. Since our model does not include planets with $P<3$ d and/or $R_p<0.5R_\oplus$, including stars with such planets could further increase the number of stars required to host multiple planet systems. Thus, explaining the dichotomy with a population of excess singles leaves very little, if any, room for stars harbouring no planets. Similarly, we could consider an alternative model in which the excess of stars observed as a single transit system is due to a population of highly excited two-planet systems. In this scenario, the fraction of stars hosting two highly inclined planets would need to be $\simeq 36\%$, similar to the fraction of stars with planetary systems drawn from the low mutual inclination population. While this rate is significantly higher than the ${\rm FSWP} \times f_{\sigma_{i,\rm high}} \simeq 24\%$ of stars in our standard fully clustered model, there are sufficient stars that the dichotomy could be explained by population of highly inclined systems with two (or more) planets. However, in order to have enough stars for such model, we had to relax the assumption that the number of cluster per stars is drawn from a Poisson distribution. Instead, we must allow for the number of zero planet systems to be decoupled from $\lambda_c$, the rate of clusters per star, which is inferred from the observed multiple planet systems. Using our clustered periods model leads to similar conclusions. Of course, these conclusions must be interpreted in the context of the parametrization of our model, e.g., the use of Poisson and Rayleigh distributions. Altogether, our results suggest that the dichotomy is most easily explained by two populations with different inclination distributions, as opposed to different multiplicity distributions. @ZCH2019 proposed that the dichotomy could be explained by a combination of geometric transit probability, a distribution of mutual inclinations, and a detection efficiency model that accounts for the order of planet detection by the pipeline. For their star and planet sample, they report that the observed ratio of double to single transit systems is 4% larger than that predicted by their model when ignoring how the detection efficiency depends on the order of detection. When using their model for how the detection efficiency differs for subsequent planets, this difference is reduced to 2%. However, @ZCH2019 did not attempt to make use of information contained in the distributions of orbital period ratios, transit depth ratios, or transit duration ratios of planets in a single planetary system, as was done in this study. We find that these provide valuable information for characterizing the distribution of planetary architectures and evidence for the dichotomy. It is also useful to compare the planet detection efficiency models of the two studies. @ZCH2019 fit separate detection efficiency curves for the first planet (technically the first threshold crossing event, TCE) to be detected by the pipeline and for subsequent planets (or other TCEs). For planets with orbital periods less than 200 d (i.e., the vast majority of detections in our sample), they find that the expected MES needed for a 50% probability of detecting the planet increases from $\simeq 8.3$ (for the first planet) to $\simeq 9.0$ (for the second planet). The detection efficiency rises rapidly with expected MES, so the more important effect is that their model for the detection probability of a transiting planet with high signal-to-noise asymptotes to 0.982 (for the first planet), but only to 0.928 for subsequent planets in the same system. In contrast, our study used a detailed planet detection and vetting efficiency model based on @H2019. It was also derived by fitting a model to the results of the @C2017 pixel-level transit injection study. One key difference is that it includes both the probability that a planet was detected by the pipeline and that it was labelled as a planet by the robovetter. A second important difference is that the detection model of @H2019 includes an explicit dependence on the number of transits observed (and an implicit dependence on the orbital period). Most detected planets have 38 or more observed transits, in which case the detection efficiency approaches 0.945 for high SNR. For planets with 5–36 transits, our model asymptotes at 0.834 to 0.890 for high SNR, depending on the number of transits observed. Therefore, it appears that accounting for dependence on the number of transits (or orbital period) is more important than the order of detection. This strengthens our findings that the data provide evidence for a high-inclination population of planetary systems to explain the observed dichotomy. Features in period ratio distribution {#secPratios} ------------------------------------- [Planets near a mean-motion resonance (MMR):]{} We find that our clustered models can produce the spikes in the period ratio distribution near first-order MMRs similar to those observed by by assigning planets from the high inclination population to the low inclination population when near an MMR.[^13] Since we do not explicitly consider models with an explicit excess near MMRs, we do not exclude that possibility. Nevertheless, our results do provide insight into how significant an excess would be needed to create such spikes. While $f_{\sigma_{i,\rm high}}$ describes the fraction of systems initially assigned to the high-inclination population, a fraction of the planets in these systems are still assigned a low mutual inclinations if they are near a (first-order) MMR. Since the distribution of period ratios for all planets is smooth (see Figures \[fig:models\_underlying\] and \[fig:clustered\_P\_R\_underlying\]), the spikes in the observed period ratio distribution come solely from this population. In our fully clustered model, we find that the fraction of all planets near an MMR (as defined in §\[Incl\]) with another planet is $f_{\rm mmr} = 0.29 \pm 0.04$. Thus, reproducing the spikes near MMRs in the observed period ratio distribution (without invoking a change in the distribution of mutual inclinations) would require a population of ${\rm FSWP} \times f_{\rm mmr} \times f_{\sigma_{i,\rm high}} \times \left<{\rm NPPS}\right> = 0.29_{-0.09}^{+0.11}$ planets per star that are near a first-order MMR. This can be compared to a rate of ${\rm FSWP} \times (1-f_{\sigma_{i,\rm high}}) \times \left<{\rm NPPS}\right> \simeq 1.31_{-0.35}^{+0.55}$ planets per star that are responsible for the “background” population of multiple planet systems, excluding spikes in the period ratio distribution. We caution that these results are based on our distance function that includes a term for the overall observed period ratio distribution, but does not know about the dynamical significance of MMRs. Using a distance function that explicitly considers the behavior of the period-ratio distribution near MMRs would be expected to provide a stronger constraint. Conversely, at least part of the observed spikes in the period ratio distribution could be due to shifting the period ratios of planet pairs from slightly less to slightly more than the value at resonance. The period ratio distribution also provides additional constraints on proposals to explain the dichotomy. Our parametrization of the mutual inclination distribution involves contributions to the period ratio distribution from both high and low mutual inclination populations. While a four-planet system has three pairs of adjacent planets, a two-planet system has only one planet pair. Therefore, reducing the mean multiplicity of the high-inclination population from greater than four to two (see §\[Dichotomy\]) would reduce the amplitude of spikes in the period ratio distribution by more than a factor of three. If the high-inclination population were to have a significantly lower mean number of planets per planetary system than the low-inclination population, then explaining the near-resonant peaks in the distribution of period ratios would likely require invoking an actual excess of planetary pairs near resonance. [Are planets evenly spaced?]{} @W2018a and @Z2019 explored the period ratios of inner–outer pairs for observed three-planet chains. @W2018a found a strong correlation between $\mathcal{P}_{\rm in} = P_{j+1}/P_{j}$ and $\mathcal{P}_{\rm out} = P_{j+2}/P_{j+1}$ (where planets are sorted so $P_j$’s are increasing) for small period ratios and concluded that planets in 3+ systems tend to be uniformly spaced. However, @Z2019 suggested that this correlation is largely due to a combination of detection biases and imposing an arbitrary upper limit on the period ratio ($\mathcal{P} < 4$ as used in @W2018a). We test these assertions by also considering $\mathcal{P}_{\rm in}$ and $\mathcal{P}_{\rm out}$ as well as the their ratio ($\mathcal{P}_{\rm out}/\mathcal{P}_{\rm in}$). Our model does not explicitly enforce any correlation between $\mathcal{P}_{\rm in}$ and $\mathcal{P}_{\rm out}$ either in the generation of systems or in the distance function. Any observed correlation will be due to a combination of observational selection effects and the “natural” peak in $\mathcal{P}$ given the process for drawing orbital periods. In Figure \[fig:pratio\_ratio\], we plot the observed distribution of the ratio of period ratios, $\mathcal{P}_{\rm out}/\mathcal{P}_{\rm in}$, for our fully clustered and non-clustered models along with that of the data. The intrinsic distributions of $\mathcal{P}_{\rm out}/\mathcal{P}_{\rm in}$ (not shown) are peaked around one in each of our models. However, the observed distribution of $\mathcal{P}_{\rm out}/\mathcal{P}_{\rm in}$ is much more highly peaked than that observed in the non-clustered model (dotted red). In contrast, the clustered models produce a peak in $\mathcal{P}_{\rm out}/\mathcal{P}_{\rm in}$ around unity, but it is slightly wider than the peak in the observed distribution of $\mathcal{P}_{\rm out}/\mathcal{P}_{\rm in}$ (solid blue). Furthermore, the distribution of the ratios of period ratios (shaded grey) is even more peaked than the predictions of our clustered models. We conclude that there is a significant correlation in the period ratios of neighbouring planet pairs in actual planetary systems. This strengthens the conclusion of @W2018a. Future research {#secFuture} --------------- Future studies can build on our model and codes to incorporate additional physical processes which are not included in our current models into our forward model. One could consider alternative parametric distributions for the distributions of intrinsic multiplicity, periods, radii, eccentricities, and mutual inclinations. For example, one possible extension would be to incorporate a more sophisticated model for the distribution of planet radii (e.g., incorporating period dependence and/or allowing for a local minimum in the frequency as a function of planet size). @ZCH2019 showed that adding a radius-valley resulted in only a modest effect on the inferred parameters for the radius and period distribution. Nevertheless, allowing for a more flexible model for the radius-period distribution would likely lead to improving the goodness-of-fit, particularly for the somewhat poorly fit distributions of transit depth and transit depth ratios. Another possible extension would be to allow for explicitly modelling a population of systems with resonant or near-resonant planet pairs or chains. The results presented here show that the observations (particularly the period ratio and transit duration ratio distributions) can be explained by a model with no excess of planets near MMRs relative to the background period ratio distribution. However, this finding does not preclude the possibility that there still might be a small excess of near-resonant planetary systems also contributing to the observed spikes in the period ratio distribution (see §\[secPratios\]). Future studies could explore models with parameters for the amplitude and/or width of such spikes in the period ratio distribution. Such additions would likely benefit from adding new terms to the distance function that explicitly compare behavior near MMRs. Improvements to the detection efficiency model that take into account the presence of transit timing variations (TTVs) may also benefit from these efforts. This paper adopted a distance function that takes into account each of the observed distributions of the population (as described in §\[Summary\_stats\] and shown in Figure \[fig:Kepler\_DR25\]), along with the overall rate of planets per observed star $f_{\rm Kepler}$. Our summary statistics are based on the marginal distributions for each observable. Therefore, it is possible that we might find models that reproduce our chosen summary statistics well, but differ in its prediction for the observed catalogue in other ways (e.g., a correlation between orbital period and planet radius, or a correlation with planet occurrence and host star metallicity). As shown in §\[Model\_params\], our choice of summary statistics and distance function already provide strong constraints on many physically interesting model parameters. Future research adding additional summary statistics and component distances could shrink the ABC posterior further and help constrain any additional model parameters. For example, future research could incorporate the detection of TTVs or transit duration variations (TDVs), so as to provide stronger constraints on the abundance and properties of non-transiting planets. Our forward model makes use of the DR25 completeness and reliability products [@BC2017a; @BC2017b; @BC2017c; @C2017; @T2018] to account for the vast majority of factors that influence the detectability of planets. One aspect in which our model of the detection efficiency could be improved relates to how the pipeline’s detection efficiency is affected by multiple transiting planets. The pipeline first detects the planet with the largest multiple event statistics (MES, analogous to combined transit signal-to-noise), masks out observations near when that planet transits, and re-searches the light curve for additional planets. Typically, only $\sim 0.4\%$ of the light curve is lost for each additional planet [@SJF2017; @ZCH2019]. Thus, masking out observations near transits of a more easily detected planet is expected to have only a modest effect on the integrated transit signal-to-noise and the probability of observing at least three transits. Indeed, @ZCH2019 showed that the multiplicity distribution is only slightly changed when accounting for the lower detection efficiency of subsequent planets. Conclusions {#Conclusions} =========== We have developed a framework for generating populations of planetary systems and simulating observed catalogues of exoplanets under the conditions of a -like mission. We compare three physically motivated models: the non-clustered model, the clustered periods model, and the clustered periods and sizes model, to the observed Q1-Q17 DR25 population of uniformly vetted planet candidates. Our analysis is limited to planets with orbital periods between 3 and 300 d and radii between 0.5 and 10 $R_\oplus$, which form the bulk of the DR25 catalogue (2137 planets in our subset). While most of our findings are consistent with previous works [@Li2011b; @FM2012; @TD2012; @F2014; @Mu2018], this work improves on previous studies by incorporating improved and more homogeneous planet and star catalogues, an improved model for detection efficiency, and by providing a forward model that simultaneously fits all of the marginal distributions described in §\[Summary\_stats\]. Our models are highly flexible and allow for the relatively fast generation ($\sim$10 s for a physical and observed catalogue with the 79 935 target stars used in this study) of simulated observed catalogues that are similar to the exoplanet catalogue in terms of many observables. We define a set of observable summary statistics and a distance function for comparing simulated models with the data. We identify model parameters that result in simulated observed catalogues that closely match the catalogue in terms of the observed planet multiplicity, period, period ratio, transit duration, period-normalized transit duration ratio ($\xi$), transit depth, and transit depth ratio distributions. We train a Gaussian process (GP) model for each of our physical models in order to build an emulator for rapidly predicting the distance as a function of the model parameters. Using the emulator, we draw from the ABC posterior and provide credible intervals for each of the model parameters as constrained by the data. We find that a non-clustered model with a Poisson distribution for the true number of planets per system, single and broken power laws for the periods and planetary radii, and a simple stability criteria, can reasonably fit most of the key observable properties of the population, including the overall rate of observed planets, as well as the period, transit depth, and transit duration distributions, although there are clear differences. However, the number of observed multiplanet systems, the period ratio distribution, and the planet radius ratio distributions are poorly modelled by the non-clustered model. In contrast, clustered models (of periods) can produce observed planet multiplicity and period ratio distributions that are significantly better fits to those observed by . The transit durations and duration ratios are also improved to an appreciable extent. Our fully clustered model, with clustered periods and sizes, performs similarly well and is the best description of the marginal distributions of the data to date. While this model allows for more similar sizes of planets within the same system, the fit to the observed radius ratio distribution is not significantly improved in terms of the KS or AD distances, suggesting that additional features in this distribution are still inadequately reproduced by any of our current models. Most notably, the observed radius ratio distribution appears to be quite asymmetric, which is likely a signature of stripped planetary atmospheres due to photoevaporation [@LFM2012; @OW2017] or core-powered mass-loss [@GS2018]. @W2018a find a very similar effect of clustering and asymmetry in the radius ratio distribution and also a slight positive correlation between radius ratio and the difference in effective temperatures between adjacent planets. Thus, we show with forward modelling that while the observed planet radii ratio distribution of the data require more than a simple clustering in intrinsic sizes to explain, the data cannot be reproduced in a non-clustered model by detection biases alone, in contrast to the results of @Z2019. Several investigations on the numbers of single and multitransiting systems, i.e. the observed multiplicity distribution, suggest that there is an apparent excess of observed singles and that this indicates the existence of more than one underlying population of planetary systems, or a so-called “ dichotomy” (e.g., @Li2011b [@J2012; @HM2013; @BJ2016]). While there is evidence for the excess of stars with a single transiting planet (relative to the predictions based on abundance of systems with multiple transiting planets), other studies (e.g., @W2018b) show that the stellar properties of the singles and multis are indistinguishable, suggesting a common origin. We account for the dichotomy by including two populations of planetary systems with separate mutual inclination dispersions (${\rm Rayleigh}(\sigma_{i,\rm high})$ for a fraction $f_{\sigma_{i,\rm high}}$ of all systems and ${\rm Rayleigh}(\sigma_{i,\rm low})$ for the remaining systems) for each of our models, and find that this is necessary to fit the multiplicity distribution. In our non-clustered model, the occurrence of multitransiting systems is significantly underproduced, due to a tendency for $f_{\sigma_{i,\rm high}}$ to be very low ($\lesssim 3\%$). In our clustered models, the multiplicity distribution is extremely well reproduced, with $1 - f_{\sigma_{i,\rm high}} \simeq 60\%$ of systems having small mutual inclinations of $\sigma_{i,\rm low} \sim 1.4^\circ$ and the remaining $\sim 40\%$ of systems with broad mutual inclinations required to account for the excess observed singles. The $\log{\xi}$ distribution for planets not near any apparent MMRs with other observed planets is also well reproduced with these mutual inclinations and a Rayleigh distribution of eccentricities, $\sigma_e \sim 0.02$. Based on models that allow for clustering in period and radius, it is unlikely that the excess of single transiting planet systems could be explained solely by a large population of intrinsically single-planet systems. Thus, our results provide new evidence in favour of the dichotomy being due to a population of planetary systems with a broader distribution of mutual inclinations than characteristic of the observed multiple planet systems. Previous studies also show that most observed systems are not near low-order MMRs (e.g., @Li2011b [@VF2012; @F2014; @SH2015]), and our model reliably reproduces this observation. Nevertheless, the small fraction of systems near MMRs are particularly interesting to dynamicists and for constraining planet formation models. Therefore, we investigated the ability of our model to reproduce the small spikes in the period ratio distribution slightly wide of first-order MMRs due purely to geometrical effects by assigning planet pairs near MMRs with mutual inclinations also drawn from the population with a narrow distribution of mutual inclinations ($\sigma_{i,\rm low} \sim 1.4^\circ$). While some but not all of the draws from our clustered models include spikes in the period ratio distribution near MMRs, these spikes are statistically robust (see the quantiles (red curves) in Figures \[fig:clustered\_P\_model\] and \[fig:clustered\_P\_R\_model\]). Thus, we show that allowing the distribution of mutual inclinations to depend upon the proximity of a planet pair to MMR provides an alternative explanation for the observed spikes in the period ratio distribution. This study did not consider and thus cannot exclude models in which the observed spikes in the period ratio distribution are due to an intrinsically higher rate (e.g., due to migration leading to resonant trapping). Previous studies have noted that most systems are not in an MMR (e.g., @VF2012), and so an additional mechanism would be required to explain why the systems are near, but not in MMR (e.g., @LW2012 [@L2013; @PMT2013; @DL2014; @GS2014; @X2014; @CF2015; @I2017]). Future studies may be able to distinguish between these two models and better constrain the properties of near-MMR systems by making use of additional observational constraints (e.g., transit timing variations and transit duration variations). The results of precision radial velocity surveys may also help constrain the rate of additional non-transiting planet companions. Our framework of simulating ensembles of planetary systems in a forward model provides a way of directly probing the underlying populations of exoplanets and their properties, including planets that are not observed or do not transit. Our clustered models also provide considerable utility for informing follow-up efforts of new exoplanet surveys such as the Transiting Exoplanet Survey Satellite (TESS) mission, which is expected to discover thousands of additional planetary candidates [@S2015; @B2017; @S2017]. By matching the observed distributions of period ratios and transit depth ratios for the exoplanets and assuming that the population is representative of planetary systems to be observed by TESS, our clustered periods and sizes model can be used to compute conditional probabilities of additional RV detectable planets in systems with already known planets of measured periods and radii. We have made the core SysSim code (<https://github.com/ExoJulia/ExoplanetsSysSim.jl>), inputs collated from numerous datafiles (<https://github.com/ExoJulia/SysSimData>), and the code specific to the clustered models (<https://github.com/ExoJulia/SysSimExClusters>, Zenodo DOI 10.5281/zenodo.3386372) available to the public via Github. We encourage other researchers to contribute model extensions via pull requests and/or additional public git repositories. Additionally, researchers can use our forward modelling pipeline to perform detailed comparisons of the results of planet formation simulations with observations, so as to improve our understanding of planet formation and the architectures of planetary systems more generally. For researchers who prefer not to run the SysSim code themselves, we provide hundreds of physical and observed catalogues in table format, each containing the simulated true and observed properties of $\sim 10^5$ planetary systems, generated from each of our three models using either the best-fitting parameter values or draws from the ABC-posterior distribution for parameter values explained in §\[Model\_params\]. Acknowledgements {#acknowledgements .unnumbered} ================ We thank the entire team for years of work leading to a successful mission and data products critical to this study. We acknowledge many valuable contributions with members of the Science Team’s working groups on multiple body systems, transit timing variations, and completeness working groups. We thank Keir Ashby, Danley Hsu, Neal Munson, Shane Marcus, and Robert Morehead for contributions to the broader SysSim project. We thank Derek Bingham, Earl Lawrence, Ilya Mandell, Oded Aahronson, Ben Bar-Oh, Dan Fabrycky, Jack Lissauer, Gijs Mulders, Aviv Ofir, and Jason Rowe for useful conversations. We thank Rebekah Dawson for reading preliminary drafts of this paper and providing detailed suggestions. MYH acknowledges the support of the Natural Sciences and Engineering Research Council of Canada (NSERC), funding reference number PGSD3-516712-2018. EBF and DR acknowledge support from NASA Origins of Solar Systems grant \# NNX14AI76G and Exoplanet Research Program grant \# NNX15AE21. EBF acknowledges support from NASA Participating Scientist Program Cycle II grant \# NNX14AN76G. MYH and EBF acknowledge support from the Penn State Eberly College of Science and Department of Astronomy & Astrophysics, the Center for Exoplanets and Habitable Worlds, and the Center for Astrostatistics. EBF acknowledges support and collaborative scholarly discussions during residency at the Research Group on Big Data and Planets at the Israel Institute for Advanced Studies. The citations in this paper have made use of NASA’s Astrophysics Data System Bibliographic Services. This research has made use of the NASA Exoplanet Archive, which is operated by the California Institute of Technology, under contract with the National Aeronautics and Space Administration under the Exoplanet Exploration Program. This work made use of the stellar catalogue from @H2019 and thus indirectly the gaia-kepler.fun crossmatch data base created by Megan Bedell. Several figures in this manuscript were generated using the `corner.py` package [@Fm2016]. We acknowledge the Institute for CyberScience (<http://ics.psu.edu/>) at The Pennsylvania State University, including the CyberLAMP cluster supported by NSF grant MRI-1626251, for providing advanced computing resources and services that have contributed to the research results reported in this paper. This study benefited from the 2013 SAMSI workshop on Modern Statistical and Computational Methods for Analysis of Data, the 2016/2017 Program on Statistical, Mathematical and Computational Methods for Astronomy, and their associated working groups. This material was based upon work partially supported by the National Science Foundation under Grant DMS-1127914 to the Statistical and Applied Mathematical Sciences Institute (SAMSI). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. Anderson, T. W. & Darling, D. A. 1952, Ann. Math. Stat., 23, 193 Andrae, R., et al. 2018, A&A, 616, A8 Ballard, S. & Johnson, J. A. 2016, ApJ, 816, 66 Batalha, N. M., et al. 2013, ApJS, 204, 24 Bezanson, J., Edelman, A., Karpinski, S., & Shah, V. B. 2014, arXiv e-prints, arXiv:1411.1607 Borucki, W. J., et al. 2010, Science, 327, 977 Borucki, W. J., et al. 2011a, ApJ, 728, 117 Borucki, W. J., et al. 2011b, ApJ, 736, 19 Bouma, L. G., Winn, J. N., Kosiarek, J., McCullough, P. R., 2017, arXiv:1705.08891 Brakensiek, J. & Ragozzine, D. 2016, ApJ, 821, 47 Burke, C. J., & Catanzarite, J. 2017a, Planet Detection Metrics: Window and One-Sigma Depth Functions for Data Release 25, Tech. rep. NASA Ames Research Center, Moffett Field, CA. Burke, C. J., & Catanzarite, J. 2017b, Planet Detection Metrics: Per-Target Flux-Level Transit Injection Tests of TPS for Data Release 25, Tech. rep. NASA Ames Research Center, Moffett Field, CA. Burke, C. J., & Catanzarite, J. 2017c, Planet Detection Metrics: Per-Target Detection Contours for Data Release 25, Tech. rep. NASA Ames Research Center, Moffett Field, CA. Carrera, D., Ford, E. B., Izidoro, A., Jontof-Hutter, D., Raymond, S. N., Wolfgang, A., 2018, ApJ, 866, 104 Chambers, J. E., Wetherill, G. W., & Boss, A. P. 1996, Icarus, 119, 261 Chatterjee, S. & Ford, E. B. 2015, ApJ, 803, 33 Chen, J. & Kipping, D. 2016, ApJ, 834, 17 Christiansen, J. L. 2017, Planet Detection Metrics: Pixel-Level Transit Injection Tests of Pipeline Detection Efficiency for Data Release 25, Tech. rep. NASA Ames Research Center, Moffett Field, CA. Ciardi, D. R., Fabrycky, D. C., Ford, E. B., Gautier, T. N., Howell, S. B., Lissauer, J. J., Ragozzine, D., Rowe, J. F., 2013, ApJ, 763, 41 Coughlin, J. L., et al. 2016, ApJS, 224, 12 Coughlin, J. L. 2017, Planet Detection Metrics: Robovetter Completeness and Effectiveness for Data Release 25, Tech. rep. NASA Ames Research Center, Moffett Field, CA. Cressie, N. & Read, T. R. C. 1984, J. R. Stat. Soc. B, 46, 440 Dawson R. I., Lee E. J., Chiang E., 2016, ApJ, 822, 54 Delisle, J.-B. & Laskar, J. 2014, A&A, 570, L7 Fabrycky, D. C., et al. 2014, ApJ, 790, 146 Fabrycky, D. C. & Winn, J. N. 2009, ApJ, 696, 1230 Fang, J., & Margot, J.-L. 2012, ApJ, 761, 92 Figueira, P., et al. 2012, A&A, 541, A139 Ford, E. B., Quinn, S. N., & Veras, D. 2008, ApJ, 678, 1407 Ford, E. B. 2014, Proc. Natl. Acad. Sci. USA, 111, 12616 Ford, E. B., He, M. Y., Hsu, D. C., & Ragozzine, D. 2018b, Planetary Systems Simulation & Model of Kepler Mission for Characterizing the Occurrence Rates of Exoplanets and Planetary Architectures, v1.0, Zenodo, doi:10.5281/zenodo.1205172. <https://doi.org/10.5281/zenodo.1205172> Foreman-Mackey, D. 2016, corner.py: Scatterplot matrices in Python, J. Open Source Softw., 1(2), 24, doi:10.21105/joss.00024 Fulton, B. J., et al. 2017, AJ, 154, 109 Gaia Collaboration, 2018, A&A, 616, A1 Ginzburg, S., Schlichting, H. E., & Sari, R. 2016, ApJ, 825, 29 Ginzburg, S., Schlichting, H. E., & Sari, R. 2016, MNRAS, 476, 759 Gladman, B. 1993, Icarus, 106, 247 Goldreich, P. & Schlichting, H. E. AJ, 147, 32 Gupta, A. & Schlichting, H. E. 2019, MNRAS, 487, 24 Hadden, S. & Lithwick, Y. 2014, ApJ, 787, 80 Hansen, B. M. S. & Murray, N. 2013, ApJ, 775, 53 Howard, A. W., et al. 2012, ApJS, 201, 15 Hsu, D. C., Ford, E. B., Ragozzine, D., & Morehead, R. C., 2018, AJ, 155, 205 Hsu, D. C., Ford, E. B., Ragozzine, D., & Ashby, K. 2019, AJ, 158, 109 Izidoro, A., Ogihara, M., Raymond, S. N., Morbidelli, A., Pierens, A., Bitsch, B., Cossou, C., Hersant, F., 2017, MNRAS, 470, 1750 Johansen, A., Davies, M. B., Church, R. P., Holmelin, V., 2012, ApJ, 758, 39 Kipping, D. M. 2010, MNRAS, 407, 301 Kolmogorov, A. N. 1933, Giornale dell’Istituto Italiano degli Attuari, 4, 83 Latham, D. W., et al. 2011, ApJL, 732, L24 Lee, M. H, Fabrycky, D., & Lin, D. N. C. 2013, ApJ, 774, 52 Lissauer, J. J., et al. 2011a, Nature, 470, 53 Lissauer, J. J., et al. 2011b, ApJS, 197, 8 Lissauer, J. J., et al. 2012, ApJ, 750, 112 Lissauer, J. J., et al. 2014, ApJ, 784, 44 Lithwick, Y. & Wu, Y. 2012, ApJL, 756, L11 Lopez, E. D. & Fortney, J. J. 2014, ApJ, 792, 1 Lopez, E. D., Fortney, J. J., & Miller, N. 2012, ApJ, 761, 59 Millholland, S., Wang, S., & Laughlin, G. 2017, ApJL, 849, L33 Mills, S. M., et al. 2019, AJ, 157, 5 Moorhead, A. V., et al. 2011, ApJS, 197, 1 Morehead, R. C. 2016, PhD Dissertation, <https://etda.libraries.psu.edu/catalog/mc87pq25t> Mulders, G. D., Pascucci, I., Apai, D., Ciesla, F. J., 2018, AJ, 156, 24 Mullally, F., et al. 2015, ApJS, 217, 31 Ning, B., Wolfgang, A., & Ghosh, S. 2018, ApJ, 869, 5 O’Hagan, A. 2004, Rel. Eng. Sys. Safety, 91, 1290. Osada, R., Funkhouser, T., Chazelle, B., & Dobkin, D. 2002, ACM Trans. Graph., Vol. 21., No. 4, 807, 832, <http://www.cs.princeton.edu/~funk/tog02.pdf> Owen, J. E. & Wu, Y. 2013, Kepler Planets: A Tale of Evaporation, ApJ, 775, 105 Owen, J. E. & Wu, Y. 2017, ApJ, 847, 29 Petigura, E. A., Marcy, G. W., & Howard, A. W. 2013b, ApJ, 770, 69 Petrovich, C., Malhotra, R., & Tremaine, S. 2013, ApJ, 770, 24 Pettitt, A. N. 1976, Biometrika, 63, 161 Price, E. M. & Rogers, L. A. 2014, ApJ, 794, 92 Pu, B. & Wu, Y. 2015, ApJ, 807, 44 Ragozzine, D. & Holman, M. J. 2010, arXiv:1006.3727 Rajpaul, V., Aigrain, S., Osborne, M. A., Reece, S., Roberts, S., 2015, MNRAS, 452, 2269 Rasmussen, C. E. & Williams, C. K. I. 2006, Gaussian Processes for Machine Learning, MIT Press, Cambridge, MA Rowe, J. F., et al. 2014, ApJ, 784, 45 Rowe, J. F., et al. 2015, ApJS, 217, 16 Seager, S. & Mallén-Ornelas, G. 2003, ApJ, 585, 1038 Shabram, M., Demory, B.-O., Cisewski, J., Ford, E. B., Rogers, L., 2015, ApJ, 820, 93 Schmitt, J. R., Jenkins, J. M., & Fischer, D. A. 2017, ApJ, 153, 180 Smirnov, N. 1948, Ann. Math. Stat., 19, 279 Stassun, K. G., Oelkers, R. J., Pepper, J., Gaudi, B. S., 2017, AJ, 156, 102 Steffen, J. H., et al. 2010, ApJ, 725, 1226 Steffen, J. H. & Hwang, J. A. 2015, MNRAS, 448, 1956 Sullivan, P. W., et al. 2015, ApJ, 809, 77 Thompson, S. E., Mullally, F., Coughlin, J., Christiansen, J. L., Henze, C. E., Haas, M. R., Burke, C. J., 2015, ApJ, 812, 46 Thompson, S. E., et al. 2018, ApJS, 235, 38 Tremaine, S. & Dong, S. 2012, AJ, 143, 94 Twicken, J. D., et al. 2016, AJ, 152, 6 Van Eylen, V. & Albrecht, S. 2015, ApJ, 808, 126 Van Eylen, V., Agentoft, C., Lundkvist, M. S., Kjeldsen, H., Owen, J. E., Fulton, B. J., Petigura, E., Snellen, I., 2018, MNRAS, 479, 4786 Van Eylen, V., et al. 2019, AJ, 157, 61 Veras, D. & Ford, E. B. 2012, MNRAS, 420, L23 Weiss, L. M., et al. 2018, AJ, 155, 48 Weiss, L. M., et al. 2018, AJ, 156, 254 Weiss, L. M. & Marcy, G. W. 2014, ApJ, 783, L6 Weissbein, A., Steinberg, E., & Sari, R. 2012, arXiv:1203.6072 Winn, J. N. & Fabrycky, D. C. 2015, ARA&A, 53, 409 Wolfgang, A. & Laughlin, G. 2012, ApJ, 750, 148 Wolfgang, A., Rogers, L. A., & Ford, E. B. 2016, ApJ, 825, 19 Wu, Y. & Lithwick, Y. 2013, ApJ, 772, 74 Wu, D.-H., Zhang, R. C., Zhou, J.-L., Steffen, J. H., 2019, MNRAS, 484, 1538 Xie, J.-W. 2014, ApJ, 786, 153 Xie, J.-W., et al. 2016, Proc. Natl. Acad. Sci. USA, 113, 11431 Youdin, A. N. 2011, ApJ, 742, 38 Zhu, W., Petrovich, C., Wu, Y., Dong, S., Xie, J., 2018, ApJ, 860, 101 Zhu, W. 2019, arXiv:1907.02074 Zink, J. K., Christiansen, J. L., & Hansen, B. M. S. 2019, MNRAS, 483, 4479 -- -- -- -- \[lastpage\] [^1]: Contact e-mail: <[email protected]> [^2]: The mean of a ZTP-distributed random variable $X$ with parameter $\lambda$ is given by $E[X] = \frac{\lambda}{1-e^{-\lambda}}$. [^3]: In practice, we first draw unscaled periods $P'_i \sim {\rm Lognormal}(0, N_p\sigma_P)$, truncated between $P'_{\rm min} = \sqrt{P_{\rm min}/P_{\rm max}}$ and $P'_{\rm max} = \sqrt{P_{\rm max}/P_{\rm min}}$ (this is to avoid drawing unscaled periods such that maximum period ratio is greater than $P_{\rm max}/P_{\rm min}$, which would have no chance fitting between $P_{\rm min}$ and $P_{\rm max}$ regardless of $P_c$). We then draw a cluster period scale $P_c$ from Equation (\[eq\_Pc\]) (truncated between $P_{\rm min}/{\rm min}\{P'_i\}$ and $P_{\rm max}/{\rm max}\{P'_i\}$) to give $P_i = P'_i{P_c}$, checking that the drawn value of $P_c$ allows the planets in the cluster to be stable with all other planets in the system; see §\[Procedure\] for a detailed outline of this process. [^4]: We adopt a value of $N_{\rm attempts} = 100$, for the purposes of computational efficiency. This is necessary to prevent certain (e.g. extremely populated but compact; large $N_p$ with a small set value of $\sigma_P$) clusters from significantly increasing the computational time for simulating a physical catalogue. On the other hand, if we attempt the drawing of planets in each cluster only a few times (or even just once), the rejection-sampling algorithm tends to produce too few planetary systems. Thus, a modest number of attempts per cluster provides a compromise between these two extremes. Also, a consequence of this procedure is that the true mean rates of clusters and planets per cluster in our samples are somewhat less than $\lambda_c$ and $E[N_p]$, respectively. [^5]: The non-clustered model could be considered as a special case of the clustered models by setting the number of planets per cluster to one, $N_p = 1$, instead of drawing from a ZTP distribution in Step 4(a). If there is exactly one planet per “cluster”, then the properties of planets in the same planetary system are effectively drawn independently. In this case, the total number of planets in any system is $N_c$. In practice, for the non-clustered model, we draw periods for all the $N_p$ planets simultaneously and accept or reject all periods at once, so as to improve computational efficiency and to minimize artefacts near $P_{\rm min}$ and $P_{\rm max}$ for systems with many planets. [^6]: We use SysSim in the single-observer mode, and do not make use of CORBITS for sky-averaging [@BR2016]. Working in single-observer model allows our forward model to reproduce the variations in the observed catalogue due to the finite number of targets (see @H2019) and avoids issues with two-planet correlations that are not calculated by CORBITS. [^7]: @Li2011a use the exact multinomial statistic, but using this statistic is intractable for our case. After significant testing of different statistics, CRPD was shown to be the best approximation of the exact statistic for distributions similar to ours. [^8]: We also considered switching the interpretation of $O_j$ and $E_j$, i.e. let $O_j$ denote the number of actual observed systems and $E_j$ denote the number of expected observed systems given a model. During exploratory analyses, we found that $\rho_{\rm CRPD}$ would occasionally give infinite values if a simulated catalogue included zero 5+ planet systems. Eliminating that term from the sum is not viable, as this can result in negative values, which violates the non-negative property of a distance function and would lead to favouring models with fewer multiplanet systems than were observed by . [^9]: This is apparent if one considers the effect of the $\frac{nm}{N}$ term in Equation (\[eq\_AD\_def\]): let $m$ be the number of observed planets in the data and $n$ be the number of observed planets given by the model. If $n \simeq m$, this term is roughly $\sim m/2$; but if $n \sim 1 \ll m$, this term becomes roughly $\frac{m}{m+1} \sim 1$. Since $m$ is relatively large, the $\frac{nm}{N}$ term is thus much smaller when $n \ll m$ than when $n \simeq m$. [^10]: @OFCD2002 prescribe the equation $P = (1-\sqrt{r_1})A + \sqrt{r_1}(1-r_2)B + {r_2}\sqrt{r_1}C$ in their Equation (1) of Section 4.2 to uniformly map the unit square to any arbitrary triangle of vertices $(A,B,C)$, i.e. by drawing $r_1$ and $r_2$ randomly between 0 and 1. We adopt this transformation and set the vertices to $A = (0^\circ, 0^\circ)$, $B = (90^\circ, 90^\circ)$, and $C = (90^\circ, 0^\circ)$ for $(\sigma_{i,\rm high}, \sigma_{i,\rm low})$. [^11]: We do not train an emulator or compute credible regions for the non-clustered model using the AD distances, because upon inspection of the observed catalogues generated using the best model parameters resulting from the optimization scheme in Section §\[Optimization\], we find that the overall rate of planets is such a poor fit with this distance function and model that it is not worthy of more detailed investigation. [^12]: While our range of periods and radii includes many Hot Jupiters which are thought to be relatively isolated, the fraction of stars with such systems is known to be quite low ($\sim$1%) and does not affect our conclusions about the origin of the dichotomy. [^13]: As expected, we do not get any statistically significant spikes if we do not assign planets near an MMR to have low mutual inclinations.	{ "pile_set_name": "ArXiv" }
--- abstract: 'Ferromagnetic materials exhibiting low magnetic damping ($\alpha$) and moderately high saturation magnetization are required from the viewpoints of generation, transmission and detection of spin wave. Since spin-to-charge conversion efficiency is another important parameter, high spin mixing conductance ($g_{r}^{\uparrow \downarrow}$) is the key for efficient spin-to-charge conversion. Full Heusler alloys e.g. $Co_2Fe_{0.4}Mn_{0.6}Si$ (CFMS), which are predicted to be 100$\%$ spin polarized, possess low $\alpha$. However, the $g_{r}^{\uparrow \downarrow}$ at the interface between CFMS and a paramagnet has not fully been understood. Here, we report the investigations of spin pumping and inverse spin Hall effect in $CFMS/Pt$ bilayers. Damping analysis indicates the presence of significant spin pumping at the interface of CFMS and Pt, which is also confirmed by the detection of inverse spin Hall voltage. We show that in CFMS/Pt the $g_{r}^{\uparrow \downarrow}$ (1.77$\times$10$^{20}$m$^{-2}$) and interface transparency (84$\%$) are higher compared to values reported for other ferromagnet/heavy metal systems.' address: - 'Laboratory for Nanomagnetism and Magnetic Materials (LNMM), School of Physical Sciences, National Institute of Science Education and Research (NISER), HBNI, Jatni-752050, India' - 'Institute for Materials Research, Tohoku University, Sendai 980-8577, Japan' - 'Center for Spintronics Research Network, Tohoku University, Sendai 980-8577, Japan' - 'Laboratory for Nanomagnetism and Magnetic Materials (LNMM), School of Physical Sciences, National Institute of Science Education and Research (NISER), HBNI, Jatni-752050, India' author: - Braj Bhusan Singh - Koustuv Roy - Pushpendra Gupta - Takeshi Seki - Koki Takanashi - Subhankar Bedanta title: 'High spin mixing conductance and spin interface transparency at $Co_2Fe_{0.4}Mn_{0.6}Si$ Heusler alloy and Pt interface' --- Introduction ============= Spin transport across interfaces in ferromagnetic (FM)/heavy metal (HM) systems are important to develop future spintronics devices \[1,2\]. Spin orbital torque \[3,4\], spin transfer torque \[1,2\], spin pumping/inverse spin Hall effect \[5$-$9\], spin Seeback effects \[[10]{}$-$[12]{}\], etc. are major phenomenon which are predominantly affected by interface spin transport in FM/HM systems. Spin pumping is an efficient method to produce pure spin current ($J_s$), which is the flow of spin angular momentum, and investigate the spin propagation across FM/HM interfaces. The efficiency of spin transport at FM/HM interfaces is understandable by the factor known as effective spin mixing conductance ($g_{r}^{\uparrow \downarrow}$), which is related to $J_s$ by the expression\[2\]. $$\label{equation1} J_s = \frac{\hbar}{4 \pi}g_{r}^{\uparrow \downarrow}\hat{m} \times \frac{d \hat{m}}{dt}$$ where $\hat{m}$ is the unit vector of magnetization. $J_s$ can be converted into transverse voltage ($V_{ISHE }$) by inverse spin Hall effect (ISHE) \[13\]: $$\label{eqiation2} V_{ISHE } \propto \theta_{SH} \Vec{J}_s \times \Vec{\sigma}$$ where $\theta_{SH}$ is the spin Hall angle which defines the conversion efficiency between the charge current ($J_c$) and $J_s$, and $\Vec{\sigma}$ is the spin matrices governed by the spin polarization direction. Therefore, in order to get high $V_{ISHE }$ and hence $g_{r}^{\uparrow \downarrow}$ in a FM/HM heterostructure, $\theta_{SH}$ of the HM needs to be large. The value of $\theta_{SH}$ mostly depends on spin orbit interaction (SOI) and conductivity of the HM \[2,14\]. Efficient interfacial spin transport critically depends on the type of interfaces and its associated FM and HM materials, FM materials with low magnetic damping ($\alpha$) are important to generate large spin current and hence high $g_{r}^{\uparrow \downarrow}$ \[2\]. In this context various low damping materials such as NiFe, CoFeB, and $Y_3Fe_5O_{12}$ have been studied. Further there is another class of half metallic materials e.g Heusler alloys which have been established as low damping systems. It is also noted here that spin pumping and the resultant spin current can be described as an accumulation of the up and down spins \[15\]. Therefore, it is expected that in Heusler alloys spin pumping efficiency will be larger due to the presence of only one type of spins at the Fermi level. In addition, low magnetic damping and expected high spin pumping makes Heusler alloys suitable for the interface spin transport study and pure spin current-based spin torque nano-oscillators \[16\]. There have been intense studies of spin dynamics with low damping materials, e.g. NiFe and CoFeB with various HM \[17$-$21\]. However, there are only a few reports on the spin dynamics of Heusler alloys with HM \[15,22\]. $Co_2Fe_{0.4}Mn_{0.6}Si$ (CFMS) is a Heusler alloy which shows low damping and 100$\%$ spin polarization \[23\]. Figure 1(a) shows a typical schematic for density of states for half metallic material. However, the spin pumping efficiency ($g_{r}^{\uparrow \downarrow}$) in CFMS/HM system has so far not been evaluated, which would help to understand its use for applications. Here, we report the spin pumping study in CFMS/Pt system with varying the thickness of Pt via (1) measurement of ISHE, (2) evaluation of effective mixing conductance $g_{r}^{\uparrow \downarrow}$ and (3) spin interface transparency. EXPERIMENTAL DETAILS ==================== The bilayer samples viz. S1$\#$ CFMS(20 nm)/Pt(3 nm), S2$\#$ CFMS(20 nm)/Pt(5 nm), S3$\#$ CFMS(20 nm)/Pt(7 nm), S4$\#$ CFMS(20 nm)/Pt(10 nm) and S5$\#$ CFMS(20 nm)/Pt(20 nm) were prepared on MgO(100) substrates using dc magnetron sputtering in a vacuum system with base pressure $\sim$ 1 $\times$ $10^{-9}$ mbar \[24\]. The prepared CFMS thin films were in-situ annealed at $600^{\circ}$C/1hr to improve its crystallinity and surface quality. Reflection high energy electron diffraction (RHEED) patterns were acquired to characterize the surface and crystalline quality of CFMS layers. After the preparation of CFMS layer, Pt layer was deposited at room temperature by dc magnetron sputtering. The thickness of the films were evaluated using x-ray reflectivity (XRR) (data not shown). Ferromagnetic resonance (FMR) measurements have been performed in the frequency range 5-17 GHz on a coplanar wave guide in the flip-chip manner \[25,26\]. ISHE measurements have been performed by connecting a nanovoltmeter over two ends of the sample (sample size: 3 mm $\times$ 2 mm). The detail of the ISHE set-up can be found elsewhere \[27\]. Results and Discussion ====================== Crystalline quality {#crystalline-quality .unnumbered} ------------------- Fig. \[fig1\] (b) and (c) show the RHEED patterns for the samples S4 and S5, respectively, observed in the MgO\[100\] and MgO\[110\] azimuths. From the streaks and spots of the RHEED patterns it is confirmed that the CFMS layer with the (001) crystalline orientation was epitaxially grown on the MgO (001) substrate. The streak lines which are elongated spots in vertical direction in the RHEED pattern implies the improvement of flatness at the CFMS surface. ![(a) A Schematic of density of states ($D(E)$) for an ideal half metal alloy. (b) and (c) are the RHEED patterns for the samples S4 and S5, respectively, on MgO (100) substrate in the \[100\] and \[110\] azimuths. (d) Schematic of the setup for ISHE measurement, where $h_{rf}$ is the rf magnetic field generated in a coplanar wave guide (CPW) perpendicular to the applied magnetic field (H).[]{data-label="fig1"}](fig1.png){width="45.00000%"} Magnetic Damping {#magnetic-damping .unnumbered} ---------------- Fig. \[fig2\](a) and (b) show the plots of resonance frequency (f) versus $H_r$ and $\Delta H$ versus f, respectively. Here, the values of $H_r$ and $\Delta H$ were evaluated using FMR spectra (see fig. A1 in supplementary information). In order to evaluate the gyromagnetic ratio ($\gamma$) and effective demagnetization ($4\pi M_{eff}$), \[fig2\](a) was fitted to Kittel’s equation \[28\] given as: $$\label{equation3} \it{f}=\frac{\gamma}{2 \pi} \sqrt{(H_K+H_r)(H_K +H_r + 4 \pi M_{eff})}$$ where $$\label{equation4} 4 \pi M_{eff} = 4 \pi M_s + \frac{2K_S}{M_st_{FM}}$$ and $H_K$, $K_s$, $M_s$ $t_{FM}$, are anisotropy field, perpendicular surface magnetic anisotropy constant, saturation magnetization, and thickness of FM layer, respectively. $\alpha$ was evaluated by fitting data of Fig. \[fig2\](b) using the following expression \[29\]: $$\label{equation5} \Delta H = \Delta H_0 +\frac{4 \pi \alpha \it{f}}{\gamma}$$ where $\Delta H_0$ is the inhomogeneous broadening of linewidth which depends on the homogeneity of the sample. There are various other effects such as interface effect, impurity, magnetic proximity effects (MPE) etc., which also can enhance the value of $\alpha$ of the system. Hence, the total $\alpha$ can be written as: $$\label{q6} \alpha = \alpha_{int} + \alpha_{impurity} + \alpha_{MPE} + \alpha_{sp}$$ where $\alpha_{int}$ is the intrinsic damping, and $\alpha_{impurity}$, $\alpha_{MPE}$, and $\alpha_{sp}$ are the contribution from impurity, magnetic proximity effect (MPE), and spin pumping to the $\alpha$, respectively \[30\]. ![(a) $\textit{f}$ vs $H_r$ and (b) $\Delta H$ vs $\textit{f}$ plots for the samples S1 (open squares) and S5 (open circles). The solid lines are the best fits which are fitted by equations \[equation3\] and \[equation5\]. []{data-label="fig2"}](fig2.png){width="45.00000%"} The linear behaviour of $\Delta H$ vs $\textit{f}$ plots implies the good homogeneity in our samples and it rules out the possibility of any kind of magnetic impurity of the FM layer. S1 0.0066 $\pm$ 0.0001 ---- --------------------- S2 0.0063 $\pm$ 0.0001 S3 0.0070 $\pm$ 0.0001 S4 0.0085 $\pm$ 0.0001 S5 0.0087 $\pm$ 0.0001 : The values of $\alpha$ for samples S1-S5 The values of $\alpha$ are larger than the reported value of single layer of CFMS ($\sim$ 0.004) \[24\]. This enhancement in the values of $\alpha$ is the indication of the spin pumping. However, we cannot rule out other effects e.g. MPE, and any impurities which may contribute in enhancing the value of $\alpha$ (see Eq. (\[q6\])). In order to investigate the MPE or magnetic dead layer formation at the interface, we measured saturation magnetization ($M_s$) for all the samples by SQUID magnetometer (data not shown). The measured values of $M_s$ for all the samples are found to be 861 emu/cc (S1), 842 emu/cc (S2), 792 emu/cc (S3), 845 emu/cc (S4) and 807 emu/cc (S5). The change in the values of Ms with the thickness of Pt may be due to MPE or dead layer formation \[31,32\]. Inverse spin Hall effect measurement {#inverse-spin-hall-effect-measurement .unnumbered} ------------------------------------ In order to confirm the spin pumping in our system, we performed ISHE measurements on all the samples as shown in schematic Fig. \[fig1\](d). The measurements are carried out at 11 mW power and 7 GHz frequency. ![Voltage ($V_{meas}$) measured across the sample with applied magnetic field along with FMR signal for sample S1 at the $\phi$ values of (a) 0$^{\circ}$, (b) 30$^{\circ}$, (c) 90$^{\circ}$, (d) 180$^{\circ}$. Open symbols are the experimental data. Solid lines are the fit to the experimental data using Eq. (\[q7\]). Short dash and dotted lines are the symmetric ($V_{sym})$ and anti-symmetric ($V_{asym})$ components of the voltage](fig3.png){width="45.00000%"} . \[fig3\] The angle $\phi$ denotes the angle between measured voltage direction and the perpendicular direction of applied DC magnetic field (H). It has been previously found that CFMS thin films exhibit a cubic anisotropy \[33\]. Sample $V_{sp}$(V)$\times$$10^{-6}$ $V_{AHE}$(V)$\times$$10^{-6}$ $V_{AMR}^{\perp}$(V)$\times$$10^{-6}$ $V_{AMR}^{\|\|}$(V)$\times$$10^{-6}$ -------- ------------------------------ ------------------------------- --------------------------------------- ------------------------------------ S1 3.93 $\pm$ 0.05 -1.14 $\pm$ 0.03 2.97 $\pm$ 0.06 0.16 $\pm$ 0.03 S2 5.41 $\pm$ 0.16 -1.28 $\pm$ 0.04 3.14 $\pm$ 0.18 0.14 $\pm$ 0.04 S3 2.88 $\pm$ 0.06 -0.78 $\pm$ 0.02 1.54 $\pm$ 0.07 0.09 $\pm$ 0.02 S4 3.26 $\pm$ 0.05 -0.67 $\pm$ 0.05 2.28 $\pm$ 0.06 0.14 $\pm$ 0.03 S5 0.89 $\pm$ 0.02 -0.64 $\pm$ 0.01 0.67 $\pm$ 0.02 0.05 $\pm$ 0.01 Angle dependent measurements of the voltage have been investigated to remove spin rectification effects e.g. anisotropic magnetoresistance (AMR), anomalous Hall effect (AHE). Fig. \[fig3\] shows the measured voltage ($V_{meas}$) (open blue symbol) versus H along with FMR signal (open black symbol) for sample S1 at the angles $\phi$= 0$^{\circ}$ (a), 30$^{\circ}$ (b), 90$^{\circ}$ (c) 180$^{\circ}$ (d). It should be noted that $\phi$= 0$^{\circ}$ means that the field was applied along the easy axis of the sample. There has been very less signal observed at $\phi$= 90$^{\circ}$ (Fig. \[fig3\](b)). This is due to the negligible amount of spin accumulation parallel to the applied magnetic field. It is evident from Fig. \[fig3\](a) and (d) that the sign of $V_{meas}$ is reversed when $\phi$ moves from 0$^{\circ}$ to 180$^{\circ}$. This indicates that the voltage is majorly produced by the spin pumping. It is well known that if the sign of the $V_{meas}$ does not reverse with angle, then the contribution solely comes from different spin rectification effects. Figure \[fig4\] shows the $v_{meas}$ versus $\it{H}$ plot for the sample S5 measured at $\phi$= 0$^{\circ}$ (a), 30$^{\circ}$ (b), 90$^{\circ}$ (c) 180$^{\circ}$ (d). The similar kind of ISHE signal was observer for all the samples (data not shown). It is observed that the strength of the $V_{meas}$ for sample S5 (20 nm thick Pt) is three times smaller than that of the sample S1 (3 nm thick Pt).This is consistent to the fact that ISHE voltage is inversely proportional to the conductivity and thickness of the HM layer \[34\]. For the separation of spin pumping contribution from the $\it{V_{meas}}$ by excluding other spurious effects, the $\it{V_{meas}}$ versus $\it{H}$ plots for the samples S1 (Fig. 3) and S5 (Fig. \[fig4\]) were fitted with Lorentzian equation \[35\] which is given by: $$\label{q7} \begin{aligned} V_{meas} = V_{sym} \frac{(\Delta H)^2}{(H-H_r)^2+(\Delta H)^2}+ \\ V_{asym} \frac{2 \Delta H (H - H_r)}{(H-H_r)^2+(\Delta H)^2} \end{aligned}$$ ![$V_{meas}$ versus $\it{H}$ and FMR signal for sample S5 at the $\phi$ values of (a) 0$^{\circ}$, (b) 30$^{\circ}$, (c) 90$^{\circ}$ (d) 180$^{\circ}$. Open symbols are representing the measured voltage. Experimental data are fitted (solid lines) using equation (\[q7\]). Dashed and dotted lines are plots for components of symmetric ($V_{sym}$) and anti-symmetric ($V_{asym}$) voltage fitted to equation (\[q7\]). []{data-label="fig4"}](fig4.png){width="45.00000%"} where $V_{sym}$ and $V_{asym}$ are the symmetric and anti-symmetric components. Solid lines are fits to the experimental data. The $V_{sym}$ consists of major contribution from spin pumping, while minor contributions from AHE, and AMR effects. The AHE contribution is zero here if the rf field and H are perpendicular to each other, which is the case in our measurement. Whereas the AHE and AMR are the major contributions in the $V_{asym}$ component. Fig. 3 and 4 also show the plot of $V_{sym}$ (dashed line) and $V_{asym}$ (dotted line) separately for the samples S1, and S5, respectively. ![Angle dependent ($\phi$) $V_{sym}$ and $V_{asym}$ measurements for samples S1 (a and b) and S5 (c and d), respectively.[]{data-label="fig5"}](fig5.png){width="50.00000%"} In-plane angle dependent measurements of $V_{meas}$ were performed at the interval of 2$^{\circ}$ to quantify spin pumping and other spin rectification contributions (Fig. \[fig5\]). It is a well-established method to decouple the individual components from the measured voltage \[30,36,37\]. The model given by Harder et.al. \[38\] has considered the rectification effects i.e., parallel AMR ($V_{asym/sym}^{AMR \|\|}$) and perpendicular AMR ($V_{asym/sym}^{AMR \perp}$) to the applied rf field and the AHE contribution due to the FM layer. The relation between the measured voltage and those rectification effects are as follows \[36\]: $$\label{q8} \begin{aligned} V_{asym}= V_{AHE}cos(\phi + \phi_0) sin (\it{\Phi})+ \\V_{asym}^{AMR \perp} cos 2(\phi + \phi_0)sin(\it{\Phi})+ \\ V_{asym}^{AMR \|\|}sin2(\phi + \phi_0)cos(\phi+\phi_0) \end{aligned}$$ $$\label{q9} \begin{aligned} V_{sym}= V_{sp}cos^3(\phi + \phi_0)+V_{AHE}cos(\phi + \phi_0)cos(\it{\Phi)}\\ + V_{sym}^{AMR \perp} cos 2(\phi + \phi_0)cos(\phi+ \phi_0)\\ + V_{sym}^{AMR \|\|}sin2(\phi + \phi_0)cos(\phi+\phi_0) \end{aligned}$$ $V_{AHE}$ and $V_{sp}$ correspond to the AHE voltage and the spin pumping contributions, respectively. $\phi$ is the angle between applied H and the rf magnetic field which is always perpendicular in the measurement. The extra factor $\phi_{0}$ is taken to incorporate the misalignment of sample positioning in defining the $\phi$ value during the measurement. The detailed fits with and without incorporation of small offset in $\phi$ value is shown in Fig. A2 in the supplementary information. Further the AMR contribution also can be quantified by the following formula \[36\] : $$\label{q10} V_{AMR}=\sqrt{(V_{Asym}^{AMR \perp,\|\|})^{2}+(V_{sym}^{AMR \perp,\|\|})^{2}}$$ The $V_{Asym}^{AMR \perp,\|\|}$ and $V_{sym}^{AMR \perp,\|\|}$ are evaluated from the in-plane angle dependent $V_{meas}$ measurements by fitting those values by equations \[q8\] and \[q9\], respectively. The extracted values of the various components are listed in the Table II. It is observed that the $V_{sp}$ is dominating over other unwanted spin rectification effects in all the samples. However, the magnitude of AHE is comparable to the spin pumping, which is decreased by one order for thicker Pt samples. It may be due to increase in conductivity of Pt layer due to increase in its thickness. It is well known that the AHE majorly depend on the magnetization of the sample due to berry curvature of the FM \[39\]. The AHE contribution is an intrinsic property of the FM layer. The Co based FM materials are always a potential candidate for the AHE phenomena \[40,41\]. The saturation magnetization measurements of all the samples, indicated the presence of MPE in Pt or dead layer formation at interface which may result in the decrease of $V_{AHE}$ contribution as the Pt thickness increases from 3 to 20 nm. However, the AMR values are of similar order in all the samples. The finite AMR contribution indicates that the samples are anisotropic in nature. The positive value of $V_{sp}$ indicates the positive spin Hall angle in Pt, which is consistent with literature \[34\]. ![$g_{\it{eff}}^{\uparrow \downarrow}$ (a) and spin Hall angle (b) as a function of Pt thickness. Solid line in (a) is the best fit using the equation (12)[]{data-label="fig6"}](fig6.png){width="50.00000%"} The lowest $\alpha$ is found to be in S2 shows the maximum spin pumping voltage which is because of smooth interface between CFMS/ Pt. $V_{sp}$ is getting dominated by the conductivity of Pt for thicker Pt sample. Thus the $V_{sp}$ decreases with higher $t_{Pt}$ value. Figure \[fig6\] shows the graph between $g_{\it{eff}}^{\uparrow \downarrow}$ and Pt thickness. $g_{\it{eff}}^{\uparrow \downarrow}$ was calculated by the following expression using damping constant\[2\]: $$\label{q11} g_{\it{eff}^{\uparrow \downarrow}}=\frac{\Delta\alpha 4\pi M_{s}t_{CFMS}}{g\mu_{B}}$$ where $\Delta\alpha$, $t_{CFMS}$, $\mu_{B}$, g are the change in the $\alpha$ due to spin pumping, the thickness of CFMS layer, Bohr magneton, Lande g- factor (2.1), respectively. In order to calculate the real part of spin mixing conductance $g_{\it{r}}^{\uparrow \downarrow}$, we used the model which considered the spin memory loss (SML) mainly due to the interfacial roughness and disorder \[44\]. In this model effective spin mixing conductance is given by the following equation \[44\]: $$\label{q12} g_{\it{eff}}^{\uparrow \downarrow}= \frac{r_{sl}cosh(\delta)+r_{sN}^{\infty}coth(\frac{t_{Pt}}{\lambda_{Pt}})sinh(\delta)}{r_{sl}[1+0.5\sqrt{\frac{3}{\epsilon}}coth(\frac{t_{Pt}}{\lambda_{Pt}}]cosh(\delta)+[r_{sN}^{\infty}coth(\frac{t_{Pt}}{\lambda_{Pt}})+0.5\frac{r_{sl}^{2}}{r_{sN}^{\infty}}\sqrt{\frac{3}{\epsilon}}]sinh(\delta)}$$ where $\epsilon$ is the ratio of the spin conserved to spin flip relaxation times. According to \[25\], we set $\epsilon$ = 0.1 for the present Pt. $r_{sI}$, $r_{sN}^{\infty}$, $\delta$,$\lambda_{Pt}$ are the interfacial spin resistance, the Pt spin resistance, the spin flip parameter for the CFMS/Pt interface, and the spin diffusion length in Pt, respectively. Fig. \[fig6\] shows the fitting (solid line) of the data of $g_{eff}^{\uparrow \downarrow}$ using Eq. \[q12\]. The fitting gives the values of $\lambda_{Pt}$ = 7.5 $\pm$ 0.5 nm, $g_{r}^{\uparrow \downarrow}$= 1.77 $\pm$ 0.03 $\pm$ $10^{20}$ $m^{-2}$, and $\delta$ = 0.1. The values of $r_{sI}$ and $r_{sN}^{\infty}$ values are found to be 0.85 f $\it{\Omega} m^2$ and 0.58 f $\it{\Omega} m^2$, respectively, which are similar to the reported values for Co/Pt systems \[42\]. In CFMS/Pt system, the SML probability (\[1-exp(-$\delta$)\]$\times$100) is found to be 9.5 $\%$. It means the disorder at the interfaces is very small since spin depolarization is mainly caused by disorder at the interfaces. The interfacial spin resistance $r_{sI}$ is given by $r_b$ / $\delta$ , where $r_b$ is the interface resistance. It indicates that most of spin current will flow through interface compared to bulk SOC of Pt if SML probability is large. However, in our case SML probability is very small (9.5$\%$), it means that most of the spin current is dissipating through bulk SOC of Pt, which produces charge current and hence to create $V_{ISHE}$. Further, we compared the $g_{r}^{\uparrow \downarrow}$ and $g_{eff}^{\uparrow \downarrow}$ values evaluated in this work to the literature of various FM/HM systems in table III. [cccc]{} ----------- Layer structure ----------- : The values of the spin diffusion length ($\lambda_{Pt}$), spin mixing conductance ($g_{r}^{\uparrow \downarrow}$), effective mixing conductance ($g_{eff}^{\uparrow \downarrow}$) from the literature and in this work. & $\lambda_{Pt}$(nm) & $g_{r}^{\uparrow \downarrow}$($m^{-2}$) & $g_{eff}^{\uparrow \downarrow}$($m^{-2}$)\ -------------------------- $NiFe/Pt$ [\[]{}47[\]]{} -------------------------- : The values of the spin diffusion length ($\lambda_{Pt}$), spin mixing conductance ($g_{r}^{\uparrow \downarrow}$), effective mixing conductance ($g_{eff}^{\uparrow \downarrow}$) from the literature and in this work. & 7.7 & 1.13$\times 10^{20}$ & 3.02$\times 10^{19}$\ ------------------------- $NiFe/Pt$ [\[]{}6[\]]{} ------------------------- : The values of the spin diffusion length ($\lambda_{Pt}$), spin mixing conductance ($g_{r}^{\uparrow \downarrow}$), effective mixing conductance ($g_{eff}^{\uparrow \downarrow}$) from the literature and in this work. & 8.4 & 2.53$\times 10^{19}$ & 2.50$\times 10^{19}$\ ------------------------- $CoFe/Pt$ [\[]{}6[\]]{} ------------------------- : The values of the spin diffusion length ($\lambda_{Pt}$), spin mixing conductance ($g_{r}^{\uparrow \downarrow}$), effective mixing conductance ($g_{eff}^{\uparrow \downarrow}$) from the literature and in this work. & 7.6 & 2.65$\times 10^{19}$ & 2.50$\times 10^{19}$\ ----------------------- $Co/Pt$ [\[]{}6[\]]{} ----------------------- : The values of the spin diffusion length ($\lambda_{Pt}$), spin mixing conductance ($g_{r}^{\uparrow \downarrow}$), effective mixing conductance ($g_{eff}^{\uparrow \downarrow}$) from the literature and in this work. & 8.0 & 1.42$\times 10^{19}$ & 1.40$\times 10^{19}$\ $Y_3Fe_5O_{12}/Pt$ [\[]{}25[\]]{} & 7.3 & 3.90$\times 10^{18}$ & 6.90$\times 10^{18}$\ $CoFeB/Pt$ [\[]{}26[\]]{} & 1.7 & 3.90$\times 10^{19}$ & 3.90$\times 10^{19}$\ $Co_2MnSi/Pt$ [\[]{}15[\]]{} & 10.0 & ...... & 1.50$\times 10^{19}$\ ----------------------------- $Co_2Fe_{0.4}Mn_{0.6}Si/Pt$ [\[]{}This work[\]]{} ----------------------------- : The values of the spin diffusion length ($\lambda_{Pt}$), spin mixing conductance ($g_{r}^{\uparrow \downarrow}$), effective mixing conductance ($g_{eff}^{\uparrow \downarrow}$) from the literature and in this work. & 7.5 & 1.77 $\times 10^{20}$ & 4.21$\times 10^{19}$\ It can be observed from Table III that the values of $g_{r}^{\uparrow \downarrow}$ and $g_{eff}^{\uparrow \downarrow}$ are higher than the available literature values for the systems with Pt. Also, it should be noted here that the values of $g_{r}^{\uparrow \downarrow}$ and $g_{eff}^{\uparrow \downarrow}$ are large compared to the other reported low damping system viz. $Y_3Fe_5O_{12}/Pt$, $CoFeB/Pt$, and $Co_2MnSi/Pt$ \[43,44\]. Therefore, $CFMS/Pt$ system can be potential system for spin transfer torque and logic devices. In addition to the $g_{r}^{\uparrow \downarrow}$, spin interface transparency (T) is another parameter which is useful for spin-orbit torque-based devices. The value of T is affected by the electronics structure matching of FM and HM layers. We used the following expression for the calculation of T \[45\] $$\label{q13} T = \frac{g_{r}^{\uparrow \downarrow} tanh(\frac{t_{Pt}}{2 \lambda_{Pt}})}{g_{r}^{\uparrow \downarrow} coth(\frac{t_{Pt}}{\lambda_{Pt}})+\frac{h\sigma_{Pt}}{2e^2\lambda_{Pt}}}$$ where $\sigma_{Pt}$ is the conductivity of Pt layer. For $t_{Pt}$=20 nm, T is calculated to be 0.84 $\pm$ 0.02 by Eq.\[q13\], which is much higher than the values reported in the literature for $NiFe/Pt$ and $Co/Pt$ systems \[46\]. Further it is also higher than the recent low damping $Co_2FeAl$/Ta layers system (68 $\%$) \[46\]. It means that in $Co_2Fe_{0.4}Mn_{0.6}Si$/Pt system, matching of electronic structure is better than the other reported systems. We also calculated the $\Theta_{SH}$ for the Pt using the following expression \[2\]: $$\label{q14} \begin{aligned} J_s \approx (\frac{g_{eff}^{\uparrow \downarrow}\hbar}{8\pi})(\frac{\mu_0 h_{rf}\gamma}{\alpha})^2\times\\ [\frac{\mu_0 M_s\gamma+\sqrt{(\mu_0 M_s\gamma)^2+16(\pi f)^2}}{(\mu_0 M_s\gamma)^2+16(\pi f)^2}](\frac{2e}{\hbar}) \end{aligned}$$ $$\label{q15} V_{ISHE}=(\frac{w_y}{\sigma_{FM} t_{FM}+\sigma_{Pt} t_{Pt}})\times \theta_{SH} l_{sd}^{Pt}tanh(\frac{t_{Pt}}{2 \lambda_{Pt}}) J_s$$ The resistivity of the samples were measured using the four probe technique. The $\sigma_{Pt}$ and $\sigma_{CFMS}$ are found to be 2.3 $\times 10^{-7} \Omega$.m and 1.7$\times 10^{-6} \Omega$.m respectively. $\sigma$ corresponds to the conductivity of the individual layers. The rf field ($\mu_0 h_{rf}$) and CPW transmission line width ($w_y$) value for our set up are 0.05 mT (at 11 mW rf power) and 200 $\mu$m, respectively. The obtained values of $\theta_{SH}$ are plotted in the Fig. \[fig6\](b). The values of $\theta_{SH}$ are comparable to the literature value \[48\]. In our case, we are observing higher SHA value for sample S1 in comparison to the sample S5. It may be due to the higher resistivity of the 3 nm Pt layers, which is consistent to the results obtained by J. Liu $\it{et al.}$ \[47\]. Further, we also performed power dependence of $V_{ISHE}$ measurements (Fig. A5, supplementary information). We observed a linear dependence of $V_{ISHE }$, which confirmed the spin pumping at $CFMS/Pt$ interface. Conclusion ========== We presented the study of the spin pumping and inverse spin Hall effect measurements in $CFMS/Pt$ bilayer samples. Angle dependent measurements of voltage were measured to quantify the various spin rectification effects. We observed a strong dependency of spin pumping voltage on the thickness of Pt. The spin pumping voltage was decreased when the thickness of Pt was increased, which may be due to increase in conductivity of Pt with thickness. The presence of substantial spin pumping keeps the damping constant values in the order of $\sim10^{-3}$. Spin mixing conductance ( $g_{r}^{\uparrow \downarrow}$ ) was obtained to be 1.77 $\times 10^{20}$ $m^{-2}$, which was higher than those for the other reported FM/Pt systems. In addition, we observed highest spin interface transparency (84 $\%$) compared to any other FM/Pt system. Low magnetic damping and large value of $g_{r}^{\uparrow \downarrow}$ with high interface transparency make the $CFMS/Pt$ system as a potential candidate for spintronic applications. ACKNOWLEDGEMENT {#acknowledgement .unnumbered} =============== The authors acknowledge DAE and DST, Govt. of India, for the financial support for the experimental facilities. KR thanks CSIR for JRF fellowship. SB acknowledges ICC-IMR fellowship to visit IMR, Tohoku University, for this collaborative work to prepare the thin films. BBS acknowledges DST for INSPIRE faculty fellowship. REFERENCES {#references .unnumbered} ========== 1. S. D. Bader, and S. S. P. Parkin, Annual Review of Condensed Matter Physics 1, 71-88 (2010). 2. Y. Tserkovnyak, A. Brataas, G. E. W. Bauer, and B. I. Halperin, Rev. Mod. Phys. 77, 1375-1421 (2005). 3. C. O. Avci, K. Garello, A. Ghosh, M. Gabureac, S. F. Alvarado, and P. Gambardella, Nature Physics 11, 570 (2015). 4. Y.-T. Chen, S. Takahashi, H. Nakayama, M. Althammer, S.T.B. Goennenwein, E. Saitoh, and G. E. W. Bauer, Phys. Rev. B 87, 144411 (2013). 5. J. C. R. Sánchez, L. Vila, G. Desfonds, S. Gambarelli, J. P. Attan, J. M. De Teresa, C. Magn, and A. Fert, Nature Communications 4, 2944 (2013). 6. X. Tao, Q. Liu, B. Miao, R. Yu, Z. Feng, L. Sun, B. You, J. Du, K. Chen, S. Zhang, L. Zhang, Z. Yuan, D. Wu, and H. Ding, Science Advances 4, eaat1670 (2018). 7. B. Heinrich, C. Burroers, E. Montoya, B. Kardasz, E. Girt, Y.-Y. Song, Y. Sun, and M. Wu, Phys. Rev. Lett. 107, 066604 (2011). 8. M. Obstbaum, M. Decker, A. K. Greitner, M. Haertinger, T. N. G. Meier, M. Kronseder, K. Chadova, S. Wimmer, D. Kdderitzsch, H. Ebert, and C. H. Back, Phys. Rev. Lett. 117, 167204 (2016). 9. K. Chen, and S. Zhang, Phys. Rev. Lett. 114, 126602 (2015). 10. K. Uchida, S. Takahashi, K. Harii, J. Ieda, E. Koshibae, K. Ando, S. Maekawa, and E. Saitoh, Nature 455, 778–781 (2008). 11. D. Meier, D. Reinhardt, M. van Straaten, C. Klewe, M. Althammer, M. Schreier, S. T. B. Goennenwein, A. Gupta, M. Schmid, C. H. Back, J.-M. Schmalhorst, T. Kuschel, and G. Reiss, Nature Communications 6, 8211 (2015). 12. W. Lin, K. Chen,S. Zhang, and C. L. Chien, Phys. Rev. Lett. 116, 186601 (2016). 13. E. Saitoh, M. Ueda,H. Miyajima, and G. Tatara, Appl. Phys. Lett. 88, 182509 (2006). 14. Y. Tserkovnyak, A. Brataas, and G. E. W.Bauer, Phys. Rev. Lett. 88, 117601 (2002). 15. H. Chudo, K. Ando, K. Saito, S. Okayasu, R. Haruki, Y. Sakuraba, H. Yasuoka, K. Takanashi, and E. Saitoh, Journal of Applied Physics 109, 073915 (2011). 16. V. E. Demidov, S. Urazhdin, G. de Loubens, O. Klein, V. Cros, A. Anane, and S.O. Demokritov, Physics Reports 673, 1–31 (2017). 17. S. Mizukami, Y. Ando, and T. Miyazaki, Phys. Rev. B 66, 104413 (2002). 18. B. F. Miao, S. Y. Huang, D. Qu, and C. L. Chien, Phys. Rev. Lett. 111, 066602 (2013). 19. L. Chen,S. Ikeda, F. Matsukura, and H. Ohno, Appl. Phys. Express 7, 013002 (2013). 20. S.-I. Kim, M.-S. Seo, Y. S. Choi, and S.-Y. Park, Journal of Magnetism and Magnetic Materials 421, 189–193 (2017). 21. M. Cecot, Karwacki, W. Skowroski, J. Kanak, J. Wrona, A. Ywczak, L. Yao, S. van Dijken, J. Barna, and T. Stobiecki, Sci Rep 7, 1–11 (2017). 22. I. Gościańska, and J. Dubowik, Acta Physica Polonica A 118, 851–853 (2010). 23. A. Hirohata, and K. Takanashi, J. Phys. D: Appl. Phys. 47, 193001 (2014). 24. S. Pan, S. Mondal, T. Seki, K. Takanashi, and A. Barman, Phys. Rev. B 94, 184417 (2016). 25. NanOsc AB. NanOsc AB Available at: http://www.nanosc.se/. (Accessed: 1st September 2019) 26. B. B. Singh, S. K. Jena, and S. Bedanta, J. Phys. D: Appl. Phys. 50, 345001 (2017). 27. B. B. Singh, S. K. Jena, M. Samanta, K. Biswas, B. Satpati, and S. Bedanta, physica status solidi (RRL) – Rapid Research Letters 13, 1800492 (2019). 28. C. Kittel, Phys. Rev. 73, 155–161 (1948). 29. B. Heinrich, J. F. Cochran, and R. Hasegawa, Journal of Applied Physics 57, 3690–3692 (1985). 30. A. Conca, S. Keller, L. Mihalceanu, T. Kehagias, G. P. Dimitrakopulos, B. Hillebrands, and E. Th. Papaioannou, Phys. Rev. B 93, 134405 (2016). 31. W. Amamou, I. V. Pinchul, A. H. Trout, R. E. A. Williams, N. Antolin, A. Goad, D. J. Ohara , A. S. Ahmed, W. Windl, D. W. McComb, and R. K. Kawakami, Phys. Rev. Materials 2, 011401 (2018). 32. X. Liang, G. Shi, L. Deng, F. Huang, J. Qin, T. Tang, C. Wang, B. Peng, C. Song, and L. Bi, Phys. Rev. Applied 10, 024051 (2018). 33. S. Mallick, S. Mondal, T. Seki, S. Sahoo, T. Forrest, F. Maccherozzi, Z. Wen, S. Barman, A. Barman, K. Takanashi, and S. Bedanta, Phys. Rev. Applied 12, 014043 (2019). 34. K. Ando, S. Takanashi, J. Ieda, Y. Kajiwara, H. Nakayama, T. Yoshino, K. Harii, Y. Fujikawa, M. Matsuo, S. Maekawa, and E. Saitoh, Journal of Applied Physics 109, 103913 (2011). 35. R. Iguchi, and E. Saitoh, J. Phys. Soc. Jpn. 86, 011003 (2016). 36. A. Conca, B. Heinz, M.R. Schweizer, S. Keller, E. Th. Papaioannou, and B. Hillebrands, Phys. Rev. B 95, 174426 (2017). 37. M. Harder, Y. Gui, and C.-M. Hu, Physics Reports 661, 1–59 (2016). 38. M. Harder, Z. X. Cao, Y. S. Gui, X. L. Fan, and C.-M. Hu, Phys. Rev. B 84, 054423 (2011). 39. N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Rev. Mod. Phys. 82, 1539–1592 (2010). 40. W. Zhang, V. Vlaminck , J. E. Pearson. R. Divan , S, D. Bader, and A. Hoffmann, Appl. Phys. Lett. 103, 242414 (2013). 41. G. Zahnd, L. Vila, V. T. Pham, M. Cosset-Cheneau, W. Lim, A. Brenac, P. Laczkowski, A. Marty, and J. P. Attan, Phys. Rev. B 98, 174414 (2018). 42. J. C. Rojas-Sánchez, N. Reyren, P. Laczkowski, W. Savero, J.-P. Attan, C. Deranlot, M. Jamet, J.-M. George, L. Vila, and H. Jaffrs, Phys. Rev. Lett. 112, 106602 (2014).\ 43. H. Wang, Graduate dissertion, The Ohio state University (2015). 44. M. Belmeguenai, K. Aitoukaci, F. Zighem, M. S. Gabor, T. Petrisor, R. B. Mos, and C. Tiusan, Journal of Applied Physics 123, 113905 (2018). 45. W. Zhang, W. Han, X. Jiang, S.-H. Yang and S. S. P. Parkin, Nature Physics 11, 496–502 (2015). 46. S. Akansel, A. Kumar, N. Behera, S. Husain, R. Brucas, S. Chaudhary, and P. Svedlindh, Phys. Rev. B 97, 134421 (2018). 47. J. Liu, T. Ohkubo, S. Mitani, K. Hono, and M. Hayashi, Appl. Phys. Lett. 107, 232408 (2015). 48. H. Nakayama, K. Ando, K. Harii, T. Yoshino, R. Takahashi, Y. Kajiwara, K. Uchida, Y, Fujikawa, and E. Saitoh, Phys. Rev. B 85, 144408 (2012).	{ "pile_set_name": "ArXiv" }
--- abstract: 'We discuss a deformation of superspace based on a hermitian twist. The twist implies a $\star$-product that is noncommutative, hermitian and finite when expanded in power series of the deformation parameter. The Leibniz rule for the twisted SUSY transformations is deformed. A minimal deformation of the Wess-Zumino action is proposed and its renormalizability properties are discussed. There is no tadpole contribution, but the two-point function diverges. We speculate that the deformed Leibniz rule, or more generally the twisted symmetry, interferes with renormalizability properties of the model. We discuss different possibilities to render a renormalizable model.' --- $\,$ [[Marija Dimitrijevi' c, Biljana Nikoli' c and\ Voja Radovanovi' c ]{}]{} University of Belgrade, Faculty of Physics\ Studentski trg 12, 11000 Beograd, Serbia\ [Keywords:]{}[ supersymmetry, hermitian twist, deformed Wess-Zumino model, renormalizability]{} .2cm Introduction ============ It is well known that Quantum Field Theory encounters problems at high energies and short distances. This suggests that the structure of space-time has to be modified at these scales. One possibility to modify the structure of space-time is to deform the usual commutation relations between coordinates; this gives a noncommutative (NC) space [@noncommspace]. Different models of noncommutativity were discussed in the literature, see [@NCbooks], [@NCbookMi] and [@NCreview] for references. A version of Standard Model on the canonically deformed space-time was constructed in [@NCSM] and its renormalizability properties were discussed in [@NCSMRenorm]. Renormalizability of different noncommutative field theory models was discussed in [@NCRenorm]. A natural further step is modification of the superspace and introduction of non(anti)commutativity. A strong motivation for this comes from string theory. Namely, it was discovered that a noncommutative superspace can arise when a superstring moves in a constant gravitino or graviphoton background [@Seiberg], [@NACStrings]. Since that discovery there has been a lot of work on this subject and different ways of deforming superspace have been discussed. Here we mention some of them. The authors of [@luksusy] combine SUSY with the $\kappa$-deformation of space-time, while in [@MWsusy] SUSY is combined with the canonical deformation of space-time. In [@Seiberg] a version of non(anti)commutative superspace is defined and analyzed. The anticommutation relations between fermionic coordinates are modified in the following way $$\{ \theta^\alpha \ds \theta^\beta\} =C^{\alpha\beta} , \quad \{ \bar{\theta}_{\dot\alpha} \ds \bar{\theta}_{\dot\beta}\} = \{ \theta^\alpha \ds \bar{\theta}_{\dot\alpha}\} = 0\ , \label{eucliddef}$$ where $C^{\alpha\beta} = C^{\beta\alpha}$ is a complex, constant symmetric matrix. This deformation is well defined only when undotted and dotted spinors are not related by the usual complex conjugation. The notion of chirality is preserved in this model, i.e. the deformed product of two chiral superfields is again a chiral superfield. On the other hand, one half of ${\cal N}=1$ supersymmetry is broken and this is the so-called ${\cal N}=1/2$ supersymmetry. Another type of deformation is introduced in [@Ferrara] and [@D-def-us]. There the product of two chiral superfields is not a chiral superfield but the model is invariant under the full supersymmetry. Renormalizability of different models (both scalar and gauge theories) has been discussed in [@RenWZ], [@Penati], [@RenYM] and [@D-def-us]. The twist approach to nonanticommutativity was discussed in [@SUSYtwist]. In our previous paper [@miSUSY] we introduced a hermitian deformation of the usual superspace. The non(anti)commutative deformation was introduced via the twist $${\cal F} = e^{\frac{1}{2}C^{\alpha\beta}\p_\alpha \otimes\p_\beta + \frac{1}{2}\bar{C}_{\dot{\alpha}\dot{\beta}} \bar{\p}^{\dot{\alpha}}\otimes\bar{\p}^{\dot{\beta}} } .\label{intro-twist}$$ Here $C^{\alpha\beta} = C^{\beta\alpha}$ is a complex constant matrix, $\bar{C}^{\dot{\alpha}\dot{\beta}}$ its complex conjugate and $\p_\alpha = \frac{\p}{\p \theta^\alpha}$ are fermionic partial derivatives. The twist (\[intro-twist\]) is hermitian under the usual complex conjugation. Due to this choice of the twist, the coproduct of the SUSY transformations becomes deformed, leading to the deformed Leibniz rule. The inverse of (\[intro-twist\]) defines the $\star$-product. It is obvious that the $\star$-product of two chiral fields will not be a chiral field. Therefore we have to use the method of projectors to decompose the $\star$-products of fields into their irreducible components. Collecting the terms invariant under the twisted SUSY transformations we construct the deformed Wess-Zumino action. Being interested in implications of the twisted symmetry on renormalizability properties, in this paper we calculate the divergent part of the one-loop effective action. More precisely, we calculate divergent parts of the one-point and the two-point functions. The plan of the paper is as follows: In the next section we summarize the most important properties of our model, more details of the construction are given in [@miSUSY]. In Section 3 we describe the method we use to calculate divergent parts of the $n$-point Green functions: the background field method and the supergraph technique. In Sections 4 the tadpole diagram and the divergent part of the two-point function are calculated. Finally, we discuss renormalizability of the model. We give some comments and compare our results with the results already present in the literature. Some details of our calculations are presented in the Appendix. Construction of the model ========================= There are different ways to realize a noncommutative and/or a nonanticommutative space and to formulate a physical model on it, see [@NCbooks] and [@NCreview]. We shall follow the approach of [@NCbookMi] and [@miSUSY]. Let us first fix the notation and the conventions which we use. The superspace is generated by supercoordinates $x^{m}$, $\theta^{\alpha}$ and $\bar{\theta}_{\dot{\alpha}}$ which fulfill $$\begin{aligned} \lbrack x^m, x^n \rbrack = \lbrack x^m, \theta^\alpha \rbrack = \lbrack x^m, \bar{\theta}_{\dot\alpha} \rbrack = 0 ,\quad \{ \theta^\alpha , \theta^\beta\} = \{ \bar{\theta}_{\dot\alpha} , \bar{\theta}_{\dot\beta}\} = \{ \theta^\alpha , \bar{\theta}_{\dot\alpha}\} = 0 , \label{undefsupsp}\end{aligned}$$ with $m=0,\dots ,3$ and $\alpha, \beta =1,2$. To $x^m$ we refer as bosonic and to $\theta^\alpha$ and $\bar{\theta}_{\dot\alpha}$ we refer as fermionic coordinates. We work in Minkowski space-time with the metric $(-,+,+,+)$ and $x^m x_m = -(x^0)^2 + (x^1)^2 + (x^2)^2 + (x^3)^2$. A general superfield ${\rm F}(x,\theta, \bar{\theta})$ can be expanded in powers of $\theta$ and $\bar{\theta}$, $$\begin{aligned} {\rm F}(x, \theta, \bar{\theta}) &=&\hspace{-2mm} f(x) + \theta\phi(x) + \bar{\theta}\bar{\chi}(x) + \theta\theta m(x) + \bar{\theta}\bar{\theta} n(x) + \theta\sigma^m\bar{\theta}v_m(x)\nonumber\\ && + \theta\theta\bar{\theta}\bar{\lambda}(x) + \bar{\theta}\bar{\theta}\theta\varphi(x) + \theta\theta\bar{\theta}\bar{\theta} d(x) .\label{F}\end{aligned}$$ Under the infinitesimal ${\cal N} = 1$ SUSY transformations it transforms as $$\delta_\xi {\rm F} = \big(\xi Q + \bar{\xi}\bar{Q} \big) {\rm F}, \label{susytr}$$ where $\xi^{\alpha}$ and $\bar{\xi}_{\dot{\alpha}}$ are constant anticommuting parameters and the SUSY generators $Q^{\alpha}$ and $\bar{Q}_{\dot\alpha}$ are given by, $$\begin{aligned} Q_\alpha = \p_\alpha - i\sigma^m_{\ \alpha\dot{\alpha}}\bar{\theta}^{\dot{\alpha}}\p_m, \quad \bar{Q}_{\dot{\alpha}} = -\bar{\p}_{\dot{\alpha}} + i\theta^\alpha \sigma^m_{\ \alpha\dot{\alpha}}\p_m .\label{q,barq}\end{aligned}$$ Transformations (\[susytr\]) close in the algebra $$[\delta_\xi, \delta_\eta] = -2i(\eta\sigma^m \bar{\xi} - \xi\sigma^m \bar{\eta})\p_m. \label{xietaalg}$$ The product of two superfields is a superfield again; its transformation law is given by $$\begin{aligned} \delta_\xi ({\rm F}\cdot {\rm G}) &=& \big(\xi Q + \bar{\xi}\bar{Q} \big) ({\rm F}\cdot {\rm G}), \nonumber\\ &=& (\delta_\xi {\rm F})\cdot {\rm G} + {\rm F}\cdot(\delta_\xi {\rm G}). \label{undefLrule}\end{aligned}$$ The last line is the undeformed Leibniz rule for the infinitesimal SUSY transformation $\delta_\xi$. Nonanticommutativity is introduced following the twist approach [@NCbookMi]. For the twist $\cal{F}$ we choose $${\cal F} = e^{\frac{1}{2}C^{\alpha\beta}\p_\alpha \otimes\p_\beta + \frac{1}{2}\bar{C}_{\dot{\alpha}\dot{\beta}}\bar{\p}^{\dot{\alpha}}\otimes\bar{\p}^{\dot{\beta}} } ,\label{twist}$$ with the complex constant matrix $C^{\alpha\beta} = C^{\beta\alpha}$. Note that $C^{\alpha\beta}$ and $\bar{C}^{\dot{\alpha}\dot{\beta}}$ are related by the usual complex conjugation. It can be shown that the twist (\[twist\]) satisfies all necessary requirements [@chpr]. The inverse of the twist (\[twist\]) $${\cal F}^{-1} = e^{-\frac{1}{2}C^{\alpha\beta}\p_\alpha \otimes\p_\beta - \frac{1}{2}\bar{C}_{\dot{\alpha}\dot{\beta}}\bar{\p}^{\dot{\alpha}}\otimes\bar{\p}^{\dot{\beta}} } ,\label{invtwist}$$ defines a new product in the algebra of superfields called the $\star$-product. For two arbitrary superfields ${\rm F}$ and ${\rm G}$ the $\star$-product is defined as follows $$\begin{aligned} {\rm F}\star {\rm G} &=& \mu_\star \{ {\rm F}\otimes {\rm G} \} \nonumber\\ &=& \mu \{ {\cal F}^{-1}\, {\rm F}\otimes {\rm G}\} \nonumber\\ &=& \mu \{ e^{-\frac{1}{2}C^{\alpha\beta}\p_\alpha \otimes\p_\beta - \frac{1}{2}\bar{C}_{\dot{\alpha}\dot{\beta}}\bar{\p}^{\dot{\alpha}} \otimes\bar{\p}^{\dot{\beta}}} {\rm F}\otimes {\rm G} \} \nonumber\\ &=& {\rm F}\cdot {\rm G} - \frac{1}{2}(-1)^{\|{\rm F}\|}C^{\alpha\beta} (\p_\alpha {\rm F})\cdot(\p_\beta {\rm G}) - \frac{1}{2}(-1)^{\|{\rm F}\|}\bar{C}_{\dot{\alpha}\dot{\beta}} (\bar{\p}^{\dot{\alpha}} {\rm F}) (\bar{\p}^{\dot{\beta}} {\rm G})\nonumber\\ &&- \frac{1}{8}C^{\alpha\beta}C^{\gamma\delta}(\p_\alpha\p_\gamma {\rm F}) \cdot(\p_\beta\p_\delta {\rm G}) - \frac{1}{8}\bar{C}_{\dot{\alpha}\dot{\beta}}\bar{C}_{\dot{\gamma}\dot{\delta}} (\bar{\p}^{\dot{\alpha}}\bar{\p}^{\dot{\gamma}} {\rm F}) (\bar{\p}^{\dot{\beta}}\bar{\p}^{\dot{\delta}} {\rm G}) \nonumber\\ &&- \frac{1}{4}C^{\alpha\beta}\bar{C}_{\dot{\alpha}\dot{\beta}} (\p_\alpha\bar{\p}^{\dot{\alpha}}{\rm F}) (\p_\beta\bar{\p}^{\dot{\beta}} {\rm G}) \nonumber\\ && + \frac{1}{16}(-1)^{\|{\rm F}\|} C^{\alpha\beta}C^{\gamma\delta}\bar{C}_{\dot{\alpha}\dot{\beta}} (\p_\alpha\p_\gamma\bar{\p}^{\dot{\alpha}} {\rm F}) (\p_\beta\p_\delta\bar{\p}^{\dot{\beta}} {\rm G}) \nonumber\\ && + \frac{1}{16}(-1)^{\|{\rm F}\|}C^{\alpha\beta}\bar{C}_{\dot{\alpha} \dot{\beta}} \bar{C}_{\dot{\gamma}\dot{\delta}} (\p_\alpha\bar{\p}^{\dot{\alpha}}\bar{\p}^{\dot{\gamma}} {\rm F}) (\p_\beta\bar{\p}^{\dot{\beta}}\bar{\p}^{\dot{\delta}} {\rm G}) \nonumber\\ && + \frac{1}{64}C^{\alpha\beta}C^{\gamma\delta} \bar{C}_{\dot{\alpha}\dot{\beta}}\bar{C}_{\dot{\gamma}\dot{\delta}} (\p_\alpha\p_\gamma\bar{\p}^{\dot{\alpha}}\bar{\p}^{\dot{\gamma}} {\rm F})(\p_\beta\p_\delta \bar{\p}^{\dot{\beta}}\bar{\p}^{\dot{\delta}} {\rm G}), \label{star}\end{aligned}$$ where $\|{\rm F}\| = 1$ if ${\rm F}$ is odd (fermionic) and $\|{\rm F}\|=0$ if ${\rm F}$ is even (bosonic) and the pointwise multiplication $\mu$ is the bilinear map from the tensor product to the space of superfields (functions). The definition of the multiplication $\mu_\star$ is given in the second line. No higher powers of $C^{\alpha\beta}$ and $\bar{C}_{\dot{\alpha}\dot{\beta}}$ appear since the derivatives $\p_\alpha$ and $\bar{\p}^{\dot{\alpha}}$ are Grassmanian. Expansion of the $\star$-product (\[star\]) ends after the fourth order in the deformation parameter. This $\star$-product is different from the Moyal-Weyl $\star$-product [@mw] where the expansion in powers of the deformation parameter leads to an infinite power series. One should also note that the $\star$-product (\[star\]) is hermitian, $$({\rm F}\star {\rm G})^ = {\rm G}^* \star {\rm F}^* , \label{complconj}$$ where $$ denotes the usual complex conjugation. The $\star$-product (\[star\]) implies $$\begin{aligned} \{ \theta^\alpha \ds \theta^\beta \} &=& C^{\alpha\beta}, \quad \{ \bar{\theta}_{\dot\alpha}\ds \bar{\theta}_{\dot\beta}\} = \bar{C}_{\dot{\alpha}\dot{\beta}}, \quad \{ \theta^\alpha \ds \bar{\theta}_{\dot\alpha} \} = 0 ,\nonumber \\ \lbrack x^m \ds x^n \rbrack &=& 0 , \quad [x^m \ds \theta^\alpha ] = 0, \quad [x^m \ds \bar{\theta}_{\dot\alpha} ] = 0 . \label{thetastar}\end{aligned}$$ Relations (\[thetastar\]) enable us to define the deformed superspace or “nonanticommutative superspace”. It is generated by the usual bosonic and fermionic coordinates (\[undefsupsp\]) while the deformation is contained in the new product (\[star\]). The next step is to apply the twist (\[twist\]) to the Hopf algebra of SUSY transformations. We will not give details here, they can be found in [@miSUSY]. We just state the most important results. The deformed infinitesimal SUSY transformation is defined in the following way $$\delta^\star_\xi {\rm F} = \big(\xi Q + \bar{\xi}\bar{Q} \big) {\rm F}. \label{defsusytr}$$ The twist (\[twist\]) leads to a deformed Leibniz rule for the deformed SUSY transformations (\[defsusytr\]). This ensures that the $\star$-product of two superfields is again a superfield. Its transformation law is given by $$\begin{aligned} \delta^\star_\xi ({\rm F}\star {\rm G}) &=& \big(\xi Q + \bar{\xi}\bar{Q} \big) ({\rm F}\star {\rm G}), \label{deftrlaw}\\ &=& (\delta^\star_\xi {\rm F}) \star {\rm G} + {\rm F} \star (\delta^\star_\xi {\rm G}) \nonumber\\ && + \frac{i}{2}C^{\alpha\beta}\Big( \bar{\xi}^{\dot{\gamma}} \sigma^m_{\ \alpha\dot{\gamma}}(\p_m {\rm F})\star (\p_\beta {\rm G}) + (\p_\alpha {\rm F})\star \bar{\xi}^{\dot{\gamma}} \sigma^m_{\ \beta\dot{\gamma}}(\p_m {\rm G}) \Big) \label{defLpravilo}\\ && -\frac{i}{2}\bar{C}_{\dot{\alpha}\dot{\beta}} \Big( \xi^\alpha\sigma^m_{\ \alpha\dot{\gamma}}\varepsilon^{\dot{\gamma}\dot{\alpha}}(\p_m {\rm F}) \star (\bar{\p}^{\dot{\beta}} {\rm G}) + (\bar{\p}^{\dot{\alpha}} {\rm F}) \star \xi^\alpha\sigma^m_{\ \alpha\dot{\gamma}} \varepsilon^{\dot{\gamma}\dot{\beta}}(\p_m {\rm G}) \Big) .\nonumber\end{aligned}$$ Note that we have to enlarge the algebra (\[xietaalg\]) by introducing the fermionic derivatives $\p_\alpha$ and $\bar{\p}_{\dot{\alpha}}$. Since these derivatives commute with the generators of Poincar' e algebra $\p_m$ and $M_{mn}$, the super Poincar' e algebra does not change. Especially, the Leibniz rule for $\p_m$ and $M_{mn}$ does not change. Being interested in a deformation of the Wess-Zumino model, we need to analyze properties of the $\star$-products of chiral fields. A chiral field $\Phi$ fulfills $\bar{D}_{\dot{\alpha}}\Phi =0$, with the supercovariant derivative $\bar{D}_{\dot{\alpha}} = -\bar{\p}_{\dot{\alpha}} - i\theta^\alpha \sigma^m_{\ \alpha\dot{\alpha}}\p_m$. In terms of component fields the chiral superfield $\Phi$ is given by $$\begin{aligned} \Phi(x, \theta, \bar{\theta}) &=& A(x) + \sqrt{2}\theta^\alpha\psi_\alpha(x) + \theta\theta F(x) + i\theta\sigma^l\bar{\theta}\p_l A(x) \nonumber\\ && -\frac{i}{\sqrt{2}}\theta\theta\p_m\psi^\alpha(x) \sigma^m_{\ \alpha\dot{\alpha}}\bar{\theta}^{\dot{\alpha}} + \frac{1}{4}\theta\theta\bar{\theta}\bar{\theta}\Box A(x).\label{chiral}\end{aligned}$$ It is easy to calculate the $\star$-product of two chiral fields from (\[star\]). It is given by $$\begin{aligned} \Phi\star\Phi &=& A^2 - \frac{C^2}{2}F^2 + \frac{1}{4}C^{\alpha\beta}\bar{C}^{\dot{\alpha}\dot{\beta}} \sigma^m_{\ \alpha\dot{\alpha}}\sigma^l_{\ \beta\dot{\beta}}(\p_m A)(\p_l A) +\frac{1}{64}C^2\bar{C}^2 (\Box A)^2\nonumber\\ && + \theta^\alpha\Big( 2\sqrt{2}\psi_\alpha A -\frac{1}{\sqrt{2}}C^{\gamma\beta}\bar{C}^{\dot{\alpha}\dot{\beta}} \varepsilon_{\gamma\alpha} (\p_m\psi^\rho)\sigma^m_{\ \rho\dot{\beta}}\sigma^l_{\ \beta\dot{\alpha}} (\p_l A) \Big) \nonumber\\ && -\frac{i}{\sqrt{2}}C^2 \bar{\theta}_{\dot{\alpha}} \bar{\sigma}^{m\dot{\alpha}\alpha}(\p_m\psi_\alpha)F + \theta\theta \Big( 2AF - \psi\psi\Big) \nonumber\\ && + \bar{\theta}\bar{\theta}\Big( -\frac{C^2}{4}\big( F\Box A - \frac{1}{2}(\p_m\psi)\sigma^m\bar{\sigma}^l(\p_l\psi) \big) \Big) \nonumber\\ && + i\theta\sigma^m\bar{\theta}\Big( (\p_m A^2) + \frac{1}{4}C^{\alpha\beta}\bar{C}^{\dot{\alpha}\dot{\beta}} \sigma_{m\alpha\dot{\alpha}}\sigma^l_{\ \beta\dot{\beta}}(\Box A) (\p_l A)\Big) \nonumber\\ && + i\sqrt{2}\theta\theta\bar{\theta}_{\dot{\alpha}} \bar{\sigma}^{m\dot{\alpha}\alpha} \big( \p_m(\psi_\alpha A)\big) + \frac{1}{4}\theta\theta\bar{\theta} \bar{\theta} (\Box A^2) , \label{phistarphi}\end{aligned}$$ where $C^2 = C^{\alpha\beta}C^{\gamma\delta}\varepsilon_{\alpha\gamma}\varepsilon_{\beta\delta}$ and $\bar{C}^2 = \bar{C}_{\dot{\alpha}\dot{\beta}}\bar{C}_{\dot{\gamma}\dot{\delta}} \varepsilon^{\dot{\alpha}\dot{\gamma}}\varepsilon^{\dot{\beta}\dot{\delta}}$. One sees that due to the $\bar{\theta}$-term and the $\bar{\theta}\bar{\theta}$-term (\[phistarphi\]) is not a chiral field. But, in order to write an action invariant under the deformed SUSY transformations (\[defsusytr\]) we need to preserve the notion of chirality. This can be done in different ways. One possibility is to use a different $\star$-product, the one which preserves chirality [@Seiberg]. However, chirality-preserving $\star$-product implies working in a space where $\bar{\theta} \neq (\theta)^$. Since we want to work in Minkowski space-time and keep the usual complex conjugation, we use the $\star$-product (\[star\]) and decompose the $\star$-products of superfields into their irreducible components using the projectors defined in [@wessbook]. In that way (\[phistarphi\]) becomes $$\Phi\star \Phi = P_1(\Phi\star \Phi) + P_2 (\Phi\star \Phi) + P_T(\Phi\star \Phi),$$ with antichiral, chiral and transversal projectors given by $$\begin{aligned} P_1 &=& \frac{1}{16} \frac{D^2 \bar{D}^2}{\Box}, \label{P1}\\ P_2 &=& \frac{1}{16} \frac{\bar{D}^2 D^2 }{\Box}, \label{P2}\\ P_T &=& -\frac{1}{8} \frac{D \bar{D}^2 D}{\Box}. \label{PT}\end{aligned}$$ Finally, the deformed Wess-Zumino action is constructed requiring that the action is invariant under the deformed SUSY transformations (\[defsusytr\]) and that in the commutative limit it reduces to the undeformed Wess-Zumino action. In addition, we require that deformation is minimal: We deform only those terms that are present in the commutative Wess-Zumino model. We do not, for the time being, add the terms whose commutative limit is zero. Taking these requirements into account we propose the following action $$\begin{aligned} S &=& \int {\mbox{d}}^4 x\hspace{1mm} \Big{\{} \Phi^+\star\Phi\Big\|_{\theta\theta\bar\theta\bar\theta} + \Big[ \frac{m}{2}P_2\big( \Phi\star\Phi\big) \Big\|_{\theta\theta} \nonumber\\ && + \frac{\lambda}{3}P_2\Big( \Phi\star P_2\big( \Phi\star\Phi\big)\Big) \Big \|_{\theta\theta} + {\mbox{ c.c }} \Big] \Big{\}}, \label{S}\end{aligned}$$ where $m$ and $\lambda$ are real constants. To rewrite (\[S\]) in terms of component fields and as compact as possible, we introduce the following notation $$\begin{aligned} C_{\alpha\beta}&=& K_{ab}(\sigma^{ab}\varepsilon)_{\alpha\beta},\label{n1}\\ \bar{C}_{\dot{\alpha}\dot{\beta}}&=& K^_{ab}(\varepsilon\bar{\sigma}^{ab})_{\dot{\alpha}\dot{\beta}},\label{n2}\end{aligned}$$ where $K_{ab}=-K_{ba}$ is an antisymmetric self-dual complex constant matrix. Then we have $$\begin{aligned} C^2 = 2K_{ab}K^{ab}, && \bar{C}^2 = 2K^_{ab}K^{ab}, \quad K^{ab}K^_{ab}= 0.\label{n4} \\ K^_{cd}K_{ab}\big( \sigma^n\bar{\sigma}^{cd}\bar{\sigma}^m \sigma^{ab}\big)_\alpha^{\ \beta} &=& -4\delta^\beta_\alpha K^{ma}K^{n}_{\ \ a} + 8 K^{ma}K^{nb}(\sigma_{ba})_\alpha^{\ \beta} ,\label{n5}\\ C^{\alpha\beta}\bar{C}^{\dot{\alpha}\dot{\beta}} \sigma^m_{\ \alpha\dot{\alpha}}\sigma^l_{\ \beta\dot{\beta}} &=& 8K^{am}K^{\ l}_a. \label{n6}\end{aligned}$$ Using these formulas and expanding (\[S\]) in component fields we obtain $$\begin{aligned} S &=& \int {\mbox{d}}^4 x\hspace{1mm} \Big{\{} \Phi^+\star\Phi\Big\|_{\theta\theta\bar\theta\bar\theta}\nonumber\\ && + \Big[ \frac{m}{2}P_2\big( \Phi\star\Phi\big) \Big\|_{\theta\theta} + \frac{\lambda}{3}P_2\Big( \Phi\star P_2\big( \Phi\star\Phi\big)\Big) \Big \|_{\theta\theta} + {\mbox{ c.c }} \Big] \Big{\}}\nonumber\\ &=& \int {\mbox{d}}^4 x\hspace{1mm} \Big{\{} A^\Box A + i(\p_m\bar\psi)\bar{\sigma}^m\psi + F^F\nonumber\\ && + \Big[ \frac{m}{2}\big( 2AF - \psi\psi \big) +\lambda\big( FA^2 - A\psi\psi \big)\nonumber\\ && -\frac{\lambda}{3}\Big(K^{m}_{\ a}K^{na}\psi(\p_n\psi) - 2K^m_{\ a}K^{n}_{\ \ b}(\p_n\psi)\sigma^{ba}\psi\Big) (\p_m A)\nonumber\\ && -\frac{\lambda}{12}K^{mn}K_{mn}F^3 +\frac{\la}{6}\KK F(\p_m A)(\p_n A) \label{Sincompfields}\\ && +\frac{\la}{3}\KK F\frac{1}{\Box}\p_m\Big( (\p_n A)\Box A\Big) +\frac{\la}{192}\KKt\KKtc F(\Box A)^2 + c.c \Big] \Big{\}}. \nonumber\end{aligned}$$ Partial integration was used to rewrite some of the terms in (\[Sincompfields\]) in a more compact way. Note that this it the complete action; there are no higher order terms in the deformation parameter $K^{ab}$. However, for simplicity in the following sections we shall keep only terms up to second order in the deformation parameter. One-loop effective action ========================= In this section we look at the quantum properties of our model. We calculate the one-loop divergent part of the one-point and the two-point functions up to second order in the deformation parameter. We use the background field method, dimensional regularization and the supergraph technique. The supergraph technique significantly simplifies calculations. However, we cannot directly apply this technique since our action (\[Sincompfields\]) is not written as an integral over the whole superspace and in terms of the chiral field $\Phi$ and its derivatives. This is a consequence of the particular deformation (\[F\]) and differs from [@D-def-us]. In order to be able to use the supergraph technique we notice the following: From (\[chiral\]), see also [@wessbook], it follows that the fields $A$, $\psi$ and $F$ can be written as $$A=\Phi\|_{\theta, \bar{\theta} =0}, \quad \psi_{\alpha}=\frac{1}{\sqrt{2}}D_{\alpha}\Phi\|_{\theta, \bar{\theta} =0}, \quad F=-\frac{1}{4}D^{2}\Phi\|_{\theta, \bar{\theta} =0}. \label{cfieldsPhi}$$ Inserting this in (\[Sincompfields\]) we obtain $$\begin{aligned} S &=& \int {\mbox{d}}^8 z\hspace{1mm} \Big{\{} \Phi^+\Phi + \Big[ -\frac{m}{8}\Phi\frac{D^2}{\Box}\Phi -\lambda\Phi^2\frac{D^2}{12\Box} \Phi\nonumber\\ && +\lambda\theta\theta\bar{\theta}\bar{\theta} \Big( \frac{1}{768} K^{mn}K_{mn}(D^{2}\Phi)^{3} \nonumber \\ & &-\frac{1}{6}\left(K^{m}_{\ a}K^{na} (D^{\alpha}\Phi)(\partial_{n}D_{\alpha}\Phi)- 2K^{m}_{\ a} K^{n}_{\ b} (\partial_{n}D^{\alpha}\Phi)(\sigma^{ba})_{\alpha}^{\ \beta}D_{\beta}\Phi\right)(\partial_{m}\Phi)\nonumber\\ & & - \frac{1}{24} K^{m}_{\ a}K^{na}(D^{2}\Phi)(\partial_{m}\Phi) (\partial_{n}\Phi) \nonumber \\ & & -\frac{1}{12}K^{m}_{\ a} K^{na}(D^{2}\Phi)\frac{1}{\Box} \partial_{m}\big( (\partial_{n}\Phi)(\square\Phi) \big) \Big) +c.c. \Big] \Big{\}}, \label{SGaction}\end{aligned}$$ with $f(x)\frac{1}{\Box}g(x) = f(x)\int {\mbox{d}}^4 y\hspace{1mm} G(x-y)g(y)$. Notice that two spurion fields $$U_{(1)\ \ ab}^{mn} = K^{m}_{\ a} K^{n}_{\ b}\theta\theta\bar{\theta}\bar{\theta}, \quad U_{(2)} = K^{mn}K_{mn}\theta\theta\bar{\theta}\bar{\theta}$$ appear in (\[SGaction\]). This is a consequence of rewriting the action (\[Sincompfields\]) as an integral over the whole superspace. Now we can start the machinery of the background field method. First we split the chiral and antichiral superfields into their classical and quantum parts $$\begin{aligned} \Phi\to \Phi+ \Phi_q ,\quad \Phi^+ \to \Phi^{+}+\Phi^{+}_q\end{aligned}$$ and integrate over the quantum superfields in the path integral. Since $\Phi_q$ and $\Phi_q^{+}$ are chiral and antichiral fields, they are constrained by $$\bar D_{\dot\a}\Phi_q= D_{\a}\Phi_q^+=0 .$$ To simplify the supergraph technique we introduce the unconstrained superfields $\S$ and $\S^+$, $$\Phi_{q} = -\frac{1}{4}\bar{D}^2\Sigma ,\quad \Phi_{q}^{+} = -\frac{1}{4}D^2\Sigma^{+}\ .\label{abel-gaug-tr}$$ Note that we do not express the background superfields $\Phi$ and $\Phi^+$ in terms of $\S$ and $\S^+$, only the quantum parts $\Phi_q$ and $\Phi_q^+$. After the integration of quantum superfields, the result is expressed in terms of the (anti)chiral superfields. This is a big advantage of the background field method and of the supergraph technique. The unconstrained superfields are determined up to a gauge transformation $$\S \to \S+\bar D_{\dot\a}\bar\Lambda^{\dot\a}, \quad \S^+ \to \S^++D^{\a}\Lambda_{\a} , \label{gf1}$$ with the gauge parameter $\Lambda$. This additional symmetry has to be fixed, so we add a gauge-fixing term to the action. For the gauge functions we choose $$\chi_\a = D_\a\S ,\quad \bar \chi_{\dot\a} = \bar D_{\dot\a}\S^+ \ . \label{gf2}$$ The product $\d(\chi)\d(\bar\chi)$ in the path integral is averaged by the weight $e^{-i\xi \int d^8z\bar{f}Mf}$: $$\int {\mbox{d}}f {\mbox{d}}\bar f \hspace{1mm} \d(\chi_\a-f_\a)\d(\bar\chi^{\dot\a}-\bar f^{\dot\a})e^{-i\xi\int d^8 z\bar f^{\dot\a}M_{\dot\a \a}f^\a}$$ where $$\bar f^{\dot \a}M_{\dot\a\a}f^\a=\frac{1}{4}\bar f^{\dot \a}(D_\a\bar D_{\dot\a}+\frac{3}{4}\bar D_{\dot\a}D_\a)f^\a$$ and the gauge-fixing parameter is denoted by $\xi$. The gauge-fixing term becomes $$S_{gf}=-\xi\int {\mbox{d}}^8 z\hspace{1mm} (\bar D_{\dot\a}\bar\S)(\frac{3}{16}\bar D^{\dot\a}D^{\a}+\frac 14 D^\a\bar D^{\dot\a})(D_\a\S) .$$ One can easily show that the ghost fields are decoupled. After the gauge-fixing, the part of the classical action quadratic in quantum superfields is given by $$S^{(2)} = S^{(2)}_0 + S^{(2)}_{int}, \label{S2}$$ with $$S^{(2)}_0=\frac{1}{2}\int {\mbox{d}}^8 z\hspace{1mm} \left(\begin{array}{cc} \Sigma & \Sigma^{+} \end{array} \right){\mathcal M} \left(\begin{array}{l}\Sigma\\ \Sigma^{+} \end{array} \right) \label{Snula}$$ and $$S^{(2)}_{int}=\frac{1}{2}\int {\mbox{d}}^8 z{\mbox{d}}^8 z'\hspace{1mm} \left(\begin{array}{cc} \Sigma & \Sigma^{+} \end{array} \right)(z){\mathcal V}(z,z') \left(\begin{array}{l}\Sigma\\ \Sigma^{+} \end{array} \right) (z'). \label{Sint}$$ Kinetic and interaction terms are collected in the matrices $\cal{M}$ and $\cal{V}$ respectively. The matrix $\cal{M}$ is given by $$\mathcal M= \left(\begin{array}{cc} -m\square^{1/2}P_{-} & \square (P_2+\xi(P_1+P_T))\\ \square (P_1+\xi(P_2+P_T)) & -m\square^{1/2}P_{+} \end{array} \right) ,\label{M}$$ with $$P_+ = \frac{D^2}{4\Box^{1/2}}, \quad P_- = \frac{{\bar D}^2}{4\Box^{1/2}}. \label{P+}$$ The interaction matrix ${\cal V}$ is $${\mathcal V}=\left(\begin{array}{cc} F & 0\\ 0 & \bar{F}\end{array}\right) .\label{V}$$ There are two types of elements in $\cal{V}$, local and nonlocal. We split them into $F_1$ and $F_2$ $$F(z,z')=F_{1}(z)\delta(z-z')+F_{2}(z,z'). \label{Fzz'}$$ Elements of $F_1$ are given by $$\begin{aligned} F_{1} (z)&=& \sum_{i=0}^{10} F^{(i)}\nonumber\\ &=& -\frac{\lambda}{2} \Phi \bar{D}^2 -\frac{\lambda}{48} K^{m}_{\ a}K^{na} \overleftarrow{\bar{D}^{2} D^{\alpha}} (\partial_{m}\Phi) \theta\theta\bar{\theta}\bar{\theta} \partial_{n}D_{\alpha}\bar{D}^{2}\nonumber \\ && - \frac{\lambda}{48} K^{m}_{\ a}K^{na} \overleftarrow{\bar{D}^{2} D^{\alpha}} (\partial_{m}D_{\alpha}\Phi) \theta\theta\bar{\theta}\bar{\theta} \partial_{n}\bar{D}^{2}\nonumber\\ && - \frac{\lambda}{48} K^{m}_{\ a}K^{na} \overleftarrow{\partial_{m}\bar{D}^{2}} (D^{\alpha}\Phi) \theta\theta\bar{\theta}\bar{\theta} \partial_{n}D_{\alpha}\bar{D}^{2} \nonumber\\ &&+\frac{\lambda}{24} K^{m}_{\ a}K^{n}_{\ b} \overleftarrow{\bar{D}^{2}D^{\alpha}}(\sigma^{ab})_{\alpha}^{\ \beta}(\partial_{m}\Phi)\theta\theta\bar{\theta}\bar{\theta} \partial_{n} D_{\beta} \bar{D}^{2} \nonumber\\ & &+\frac{\lambda}{24} K^{m}_{\ a}K^{n}_{\ b} \overleftarrow{\partial_{m}\bar{D}^{2}}(D^{\alpha}\Phi)(\sigma^{ab})_{\alpha}^{\ \beta}\theta\theta\bar{\theta}\bar{\theta} \partial_{n} D_{\beta} \bar{D}^{2} \nonumber\\ &&+\frac{\lambda}{24} K^{m}_{\ a}K^{n}_{\ b} \overleftarrow{\partial_{m}\bar{D}^{2}}(\partial_{n}D^{\alpha}\Phi)(\sigma^{ba})_{\alpha}^{\ \beta}\theta\theta\bar{\theta}\bar{\theta} D_{\beta} \bar{D}^{2} \nonumber\\ && -\frac{\lambda}{512} K^{mn}K_{mn} \overleftarrow{\bar{D}^{2}D^{2}} \Phi \bar{\theta}\bar{\theta} D^{2}\bar{D}^{2} \label{F1} \\ && -\frac{\lambda}{96} K^{m}_{\ a}K^{na} \overleftarrow{\partial_{m}\bar{D}^2} (\partial_{n}\Phi) \theta\theta\bar{\theta}\bar{\theta} D^{2}\bar{D}^2 \nonumber\\ && -\frac{\lambda}{192}K^{m}_{\ a}K^{na} \overleftarrow{\partial_{m}\bar{D}^{2}} (D^{2}\Phi)\theta\theta\bar{\theta}\bar{\theta} \partial_{n}\bar{D}^2\nonumber\\ && +\frac{\lambda}{96} K^{m}_{\ a}K^{na} \overleftarrow{\square\bar{D}^{2}} \Big(\int {\mbox{d}}^8 z'\hspace{1mm} (\partial_{m}D^{2}\Phi)(z') \frac{1}{\Box_{z'}}\delta(z'-z)\Big) \theta\theta\bar{\theta}\bar{\theta} \partial_{n}\bar{D}^2, \nonumber\end{aligned}$$ while the elements of $F_2$ read $$\begin{aligned} F_{2}(z,z') &=& \sum_{i=11}^{12} F^{(i)}\nonumber\\ &=& \frac{\lambda}{96} K^{m}_{\ a}K^{na} \overleftarrow{\partial_{m}\bar{D}^{2} D^{2}} \frac{1}{\Box_{z'}}\delta(z'-z)\theta\theta\bar{\theta}\bar{\theta} ((\partial_{n}\Phi)\square\bar{D}^{2})(z') \nonumber\\ && + \frac{\lambda}{96} K^{m}_{\ a}K^{na} \overleftarrow{\partial_{m}\bar{D}^{2} D^{2}} \frac{1}{\Box_{z'}}\delta(z'-z)\theta\theta\bar{\theta}\bar{\theta} (\Box\Phi \partial_{n}\bar{D}^{2})(z') .\label{F2'}\end{aligned}$$ The one-loop effective action is then $$\G=S_0+S_{int}+\frac{i}{2}\tr\log(1+{\mathcal M}^{-1}{\mathcal V}). \label{ED-opsta}$$ The last term in (\[ED-opsta\]) is the one-loop correction to the effective action and ${\cal M}^{-1}$ is the inverse of (\[M\]) given by $${\mathcal M}^{-1}=\left(\begin{array}{cc} A & B\\ \bar{B} & \bar{A} \end{array}\right)=\left(\begin{array}{cc} \frac{mD^2}{4\square (\square-m^2)} & \frac{D^2\bar{D}^2}{16\square (\square-m^2)}+\frac{\bar{D}^2 D^2-2\bar{D}D^2\bar{D}}{16\xi\square^2}\\ \frac{\bar{D}^2 D^2}{16\square (\square-m^2)}+\frac{D^2\bar{D}^2-2D\bar{D}^2 D}{16\xi\square^2} & \frac{m\bar{D}^2}{4\square (\square-m^2)} \end{array}\right) .\label{Minv}$$ Expansion of the logarithm in (\[ED-opsta\]) leads to the one-loop corrections $$\Gamma_1=\frac{i}{2}\tr\sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n}({\mathcal M}^{-1}\mathcal{V})^{n}=\sum^{\infty}_{n=1} \Gamma_{1}^{(n)}. \label{1LoopExp}$$ One-point and two-point functions ================================= The first term in the expansion (\[1LoopExp\]) gives the divergent part of the one-point functions, the tadpole contribution. We obtain $$\Gamma^{(1)}_{1}=\frac{i}{2} \tr ({\mathcal M}^{-1}{\mathcal V})=\frac{i}{2} \tr \big( AF+\bar{A}\bar{F}\big) = 0. \label{1point}$$ Therefore just like in the commutative Wess-Zumino model there is no tadpole contribution. Next we calculate the divergent part of the two point functions. It is given by $$\begin{aligned} \Gamma^{(2)}_{1} &=& -\frac{i}{4} \tr ({\mathcal M}^{-1}{\mathcal V})^{2} \nonumber\\ &=& -\frac{i}{4} \tr \big( AFAF + 2 \bar{B}FB\bar{F} + \bar{A}\bar{F}\bar{A}\bar{F} \big) .\label{2point1}\end{aligned}$$ First we calculate the $AFAF$ contributions. They are given by (remember that $F^{(i)}$ is the i-th element of the expansions (\[F1\]) and (\[F2’\])) $$\begin{aligned} {\rm Tr} (AF^{(0)}AF^{(0)})&=&0 ,\nonumber\\ {\rm Tr} (AF^{(1)}AF^{(0)})\bigg \|_{d.p.}&=& -\frac{i m^{2} \lambda^{2}K^{m}_{\ a}K^{na}}{6\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \partial_{m}A\partial_{n}A ,\nonumber\\ {\rm Tr} (AF^{(2)}AF^{(0)})&=&0 ,\nonumber\\ {\rm Tr} (AF^{(3)}AF^{(0)})&=&0 ,\nonumber\\ {\rm Tr} (AF^{(4)}AF^{(0)})&=&0 ,\nonumber\\ {\rm Tr} (AF^{(5)}AF^{(0)})&=&0 ,\nonumber\\ {\rm Tr} (AF^{(6)}AF^{(0)})&=&0 ,\nonumber\\ {\rm Tr} (AF^{(7)}AF^{(0)})\bigg \|_{d.p.}&=& - \frac{i m^{2}\lambda^{2} K^{mn}K_{mn}} {8\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} F^{2}, \nonumber\\ {\rm Tr} (AF^{(8)}AF^{(0)})\bigg \|_{d.p.}&=& \frac{i m^{2}\lambda^{2} K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \partial_{m}A\partial_{n}A ,\nonumber\\ {\rm Tr} (AF^{(9)}AF^{(0)}) &=&0 ,\nonumber\\ {\rm Tr} (AF^{(10)}AF^{(0)})&=&0 ,\nonumber\\ {\rm Tr} (AF^{(11)}AF^{(0)})\bigg \|_{d.p.}&=&0 ,\nonumber\\ {\rm Tr} (AF^{(12)}AF^{(0)})\bigg \|_{d.p.}&=&0.\nonumber\end{aligned}$$ Adding these terms we obtain $$\begin{aligned} \tr (AFAF)\bigg \|_{d.p.}&=& \tr (AF^{(0)}AF^{(0)})\bigg \|_{d.p.}+ 2\sum_{i=1}^{12}\tr (AF^{(i)}AF^{(0)})\bigg \|_{d.p.}\nonumber \\ &=& -\frac{i m^2 \lambda^2 K^{m}_{\ a}K^{na}}{6\pi^2 \varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \partial_{m}A \partial_{n}A \nonumber \\ && -\frac{i m^2 \lambda^2 K^{mn}K_{mn}}{4\pi^2\varepsilon}\int {\mbox{d}}^4 x\hspace{1mm} F^2. \label{2pointA}\end{aligned}$$ The $\bar{B}FB\bar{F}$ term is more difficult to calculate. Some of the identities we use are given in the Appendix. We obtain the following contributions: $$\begin{aligned} \tr (\bar{B}F^{(0)}B\bar{F}^{(0)})\bigg \|_{d.p.} &=& \frac{i \lambda^{2}}{2\pi^{2}\varepsilon}\int {\mbox{d}}^8 z\hspace{1mm} \Phi^{\dagger}\Phi ,\nonumber\\ \tr (\bar{B}F^{(1)}B\bar{F}^{(0)})\bigg \|_{d.p.} &=& -\frac{i \lambda^{2}K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} A^(\square - 4 m^2)\partial_{m} \partial_{n} A ,\nonumber\\ \tr (\bar{B}F^{(2)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& -\frac{\lambda^{2}K^{m}_{\ a}K^{na}}{36\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi} \bar{\sigma}^{l}\partial_{l} \partial_{m}\partial_{n}\psi \nonumber \\ && + \frac{\lambda^{2}K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi} \bar{\sigma}_{n} \bigg(m^2-\frac{\square}{6}\bigg)\partial_{m}\psi ,\nonumber\\ \tr (\bar{B}F^{(3)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& \frac{\lambda^{2}K^{m}_{\ a}K^{na}}{72\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi} \bar{\sigma}^{l}\partial_{l} \partial_{m}\partial_{n}\psi ,\nonumber\\ \tr (\bar{B}F^{(4)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& -\frac{i \lambda^{2}K^{m}_{\ a}K^{na}}{2\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} A^\big(m^{2}-\frac{\square}{6}\big) \partial_{m}\partial_{n} A ,\nonumber\\ \tr (\bar{B}F^{(5)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& -\frac{\lambda^{2}K^{m}_{\ a}K^{n}_{\ b}}{72\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}(\bar{\sigma}^{b}\partial^{a} - \bar{\sigma}^{a}\partial^{b} + i\varepsilon^{abcd}\bar{\sigma}_{d} \partial_{c})\partial_{m}\partial_{n}\psi \nonumber\\ &&+ \frac{\lambda^2 K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}\bar{\sigma}_{n} \bigg(m^2-\frac{\square}{6}\bigg)\partial_{m}\psi ,\nonumber\\ \tr (\bar{B}F^{(6)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& -\frac{\lambda^{2}K^{m}_{\ a}K^{n}_{\ b}}{36\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}(\bar{\sigma}^{b}\partial^{a} - \bar{\sigma}^{a}\partial^{b} + i\varepsilon^{abcd}\bar{\sigma}_{d} \partial_{c})\partial_{m}\partial_{n}\psi \nonumber\\ &&- \frac{\lambda^2 K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}\bar{\sigma}_{n} \bigg(m^2-\frac{\square}{6}\bigg)\partial_{m}\psi ,\nonumber\\ \tr (\bar{B}F^{(7)}B\bar{F}^{(0)})&=&0 ,\nonumber\\ \tr (\bar{B}F^{(8)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& \frac{i m^{2} \lambda^{2}K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \partial_{m}A^\partial_{n} A ,\nonumber\\ \tr (\bar{B}F^{(9)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& \frac{i \lambda^{2}K^{m}_{\ a}K^{na}}{72\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} F^\partial_{m}\partial_{n}F ,\nonumber\\ \tr (\bar{B}F^{(10)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& \frac{i m^{2} \lambda^{2}K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x{\mbox{d}}^4 y\hspace{1mm} \partial_{m}\partial_{n} F(x)\square^{-1}_{x}\delta(x-y) F^(y) ,\nonumber\\ \tr (\bar{B}F^{(11)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& - \frac{i m^{2} \lambda^{2}K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \partial_{m}A^\partial_{n} A ,\nonumber\\ \tr (\bar{B}F^{(12)}B\bar{F}^{(0)})\bigg \|_{d.p.}&=& -\frac{i \lambda^{2}K^{m}_{\ a}K^{na}}{36\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} A^\partial_{m}\partial_{n}\square A .\nonumber\end{aligned}$$ Collecting all contributions we have $$\begin{aligned} \tr (\bar{B}FB\bar{F})\bigg \|_{d.p.} &=& \tr (\bar{B}F^{(0)}B\bar{F}^{(0)})\bigg \|_{d.p.} + 2 \sum_{i=1}^{12}\tr (\bar{B}F^{(i)}B\bar{F}^{(0)})\bigg \|_{d.p.} \nonumber\\ &=& \frac{i \lambda^{2}}{2\pi^{2}\varepsilon}\int {\mbox{d}}^8 z\hspace{1mm} \Phi^{+}\Phi \nonumber\\ && - \frac{i \lambda^{2}K^{m}_{\ a}K^{na}}{3\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} A^\bigg(m^{2} + \frac{\square}{6}\bigg)\partial_{m}\partial_{n} A \nonumber\\ && - \frac{\lambda^2 K^{m}_{\ a}K^{n}_{\ b}}{12\pi^{2}\varepsilon} \int d^{4}x \bar{\psi}(\bar{\sigma}^{b}\partial^{a}-\bar{\sigma}^{a} \partial^{b}+ i\varepsilon^{abcd}\bar{\sigma}_{d}\partial_{c}) \partial_{m}\partial_{n} \psi \nonumber \\ && + \frac{\lambda^2 K^{m}_{\ a}K^{na}}{6\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}\bar{\sigma}_{n} \partial_{m}\bigg(m^{2}-\frac{\square}{6}\bigg)\psi \nonumber \\ && - \frac{\lambda^2 K^{m}_{\ a}K^{na}}{36\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}\bar{\sigma}^{l}\partial_{l}\partial_{m}\partial_{n} \psi \nonumber \\ && + \frac{i \lambda^{2}K^{m}_{\ a}K^{na}}{36\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} F^\partial_{m}\partial_{n} F \label{2pointB}\\ && +\frac{i m^{2}\lambda^{2}K^{m}_{\ a}K^{na}}{6\pi^{2}\varepsilon} \int {\mbox{d}}^4 x {\mbox{d}}^4 y\hspace{1mm} \partial_{m}\partial_{n}F(x) \square^{-1}_{x}\delta(x-y) F^(y). \nonumber\end{aligned}$$ Finally, adding (\[2pointA\]) and (\[2pointB\]) we obtain the divergent part of the two-point function $$\begin{aligned} \Gamma^{(2)}_{1}\bigg\|_{d.p.}&=& -\frac{m^2 \lambda^2 K^{m}_{\ a}K^{na}}{24\pi^2 \varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} (\partial_{m}A \partial_{n}A + \partial_{m}A^* \partial_{n}A^) \nonumber \\ && -\frac{m^2 \lambda^2 K^{mn}K_{mn}}{16\pi^2\varepsilon}\int {\mbox{d}}^4 x\hspace{1mm} F^2 \nonumber\\ && -\frac{m^2 \lambda^2 K^{mn}K^{}_{mn}}{16\pi^2\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} F^{2} \nonumber\\ && +\frac{\lambda^{2}}{4\pi^{2}\varepsilon} \int {\mbox{d}}^8 z\hspace{1mm} \Phi^{+}\Phi \nonumber\\ && - \frac{\lambda^{2}K^{m}_{\ a}K^{na}}{6\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} A^\bigg(m^{2} + \frac{\square}{6}\bigg)\partial_{m}\partial_{n} A \nonumber\\ && + \frac{i\lambda^2 K^{m}_{\ a}K^{n}_{\ b}}{24\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}(\bar{\sigma}^{b}\partial^{a} -\bar{\sigma}^{a}\partial^{b}+ i\varepsilon^{abcd}\bar{\sigma}_{d} \partial_{c})\partial_{m}\partial_{n} \psi \nonumber \\ && - \frac{i\lambda^2 K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}\bar{\sigma}_{n} \partial_{m}\bigg(m^{2}-\frac{\square}{6}\bigg)\psi \nonumber \\ && + \frac{i\lambda^2 K^{m}_{\ a}K^{na}}{72\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} \bar{\psi}\bar{\sigma}^{l}\partial_{l}\partial_{m}\partial_{n} \psi \nonumber \\ && + \frac{\lambda^{2}K^{m}_{\ a}K^{na}}{72\pi^{2}\varepsilon} \int {\mbox{d}}^4 x\hspace{1mm} F^\partial_{m}\partial_{n} F \nonumber\\ && + \frac{m^{2}\lambda^{2}K^{m}_{\ a}K^{na}}{12\pi^{2}\varepsilon} \int {\mbox{d}}^4 x {\mbox{d}}^4 y\hspace{1mm} \partial_{m}\partial_{n}F(x) \square^{-1}_{x}\delta(x-y) F^(y). \label{2pointAll}\end{aligned}$$ We immediately see that the divergences appearing in (\[2pointAll\]) cannot be absorbed by counterterms since the terms appearing in (\[2pointAll\]) do not exist in the classical action. All terms in (\[2pointAll\]) quadratic in the deformation parameter are also quadratic in fields. However, the deformation of the classical action (\[SGaction\]) is only present in the interaction term, and terms in the action quadratic in the deformation parameter will always be of the third order in fields. We have to conclude that our model, as it stands, is not renormalizable. Discussion and conclusions ========================== Let us now summarize what we have done so far and discuss the obtained results in more detail. In order to see how different deformations (different twists) affect renormalizability of the Wess-Zumino model, we considered one special example of twist, (\[twist\]). The main adventage of this twist is that it is hermitian and therefore implies the hermitian $\star$-product. Compared with the undeformed SUSY Hopf algebra, the twisted SUSY Hopf algebra changes. In particular, the Leibniz rule (\[deftrlaw\]) becomes deformed. The notion of chirality is lost and we had to apply the method of projectors introduced in [@miSUSY] to obtain the action. A nonlocal deformation of the commutative Wess-Zumino action invariant under the deformed SUSY transformations (\[defsusytr\]) and with a good commutative limit was introduced and its renormalizability properties were investigated. Notice that the nonlocality comes from the application of the chiral projector $P_2$[^1]. To calculate the divergent part of the effective action we used the background field method and the supergraph technique. Like in the commutative Wess-Zumino model, there is no tadpole contribution. There is no mass counterterm which is again the same as in the undeformed Wess-Zumino model. However, the divergent part of the two-point function cannot be canceled and we have to conclude that our model is not renormalizable. Calculating divergent parts of the three-point and higher functions does not make sense and it is technically very demanding. Having in mind results of [@on-shell], we also investigated on-shell renormalizability of our model. In general, on-shell renormalizability leads to a one-loop renormalizable $S$-matrix. On the other hand, one-loop on-shell renormalizable Green functions may spoil renormalizability at higher loops. After using the equations of motion which follow form the action (\[SGaction\]) to obtain the on-shell divergent terms we see that the divergences in the two-point function remain and therefore the model is also not on-shell renormalizable. In our previous work [@D-def-us] we had a similar problem, a deformed model which was not renormalizable. To obtain a renormalizable model we had to relax the condition of minimality of deformation and to include non-minimal terms. Also, in [@Penati] new terms of the form $\int {\mbox{d}}^8 z \theta\theta\bar{\theta}\bar{\theta} D^2 \Phi$ and $\int {\mbox{d}}^8 z \theta\theta\bar{\theta}\bar{\theta} (D^2 \Phi)^2$ were added in order to absorb divergences produced by $\int {\mbox{d}}^4 x F^3 = \int {\mbox{d}}^8 z \theta\theta\bar{\theta}\bar{\theta} (D^2 \Phi)^3$ term. Since the model we work with is more complicated than the models of [@D-def-us] and [@Penati], it is not obvious which terms should be added. Let us list possible terms. Note that the new terms have to be invariant under the deformed SUSY transformations (\[defsusytr\]). This requirement gives three possibilities: $$\begin{aligned} T_1 &=& \int {\mbox{d}}^4 x\hspace{1mm} P_1(\Phi\star\Phi)\Big\|_{\bar{\theta}\bar{\theta}} \label{T1}\\ &=& \frac{1}{2} K^{ab}K_{ab}\int {\mbox{d}}^4 x\hspace{1mm} \big( \frac{1}{2}(\p_m\psi)\sigma^m\bar{\sigma}^n(\p_n\psi)-F\Box A \big). \nonumber\\ T_2 &=& \int {\mbox{d}}^4 x\hspace{1mm} P_1(\Phi\star P_2(\Phi\star\Phi))\Big\|_{\bar{\theta}\bar{\theta}} \label{T2}\\ &=& \frac{1}{4}K^{ab}K_{ab}\int {\mbox{d}}^4 x\hspace{1mm} \big( -AF \Box A - \frac{1}{2}F\Box A^2 \nonumber\\ && + \frac{1}{2}\psi\psi\Box A + \p_m(A\psi)\sigma^m\bar{\sigma}^n(\p_n\psi) \big). \nonumber\\ T_3 &=& \int {\mbox{d}}^4 x\hspace{1mm} \Phi\star P_1(\Phi\star\Phi)) \Big\|_{\theta\theta\bar{\theta}\bar{\theta}} \label{T3}\\ &=& \frac{3}{4}K^{ab}K_{ab} \int {\mbox{d}}^4 x\hspace{1mm} \Big( F(\p_m\psi)\sigma^m\bar{\sigma}^l(\p_l\psi) - F^2\Box A \Big)\nonumber\\ && + K^{m}_{\ a}K^{na} \int {\mbox{d}}^4 x\hspace{1mm} \Big( A(\Box A)(\p_m\p_n A) + A(\p_m\p_l A)(\p_n\p^l A) \Big). \nonumber\end{aligned}$$ The term $T_1$ produces divergences of the type $\int {\mbox{d}}^4 x\hspace{1mm} \Phi^+ \Phi$ so it would not spoil the renormalizability of the model. However, it cannot improve renormalizability since divergences appearing in (\[2pointAll\]) are not of the type $T_1$. The term $T_2$ produces additional divergences that cannot be absorbed, so we have to ignore it. The $T_3$ term does not cancel any of the terms present in the action (\[Sincompfields\]). Additionally, it produces new divergent terms. However these terms might look like, they can never cancel all the divergences in (\[2pointAll\]) as divergences proportional to $K^{m}_{\ a}K^{n}_{\ b}$ will remain. This analysis forces us to conclude that even with a non-minimal deformation our model remains nonrenormalizable. Let us make a remark about the nonrenormalization theorem and its modifications in the case of deformed SUSY. One easily sees that the divergent terms of the effective action (\[2pointAll\]) can be rewritten as $$\Gamma^{(2)}_1\bigg\|_{d.p.} = \int {\mbox{d}}^4 x_1 \hspace{1mm} {\mbox{d}}^4 x_2\hspace{1mm} {\mbox{d}}^2 \theta {\mbox{d}}^2 \bar{\theta}\hspace{1mm} G_2(x_1, x_2, U_{(1)}, U_{(2)}) f_1(x_1, \theta, \bar{\theta}) f_2(x_2, \theta, \bar{\theta}), \label{NRTh}$$ with $f_i = f_i (\Phi, \Phi^+, D\Phi, \bar{D}\Phi, D\Phi^+, \bar{D}\Phi^+,\dots)$. The nonlocal term in (\[NRTh\]) appears as a consequence of nonlocality in the classical action (\[SGaction\]). The result (\[NRTh\]) confirms the modified nonrenormalization theorem [@Penati]. The appearance of the spurion fields in (\[NRTh\]) signals breaking of the undeformed SUSY. In our case, symmetry which remains after the breaking is the twisted SUSY (\[defsusytr\]). However, it seems that the twisted SUSY is not enough to guarantee renormalizability. It is obvious that different deformations obtained from different twists lead to models with different quantum properties. In our previous work [@D-def-us] we studied a deformation which preserves the full undeformed SUSY. There, after relaxing the condition of minimality of deformation, we obtained a renormalizable Wess-Zumino model. In this paper we work with a deformation given in terms of the non-SUSY-covariant derivatives. The Leibniz rule for the SUSY transformation (\[defsusytr\]) changes and the deformed action (\[Sincompfields\]) though invariant under twisted SUSY transformations, is not invariant under the undeformed SUSY transformations. For example, the term $K^{mn}K_{mn}F^3$ breaks the undeformed SUSY. On the other hand, the twisted SUSY allows this term as a part of the invariant term $P_2(\Phi\star P_2(\Phi\star \Phi))\Big\|_{\theta\theta}$, see Equations (5.13) and (5.14) in [@miSUSY]. The classical properties of theories with twisted symmetries are not fully understood [@NCbookMi], [@TwistedNoeth]. For example, one cannot apply standard methods to find conserved charges and the modification of Noether theorem in the case of twisted symmetry has not been formulated yet. In this paper we analyze quantum properties of the theory with the twisted SUSY. This is the first time that renormalizability of a theory with a twisted symmetry is analyzed. Even after relaxing the condition of minimality of deformation our model remains nonrenormalizable. This indicates that theories with twisted symmetries do not have the same quantum properties as theories with undeformed symmetries. In our example, we see that the twisted SUSY is not enough to guarantee renormalizability of the Wess-Zumino model. It is obvious that a better understanding of the twisted symmetry and its consequences, both classical and quantum is needed. Calculation =========== In this appendix we collect details of some calculations and some important side results. - Transformation laws of the component fields of the superfield ${\rm F}$ (\[F\]): $$\begin{aligned} \delta_\xi f &=& \xi^\alpha \phi_\alpha + \bar{\xi}_{\dot\alpha}\bar{\chi}^{\dot\alpha}, \label{susytrf} \\ \delta_\xi \phi_\alpha &=& 2\xi_\alpha m + \sigma^m_{\ \alpha\dot{\alpha}}\bar{\xi}^{\dot\alpha}\big( v_m + i (\p_m f) \big), \label{susytrphi} \\ \delta_\xi \bar{\chi}^{\dot\alpha} &=& 2\bar{\xi}^{\dot\alpha}n + \bar{\sigma}^{m\dot{\alpha}\alpha}\xi_\alpha\big( -v_m + i (\p_m f) \big), \label{susytrchi} \\ \delta_\xi m &=&\bar{\xi}_{\dot\alpha}\bar{\lambda}^{\dot\alpha} + \frac{i}{2}\bar{\xi}_{\dot\alpha}\bar{\sigma}^{m\dot{\alpha}\alpha}(\p_m \phi_\alpha) , \label{susytrm} \\ \delta_\xi n &=& \xi^\alpha \varphi_\alpha + \frac{i}{2}\xi^\alpha\sigma^m_{\ \alpha\dot{\alpha}}(\p_m \bar{\chi}^{\dot\alpha}), \label{susytrn} \\ \sigma^m_{\ \alpha\dot{\alpha}}\delta_\xi v_m &=& -i(\p_m\phi_\alpha)\xi^\beta\sigma^m_{\ \beta\dot{\alpha}} + 2\xi_\alpha\bar{\lambda}_{\dot\alpha} \nonumber\\ &&+ i\sigma^m_{\ \alpha\dot{\beta}}\bar{\xi}^{\dot\beta}(\p_m \bar{\chi}_{\dot\alpha}) + 2\varphi_\alpha\bar{\xi}_{\dot\alpha} , \label{susytrvm} \\ \delta_\xi \bar{\lambda}^{\dot\alpha} &=& 2\bar{\xi}^{\dot\alpha}d + i\bar{\sigma}^{l\dot{\alpha}\alpha}\xi_\alpha(\p_l m) + \frac{i}{2}\bar{\sigma}^{l\dot{\alpha}\alpha}\sigma^m_{\ \alpha\dot{\beta}}\bar{\xi}^{\dot\beta}(\p_m v_l) , \label{susytrlambda} \\ \delta_\xi \varphi_\alpha &=& 2\xi_\alpha d + i\sigma^l_{\ \alpha\dot{\alpha}}\bar{\xi}^{\dot\alpha}(\p_l n) -\frac{i}{2}\sigma^l_{\ \alpha\dot{\alpha}}\bar{\sigma}^{m\dot{\alpha}\beta}\xi_\beta(\p_m v_l) , \label{susytrvarphi} \\ \delta_\xi d &=& \frac{i}{2}\xi^\alpha \sigma^m_{\ \alpha\dot{\alpha}}(\p_m \bar{\lambda}^{\dot\alpha}) -\frac{i}{2}(\p_m \varphi^\alpha)\sigma^m_{\ \alpha\dot{\alpha}}\bar{\xi}^{\dot\alpha} . \label{susytrd}\end{aligned}$$ - Irreducible components the superfield ${\rm F}$: $$\begin{aligned} P_2 {\rm F} &=& \frac{1}{16} \frac{\bar{D}^2 D^2 }{\Box} {\rm F} \nonumber\\ &=& \frac{1}{\Box}\Big( d - \frac{i}{2}(\p_m v^m) + \frac{1}{4}\Box f\Big) + \sqrt{2}\theta^\alpha\Big( \frac{i}{\sqrt{2}\Box}\sigma^m_{\ \alpha\dot{\alpha}} (\p_m \bar{\lambda}^{\dot{\alpha}}) + \frac{1}{2\sqrt{2}}\phi_\alpha \Big) \nonumber\\ && + \theta\theta m + i\theta\sigma^l \bar{\theta}\p_l\Big( \frac{d}{\Box} - \frac{i}{2\Box}(\p_m v^m) + \frac{1}{4}f \Big) \label{P2F'}\\ && + \frac{1}{\sqrt{2}}\theta\theta\bar{\theta}_{\dot\alpha}\Big( \frac{1}{\sqrt{2}}\bar{\lambda}^{\dot{\alpha}} + \frac{i}{2\sqrt{2}}\bar{\sigma}^{m\dot{\alpha}\alpha}(\p_m\phi_\alpha) \Big) \nonumber\\ && +\frac{1}{4}\theta\theta\bar{\theta}\bar{\theta}\Big( d - \frac{i}{2}(\p_m v^m) + \frac{1}{4}\Box f \Big) .\nonumber\end{aligned}$$ $$\begin{aligned} P_1 {\rm F}&=& \frac{1}{16} \frac{D^2\bar{D}^2}{\Box} {\rm F} \nonumber\\ &=& \frac{1}{\Box}\Big( d + \frac{i}{2}(\p_m v^m) + \frac{1}{4}\Box f\Big) + \sqrt{2}\bar{\theta}_{\dot\alpha}\Big( \frac{i}{\sqrt{2}\Box}\bar{\sigma}^{m\dot{\alpha}\alpha} (\p_m \varphi_\alpha) + \frac{1}{2\sqrt{2}}\bar{\chi}^{\dot\alpha} \Big) \nonumber\\ && + \bar{\theta}\bar{\theta} n - i\theta\sigma^l \bar{\theta}\p_l\Big( \frac{d}{\Box} + \frac{i}{2\Box}(\p_m v^m) + \frac{1}{4}f \Big) \label{P1F'}\\ && + \frac{1}{\sqrt{2}}\bar{\theta}\bar{\theta}\theta^\alpha\Big( \frac{1}{\sqrt{2}}\varphi_\alpha + \frac{i}{2\sqrt{2}}\sigma^m_{\ \alpha\dot{\alpha}}(\p_m\bar{\chi}^{\dot\alpha}) \Big) \nonumber\\ &&+\frac{1}{4}\theta\theta\bar{\theta}\bar{\theta}\Big( d + \frac{i}{2}(\p_m v^m) + \frac{1}{4}\Box f \Big) ,\nonumber\end{aligned}$$ $$\begin{aligned} P_T {\rm F} &=& \frac{1}{2}f - \frac{2}{\Box}d +\theta^\alpha\Big( \frac{1}{2}\phi_\alpha- i\frac{1}{\Box}\sigma^m_{\ \alpha\dot{\alpha}}\p_m\bar{\lambda}^{\dot\alpha}\Big) \nonumber\\ && +\bar\theta_{\dot{\alpha}}\Big( \frac{1}{2}\bar\chi^{\dot{\alpha}}-i\frac{1}{\Box}\bar{\sigma}^{m\dot{\alpha}\alpha} \p_m\varphi_{\alpha} \Big) +\theta\sigma^m\bar{\theta} \Big( v_m-\frac{1}{\Box}\p_m\p_l v^l \Big)\nonumber\\ && +\theta\theta\bar{\theta}_{\dot{\alpha}}\Big( \frac{1}{2} \bar{\lambda}^{\dot\alpha} - \frac{i}{4}\bar{\sigma}^{m\dot{\alpha}\alpha}(\p_m\phi_\alpha)\Big) + \bar{\theta}\bar{\theta}\theta^\alpha\Big( \frac{1}{2} \varphi_\alpha - \frac{i}{4}\sigma^m_{\ \alpha\dot{\alpha}}(\p_m\bar{\chi}^{\dot\alpha})\Big)\nonumber\\ && +\frac{1}{4}\theta\theta\bar{\theta}\bar{\theta} \Big( 2d - \frac{1}{2}\Box f\Big) .\label{PTF}\end{aligned}$$ The following identity holds $$P_T = I-P_1-P_2 .\label{PT'}$$ - Some general formulas for the divergent parts of traces, where $K=\Box-m^2$ $$\begin{aligned} \tr(K^{-1}f) &=& \frac{i}{8\pi^2\epsilon}m^2\int{\mbox{d}}^4 x\hspace{1mm} f ,\label{f1} \\ \tr(\partial_aK^{-1}f) &=&0 , \label{f2}\\ \tr(\Box K^{-1}f) &=& \frac{im^4}{8\pi^2\epsilon}\int{\mbox{d}}^4 x \hspace{1mm} f ,\label{f3}\\ \tr(\Box^2K^{-1}f) &=& \frac{im^6}{16\pi^2\epsilon}\int{\mbox{d}}^4 x\hspace{1mm} f , \label{f4}\\ \tr(K^{-1}fK^{-1}g)&=&\frac{i}{8\pi^2\epsilon}\int{\mbox{d}}^4 x\hspace{1mm} fg ,\label{f5}\\ \tr(\partial_nK^{-1}fK^{-1}g) &=&\frac{i}{16\pi^2\epsilon} \int{\mbox{d}}^4 x\hspace{1mm} \partial_nfg ,\label{f6}\\ \tr(\partial_nK^{-1}f\partial_mK^{-1}g) &=& -\frac{i}{16\pi^2\epsilon} \int{\mbox{d}}^4 x\hspace{1mm} \label{f7}\\ &&\hspace{-4cm}f\Big( \frac{1}{3} \partial_n\partial_m+\frac{1}{6}\eta_{mn}\Box -\eta_{mn}m^2\Big)g , \nonumber\\ \tr(\partial_nK^{-1}f\partial_m\partial_pK^{-1}g) &=& -\frac{i}{32\pi^2\epsilon} \int{\mbox{d}}^4 x\hspace{1mm} \label{f8}\\ &&\hspace{-4cm}f\Big( \frac{1}{3} \partial_n\partial_m\partial_p+(\eta_{mp}\partial_{n}-\eta_{np}\partial_{m}- \eta_{nm}\partial_{p})(m^{2}-\frac{1}{6}\Box)\Big)g . \nonumber\end{aligned}$$ The work of the authors is supported by the project $141036$ of the Serbian Ministry of Science. [99]{} (1930), in: Wolfgang Pauli, Scientific Correspondence, vol. II, 15, Ed. Karl von Meyenn, Springer-Verlag 1985. H. S. Snyder, [Quantized spacetime]{}, Phys.Rev. [71]{}, 38 (1947). Y. I. Manin, [Multiparametric quantum deformation of the general linear supergroup]{}, Commun. Math. Phys. [123]{}, 163 (1989). J. Wess, [q-deformed Heisenberg Algebras]{}, Lecture Notes in Physics 543, Berlin, Springer (2000), \[math.ph/9910013\]. G. Landi, [An introduction to noncommutative spaces and their geometry]{}, Springer, New York, 1997; hep-th/9701078. J. Madore, [An Introduction to Noncommutative Differential Geometry and its Physical Applications]{}, 2nd Edition, Cambridge Univ. Press, 1999. P. Aschieri, M. Dimitrijevi' c, P. Kulish, F. Lizzi and J. Wess [Noncommutative spacetimes: Symmetries in noncommutative geometry and field theory]{}, Lecture notes in physics [774]{}, Springer (2009). L. Castellani, [Noncommutative geometry and physics: A review of selected recent results]{}, Class. Quant. Grav. [17]{}, 3377 (2000), \[hep-th/0005210\]. M. R. Douglas and N. A. Nekrasov, [Noncommutative field theory]{}, Rev. Mod. Phys. [73]{}, 977 (2001), \[hep-th/0106048\]. R. J. Szabo, [Quantum field theory on noncommutative spaces]{}, Phys. Rept. [378]{}, 207 (2003), \[hep-th/0109162\]. R. J. Szabo, [Symmetry, Gravity and Noncommutativity]{}, Class. Quant. Grav. [23]{}, R199-R242 (2006), \[hep-th/0606233\]. X. Calmet, B. Jurčo, P. Schupp, J. Wess and M. Wohlgenannt, [The Standard Model on noncommutative spacetime]{}, Eur. Phys. J. [C23]{}, 363 (2002), \[hep-ph/0111115\]. P. Aschieri, B. Jurčo, P. Schupp and J. Wess, [Noncommutative GUTs, standard model and C, P, T]{}, Nucl. Phys. B [651]{}, 45 (2003), \[hep-th/0205214\]. M. Buri' c, V. Radovanovi' c and J. Trampeti' c, [The one-loop renormalization of the gauge sector in the noncommutative standard model]{}, JHEP [0703]{}, 030 (2007), \[hep-th/0609073\]. M. Buri' c, V. Radovanovi' c and J. Trampeti' c, [Nonzero $Z \to \gamma \gamma$ decays in the renormalizable gauge sector of the noncommutative standard model]{}, Phys. Rev. [D 75]{}, 097701 (2007). H. Grosse and R. Wulkenhaar, [Renormalisation of $\phi^4$-theory on noncommutative $R^4$ in the matrix base]{}, Commun. Math. Phys. [256]{}, 305-374 (2005), \[hep-th/0401128\]. M. Buri' c, D. Latas and V. Radovanovi' c, [Renormalizability of noncommutative $SU(N)$ gauge theory]{}, JHEP [0602]{}, 046 2006, \[hep-th/0510133\]. C. P. Martin and C. Tamarit, [Renormalisability of noncommutative GUT inspired field theories with anomaly safe groups]{}, JHEP [0912]{}, 042 (2009), 0910.2677\[hep-th\]. N. Seiberg, [Noncommutative superspace, $N = 1/2$ supersymmetry, field theory and string theory]{}, JHEP [0306]{} 010 (2003), \[hep-th/0305248\]. J. de Boer, P. A. Grassi and P. van Nieuwenhuizen, [Noncommutative superspace from string theory]{}, Phys. Lett. B [574]{}, 98 (2003), \[hep-th/0302078\]. B. Nikoli' c and B. Sazdovi' c, [Noncommutativity relations in type $IIB$ theory and their supersymmetry]{}, JHEP [0008]{}, 037 (2010), 1005.1181\[hep-th\]. P. Kosi[' n]{}ski, J. Lukierski and P. Ma[' s]{}lanka, [Quantum Deformations of Space-Time SUSY and Noncommutative Superfield Theory]{}, hep-th/0011053. P. Kosi[' n]{}ski, J. Lukierski, P. Ma[' s]{}lanka and J. Sobczyk, [Quantum Deformation of the Poincare Supergroup and $\kappa$-deformed Superspace]{}, J. Phys. A [27]{} (1994) 6827, \[hep-th/9405076\]. Chong-Sun Chu and F. Zamora, [Manifest supersymmetry in noncommutative geometry]{}, JHEP [0002]{}, 022 (2000), \[hep-th/9912153\]. S. Ferrara and M. A. Lledo, [Some aspects of deformations of supersymmetric field theories]{}, JHEP [05]{}, 008 (2000), \[hep-th/0002084\]. D. Klemm, S. Penati and L. Tamassia, [Non(anti)commutative superspace]{}, Class. Quant. Grav. [20]{} (2003) 2905, \[hep-th/0104190\]. S. Ferrara, M. Lledo and O. Macia, [Supersymmetry in noncommutative superspaces]{}, JHEP [09]{} (2003) 068, \[hep-th/0307039\]. M. Dimitrijevi' c and V. Radovanovi' c, [D-deformed Wess-Zumino model and its renormalizability properties]{}, JHEP [0904]{}, 108 2009, 0902.1864\[hep-th\]. M. Dimitrijevi' c, B. Nikoli' c and V. Radovanovi' c, [(Non)renormalizability of the $D$-deformed Wess-Zumino model]{}, Phys. Rev. D [81]{}, 105020-105032 (2010), \[arxiv:1001.2654\]. S. Terashima and J. T. Yee, [Comments on Noncommutative Superspace]{}, JHEP [0312]{}, 053 (2003), \[hep-th/0306237\]. I. Jack, D. R. T. Jones and R. Purdy, [The non-anticommutative supersymmetric Wess-Zumino model]{}, 0808.0400\[hep-th\]. R. Britto and B. Feng, [$N=1/2$ Wess-Zumino model is renormalizable]{}, Phys. Rev. Lett. [91]{}, 201601 (2003), \[hep-th/0307165\]. A. Romagnoni, [Renormalizability of N = 1/2 Wess-Zumino model in superspace]{}, JHEP [0310]{}, 016 (2003), \[hep-th/0307209\]. R. Britto, B. Feng, Soo-Jong Rey, [Non(anti)commutative superspace, UV/IR mixing and open Wilson lines]{}, JHEP [0308]{}, 001 (2003), \[hep-th/0307091\]. R. Britto, B. Feng, Soo-Jong Rey, [Deformed superspace, N = 1/2 supersymmetry and nonrenormalization theorems]{}, JHEP [0307]{}, 067 (2003), \[hep-th/0306215\]. M. T. Grisaru, S. Penati and A. Romagnoni, [Two-loop Renormalization for Nonanticommutative $N=1/2$ Supersymmetric WZ Model]{}, JHEP [0308]{}, 003 (2003), \[hep-th/ 0307099\]. C. P. Martin and C. Tamarit, [The Seiberg-Witten map and supersymmetry]{}, JHEP [0811* ]{}, 087 (2008), 0809.2684\[hep-th\]. C. P. Martin and C. Tamarit, [Noncommutative $N=1$ super Yang-Mills, the Seiberg-Witten map and UV divergences]{}, JHEP [0911]{}, 092 (2009), 0907.2437\[hep-th\]. O. Lunin and S. J. Rey, [Renormalizability of Non(anti)commutative Gauge Theories with $N =1/2$ Supersymmetry]{}, JHEP [0309]{}, 045 (2003), \[hep-th/0307275\]. I. Jack, D. R. T. Jones and L. A. Worthy, [One-loop renormalisation of $N = 1/2$ supersymmetric gauge theory with a superpotential]{}, Phys. Rev. [D75]{}, 045014 (2007), \[hep-th/0701096\]. D. Berenstein, S. J. Rey, [Wilsonian proof for renormalizability of N=1/2 supersymmetric field theories]{}, Phys. Rev. D68, 121701, (2003), \[hep-th/0308049\]. B. M. Zupnik, [Twist-deformed supersymmetries in non-anticommutative super- spaces]{}, Phys. Lett. [B 627]{} 208 (2005) \[hep-th/0506043\]. M. Ihl and C. S" amann, [Drinfeld-twisted supersymmetry and non-anticommutative superspace]{}, JHEP [0601]{} (2006) 065, \[hep-th/0506057\]. M. Irisawa, Y. Kobayashi and S. Sasaki, [Drinfel’d Twisted Superconformal Algebra and Structure of Unbroken Symmetries]{}, Prog. Theor. Phys. [118]{} (2007) 83-96, \[hep-th/0606207\]. M. Dimitrijevi' c, V. Radovanovi' c and J. Wess, [Field Theory on Nonanticommutative Superspace]{}, JHEP [0712]{}, 059 (2007), 0710.1746\[hep-th\]. P. Aschieri, M. Dimitrijevi' c, F. Meyer and J. Wess, [Noncommutative Geometry and Gravity]{}, Class. Quant. Grav. [23]{}, 1883-1912 (2006), \[hep-th/0510059\]. V. G. Drinfel’d, [Quasi-Hopf algebras]{}, Leningrad Math. J. 1, 1419 (1990). V. Chari and A. Pressley, [A Guide to Quantum Groups]{}, Cambridge University Press, Cambridge (1995). H. Weyl, [Quantenmechenik und Gruppentheorie]{}, Z. Phys. [46]{}, 1 (1927). J. E. Moyal, [Quantum mechanics as a statistical theory]{}, Proc. Cambridge Phil. Soc. [45]{}, 99 (1949). J. Wess and J. Bagger, [Supersymmetry and Supergravity]{}, Princton, USA: Univ. Pr. (1992). C. Tamarit, [Noncommutative GUT inspired theories with $U(1)$, $SU(N)$ groups and their renormalisability]{}, 0910.5195\[hep-th\]. G. ’t Hooft and M. J. G. Veltman, [One loop divergencies in the theory of gravitation]{}, Annales Poincare Phys. Theor. [A20]{}, 69 (1974). R. E. Kallosh, O. V. Tarasov and I. V. Tyutin, [One Loop Finiteness Of Quantum Gravity Off Mass Shell]{}, Nucl. Phys. [B137]{}, 145 (1978). S. Giller, C. Gonera, P. Kosinski and P. Maslanka, [On the consistency of twisted gauge theory]{}, Phys. Lett. [B 655]{}, 80–83 (2007), \[hep-th/0701014\]. C. Gonera, P. Kosinski, P. Maslanka and S. Giller, [Space-time symmetry of noncommutative field theory]{}, Phys. Lett. [B 622]{}, 192 (2005), \[hep-th/0504132\]. ´ P. Aschieri, L. Castellani and M. Dimitrijevi' c, [Dynamical noncommutativity and Noether theorem in twisted $\Phi^4$ theory]{}, Lett. Math. Phys. [85]{}, 39–53 (2008), 0803.4325 \[hep-th\]. [^1]: Unlike the Moyal-Weyl $\star$-product, the $\star$-product (\[star\]) is finite and it does not introduce non-locality.	{ "pile_set_name": "ArXiv" }
--- abstract: 'The slow decay of charge carriers in polymer–fullerene blends measured in transient studies has raised a number of questions about the mechanisms of nongeminate recombination in these systems. In an attempt to understand this behavior, we have applied a combination of steady-state and transient photoinduced absorption measurements to compare nongeminate recombination behavior in films of neat [poly(3-hexyl thiophene)]{}(P3HT) and P3HT blended with [\[6,6\]-phenyl-C$_{61}$ butyric acid methyl ester]{}(PCBM). Transient measurements show that carrier recombination in the neat P3HT film exhibits second-order decay with a recombination rate coefficient that is similar to that predicted by Langevin theory. In addition, temperature dependent measurements indicate that neat films exhibit recombination behavior consistent with the Gaussian disorder model. In contrast, the P3HT:PCBM blend films are characterized by a strongly reduced recombination rate and an apparent recombination order greater than two. We then assess a number of previously proposed explanations for this behavior, including phase separation, carrier concentration dependent mobility, non-encounter limited recombination, and interfacial states. In the end, we propose a model in which pure domains with a Gaussian density of states are separated by a mixed phase with an exponential density of states. We find that such a model can explain both the reduced magnitude of the recombination rate and the high order recombination kinetics and, based on the current state of knowledge, is the most consistent with experimental observations.' author: - Julien Gorenflot - 'Michael C. Heiber' - Andreas Baumann - Jens Lorrmann - Matthias Gunz - Andreas Kämpgen - Vladimir Dyakonov - Carsten Deibel bibliography: - 'Papers.bib' title: 'Nongeminate recombination in neat P3HT and P3HT:PCBM blend films' --- Introduction ============ Organic solar cells based on polymer–fullerene blends have recently reached power conversion efficiencies as high as 10%. [@green2012review] Paradoxically, those high performances are achieved due to an only marginally understood peculiarity: the inefficiency of nongeminate charge carrier recombination. Such inefficient recombination has been observed in a number of polymer–fullerene blends.[@pivrikas2005a; @parkinson2008; @clarke2011; @zusan2014] The nongeminate recombination of two oppositely charged particles is predicted by the Langevin theory.[@langevin1903] This theory states that the recombination rate of electrons and holes ($R_\text{L}$) is governed by a second-order process, $$R_\text{L} = k_\text{L}np. \label{eqn:bimol}$$ where $n$ is the electron concentration, $p$ is the hole concentration, and $k_\text{L}$ is the Langevin rate constant. This rate constant is defined by assuming that the recombination event is much faster than the rate at which electrons and holes encounter one another. As a result, the Langevin rate constant depends on the charge carrier mobility of each species, $$k_{\text{L}} = \frac{e}{\varepsilon} (\mu_e + \mu_h) . \label{eqn:prefact}$$ where $e$ is the elementary charge, $\epsilon$ the dielectric constant, $\mu_e$ is the electron mobility, and $\mu_h$ is the hole mobility. This implies that for equal densities of electrons and holes, $R \propto n^2$, and an initial carrier density ($n_0$) will decay as $n(t) = n_0/(1+k_\text{L}n_0t)\propto t^{-1}$ in the absence of further photogeneration. For conditions typical of an organic solar cell, with an initial polaron density of 10$^{18}$ cm$^{-3}$, a mobility of 10$^{-4}$ cm$^2$/Vs, and a relative dielectric constant of 3.5, Langevin theory predicts that almost 85$\%$ of charge carriers should recombine nongeminately within 100 ns, which is the minimum time required for the charge extraction.[@guo2010a; @deibel2010review] Yet, external quantum efficiencies of 80$\%$ and higher have been reported under appropriate illumination in polymer–fullerene blends. [@deibel2010review] In [poly(3-hexyl thiophene)]{}(P3HT):[\[6,6\]-phenyl-C$_{61}$ butyric acid methyl ester]{}(PCBM) blends, the actual recombination rate is found to be up to 10$^{4}$ times slower than predicted by the Langevin theory,[@pivrikas2005a; @deibel2009] resulting in the characterization of a reduction factor ($\zeta$),[@deibel2008b] $$\zeta = \frac{R_\text{exp}}{R_\text{L}}. \label{eqn:zeta}$$ Furthermore, a variety of experimental methods have shown that the recombination rate in polymer–fullerene blends does not have the expected second-order kinetics. Instead, higher orders between 2.3 and 2.8 have been found at room temperature.[@nelson2003; @guo2010a; @foertig2009; @shuttle2010] Although under these conditions it is not possible to describe the higher order decay by reduced Langevin recombination based on Eqn. (\[eqn:bimol\]), such reduction factors are still being reported.[@kniepert2011; @mingebach2012; @wetzelaer2013] This discrepancy needs to be resolved in order to understand the detailed processes involved in nongeminate recombination. In this paper, we compare the dynamics of nongeminate recombination in a polymer–fullerene blend (P3HT:PCBM) to those in the neat polymer (P3HT) in order to better understand the origins of the reduced recombination rate and super-second order kinetics. Using pump-probe transient absorption spectroscopy (TA), we measure the polaron decay dynamics from the 10 ns to the 100 s timescale from 59-300 K. In spite of numerous studies concerning charge generation in neat P3HT and P3HT:PCBM blends, as well as charge recombination in P3HT:PCBM blends,[@piris2009; @guo2010a; @grzegorczyk2010; @howard2010; @ohkita2008] TA studies of the nongeminate recombination dynamics in neat P3HT are so far missing. Based on these results, we discuss the feasibility of several proposed models for nongeminate recombination in polymer–fullerene blends. Experimental Methods ==================== The experimental setup for steady state photo-induced absorption (PIA), as well as sample preparation, have been described elsewhere. [@delgado2009] P3HT was purchased from BASF (Sepiolid P200) and PCBM from Solenne. All materials were used without further purification. Solar cells prepared with these batches typically reach efficiencies over 3$\%$. [@schafferhans2010] All materials were dissolved in chlorobenzene at a concentration of 20 mg/ml. The films were deposited onto sapphire substrates by spin-coating and annealed at 140[[$^{\circ}$]{}]{}C for 10 min. Blends with a 1:1 weight ratio were studied. Films were prepared under a nitrogen atmosphere in a glove box. The thickness of the TA films was measured at approximately 300 nm by a profilometer. For TA experiments, samples were excited by a 5 ns pulse of a nitrogen/dye laser at a wavelength of 500 nm with a pulse energy of 25 J/cm$^2$. The generated polarons were probed using their characteristic absorption at 980 nm by an 80 mW cw laser. The decay of this absorption was measured using a FEMTO HCA-S-400M-IN preamplified InGaAs photodiode and recorded by a Tektronix oscilloscope.[@deibel2010review; @guo2010a] The change in optical density ($\Delta OD$) was computed from the transient signal and is directly related to the density of the absorbing species by the absorption cross section.[@kroeze2004] The hole mobility of neat P3HT films was also measured using the charge carrier extraction technique, OTRACE, as described in more detail elsewhere.[@baumann2012] OTRACE samples were prepared by spin-coating a solution of P3HT dissolved in chlorobenzene (30 mg/ml), resulting in 200 nm films as measured by a profilometer. For OTRACE measurements, the sample was directly transfered to a closed-cycle He cryostat without any exposure to air. A pulsed 10 W neutral white Rebel-LED was then used to generate charge carriers in the bulk of the P3HT film. The waveform was applied by an Agilent A81150A waveform generator, and the current transients were amplified by a FEMTO DHPCA-100 current amplifier and then recorded using an Agilent DSO90254A digital storage oscilloscope. Experimental Results ==================== Steady state photoinduced absorption ------------------------------------ In neat P3HT films, several species can coexist due to the lower efficiency of charge carrier photogeneration.[@deibel2010] At low temperatures, PIA spectra indeed exhibits bands due to several species as shown in Fig. \[fig:PIA\]. In addition to the polaronic features visible in the P3HT:PCBM blend films, there is a peak at 1170 nm (1.06 eV) that has been attributed to neutral species in P3HT.[@jiang2002a; @vanhal1999] At 30 K, the tail of this peak is overlapping with the P2 polaron peak (see Fig. \[fig:PIA\] inset). Yet, at room temperature, the lifetime of those excitonic species becomes too short to contribute to PIA after 1 ns.[@piris2009] We find that their contribution to the absorption at 980 nm continuously decreases when increasing the temperature over 30 K and vanishes above 142 K (Fig. \[fig:PIA\]). ![(Color online) Steady-state PIA spectra of neat P3HT (normalized to absorption at 1.2 eV). The blue line indicates the probe wavelength used for TA; above 140 K, absorption at this wavelength is proportional to the density of polarons in both P3HT and P3HT:PCBM. Inset: Steady-state PIA spectra of neat P3HT (continuous line) and P3HT:PCBM blend (broken line) at a temperature of 30 K. \[fig:PIA\]](PIA.pdf){width="8cm"} In contrast, the density of the neutral species in P3HT:PCBM blends decays on the sub-nanosecond scale even at low temperatures.[@grzegorczyk2010] This decay is explained by efficient charge carrier photogeneration as revealed by the high external quantum efficiency measurements in solar cells based on this blend.[@brabec2001] It is therefore safe to assume that polarons are the only absorbing species at 980 nm in the blend over the time range $10^{-8}$ to $10^{-4}$ s. We conclude that, above 140 K, the absorption at 980 nm is representative of the polaron density in both the neat and blend films. Transient absorption: absorption cross section and photogeneration ------------------------------------------------------------------ TA was probed at 980 nm, which corresponds to the maximum of the so-called P2 peak (see inset of Fig. \[fig:PIA\]). A number of spectroscopic studies including spin-sensitive methods have assigned this peak to P3HT polaron absorption.[@jiang2002a; @brown2001] The corresponding absorption cross section was determined as follows. For a P3HT:PCBM blend film excited by a 25 J/cm$^2$ pulse of 500 nm light, 10.3$\%$ of the incident light was reflected by the sample and 3.6$\%$ was transmitted, yielding an upper limit for the generated exciton density of 1.8$\times 10^{18}$ cm$^{-3}$. By solving the rate equations described by @howard2010,[@howard2010] including the nonlinear losses due to polaron–exciton annihilation at high excitation intensity, [@ferguson2011] and implementing a Gaussian exciton generation term to represent the laser pulse, we estimated the number of generated positive polarons to be (7.8$\pm$2.0)$\times 10^{17}$ cm$^{-3}$. We also obtained a polaron–exciton annihilation rate of (1.7$\pm$0.7)$\times 10^{-7}$ cm$^3$/s from the excitation intensity dependence of the initial change of optical density ($\Delta OD_0$). From the corresponding change in the optical density ($\Delta OD$), the absorption cross section of P3HT polarons at 980 nm in blend films was evaluated at (1.9$\pm$0.5)$\times 10^{-16}$ cm$^2$, which is in a similar range as recently reported for neat P3HT films.[@leijtens2013] The transient decays of the change in optical density for the neat P3HT and P3HT:PCBM blend films are shown in Fig. \[fig:transients\]. The initial change of optical density at 10 ns ($\Delta OD_0$), which includes geminate recombination and exciton–polaron annihilation, [@howard2010] is virtually temperature independent and is only slightly lowered at temperatures approaching 300 K due to a faster onset of nongeminate losses at higher temperatures. This finding is consistent with earlier reports of temperature independent charge carrier photogeneration in polymer–fullerene blends.[@pensack2009; @grzegorczyk2010; @street2010a] For both neat and blend films, the polaron decay beyond 10 ns is due to nongeminate recombination, with an increasing recombination rate for higher temperatures. In the following sections, we will focus on these nongeminate losses. ![(Color online) (a) Transient absorption decays in neat P3HT for different temperatures (solid lines). The dashed lines show fits including only second-order recombination and the dotted lines show fits accounting for contributions by both first and second-order decay. The asymptotes corresponding to a purely second-order decay are shown for comparison for 111 K, 91 K and 59 K (dashed lines). (b) Transient absorption decays in P3HT:PCBM (solid lines) and power law fits (dashed lines). Typical shape of first and second-order decays (dotted lines) are shown for comparison. \[fig:transients\]](transients.pdf){width="8cm"} Transient absorption: neat P3HT {#sec:exp-p3ht} ------------------------------- In order to gain a deeper understanding of the polaron dynamics, we compare the experimental decays to analytical models based on continuity equations. In the absence of any external contributions (injection or photoexcitation from the ground state) after exciton generation by the laser pulse at $t=0$, the continuity equation describing the total polaron density ($n$) is $$\frac{dn}{dt} = -R \label{eqn:continuity}$$ for $t>0$, where $R$ is the recombination rate. Although Langevin recombination (Eqn. (\[eqn:bimol\])) is the expected loss mechanism for separated polarons, we also consider a first order decay.[@deibel2010review] The sum of a first and a second-order term is able to perfectly fit the decays observed in neat P3HT at temperatures below 140 K, whereas at higher temperatures, the decays are found to be purely second-order. These findings are in agreement with the assignment of the absorption signal in neat P3HT at 980 nm to polarons at temperatures above 140 K and to a sum of contributions from both polarons and neutral species at lower temperatures. These neutral species could be triplet excitons,[@vanhal1999] interchain singlet excitons,[@jiang2002a] or polaron pairs [@howard2010] and are outside the scope of this article. As predicted by Langevin theory, the recombination of polarons in neat P3HT exhibits second-order kinetics as shown in Fig. \[fig:rec\_order\]b. In addition, the temperature dependence of the second-order recombination coefficient ($k_\text{br}$) obtained from the fits in Fig. \[fig:transients\]a is compared to Langevin recombination coefficients ($k_\text{L}$) calculated from temperature dependent mobility measurements in Fig. \[fig:rec\_order\]a. Assuming equal electron and hole mobilities and a dielectric constant of 3.5, Eqn. (\[eqn:prefact\]) was used to calculate the Langevin recombination rate coefficient from several different mobility measurements. The recombination coefficient derived from the transients is very similar to Langevin theory at 250 K when compared to coefficients derived from our OTRACE experiments and from previous CELIV measurements.[@mozer2005a] ToF measurements also indicate similar magnitudes and temperature dependencies as the OTRACE and CELIV measurements shown here.[@mozer2005a; @mauer2010] However, the rate coefficients determined from the transients demonstrate a much weaker temperature dependence than observed in mobility measurements. At 150 K the measured recombination coefficient is more than one order of magnitude greater than expected from Langevin theory, although it remains far below the calculated Langevin recombination rate using the temperature independent local mobility determined by time-resolved microwave conductivity (TRMC).[@grzegorczyk2010] ![(Color online) (a) Recombination coefficients extracted from neat P3HT transients (Fig. \[fig:transients\]) compared to Langevin coefficients calculated from experimental mobility measurements using Eqn. (\[eqn:prefact\]) (b) Apparent recombination order as function of temperature for neat P3HT and P3HT:PCBM blend films.[]{data-label="fig:rec_order"}](recombination_analysis.pdf) Transient absorption: P3HT:PCBM blend ------------------------------------- While the dynamics of charge recombination in neat P3HT appear Langevin-like, the transient absorption signal in the P3HT:PCBM blend exhibits a much slower decay (Fig. \[fig:transients\]b), which is similar to previous reports. [@guo2010a; @foertig2009; @shuttle2010] It is characterized by a reduced recombination rate that does, however, depend on time (or carrier concentration). The recombination order exceeds 2 already at room temperature, increasing to about 7 at 30 K as shown in Fig. \[fig:rec\_order\]b. Since the recombination mechanism is still assumed to be between one electron and one hole, the resulting experimentally determined nongeminate recombination rate is expressed as $$\begin{aligned} R_\text{exp} = k_{\text{exp}}(n) n^2 \propto n^{\lambda+1}, \label{eqn:R_exp}\end{aligned}$$ where $k_\text{exp}$(n) is the carrier concentration dependent rate coefficient and $\lambda+1$ is the recombination order. This form is convenient because the slope of the polaron decay on the log-log plots shown in Fig. \[fig:transients\] is equal to $-1/\lambda$. To produce a recombination order of $\lambda+1$, the recombination rate coefficient must then take on the form $$\begin{aligned} k_\text{exp}(n) \propto n^{\lambda-1}. \label{eqn:k_exp}\end{aligned}$$ Discussion ========== Multiple Trapping and Release (MTR) Model ----------------------------------------- It is well known that charge carrier trapping plays a significant role in P3HT and other organic semiconductors.[@schafferhans2008] Therefore, we approach the interpretation of our experimental data using the multiple-trapping-and-release (MTR) model.[@schmidlin1977; @noolandi1977; @oelerich2012] With this model, the overall charge carrier density ($n$) is split into two populations, free carriers ($n_c$) and trapped carriers ($n_t$). The free carriers are assumed to move at a speed defined by the free carrier mobility ($\mu_c$), and the trapped carriers are assumed to be immobile. Over time, trapped carriers are thermally excited to become free carriers, and free carriers relax into trap states. However, under steady state conditions, the ratio of free to trapped carriers is assumed to be constant. To start our analysis, we assume that there is a very low concentration of intrinsic dark carriers, that the concentrations of electrons and holes are equal due to the symmetric nature of photogeneration, and that mobilities of electrons and holes are equal. Neat P3HT: Langevin-like recombination -------------------------------------- To analyze the second-order decay observed in the neat P3HT measurements, it is then assumed that carrier recombination can only occur between free electrons and free holes, free electrons and trapped holes, or trapped electrons and free holes.[@kirchartz2011; @foertig2012; @deibel2013] As a result, the recombination rate is given by $$\begin{aligned} R \approx \frac{e}{\varepsilon} (2 \mu_{c} n_c^2 + 2 \mu_{c} n_c n_t) = \frac{e}{\varepsilon} 2 \mu_{c} n_c (n_c + n_t) \label{eqn:R_neat}\end{aligned}$$ We then define the fraction of free charge carriers, $\Theta$, where $\Theta=n_c/n$, resulting in a final recombination rate $$\begin{aligned} R \approx 2\frac{e}{\varepsilon}\mu_c \Theta n^2. \label{eqn:R_theta}\end{aligned}$$ As a result, the effective macroscopic mobility ($\mu$) is governed by the free carrier mobility and the fraction of free carriers, $\mu = \Theta\mu_c$. Under these conditions, Eqn. (\[eqn:R\_theta\]) is equivalent to Eqn. (\[eqn:bimol\]), and within the MTR model, $k_\text{L}$ can then be expressed as $$k_\text{L} = 2 \frac{e}{\epsilon}\mu_c \Theta. \label{eqn:MTRprefact}$$ Within this framework, we note that if $\Theta(n)$ is not constant, the macroscopic mobility would be expected to be carrier concentration dependent, resulting in super-second-order recombination. However, if $\Theta(n)$ is constant, the macroscopic mobility would be independent of the charge carrier concentration, and second-order recombination is expected. Considering the charge carrier dynamics measured here, we observe second-order decay between 140 and 300 K (Fig. \[fig:rec\_order\]), implying that neat P3HT can be well described by Eqn. (\[eqn:bimol\]) and that both the macroscopic mobility ($\mu$) and $\Theta$ are independent of the carrier concentration. We assume for now, in accordance with the findings of Oelerich et al.,[@oelerich2012] that P3HT has a Gaussian density of states (DOS) distribution. Then, within the framework of the Gaussian disorder model (GDM), in which $\mu(T) \propto \exp( -( 2\sigma/3k_BT)^2)$,[@bassler1993] we determined the standard deviation of the DOS ($\sigma$) to be 37 meV. Previously, a value of 56 meV was found for holes by photocurrent transient measurements on much thicker samples (c.f. Fig. \[fig:rec\_order\]a).[@mauer2010] Using the GDM, the macroscopic mobility in neat disordered materials has been considered previously by hopping master equation[@pasveer2005] and the MTR model.[@oelerich2012] For $\sigma=$37 meV and temperatures between 140 and 300 K, the mobility depends on the carrier concentration only if more than $10^{-3}$ of the states are occupied.[@pasveer2005] Below that limit, the mobility is predicted to be independent of the carrier concentration. In contrast, if neat P3HT has an exponential DOS, the mobility would be expected to depend on the carrier concentration in all regimes.[@oelerich2012] Therefore, the observed second-order decay implies a carrier concentration independent mobility that is consistent with a Gaussian DOS but not with an exponential DOS. This finding is consistent with the analysis of mobility measurements by Oelerich et al.[@oelerich2012] However, as shown in Fig. \[fig:rec\_order\], we find that the measured recombination coefficient has a weaker temperature dependence than expected from OTRACE and CELIV hole mobility measurements, although not temperature independent as shown in the TRMC measurements. TRMC measures the high frequency photoconductivity, which is assumed to result from the motion of free carriers and to be proportional to $n_c \mu_c$, in contrast to the macroscopic mobility ($\mu$) measured by the other techniques, which calculate the mobility based on long-range charge transport of all carriers. As a result, mobility values derived from TRMC are usually much higher, but it is unclear why the temperature dependence is so weak. We would expect that the free carrier concentration ($n_c$) should be temperature dependent and cause the TRMC signal to have a stronger temperature dependence. This discrepancy makes it difficult to rely on mobility measurements derived from TRMC experiments when describing mechanisms that require longer range charge transport until further studies clarify the nature of the mobility measured by TRMC. In any case, the experimentally determined rate coefficients are greater than the Langevin rates derived from macroscopic mobility measurements at lower temperatures and have a weaker temperature dependence. It is plausible that the mobility that characterizes the charge motion required for recombination is different than that for long range macroscopic charge transport. For example, charge trapping may be effectively shallower for transport on the [.17ex]{}10 nm length scale than on the [.17ex]{}100 nm length scale due to spatial homogeneity arising from the presence of crystalline and amorphous domains. Another possibility is that the electron and hole mobilities have different temperature dependencies. However, further detailed studies are needed to test these concepts. As a result, while Langevin theory works fairly well to describe nongeminate recombination near room temperature, questions remain as to why the temperature dependence is weaker than expected from macroscopic mobility measurements. Nevertheless, the recombination rate has no major reduction factors and is actually slightly greater than expected at lower temperatures. This implies that P3HT is sufficiently homogeneous such that mobile charge carriers can reach their recombination partners everywhere. Comparing to previous recombination measurements on neat P3HT films, our observation of second-order recombination dynamics in neat P3HT films is in contrast to the first-order decays observed in TRMC measurements by Ferguson et al.[@ferguson2011] In their study, they attribute the first-order decay to the presence of a significant dark carrier concentration ([.17ex]{}$10^{19}$ cm$^{-3})$. If the observed behavior is indeed due to the presence of dark carriers, our measurements suggest that the neat P3HT samples that we have tested have a significantly lower dark carrier concentration. Our observation of second-order kinetics indicates that the dark carrier concentration in our samples is less than the range of photogenerated carrier concentrations tested. For an estimated initial exciton concentration of [.17ex]{}$10^{18}$ cm$^{-3}$ due to the excitation laser pulse and an upper bound of [.17ex]{}10% carrier yield, we estimate the photogenerated carrier concentration tested here is in the range of [.17ex]{}$10^{15}$ cm$^{-3}$ to [.17ex]{}$10^{17}$ cm$^{-3}$. As a result, we estimate that the dark carrier concentration in our samples is less than [.17ex]{}$10^{15}$ cm$^{-3}$. P3HT:PCBM blends: Langevin recombination? ----------------------------------------- In contrast to neat P3HT, the P3HT:PCBM blend films clearly display slower recombination and super-second-order decay (Fig. \[fig:rec\_order\]b). Under these conditions, the reduction factor defined in Eqn. \[eqn:zeta\] and calculated using Eqn. \[eqn:bimol\] and Eqn. \[eqn:R\_exp\] becomes carrier concentration dependent with the form $$\zeta = \frac{R_\text{exp}}{R_\text{L}} = \frac{k_\text{exp}(n)}{k_\text{L}} \propto n^{\lambda-1}. \label{eqn:zeta2}$$ To understand the origin of this reduction factor, we discuss several previously proposed hypotheses. Based on the measurements presented here and in previous studies in the literature, we attempt to eliminate those that are inconsistent with the current state of experimental knowledge, highlight those that are still feasible, and finally direct researchers to areas where further measurements are needed. We emphasize that any well-suited model must be able to account for both the magnitude and the carrier concentration dependence of the reduction factor. Previously, it has been suggested that due to the presence of lamellar crystals in P3HT domains, which may promote two-dimensional transport, the resulting recombination behavior in P3HT:PCBM blends is more accurately represented by a two-dimensional Langevin recombination model.[@juska2009] However, our observation that the neat P3HT films, which also have lamellar crystalline domains, do not demonstrate these same characteristics suggests that two-dimensional transport is not a dominant factor and that a two-dimensional Langevin recombination model is not appropriate. In addition, kinetic Monte Carlo simulations implementing anisotropic mobility also conclude that the anisotropy effect is likely to be too weak to be the dominant factor in P3HT:PCBM blends.[@groves2008b] Another previously suggested explanation for the super-second order recombination kinetics is the presence of carrier concentration gradients near the electrodes in operational devices.[@deibel2009] While carrier concentration gradients may enhance this effect in devices, our observation here of similar kinetics on samples without electrodes suggests that is not likely be the dominant cause. With these hypotheses ruled out, we move now to a more detailed discussion of the remaining concepts in the following subsections. ### Effect of phase separation One important difference between traditional Langevin theory and the P3HT:PCBM blend system is the presence of a complex nanoscale phase separated morphology[@erb2005] that spatially limits the motion of the electrons and holes and limits the possible places where recombination can occur. The reduction factor has been previously attributed to the presence of phase separation.[@koster2006; @baumann2011] To assess the effect of phase separation, we derive and compare the expected recombination rate equations for a homogeneous blend and a phase separated blend with pure domains. For a homogeneous blend, similar to neat P3HT, in a system with charge carrier trapping due to energetic disorder,[@schafferhans2010; @foertig2012] the recombination rate can be approximated by Eqn. (\[eqn:R\_theta\]) and, equivalently, by Eqn. (\[eqn:bimol\]). As a result, the standard Langevin rate equation would be expected in a homogeneous blend. Moving now to a phase separated blend, carriers trapped in the interior of the domains are unable to undergo recombination, which should reduce the overall recombination rate. If we assume that very few charge carriers are trapped close to the interface, the recombination rate should be dominated by reactions between free electrons and free holes. In this framework, the resulting recombination rate is $$R \approx 2\frac{e}{\varepsilon} \mu_{c} n_c^2 \approx \Theta k_\text{L} n^2. \label{eqn:R_ps}$$ Here, the trapped charge carriers in the bulk of the domains are protected from recombination as long as they are trapped,[@baumann2011] lowering the overall recombination rate by a factor of $\Theta$. However, given the nanoscale dimensions of the domains ($\approx15$ nm),[@vanbavel2009; @pfannmoeller2011] there can actually be a large fraction of the P3HT volume near the interface. As a simple example, given a spherical domain with a 15 nm diameter, 35% of the volume is within 1 nm of the interface, and it can be expected that 35% of the trapped carriers can participate in recombination. Therefore, the amount of carriers trapped close to interface is, in fact, not likely to be negligible. By including recombination between free carriers and these carriers trapped near the interface, the recombination rate equation becomes $$R \approx \Theta k_\text{L} n^2 + \chi (1-\Theta) k_\text{L} n^2, \label{eqn:R_phi_i}$$ where $\chi$ is the interfacial volume fraction, the fraction of the donor and acceptor volume at the interface with respect to the total volume. Furthermore, if we assume that most carriers at any given time are trapped ($\Theta\ll1$) and that $\Theta\ll\chi$, then $$R \approx \chi k_\text{L} n^2. \label{eqn:R_phi}$$ As a result, this scenario predicts that the reduction factor is approximately equal to $\chi$. However, the magnitude of $\chi$ expected in a nanostructured morphology ($>10^{-1}$) is closer to unity than previously measured reduction factors ([.17ex]{}$10^{-3}$) at room temperature.[@deibel2008b] In addition, a simple phase separation model has no way of explaining the super-second order kinetics. As a result, we find it unlikely that phase separation inherently causes the apparent major deviations from Langevin theory. ### Carrier concentration dependence of mobility Assuming that carrier recombination is still encounter-limited as assumed in Langevin theory, the recombination rate should still depend on the carrier mobility. Up to now, we have assumed that the carrier mobility is independent of the carrier concentration, but a more complex carrier concentration dependence must be considered. @shuttle2010 have attempted to explain the super-second order decay by assuming a carrier concentration dependent mobility in which $\mu(n) \propto n^{\lambda-1}$.[@shuttle2010] Such behavior would only be expected if the materials were to have an exponential DOS. However, we have recently shown that this explanation may not generally hold,[@rauh2012] but we point out here that neither of these studies used a method that probes the charge carrier mobility directly. Therefore, these previous conclusions need to be verified. In addition, @savenije2011 compared TRMC and TA measurements on P3HT:PCBM thin films and concluded that the mobility in P3HT:PCBM blends is time independent on the timescale of tens of nanoseconds onwards, indicating a carrier concentration independent mobility for the range tested.[@savenije2011] However, as discussed in section IV.B on neat P3HT, TRMC mobility measurements may probe behavior that is significantly different than the more macroscopic mobility important for describing nongeminate recombination. As a result, further concentration dependent mobility studies are needed to completely rule out carrier concentration dependent mobility as a main cause of the super-second order recombination kinetics. If a carrier concentration dependent mobility is to be the dominant cause of the observed recombination kinetics, this relationship should be proportional to the concentration dependence of the experimental recombination prefactor, $k_\text{exp}$, shown in Fig. \[fig:k\_vs\_n\]. However, given that neat P3HT mobility is only weakly carrier concentration dependent, consistent with a Gaussian DOS, it is a reasonable assumption that at least the pure P3HT domains should demonstrate similar behavior. But even if the P3HT domains do have a Gaussian DOS, another possibility is that the PCBM domains have an exponential DOS and a mobility that dominates the recombination rate. However, space-charge limited current (SCLC) measurements on PCBM have indicated behavior consistent with a Gaussian DOS,[@mihailetchi2003] but SCLC measurements are also performed at much higher carrier concentrations than are present in working solar cells. To clarify this behavior further, carrier concentration dependent mobility measurements on neat PCBM films are needed as well. ![(Color online) Predicted carrier concentration dependence of the mobility when assuming super-second order recombination is caused only by a carrier concentration dependent mobility. \[fig:k\_vs\_n\]](k_vs_n.pdf) ### Effect of non-encounter limited recombination It has also been proposed that nongeminate recombination is not encounter limited as assumed in Langevin theory.[@hilczer2010; @ferguson2011] In this case, if the actual recombination mechanism itself is slow, the decay of charge carriers in the blend does not depend solely on the encounter probability. Instead of the electron and hole recombining immediately when reaching each other, they form an intermediate polaron pair state that can recombine after some time but can also re-dissociate into free charges. The experimental recombination rate is then defined $$R \approx -\frac{dn}{dt} = k_L n^2 - k_d [PP], \label{eqn:non-encounter}$$ where $k_d$ is the polaron pair dissociation rate and $[PP]$ is the polaron pair concentration. For this to have a major effect, the second term must be much larger than the Langevin term, which means the polaron concentration must persist over long timescales and the polaron pair dissociation rate must be fairly fast. To determine the conditions in which this would occur, we first describe the polaron pair rate equation as $$\frac{d[PP]}{dt} = k_L n^2 - k_r [PP] - k_d [PP], \label{eqn:polaron-pair_rate}$$ where $k_r$ is the rate of the final polaron pair recombination event by which the electron finally returns to the ground state. For polaron pairs to persist in the system, $d[PP]/dt$ must not be much less than zero, and in a special case, when the polaron pair concentration is constant ($d[PP]/dt=0$) and the reduction factor ($\zeta$) is very small, $$\zeta \approx \frac{k_r}{k_d} \label{eqn:zeta_non-encounter}$$ However, if the polaron pair concentration is changing over time, a more complicated expression is necessary. Nonetheless, given the highly efficient charge separation that occurs in P3HT:PCBM blends, it is likely that the dissociation rate is significantly faster than the recombination rate. As a result, while a significant reduction in the observed polaron decay rate could be expected, this model still does not explain the origins of the super-second-order kinetics. ### Effect of interfacial states Another important aspect to consider is the presence of a third mixed phase with unique materials properties compared to the pure phases. P3HT:PCBM blends have been shown to have a more complex morphology than simply pure donor and pure acceptor domains,[@collins2010] and the presence of a mixed amorphous phase has been clearly identified.[@pfannmoeller2011] In addition, the presence of deep states that go beyond a superposition of the tail states of the separate pure materials has been experimentally measured.[@vandewal2009a] These deeper states originate from the close interaction of donor and acceptor molecules at the heterointerface. Taking this into account, Street et al. have proposed recombination via interfacial states.[@street2010; @deibel2010b; @street2010b] With this in mind, we consider a model in which pure domains are separated by an interfacial mixed region that contains a DOS that is different from those present in either of the pure domains. We note that implementing interfacial mixing without a separate DOS simply increases $\chi$, as previously defined in subsection 1, and cannot explain the observed behavior. With separate DOS distributions, however, it is plausible that the macroscopic mobility would be dominated more so by charge transport within the pure domains containing a Gaussian DOS, as indicated by our measurements on neat P3HT, but that the actual recombination event is governed mainly by the spatial and energetic properties of the interfacial regions containing an exponential DOS, which has been indicated by defect spectroscopy.[@foertig2012; @presselt2012] To determine the expected recombination rate in this more complex scenario, we need to consider two separate contributions to the recombination rate, the behavior of the carriers in the pure domains and in the interfacial regions. First, we assume that the majority of the carriers are trapped and that the contribution from free–free recombination is negligible. As a result, the dominant recombination mechanism occurs when free carriers from the pure domains ($n_{c,p}$) are transported to the interfacial regions and recombine with carriers that are already present within the interfacial regions ($n_i$). In this case, the resulting recombination rate becomes $$R \approx 2\frac{e}{\varepsilon}\mu_{c,p} n_{c,p} n_i, \label{eqn:R_interfacial}$$ where $\mu_{c,p}$ is the mobility of the free carriers in the pure phases. Rewriting this in terms of the overall carrier concentration ($n$), where $n=n_{c,p}+n_{t,p}+n_{c,i}+n_{t,i}$, the recombination rate becomes $$R \approx \Phi (1-\Phi) k_L n^2, \label{eqn:R_interfacial_n}$$ where $\Phi$ is the fraction of carriers in the pure phase with respect to all carriers, $n_p/(n_p+n_i)$, and $k_L$ is derived from the mobility of the pure phases. This model can explain both the magnitude and the carrier concentration dependence of the reduction factor when $\Phi$ is large and carrier concentration dependent. Given studies that have indicated an energetic driving force for carriers to diffuse from amorphous mixed regions to more ordered pure domains,[@mcmahon2011; @jamieson2012] it is probable that $\Phi$ would be large, and given different DOS distributions, it is possible that the density of occupied states would be populated in different proportions at different overall carrier concentrations. Here, to give super-second order kinetics, interfacial states would have to fill up proportionally faster than the states in the pure phases when increasing the overall carrier concentration. Further theoretical and experimental studies are needed to test this model in more detail, but given the current state of knowledge and the critical analysis presented here, we find it to provide the most complete explanation to date. Conclusions =========== To conclude, we have used transient absorption spectroscopy to monitor the nongeminate polaron decay in neat P3HT and P3HT:PCBM blend films. In the neat polymer, we observed Langevin-like recombination at temperatures above 140 K with second-order kinetics and a recombination coefficient that is slightly less temperature dependent than macroscopic mobility measurements. To analyze the results, we have used a multiple trapping and release (MTR) model to derive the expected recombination rate equations. For neat materials, the MTR model predicts recombination dynamics consistent with Langevin theory, and the neat P3HT measurements appear to be mostly consistent with this model, aside from the weaker temperature dependence. Most importantly, though, no significant reduction factor was observed in neat P3HT, and dark carriers were not found to play a role in the recombination kinetics. In contrast, the recombination dynamics in the blend films were characterized by a reduced recombination rate and super-second-order recombination kinetics. To narrow down the possible explanations for this behavior, we first eliminated several different models previously proposed, including a two-dimensional Langevin model and carrier concentration gradients. To understand the effect of phase separation, the MTR model was used to derive the expected recombination rate for a homogeneous and phase separated blend. However, phase separation alone was shown to only slightly reduce the recombination rate. In addition, we argued that the mobility is not likely to be strongly carrier concentration dependent but have identified that further measurements are needed to rule it out as the sole contributor to the higher order recombination kinetics. We then considered the idea that nongeminate recombination is not encounter limited as assumed by Langevin theory and found that the recombination rate could indeed be significantly reduced from the rate predicted by Langevin theory. However this would still be unable to explain the origins of the recombination order. Finally, we considered the effects of interfacial states, which have been previously identified and proposed to play a significant role in the recombination behavior. Using the MTR model, we then derived the recombination rate expected when there are pure domains with a Gaussian density of states that are separated by a mixed interfacial phase with an exponential density of states. This scenario is expected to produce both a reduced recombination rate and super-second-order recombination kinetics. While still a qualitative model, we propose that it is most consistent with the available experimental data to date. The current work is supported by the Bundesministerium f[ü]{}r Bildung und Forschung in the framework of the GREKOS project (contract no. 03SF0356B) and the European Commission through the Human Potential Program (Marie-Curie RTN SolarNType contract no. MRTN-CT-2006-035533) and the Deutsche Forschungsgemeinschaft, DFG under the contract INST 93/623-1 FUGG. C.D. gratefully acknowledges the support of the Bavarian Academy of Sciences and Humanities.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We have investigated the effect of sheared equilibrium flows on the $m = 1, n = 1$ resistive internal kink mode in the framework of a reduced magnetohydrodynamic model in a periodic cylindrical geometry. Our numerical studies show that there is a significant change of the scaling dependence of the mode growth rate on the Lundquist number in the presence of axial flows compared to the no flow case. Poloidal flows do not influence the scaling. We have further found that viscosity strongly modifies the effect of flows on the (1,1) mode both in the linear and nonlinear regime. Axial flows increase the linear growth rate for low viscosity values, but they decrease the linear growth rate for higher viscosity values. In the case of poloidal flows the linear growth rate decreases in all cases. Additionally at higher viscosity, we have found strong symmetry breaking in the behaviour of linear growth rates and in the nonlinear saturation levels of the modes as a function of the helicities of the flows. For axial, poloidal and most helical flow cases, there is flow induced stabilisation of the nonlinear saturation level in the high viscosity regime and destabilisation in the low viscosity regime.' author: - 'J. Mendonca' - 'D. Chandra' - 'A. Sen' - 'A.Thyagaraja' bibliography: - 'new\_rev\_unmarked.bib' title: 'Visco-resistive MHD study of internal kink(m=1) modes' --- Introduction ============ The $m=1, n=1$ internal kink instability is of great importance in tokamaks and has beeen extensively studied in the past by several authors, notably[@VonGoeler1974; @Kadomtsev; @rosenblu; @Porcelli1996; @Monticello1986]. The (1,1) mode arises within the q=1 rational surface (where q is the safety factor), when the q at the axis is smaller than 1. It can trigger sawtooth oscillations which can influence plasma quality and confinement[@VonGoeler1974; @Kadomtsev]. Monticello et. al. [@Monticello1986] have given an overview of the research on the (1,1) mode and its importance, particularly in the context of research on the sawtooth oscillations. It is well known that flows are a common occurence in a tokamak, which can be generated intrinsically[@Rice2007] or induced externally e.g. by unbalanced NBI injection [@Menard2003; @Menard2005]. Experiments on NSTX have shown a significant increase of sawtooth period that is attributed to a fast rotation of the plasma[@Menard2003; @Menard2005]. Experimental studies on sawteeth phenomena in presence of NBI in JET [@Chapman2007a; @Nave2006], MAST [@Chapman2006a], and TEXTOR [@Chapman2008c] have further shown that there is an asymmetry in sawtooth period depending on the direction of the NBI. The sawtooth period increases with an increase in co-NBI power, and decreases with an increase in counter-NBI power. Thus, these experiments have shown that flow can have a stabilising or destabilising effect on the kink mode depending on the direction of flow. However, there still does not exist a full understanding of the effect of flows on the $m=1,n=1$ kink instability. A number of past studies have addressed this question. In one of the earliest such studies carried out in a slab geometry, Ofman et. al.[@Morrison12] have shown that small flow shear has a stabilising influence on the $m=1$ resistive mode. Numerical studies by Kleva and Guzdar[@Guzdar1234] show that toroidal sheared flow close to the sound speed can completely stabilise the (1,1) mode. Shumlak et al.[@Shumlak1995] have also found a similar stabilising effect due to a sheared axial flow on the (1,1) mode in a cylindrical Z-pinch. On the other hand, Gatto et. al. [@gatto] have found sheared axial flows to have a destabilising effect on the $m=1$ mode in a reverse field pinch configuration. Naitou et al. [@naitou_kobayashi_tokuda_1999] have studied the effect of poloidal flow on the kink mode in kinetic and two fluid regimes, and noted a stabilisation of the kink mode that can possibly be related to sawtooth stabilisation. Studies by Mikhailovskii et. al.[@Mikhailovskii2008aa], Wahlberg et. al.[@wahl_bond] and Waelbrock[@wael12] show that toroidal and poloidal rotations are a stabilising factor for the internal kink mode. Chapman et. al.[@Chapman2006a] have explained the asymmetry in sawtooth period in terms of the relative direction of the plasma flow with respect to the diamagnetic drift. They postulated that the toroidal component of the diamagnetic drift adds to the toroidal rotation for co-current flow but it reduces the toroidal rotation for counter current flow. Therefore, there are conflicting results in the literature regarding the nature of stabilisation due to flows depending on the parameter regime of the studies. Recent analytic calculations by Brunetti et. al.[@Brunetti2017] have found that small flow shear has a destabilising effect on the (1,1) mode, but a large flow shear can stabilise it. It may be noted that most of the past flow studies have been done in the low viscosity regime. However, viscosity can be high in tokamak operations, particularly due to enhancements from turbulent effects and could therefore significantly influence the effect of flow shear on the internal kink mode. For example, Maget et. al.[@Maget12], Wang et. al.[@Wang2015], Tala et. al.[@Tala2011] and Takeda et. al.[@Takeda2008] have shown that Magnetic Prandtl number in advanced tokamak scenarios can be as high as 100, and stability results in the high viscosity regime can be significantly different from results of the low viscosity regime. Chen et. al.[@Chen1990a] and Ofman et. al.[@Ofman1991] have given detailed analytical calculations as to how viscosity can modify shear flow effects for constant-$\psi$ and nonconstant-$\psi$ for the resistive tearing mode instability. Wang et. al.[@Wang2015] have also reported from their simulation studies of (2,1) tearing modes in the presence of flows, which showed a destabilisation at lower viscosity and stabilisation at higher viscosity. They predict that this is due to the distortion of magnetic island structures at higher viscosity as reported by Ren et. al.[@Ren1999] and La Haye et. al.[@Haye2009]. Thus, viscosity is an important contributing factor and can change the nature of the effect of flows significantly. In this paper we have addressed this issue and investigated the stability of the (1,1) mode in the presence of sheared flows over a range of viscosity regimes. We indeed find that the high viscosity results are often very different from the low viscosity results. In our study, we have systematically examined the effects of several kinds of sheared flows on the (1,1) mode, namely axial, poloidal and combinations of both kinds of flows in the linear as well as nonlinear regimes.Various non-dimensional parameters are used to characterise our results, such as S (the Lundquist number, which measures resistivity), Pr (Prandtl Number, which measures viscosity), M (Alfvén Mach number) and $\lambda$(a measure of the equilibrium flow shear). These linear and nonlinear studies on the (1,1) mode were obtained using the CUTIE code [@Thyagaraja2000]. Our principal findings are as follows. To begin with, we have done the linear scaling studies of the $m=1,n=1$ mode in the absence of flow. Here, the variation of linear growth rates have been studied for different S and Pr values. The obtained scalings are in agreement with past analytic theory results in the no flow case[@Porcelli1987]. With the application of sheared axial flows, a significant change in the scaling of the growth rates is observed. However, in the presence of poloidal flow, there is no such change in scaling as compared to the no flow case. In our linear studies we have noticed that axial flows destabilise the mode in the low viscosity regime, but it stabilises in the high viscosity regime as compared to the no flow case. On the other hand, poloidal flow always tends to stabilise the linear growth rate. For pure axial and poloidal flows, the results do not change if we change the direction of the flow. This symmetry is broken for helical flows where the time evolution of the modes show a significant dependence on the helicity of the flows even in the linear regime. In the nonlinear regime, there is mostly a reduction of the nonlinear saturation level of the (1,1) mode for both sheared axial and poloidal flows in the high viscosity regime, while in the low viscosity regime, the poloidal and axial flows are destabilising in nature. Helical flows show a strong stabilisation for positive helicity and in most cases, weak stabilisation for negative helicity in the high viscosity regime. In the low viscosity regime, this symmetry breaking of helical flow results gets significantly diminished. This paper is organised in the following manner. In section \[model\], we have described the reduced magnetohydrodynamic(RMHD) model of plasma in a cylindrical geometry. In section \[linear results\], we have studied the (1,1) mode in the linear regime. Here, we have described studies of the growth rate scaling in the absence of flow, and compared our results with analytical results from the literature. Then we have repeated these studies in the presence of flow. We have done these studies both in the low and high viscosity regimes. In section \[nonlinear results\] we have studied the (1,1) mode in the nonlinear regime in the absence of flow as well as in presence of axial, poloidal and helical flows. Section \[discussion\] provides a brief summary and a discussion of the results. Model ===== Our numerical investigations have been carried out in the framework of a reduced MHD model in a periodic cylinder geometry $(\rho,\theta,z)$,($\rho$ being the radial coordinate, $\theta$ being the azimuthal coordinate, and $z$ being the axial coordinate) defined in terms of the minor radius, $a$, and the major radius, $R_{0}$. Using normalised coordinates, we set $\rho = r/a$, $r$ being the radial distance, namely $0\leqslant\rho\leqslant1;0\leqslant\theta,\zeta\leqslant2\pi;\zeta=z/R_{0}$. The model utilises CGS electrostatic units. We have Fourier expanded the fields, and split the equations into a set of “mean" equations consisting of the (0,0) components of the field and a set for the “fluctuating" components consisting of the remaining terms. The mean equations can alternatively be obtained by averaging the full equations over the flux surface. In the linear runs we have solved the mean equations once, while in the nonlinear runs, the mean equations are co-evolved with the equations for fluctuating quantities. The fluctuation equations are : $$\label{Eq.1} \begin{aligned} & \frac{\partial\tilde{W}}{\partial t}+ \mathrm{\mathbf{v_{0}}}\cdot\nabla\tilde{W} +v_{A}\nabla_{\parallel}\rho_{s}^{2}\nabla_{\perp}^{2}\bar{\psi} \\ = & v_{A}\rho_{s}\frac{1}{r}\frac{\partial\tilde{\psi}}{\partial\theta}\frac{4\pi\rho_{s}}{cB_{0}}j_{0}^{'} + \frac{v_{th}\rho_{s}}{r}\left\{ \tilde{\psi},\rho_{s}^{2}\nabla_{\perp}^{2}\tilde{\psi}\right\} +\frac{v_{th}\rho_{s}}{r}\left\{ \tilde{W},\tilde{\phi}\right\} \\ & +\frac{\rho_{s}^{2}W_{0}^{'}}{r}\frac{\partial\tilde{\phi}}{\partial\theta} +\nu\nabla_{\perp}^{2}\tilde{W} \\ \end{aligned}$$ $$\label{Eq.2} \frac{\partial\tilde{\psi}}{\partial t}+\mathrm{\mathbf{v_{0}}}\cdot\nabla\tilde{\psi}+v_{A}\nabla_{\parallel}\tilde{\phi}=\frac{v_{th}\rho_{s}}{r}\left\{ \tilde{\psi},\tilde{\phi}\right\} +\frac{c^{2}\eta}{4\pi}\nabla_{\perp}^{2}\tilde{\psi}$$ where, $$\tilde{W}=\rho_{s}^{2}\nabla\cdot\left(\frac{n_{0}\left(\rho\right)}{n_{0}\left(0\right)}\nabla_{\perp}\tilde{\phi}\right)$$ Equation\[\[Eq.1\]\] is the vorticity equation, where $\tilde{W}$ is the perturbed vorticity. Equation\[\[Eq.2\]\] describes the evolution of the perturbed poloidal flux function $\tilde{\psi}$. The resistivity $\eta$ and viscosity $\nu$ are specified quantities and are held constant during our calculations. Additionally, $\rho_{s}=\frac{v_{th}}{\omega_{ci}}$, where, $v_{th}^{2}=\left(T_{0i}+T_{0e}\right)/m_{i}$, $\omega_{ci}=\left(eB_{0}/m_{i}c\right)$, with $T_{0i},T_{0e}$ being ion and electron temperatures respectively. $m_{i}$ is the ion mass, $e$ is the elemenatary charge. $\Phi_{0}(r), \Psi_{0}(r)$ denote the mean electrostatic and magnetostatic potentials respectively. Also, we have used fixed boundary conditions, along with a conducting boundary. This comes from $$\frac{\delta\mathbf{E}}{B_{0}}=-\nabla\tilde{\phi}-\frac{1}{c}\frac{\partial\tilde{\phi}}{\partial t}\mathbf{e_{\zeta}}$$ where, $\mathbf{E}$ is the electric field, $\phi$ is the electrostatic potential, and $\mathbf{B_{0}\simeq} B_{0z}{\bf e}_{\zeta}+B_{0\theta}(\rho){\bf e}_{\theta}$ is the equilibrium field. The fluctuating electric field, $\delta E$, is related to $\tilde{\phi}$ \[this has dimensions of length\]. We set, $\epsilon=a/R_{0}$, the inverse aspect ratio, $v_{0}=V_{0z}\left(\rho\right)\mathbf{e}_{\zeta}+a\rho\Omega\left(\rho\right)\mathbf{e}_{\theta}$. The equilibrium axial and poloidal, sub-Alfvénic sheared flows are: $M_{z}=V_{0z}/v_{A}$ is the Axial Mach number; $M_{\theta}=\rho\Omega\left(\rho\right)\tau_{A}$ is the poloidal Mach number; $\tau_{A}=a/v_{A}$ the Alfvén time; $\tau_{\eta}=(4\pi a^{2}/c^{2}\eta)$ the resistive time; $\tau_{\nu}=(a^{2}/\nu)$ the viscous time. We will use in the following the Lundquist Number, $S=\frac{\tau_{\eta}}{\tau_{A}}$, and the Prandtl Number, $Pr=\frac{\tau_{\eta}}{\tau_{\nu}}$. The velocity perturbations are non-dimensionalised relative to the Alfven speed, $v_{A}=\frac{B_{0}}{\left(4\pi m_{i}n_{0}\right)^{1/2}}$. The magnetic field perturbations are normalised by the equilibrium axial magnetic field $B_{0z}$. The fluctuations of magnetic field and velocity are incompressible in the $(r- \theta)$ plane. Together, these equations constitute the V-RMHD(visco-resistive MHD) model and we solve them using the CUTIE (CUlham Transporter of Ions and Electrons) code [@Thyagaraja2000; @Chandra2015]. CUTIE is a nonlinear, global, electromagnetic, quasi-neutral, two fluid initial value code. It has been used earlier for studies of tearing modes, ELMs, L to H transitions, internal transport barriers and other problems [@Thyagaraja2000; @Thyagaraja2010; @Chandra2015; @Chandra2017]. Linear Results {#linear results} ============== In this section we describe the results of our linear studies carried out for a q profile of the following form: $$\label{Eq.3} q(\rho)=q_{0}\left(1+\left(\frac{\rho}{\rho_{0}}\right)^{2\Lambda}\right)^{\frac{1}{\Lambda}}$$ with the safety factor, $q(\rho)=\frac{\epsilon \rho B_{0z}}{B_{0\theta}(\rho)}$, $q_{0}=0.9$, $\Lambda=1$, $\rho_{0}=0.6a$, $a=$ radius of the cylinder. Fig. \[fig:q\_profile\] shows the q profile used in the simulations and the q=1 surface. ![q profile[]{data-label="fig:q_profile"}](q-profile.pdf) Scaling with S and Pr --------------------- At first, we have studied the scaling of the growth rate of the (1,1) mode with S(Lundquist number) and Pr(Prandtl Number). For most of our simulations we have used a flat $\eta$ profile but we get similar results when we use a self-consistent $\eta$ such that $E_{0z}=\eta j_{0z}; E_{0z}=V/(2\pi R_{0})$, where $V$ is the constant loop-voltage. In Fig. \[fig:res\_nf\_high\], we have plotted the normalised growth rate $\gamma\tau_{A}$ with $S$, at a fixed $Pr$ of 0.1. We have found a scaling of the variation of $\gamma\tau_{A}$ with $S$ to be of the form $S^{-1/3}$ . These results are similar to those obtained by Porcelli[@Porcelli1987] for the resistive internal kink mode. At low $S$, we notice a deviation from the scaling that can be attributed to local asymmetries of the equilibrium current density[@Militello2004a]. ![Linear Resistivity scaling without flow for the m=1, n=1 mode at $Pr=0.1$[]{data-label="fig:res_nf_high"}](no_flow_flat_res_scaling.pdf) In Fig. \[fig:vis\_nf\_l\], we observe a $Pr^{-1/3}$ scaling of $\gamma\tau_{A}$ as we vary $Pr$ by keeping $S$ fixed. This scaling agrees with that reported earlier by Porcelli[@Porcelli1987]. However, as we increase the viscosity further, the growth rate scaling changes to $Pr^{-5/6}$, which is a new result that has not been reported earlier. It shows that high viscosity can strongly influence the linear growth rate of the modes. These results can be qualitatively understood by a standard dominant balance analysis of the dynamical equations of the mode in the inner layer. From the set of model equations (\[Eq.1\]) and (\[Eq.2\]) one can obtain the following set of linear inner layer equations:\ $$\begin{aligned} (\gamma +iv_{0,res}^{\prime}x)\frac{d^2 \phi}{dx^2} +i\frac{q^{\prime}_{res}}{q_{res}}x \frac{d^2\psi}{dx^2} &=& \nu \frac{d^4 \phi}{dx^4} \label{vort}\\ (\gamma +iv_{0,res}^{\prime}x)\psi + i\frac{q^{\prime}_{res}}{q_{res}}x \phi &=& \eta \frac{d^2 \psi}{dx^2} \label{Ohm}\end{aligned}$$ where all quantities are suitably made non-dimensional and where $\nu$, $\eta$ are non-dimensional viscous and resistive diffusivities respectively. $v_{0,res}^{\prime}$ and $q^{\prime}_{res}$ are the derivatives of the flow terms and q profile respectively at the resonant surface and $\gamma$ is the normalized growth rate. In the absence of flow and in the regime where both the viscous and resistive contributions are important, the term proportional to $\nu$ with the highest derivative dominates over the term proportional to $\gamma$ in (\[vort\]), while in (\[Ohm\]) all terms contribute equally. Using the dominant balance argument one then gets, $$\gamma \sim \eta^{2/3}\nu^{-1/3} \sim Pr^{-1/3}$$ and the layer width goes as $$x \sim \eta^{1/6}\nu^{1/6}$$ This agrees with the numerical scaling obtained in Fig. 3 for moderate values of Pr and is also in accordance with the scaling discussed by Porcelli [@Porcelli1987]. For higher values of viscosity, when viscous effects dominate over resistive contributions, the term on the R.H.S. of (\[Ohm\]) may be ignored in the dominant balance calculation. In this limit the layer width also has a very weak dependence on viscosity and can be nearly taken to be a constant. The balance arguments then lead to $ \gamma \nu \sim x^4$ and hence $\gamma \sim \nu^{-1}$. This scaling is close to the $\gamma \sim Pr^{-5/6}$ obtained from our numerical solutions. Such a scaling has also been alluded to by Porcelli [@Porcelli1987] for the so-called visco-ideal limit. The growth rate becomes nearly constant in the low Pr regime, as we would expect the plasma to be nearly inviscid. We next consider the effect of flows on the linear growth rates. ![Linear Viscosity scaling without flow for the m=1, n=1 mode at $S=10^6$[]{data-label="fig:vis_nf_l"}](no_flow_flat_vis_scaling.pdf) ### Axial Flow {#linear axial} We next present scaling results in the presence of a sheared axial flow. We have used an axial flow profile of the form: $$\frac{V_{0z}}{v_{A}}=M_{z}\tanh[\lambda(\rho-\rho_{res})]$$ where, $V_{0z}$ is the equilibrium axial flow, $M_{z}$ is the axial Mach number, $\lambda$ is the shear parameter and $\rho_{res}$ is the location of the mode resonant surface. The flow profile is shown in Fig. \[axial\_linear\_profile\] ![Axial flow profile(tanh profile)[]{data-label="axial_linear_profile"}](axial_flow_profile.pdf) This profile has the property that it has zero flow and non-zero shear at the resonant surface and hence is a useful profile to study the effect of shear on the mode. It has been used in the past to understand the effect of flow shear on tearing modes[@Ofman1991; @Chandra2015]. For our linear scaling studies we have taken several different values $M_{z}=0.05$ that are within a physically reasonable range of values for experimental observations. In general, the presence of an axial sheared flow has a destabilizing influence on the $m=1$ resistive kink mode, as has been noted before [@Gimblett1996] and is due to the additional ideal free energy arising from the nature of the flow profile. A principal consequence of this is an increase in the growth rate of the kink mode compared to the no flow case. This is clearly seen in Fig. \[res\_ax\_flow\_lin\] where we have marked the values of the growth rate for the no flow case and for several different finite values of the axial flow (and correspondingly different velocity shears) in a single plot. It is also seen that there is a near independence of the growth rate on $S$ for higher values of $S$. This can be physically understood as follows: as the resistivity decreases (S increases) the growth rate of the classical resistive kink mode decreases and the growth is dominated by the ideal driving term of the flow shear. This term is independent of $S$ and hence at higher values of $S$ the growth rate becomes independent of $S$. In Fig. (\[res\_ax\_flow\_lin\]) we thus see how an increase in $M_{z}$ progressively changes the $S^{-1/3}$ scaling in the “no-flow” case to one independent of $S$. ![Linear Resistivity scaling with axial flow for the m=1, n=1 mode at $Pr=0.1$ with $M_{z}=(0.0,0.01,0.025,0.05)$[]{data-label="res_ax_flow_lin"}](axial_flow_scaling.pdf) We have similarly observed a change in the viscosity scaling due to the presence of axial flow and this is shown in Fig. \[vis\_ax\_flow\_lin\]. As we go from $Pr =1$ to $Pr =10 $, the scaling of the linear growth rate gradually changes from $Pr^{-1/5}$ to $Pr^{-3/5}$ and beyond $Pr =10$, the growth rate becomes negative. If we compare it with the no flow case, there the scaling goes as $Pr^{-1/3}$ up to $Pr=10$, beyond which it changes to $Pr^{-5/6}$. Thus in the presence of an axial flow, the stabilising influence of viscosity is enhanced and can lead to complete stabilisation of the $m=1$ visco-resistive mode at high $Pr$ numbers. ![Linear Viscosity scaling with axial flow for the m=1, n=1 mode at $S=10^6$, $M_{z}=0.05$[]{data-label="vis_ax_flow_lin"}](axial_flow_flat_vis_scaling.pdf) ### Poloidal flow For our poloidal flow studies, we have used the following flow profile, $$\frac{V_{0\theta}}{v_{A}}=M_{\theta}(\rho)$$ where, $$M_{\theta}(\rho)=\Omega\tau_{A}\;\rho\left(1+k\rho\right)$$ Here, $V_{0\theta}$ is the equilibrium poloidal flow, and $\Omega$ is the poloidal angular frequency and $k$ measures the shear in the flow. The profile is plotted in Fig. \[poloidal\_profile\] ![Poloidal flow profile[]{data-label="poloidal_profile"}](poloidal_flow_profile.pdf) We have obtained the growth rates for several different values of $M_{\theta}$ and the results are plotted in Fig. ( \[rmhd\_pol\_resis\_scal\]). In contrast to the sheared axial flow the sheared poloidal flow has a stabilizing influence on the resistive kink mode so that it decreases the value of the growth rate. This stabilizing influence is independent of $S$ and hence the net effect is a shift in the value of the growth rate without a change in the scaling dependence on $S$ which remains the same as the no-flow case, namely $\gamma\tau_{A}\propto S^{-1/3}$. As seen in Fig. (\[rmhd\_pol\_resis\_scal\]) as the poloidal Mach number is increased from zero, the scaling curve shifts downwards without changing shape. For very low values of $M_{\theta}$, the curve almost coincides with the no-flow case, as expected. ![Linear Resistivity scaling with poloidal flow for the m=1,n=1 mode at $Pr=0.1$, $M_{\theta}=(0.0,0.0018,0.009,0.018)$[]{data-label="rmhd_pol_resis_scal"}](poloidal_flow_scaling.pdf) In Fig. \[rmhd\_pol\_vis\_scal\] we have displayed the scaling of the growth rate with viscosity for a fixed value of the resistivity. We find that the scaling is similar to the no flow case, namely $\gamma\tau_{A} \propto Pr^{-1/3}$ at a lower viscosity and $\gamma\tau_{A} \propto Pr^{-5/6}$ at a higher viscosity. ![Linear Viscosity scaling with poloidal flow for the m=1,n=1 mode at $S=10^6$, $M_{\theta}=0.009$[]{data-label="rmhd_pol_vis_scal"}](pol_flow_flat_vis_scaling.pdf) Effect of flows in different viscosity regimes ---------------------------------------------- In Fig. \[high\_vis\_flow\_comparison\], we compare the linear growth rate changes of the $m=1, n=1$ mode with axial and poloidal flows as we go from low to high viscosity. We note that the nature of stabilisation for axial flows changes as we increase the viscosity. While axial flows [destabilise]{} the mode at low viscosity, they [stabilise]{} it at higher viscosities. On the other hand, we find the poloidal flow to be always stabilising in contrast to the no-flow case. ![Effect of viscosity on the linear growth rate of the m=1, n=1 mode with and without axial flow at $S=10^6$[]{data-label="high_vis_flow_comparison"}](viscosity_effect_flow.pdf) In Fig. \[axial\_flow\_diff\_Pr\], we have shown how the linear growth rate of the $m=1, n=1$ mode changes as we go from low to high axial flow shear, $\frac{a}{V_{a}}\frac{dv_{0z}}{dr}$, for different viscosity regimes. We see that for a fixed Pr, the nature of stabilisation of flow does not change with the amount of flow shear. However, the nature of stabilisation changes depending on the viscosity regime irrespective of the amount of flow shear. ![Linear growth rate of (1,1) mode v/s axial flow shear for different Pr values and $S=10^6$[]{data-label="axial_flow_diff_Pr"}](ax_flow_vs_viscosity.pdf) Nonlinear Results {#nonlinear results} ================= Next we report on our nonlinear results for the (1,1) mode in the absence and presence of flows. We have continued with the same parameters and related profiles that we have used for the linear runs. Here, we have a slight difference in the method of calculation as compared to the linear case. We set up an equilibrium from the given initial parameters and in every iteration we solve both the mean, i.e., (0,0) Fourier components and the perturbed components. As a result, the equilibrium evolves with time, while in the linear case, the equilibrium was held fixed. Axial flow {#axial-flow} ---------- In this section, we describe the nonlinear evolution of (1,1) modes in the presence of axial flows. Here, two different flow profiles are employed to elucidate the dependence of the results on the profile. At first, to understand the effect of flow shear we have used a tanh flow profile. The form of the tanh profile has been described in section \[linear axial\], but here we have used $M_{z}=0.01$ Next, we use a Gaussian flow profile which is more realistic from an experimental point of view. This profile has the form (illustrated in Fig. \[axial\_profile\_gaussian\]) : $$V_{0z}/V_{A}=M_{z}e^{-\rho^{2}}$$ where, $\rho_{res}$ is the location of the mode resonant surface and $M_{z}=0.05$. ![Axial flow profile(Gaussian profile)[]{data-label="axial_profile_gaussian"}](ax_flow_gauss_profile.pdf) The Figs. \[fig:tanh\_nl\_12\] and \[tanh\_nl\_pr30\] illustrate the time evolution of $\tilde{\|\psi\|}_{max}$ with a tanh flow profile for $Pr=100$ and $Pr=30$ respectively. For the high viscosity case, we notice a strong stabilisation of the (1,1) mode in the presence of axial flow both in the linear growth rate as well as in the nonlinear saturation level. However, for the low viscosity case, there is a slight increase of nonlinear saturation level of the modes in the presence of axial flow. Similar to the linear runs, the nonlinear evolution runs also show the destabilising trend of the mode for lower viscosity and a stabilising influence for higher viscosity compared to the no flow case. ![$\tilde{\|\psi\|}_{max}$ evolution with axial flow with tanh profile, $M_{z}=0.01$, $Pr=100$ and $S=10^6$.[]{data-label="fig:tanh_nl_12"}](axial_flow_nl_tanh_pr_100.pdf) ![$\tilde{\|\psi\|}_{max}$ evolution with axial flow with tanh profile, $M_{z}=0.01$, $Pr=30$ and $S=10^6$.[]{data-label="tanh_nl_pr30"}](axial_flow_nl_tanh_pr_30.pdf) Figs. \[gau\] and \[fig:gauss\_nl\_pr30\] show the nonlinear evolution of the mode with a Gaussian flow profile for $Pr=100$ and $Pr=30$ respectively. In this case, the nature of the effects is qualitatively similar to that of the tanh flow case. However, the changes in the growth rates compared to the no flow case are smaller even if the amount of flow is higher in this case. We have seen that even if the linear evolution does not depend on the sign of the flow, the nonlinear saturation levels of $\tilde{\|\psi\|}_{max}$ are different for different signs of flows except for the $Pr=100$ tanh flow profile case(cf. Fig. \[fig:tanh\_nl\_12\]), where the difference is very small. ![$\tilde{\|\psi\|}_{max}$ evolution with axial flow with gaussian profile, $M_{z}=0.05$, $Pr=100$ and $S=10^6$.[]{data-label="gau"}](axial_flow_nl_gauss_pr_100.pdf) ![$\tilde{\|\psi\|}_{max}$ evolution with axial flow with gaussian profile, $M_{z}=0.05$, $Pr=30$ and $S=10^6$[]{data-label="fig:gauss_nl_pr30"}](axial_flow_nl_gauss_pr_30.pdf) Poloidal Flow ------------- In Fig. \[pol\_flow\_nl\], we display the effects of poloidal flow upon the nonlinear evolution of the amplitude of the (1,1) mode. Here, we notice that the poloidal flow stabilises the mode, and the final saturation levels are nearly equal for such small amounts of flow. For higher values of $M_{\theta}$ the saturation levels do differ significantly as a function of the direction of flow. ![$\tilde{\|\psi\|}_{max}$ evolution with poloidal flow, $M_{\theta}=0.0018$ $Pr=100$ and $S=10^6$.[]{data-label="pol_flow_nl"}](pol_flow_nl_pr_100.pdf) We have repeated this study with a $Pr=30$ in Fig. \[pol\_flow\_nl\_pr30\], and we notice here that the nonlinear saturated levels show a different behaviour from that at a higher $Pr$, in that poloidal flow now slightly destabilises the mode. The implication of this effect is clearly reflected for helical flows which we will discuss next. ![$\tilde{\|\psi\|}_{max}$ evolution with poloidal flow, $M_{\theta}=0.0018$, $Pr=30$ and $S=10^6$.[]{data-label="pol_flow_nl_pr30"}](pol_flow_nl_pr_30.pdf) Helical Flow ------------ In this subsection we discuss the combined effect of axial and poloidal flows on the stability of the (1,1) mode. We have considered all four sign combinations of the axial and poloidal flows to understand the effect of flow helicity on the evolution of the mode. In Fig. \[hel\_flow\_tanh\_nl\], we show the effect of a sheared axial flow with a tanh profile combined with a sheared poloidal flow for $Pr=100$. Here we can have two types of flow helicity depending on the signs of the axial flow and the poloidal flow. We find that although both the flow helicity cases impart a stabilising effect compared to the no flow case, the degree of stabilisation is very different for different flow helicities. For example, having kept the poloidal flow sign to be positive but changing the direction of the axial flow from positive to negative, both the linear growth rates and the nonlinear saturation levels have increased significantly to a much higher value. Thus, we find an asymmetry in the nature of the stabilisation of the (1,1) mode in the presence of helical flows that depends on the type of flow helicity. The change in the degree of stabilization for different helicities arises from the relationship between the flow direction and the direction of the magnetic field which essentially changes the relative sign between $q_{res}^{\prime}$ (the magnetic shear) and $v_{0,res}^{\prime}$ (the flow shear) near the mode resonant surface [@Haye2009]. ![$\tilde{\|\psi\|}_{max}$ evolution with helical flow using tanh profile, $M_{z}=0.01$, $M_{\theta}=0.0018$, $Pr=100$ and $S=10^6$.[]{data-label="hel_flow_tanh_nl"}](helical_flow_nl_tanh_pr_100.pdf) We have carried out a similar study at a lower viscosity of $Pr=30$ as shown in Fig. \[hel\_flow\_tanh\_nl\_pr30\]. Here, the nonlinear saturation levels in all flow cases are slightly higher compared to the no flow case. Also, the symmetry breaking for two different flow helicities are so small both in the linear and nonlinear regime that it cannot be distinguished from the figure, but can be distinguished from numerical values of the linear growth rates and nonlinear saturation levels. In fact, the symmetry breaking effect begins to manifest itself in the linear stage itself as can be clearly seen in the difference of the slopes of the time evolution of $\mid \psi \mid_{max}$ for the two helicities of the flow. A comparison of Figs. (\[hel\_flow\_tanh\_nl\]) and (\[hel\_flow\_tanh\_nl\_pr30\]) also raises the interesting question whether there occurs a “bifurcation” of the saturated states at some value of the Prandtl number between 30 and 100. To check this interesting question we have numerically determined the linear growth rates for a number of different magnitudes of the helical flow and plotted their values for two different helicities in Fig. (\[gr\_heli\_Pr\]). As can be clearly seen there is a continuous transition in the behaviour as a function of $Pr$ that is indicative of an absence of any bifurcation phenomenon. ![$\tilde{\|\psi\|}_{max}$ evolution with helical flow using tanh profile, $M_{z}=0.01$, $M_{\theta}=0.0018$, $Pr=30$ and $S=10^6$.[]{data-label="hel_flow_tanh_nl_pr30"}](helical_flow_nl_tanh_pr_30.pdf) ![$\tilde{\|\psi\|}_{max}$ Linear growth rates with helical flows for different Prandtl numbers using tanh profile and $S=10^6$.[]{data-label="gr_heli_Pr"}](growth_rate_vs_prandtl_number.pdf) In Fig. \[hel\_flow\_gauss\_nl\] and Fig. \[hel\_flow\_gauss\_nl\_pr30\], we have shown the effect of a sheared axial flow with a Gaussian profile combined with a sheared poloidal flow for $Pr=100$ and $Pr=30$ respectively. Here, we notice that the effects are very similar to the tanh flow case, but the changes in the linear growth rates and nonlinear saturation levels are relatively smaller. ![$\tilde{\|\psi\|}_{max}$ evolution with Helical flow using gaussian profile, $M_{z}=0.05$,$M_{\theta}=0.0018$, $Pr=100$ and $S=10^6$.[]{data-label="hel_flow_gauss_nl"}](helical_flow_nl_gauss_pr_100.pdf) On comparison of the helical flow results obtained here with those using pure axial and poloidal flows, as discussed in the previous sections, there is no symmetry breaking in the linear growth rates of the (1,1) mode if we change the direction of the flow. However, there is a difference in the nonlinear saturation levels even for those cases where we use a pure axial or poloidal flow. This is due to the self-generation of nonlinear helical terms even if we start with pure flows, as discussed in Chandra et. al. [@Chandra2015]. ![$\tilde{\|\psi\|}_{max}$ evolution with helical flow using gaussian profile, $M_{z}=0.05$,$M_{\theta}=0.0018$, $Pr=30$ and $S=10^6$.[]{data-label="hel_flow_gauss_nl_pr30"}](helical_flow_nl_gauss_pr_30.pdf) Summary and Discussion {#discussion} ====================== To summarise, we have carried out linear and nonlinear studies of the (1,1) resistive internal kink mode using a V-RMHD version of the CUTIE code, in a cylindrical geometry with periodic boundary conditions. We have studied the effect of equilibrium sheared flows on the (1,1) mode and the role of viscosity in modifying the effect of flows. Viscosity can be significantly enhanced due to turbulence in tokamaks and it is expected that the $Prandtl$ $number$ can be as high as 100 [@Maget12; @Takeda2008] in advanced scenarios for JET and ITER. Our results can be summarised as follows. In the linear regime our scaling studies in the absence of flow agree with analytical results in the literature[@Porcelli1987]. The presence of poloidal flow does not change the linear scaling results but axial flows do bring about a significant change. We further find that the effect of viscosity on the growth rate of the mode can be significantly altered by the presence of flows. Helical flows exhibit a strong symmetry breaking with respect to the direction of the flow at high Pr but such an effect weakens at low Pr. In the nonlinear regime, for axial flows the saturation level of the mode decreases at a higher viscosity compared to the case of no flow but slightly increases at lower viscosity. Similar results are found for the poloidal flow case. In the case of helical flows at high viscosity, there is a significant change in the nonlinear saturation level depending on the flow helicity. Such an asymmetry effect is much weaker in the low viscosity case. It might be worth mentioning that similar asymmetric effects in the sawteeth time period have been observed in tokamak experiments with a change in the direction of the equilibrium flow induced by neutral beam injections[@Chapman2007a; @Nave2006; @Chapman2006a; @Chapman2008c]. Our results can prove useful in developing appropriate theoretical models for sawteeth behaviour in the presence of sheared flows and high viscosity.	{ "pile_set_name": "ArXiv" }
--- abstract: \| We study the problem of Byzantine-robust topology discovery in an arbitrary asynchronous network. We formally state the weak and strong versions of the problem. The weak version requires that either each node discovers the topology of the network or at least one node detects the presence of a faulty node. The strong version requires that each node discovers the topology regardless of faults. We focus on non-cryptographic solutions to these problems. We explore their bounds. We prove that the weak topology discovery problem is solvable only if the connectivity of the network exceeds the number of faults in the system. Similarly, we show that the strong version of the problem is solvable only if the network connectivity is more than twice the number of faults. We present solutions to both versions of the problem. The presented algorithms match the established graph connectivity bounds. The algorithms do not require the individual nodes to know either the diameter or the size of the network. The message complexity of both programs is low polynomial with respect to the network size. We describe how our solutions can be extended to add the property of termination, handle topology changes and perform neighborhood discovery. author: - 'Mikhail Nesterenko[^1]' - 'Sébastien Tixeuil[^2]' bibliography: - 'topology.bib' title: \| Discovering Network Topology in\ the Presence of Byzantine Faults[^3] --- Introduction ============ In this paper, we investigate the problem of Byzantine-tolerant distributed topology discovery in an arbitrary network. Each node is only aware of its neighboring peers and it needs to learn the topology of the entire network. Topology discovery is an essential problem in distributed computing (e.g. see [@SG89]). It has direct applicability in practical systems. For example, link-state based routing protocols such as OSPF use topology discovery mechanisms to compute the routing tables. Recently, the problem has come to the fore with the introduction of ad hoc wireless sensor networks, such as Berkeley motes [@Hill:2002:MWP], where topology discovery is indispensable for routing decisions. As reliability demands on distributed systems increase, the interest in developing robust topology discovery programs grows. One of the strongest fault models is Byzantine [@LSP82]: the faulty node behaves arbitrarily. This model encompasses rich set of fault scenarios. Moreover, Byzantine fault tolerance has security implications, as the behavior of an intruder can be modeled as Byzantine. One approach to deal with Byzantine faults is by enabling the nodes to use cryptographic operations such as digital signatures or certificates. This limits the power of a Byzantine node as a non-faulty node can verify the validity of received topology information and authenticate the sender across multiple hops. However, this option may not be available. For example, wireless sensors may not have the capacity to manipulate digital signatures. Another way to limit the power of a Byzantine process is to assume synchrony: all processes proceed in lock-step. Indeed, if a process is required to send a message with each pulse, a Byzantine process cannot refuse to send a message without being detected. However, the synchrony assumption may be too restrictive for practical systems. \ Our contribution. In this study we explore the fundamental properties of topology discovery. We select the weakest practical programming model, establish the limits on the solutions and present the programs matching those limits. Specifically, we consider arbitrary networks of arbitrary topology where up to fixed number of nodes $k$ is faulty. The execution model is asynchronous. We are interested in solutions that do not use cryptographic primitives. The solutions should be terminating and the individual processes should not be aware of the network parameters such as network diameter or its total number of nodes. We state two variants of the topology discovery problem: weak and strong. In the former — either each non-faulty node learns the topology of the network or one of them detects a fault; in the latter — each non-faulty node has to learn the topology of the network regardless of the presence of faults. As negative results we show that any solution to the weak topology discovery problem can not ascertain the presence of an edge between two faulty nodes. Similarly, any solution to the strong variant can not determine the presence of a edge between a pair of nodes at least one of which is faulty. Moreover, the solution to the weak variant requires the network to be at least $(k+1)$-connected. In case of the strong variant the network must be at least $(2k+1)$-connected. The main contribution of this study are the algorithms that solve the two problems: Detector and Explorer. The algorithms match the respective connectivity lower bounds. To the best of our knowledge, these are the first asynchronous Byzantine-robust solutions to the topology discovery problem that do not use cryptographic operations. Explorer solves the stronger problem. However, Detector has better message complexity. Detector either determines topology or signals fault in $O(\delta n^3)$ messages where $\delta$ and $n$ are the maximum neighborhood size and the number of nodes in the system respectively. Explorer finishes in $O(n^4)$ messages. We extend our algorithms to (a) terminate (b) handle topology changes (c) discover neighbors if ports are known (d) discover a fixed number of routes instead of complete topology and (e) reliably propagate arbitrary information instead of topological data. \ Related work. A number of researchers employ cryptographic operations to counter Byzantine faults. Avromopolus et al [@AKWK04] consider the problem of secure routing. Therein see the references to other secure routing solutions that rely on cryptography. Perrig et al [@Perrig:2004:SWS] survey robust routing methods in ad hoc sensor networks. The techniques covered there also assume that the processes are capable of cryptographic operations. A naive approach of solving the topology discovery problem without cryptography would be to use a Byzantine-resilient broadcast [@BV05; @D82; @K04; @PP05]: each node advertises its neighborhood. However all existing solutions for arbitrary topology known to us require that the graph topology is a priori known to the nodes. Let us survey the non-cryptography based approaches to Byzantine fault-tolerance. Most programs described in the literature [@AW98; @MRRS01; @MMR03; @MS03] assume completely connected networks and can not be easily extended to deal with arbitrary topology. Dolev [@D82] considers Byzantine agreement on arbitrary graphs. He states that for agreement in the presence of up to $k$ Byzantine nodes, it is necessary and sufficient that the network is $(2k+1)$-connected and the number of nodes in the system is at least $3k+1$. However, his solution requires that the nodes are aware of the topology in advance. Also, this solution assumes the synchronous execution model. Recently, the problem of Byzantine-robust reliable broadcast has attracted attention [@BV05; @K04; @PP05]. However, in all cases the topology is assumed to be known. Bhandari and Vaidya [@BV05] and Koo [@K04] assume two-dimensional grid. Pelc and Peleg [@PP05] consider arbitrary topology but assume that each node knows the exact topology a priori. A notable class of algorithms tolerates Byzantine faults locally [@MT05r; @NA02; @SOM04]. Yet, the emphasis of these algorithms is on containing the fault as close to its source as possible. This is only applicable to the problems where the information from remote nodes is unimportant such as vertex coloring, link coloring or dining philosophers. Thus, local containment approach is not applicable to topology discovery. Masuzawa [@M95] considers the problem of topology discovery and update. However, Masuzawa is interested in designing a self-stabilizing solution to the problem and thus his fault model is not as general as Byzantine: he considers only transient and crash faults. \ The rest of the paper is organized as follows. After stating our programming model and notation in Section \[SecNotation\], we formulate the topology discovery problems, as well as state the impossibility results in Section \[SecBounds\]. We present Detector and Explorer in Sections \[SecDetector\] and \[SecExplorer\] respectively. We discuss the composition of our programs and their extensions in Section \[SecExtensions\] and conclude the paper in Section \[SecEnd\]. Notation, Definitions and Assumptions {#SecNotation} ===================================== Graphs. A distributed system (or program) consists of a set of processes and a neighbor relation between them. This relation is the system topology. The topology forms a graph $G$. Denote $n$ and $e$ to be the number of nodes[^4] and edges in $G$ respectively. Two processes are neighbors if there is an edge in $G$ connecting them. A set $P$ of neighbors of process $p$ is neighborhood of $p$. In the sequel we use small letters to denote singleton variables and capital letters to denote sets. In particular, we use a small letter for a process and a matching capital one for this process’ neighborhood. Since the topology is symmetric, if $q \in P$ then $p \in Q$. Denote $\delta$ to be the maximum number of nodes in a neighborhood. A node-cut of a graph is the set of nodes $U$ such that $G \setminus U$ is disconnected or trivial. A node-connectivity (or just connectivity) of a graph is the minimum cardinality of a node-cut of this graph. In this paper we make use of the following fact about graph connectivity that follows from Menger’s theorem (see [@YG98]): if a graph is $k$-connected (where $k$ is some constant) then for every two vertices $u$ and $v$ there exists at least $k$ internally node-disjoint paths connecting $u$ and $v$ in this graph. \ Program model. A process contains a set of variables. When it is clear from the context, we refer to a variable $var$ of process $p$ as $var.p$. Every variable ranges over a fixed domain of values. For each variable, certain values are initial. Each pair of neighbor processes share a pair of special variables called channels. We denote $Ch.b.c$ the channel from process $b$ to process $c$. Process $b$ is the sender and $c$ is the receiver. The value for a channel variable is chosen from the domain of (potentially infinite) sequences of messages. A state of the program is the assignment of a value to every variable of each process from its corresponding domain. A state is initial if every variable has initial value. Each process contains a set of actions. An action has the form $\langle name \rangle : \langle guard \rangle \longrightarrow \langle command \rangle$. A guard is a boolean predicate over the variables of the process. A command is sequence of assignment and branching statements. A guard may be a receive-statement that accesses the incoming channel. A command may contain a send-statement that modifies the outgoing channel. A parameter is used to define a set of actions as one parameterized action. For example, let $j$ be a parameter ranging over values 2, 5 and 9; then a parameterized action $ac.j$ defines the set of actions $ac.(j=2)\ \MYBOX\ ac.(j=5)\ \MYBOX\ ac.(j=9)$. Either guard or command can contain quantified constructs [@dijkstra-scholten:1990a] of the form: $(\langle quantifier \rangle \langle bound\ variables \rangle : \langle range \rangle : \langle term \rangle)$, where range and term are boolean constructs. \ Semantics. An action of a process of the program is enabled in a certain state if its guard evaluates to true. An action containing receive-statement is enabled when appropriate message is at the head of the incoming channel. The execution of the command of an action updates variables of the process. The execution of an action containing receive-statement removes the received message from the head of the incoming channel and inserts the value the message contains into the specified variables. The execution of send-statement appends the specified message to the tail of the outgoing message. A computation of the program is a maximal fair sequence of states of the program such that the first state $s_0$ is initial and for each state $s_i$ the state $s_{i+1}$ is obtained by executing the command of an action whose state is enabled in $s_i$. That is, we assume that the action execution is atomic. The maximality of a computation means that the computation is either infinite or it terminates in a state where none of the actions are enabled. The fairness means that if an action is enabled in all but finitely many states of an infinite computation then this action is executed infinitely often. That is, we assume weak fairness of action execution. Notice that we define the receive statement to appear as a standalone guard of an action. This means, that if a message of the appropriate type is at the head of the incoming channel, the receive action is enabled. Due to weak fairness assumption, this leads to fair message receipt assumption: each message in the channel is eventually received. Observe that our definition of a computation considers asynchronous computations. To reason about program behavior we define boolean predicates on program states. A program invariant is a predicate that is true in every initial state of the program and if the predicate holds before the execution of the program action, it also holds afterwards. Notice that by this definition a program invariant holds in each state of every program computation. \ Faults. Throughout a computation, a process may be either Byzantine (faulty) or non-faulty. A Byzantine process contains an action that assigns to each local variable an arbitrary value from its domain. This action is always enabled. Yet, the weak fairness assumption does not apply to this action. That is, we consider computations where a faulty process does not execute any actions. Observe that we allow a faulty node to send arbitrary messages. We assume, however, that messages sent by such a node conform to the format specified by the algorithm: each message carries the specified number of values, and the values are drawn from appropriate domains. This assumption is not difficult to implement as message syntax checking logic can be incorporated in receive-action of each process. We assume oral record [@LSP82] of message transmission: the receiver can always correctly identify the message sender. The channels are reliable: the messages are delivered in FIFO order and without loss or corruption. Throughout the paper we assume that the maximum number of faulty nodes in the system is bounded by some constant $k$. \ Graph exploration. The processes discover the topology of the system by exchanging messages. Each message contains the identifier of the process and its neighborhood. Process $p$ explored process $q$ if $p$ received a message with $(q,Q)$. When it is clear from the context, we omit the mention of $p$. An explored subgraph of a graph contains only explored processes. A Byzantine process may potentially circulate information about the processes that do not exist in the system altogether. A process is fake if it does not exist in the system, a process is real otherwise. The Topology Discovery Problem: Statement and Solution Bounds {#SecBounds} ============================================================= Problem statement. A program is a solution to the weak topology discovery problem if each of the program’s computation satisfies the following properties: termination* — either all non-faulty processes determine the system topology or at least one process detects a fault; safety — for each non-faulty process, the determined topology is a subset of the actual system topology; validity — the fault is detected only if there are faulty processes in the system.* A program is a solution to the strong topology discovery problem if each of the program’s computations satisfies the following properties: termination* — all non-faulty processes determine the system topology; safety — the determined topology is a subset of the actual system topology.* According to the safety property of both problem definitions each non-faulty process is only required to discover a subset of the actual system topology. However, the desired objective is for each node to discover as much of it as possible. The following definitions capture this idea. A solution to a topology discovery problem is complete if every non-faulty process always discovers the complete topology of the system. A solution to the problem is node-complete if every non-faulty process discovers all nodes of the system. A solution is adjacent-edge complete if every non-faulty node discovers each edge adjacent to at least one non-faulty node. A solution is two-adjacent-edge complete if every non-faulty node discovers each edge adjacent to two non-faulty nodes. \ Solution bounds. To simplify the presentation of the negative results in this section we assume more restrictive execution semantics. Each channel contains at most one message. The computation is synchronous and proceeds in rounds. In a single round, each process consumes all messages in its incoming channels and outputs its own messages into the outgoing channels. Notice that the negative results established for this semantics apply for the more general semantics used in the rest of the paper. \[NoWeakComplete\]There does not exist a complete solution to the weak topology discovery problem. Assume there exists a complete solution to the problem. Consider $k \geq 2$ and topology $G_1$ that is not completely connected. Let none of the nodes in $G_1$ be faulty. By the validity property, none of the nodes may detect a fault in such topology. Consider a computation $s_1$ of the solution program where each node discovers $G_1$. Let $p \in G_1$, $q \neq p$, and $r \neq p$ be three nodes in $G_1$, with $q$ and $r$ being non-neighbor nodes in $G_1$. Since $G_1$ is not completely connected we can always find two such nodes. We form topology $G_2$ by connecting $q$ and $r$ in $G_1$. Let $q$ and $r$ be faulty in $G_2$. We construct a computation $s_2$ which is identical to $s_1$. That is, $q$ and $r$, being faulty, in every round output the same messages as in $s_1$. Since $s_2$ is otherwise identical to $s_1$, process $p$ determines that the topology of the system is $G_1 \neq G_2$. Thus, the assumed solution is not complete. \[NoWeak\]There exists no node- and adjacent-edge complete solution to the weak topology problem if the connectivity of the graph is lower or equal to the total number of faults $k$. Assume the opposite. Let there be a node- and adjacent-edge complete program that solves the problem for graphs whose connectivity is $k$ or less. Let $G_1$ and $G_2$ be two graphs of connectivity $k$. This means that $G_1$ and $G_2$ contain the respective cut node sets $A_1$ and $A_2$ whose cardinality is $k$. Rename the processes in $G_2$ such that $A_1=A_2$. By definition $A_1$ separates $G_1$ into two disconnected sets $B_1$ and $C_1$. Similarly, $A_2$ separates $G_2$ into $B_2$ and $C_2$. Assume that $B_1 \not \subseteq B_2$. Since $A_1=A_2$ we can form graph $G_3$ as $A_1 \cup B_2 \cup C_1$. Let $s_1$ be any computation of the assumed program in the system of topology $G_1$ and no faulty nodes. Since the program solves the weak topology problem, the computation has to comply with all the properties of the problem. By validity property, no fault is detected in $s_1$. By termination property, each node in $G_1$, including some node $p \in C_1$, eventually discovers the system topology. By safety property the topology discovered by $p$ is a subset of $G_1$. Since the solution is complete the discovered topology is $G_1$ exactly. Let $s_2$ be any computation of the assumed program in the system of topology $G_2$ and no faulty nodes. Again, none of the nodes detects a fault and all of them discover the complete topology of $G_2$ in $s_2$. We construct a new computation $s_3$ of the assumed program as follows. The system topology for $s_3$ is $G_3$ where all nodes in $A_1$ are faulty. Each faulty node $q \in A_1$ behaves as follows. In the channels connecting $q$ to the nodes of $C_1 \subset G_3$, each round $q$ outputs the messages as in $s_1$. Similarly, in the channels connecting $q$ to the nodes of $B_2 \subset G_3$, $q$ outputs the messages as in $s_2$. The non-faulty nodes of $B_2$ and $C_1$ behave as in $s_1$ and $s_2$ respectively. Observe that for the nodes of $B_2$, the topology and communication is indistinguishable from that of $s_2$. Similarly, for the nodes of $C_1$ the topology and communication is indistinguishable from that of $s_1$. Notice that this means that none of the non-faulty nodes detect a fault in the system. Moreover, node $p \in C_1$ decides that the system topology is the subset of $G_1$. Yet, by construction, $G_1 \neq G_3$. Specifically, $B_1 \not \subseteq B_2$. Moreover, none of the nodes in $B_2$ are faulty. If this is the case then either $s_3$ violates the safety property of the problem or the assumed solution is not adjacent-edge complete. The theorem follows. Observe that for $(k+1)$-connected graphs an adjacent-edge complete solution is also node complete. \[NoStrongEdgeComplete\] There does not exist an adjacent-edge complete solution to the strong topology discovery problem. Assume such a solution exists. Consider system graph $G_1$ that is not completely connected. Let $p \in G_1$ be an arbitrary node. Let $q \neq p$ and $r \neq p$ be two non-neighbor nodes of $G_1$. We form topology $G_2$ by connecting $q$ and $r$ in $G_1$. We construct computations $s_1$ and $s_2$ as follows. Let $s_1$ and $s_2$ be executed on $G_1$ and $G_2$ respectively. And let $q$ be faulty in $s_1$ and $r$ be faulty in $s_2$. Set the output of $q$ in each round to be identical in $s_1$ and $s_2$. Similarly, set the output of $r$ to be identical in both computations as well. Since the output of $q$ and $r$ in both computations is identical, we construct the behavior of the rest of the nodes in $s_1$ and $s_2$ to be the same. Due to termination property, $p$ has to decide on the system topology in both computations. Due to the safety property, in $s_1$ process $p$ has to determine that the topology of the graph is a subset of $G_1$. However, since the behavior of $p$ in $s_2$ is identical to that in $s_1$, $p$ decides that the topology of the system graph is $G_1$ in $s_2$ as well. This means $p$ does not include the edge between $q$ and $r$ to the explored topology in $s_2$. Yet, one of the nodes adjacent to this edge, namely $q$, is not faulty. An adjacent-edge complete program should include such edges in the discovered topology. Therefore, the assumed program is not adjacent-edge complete. \[NoStrong\]There exists no node- and two-adjacent-edge complete solution to the strong topology problem if the connectivity of the graph is less than or equal to twice the total number of faults $k$. Assume that there is a program that solves the problem for graphs whose connectivity is $2k$ or less. Let $G_1$ and $G_2$ be two different graphs whose connectivity is $2k$. Similar to the the proof of Theorem \[NoWeak\], we assume that $G_1 = A_1 \cup B_1 \cup C_1$ and $G_2 = A_2 \cup B_2 \cup C_2$ where the cardinality of $A_1$ and $A_2$ are $2k$, $A_1 = A_2$, $B_1 \cap C_1 = \varnothing$, $B_2 \cap C_2 = \varnothing$, and $B_1 \not \subseteq B_2$. Form $G_3 = A_1 \cup B_2 \cup C_1$. Divide $A_1$ into two subsets $A'_1$ and $A''_1$ of the same number of nodes. Construct a computation $s_1$ with system topology $G_1$ where all nodes in $A'_1$ are faulty; and another computation $s_3$ with system topology $G_3$ where all nodes in $A''_1$ are faulty. The faulty nodes in $s_1$ in the channels connecting $A'_1$ to $C_1$ communicate as the (non-faulty) nodes of $A'_1$ in $s_3$. Similarly, the faulty nodes in $s_3$ in the channels connecting $A''_1$ to $C_1$ communicate as the nodes of $A''_1$ in $s_1$. Observe that $s_1$ and $s_3$ are indistinguishable to the nodes in $C_1$. Let the nodes in $C_1$, including $p \in C_1$ behave identically in both computations. According to the termination property of the strong topology discovery problem every node, including $p$ has to determine the system topology in both $s_1$ and $s_3$. Due to safety, the topology that $p$ determines in $s_1$ is a subset of $G_1$. However, $p$ behaves identically in $s_3$. This means that $p$ decides that the system topology in $s_3$ is also a subset of $G_1$. Since $G_1 \neq G_3$ (specifically, $B_1 \not \subseteq B_2$), and that none of the nodes in $B_2$ are faulty, this implies that either $s_3$ violates the safety property of the problem or the assumed solution is not adjacent-edge complete. The theorem follows. Detector {#SecDetector} ======== Outline. Detector solves the weak topology discovery problem for system graphs whose connectivity exceeds the number of faulty nodes $k$. The algorithm leverages the connectivity of the graph. For each pair of nodes, the graph guarantees the presence of at least one path that does not include a faulty node. The topology data travels along every path of the graph. Hence, the process that collects information about another process can find the potential inconsistency between the information that proceeds along the path containing faulty nodes and the path containing only non-faulty ones. Care is taken to detect the fake nodes whose information is introduced by faulty processes. Since the processes do not know the size of the system, a faulty process may potentially introduce an infinite number of fake nodes. However, the graph connectivity assumption is used to detect fake nodes. As faulty processes are the only source of information about fake nodes, all the paths from the real nodes to the fake ones have to contain a faulty node. Yet, the graph connectivity is assumed to be greater than $k$. If a fake node is ever introduced, one of the non-faulty processes eventually detects a graph with too few paths leading to the fake node. 1123451234512345123451234512345123451234512345= $\textbf{process}\ p$\ $\textbf{const}$\ $P$: set of neighbor identifiers of $p$\ $k$: integer, upper bound on the number of faulty processes\ $\textbf{parameter}$\ $q: P$\ $\textbf{var}$\ $detect:$ boolean, initially false, signals fault\ $start:$ boolean, initially true, controls sending of $p$’s neighborhood info\ $TOP:$ set of tuples, initially $\{(p,P)\}$, (process ids, neighbor id set)\ received by $p$\ $[$\ init:$start \longrightarrow $\ $start := \textbf{false}$,\ $(\forall j: j \in P :\textbf{send}\ (p,P)\ \textbf{to}\ j)$\ \ accept:$\textbf{receive}\ (r,R)\ \textbf{from}\ q \longrightarrow$\ $\textbf{if}\ (\exists s,S: (s,S) \in TOP: s=r \wedge S \neq R)\ \vee$\ $(\textbf{path\_number}(TOP \cup \{(r,R)\}) < k+1)$\ $\textbf{then}$\ $detect:=\textbf{true}$\ $\textbf{else}$\ $\textbf{if}\ (\nexists s,S: (s,S) \in TOP : s=r)\ \textbf{then}$\ $TOP := TOP \cup \{(r,R)\}$,\ $(\forall j: j \in P :\textbf{send}\ (r,R)\ \textbf{to}\ j)$\ $]$\ \ Detailed Description.* The program is shown in Figure \[FigDetector\]. Each process $p$ stores the identifiers of its immediate neighbors. They are kept in set $P$. Each process keeps the upper bound $k$ on the number of faulty processes. Process $p$ maintains the following variables. Boolean variable $detect$ indicates if $p$ discovers a fault in the system. Boolean variable start guards the execution of the action that sends $p$’s neighborhood information to its neighbors. Set $TOP$ (for topology) stores the subgraph explored by $p$; $TOP$ contains tuples of the form: (process identifier, its neighborhood). In the initial state, $TOP$ contains $(p,P)$. Function path\_number evaluates the topology of the subgraph stored in $TOP$. Recall that a node $u$ is unexplored by $p$ if for every tuple $(s,S) \in TOP$, $s$ is not the same as $u$. That is $u$ may appear in $S$ only. We construct graph $G'$ by adding an edge to every pair of unexplored processes present in $TOP$. We calculate the value of path\_number as follows. If the information of $TOP$ is inconsistent, that is: $$\begin{split} & (\exists u,v, U,V: ((u,U) \in TOP) \wedge ((v,V) \in TOP) : \\ & (u \in V) \wedge (v \not\in U)) \end{split}$$ then path\_number returns $0$. If there is exactly one explored node in $TOP$, path\_number returns $k+1$. Otherwise the function returns the minimum number of internally node disjoint paths between two explored nodes in $G'$. In the correctness proof for this program we show that unless there is a fake node, the path\_number of $G'$ is no smaller than the connectivity of $G$. Processes exchange messages of the form (process identifier, its neighborhood id set). A process contains two actions: init and accept. Action init starts the propagation of $p$’s neighborhood throughout the system. Action accept receives the neighborhood data of some process, records it, checks against other data already available for $p$ and possibly further disseminates the data. If the data received from neighbor $q$ about a process $r$ contradicts what $p$ already holds about $r$ in $TOP$ or if the newly arrived information implies that $G$ is less than $(k+1)$-connected $p$ indicates that it detected a fault by setting $detect$ to $\textbf{true}$. Alternatively, if $p$ did not previously have the information about $r$, $p$ updates $TOP$ and sends the received information to all its neighbors. Observe that the propagation of information about the neighborhood of a certain process is independent of the information propagation of another process. Thus, we will focus on the propagation of the information about a particular non-faulty process $a$. Let $COR$ contain each process $b$ such that $b$ is not faulty and $TOP.b$ holds $(a,A)$. Let $a$ itself belong to $COR$ if $start.a$ is false. \[ProveInvDetect\]The following predicate is an invariant of Detector. $$\label{InvDetect} \begin{split} &( \forall\ \textup{non-faulty}\ b,c: b \in COR, c \in B: \\ &(c \in COR) \vee \\ &((a,A) \in Ch.b.c))\ \vee \\ &(\exists\ \textup{non-faulty}\ j: j \in N: detect.j = \textbf{\textup{true}}) \end{split}$$ The predicate states that unless one of the non-faulty processes in the program detects a fault, if a process $b$ belongs to $COR$ then each neighbor $c$ of $b$ either belongs to $COR$ as well or the channel from $b$ to $c$ contains $(a,A)$. \ To prove that Predicate \[InvDetect\] is an invariant of the program, we need to show that it holds in the initial state of any computation and it is closed under the execution of actions of Byzantine as well as non-faulty processes. The predicate holds initially as the first disjunct is vacuously true. Note that no action of a Byzantine process immediately affects the validity of the predicate. Observe also that a non-faulty process can only set $detect$ to $\textbf{true}$. Thus, once this happens the predicate holds throughout the rest of the computation. Suppose $detect$ is $\textbf{false}$ in all processes of the program. Then the predicate is violated only if there is a non-faulty pair of neighbors $b$ and $c$ such that $b$ belongs to $COR$, $c$ does not and there is no message $(a,A)$ in the channel from $b$ to $c$. Notice that a non-faulty process adds the first value $(r,R)$ to $TOP$ and never changes it afterwards. Thus, provided that $detect=\textbf{false}$, to violate the predicate, a process has to join $COR$ without sending $(a,A)$ to its neighbors or consume a message with $(a,A)$ without joining $COR$. Let us examine the actions of a non-faulty process and ensure that neither of this happens. Observe that init is only of interest in $a$. This action sets $start.a = \textbf{false}$ which, by definition, adds $a$ to $COR$. Also, init atomically sends $(a,A)$ to all neighbors of $a$. Thus, the predicate is not violated by the execution of init. Let us now consider accept in an arbitrary non-faulty process $u$. Let the message received by $u$ carry $(r,R)$. Observe that accept affects Predicate \[InvDetect\] only if $r=a$. accept may make $u$ join $COR$ or consume a message with $(a,A)$. Notice, that if $u$ is already in $COR$ the receipt of a message with $(a,A)$ does not violate the predicate. Also, $u$ joins $COR$ only if it receives $(a,A)$. Hence, the only case we have to consider is when $u$ does not belong to $COR$ before the execution of accept, $u$ receives $(a,A)$ and joins $COR$. The behavior of $u$ in this case depends on whether it has an element $(s,S)$ in $TOP.u$ such that $s=a$. Since $u \not \in COR$, if $(a,S) \in TOP.u$, then $S$ differs from $A$. In this case if $u$ receives $(a,A)$ then it sets $detect=\textbf{true}$. This preserves the validity of the predicate. Alternatively, if such an entry in $TOP.u$ does not exist, then the receipt of $(a,A)$ causes $u$ to join $COR$ and forward $(a,A)$ to all its neighbors. This preserves the predicate as well. Thus, Predicate \[InvDetect\] holds in the initial state of every computation of the program and is preserved by its every action. Which means that this predicate is an invariant of the program. \[NeighborJoin\]If a computation of Detector* contains a state where there is a process $u$ that belongs to $COR$ that has a non-faulty neighbor $v$ that does not, then further in the computation, either some non-faulty process sets $detect=\textbf{\textup{true}}$ or $v$ joins $COR$.* According to Lemma \[ProveInvDetect\], Predicate \[InvDetect\] is an invariant of the program. Hence, if $u$ belongs to $COR$ and its non-faulty neighbor $v$ does not, then channel $Ch.u.v$ contains a message with $(a,A)$. Due to fair message receipt assumption, $(a,A)$ is received. Observe that if $v$ is not in $COR$ and it receives $(a,A)$, then either $v$ sets $detect=\textbf{true}$ or joins $COR$. \[AllJoin\]Every computation of Detector* contains a state where either $detect=\textbf{\textup{true}}$ in some non-faulty process or every non-faulty process belongs to $COR$.* The proof is by induction on the number of non-faulty processes in the program. As a base case, we show that $a$ itself eventually joins $COR$. Recall, that we assume that $a$ itself is not faulty. Observe that the program starts in a state where $start.a$ is $\textbf{true}$. If this is so, init is enabled. Moreover, init is the only action that sets $start.a$ to $\textbf{false}$. Thus, init stays enabled until executed. By weak fairness assumption, init is eventually executed. When this happens, $a$ joins $COR$. Assume that $COR$ contains $i$: $1 \leq i< n$ processes at some state of a computation and there is a non-faulty process that does not belong to $COR$. We assume that the connectivity of the graph exceeds the maximum number of faulty processes. Thus, there is a non-faulty process $u \in COR$ that has a non-faulty neighbor $v \not \in COR$. According to Lemma \[NeighborJoin\], this computation contains a state where $COR$ contains $v$. Thus, every non-faulty process eventually joins $COR$. \[NoFake\]If a computation of Detector* contains a state where non-faulty process $u$ explores a fake process $v$, then this computation contains a state where $detect=\textbf{\textup{true}}$ in some non-faulty process.* Observe that the only source of fake process information is a Byzantine process. Hence, if $u$ explores a fake process $v$, then every path to $v$ leads through a Byzantine process. Thus, in a graph with a fake node, the maximum number of node-disjoint paths between a real and a fake node is no more than $k$. According to Lemma \[AllJoin\], eventually, either $detect=\textbf{\textup{true}}$ at a non-faulty process or $u$ explores every non-faulty process in the system. In this case $u$ detects that all paths to the fake node $v$ lead through no more than $k$ processes and sets $detect=\textbf{\textup{true}}$. \[PathNumberOK\]If the system does not have a faulty process, then in every computation, for each process, the path\_number* of the explored subgraph $G'$ is greater than $k$.* Observe that if there are no faulty processes, only correct topology information is circulated in the system. Hence, for each process $u$, $TOP.u$ contains the subgraph of the system graph $G$. In this case, $G'.u$ is an arbitrary set of explored processes from $G$ and the unexplored members of their neighborhoods. By the construction of $G'.u$, every pair of unexplored processes is connected by an edge. Let $v$ and $w$ be an arbitrary pair of explored nodes in $G'.u$. And let $P$ be a path connecting $v$ and $w$ in $G$. We claim that there exists a path $P'$ in $G'.u$ connecting $v$ and $w$ that is also a node-subset of $P$. That is, every node that belongs to $P'$ also belongs to $P$. See Figure \[PathNumberOKFig\] for the illustration. If $P$ contains only the nodes explored in $G'.u$, our claim holds since $P'=P$. Let $P$ contain unexplored nodes as well. In general, $P$ contains alternating segments of explored and unexplored nodes. Let $\langle x_i,y_i,\cdots,y_{i+1},x_{i+1}\rangle$ be any such unexplored segment, where $x_i,x_{i+1}$ are explored and $y_i,\cdots,y_{i+1}$ are not. Observe that $y_i$ and $y_{i+1}$ have explored neighbors — $x_i$ and $x_{i+1}$ respectively. This means that both $y_i$ and $y_{i+1}$ belong to $G'.u$. Since $y_i$ and $y_{i+1}$ are unexplored, $G'.u$ contains an edge connecting them. We construct $P'$ to contain every explored segment of $P$; we replace every unexplored segment by the edge that links unexplored nodes in $G'.u$. Observe that by construction, $P' \in G'.u$ and $P'$ contains a subset of the nodes of $P$. Thus, our claim holds. Let $P_1$ and $P_2$ be two internally node disjoint paths connecting $v$ and $w$ in $G$. According to the just proved claim, there exist $P'_1$ and $P'_2$ belonging $G'.u$ that connect $v$ and $w$. Moreover, $P'_1$ contains a subset of nodes of $P_1$ and $P'_2$ contains a subset of nodes of $P_2$. Since $P_1$ and $P_2$ are internally node disjoint, so are $P'_1$ and $P'_2$. Recall that $G$ is assumed to be $(k+1)$-connected. This means that for every two vertices $v$ and $w$ there exist $k+1$ internally node disjoint paths between $v$ and $w$. Thus, the number of internally node disjoint paths for $v$ and $w$ in $G'.u$ is at least $k+1$. Hence, the path\_number of $G'.u$ is greater than $k$. \[DValidity\] Any computation of a detector program contains a state where a Byzantine process is detected only if there indeed is a Byzantine process in the system. A non-faulty process sets $detect$ to $\textbf{true}$ if it encounters divergent information about some node’s neighborhood or when it detects that path\_number is less than $k+1$. However, a non-faulty process never modifies the neighborhood information about other processes. Hence, if the program does not have a faulty process, all the information about a particular neighborhood that is circulated in the system is identical. Also, according to Lemma \[PathNumberOK\] if there are no faulty processes in the system, the path\_number never falls below $k+1$. Hence, $detect$ is set to $\textbf{true}$ only if indeed the system contains a faulty process. *Detector is an adjacent-edge complete solution to the weak topology discovery problem in case the connectivity of system topology graph exceeds the number of faults.* To prove the theorem we show that every computation of Detector conforms to the properties of the problem. We then show that the discovered topology is adjacent-edge complete. Termination property follows from Lemma \[AllJoin\], safety — from Lemma \[NoFake\], while validity follows from Lemma \[DValidity\]. Notice that Lemma \[AllJoin\] states that unless a fault is detected, the neighborhood of every non-faulty process is added to $COR$. That is, edges adjacent to a non-faulty processes are detected by every non-faulty processes. Thus, Detector is adjacent-edge complete. Hence the theorem. \ Efficiency evaluation. Since we consider an asynchronous model, the number of messages a Byzantine process can send in a computation is infinite. To evaluate the efficiency of Detector we assume that each process is familiar with the upper bound on the number of processes in the system and this upper bound is in $O(n)$. A non-faulty process then detects a fault if the number of processes it explores exceeds this bound or if it receives more than one identical message from the same neighbor. We assume that the process stops and does not send or receive any more messages if it detects a fault. In this case we can estimate the number of messages that are received by non-faulty processes before one of them detects a fault or before the computation terminates. To make the estimation fair, the assume that the unit is $log(n)$ bits. Since it takes that many bits to assign unique process identifiers to $n$ processes, we assume that one identifier is exactly one unit of information. A message in Detector carries up to $\delta+1$ identifiers, where $\delta$ is the maximum number of nodes in the neighborhood of a process. Observe that a process can receive at most $n$ messages from each incoming channel. Thus, the total number of messages that can be sent by Detector is $2en$, where $e$ is the number of edges in the graph. The message complexity of the program is in $O(2en\delta)$. If $e$ is proportional to $n^2$, then the complexity of the program is in $O(\delta n^3)$. Explorer {#SecExplorer} ======== Outline. The main idea of Explorer is for each process to collect information about some node’s neighborhood such that the information goes along more than twice as many paths as the maximum number of Byzantine nodes. While the paths are node-disjoint, the information is correct if it comes across the majority of the paths. In this case the recipient is in possession of confirmed information. It turns out that the topology information does not have to come directly from the source. Instead it can come from processes with confirmed information. The detailed description of Explorer follows. To simplify the presentation, we describe and prove correct the version of Explorer that tolerates only one Byzantine fault. We describe how this version can be extended to tolerate multiple faults in the end of the section. \ Description. Since we first describe the 1-fault tolerant version of Explorer we assume that the graph is $3$-connected. The program is shown in Figure \[FigExplorer\]. Similar to Detector, each process $p$ in Explorer, stores the ids of its immediate neighbors. Process $p$ maintains the variable start, whose function is to guard the execution of the action that initiates the propagation of $p$’s own neighborhood. Unlike Detector, however, $p$ maintains two sets that store the topology information of the network: $uTOP$ and $cTOP$. Set $uTOP$ stores the topology data that is unconfirmed; $cTOP$ stores confirmed topology data. Set $uTOP$ contains the tuples of neighborhood information that $p$ received from other nodes. Besides the process id and the set of its neighbor ids, each such tuple contains a set of process identifiers, that relayed the information. We call it visited set. The tuples in $cTOP$ do not require a visited set. Processes exchange messages where, along with the neighbor identifiers for a certain process, a visited set is propagated. A process contains two actions: init and accept. The purpose of init is similar to that in the process of Detector. Action accept receives the neighborhood information of some process $r$, its neighborhood $R$ which was relayed by nodes in set $S$. The information is received from $p$’s neighbor — $q$. First, accept checks if the information about $r$ is already confirmed. If so, the only manipulation is to record the received information in $uTOP$. Actually, this update of $uTOP$ is not necessary for the correct operation of the program, but it makes the its proof of correctness easier to follow. If the received information does not concern already confirmed process, accept checks if this information differs from what is already recorded in $uTOP$ either in $r$ or in $R$. In either case the information is broadcast to all neighbors of $p$. Before broadcasting, $p$ appends the sender — $q$ to the visited set $S$. If the information about $r$ and $R$ has already been received and recorded in $uTOP$, accept checks if the previously recorded information came along an internally node disjoint path. If so, the information about $r$ is added to $cTOP$. In this case, this information is also broadcast to all $p$’s neighbors. Note, however, that $p$ is now sure of the information it received. Hence, the visited set of nodes in the broadcast message is empty. 1123451234512345123451234123412341234512345= $\textbf{process}\ p$\ $\textbf{const}$\ $P$, set of neighbor identifiers of $p$\ $\textbf{parameter}$\ $q: P$\ $\textbf{var}$\ $start:$ boolean, initially true, controls sending of $p$’s neighbor ids\ $cTOP:$ set of tuples, initially $\{(p,P)\}$,\ (process id, neighbor id set) confirmed topology info\ $uTOP:$ set of tuples, initially $\varnothing$,\ (process id, neighbor id set, visited id set)\ unconfirmed topology info\ $[$\ init:$start \longrightarrow $\ $start := \textbf{false}$,\ $(\forall j: j \in P :\textbf{send}\ (p,P,\varnothing)\ \textbf{to}\ j)$\ \ accept:$\textbf{receive}\ (r,R,S)\ \textbf{from}\ q \longrightarrow$\ $\textbf{if}\ (\forall t,T: (t,T) \in cTOP : t \neq r)\ \textbf{then}$\ $\textbf{if}\ (\forall t,T,U: (t,T,U) \in uTOP: t \neq r \vee T \neq R) \ \textbf{then}$\ $(\forall j: j \in P: \textbf{send}\ (r, R, S \cup \{q\})\ \textbf{to}\ j)$\ $\textbf{elsif}\ (\exists t,T,U: (t,T,U) \in uTOP: $\ $\;t=r \wedge R=T \wedge ((U \cap (S \cup \{q\}))) \subset \{r\})) $\ $\textbf{then}$\ $cTOP :=cTOP \cup \{(r,R)\}$,\ $(\forall j: j \in P: \textbf{send}\ (r, R, \varnothing)\ \textbf{to}\ j)$\ $uTOP := uTOP \cup \{(r, R, S \cup \{q\})\}$\ $]$\ \ Correctness proof. Just like for the Detector* algorithm, we are focusing on the propagation of the neighborhood information $A$ of a singular non-faulty process $a$. Notice that we use $A$ to denote the correct neighborhood info. We use $A'$ for the neighborhood information of $a$ that may not necessarily be correct. To aid us in the argument, we introduce an auxillary set SENT to be maintained by each process. Since this set does not restrict the behavior of processes, we assume that the Byzantine process maintains this set as well. SENT contains each message sent by the process throughout the computation. Notice that $uTOP$ records every message received by the process in the computation. Hence, the comparison of $uTOP$ and $SENT$ allows us to establish the channel contents. Since, a message cannot be received without being sent and vice versa, the following proposition states the invariant of the predicate that affirms it. \[ProveInvExplore\]The following predicate is an invariant of the Explorer* program. $$\label{InvExplore} \begin{split} &(\forall b, \textup{non-faulty}\ c, A', V : c \in B : \\ & (((a,A',V) \in Ch.b.c) \vee \\ & ((a,A',V \cup \{b\}) \in uTOP.c)) \Leftrightarrow \\ & ((a,A',V) \in SENT.b)) \end{split}$$* The predicate states that for any process $b$ and its non-faulty neighbor $c$ the information about the neighborhood of $a$ is recorded in $SENT.b$ if and only if this information is en route from $b$ to $c$ or is recorded in $uTOP.c$ with $b$ appended to the sequence of visited nodes $V$. Before we proceed with the correctness argument we have to introduce additional notation. We say that some process $c$ confirms $(a,A')$ if it adds this tuple to $cTOP.c$. We view the propagation of $A'$ as construction of a tree of processes that relayed $A'$. This tree carries $A'$. A tree contains two types of nodes: a root and non-root. If process $c$ is non-root, then for some $V$, $(a,A',V) \in SEND.c$ and $(a,A',V) \in uTOP.c$. That is, a non-root is a process that forwarded the information received from elsewhere without alteration. If $c$ is a root, then $(a,A',V) \in SEND.c$ but $(a,A',V) \not\in uTOP.c$. Node $c$’s ancestor in a tree is the node that lies on a path from $c$ to the root. Observe that the root of a tree can only be the process $a$ itself, the Byzantine node or a node that confirms $(a,A')$. Notice also that since each non-faulty process $c$ sends a message about $a$’s information at most twice, $c$ can belong to at most two trees. Moreover, $c$ has to be the root of one of those trees. The proposition below follows from Proposition \[ProveInvExplore\]. \[TreeOK\]If some process $d$ is the ancestor of another process $c$ in a tree carrying $(a,A')$ and $(a,A',V) \in uTOP.c$, then $d \in V$. \[GoodInfoInCTOP\]If a non-faulty node $c$ confirms $(a,A')$, then $A'=A$ and $a$ is real. Let us first suppose that $a$ is real. Further, suppose $c$ is the first non-faulty process in the system, besides $a$, to confirm $(a,A')$. To add $(a,A')$ to $cTOP.c$ any process $c \neq a$ has to contain $(a, A', V) \in uTOP.c$ and receive a message from one of its neighbors $b$ carrying $(a,A',V')$ such that $V \cap V' \subset \{a\}$. In our notation this means that $c$ belongs to a tree that carries $(a,A')$ and receives a message from $b$ (possibly belonging to a different tree) that carries the same information: $(a,A')$. Let us consider if $b$ and $c$ belong to the same or different trees. Suppose $b$ and $c$ belong to the same tree. If this is the case the messages that $c$ receives have to share nodes in the visited sets $V$ and $V'$. However, for $c$ to confirm $(a,A')$ the intersection of $V$ and $V'$ has to be a subset of $\{a\}$. That is, the only common node between the two sets is $a$. Observe that $a$ does not forward the information about its own neighborhood if it receives it from elsewhere. Thus, if $a$ belongs to a tree then $a$ is its root. In this case $A'=A$. Suppose $b$ and $c$ belong to different trees. Recall that for $c$ to confirm $(a,A')$, both of these trees have to carry $(a,A')$. However, if $A' \neq A$ then the root of the tree is either the faulty node or another node that confirmed $(a,A')$. Yet, we assumed that $c$ is the first node to do so. Thus, if $c$ receives a message from $b$, the only tree that carries the information $(a,A')$ such that $A' \neq A$ is rooted in the faulty node. Thus, even if $b$ and $c$ belong to different trees, $A'=A$. Similarly, if $a$ is fake, unless another node confirms $(a,A')$ there is only one tree that carries $(a,A')$ and it is rooted in the faulty node. In this case, no other node confirms $(a,A')$. \[AllJoinTree\]Every computation of Explorer* contains a state where each non-faulty process belongs to at least one tree carrying $(a,A)$.* We prove the lemma by induction on the number of nodes in the system. To prove the base case we observe that the init action is enabled in $a$ in the beginning of every computation. This action stays enabled unless executed. Thus, due to weak-fairness of action execution assumption, init is eventually executed in $a$. When it is executed, $a$ forms a tree carrying $(a,A)$. Let us assume that there are $i$: $1 \leq i <n$ non-faulty nodes that belong to trees carrying $(a,A)$. Since the network is at least $3$-connected, there exists a non-faulty process $c$ that does not belong to such a tree but has a neighbor $b$ that does. If $b$ belongs to a tree carrying $(a,A)$ then $SEND.b$ contains an entry $(a,A,V)$ for some set of visited nodes $V$. If $c$ does not belong to such a tree then, by definition, $(a,A,V') \not\in uTOP.c$. In this case, according to Proposition \[ProveInvExplore\], $Ch.b.c$ contains $(a,A,V)$. Similar argument applies to the other neighbors of $c$ that belong to trees carrying $(a,A)$. That is, $c$ has incoming messages from every such neighbor. According to the fair message receipt assumption, these messages are eventually received. We can assume, without loss of generality, that $c$ receives a message from $b$ first. Since $c$ does not contain an entry $(a,A,V')$ in $uTOP.c$, upon receipt of the message from $b$, $c$ sends a message with $(a,A,V \cup \{b\})$, attaches this message to $SEND.c$ and includes it in $uTOP.c$. This means that $c$ joins the tree carrying $(a,A)$. Thus, every non-faulty node eventually joins a tree carrying correct neighborhood information about $a$. A branch of a tree is either a subtree without the root or the root process alone. The following proposition follows from Proposition \[ProveInvExplore\]. \[2TreesMerge\]If a computation of Explorer* contains a state where a non-faulty node $c$ and its neighbor $b$ either belong to two different trees carrying the same information $(a,A)$ or to two different branches of the tree rooted in $a$, then this computation also contains a state where $c$ confirms $(a,A)$.* \[AllConfirm\]Every non-faulty process $c$ eventually confirms $(a,A)$. The proof is by induction on the number of nodes in the system. The base case trivially holds as $a$ itself confirms $(a,A)$ in the beginning of every computation. Assume that $i$ non-faulty processes have $(a,A)$ in $cTOP$, where $1 \leq i < n$. We show that if there exists another non-faulty process $c$, it eventually confirms $(a,A)$. Two cases have to be considered: there exists only one tree carrying $(a,A)$, and there are multiple such trees. Let us consider the first case. Notice, that in every computation there eventually appears a tree rooted in $a$. In this case, we may only consider a tree so rooted. Since the network is at least $3$-connected, there exists a simple cycle containing $a$ and not containing the faulty process. According to Lemma \[AllJoinTree\], every process in the cycle eventually joins this tree. Observe that, by our definition of a tree branch, there always is a pair of neighbor processes $b$ and $c$ that belong to different branches of a tree rooted in $a$ and carrying $(a,A)$. In this case, according to Proposition \[2TreesMerge\], one of the two nodes eventually confirms $(a,A)$. Let us now consider the case of multiple trees carrying $(a,A)$. Again, according to Lemma \[AllJoinTree\], each non-faulty process in the system joins at least one of these trees. Since the network is at least $3$-connected there exists a non-faulty process $c$ belonging to one tree that has a neighbor $b$ belonging to a different tree. In this case, according to Proposition \[2TreesMerge\], $c$ confirms $(a,A)$. By induction, every non-faulty process in the system eventually confirms $(a,A)$. *Explorer is a two-adjacent-edge complete solution to the strong topology discovery problem in case of one fault and the system topology graph is at least $3$-connected.* Explorer conforms to the termination and safety properties of the problem as a consequence of Lemmas \[AllConfirm\] and \[GoodInfoInCTOP\] respectively. Observe that a non-faulty node may potentially confirm incorrect neighborhood information about a Byzantine node. That is, an edge reported by the faulty process is either missing or fake. However, due to the two above lemmas, if two nodes are non-faulty the information whether there is an adjacent edge between them is discovered by every non-faulty node. Hence Explorer is two-adjacent-edge complete. \ Modification to handle $k>1$ faults. Observe that Explorer confirms the topology information about a node’s neighborhood, when it receives two messages carrying it over internally node disjoint paths. Thus, the program can handle a single Byzantine fault. Explorer can handle $k>1$ faults, if it waits until it receives $k+1$ messages before it confirms the topology info. All the messages have to travel along internally node disjoint paths. For the correctness of the algorithm, the topology graph has to be $(2k+1)$-connected. *Explorer is a two-adjacent-edge complete solution to the strong topology discovery problem in case of $k$ faults and the system topology graph is at least $(2k+1)$-connected.* Efficiency evaluation. Unlike Detector, Explorer does not quit when a fault is discovered. Thus, the number of messages a faulty node may send is arbitrary large. However, we can estimate the message complexity of Explorer in the absence of faults. Each message carries a process identifier, a neighborhood of this process and a visited set. The number of the identifiers in a neighborhood is no more than $\delta$, and the number of identifiers in the visited set can be as large as $n$. Hence the message size is bounded by $\delta+n+1$ which is in $O(n)$. Notice, that for the neighborhood $A$ of each process $a$, every process broadcasts a message twice: when it first receives the information, and when it confirms it. Thus, the total number of sent messages is $4e \cdot n$ and the overall message complexity of Explorer if no faults are detected is in $O(n^4)$. Composition and Extensions {#SecExtensions} ========================== *Composing Detector* and Explorer*. Observe that Detector* has better message complexity than Explorer if the neighborhood size is bounded. Hence, if the incidence of faults is low, it is advantageous to run Detector and invoke Explorer only if a fault is detected. We assume that the processes can distinguish between message types of Explorer and Detector. In the combined program, a process running Detector switches to Explorer if it discovers a fault. Other processes follow suit, when they receive their first Explorer messages. They ignore Detector messages henceforth. A Byzantine process may potentially send an Explorer message as well, which leads to the whole system switching to Explorer. Observe that if there are no faults, the system will not invoke Explorer. Thus, the complexity of the combined program in the absence of faults is the same as that of Detector. Notice that even though Detector alone only needs $(k+1)$-connectivity of the system topology, the combined program requires $(2k+1)$-connectivity. \ Message Termination. We have shown that Detector and Explorer comply with the functional termination properties of the topology discovery problem. That is, all processes eventually discover topology. However, the performance aspect of termination, viz. message termination, is also of interest. Usually an algorithm is said to be message terminating if all its computations contain a finite number of sent messages [@DS80]. However, a Byzantine process may send messages indefinitely. To capture this, we weaken the definition of message termination. We consider a Byzantine-tolerant program message terminating if the system eventually arrives at a state where: (a) all channels are empty except for the outgoing channels of a faulty process; (b) all actions in non-faulty processes are disabled except for possibly the receive-actions of the incoming channels from Byzantine processes, these receive-actions do not update the variables of the process. That is, in a terminating program, each non-faulty process starts to eventually discard messages it receives from its Byzantine neighbors. Making Detector terminating is fairly straightforward. As one process detects a fault, the process floods the announcement throughout the system. Since the topology graph for Detector is assumed $(k+1)$-connected, every process receives such announcement. As the process learns of the detection, it stops processing or forwarding of the messages. Notice that the initiation of the flood by a Byzantine node itself, only accelerates the termination of Detector as the other processes quickly learn of the faulty node’s existence. The addition of termination to Explorer is more involved. To ensure termination, restrictions have to be placed on message processing and forwarding. However, the restrictions should be delicate as they may compromise the liveness properties of the program. By the design of Explorer, each process may send at most one message about its own neighborhood to its neighbors. Hence, the subsequent messages can be ignored. However, a faulty process may send messages about neighborhoods of other processes. These processes may be real or fake. We discuss these cases separately. Note that each process in Explorer can eventually obtain an estimate of the identities of the processes in the system and disregard fake process information. Indeed, a path to a fake node can only lead through faulty processes. Thus, if a process discovers that there may be at most $k$ internally node disjoint paths between itself and a certain node, this node is fake. Therefore, the process may cease to process messages about the fake node’s neighborhood. Notice, that since the system is $(2k+1)$-connected, messages about real nodes will always be processed. Therefore, the liveness properties of Explorer are not affected. As to the real processes, they can be either Byzantine or non-faulty. Recall that each non-faulty process of Explorer eventually confirms neighborhoods of all other non-faulty processes. After the neighborhood of a process is confirmed, further messages about it are ignored. The last case is a Byzantine process $u$ sending a message to its correct neighbor $v$ about the neighborhood of another Byzantine process $w$. By the design of Explorer, $v$ relays the message about $w$ provided that the neighborhood information about $w$ differs from what previously received about $w$. As we discussed above, eventually $v$ estimates the identities of all real processes in the system. Therefore, there is a finite number of possible different neighborhoods of $w$ that $u$ can create. Hence, eventually they will be exhausted, and $v$ starts ignoring further messages form $u$ about $w$. Thus, Explorer can be made terminating as well. \ Handling topology updates. In the topology discovery problem statement, it is assumed that the system topology does not change. However, Detector and Explorer can be adapted to manage topology changes as well. There are two aspects of topology change: the notification and the transport. For notification, a node should inform the others of its most up-to-date neighborhood. The transport aspect should ensure that this notification is delivered to all nodes despite of topology changes. We implement the transport aspect as follows. If a node $p$, due to the change in topology, obtains a new neighbor $q$. Then $p$ sends to $q$ the most recent neighborhood information about all nodes that $p$ is aware of. Thus, the most recent information gets propagated regardless of topology changes. The satisfaction of the notification aspect is more involved. Observe, however, that apart from detecting fake nodes in Explorer, both algorithms propagate the information of one process neighborhood independently of the others. We first describe how this propagation can be done in case the topology changes and then address the fake node detection. Each time the neighborhood of a process $p$ changes, $p$ starts a new version of the topology discovery algorithm for its neighborhood. Observe that a faulty process may also start a new version for $p$. The versions are distinguished by version numbers. Each process maintains the version numbers of $p$. Each related message carries the version number. Each process outputs the discovered neighborhood of $p$ with the highest received version number. Observe that in the case of Explorer the processes only output confirmed information. Notice that if a faulty process sends incorrect information about $p$’s neighborhood with a certain version number, this incorrect info will be handled by the basic Detector or Explorer within that version. For example, the faulty messages of version $i$ about $p$’s neighborhood will be countered by the correct messages of the same version. Notice that a faulty process in Explorer may start a version $j$ for $p$’s neighborhood such that it is higher than the highest version $i$ that $p$ itself started. However, according to the basic Explorer, the incorrect information in version $j$ will not be confirmed. There are two specific modifications to the basic Detector. If the faulty process sends a message concerning $p$ with the version number higher than that of $p$, $p$ itself detects the fault. To detect fake nodes generated by a faulty process, each node has to compile the topology $TOP$ graph of the highest version number for each node in the system and ensure that its connectivity does not fall below $k+1$. Observe that Detector is unable to differentiate between temporary lack of connectivity from malicious behavior of the faulty nodes. Therefore, the connectivity of the discovered network at each node should never fall below $k+1$. For that, we assume that throughout a computation the intersection of all system topologies is $k+1$-connected. This assumption is not necessary for Explorer. The notification mechanism can be optimized in obvious ways. For Detector, each process has to keep the information for $p$ with only the highest version number. Obsolete information can be safely discarded. For Detector, the process may keep the latest version of confirmed neighborhood information. Observe that this extension of the topology discovery algorithms assumes infinite-size counters. Care must be taken when implementing these counters in the actual hardware, as the faulty processes may try to compromise topology discovery if the counter values are reused. Hence, such an implementation would require a Byzantine-robust counter synchronization algorithm. Lamport and Melliar-Smith [@LM86] proposed such algorithm for completely connected systems. Extending it to arbitrary topology systems is an attractive avenue of future research. \ Discovering neighbors. As described, in the initial state of Detector and Explorer, each process has access to correct information about its immediate neighborhood. Note that, in general, obtaining this information in the presence of Byzantine processes may be difficult as they can mount a Sybil attack [@D02]. In such an attack, a faulty process is able to send a message and put an arbitrary process identifier as the sender of this message. That is, a faulty process assumes the identity of this process. Sybil attack is difficult to handle. However, Detector and Explorer can be modified to handle neighborhood discovery with known ports. That is, each process does not know the identities of its neighbors but can determine if a message is coming from the same process. The modified algorithms contain two phases: neighborhood discovery phase and topology discovery proper phase. In the first phase, each process broadcasts its identifier to its neighbors. Observe that faulty processes may not send these initial messages at all. Thus, the process should not wait for a message from every possible neighbor. Instead, as soon as each process $p$ gets a message with $q$ in its identifier, $p$ may start the second phase with $\{q\}$ as its neighborhood. Every time $p$ gets a new distinct identity, $p$ treats it as topology update, increments its counter and re-initiates the topology discovery. This procedure can be further streamlined. Recall that for Detector and Explorer the topology graph has to be respectively $k+1$ and $2k+1$-connected. Thus, depending on the algorithm, each process is guaranteed to have $k+1$ or $2k+1$ non-faulty neighbors. Therefore, each process may delay initiating topology discovery until it gets this minimum number of distinct identities. Observe that due to known ports a faulty process may not be able to use more than one identifier per neighbor without being detected. However, the modified algorithms may not be able to determine the identifier of a faulty process as it may select an arbitrary one, including the identifier of an already existing process. Thus, a pair of colluding faulty nodes may deceive their non-faulty neighbors into believing that they share an edge. This behavior is illustrated in Figure \[blackhole\]. When communicating to a non-faulty node $a$, its faulty neighbor $b$ assumes the identity of another non-faulty node $d$. Similarly, a faulty neighbor $c$ of $d$ assumes the identity of $a$. This way, non-faulty nodes $a$ and $d$ are led to believe they share an edge. \ Other extensions. Observe that Explorer is designed to disseminate the information about the complete topology to all processes in the system. However, it may be desirable to just establish the routes from all processes in the system to one or a fixed number of distinguished ones. To accomplish this Explorer needs to be modified as follows. No neighborhood information is propagated. Instead of the visited set, each message carries the propagation path of the message. That is, the order of the relays is significant. Only the distinguished processes initiate the message propagation. The other processes only relay the messages. Just as in the original Explorer, a process confirms a path to another process only if it receives $2k+1$ internally process disjoint paths from the source or from other confirming processes. Again, like in Explorer, such process rebroadcasts the message, but empties the propagation path. In the outcome of this program, for every distinguished process, each non-faulty process will contain paths to at least $2k+1$ processes that lead to this distinguished process. Out of these paths, at least $k+1$ ultimately lead to the distinguished process. In Explorer, for each process the propagation of its neighborhood information is independent of the other neighborhoods. Thus, instead of topology, Explorer can be used for efficient fault-tolerant propagation of arbitrary information from the processes to the rest of the network. Conclusion {#SecEnd} ========== In conclusion, we would like to outline a couple of interesting research directions. The existence of Byzantine-robust topology discovery solutions opens the question of theoretical limits of efficiency of such programs. The obvious lower bound on message complexity can be derived as follows. Every process must transmit its neighborhood to the rest of the nodes in the system. Transmitting information to every node requires at least $n$ messages, so the overall message complexity is at least $\delta n^2$. If $k$ processes are Byzantine, they may not relay the messages of other nodes. Thus, to ensure that other nodes learn about its neighborhood, each process has to send at least $k+1$ messages. Thus, the complexity of any Byzantine-robust solution to the topology discovery problem is at least in $\Omega(\delta n^2 k)$. Observe that Explorer and Detector may not explicitly identify faulty nodes or the inconsistent view of the their immediate neighborhoods. We believe that this identification can be accomplished using the technique used by Dolev [@D82]. In case there are $3k+1$ non-faulty processes, they may exchange the topologies they collected to discover the inconsistencies. This approach, may potentially expedite termination of Explorer at the expense of greater message complexity: if a certain Byzantine node is discovered, the other processes may ignore its further messages. [^1]: This author was supported in part by DARPA contract OSU-RF\#F33615-01-C-1901 and by NSF CAREER Award 0347485. Part of this work was done while the author was visiting Paris Sud University. [^2]: This author was supported in part by the FNS grants FRAGILE and SR2I from ACI “Sécurité et Informatique”, and ANR grant SOGEA from ARA program “Sécurité, Systèmes Embarqués et Intelligence Ambiante”. Part of this work was done while the author was visiting Kent State University. [^3]: Some of the results in this article were presented at the 13th Colloquium on Structural Information and Communication Complexity, Chester, UK in July 2006 [^4]: We use terms process and node interchangeably.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We analyze a class of piecewise linear parabolic maps on the torus, namely those obtained by considering a linear map with double eigenvalue one and taking modulo one in each component. We show that within this two parameter family of maps, the set of noninvertible maps is open and dense. For cases where the entries in the matrix are rational we show that the maximal invariant set has positive Lebesgue measure and we give bounds on the measure. For several examples we find expressions for the measure of the invariant set but we leave open the question as to whether there are parameters for which this measure is zero.' author: - \| Peter Ashwin and Xin-Chu Fu\ Department of Mathematics and Statistics\ University of Surrey\ Guildford GU2 5XH, UK - \| Takashi Nishikawa\ Institute for Plasma Research, University of Maryland\ College Park, MD 20742, USA - \| Karol [Ż]{}yczkowski\ Instytut Fizyki im. Smoluchowskiego, Uniwersytet Jagiello[ń]{}ski\ ul. Reymonta 4, 30-059 Krak[ó]{}w, Poland\ and\ Centrum Fizyki Teoretycznej PAN,\ Al. Lotnik[ó]{}w 32/46, 02-668 Warszawa, Poland\ title: 'Invariant sets for discontinuous parabolic area-preserving torus maps' --- [Revised version]{} PACS: [05.45.+b]{} Keywords: [Parabolic maps on the torus, non-invertibility, measure of invariant set, Interval translation map]{} Introduction ============ We consider a class of maps of the torus $X=[0,1]^2$ of the form $$\label{eqmap} \begin{array}{l} x'=ax+by ~(\bmod ~1)\\ y'=cx+dy ~(\bmod ~1) \end{array}$$ where $(x,y)\in[0,1]^2$. This can be thought of as a map $f=g\circ M$ where $$M={\left(\begin{array}{cc} {a} & {b} \\ {c} & {d} \end{array}\right)}$$ (we also write $M=(a,b;c,d)$ for convenience) and $g(x)=x-\lfloor x\rfloor$ is a map that takes modulo $1$ in each component. Although such maps are linear except at the discontinuity induced by the map $g$, their dynamical behaviour can be quite complicated. Depending on the eigenvalues $\lambda_{1,2}$ of $M$ we refer to the map as elliptic ($\lambda_1=\overline{\lambda}_2\neq \lambda_1$), hyperbolic ($\lambda_1>\lambda_2$) or parabolic ($\lambda_1=\lambda_2$). In this paper we focus on the area-preserving parabolic case with determinant $ad-bc=1$ and trace $a+d=2$. Such maps arise naturally on examining linear maps with a periodic overflow. In particular, suppose that one would like to iterate a matrix using a digital representation with very small discretization error but a finite range which we set to be $(0,1)$. If the calculation overflows such that the fractional part remains, we will get the map (\[eqmap\]) (compare, for example, with [@ACP97]). In this case $f$ is a piecewise continuous map of ${{\bf T}}^2={{\bf R}}^2/{{\bf Z}}^2$ to itself that is area-preserving and such that the linearization is at almost every point in ${{\bf T}}^2$ is $M$. This matrix has two eigenvalues equal to one. The map is continuous everywhere on the torus except on a one-dimensional discontinuity $D$ in ${{\bf T}}^2$. The behaviour of $f$ on $D$ does not affect a full measure set of $X$. Parabolic area-preserving maps are not typical in the set of almost-everywhere linear maps on the torus [@Be]. However, their dynamical properties are of a particular interest, since such maps can be considered as an interpolating case between the hyperbolic maps ($\|t\|>2$) and the elliptic maps ($\|t\|<2$) in the area-preserving case. They are in some sense generalisations of interval exchange maps [@Ke75; @Ve78; @Ka80], piecewise rotations [@Go2; @ACP97] and interval translation maps to two dimensions. In fact, in Section \[secmaximal\] and the subsequent sections, our results use the fact that for rational coefficients of $M$ the map (\[eqmap\]) may be decomposed into a one-parameter family of 1 dimensional interval translation maps. Hyperbolic area-preserving maps are characterized by a positive Liapunov exponents, and are often studied as model chaotic area-preserving dynamical systems (see eg. [@Be; @Ott94] and the Baker’s transformation). Elliptic area-preserving maps correspond (in an appropriate eigenbasis) to rigid rotation, where the presence of a discontinuity caused by $g$ will lead to very complicated dynamics [@As96; @As97; @Go1; @ACP97]. The elliptic–hyperbolic transition for the linear maps on the torus was studied by Amadasi and Casartelli [@AC91], while certain properties of linear parabolic maps were analyzed in [@Zyc98]. In particular, a generic parabolic map displays some sort of sensitive dependence on initial conditions. However, this property is not related to the Lyapunov exponents (which are always zero for parabolic maps), but is due to the discontinuity of the map. Summary of main results ----------------------- We parametrize the set of parabolic area-preserving maps by $A$ and $\alpha$, namely $$\label{eqmapAalpha} \begin{array}{l} {\left(\begin{array}{c} {x'} \\ {y'} \end{array}\right)}=f{\left(\begin{array}{c} {x} \\ {y} \end{array}\right)}= {\left(\begin{array}{c} {(1+A)x+\frac{A}{\alpha}y ~(\bmod 1)} \\ {-\alpha Ax+(1-A)y~(\bmod 1)} \end{array}\right)} \end{array}.$$ Section \[secinvert\] characterises (in Theorem \[thmnoninvert\]) the open dense set of $(A,\alpha)$ such that the mapping (\[eqmapAalpha\]) is not invertible and discuss some properties of the invertible maps. Section \[secmaximal\] examines properties of the [maximal invariant set]{} [@Zyc98] $$X^+=\bigcap_{n=0}^{\infty} f^n(X)$$ which by the previous result is strictly smaller than $X$ for most $(A,\alpha)$. Note that any $A\subset X$ with $f(A)=A$ will have $A\subset X^+$ and so in this sense the set is maximal. We say a map (\[eqmapAalpha\]) is [semirational]{} if $\alpha$ is rational. It is [rational]{} if both $A$ and $\alpha$ are rational. If $f$ is not semirational we say it is [irrational]{}. Note that the map is rational if and only if $a$, $b$, $c$ and $d$ are rational. We investigate the two dimensional Lebesgue measure $\ell(\cdot)$ of this set, showing in Theorem \[thmposmeas\] that for rational parabolic maps we have $\ell(X^+)>0$. We give explicit lower and upper bounds for $\ell(X^+)$ depending on $A$ and $\alpha$. We discuss two particular examples of rational maps where we can compute $\ell(X^+)$, namely with $A=\frac{1}{2}$ and $\alpha=1$ or $\frac{1}{2}$. In the former case we show that $\ell(X^+)=\frac{1}{2}$ and in the latter case we show that $\ell(X^+)=\frac{1}{2}-\frac{\sqrt{3}\pi}{72}-\frac{1}{8}\ln 3\sim 0.28710$. In some special cases, e.g., $-1\leq \alpha<0,~ \|A\| \le 1 $ and $ A = 1, ~ 0< \alpha < 1$ , we can compute $X^+$, and thus we can obtain exact values of $\ell(X^+)$. Other than these cases we have not been able to compute exact values of $\ell(X^+)$ other than by numerical approximation, even for rational $A$ and $\alpha$. We suspect that for many $\alpha>0$ and $A>0$ (both irrational) then $\ell(X^+)=0$, whereas $\ell(X^+)>0$ for $\alpha$ rational. Section \[secnonlinexample\] briefly discusses examples of nonlinear parabolic maps, and we conclude with some remarks on open problems in Section \[secdiscuss\]. Invertibility of linear maps on the torus {#secinvert} ========================================= Consider a matrix $M$ corresponding to a parabolic map on the torus (\[eqmap\]). Apart from the special [horocyclic]{} [@CFS82] cases $${\left(\begin{array}{cc} {1} & {B} \\ {0} & {1} \end{array}\right)}~~~~{\left(\begin{array}{cc} {1} & {0} \\ {B} & {1} \end{array}\right)}, \label{horo}$$ any such matrix can be written in the form $$M=M_{A,\alpha}={\left(\begin{array}{cc} {1+A} & {A/\alpha} \\ {-\alpha A} & {1-A} \end{array}\right)}, \label{mapM}$$ where $A$ and $\alpha$ are real parameters. Observe that the cases $\alpha, A \rightarrow 0$ (with $A/\alpha$ held constant) and $\alpha\rightarrow\infty$, $A\rightarrow 0$ (with $\alpha A$ held constant) correspond to the cases (\[horo\]). Observe that $$M_{A,\alpha}^{-1}=M_{-A,\alpha},$$ and if $S(x,y)=(y,x)$ is transposition then $S^2=I$ and $$SM_{A,\alpha}=M_{-A,1/\alpha} S.$$ Moreover, in the absence of the rounding $g$ we have $$M_{A,\alpha}^p=M_{pA,\alpha}$$ for any $p\in{{\bf Z}}$. Noninvertibility of generic parabolic maps ------------------------------------------ The next lemma gives necessary and sufficient conditions that (\[eqmap\]) is invertible (ignoring points that land on the discontinuity). We say the map is invertible if there is a full measure subset on which it is invertible. Since the mapping is a composition of an invertible linear map and a discontinuous map that maps open sets to open sets, both of which preserve Lebesgue measure, a map is non-invertible if and only if there is an open set of $(x,y)$ that has two or more preimages. Let $a,b,c,d$ be defined as previously. \[lem1\] Any map (\[eqmap\]) with $ad-bc=1$ will be noninvertible on an open set if and only if there are $(K,L)\in{{\bf Z}}^2\setminus(0,0)$ integers such that $$\|Kc-La\|<1~\mbox{ and }~\|Kd-Lb\|<1.$$ [[Proof]{} ]{}Suppose that we have $(x,y)$ and $(u,v)\in[0,1)^2$ such that $(x,y)\neq (u,v)$ but $f(x,y)=f(u,v)$. Then there is $(K,L)\in {{\bf Z}}^2\setminus (0,0)$ such that $ax+by=au+bv+K$ and $cx+dy=cu+dv+L$. These hold if and only if $$a(x-u)+b(y-v)=K~\mbox{ and }~c(x-u)+d(y-v)=L.$$ Using the fact that $a,b,c,d$ are non-zero and $ad-bc=1$ this implies that this holds if and only if $$x-u=Kc-La~\mbox{ and }~y-v=Kd-Lb.$$ Thus the mapping is many-to-one if and only if there are $K$ and $L$ such that $$\|Kc-La\|<1~\mbox{ and }~\|Kd-Lb\|<1$$ and the result follows. [\ ]{} It is quite possible that there are simultaneously many solutions to the inequalities of Lemma \[lem1\]; for example, near $A=-1$ and $\alpha$ large one can find arbitrarily large numbers of integers $(K,L)$ satisfying both inequalities. Using the previous Lemma we obtain the main result of this section. \[thmnoninvert\] The only invertible parabolic maps $f$ have $M$ equal to one of $${\left(\begin{array}{cc} {k} & {-(k-1)^2/l} \\ {l} & {2-k} \end{array}\right)},~~ {\left(\begin{array}{cc} {2+l} & {-k} \\ {(l+1)^2/k} & {-l} \end{array}\right)},~~ {\left(\begin{array}{cc} {1} & {b} \\ {0} & {1} \end{array}\right)}~~\mbox{ or }~~ {\left(\begin{array}{cc} {1} & {0} \\ {b} & {1} \end{array}\right)},$$ where $(k,l)$ are integers and $b$ is real. [[Proof]{} ]{}The square $V_\epsilon=(-1-\epsilon,1+\epsilon)^2$ is a convex region symmetrical about the origin with area $4(1+\epsilon)^2$. A variant of Minkowski’s theorem says that for any $\epsilon>0$ there will be a point on the lattice $\Lambda=\{(Kc-La,Kd-Lb)~:~(K,L)\in{{\bf Z}}^2\}$ other than zero inside $V_\epsilon$, if the lattice has determinant $\Delta=ad-bc\geq 1$. Taking the limit $\epsilon\rightarrow 0$ and using compactness we see that there must be a point $(\kappa,\lambda)\in \Lambda$ other than zero in $\overline{V_0}$. If there is a lattice point $(\kappa,\lambda)$ in the interior of $V_0$, we are done. If not, assume without loss of generality that $(\kappa,\lambda)=(1,y)$ with $\|y\|\leq 1$. Thus exactly one point in the line segment from (but not including) $(-1,1-y)$ to $(0,1)$ that is in the lattice (one can generate the same lattice by adding multiples of $(1,y)$ to any lattice point generating the lattice with $(1,y)$). Therefore there will be a point in the lattice in the interior of $V_0$ unless this other point is $(0,1)$. Therefore (considering the other case by interchanging $x$ and $y$) there will be integers $K$ and $L$ such that either $$(Kc-La=1~~\mbox{ and}~~Kd-Lb=0)~~~\mbox{ or }~~~ (Kc-La=0~~\mbox{and}~~Kd-Lb=1).$$ Thus, in the invertible case the matrix $M$ is determined by which of these cases occurs. One can easily solve to show that either of these cases can occur. If $k=1$ or $l=-1$ then the other (horocyclic) cases occur. [\ ]{} It follows from this result that most parabolic maps are noninvertible in the following sense: The set of noninvertible maps is open and dense within the set of parabolic maps. Because the $(\bmod ~1)$ map $g$ commutes with integer matrix and $g^2 = g$, all conjugations of parabolic maps with integer coefficients by automorphisms of the torus ${{\bf T}}^2$ remain integer parabolic maps. For an integer parabolic map with rational $\alpha = r/s$, where $(r, ~s) = 1$, since there exist $m,~n \in {{\bf Z}}$ such that $mr+ns=1$, there exists an automorphism whose $GL(2,{{\bf Z}})$ matrix has bottom row $(r,~s)$. Conjugation by this automorphism gives a parabolic map whose matrix has bottom row $(0,~)$. Since conjugation preserves integrality, trace and determinant, the resulting matrix must be of the form $(1,~B;~0,~1)$ for some $B$. Therefore we conclude that all parabolic maps with integer coefficients can be reduced to one of the horocyclic cases. Maximal invariant sets {#secmaximal} ====================== By Poincaré recurrence the maximal invariant set $X^+=X$ (up to a set of zero measure) if and only if $f$ is invertible at almost every point. In fact the maximal invariant set $X^+$ is an upper semicontinuous function of the system parameters $A,\alpha$ in the Hausdorff metric. However, the measure (and dimension) of $X^+$ can (and does) change discontinously with parameters; see Proposition \[propsimplecases\]. We now consider the structure of $X^+$ for the semirational case (ie $\alpha$ rational). Consider the family of lines $L_B$ defined by $$y = B- \alpha x$$ with $B$ a fixed real number. projected onto the torus by taking modulo $1$ in $x$ and $y$. This can be thought of as the set $$L_B=\{ (x,y)~:~ y=B+K-\alpha (x+L)~~\mbox{with}~~(K,L)\in{{\bf Z}}^2\}\cap X.$$ Note that if $$y=B+K-\alpha (x+L)$$ with $L$ and $K$ integers such that $x,y$ in $[0,1]$ then $y'=-\alpha A x+ (1-A)(B+K-\alpha (x+L) +N$ and $x'=(1+A) x + A/\alpha (B+K-\alpha (x+L))+M$ with $M,N$ integers. Rearranging this we have $$y'= B+(K+M+N)-\alpha(x+L)$$ and so the family of lines $L_B$ is invariant under the map for any given $B$. We define the maximal invariant set within $L_B$ as $$X_B^+= X^+ \cap L_B.$$ Semirational and rational parabolic maps {#secrational} ---------------------------------------- For semirational maps ($\alpha=r/s>0$) and for any given value of $B$ the set $L_B$ consists of $r+s$ (or exceptionally $r+s-1$ if it contains the origin) intervals that are parallel to the eigenvector $v=(1,-\alpha)$ of the matrix $M$. We can parametrise any particular $L_B$ by $\theta\in[0,s)$. We will consider $\alpha>0$ from here on. More precisely we re-parametrize $X$ by $$\begin{array}{l} x = \theta - [\theta]\\ y = B- \alpha\theta -[B-\alpha\theta] \end{array}$$ For $\theta\in[0,s)$ and $B\in [\frac{s-1}{s},1)$ the mapping $(x,y) \leftrightarrow (\theta,B)$ is one-to-one and has unit Jacobian everywhere. Since each $L_B$ is invariant the maximal invariant set must nontrivially intersect $L_B$ and so no single trajectory is dense in $X^+$. In this case we need to approximate $X^+$ using a distribution of trajectories on a dense set of lines. The relation between $ \theta $ and $ x $ is $ x=\theta -[\theta ]. $ The variable $ y $ is related to $ \theta $ and $ B $ by the relation $ y=B-\alpha \theta -[B-\alpha \theta ] $. So the point $(\theta ,B)$ maps to $$(\theta',B')= (\theta +AB/\alpha -A[\theta ]-A/\alpha [B-\alpha \theta ],B)$$ since $$\begin{aligned} {\left(\begin{array}{c} {\theta'} \\ {v} \end{array}\right)} &=& \left( \begin{array}{cc} 1+A & A/\alpha \\ -A\alpha & 1-A \end{array}\right) \left( \begin{array}{c} \theta -[\theta ]\\ B-\alpha \theta -[B-\alpha \theta ] \end{array}\right) +\left( \begin{array}{c} [\theta ]\\ \left[ B-\alpha \theta \right] \end{array}\right) \\ &=& \left( \begin{array}{c} \theta +AB/\alpha -A[\theta ]-A/\alpha [B-\alpha \theta ]\\ -\alpha \theta +B-AB+A\alpha [\theta ]+A[B-\alpha \theta ] \end{array}\right). \end{aligned}$$ We have added integers $ [\theta ] $ and $ [B-\alpha \theta ] $ so that $\theta'\in[0,s)$ and $v=B-\alpha\theta$ (see eg [@Zyc98]). Thus the action of the map (\[eqmap\]) on the coordinate $\theta$ is simply $$\theta'=T_B(\theta)=\left(\theta+\frac{s}{r}AB-A[\theta]-A\frac{s}{r} [B-\frac{r}{s}\theta] \right) ~ (\bmod s). \label{map1d}$$ Because the variable $B\in [\frac{1-s}{s},1)$ is invariant during the dynamics it is treated as a parameter. Again, the slope of these 1D maps is equal to one implying that all Liapunov exponents are zero. In the case of a rational map, ie where both $A=p/q$ and $\alpha=r/s$ are rational with $(r,s)=(p,q)=1$ we can understand a lot about the mapping $T$ using the following factor map. Let $$\pi(\theta)= \theta-\frac{1}{qr}[qr\theta]$$ be a projection of $I=[0,s)$ onto $J=[0,\frac{1}{qr})$: this maps $qrs$ points onto one point. If $A=\frac{p}{q}$ and $\alpha=\frac{r}{s}$ then the diagram $$\begin{array}{rcl} I & \stackrel{T_B}{\rightarrow} & I \\ \pi \downarrow & & \downarrow \pi \\ J & \stackrel{S_B}{\rightarrow} & J \end{array}$$ commutes where $$S_B(\psi)=\psi+ \frac{sp}{rq}B ~ (\bmod \frac{1}{qr})$$ is a rotation. In other words $T_B$ has a factor that is a rotation. [[Proof]{} ]{}For these assumptions we can write (\[map1d\]) as $$\theta'=\left( \theta+\frac{sp}{rq}B-\frac{p}{q}[\theta] -\frac{ps}{qr}[B-\frac{r}{s}\theta]\right) ~(\bmod s). \label{map1drational}$$ Defining $\psi=\pi(\theta)$ and $\psi'=\pi(\theta')$, note that $$\psi'=\psi+ C ~(\bmod \frac{1}{qr})$$ where $C=\pi(AB/\alpha)=\pi(\frac{sp}{rq}B)$. [\ ]{} Since $S_B$ is invertible, $\cap_{n=0}^{\infty} S^n(J)=J$. The map $\pi$ maps $M=qrs$ points to one point and so $\pi^{-1}$ must be understood as a set-valued function. We have $$\pi\pi^{-1}(\psi)=\psi,~~ \pi T (\theta) = S \pi(\theta)$$ for all $\psi\in J$ and $\theta\in I$, while for any set $K\subset I$ $$\pi^{-1}\pi K \supseteq K.$$ Let $$\begin{aligned} N(\psi) &=& \{ \theta\in X^+_B~:~ \pi(\theta)=\psi\} = \pi^{-1}(\psi) \cap X_B^+\end{aligned}$$ be the set of $\theta\in X_B^+$ with $\pi(\theta)=\psi$ (we suppress the dependence of $S$ and $T$ on $B$ for the next result). For any $\psi\in J$, the set $N$ satsifies $$N( S(\psi)) = T ( N(\psi) ).$$ [[Proof]{} ]{}To show that $T(N(\psi))\subseteq N(S(\psi))$, note that $$\begin{aligned} T(N(\psi)) & \subseteq & \pi^{-1} \pi T(N(\psi)) = \pi^{-1} S \pi (N(\psi))\\ & = & \pi^{-1} S \pi (\pi^{-1} (\psi) \cap X_B^+) \\ & \subseteq & \pi^{-1} S(\psi)) \cap X_B^+ \\ & = & N(S(\psi)).\end{aligned}$$ For the other direction, suppose that $\theta\in N(S(\psi))$ and so $\pi(\theta)=S(\psi)$ and $\theta\in \cap_{n\geq 0} T^n(I)$. In particular, $\theta=T(\theta')$ for some $\theta'$. Now $\pi(\theta) = \pi T(\theta')= S \pi(\theta')$ and so $S(\psi)= S \pi(\theta')$. Invertibility of $S$ gives $\psi=\pi(\theta')$ and so $\theta'\in N(\psi)$. Hence $$T(N(\psi)) \supseteq N(S(\psi))$$ and we have the result. [\ ]{} The previous Lemma relies crucially on the fact that $S$ is invertible. The next result implies that the number of points in $N$ is a constant almost everywhere. Let $$\hat{N}=\|N(\psi)\|$$ be the cardinality of $N(\psi)$. If $B$ is irrational then $\hat{N}$ is constant for a set of $\psi$ with full Lebesgue measure. [[Proof]{} ]{}It is a standard result that $S_B$ is ergodic for Lebesgue measure if and only if $B$ is irrational. If we look at the set of $\psi$ that give a certain value of $N(\psi)$ this is invariant and therefore must have Lebesgue measure $0$ or $1$. [\ ]{} One consequence of this is that the measure of $X_B^+$ can only take a finite number of values. We write $\hat{N}(B)$ to show the dependence on $B$ explicitly. \[thmxb+meas\] Suppose that $A=p/q$ and $\alpha=r/s$. For Lebesgue almost all $B$ we have $$\ell(X_B^+) = \frac{\hat{N}(B)}{qr}$$ where $\hat{N}(B)$ is an integer and $1\leq \hat{N}(B)\leq qrs$. If there is an interval on which $T$ has $L$ preimages, then $$\hat{N}(B)\leq qrs+1-L.$$ [[Proof]{} ]{}As preimages of $\pi^{-1}(J)$ are disjoint and all have length $\frac{1}{qr}$ we compute $$\ell(X_B^+) = \sum_{k=1}^{\hat{N}(B)} \frac{1}{qr} = \frac{\hat{N}(B)}{qr}.$$ Note that $\hat{N}(B)$ must take an integer value less than or equal to $qrs$, being the number of preimages $\pi^{-1}(\psi)$. Since $T$ will be invertible on $X_B^+$, if there is an interval with more than one preimage, only one of these will be in $X_B^+$ and hence $\hat{N}(B)\leq qrs+1-L$ where $L$ is the number of preimages. [\ ]{} We can now prove the main result in this section. \[thmposmeas\] Suppose that $\alpha=r/s$ and $A=p/q$ are both rational, with $(r,s)=(p,q)=1$. Then the maximal invariant set $X^+$ has Hausdorff dimension $2$ and positive Lebesgue measure, more precisely, if there is an open set on which $f$ has $K$ preimages then $$\frac{1}{qrs}\leq \ell(X^+) \leq 1-\frac{K-1}{qrs}.$$ [[Proof]{} ]{}Note that under the hypotheses of the theorem, $$\begin{aligned} \ell(X^+) &=& \int_{B\in[1-\frac{1}{s})}\int_{s\in X_B^+} \,d\theta\,ds \\ &=& \int_{B\in[1-\frac{1}{s})}\ell(X_B^+)\,ds.\end{aligned}$$ Upon integrating the estimates in Theorem \[thmxb+meas\] we obtain the result. [\ ]{} We can characterise the dynamics on the lines $L_B$ in the following way: For any rational map, almost all $B$ and all $x\in L_B$, $\omega(x)$ consists of a set with positive one-dimension measure in $L_B$. Conversely, for a countable set of $B$ and all $x\in L_B$, the trajectory going through $x$ is eventually periodic. [[Proof]{} ]{}In the proof of Theorem \[thmposmeas\], the reduced map $S_B$ will be an irrational rotation for all irrational $B$ and a rational rotation for all rational $B$. The result follows. [\ ]{} Density of irrational $B$ simply implies the following result for the original map. For any rational parabolic map, the preimages of the discontinuity are dense in $X^+$. We do not have a precise analytical expression for $\ell(X^+)$ even for rational parabolic maps. Numerically one can approximate the measure and obtain various values of $\ell(X^+)$; see for example Table \[tablevals\]. In certain simple cases one can obtain exact values of $\ell(X^+)$ by constructing the maximal invariant set explicitly. \[propsimplecases\] One can compute $$\ell(X^+)= \left\{ \begin{array}{ll} 1 & \mbox{ if the map is invertible,}\\ \alpha & \mbox{ if $A=1$ and $0<\alpha\leq 1$,}\\ \alpha^{-1} & \mbox{ if $A=-1$ and $\alpha\geq 1$, and}\\ \|A\alpha\| & \mbox{ if $0<A\leq 1$ and $-1\leq \alpha <0$}. \end{array}\right.$$ The invertible case is trivial, while the cases where $A=1$ follow because the map reduces to an invertible map on a strip height $\alpha$. The appendix gives a constructive proof for the case $0<A\leq 1$, $-1\leq \alpha<0$. Because $f_{-A,-1}$ is conjugate to $f_{A,-1}$ by interchanging $x$ and $y$ we have $$\ell(X^+)=\left\{ \begin{array}{cc} \|A\|~, & ~~ 0<\|A\|\le 1~, \alpha=-1\\ 1~, & ~~ A=0 ~, \alpha=-1. \end{array}\right.$$ Note in particular that there is a discontinuity in $\ell(X^+)$ at $(\alpha,A)=(1,0)$. The following examples (especially that in Section \[secegii\]) show that a general expression, if it can be obtained, is likely to be non-trivial. $$\begin{array}{l\|llll} A & \alpha=2/3 & 1/2 & 1/3 & 1/5 \\ \hline 1/4 & 0.261 & 0.337 & 0.251 & 0.196 \\ 1/3 & 0.235 & 0.326 & 0.265 & 0.199 \\ 2/3 & 0.215 & 0.342 & 0.257 & 0.149 \\ 5/4 & 0.628 & 0.626 & 0.419 & 0.252 \\ 3/2 & 0.338 & 0.562 & 0.505 & 0.303 \end{array}$$ Example I --------- The simplest nontrivial rational map is defined by $A=1/2$ and $\alpha=1$. In this case, we can approximate the maximal invariant set numerically as in Figure \[figa1\_2alp1\]. The map in this case is $$\label{eqeg0} \begin{array}{l} x'=\frac{3}{2}x+\frac{1}{2}y ~(\bmod 1)\\ y'=-\frac{1}{2}x+\frac{1}{2}y~(\bmod 1) \end{array}$$ where $(x,y)\in X=[0,1]^2$. In this case, as noted in [@Zyc98], there appears to exist a symmetry between $X^+$ and its complement. In fact, we can use the results in Section \[secrational\] to get the result directly. For the map (\[eqeg0\]) with $A=\frac{1}{2}$ and $\alpha=1$, we have $\ell(X^+)=\frac{1}{2}$. [[Proof]{} ]{}Note that $r=s=p=1$ and $q=2$. Moreover, on $L_B$ the map (\[map1d\]) can be written $$\theta'=h(\theta)=\theta+\frac{B}{2} - \frac{1}{2}[B-\theta]~(\bmod 1)$$ and this map is two-to-one on an open set of $\theta$ for almost all $B$. Therefore, for almost all $B$, the upper and lower bounds of Theorem \[thmxb+meas\] agree and $$\ell(X_B^+)=\frac{1}{2}$$ for almost all $B$. This implies that $\ell(X^+)=\frac{1}{2}$. [\ ]{} Example II {#secegii} ---------- We now consider the case $A=\alpha=1/2$; this shows that $\ell(X^+)$ may be irrational for a rational map. In this case the map is $$\label{eqeg1} \begin{array}{l} x'=\frac{3}{2}x+y ~(\bmod 1)\\ y'=-\frac{1}{4}x+\frac{1}{2}y~(\bmod 1) \end{array}$$ On the invariant lines $L_B$ with $B\in[\frac{1}{2},1)$ (\[eqeg1\]) reduces to (\[map1d\]) with $\theta\in[0,2)$ to $$\theta' = T_B(\theta)=\left(\theta+B-\frac{1}{2}[\theta]- [B-2\theta] \right) ~ (\bmod 2). \label{pmap}$$ (6,6)(0,0) (0.5,0.53) (0.2,0.5) (0.45,0.25) (0.25,5.5) (5.5,0.25) (0.5,0.5)[(5,5)]{} (0.5,5.5)[(0,1)[0.25]{}]{} (5.5,0.5)[(1,0)[0.25]{}]{} (0.45,5.85) (5.8,0.45) \[fig:fig1\] The maximal invariant set $X^+$ for the map (\[eqeg1\]) is shown in Fig. 2. The results in Theorem \[thmposmeas\] imply that $$\frac{1}{4}\leq \ell(X^+)\leq \frac{3}{4}$$ but this is clearly not very precise. We have obtained a numerical estimate by iterating a grid of initial points distributed uniformly on the square, dividing the unit square into boxes and counting the occupied cells. To avoid the transient effects some number (say, the first hundred) of images are not marked on the graph. Applying this method we obtained $V\sim 0.287$, however the precision of this result is limited by a highly complicated structure of some fragments of the invariant set and a very slow convergence of transients. (6,6)(0,0) (0.5,0.53) (0.2,0.5) (0.45,0.25) (0.25,5.5) (5.5,0.25) (0.5,0.5)[(5,5)]{} (0.5,5.5)[(0,1)[0.25]{}]{} (5.5,0.5)[(1,0)[0.25]{}]{} (0.45,5.85) (5.8,0.45) \[fig:bif\] By using the structure noted in Section \[secmaximal\] we show numerical approximations of $X_B^+$ as $B$ changes in the map (\[pmap\]) are shown in Figure. \[fig:bif\] and notice that the bifurcation diagram of $X_B^+$ has a regular structure $$\begin{array}{cc} {1 \over 2} < B < {5 \over 6} & \ell(X^+_B) = {1 \over 2}\\ {5 \over 6} < B < {7 \over 8} & \ell(X^+_B) = 1\\ {7 \over 8} < B < {11 \over 12} & \ell(X^+_B) = {1 \over 2}\\ {11 \over 12} < B < {13 \over 14} & \ell(X^+_B) = 1\\ {13 \over 14} < B < {17 \over 18} & \ell(X^+_B) = {1\over 2}\\ \end{array}$$ These values agree with the predictions of Theorem \[thmposmeas\] that $\ell(X^+_B)\in \frac{1}{4}\{1,2,3\}$. More generally for any $n \in {{\bf N}}$ we observe that $$\begin{array}{ll} {6n-1 \over 6n} < B < {6n+1 \over 6n+2} & \ell(X^+_B) = {1}\\ {\rm else} & \ell(X^+_B) = {1 \over 2}.\\ \end{array}$$ Therefore we have $$\begin{aligned} \ell(X^+) &=& \int_{1/2}^1 \ell(X_B^+) dB \\ &=& \frac{1}{4}+\frac{1}{2}\sum_{n=1}^\infty \left({6n+1 \over 6n+2} - {6n-1 \over 6n}\right) \\ &=& {1 \over 4} + {1 \over 12} \sum_{n=1}^\infty {1 \over 3n^2+n} \\ &=& {1 \over 2} - {\sqrt{3}\pi \over 72} - {1 \over 8} \ln 3 \\ &\approx& 0.28709849.\end{aligned}$$ Thus we can obtain a series expression for the measure and in this case we can sum it explicitly. Unfortunately we have not been able to extend this method to one that applies generally. Irrational parabolic maps ------------------------- We have not yet obtained any significant results for irrational parabolic area-preserving maps and, in fact, for the all cases where we can compute it, the maximal invariant set has positive measure. Nonetheless, there is numerical evidence ([@Zyc98]) that suggests that - There are $(\alpha,A)$ for which the Hausdorff dimension of $X^+$ is less than 2 (in particular, such that $\ell(X^+)=0$). By results here, this can only occur in cases where both parameters are irrational. - This is common for $A>0$ and $\alpha>0$. Note that not all irrational parabolic maps have zero measure; for example Proposition \[propsimplecases\] shows that there are open regions in $(A,\alpha)$ with $\alpha<0$ where $\ell(X^+)>0$. For semirational cases ($\alpha=r/s$) we can reduce to maps of the form (\[map1d\]), however we cannot find factors that are rotations if $A$ is not rational. Results of suggest that the Hausdorff dimension of $X^+$ may vary between $1$ and $2$. The Hausdorff dimension is a rather subtle characteristic of a set. For example, two sets may be arbitrarily close in the sense of the Hausdorff metric, but their dimensions may differ by an arbitrarily large number. Therefore, small changes of the parameter $\alpha$, which do not change much the structure of the maximal invariant set (in sense of the Hausdorff metric) may induce wild fluctuations of its dimension. Note also that the cases $\alpha>0$ and $\alpha<0$ seem to be fundamentally different; we do not have a way of conjugating one with the other. Nonlinear parabolic maps {#secnonlinexample} ======================== One can also construct nonlinear parabolic maps that display similar structure in their maximal invariant sets as the piecewise linear case. Consider the two parameter family of the maps on the torus defined by $$\label{nonlin} \begin{array}{l} x'=x+ \beta(e^{\gamma (y-x)} -1)~~(\bmod ~1)\\ y'=y+ \beta(e^{\gamma (y-x)} -1)~~(\bmod ~1)\\ \end{array}$$ For any non-zero values of the real parameters $\beta$ and $\gamma$ the Jacobian $\partial(x',y')/\partial(x,y)$ is constant and equal to one. Moreover, for any point $(x,y)$ the trace of the map $t$ is equal to $2$, which makes the above map similar to the linear case (\[eqmap\]). Observe that the diagonal of the unit square $y=x$ is invariant with respect to this map. This map acting on the plane transforms the unit square into a set confined between four exponential functions. After folding it back into the unit square some fragments of it will overlap as (\[nonlin\]) is typically noninvertible. (6,6)(0,0) (0.5,0.5) (0.2,0.5) (0.45,0.25) (0.25,5.5) (5.5,0.25) (0.5,0.5)[(5.0,5.0)]{} (0.5,5.5)[(0,1)[0.25]{}]{} (5.5,0.5)[(1,0)[0.25]{}]{} (0.45,5.85) (5.8,0.45) \[fignonlin\] Figure \[fignonlin\] shows the invariant set of the nonlinear map in the case $\beta=\gamma=1$. There is numerical evidence that the maximal invariant set has positive volume in this case even though the Jacobian changes between being a rational and an irrational parabolic map at different points. Discussion {#secdiscuss} ========== In this paper we have considered some basic properties of the maps (\[eqmap\]) in the parabolic area-preserving regime, and show a surprising degree of sensitivity of the dynamical behaviour on rationality of the parameters $\alpha$ and $A$. Our investigations raise some intriguing questions concerning how the Lebesgue measure of the maximal invariant set varies with parameters. For several examples do we have analytical values of this measure. For general rational parabolic maps we do have a result that gives upper and lower bounds for this measure. However, these bounds become weak as one examines higher denominator rational mapsand do not easily give insight into the measure for irrational parabolic maps. It would be very informative to understand the structure of invariant measures of the map (\[eqmap\]). Note that $\ell(.)$ (Lebesgue measure) restricted to $X^+$ is invariant under this mapping but it is not ergodic. In particular, it may be the case that this restriction is trivial in which case empirical measures can still be defined; by analogy with the Interval Translation Maps of we surmise that there are cases where a Hausdorff measure restricted to $X^+$ is invariant, in particular in cases where $\ell(X^+)=0$. Acknowledgements {#acknowledgements .unnumbered} ================ The research of PA and XF is supported by EPSRC grant GR/M36335, while K[Ż]{} acknowledges the support by a KBN Grant. PA, TN and K[Ż]{} thank the Max-Planck-Institute for the Physics of Complex Systems, Dresden for providing an opportunity to meet and discuss this paper. We also thank the anonymous referees for their invaluable comments, in particular to one who pointed out a considerable simplification of our argument in Section 2. Appendix: Proof of Proposition \[propsimplecases\] {#secexample3 .unnumbered} ================================================== We consider the case $-1\leq \alpha<0$ and $0<A<1$; figure \[fig\_paraconst\] shows how the linear map $M_{A, \alpha}$ maps the square $[0,1]^2$ given by $OPRQ$ into the maximal invariant set consisting of the union of the line $OS$ and the triangles $OAB$ and $CSQ$, where $$A=(0,-\alpha),~~B=(A,-\alpha A),~~C=(1-A,-\alpha(1-A))~~\mbox{ and }~~ S=(1.-\alpha).$$ To see this is the maximal invariant set, note that everything in $OPRS$ decreases in its $y$- component unless it lands in the triangle $OAB$. Similarly, all points in $OQS$ must increase in $y$-component unless it lands in the triangle $CQS$. The union of the two triangles is invariant as the dotted images $ODB$ and $CSQ$ show, as is the line of fixed points $OS$. Hence in this case we have $$\ell(X^+)=-\alpha A= \|\alpha A\|.$$ A similar argument (with extra triangles) can be used to show that for $0<A \leq 1$ and $$-2 < \frac{-2}{1+A} \leq \alpha<-1$$ the maximal invariant set has measure $$\ell(X^+)=-\frac{\alpha A}{2},$$ but we omit this for conciseness. [99]{} L. Amadasi and M. Casartelli, J. Stat. Phys. [65]{}, 363 (1991). P. Ashwin, Non–smooth invariant circles in digital overflow oscillations, Proceedings of workshop on nonlinear Dynamics of Electronic Systems, Sevilla 1996. P. Ashwin, [Phys. Lett.]{} [A 232]{}, 409 (1997). P. Ashwin, W. Chambers and G. Petkov, [Intl. J. Bifn. Chaos]{} [7]{}, 2603 (1997). M.V. Berry, in [Hamiltonian Dynamical Systems]{}, ed. R.S. MacKay and J.D. Meiss, Adam Hilger, Bristol 1987. M. Boshernitzan and I. Kornfeld, [Ergod. Th. & Dynam. Sys.]{} [15]{}, 821 (1995). I.P. Cornfeld, S.V. Fomin and Ya. G. Sinai, [Ergodic Theory]{}, Springer-Verlag, New York 1982. A. Goetz, [Discrete Cont. Dyn. Sys.]{} [4]{}, 593 (1998). A. Goetz, [Int. J. Bif. Chaos]{} [8]{}, 1937 (1998). G.H. Hardy and E.M. Wright Clarendon Press, Oxford (1979). A. Katok, [Israel J. Math.]{} [35]{}, 301 (1980). A. Katok and B. Hasselblat, [Introduction to Modern Theory of Dynamical Systems]{}, Cambridge University Press, Cambridge 1995. M. Keane, [Mathemat. Zeit.]{} [141]{}, 25 (1975). E. Ott, [Chaos in Dynamical Systems]{}, Cambridge University Press, Cambridge, 1994. J. Schmeling and S Troubetzkoy, Interval Translation Mappings, In [Dynamical Systems, from crystals to chaos, Marseille 1998]{}, World Scientific Publishers (to appear). W. Veech, [J. D’Analyse Math.]{} [33]{}, 222 (1978). K. [Ż]{}yczkowski and T. Nishikawa [Linear Parabolic Maps on the Torus]{} [Physics Letters]{} [A 259*]{}, 377 (1999).	{ "pile_set_name": "ArXiv" }
--- abstract: 'Using geometric methods and without invoking deep results from group theory, we prove that a classical unital of even order $n\geq4$ is characterized by two conditions (I) and (II): (I) is the absence of O’Nan configurations of four distinct lines intersecting in exactly six distinct points; (II) is a notion of parallelism. This was previously proven by Wilbrink (1983), where the proof depends on the classification of finite groups with a split BN-pair of rank 1.' author: - 'Alice M. W. HUI' title: 'A geometric proof of Wilbrink’s characterization of even order classical unitals' --- Keywords: unital; classical unital; Hermitian curve; spread\ MSC(2000): 51E20, 05B25, 05B25, 51E21, 51E23\ Introduction {#st 1} ============ A unital of order $n>2$ is a design with parameters $2$–$(n^3 + 1, n + 1, 1)$ (see [@Dem2; @BE]). If $\pi$ is a projective plane of order $m$, i.e. a $2$–$(m^2 + m + 1, m + 1, 1)$ design, and if a unital $U$ is an induced substructure of $\pi$, then we call $U$ an [embedded unital]{}. Some embedded unitals $U$ of order $n$ are the incidence structure formed from the absolute points and non-absolute lines of a unitary polarity in a projective plane $\pi$ of order $n^2$. Any unital which is isomorphic to such a unital $U$ as a design is called a [polar unital]{}. Further if the ambient plane is ${\mbox{\rm PG}}(2,n^2)$, then the unital is called [classical]{}. The set of absolute points of a unitary polarity in ${\mbox{\rm PG}}(2,n^2)$ is called the [Hermitian curve]{} (see [@Hir1; @HP]). In 1972 [@ON], O’Nan showed that the classical unital does not contain a configuration of four lines meeting in six points (an [O’Nan configuration]{}). Piper (1981) [@Pi] conjectured that this property characterizes the classical unital. Wilbrink (1983) [@Wil] characterized the classical unital by three conditions (I), (II) and (III). His proof depends on a result in the classification of finite groups with a split BN-pair of rank 1. Wilbrink [@Wil] further proved that when the order of unital is even, (III) is a necessary condition of (I) and (II). Let ${{\mathcal U}}$ be a unital of even order $n \geq 4$ satisfying Wilbrink’s conditions (I) and (II). In this article, we give an alternative proof that ${{\mathcal U}}$ is classical without invoking deep results from group theory, as follows. We construct from ${{\mathcal U}}$ a hyperbolic Buekenhout unital ${{\mathcal U}}'$ [@Bkt] in ${\mbox{\rm PG}}(2, n^2)$, via the Bruck-Bose construction of projective plane [@BB1; @BB2]. Then we prove that ${{\mathcal U}}$ is isomorphic to ${{\mathcal U}}'$, and hence is classical by a result of Barwick [@Bar]. To construct ${{\mathcal U}}'$, we shall consider some inversive planes and a generalized quadrangle derived from ${{\mathcal U}}$ (Wilbrink [@Wil]), and the special spreads of ${{\mathcal U}}$ (Hui and Wong [@HW2]). We also need a theorem of Cameron and Knarr [@CK] on how to build a regular spread of ${\mbox{\rm PG}}(3,q)$ from a tube in ${\mbox{\rm PG}}(3,q)$, and a theorem of Hui [@H] on when two inversive planes are identical. In Section \[st 2\], we follow Wilbrink’s [@Wil] construction of the inversive planes ${{\mathcal I}}(x)$ at each point $x$ of ${{\mathcal U}}$. Then following the work in [@HW2], we construct a special spread $\mathcal S_L$ for each line $L$ of ${{\mathcal U}}$ using these inversive planes. As a consequence, ${{\mathcal U}}$ can be embedded in a projective plane $\pi$ as a polar unital [@HW2]. This enables us to define self-polar triangles with respect to ${{\mathcal U}}$ intrinsically in terms of Wilbrink’s $x$-parallelism (Theorem \[thm LMN\]). In Section \[st 3\], by studying the inversive planes ${{\mathcal I}}(x)$ for various $x$’s, we prove that the set of points of ${{\mathcal U}}$ is partitioned into a self-polar triangle, and $n - 2$ subsets of $(n+1)^2$ points triply ruled by lines through the vertices of the triangle (Theorem \[thm bigP\]). This describes how unital lines in $\pi$ through distinct non-unital points intersect. In Section \[st 4\], we fix one line $L$ of ${{\mathcal U}}$ and consider the generalized quadrangle $GQ(L)$ as in Wilbrink [@Wil]. Through $GQ(L)$, we associate ${{\mathcal U}}$ with $Q(4,n)$ formed by the set of points and lines of a parabolic quadric ${{\mathcal P}}$ in ${\mbox{\rm PG}}(4,n)$ [@PT]. Wilbrink’s construction gives naturally a 3-dimensional subspace $\Sigma$ of ${\mbox{\rm PG}}(4,n)$. We find a spread ${{\mathscr S}}$ in $\Sigma$ by studying the special spread $\mathcal S_L$. We then prove that ${{\mathscr S}}$ is regular (Theorem \[thm regular\]) using a result on tubes in ${\mbox{\rm PG}}(3,n)$ (Cameron and Knarr [@CK]) and properties of self-polar triangles with respect to ${{\mathcal U}}$. In Section \[st 5\], we prove that the partition of ${{\mathcal U}}$ into a self-polar triangle and triply ruled sets corresponds to a pencil of quadrics in $\Sigma$ of two lines and $n - 1$ hyperbolic quadrics (Theorem \[thm pencil\]). In particular, this gives a correspondence between the structure of ${{\mathcal U}}$ and that of $\Sigma$. The regularity of ${{\mathscr S}}$ is essential for proving Theorem \[thm pencil\], because we have to describe the reguli of ${{\mathscr S}}$ in terms by geometry of ${{\mathcal U}}$ by applying a result of Hui [@HW2] to the Miquelian inversive plane formed by the lines and reguli of ${{\mathscr S}}$ (Bruck [@Bru]). In Section \[st 6\], by considering the spread ${{\mathscr S}}$ of $\Sigma$ in ${\mbox{\rm PG}}(4,n)$, we construct a projective plane $\overline{\pi({{\mathscr S}})}$ by the Bruck-Bose construction [@BB1]. Since ${{\mathscr S}}$ is regular, $\pi({{\mathscr S}})$ is ${\mbox{\rm PG}}(2,n^2)$ [@BB2]. By Buekenhout [@Bkt], ${{\mathcal P}}$ defines a hyperbolic Buekenhout unital ${{\mathcal U}}'$ in $\overline{ \pi({{\mathscr S}})}$ [@Bkt] (also known as a nonsingular Buekenhout unital). By Barwick [@Bar], since $\overline{\pi({{\mathscr S}})}\cong{\mbox{\rm PG}}(2,n^2)$, ${{\mathcal U}}'$ is the classical unital. With the help of Theorem \[thm pencil\], we write down an isomorphism between ${{\mathcal U}}$ and the classical unital ${{\mathcal U}}'$ (Theorem \[thm main\]). Self-polar triangles and Parallelism {#st 2} ==================================== Let ${{\mathcal U}}$ be a unital of even order $n \geq 4$, satisfying Wilbrink’s first two conditions (I) and (II) [@Wil]: 1. ${{\mathcal U}}$ contains no O’Nan configurations. 2. Let $x$ be a point, $L$ be a line through $x$, and $M$ be a line missing $x$, such that $L$ and $M$ meets. For any point $y'\in L \setminus \{x\}$, there is a line $M'$ through $y'$ but not $x$ meeting all lines from $x$ which meet $M$. Following Wilbrink [@Wil], we introduce [$x$-parallelism]{} in ${{\mathcal U}}$ [@Wil]: - Let $x$ be a point, and $M,M'$ be two lines missing $x$. $M,M'$ are said to be $x$-parallel if $M,M'$ intersect the same lines through $x$. We write $M\\|_x M'$. $\\|_x$ defines an equivalence relation on the set of all lines missing $x$ [@Wil]. We denote the equivalence class of a line $M$ under $\\|_x$ by ${\overline{M} }^x$, or simply ${\overline{M} }$ if there is no confusion. Further following Wilbrink [@Wil], we introduce an inversive plane ${{\mathcal I}}(x)$ of order $n$ for every point $x$ in ${{\mathcal U}}$ ([@Wil], Lemmas 1, 2 and Corollary 3; see also [@HW2]). The points of ${{\mathcal I}}(x)$ are the lines of ${{\mathcal U}}$ through $x$ together with a symbol $\infty_x$. The circle set of ${{\mathcal I}}(x)$ is $\mathcal C^x \cup \mathcal C_x$. Here $\mathcal C^x$, $ \mathcal C_x$ are given by: $\mathcal C^x$ is the set of $\\|_x$-equivalence classes on the set of lines of ${{\mathcal U}}$ missing $x$; $\mathcal C_x$ consists of blocks of the form $C_x(L, L') \cup \{\infty_x\}$, where for any lines $L,L'$ on $x$, $C_x(L,L')$ = $\{L,L'\} \cup \{L'' \mid L''$ is a line through $x$ such that no line of through $x$ meets $L, L'$ and $L''\}$. The incidence in ${{\mathcal I}}(x)$ is defined as follows: Whenever $L$ is a line of ${{\mathcal U}}$ through $x$ and $M$ is a line of ${{\mathcal U}}$ missing $x$, $L$ is incident with ${\overline{M} }$ in ${{\mathcal I}}(x)$ if and only if $L$ meets $M$ in ${{\mathcal U}}$; whenever $L, L', L''$ are lines of ${{\mathcal U}}$ through $x$, $L$ is incident with $C_x(L', L'')$ in ${{\mathcal I}}(x)$ if and only if $L \in C_x (L',L'')$; the point $\infty_x$ is incident with all circles in ${{\mathcal C}}^x$ but none in ${{\mathcal C}}_x$. According to Dembowski [@Dem1], ${{\mathcal I}}(x)$ is egglike. By Thas [@Thas1], every flock in ${{\mathcal I}}(x)$ is uniquely determined by its carriers. We denote the unique flock in ${{\mathcal I}}(x)$ with carriers $p_1$ and $p_2$ by ${{\mathcal F}}(p_1,p_2)$. With the help of flocks of the form ${{\mathcal F}}(L,\infty_x)$, ${{\mathcal U}}$ can be shown to satisfy condition ($P$) [@HW2 Theorem 1.6], formulated in terms of [special spreads]{} in ${{\mathcal U}}$. Let ${{\mathcal S}}$ be a spread of ${{\mathcal U}}$, with $L \in {{\mathcal S}}$. Then ${{\mathcal S}}$ is [special with respect to $L$]{} if the following condition is satisfied: - for any point $x$ on $L$, ${{\mathcal S}}\setminus \{ L\}$ can be partitioned into $n - 1$ subsets ${{\mathcal L}}_1^x, \cdots, {{\mathcal L}}_{n - 1}^x$, each of cardinality $n$, and the set of lines on $x$, except $L$, can be partitioned into $n - 1$ subsets ${{\mathcal K}}_x^1, \cdots, {{\mathcal K}}_x^{n - 1}$, each of cardinality $n + 1$, such that whenever $L' \in {{\mathcal L}}_i^x$ and $K \in {{\mathcal K}}_x^j$, $L'$ and $K$ intersect if and only if $i = j$. Now, [@HW2 Theorem 1.6] says that ${{\mathcal U}}$ satisfies condition ($P$), which is a strengthened version of condition ($p$): 1. Let $L_1, L_2, \cdots, L_{n^4-n^3 + n^2}$ be the lines of ${{\mathcal U}}$. There exists a family of lines $\mathscr F = \{ {{\mathcal S}}_{L_1}$, ${{\mathcal S}}_{L_2}$, $\cdots$, ${{\mathcal S}}_{L_{n^4-n^3 + n^2}} \}$ such that: 1. For $i = 1, 2, \cdots, n^4-n^3 + n^2$, ${{\mathcal S}}_{L_i}$ is a spread containing $L_i$. 2. For $i\neq j$, $L_i \in {{\mathcal S}}_{L_j} \setminus \{L_j\}$ if and only if $L_j \in {{\mathcal S}}_{L_i} \setminus \{L_i\}$. 3. For any two lines $L_i$ and $L_j$ missing each other, there exists a line $L_k$ such that $ L_k \in {{\mathcal S}}_{L_i} \setminus \{L_i\}$ and $ L_k \in {{\mathcal S}}_{L_j} \setminus \{L_j\}$. 2. ($p$) holds such that for $i = 1,2,\cdots,n^4-n^3 + n^2$, ${{\mathcal S}}_{L_i}$ is a special spread with respect to $L_i$. In the above statement, the set ${{\mathcal S}}_L$ is given explicitly by this construction: - Pick some point $x$ on $L$. ${{\mathcal S}}_L$ is given by ${\overline{L_1} }$ $\cup {\overline{L_2} }$ $\cup \cdots$ $\cup {\overline{L_{n - 1}} }$ $\cup \{L\}$, where ${\overline{L_1} }$, ${\overline{L_2} }$, $\cdots$, ${\overline{L_{n - 1}} }$ are the circles of ${{\mathcal F}}(L, \infty_{x})$ in ${{\mathcal I}}(x)$. ${{\mathcal S}}_L$ is shown to be independent of $x$ [@HW2 Lemma 5.4]. For each line $L$ of ${{\mathcal U}}$, denote by ${\mathcal S^{}_{L} }$ the set ${{\mathcal S}}_L \setminus \{L\}$. By [@HW2 Theorem 1.1], since ${{\mathcal U}}$ satisfies ($p$), ${{\mathcal U}}$ can be embedded in a projective plane $\pi$ as a polar unital, so that in $\pi$, for each unital line $J$, the unital lines of $\pi$ through the pole of $J$ are exactly the lines in ${\mathcal S^{}_{J} }$. Since two distinct points in $\pi$ determine a unique line, ${\mathcal S^{}_{J} }\cap {\mathcal S^{}_{J'} }$ contains at most one (unital) line for any distinct unital lines $J,J'$. Thus there are at most three lines in ${{\mathcal S}}_{J}\cap {{\mathcal S}}_{J'}$. Whenever $L,M,N$ are three distinct lines of ${{\mathcal U}}$ satisfying $\{L,M,N\}={{\mathcal S}}_{L}\cap {{\mathcal S}}_{M} = {{\mathcal S}}_{L}\cap {{\mathcal S}}_{N} ={{\mathcal S}}_{M}\cap {{\mathcal S}}_{N}$, we say that $L,M,N$ form a [self-polar triangle with respect to ${{\mathcal U}}$]{}. \[lemma polartri\] Let $L,M$ be disjoint lines of ${{\mathcal U}}$. If $M\in {\mathcal S^{}_{L} }$ (or equivalently $L \in {\mathcal S^{}_{M} }$), then there exists a unique line $N$ of ${{\mathcal U}}$ such that $L,M,N$ form a self-polar triangle with respect to ${{\mathcal U}}$. In $\pi$, $L$ and $M$ meet in a unique non-unital point $a$. Let $N$ be the polar line of $a$. By construction of $\pi$, $L,M\in {\mathcal S^{}_{N} }$ and $N\neq L,M$. Since ${{\mathcal U}}$ satisfies ($p$), $N\in {{\mathcal S}}_{L} \cap {{\mathcal S}}_{M}$ by ($p$)(ii). By ($p$)(i), $L \in {{\mathcal S}}_{L}$, $M \in {{\mathcal S}}_{M}$ and $N \in {{\mathcal S}}_{N}$. The result follows from the fact the there are at most three lines in ${{\mathcal S}}_{L}\cap {{\mathcal S}}_{M}$, ${{\mathcal S}}_{L}\cap {{\mathcal S}}_{N}$ and ${{\mathcal S}}_{M}\cap {{\mathcal S}}_{N}$. \[thm LMN\] Let $L,M,N$ be disjoint lines of ${{\mathcal U}}$. Then the following are equivalent. 1. \[item LMN1\]$L,M,N$ form a self-polar triangle with respect to ${{\mathcal U}}$. 2. \[item LMN2\]$L \\|_z M$ for any point $z \in N$. 3. \[item LMN4\] Any line meeting two of $L,M,N$ meet all of $L,M,N$. 4. \[item LMN3\]$M \in {\mathcal S^{}_{L} }$, and there exist distinct points $z_1,z_2 \in N$ such that $L \\|_{z_1} M$ and $L \\|_{z_2} M$. Since lines in a self-polar triangle play the same role, in Theorem \[thm LMN\], statement is also equivalent to $'$: $M \\|_z N$ for any point $z \in L$, or other statement obtained by permuting $L,M,N$ in and . Let $x$ be a point on $L$, and $K_1, K_2, \cdots, K_{n + 1}$ be the lines through $x$ meeting $M$. $\Rightarrow$ : Since $M\in {\mathcal S^{}_{L} }$, there is a point $y_i \in K_i$ such that $L \\|_{y_i} M$ for $i = 1, 2, \cdots, n + 1$, by Lemmas 5.3 and 5.4 of [@HW2]. Hence, for each $i$, ${\overline{L} } = {\overline{M} }$ in ${{\mathcal I}}(y_i)$. Then there is a unique point $N_i$ such that ${\overline{L} } \in {{\mathcal F}}(N_i,\infty_{y_i})$. This $N_i$ is a line of ${{\mathcal U}}$ through $y_i$ such that $L\in {\mathcal S^{}_{N_i} } $ and $M \in {\mathcal S^{}_{N_i} }$. By condition ($p$), $N_i \in{\mathcal S^{}_{L} } \cap {\mathcal S^{}_{M} }$. Since there is at least one line in ${\mathcal S^{}_{L} } \cap {\mathcal S^{}_{M} }$, we have $N_1 = N_2 = \cdots = N_{n + 1}=N$. $\Rightarrow$ : For each point $z\in N$, since $L \\|_z M$, there are $n+1$ lines through $z$ meeting $L$ and $M$. Thus there are $(n+1)^2$ lines meeting $L,M,N$. They are the lines meeting at least two of $L,M,N$. $\Rightarrow$ follows from the definition of $z$-parallelism. $\Rightarrow$ : It suffices to show $M \in {\mathcal S^{}_{L} }$. Let $z'\in N$. By , any line through $z'$ meeting $M$ meets $L$. Thus, $L\\|_{z'} M$. By , $K_i$ meets $N$ for $i=1,2,\cdots,n+1$. Thus $z'$ is the point on $K_i$ such that $L\\|_{z'} M$. By Lemmas 5.3 and 5.4 of [@HW2], $M \in {\mathcal S^{}_{L} }$. $\Rightarrow$ : Let $N'$ be a line such that $L,M,N'$ form a self-polar triangle. Suppose $z_1\notin N'$. Let $x_1,x_2 \in L$ be distinct points. For $i = 1,2$, let $J_i$ be the line passing through $x_i$ and $z_1$. Then $J_i$ meets $M$ by . Since implies , $J_1$ meets $N'$ at a point, say $w$. Since implies , we have $L \\|_w M$. Thus, the four lines $J_1$, $J_2$, $M$, $w.x_2$ form an O’Nan configuration, which is a contradiction. ($w.x_2$ denotes the line through $w$ and $x_2$.) Hence $z_1 \in N'$. Similarly, $z_2 \in N'$. So $N=N'$. With Theorem \[thm LMN\], we characterize ${\mathcal S^{}_{M} }$ by $z$-parallelism. \[lemma LJzparallel\] Let $L,M$ be distinct lines of ${{\mathcal U}}$. Suppose $M\in {\mathcal S^{}_{L} }$. Then $${\mathcal S^{}_{M} }=\{J \mid J \mbox{ is a line of ${{\mathcal U}}$ such that there is a point $y\in M$ such that }L \\|_y J\}.$$ Let $J$ be a line. Suppose there is a point $y\in M$ such that $L \\|_y J$. Since $L \in {\mathcal S^{}_{M} }$ by condition ($p$) and $L \\|_y J$, we have ${\overline{J} } = {\overline{L} } \in {{\mathcal F}}(M,\infty_y)$ in ${{\mathcal I}}(y)$. Thus, $J \in {\mathcal S^{}_{M} }$ by the construction of ${\mathcal S^{}_{M} }$. To prove the reverse inclusion, it suffices to show that there are $\|{\mathcal S^{}_{M} }\| = n^2 - n $ $J$’s such that $L \\| _z J$ for some $z \in M$. Let $N$ be the line such that $L,M,N$ form a self-polar triangle. For each point $z$ on $M$, there are exactly $n$ lines $z$-parallel to $L$. By Theorem \[thm LMN\], $L$ and $N$ are two of these $n$ lines. Apart from $L$ and $N$, no line is $z$-parallel to $L$ for distinct $z$’s on $M$ by Theorem \[thm LMN\]. Hence, there are $2 + (n + 1) (n - 2) = n^2 - n$ lines $z$-parallel to $L$ for some $z$ on $M$, as desired. Partition ${{\mathcal U}}$ into a self-polar triangle and triply ruled sets {#st 3} =========================================================================== In this section, we prove that ${{\mathcal U}}$ can be partitioned into a self-polar triangle and $n-2$ triply ruled sets of $(n+1)^2$ points (Theorem \[thm bigP\]). This result will be used at the end of Section \[st 5\], where we relate the partition to a pencil of quadrics in a projective space. In Dover [@Dov Theorem 3.2] (see also Section 4 of Baker et al. [@BEKS]), it is proved that any classical unital admits such a partition by considering coordinates and its automorphism group. The argument in this section will yield a synthetic proof for Dover’s result once we prove that ${{\mathcal U}}$ is classical in Section \[st 6\]. We describe how lines of ${{\mathcal S}}_L$ correspond to circles tangent to ${\overline{L} }$ in ${{\mathcal I}}(y)$, whenever $y$ is not a point on $L$. \[lemma SLinIy\] Let $L$ be a line of ${{\mathcal U}}$, and $y$ be a point not on $L$. Let $M \in {{\mathcal S}}_L$ be the line of ${{\mathcal U}}$ through $y$. Let $N$ be the line of ${{\mathcal U}}$ such that $L,M,N$ form a self-polar triangle. Then the following statements hold: 1. \[itemSLinIy1\] For every $J \in {{\mathcal S}}_L \setminus \{L,M,N\}$, ${\overline{J} }$ is tangent to ${\overline{L} }$ in ${{\mathcal I}}(y)$. 2. \[itemSLinIy2\] Every circle of type ${{\mathcal C}}^y$ tangent to ${\overline{L} }$ in ${{\mathcal I}}(y)$ is ${\overline{J'} }$ for a unique line $J'$ in ${{\mathcal S}}_L \setminus \{L,M,N\}$. \(1) Let $J \in {{\mathcal S}}_L \setminus \{L,M,N\}$. It suffices to show that there is a unique line of ${{\mathcal U}}$ through $y$ meeting both $J$ and $L$. By Lemma \[lemma LJzparallel\], since $L \in {\mathcal S^{}_{M} }$ and $J\in {\mathcal S^{}_{L} }$, we have $M\\|_x J$ for some point $x \in L$. Let $K$ be the line of ${{\mathcal U}}$ through $x$ and $y$. Then $K$ meets both $J$ and $L$. Suppose there is a line $K' \neq K$ through $y$ meeting both $J$ and $L$. Let $x'$ be the intersection of $K'$ and $L$. By Theorem \[thm LMN\], $K'$ meets $N$, and $K$ meets $N$. By Theorem \[thm LMN\], $M \\|_{x} N$. Since $M \\|_{x} N$ and $M\\|_x J$, we have $J\\|_{x}N $ and so the line $x.(J\cap K')$ meets $N$. ($J\cap K'$ denotes the intersection point of $J$ and $K'$.) Then the four lines $K$, $K'$, $N$, $x.(J\cap K')$ form an O’Nan configuration. This contradicts (I). \(2) $\|{{\mathcal S}}_L \setminus \{L,M,N\}\| = (n^2-n+1)-3 =(n + 1)(n - 2)$. This is also the number of circles of type ${{\mathcal C}}^y$ tangent to ${\overline{L} }$ in ${{\mathcal I}}(y)$. Since we have , to show , it suffices to show ${\overline{J_1} } \neq {\overline{J_2} }$ for any distinct $J_1,J_2 \in {{\mathcal S}}_L \setminus \{L,M,N\}$. Suppose not. Then $J_1 \\|_y J_2$. By (1), there is a line $K''$ through $y$ meeting $J_1,J_2,L$. Let $x''$ be the intersection of $K''$ and $L$. Since $J_1,J_2 \in {\mathcal S^{}_{L} }$ and ${{\mathcal S}}_L$ is a special spread, we have $J_1 \\|_{x''} J_2$ by the definition of special spread. Let $K'''$ be the line through $y$ and a point of $J_2$ not on $K$. Since $J_1 \\|_y J_2$ and $J_1 \\|_{x''} J_2$, the four lines $K''$, $K'''$, $J_1$, $x''.(J_3\cap K''')$ form an O’Nan configuration. A contradiction. Using Lemma \[lemma SLinIy\] and the characterization of flocks in terms of bundles in even order inversive planes by Dembowski and Hughes [@DH] (see also (6.2.11), (6.2.12) with footnote on p.267, and (6.2.13), of [@Dem2]), we deduce Lemma \[lemma FMN\]. Lemma \[lemma FMN\] describes when lines of ${\mathcal S^{}_{L} }$ meet two intersecting lines which do not belong to ${\mathcal S^{}_{L} }$ and which miss $L$. \[lemma FMN\] Let $L$ be a line of ${{\mathcal U}}$. Let $M,N$ be distinct lines not in ${{\mathcal S}}_{L}$. Suppose $M$ and $N$ intersect at a point $y$ of ${{\mathcal U}}$. If ${\overline{L} }$ belongs to ${{\mathcal F}}(M,N)$ in ${{\mathcal I}}(y)$, then the following statements hold: 1. \[itemFMN2\] Let $L_1$ be the line in ${\mathcal S^{}_{L} }$ through $y$. In ${{\mathcal I}}(y)$, $L_1$ is a point on the circle $C$ determined by $M$, $N$ and $\infty_y$. 2. \[itemFMN1\] In ${{\mathcal U}}$, every line in ${\mathcal S^{}_{L} }$ meeting $M$ meets $N$ as well. 3. \[itemFMN4\] In ${{\mathcal U}}$, there is a unique point $x$ on $L$ such that any line from $x$ meeting $L_1$ misses all other lines in ${\mathcal S^{}_{L} }$ that meet both $M$ and $N$. Furthermore, in ${{\mathcal I}}(y)$, the unital line $x.y$ through $x$ and $y$ is a point on $C$. By Dembowski and Hughes [@DH], in ${{\mathcal I}}(y)$, the flock ${{\mathcal F}}(M,N)$ is the set of circles tangent to every circle in the bundle ${{\mathcal B}}(M,N)$, i.e. the set of circles in ${{\mathcal I}}(y)$ through the points $M$ and $N$. Suppose ${\overline{L} }$ is in ${{\mathcal F}}(M,N)$ in ${{\mathcal I}}(y)$. Then ${\overline{L} }$ is tangent to every circle in ${{\mathcal B}}(M,N)$. By Lemma \[lemma SLinIy\], ${{\mathcal B}}(M,N) = \{{\overline{L_2} },{\overline{L_3} },\cdots,{\overline{L_{n+1}} },C \}$ for some distinct lines $L_2,L_3,\cdots,L_{n+1} \in {\mathcal S^{}_{L} }$. Since ${\mathcal S^{}_{L} }$ is a spread, $L_1$ does not meet $L_2,L_3,\cdots, L_{n+1}$. Hence in ${{\mathcal I}}(y)$, $L_1$ is not on ${\overline{L_2} },{\overline{L_3} },\cdots,{\overline{L_{n+1}} }$, but on the remaining circle $C$ of ${{\mathcal B}}(M,N)$. This proves . The lines in ${\mathcal S^{}_{L} }$ meeting $M$ are $L_1,L_2,\cdots, L_{n+1}$, and they meet $N$. This proves . Let $K$ be the unital line on $y$ such that $K$ is the intersection point of $C$ and ${\overline{L} }$ in ${{\mathcal I}}(y)$. Since $K$ is on ${\overline{L} }$ in ${{\mathcal I}}(y)$, $K$ meets $L$ at a point, say $x$, in ${{\mathcal U}}$. Since $K$ is not on ${\overline{L_2} },{\overline{L_3} },\cdots,{\overline{L_{n+1}} }$ in ${{\mathcal I}}(y)$, $K$ does not meet $L_2,L_3,\cdots, L_{n+1}$ in ${{\mathcal U}}$. Since ${\mathcal S^{}_{L} }$ is a special spread and $K$ is a line from $x$ meeting $L_1\in {\mathcal S^{}_{L} }$ but not $L_2,L_3,\cdots, L_{n+1}\in {\mathcal S^{}_{L} }$, $x$ is a point on $L$ satisfying the property in . Uniqueness of $x$ follows from that fact that in ${{\mathcal I}}(y)$, $K$ is the unique point on ${\overline{L} }$ not on ${\overline{L_2} },{\overline{L_3} },\cdots,{\overline{L_{n+1}} }$. This proves . We introduce the notion of [triply ruled set]{} for a general unital: in a unital of order $m$, a set of $(m + 1)^2$ points is [triply ruled]{} if there are three partitions of the $(m + 1)^2$ points by lines. The following lemma suggests a method to find a triply ruled set. \[lemma parallel class\] Let $L$ and $M_1$ be disjoint lines of ${{\mathcal U}}$ with $M_1 \notin {{\mathcal S}}_L$. Let $L_1$, $L_2$, $\cdots$, $L_{n + 1}$ be the lines of ${\mathcal S^{}_{L} }$ meeting $M_1$. Then there are lines $M_2$, $M_3$, $\cdots$, $M_{n + 1}$, $N_1$, $N_2$, $\cdots$, $N_{n + 1} $ such that $\{ M_1,M_2,\cdots,M_{n + 1}\}$ and $\{ N_1,N_2,\cdots,N_{n + 1}\}$ are partitions of the set of points covered by the disjoint lines $L_1,L_2,\cdots,L_{n + 1}$. Let $y$ be a point on $M_1$. Since $M_1$ and $L$ are disjoint, the point $M_1$ is not incident with ${\overline{L} }$ in ${{\mathcal I}}(y)$. Since $M_1 \notin {\mathcal S^{}_{L} }$, condition ($p$) implies $L \notin {\mathcal S^{}_{M_1} }$ and hence ${\overline{L} } \notin {{\mathcal F}}(M_1,\infty_y)$. Thus, there is a unital line $N_1$ through $y$ such that ${\overline{L} } \in {{\mathcal F}}(M_1,N_1)$. By Lemma \[lemma FMN\], $L_1,L_2,\cdots,L_{n + 1}$ meet $N_1$. For $i=2,3,\cdots,n+1$, let $y_i$ be the intersection point of $N_1$ and $L_i$. By a similar argument, for each $i=2,3,\cdots,n+1$, there is a unital line $M_i$ through $y_i$ such that ${\overline{L} } \in {{\mathcal F}}(M_i,N_1)$, and there is a unital line $N_i$ through $y_i$ such that ${\overline{L} } \in {{\mathcal F}}(M_i,N_i)$. By Lemma \[lemma FMN\], $L_1,L_2,\cdots,L_{n + 1}$ meet $M_i,N_i$, for all $i=2,3,\cdots,n+1$. We are going to show that $M_1,M_2,\cdots,M_{n + 1}$ are mutually disjoint. Suppose $M_{i_1}$ and $M_{i_2}$ intersect at a point $z$ for some distinct $i_1,i_2 \in \{1,2,\cdots,n + 1\}$. Then $z$ is on $L_{i_3}$ for some $i_3 \in \{1,2,\cdots,n + 1\} \setminus \{i_1,i_2\}$. Let $i_4 \in \{1,2,\cdots,n + 1\} \setminus \{i_1,i_2,i_3\}$. Then the four lines $M_{i_1}$, $M_{i_2}$, $N_1$, $L_{i_4}$ form an O’Nan configuration. A contradiction. Hence $\{ M_1,M_2,\cdots,M_{n + 1}\}$ is a partition of the set of points covered by $L_1,L_2,\cdots,L_{n + 1}$. By a similar argument, the same conclusion can be drawn for $N_i$’s. Note that there cannot be a forth partition of the set of points covered by $L_1$, $L_2$, $\cdots$, $L_{n + 1}$; otherwise, there would be an O’Nan configuration constituted by lines of different partitions. This suggests the following notion of parallelism: - Let $M$, $N$ be lines missing $L$ and not in ${{\mathcal S}}_{L}$. We say that $M$ and $N$ are [$L$-parallel]{}, denoted by $M \\|_{L} N$, if they are identical or they are non-intersecting and meet the same lines in ${\mathcal S^{}_{L} }$. We say that $M$ and $N$ are [$L$-non-parallel]{} if they intersect and meet the same lines in ${\mathcal S^{}_{L} }$. \[lemma equiv relation\] Let $L$ be a line of ${{\mathcal U}}$. Then $\\|_L$ defines an equivalence relation in the set of lines missing $L$ and not in ${{\mathcal S}}_{L}$. Each class has $n + 1$ lines. There are $n(n - 1)(n - 2)$ equivalence classes under the equivalence relation $\\|_L$. It is clear that $\\|_{L}$ is reflexive and symmetric. It is transitive because of Lemma 3.3 and the fact that two points determine a line. Since non-equal $L$-parallel line are non-intersecting, there are at most $n + 1$ lines in an equivalence class. By Lemma \[lemma parallel class\], this upper bound is achieved. The number of lines missing $L$ is $n(n - 1)(n^2-n - 1)$. Among these lines, $n(n - 1)$ are in ${\mathcal S^{}_{L} }$. The result follows from simple counting. By Lemma \[lemma FMN\], for any distinct lines $M'$ and $N'$ through $y$, $N'$ is $L$-non-parallel to $M'$ if ${\overline{L} } \in {{\mathcal F}}(M',N')$ in ${{\mathcal I}}(y)$. Furthermore, the set of lines which are $L$-non-parallel to $M'$ is an $L$-parallel class. In the context of Lemma \[lemma parallel class\], it is natural to ask whether $M_1,M_2,\cdots,M_{n + 1}$ are in ${\mathcal S^{}_{M} }$ for some line $M$. The answer is yes: \[lemma concurrent\] Refer to the set-up in Lemma \[lemma parallel class\]. Let $M$ be the line in ${\mathcal S^{}_{L} }$ such that $M_1\in {\mathcal S^{}_{M} }$. Then for each $i=2,3,\cdots,n+1$, $M_i$ belongs to ${\mathcal S^{}_{M} }$. By Lemma \[lemma LJzparallel\], there is a point $z_1\in M$ such that $L \\|_{z_1} M_1$. Let $z_2,z_3,\cdots,z_{n + 1}$ be the points on $M$ other than $z_1$. Let $M_{z_1} = M_1$. Let $N$ be a line such that $L,M,N$ form a self-polar triangle. For $i = 2,3,\cdots, n + 1$, we claim that [there is a line $M_{z_i} \in {\mathcal S^{}_{M} }$ distinct from $L$ and $N$, such that $M_{z_i}$ is $z_i$-parallel to $L$, and $M_{z_i}$ meets $L_1,L_2,\cdots,L_{n + 1}$.]{} It the claim is true, then $\{M_{z_i} \| z_i \in M \}$ is the $L$-parallel class containing $M_{z_1}$. Indeed, if $M_{z_i} = M_{z_j}$ for some $i\neq j$, then $M_{z_i}$, $L$ and $N$ would be three lines both $z_i$-parallel and $z_j$-parallel, giving an O’Nan configuration. Hence $M_{z_i} \neq M_{z_j}$ for distinct $i,j$. Furthermore, for $i = 2,3,\cdots,n + 1$, since $M_{z_i}$ is $z_i$-parallel to $L$, $M_{z_i}$ is in ${\mathcal S^{}_{M} }$ by Lemma \[lemma LJzparallel\]. Thus $M_{z_i}$’s are non-intersecting. Hence $\{M_{z_i} \| z_i \in M \}$ is the $L$-parallel class containing $M_{z_1}$. By uniqueness of such a class, $\{M_1,M_2,\cdots, M_{n+1} \}=\{M_{z_i} \| z_i \in M \}$. The result follows. We now prove the claim. Let $k \in \{2,3,\cdots, n + 1\}$. For $i = 1,2,\cdots,n + 1$, let $K_i = (M_1 \cap L_i).z_1$. Since $L \\|_{z_1} M_1$, we can label the points on $L$ as $x_1,x_2,\cdots,x_{n + 1}$ such that $x_i \in K_i$. Since ${{\mathcal S}}_L$ is a special spread, and $K_i$ passes through $x_i \in L$ and meets both $M\in{\mathcal S^{}_{L} }$ and $L_i\in{\mathcal S^{}_{L} }$, we have $M \\|_{x_i} L_i$. For $i = 1,2,\cdots,n + 1$, let $K_i' = z_k.x_i$. Since $M \\|_{x_i} L_i$ and $K_i'$ meets $M$, we conclude that $K_i'$ meets $L_i$, say at $w_i$. If $w_1,w_2,\cdots,w_{n + 1} $ are collinear, then take $M_{z_k}$ to be the line that they are on and we find $M_{z_k}$ (Figure \[lemma35\]). \[lemma35\] It remains the case when some of $w_1,w_2,\cdots,w_{n + 1}$ are not collinear. For $i = 1,2,\cdots,n + 1$, let $J_i$ be the line on $w_i$ that is $z_k$-parallel to $L$. At least two of $J_1,J_2,\cdots,J_{n + 1}$ are the same line. For, otherwise, there would be $n + 1$ lines $z_k$-parallel to $L$. Without loss of generality, assume $J_1 = J_2$. We are going to prove that $J_1$ meets $L_1,L_2,\cdots,L_{n + 1}$ by contradiction. Suppose $J_1$ misses some of $L_3,L_4,\cdots,L_{n + 1}$. By Lemma \[lemma parallel class\], there are lines $M_2$ and $N_2$ on $w_1$ respectively $L$-parallel to $M_1$ and $L$-non-parallel to $M_1$. Thus, in ${{\mathcal I}}(w_1)$, ${{\mathcal B}}(M_2,N_2) = \{{\overline{L_2} },{\overline{L_3} },\cdots,{\overline{L_{n + 1}} }, C\}$ for some circle $C$ of type ${{\mathcal C}}_{w_1}$. Since $L_1$ are disjoint from $L_2,L_3,\cdots,L_{n+1}$ in ${{\mathcal U}}$, $L_1$ is a point on $C$ in ${{\mathcal I}}(w_1)$. Since $J_1 = J_2$ meets $L_2$ in ${{\mathcal U}}$, $J_1$ is a point on ${\overline{L_2} }$ in ${{\mathcal I}}(w_1)$. Since $J_1 \neq M_2$ and $J_1 \neq N_2$ by hypothesis, and $J_1\in {\overline{L_2} }$, we conclude $J_1 \notin C$. On the other hand, since $J_1\\|_{z_k} L$, ${{\mathcal B}}(K_1',J_1) = \{{\overline{K_2'} },{\overline{K_3'} },\cdots,{\overline{K_{n + 1}'} }, D\}$ for some circle $D$ of type ${{\mathcal C}}_{w_1}$. Note that $L_1 \in D$. Indeed, if $L_1$ is not on $D$, then $L_1$ meets $K_i'$ in ${{\mathcal U}}$ for some $i \neq 1$. By Theorem \[thm LMN\], $K_i'$ meets $N$. Since $K_i'$ meets both $N \in {\mathcal S^{}_{L} }$ and $L_1\in {\mathcal S^{}_{L} }$, we have $N \\|_{x_i}L_1$ because ${{\mathcal S}}_{L}$ is a special spread. Thus the four lines $K_1'$, $K_i'$, $L_1$, $x_i. (K_1 \cap N)$ form an O’Nan configuration. Hence $L_1 \in D$. Since $J_1\in D$ but $J_1\notin C$, we have $C\neq D$. Since $\infty_{w_1}$ and $L_1$ are points on both $C$ and $D$, $K_1' \notin C$. Then $K_1'$ meets $L_j$ in ${{\mathcal U}}$ for some $j \neq 1$. By Theorem \[thm LMN\], $K_i'$ meets $N$. Since $K_1'$ meets both $N\in {\mathcal S^{}_{L} }$ and $L_j \in {\mathcal S^{}_{L} }$, we have $N \\|_{x_1} L_j$ and so the four lines $K_1'$, $K_j'$, $L_j$, $x_1. (K_j'\cap N)$ form an O’Nan configuration. A contradiction. Thus $J_1$ meets $L_1,L_2,\cdots,L_{n + 1}$. Take $M_{z_k}$ to be $J_1$ and we prove our claim. \[thm bigP\] Let $L,M,N$ be a self-polar triangle with respect to ${{\mathcal U}}$. Then ${{\mathcal S}}_L \setminus \{L,M,N\}$, ${{\mathcal S}}_M \setminus \{L,M,N\}$, ${{\mathcal S}}_N \setminus \{L,M,N\}$ can be respectively partitioned into $n - 2$ subsets $$\{L^1_{1},L^1_{2},\cdots,L^1_{n + 1}\}, \cdots, \{L^{n - 2}_{1},L^{n - 2}_{2}, \cdots, L^{n - 2}_{n + 1}\};$$ $$\{M^1_{1},M^1_{2},\cdots,M^1_{n + 1}\}, \cdots, \{M^{n - 2}_{1},M^{n - 2}_{2}, \cdots, M^{n - 2}_{n + 1}\};$$ $$\{N_1^1,N^1_{2}, \cdots, N^1_{n + 1}\}, \cdots, \{N^{n - 2}_{1},N^{n - 2}_{2}, \cdots, N^{n - 2}_{n + 1}\};$$ each of cardinality $n + 1$, such that for $i = 1,2,\cdots,n - 2$, the sets of points incident respectively on the lines of $\{L^i_1,L^i_{2},\cdots,L^i_{n + 1}\}$, $\{M^i_{1},M^i_{2}$, $\cdots$, $M^i_{n + 1}\}$ and $\{N^i_{1},N^i_{2},\cdots,N^i_{n + 1}\}$ are the same. Take a line $M^1_1 \in {{\mathcal S}}_M \setminus \{L,M,N\}$. Note that $M^1_1 \notin {\mathcal S^{}_{L} }$ because $ {\mathcal S^{}_{L} } \cap {\mathcal S^{}_{M} } = \{N\}$. Let $L^1_1,L^1_2,\cdots,L^1_{n + 1}$ be the lines of ${\mathcal S^{}_{L} }$ meeting $M^1_1$. By Lemmas \[lemma parallel class\] and \[lemma concurrent\], there is an $L$-parallel class $\{M^1_1$, $M^1_2$, $\cdots$, $M^1_{n + 1}\}$$\subset {\mathcal S^{}_{M} }$ and its $L$-non-parallel class $\{N^1_1,N^1_2$, $\cdots$, $N^1_{n + 1}\}$ both partitioning the set of points covered by $L^1_1$, $L^1_2$, $\cdots$, $L^1_{n + 1}$. Since $\{N^1_1,N^1_2$, $\cdots$, $N^1_{n + 1}\}$ is an $L$-parallel class and an $M$-parallel class, it is a subset of ${\mathcal S^{}_{N'} }$ where $N' \in {\mathcal S^{}_{L} } \cap {\mathcal S^{}_{M} }$ by applying Lemma \[lemma concurrent\] twice. Hence $N' = N$. Repeat the process $n-3$ times by taking a line $M^i_1 \in {{\mathcal S}}_M \setminus ( \{L,M,N\} \cup \{ M^k_l \| k = 1,2,\cdots,i-1,l = 1,2,\cdots,n + 1\})$. This finishes the proof. We may interpret Theorem \[thm bigP\] as follows. If ${{\mathcal U}}$ is embedded in a projective plane $\pi$ as a polar unital via the construction of [@HW2 Theorem 1.1], then ${\mathcal S^{}_{L} }$, ${\mathcal S^{}_{M} }$, ${\mathcal S^{}_{N} }$ are respectively the set of lines through the pole of $L$, $M$, and $N$. Thus, Theorem \[thm bigP\] says that in $\pi$, the set of points of ${{\mathcal U}}$ is partitioned into a self-polar triangle, and $n - 2$ subsets of $(n+1)^2$ points triply ruled by lines through the vertices of the triangle. In the setting in Theorem \[thm bigP\], for any disjoint index sets $I_L, I_M, I_N$ such that $I_L \cup I_M \cup I_N = \{1,2,\cdots,n + 1\}$, the set $\{L,M,N\}\cup$$\{L^i_j \mid i\in I_L, j = 1,2,\cdots,n + 1\}$$\cup \{M^i_j \mid i\in I_M, j = 1,2,\cdots,n + 1\}$$\cup \{N^i_j \mid i\in I_N, j = 1,2,\cdots,n + 1\}$ is a spread of ${{\mathcal U}}$. These spreads are the subregular spreads studied by Dover [@Dov]. From a special spread of a unital to a regular spread of ${\mbox{\rm PG}}(3,n)$ {#st 4} =============================================================================== From now on, we fix a line $L$, and let $x_1,x_2,\cdots, x_{n+1}$ be the points on $L$. Following Wilbrink [@Wil], we are going to construct a generalized quadrangle $GQ(L)$, which is isomorphic to $Q(4,n)$ [@Wil]. We will then embed $Q(4,n)$ into ${\mbox{\rm PG}}(4,n)$ and choose a 3-dimensional projective space $\Sigma$ in ${\mbox{\rm PG}}(4,n)$. It turns out that the special spread ${{\mathcal S}}_{L}$ of ${{\mathcal U}}$ introduced in Section \[st 2\] defines a regular spread ${{\mathscr S}}$ of $\Sigma$ (Theorem \[thm regular\]). Using the Bruck-Bose construction [@BB1; @BB2], we will construct a projective plane in Section \[st 6\] using this regular spread ${{\mathscr S}}$, such that ${{\mathcal U}}$ is embedded in a way into this projective plane as a classical unital. To construct $GQ(L)$, we recall the definition of the sets ${{\mathcal A}}_{ij}, 1 \leq i, j \leq n + 1$ of ${{\mathcal U}}$ ([@Wil], also see [@HW2]). Considering $\mathcal I(x_1)$, denote the circles in the bundle ${{\mathcal B}}(L, \infty_{x_1})$ by $\{L, \infty_{x_1}\} \cup {{\mathcal A}}_{1j}$, where $j = 1, 2, \cdots, n + 1$. We have defined ${{\mathcal A}}_{11}, {{\mathcal A}}_{12}, \cdots, {{\mathcal A}}_{1, n + 1}$. Next, for each $j \in \{1, 2, \cdots, n + 1\}$, consider the pencil $<L, \{L, \infty_{x_1}\} \cup {{\mathcal A}}_{1j}>$ in ${{\mathcal I}}(x_1)$, i.e. the maximal set of mutually tangent circles through $L$ with a member the circle $\{L, \infty_{x_1}\} \cup {{\mathcal A}}_{1j}$. For $k = 1, 2, \cdots, n - 1$, denote by $C_{jk}$ the remaining circles in the pencil. For $i \in \{2, 3, \cdots, n + 1\}$, consider the $n - 1$ lines on $x_i$ which correspond respectively to these $n - 1$ circles $C_{jk}$’s. Denote this set of lines by ${{\mathcal A}}_{ij}$. We have defined ${{\mathcal A}}_{2j}, {{\mathcal A}}_{3j}, \cdots, {{\mathcal A}}_{n + 1,j}$, for $j = 1, 2, \cdots, n + 1$. The definition of ${{\mathcal A}}_{ij}$ is independent on the choice of the point $x_1 \in L$. The set of points of $GQ(L)$ is $$\{ {{\mathcal A}}_{ij} \mid i,j = 1,2 \cdots n + 1 \} \cup \{\ y \mid y \mbox{ is a point of }{{\mathcal U}}\mbox{ not on }L\}.$$ The set of lines of $GQ(L)$ is $$\{ A_i \mid i = 1,2 \cdots n + 1 \} \cup \{ B_i \mid i = 1,2 \cdots n + 1 \} \cup \{ K \mid K\neq L\mbox{ is a line of }{{\mathcal U}}\mbox{ meeting }L\}.$$ The incidence of $GQ(L)$ is as follows. ${{\mathcal A}}_{ij}$ is incident with $A_k$ if and only if $i = k$; ${{\mathcal A}}_{ij}$ is incident with $B_k$ if and only if $j = k$; for any line $K$ of ${{\mathcal U}}$ meeting $L$, ${{\mathcal A}}_{ij}$ is incident with $K$ if and only if $K \in {{\mathcal A}}_{ij}$; a point $y$ of ${{\mathcal U}}$ is never incident with $A_i$ or $B_j$ for $i,j = 1,2,\cdots,n + 1$; incidence between a point and a line of ${{\mathcal U}}$ is the natural incidence. Consider a parabolic quadric ${{\mathcal P}}$ in ${\mbox{\rm PG}}(4,n)$. The points and lines of ${{\mathcal P}}$ form a generalized quadrangle $Q(4,n)$ [@PT]. By [@Wil], $GQ(L)$ is isomorphic under some GQ isomorphism $$\label{eqn varphi} \varphi: GQ(L) \longrightarrow Q(4,n)$$ to $Q(4,n)$. Consider the 3-dimensional subspace $\Sigma$ of ${\mbox{\rm PG}}(4,n)$ determined by the skew lines $\varphi(A_1)$ and $\varphi(A_2)$. Then $\Sigma \cap {{\mathcal P}}= \{ \varphi({{\mathcal A}}_{ij}) \mid i,j = 1,2 \cdots n + 1 \}$, and is a hyperbolic quadric ${{\mathcal H}}$ with regulus $$\label{eqn R0} {{\mathcal R}}_0 = \{ \varphi(A_i) \mid i = 1,2,\cdots,n + 1 \}$$ and opposite regulus $ \{ \varphi(B_i) \mid i = 1,2,\cdots,n + 1 \}$. ${{\mathcal H}}$ defines a polarity $$\label{eqn alpha} \alpha: \Sigma\longrightarrow \Sigma$$ of $\Sigma$. The tangent spaces of ${{\mathcal P}}$ are concurrent at a point [nucleus]{} ${\bf N}$ of ${{\mathcal P}}$. Let $$\label{eqn mu} \mu: {{\mathcal P}}\longrightarrow \Sigma$$ be the function defined as follows: for any point ${\bf V}$ of ${{\mathcal P}}$, $\mu({\bf V})$ is the intersection point of $\Sigma$ and the line joining ${\bf N}$ and ${\bf V}$. Since ${{\mathcal H}}\cap \Sigma = {{\mathcal H}}$, $\mu$ is identity on ${{\mathcal H}}$. Since $\|{{\mathcal P}}\| = n^3 + n^2 + n + 1 = \|\Sigma\|$, $\mu$ is a bijection. Hence, we have a 1-1 correspondence between points of $GQ(L)$ and that of $\Sigma$ via the composition function $\mu \varphi$. Furthermore, some quadratic cones in ${{\mathcal P}}$ are mapped to planes of $\Sigma$ because $n$ is even: \[lemma projcone\] Let ${\bf V}\in {{\mathcal P}}\setminus \Sigma$. Let ${{\mathcal Q}}$ be the quadratic cone formed by the intersection of ${{\mathcal P}}$ and the tangent space of ${{\mathcal P}}$ at ${\bf V}$. Then $\mu$ maps ${{\mathcal Q}}$ onto the plane $\alpha(\mu(\bf V))$ in $\Sigma$. Furthermore, the plane $\alpha(\mu(\bf V))$ meets ${{\mathcal H}}$ in an irreducible conic with nucleus $\mu(\bf V)$. Since ${\bf V}\notin \Sigma$, every generator of ${{\mathcal Q}}$ meets $\Sigma$ in a unique point (of ${{\mathcal H}}$). Hence, for any generator $l$ of ${{\mathcal Q}}$, $\mu(l)$ is tangent to ${{\mathcal H}}$ and passes through $\mu(\bf V)$. By a property of hyperbolic quadric of even order, $\{ \mu(l) \mid l$ is a generator of ${{\mathcal Q}}\}$ is on the plane $\alpha(\mu(\bf V))$. Since $\|{{\mathcal Q}}\| = n^2 + n + 1$, the image set $\mu({{\mathcal Q}})$ is a plane. Furthermore, $\mu(\bf V)$ is the nucleus of the irreducible conic formed by the intersection of $\mu({{\mathcal Q}})$ and ${{\mathcal H}}$. Using Lemma \[lemma projcone\] and the GQ isomorphism $\varphi$, we prove that every line of ${\mathcal S^{}_{L} }$ is mapped to a line of $\Sigma$ under $\mu \varphi$: \[lemma Mvarphimu\] Let $M$ be a line of ${{\mathcal U}}$ in ${\mathcal S^{}_{L} }$. Then $\mu(\varphi(M))$ is a line in $\Sigma$. By Theorem \[thm LMN\], $L \\|_{z_i} M$ for $i = 1,2,\cdots,n + 1$. Let ${{\mathcal Q}}_i$ be the quadratic cone formed by the intersection of ${{\mathcal P}}$ and the tangent space of ${{\mathcal P}}$ at $\varphi(z_i)$. Hence $\varphi(M) \subset {{\mathcal Q}}_i$. By Lemma \[lemma projcone\], $\mu(\varphi(M)) \subset \bigcap^{n + 1}_{i = 1} \alpha(\mu(\varphi(z_i)))$ and $\alpha(\mu(\varphi(z_i)))$’s are planes in $\Sigma$. Since $\| \varphi(M) \| = n + 1$ and $\mu(\varphi(M))$ is in the intersection of $n + 1$ planes, $\mu(\varphi(M))$ is a line. Since ${{\mathcal S}}_L$ is a spread of ${{\mathcal U}}$ and $\mu$ is a bijection, the set $\{\mu(\varphi(L')) \mid L'\in {\mathcal S^{}_{L} }\}$ consists of disjoint lines. Let $$\label{eqn S} {{\mathscr S}}= {{\mathcal R}}_0 \cup \{\mu(\varphi(L')) \mid L'\in {\mathcal S^{}_{L} }\}.$$ Then ${{\mathscr S}}$ is a spread. We claim that ${{\mathscr S}}$ is regular (Theorem \[thm regular\]). The justification of this claim requires the notion of tube [@CK]: When $q$ is even, a [tube]{} in ${\mbox{\rm PG}}(3,q)$ is a pair $\mathcal T = \{l,\mathcal B\}$, where $\{l\} \cup \mathcal B$ is a collection of mutually disjoint lines of ${\mbox{\rm PG}}(3,q)$ such that for each plane $\Pi$ of ${\mbox{\rm PG}}(3,q)$ containing $l$, the intersection of $\Pi$ with the lines of $\mathcal B$ is a hyperoval. For any mutually skew lines $l_1,l_2,l_3$ in $\Sigma$, denote by ${{\mathcal R}}(l_1,l_2,l_3)$ the unique regulus determined by them. According to Cameron and Knarr [@CK], if $\{l,\{l_0,l_1,\cdots,l_{q + 1}\}\}$ is a tube, then the union $\bigcup_{i = 1}^{n + 1} {{\mathcal R}}(l,l_0,l_i)$ is a regular spread in ${\mbox{\rm PG}}(3,q)$. \[lemma tube\] Let $M,N$ be lines of ${{\mathcal U}}$. Suppose $L,M,N$ form a self-polar triangle. Then the pair $\mathcal T = \{ \mu(\varphi(M))$, $\{ \mu(\varphi(N)) \} \cup {{\mathcal R}}_0 \}$ is a tube in $\Sigma$. Let $z_1,z_2,\cdots,z_{n + 1}$ be the points of $N$. By Lemma \[lemma projcone\], the $n + 1$ planes in $\Sigma$ containing $\mu(\varphi(M))$ are $\alpha(\mu(\varphi(z_i)))$, $i = 1,2,\cdots,n + 1$. Since ${{\mathcal R}}_0$ is a regulus of ${{\mathcal H}}$, $\alpha(\mu(\varphi(z_i)))$ meets the points covered by ${{\mathcal R}}_0$ in an irreducible conic with nucleus $\mu(\varphi(z_i))$ by Lemma \[lemma projcone\]. The result follows. \[thm regular\] ${{\mathscr S}}$ is a regular spread in $\Sigma$. Let $M,N$ be lines such that $L,M,N$ form a self-polar triangle. For $i = 1,2,\cdots,n + 1$, let ${{\mathcal R}}_i = {{\mathcal R}}( \mu(\varphi(M))$, $ \mu(\varphi(N))$, $\mu(\varphi(A_i)) )$. By [@CK] mentioned above, the union $\bigcup_{i = 1}^{n + 1} {{\mathcal R}}_i$ is a regular spread. We are done if we show ${{\mathscr S}}= \bigcup_{i = 1}^{n + 1} {{\mathcal R}}_i$. Thus it suffices to show ${{\mathscr S}}\subset \bigcup_{i = 1}^{n + 1} {{\mathcal R}}_i $. Let $L_1 \in {{\mathcal S}}_L \setminus \{L,M,N\}$. Since $L\in {\mathcal S^{}_{M} }$ and $L_1\in {\mathcal S^{}_{L} }$, there is a point $x_i \in L$ such that $M \\|_{x_i} L_1$ by Lemma \[lemma LJzparallel\]. Let $K_1,K_2,\cdots,K_{n + 1}$ be the lines of ${{\mathcal U}}$ on $x_i$ meeting $M$ (and hence meeting $L_1$). Then they meet $N$ by Theorem \[thm LMN\]. Hence, $\mu(\varphi(K_1))$, $\mu(\varphi(K_2))$, $\cdots$, $\mu(\varphi(K_{n + 1}))$ are lines in $\Sigma$ meeting $\mu(\varphi(M))$, $\mu(\varphi(N))$ and $\mu(\varphi(A_i))$. Furthermore, these $n + 1$ lines are disjoint. Indeed, the points $K_1,K_2,\cdots,K_{n + 1}$ are incident with ${\overline{M} } \in {{\mathcal F}}(L,\infty_{x_i})$ in ${{\mathcal I}}(x_i)$, where ${\overline{M} }$ is tangent to each circle in the bundle ${{\mathcal B}}(L, \infty_{x_i})$ [@DH], and so we may assume $K_j\in {{\mathcal A}}_{ij}$ for $j = 1,2,\cdots n + 1$ . Thus, $\{\mu(\varphi(K_j)) \mid j = 1,2,\cdots,n + 1\}$ is the opposite regulus of ${{\mathcal R}}_i$. Since $\mu(\varphi(L_1))$ meets every line in the opposite regulus of ${{\mathcal R}}_i$, it is in ${{\mathcal R}}_i$. From a partition of $\mathcal S^{}_L$ to a pencil of quadrics in ${\mbox{\rm PG}}(3,n)$ {#st 5} ======================================================================================== We use the notations in Section \[st 4\] and continue to prove that ${{\mathcal U}}$ is classical. The key result in this section is Theorem \[thm pencil\]. It says that the image of the partition of ${{\mathcal S}}_{L}$ in Theorem \[thm bigP\] under $\mu\varphi$ corresponds to a pencil of quadrics of two lines and $n-1$ hyperbolic quadrics of the projective space $\Sigma$, where $\mu$, $\varphi$ and $\Sigma$ are defined Section \[st 4\]. To prove Theorem \[thm pencil\], we have to describe every regulus of the spread ${{\mathscr S}}$ defined in , in terms of the geometry of ${{\mathcal U}}$. A regulus of ${{\mathscr S}}$ has two, one, or no common lines with ${{\mathcal R}}_0$. We consider these cases separately. We first consider the reguli of ${{\mathscr S}}$ with exactly one common line with ${{\mathcal R}}_0$. They can be described using $x$-parallelism introduced in Section \[st 2\], where the $x$’s are the points of the line $L$. \[lemma C1\] Let $i \in \{1,2, \cdots, n + 1\}$ and $L_1, L_2,\cdots,L_n \in {\mathcal S^{}_{L} }$. If ${{\mathcal R}}= \{ \mu(\varphi(A_i))$, $\mu(\varphi(L_1))$, $\mu(\varphi(L_2))$, $\cdots$, $\mu(\varphi(L_n))\}$ forms a regulus in ${{\mathscr S}}$, then $L_j \\|_{x_i} L_k$ for any $j,k \in \{ 1,2,\cdots,n\}$. Let $l$ be a line in the opposite regulus of ${{\mathcal R}}$. By the definition of $\mu$ and the construction of $GQ(L)$, there is a line $K$ of ${{\mathcal U}}$ on $x_i$ such that $l = \mu(\varphi(K))$. Since $l$ meets $\mu(\varphi(L_1)), \mu(\varphi(L_2)) ,\cdots, \mu(\varphi(L_n))$, the line $K$ meets $L_1, L_2,\cdots,L_n \in {\mathcal S^{}_{L} }$ in ${{\mathcal U}}$. Since ${{\mathcal S}}_L$ is a special spread, $L_1, L_2,\cdots,L_n$ are $x_i$-parallel. Lemma \[lemma C2\] describes the reguli of ${{\mathscr S}}$ with exactly two common lines with ${{\mathcal R}}_0$. To prove Lemma \[lemma C2\], we need Lemma \[lemma F(K,K’)\]. \[lemma F(K,K’)\] Let $i_1,i_2,j_1,j_2 \in \{1,2, \cdots, n + 1\}$ with $i_1 \neq i_2$ and $j_1 \neq j_2$. Let $K_1 \in {{\mathcal A}}_{i_1 j_1}$ and $K_2 \in {{\mathcal A}}_{i_2 j_2}$ be lines of ${{\mathcal U}}$ meeting at a point $y$ of ${{\mathcal U}}$. Let $M$ be a line in ${\mathcal S^{}_{L} }$ not through $y$. Then ${\overline{M} }$ is in the flock ${{\mathcal F}}(K_1,K_2)$ in ${{\mathcal I}}(y)$ if and only if there is a point $z\in M$ such that $z.x_k\in {{\mathcal A}}_{i_k j_k}$ for $k=1,2$. Without loss of generality, assume $i_1=j_1=1$ and $i_2=j_2=2$. Suppose ${\overline{M} }\in {{\mathcal F}}(K_1,K_2)$ in ${{\mathcal I}}(y)$. For $k=1,2$, let $N_k$ be a line through $x_k$ such that ${\overline{M} }\in {{\mathcal F}}(K_k,N_k)$ in ${{\mathcal I}}(x_k)$. Applying Lemma \[lemma FMN\] respectively to ${\overline{M} }\in {{\mathcal F}}(K_1,K_2)$ in ${{\mathcal I}}(y)$, ${\overline{M} }\in {{\mathcal F}}(K_1,N_1)$ in ${{\mathcal I}}(x_1)$ and ${\overline{M} }\in {{\mathcal F}}(K_2,N_2)$ in ${{\mathcal I}}(x_2)$, any line of ${\mathcal S^{}_{M} }$ meeting $K_1$ meets $K_2$,$N_1$ and $N_2$. Since $M \in {\mathcal S^{}_{L} }$, we have $L \in {\mathcal S^{}_{M} }$ by condition ($p$). Hence, $L$ is the line in ${\mathcal S^{}_{M} }$ that meet $K_1,K_2,N_1,N_2$. By Lemma \[lemma FMN\], there is a point $z_k$ on $M$ such that any line from $z_k$ meeting $L$ miss all other lines in ${\mathcal S^{}_{M} }$ that meet $K_1,K_2,N_1,N_2$. By uniqueness in Lemma \[lemma FMN\], $z_1=z_2$. By Lemma \[lemma FMN\] and , in ${{\mathcal I}}(x_k)$, the points $z_k.x_k$, $K_k$, $L$, $\infty_z$ are concircular. By definition of ${{\mathcal A}}_{kk}$, we have $z_k.x_k \in {{\mathcal A}}_{k k}$ for $k=1,2$. Conversely, note that there are exactly $n-2$ circles in ${{\mathcal F}}(K_1,K_2)$ in ${{\mathcal I}}(y)$ is of type ${{\mathcal C}}^y$, and each of these circles gives one $z\neq y$ such that $z.x_1 \in {{\mathcal A}}_{11}$ and $z.x_2 \in {{\mathcal A}}_{22}$. To prove the converse, it suffices to show there are exactly $n - 2$ such $z$’s. By definition of ${{\mathcal A}}_{k k}$, in ${{\mathcal I}}(x_1)$, lines of ${{\mathcal A}}_{22}$ correspond to circles of a pencil with carrier $L$, and lines of ${{\mathcal A}}_{11}$ correspond to points on a circle through $L$ not in that pencil. Hence, each line of ${{\mathcal A}}_{11}$ meets exactly one line of ${{\mathcal A}}_{2 2}$ in ${{\mathcal U}}$. Since $\|{{\mathcal A}}_{11}\|=n-1$ and there is only one line of ${{\mathcal A}}_{11}$ passing through $y$, there are exactly $n - 2$ such $z$’s. \[lemma C2\] Let $i_1,i_2 \in \{1,2, \cdots, n + 1\}$ with $i_1 \neq i_2$. Let $L_1, L_2,\cdots,L_{n-1}\in {\mathcal S^{}_{L} }$. Suppose ${{\mathcal R}}$ is a regulus of ${{\mathscr S}}$ containing $\mu(\varphi(A_{i_1}))$, $\mu(\varphi(A_{i_2}))$, $\mu(\varphi(L_1))$, $\mu(\varphi(L_2))$, $\cdots$, $\mu(\varphi(L_{n-1}))$. Then for any point $z_1\in L_1$, if $K_1$ is the line passing through $z_1$ and $x_{i_1}$, and if $K_2$ is the line passing through $z_1$ and $x_{i_2}$, then in ${{\mathcal I}}(z_1)$, $${{\mathcal F}}(K_1, K_2) = \{ {\overline{L_2} },{\overline{L_3} },\cdots,{\overline{L_{n-1}} },C\}$$ for some circle $C$ through $\infty_{z_1}$. Let $z$ be a point on $L_1$. Let $l$ be the line in the opposite regulus of ${{\mathcal R}}$ through $\mu(\varphi(z))$. Let $z' \in L_2$ be the unital point such that $\mu(\varphi(z')) \in l$. Consider $\alpha$ defined in . Then $\alpha(l)$ meets ${{\mathcal H}}$ at $\mu(\varphi({{\mathcal A}}_{i_1 j_1}))$ and $\mu(\varphi({{\mathcal A}}_{i_1 j_2}))$ for some $j_1,j_2$ with $j_1 \neq j_2$, and $\alpha(l)$ lies on the plane $\alpha(\mu(\varphi(z')))$. Let ${{\mathcal Q}}$ be the quadratic cone formed by the intersection of ${{\mathcal P}}$ and the tangent space of ${{\mathcal P}}$ at $\varphi(z')$. By Lemma \[lemma projcone\], $\mu^{-1}(\alpha(l)) \in {{\mathcal Q}}$ and so $\varphi({{\mathcal A}}_{i_1 j_1}),\varphi({{\mathcal A}}_{i_1 j_2}) \in {{\mathcal Q}}$. Hence $z'$ lies on some unital lines $K_1 \in {{\mathcal A}}_{i_1 j_1}$ and $K_2 \in {{\mathcal A}}_{i_1 j_2}$. By Lemma \[lemma F(K,K’)\], ${\overline{L_2} }\in {{\mathcal F}}(K_1,K_2)$ in ${{\mathcal I}}(z)$. Similarly, ${\overline{L_3} },\cdots,{\overline{L_{n-1}} }\in {{\mathcal F}}(K_1,K_2)$. The result follows by Lemma \[lemma SLinIy\]. Lemma \[lemma C0\] gives a characterization of reguli of ${{\mathscr S}}$ with no common line with ${{\mathcal R}}_0$, by considering inversive planes whose blocks are defined by the reguli of ${{\mathscr S}}$. For each line $J$ of ${{\mathcal U}}$ that misses $L$ and not in ${{\mathcal S}}_{L}$, let $C(J)$ be the set of images of the $n + 1$ lines of ${\mathcal S^{}_{L} }$ meeting $J$ under $\mu\varphi$. \[lemma C0\] A set of $n+1$ lines of ${{\mathscr S}}\setminus {{\mathcal R}}_0$ is a regulus if and only if it is $C(J)$ for some line $J$ of ${{\mathcal U}}$ missing $L$ and not in ${{\mathcal S}}_{L}$. Consider the incidence structure $${{\mathcal I}}_1 = ({{\mathscr S}}, {{\mathcal C}})$$ where ${{\mathcal C}}$ is the set of the reguli of ${{\mathscr S}}$. By Theorem 4.5 (iv) of Bruck [@Bru], ${{\mathcal I}}_1$ is the Miquelian inversive plane of order $n$. Note that ${{\mathcal R}}_0$ is a circle of ${{\mathcal I}}_1$. We denote by ${{\mathcal C}}_0$ the set of those circles in ${{\mathcal I}}_1$ disjoint from ${{\mathcal R}}_0$; by ${{\mathcal C}}_1$ the set of those circles in ${{\mathcal I}}_1$ tangent to ${{\mathcal R}}_0$; by ${{\mathcal C}}_2$ the set of those circles in ${{\mathcal I}}_1$ secant to ${{\mathcal R}}_0$. Let ${{\mathcal C}}_0^* = \{ C(J) \mid J$ is a line of ${{\mathcal U}}$ missing $L$ and not in ${{\mathcal S}}_{L} \}$. Now considering the incidence structure $${{\mathcal I}}_2 = ({{\mathscr S}}, ({{\mathcal C}}\setminus {{\mathcal C}}_0) \cup {{\mathcal C}}_0^* ).$$ By Theorem 2 of [@H], provided that ${{\mathcal I}}_2$ is an inversive plane of order $n$, we will have ${{\mathcal I}}_1 = {{\mathcal I}}_2$ and thus ${{\mathcal C}}_0={{\mathcal C}}_0^$. Hence, to prove Lemma \[lemma C0\], [it suffices to prove that ${{\mathcal I}}_2$ is a $3$–$(n^2 + 1,n + 1,1)$ design.]{} Since ${{\mathscr S}}$ has $n^2 + 1$ lines, ${{\mathcal I}}_2$ has $n^2 + 1$ points. A block in ${{\mathcal C}}_0^$ has exactly $n + 1$ points because every line which is not in ${{\mathcal S}}_{L}$ and which misses $L$ meets exactly $n + 1$ lines of ${\mathcal S^{}_{L} }$. Other blocks of ${{\mathcal I}}_2$ has exactly $n+1$ points because every regulus of ${{\mathscr S}}$ consists of $n+1$ lines. $\|{{\mathcal C}}_0^\| = n(n - 1)(n - 2)/2$ because there are $n(n - 1)(n - 2)$ $L$-parallel classes by Lemma \[lemma equiv relation\], and any class and its $L$-non-parallel class define a same block of ${{\mathcal I}}_2$. Since ${{\mathcal I}}_1$ is an inversive plane, $\|{{\mathcal C}}\setminus {{\mathcal C}}_0\|=n^2(n+3)/2$. Thus ${{\mathcal I}}_2$ has $n (n^2 + 1)$ blocks. It remains to show that any two distinct blocks of ${{\mathcal I}}_2$ have at most two common points. Since ${{\mathcal I}}_2$ only differs from the inversive plane ${{\mathcal I}}_1$ in blocks missing ${{\mathcal R}}_0$, we only need to consider the case when one of the two blocks belongs to ${{\mathcal C}}_0^$. Let $C(J)$ be a block in ${{\mathcal C}}_0^$. Let $\mu(\varphi(L_k)),k=1,2,3$ be distinct points of $C(J)$, where $L_1,L_2,L_3 \in {\mathcal S^{}_{L} }$. Suppose $\mu(\varphi(L_k)),k=1,2,3$, are on a block in ${{\mathcal C}}_1$. By Lemma \[lemma C1\], $L_1$, $L_2$, $L_3$ are $x$-parallel for some $x$ on $L$. Since $J$ meets $L_1,L_2,L_3$ but does not pass though $x$, there is an O’Nan configuration, contradicting (I). Suppose $\mu(\varphi(L_k)),k=1,2,3$, are on a block in ${{\mathcal C}}_2$ which contains $\mu(\varphi(A_i))$ and $\mu(\varphi(A_j))$. Let $z$ be a point intersection of $J$ and $L_1$. Let $K_i, K_j$ be the lines through $z$ that pass through $x_i$ and $x_j$ respectively. By Lemma \[lemma C2\], ${\overline{L_2} }, {\overline{L_3} }$ are distinct circles in ${{\mathcal F}}(K_i, K_j)$ in ${{\mathcal I}}(z)$. Since circles in a flock are disjoint, ${\overline{L_2} }$ and ${\overline{L_3} }$ are disjoint. This contradicts that $J$ is a line through $z$ meeting $L_2$ and $L_3$. Suppose $\mu(\varphi(L_k)),k=1,2,3$, are on a block $C(J_2)$ in ${{\mathcal C}}_0^$. Suppose $J$ and $J'$ are not $L$-parallel or $L$-non-parallel. We claim that $L_2 \\|_y L_3$ for any point $y \in L_1$. If the claim is true, then $L_1,L_2,L_3$ is a self-polar triangle by Theorem \[thm LMN\] and so ${\mathcal S^{}_{L_2} } \cap {\mathcal S^{}_{L_3} } = \{L_1\}$. Since $L\in {\mathcal S^{}_{L_2} } \cap {\mathcal S^{}_{L_3} }$ by condition ($p$) and $L \neq L_1$, a contradiction rises. It follows that $J$ and $J'$ are $L$-parallel or $L$-non-parallel, and $C(J) = C(J')$. To see that $L_2 \\|_y L_3$ for any $y \in L_1$. Note that, if $y$ is a point on $L_1$, then the four lines on $y$ which are respectively $L$-parallel to $J_1$, $L$-non-parallel to $J_1$, $L$-parallel to $J_2$ and $L$-non-parallel to $J_2$, meet both $L_2$ and $L_3$. Thus by Wilbrink [@Wil Lemma 1], $L_2 \\|_y L_3$. Hence, a block in ${{\mathcal C}}_0^$ has at most two common points with any other block. Thus, ${{\mathcal I}}_2$ is an inversive plane of order $n$. To prove the main theorem in this section, we need Lemma \[lemma C0\] and that fact that when $q$ is even, if a regular spread of ${\mbox{\rm PG}}(3,q)$ is partitioned into two lines and $q-1$ reguli, then the two lines and the hyperbolic quadrics containing the reguli lie in a pencil of quadrics (Hirschfeld [@Hir2 Lemma 17.1.5, Corollary of Theorem 17.1.6]). \[thm pencil\] Let $M$, $N$ be lines in ${\mathcal S^{}_{L} }$. Suppose $L,M,N$ form a self-polar triangle with respect to ${{\mathcal U}}$. Consider the lines $L^1_{1},L^1_{2},\cdots,L^1_{n + 1}$, $\cdots$, $L^{n - 2}_{1},\cdots,L^{n - 2}_{n + 1}$, $M^1_{1},M^1_{2},\cdots,M^1_{n + 1}$ as obtained in the construction described in Theorem \[thm bigP\]. For $i=1,2,\cdots,n-1$, let ${{\mathcal H}}_i$ be the image set of the points on $L^i_1,L^i_{2},\cdots,L^i_{n + 1}$ under $\mu\varphi$. Consider ${{\mathcal H}}$ is defined in Section \[st 4\]. Then $\{ {{\mathcal H}}_i \mid i=1,2,\cdots,n-2\}\cup \{{{\mathcal H}}\} \cup \{ \mu(\varphi(M)),\mu(\varphi(N)) \}$ is a pencil of quadrics in $\Sigma$ of two lines and $n-1$ hyperbolic quadrics. By Lemma \[lemma C0\], for $i=1,2,\cdots,n-2$, the block $C(M^i_1)=\{\mu(\varphi(L^i_1))$, $\mu(\varphi(L^i_{2}))$, $\cdots$, $\mu(\varphi(L^i_{n + 1})) \}$ is a regulus. Thus ${{\mathcal H}}_i$ is a hyperbolic quadric. By the definition of ${{\mathscr S}}$, ${{\mathscr S}}= \Big(\bigcup_{i = 1}^{n - 2} C(M^i_1) \Big) \cup {{\mathcal R}}_0 \cup \{\mu(\varphi(M)), \mu(\varphi(N))\}$. Since ${{\mathscr S}}$ is regular by Theorem \[thm regular\], the result follows by [@Hir2] mentioned above. Completion of the proof that ${{\mathcal U}}$ is classical {#st 6} ========================================================== In this section, we use the notations in Section \[st 4\] and complete the proof that ${{\mathcal U}}$ is classical. Recall from Section \[st 1\] that we wish to show that ${{\mathcal U}}$ is isomorphic to the hyperbolic Buekenhout unital ${{\mathcal U}}'$ in ${\mbox{\rm PG}}(2,n^2) \cong \overline{ \pi({{\mathscr S}})}$ defined by ${{\mathcal P}}$ in Section \[st 4\] under the Bruck-Bose representation [@BB1; @BB2]. To this end, we define an isomorphism $\varphi'$ from $\varphi$, where $\varphi$ is the isomorphism between $GQ(L)$ and $Q(4,n)$ defined in . \[lemma Jcoplanar\] Let $J$ be a line of ${{\mathcal U}}$ missing $L$ and not in ${{\mathcal S}}_L$. Then $\varphi(J)$ lies on a plane which contains a line of ${{\mathscr S}}$. We locate $\varphi(J)$ as a subset of an intersection of two quadratic cones, as follows. Let $M$ be the line in ${\mathcal S^{}_{L} }$ such that $J\in {\mathcal S^{}_{M} }$. By Lemma \[lemma LJzparallel\], there is a point $z \in M$ such that $L\\|_z J$. Let ${{\mathcal Q}}_0$ be the quadratic cone formed by the intersection of ${{\mathcal P}}$ and the tangent space of ${{\mathcal P}}$ at $\varphi(z)$. By the definition of $GQ(L)$, $$\label{eqn 2} \varphi(J) \subset {{\mathcal Q}}_0.$$ By Lemma \[lemma projcone\], $\mu({{\mathcal Q}}_0)\cap {{\mathcal H}}$ is a base of ${{\mathcal Q}}_0$ and its nucleus is $\mu(\varphi(z))$. On the other hand, let ${{\mathcal Q}}_1$ be the cone with vertex $\bf N$ and a base $\mu(\varphi(J))$. By the definition of $\mu$, we have $$\label{eqn 3} \varphi(J) \subset {{\mathcal Q}}_1.$$ We are going to show ${{\mathcal Q}}_1$ is a quadratic cone by studying $\mu(\varphi(J))$. By Lemma \[lemma projcone\] and , $\mu(\varphi(J))$ lies on the plane $\mu({{\mathcal Q}}_0)$. Let ${{\mathcal H}}_1 = \{ \mu(\varphi(y)) \mid y$ is a point of ${{\mathcal U}}$ on a line of ${\mathcal S^{}_{L} }$ meeting $J\}$. By Lemma \[lemma C0\], ${{\mathcal H}}_1$ is a hyperbolic quadric in $\Sigma$. Note that $\mu(\varphi(J)) \subset {{\mathcal H}}_{1}$. To see that the plane $\mu({{\mathcal Q}}_0)$ is secant to ${{\mathcal H}}_1$, consider the line $N$ of ${{\mathcal U}}$ such that $L,M,N$ form a self-polar triangle with respect to ${{\mathcal U}}$. By Theorem \[thm LMN\], $L \\|_z N$. Thus, ${{\mathcal Q}}_0$ contains $\varphi(N)$, and the plane $\mu({{\mathcal Q}}_0)$ contains $\mu(\varphi(N))$. Since $L\\|_z J \\|_z N$ and $N\in {\mathcal S^{}_{L} }$, we conclude that $N$ is disjoint from any line of ${\mathcal S^{}_{L} }$ that meets $J$. Thus, $\mu(\varphi(N))$ is external to ${{\mathcal H}}_{1}$. Thus, $\mu(\varphi(J))=\mu({{\mathcal Q}}_0)\cap {{\mathcal H}}_1$ and the base $\mu(\varphi(J))$ of ${{\mathcal Q}}_1$ is an irreducible conic (Figure \[fig lemma61\]). \[fig lemma61\] We then study ${{\mathcal Q}}_0 \cap {{\mathcal Q}}_1$. Let $\Pi$ be the plane determined by three distinct points in ${{\mathcal Q}}_0 \cap {{\mathcal Q}}_1$. Then ${{\mathcal Q}}_0 \cap \Pi$ and ${{\mathcal Q}}_1 \cap \Pi$ are irreducible conics. Since $L$ is $z$-parallel to $J$, the nucleus of $\mu(\varphi(J))$ is $\mu(\varphi(z))$. Since the nuclei of ${{\mathcal Q}}_0 \cap \mu ({{\mathcal Q}}_0)$ and ${{\mathcal Q}}_1 \cap \mu ({{\mathcal Q}}_0)$ are both $\mu(\varphi(z))$, the conics ${{\mathcal Q}}_0 \cap \Pi$ and ${{\mathcal Q}}_1 \cap \Pi$ have a same nucleus, namely the intersection of $\Pi$ and the line through $\bf N$, $\varphi(z)$ and $\mu(\varphi(z))$. Since ${{\mathcal Q}}_0 \cap \Pi$ and ${{\mathcal Q}}_1 \cap \Pi$ are irreducible conics with the same nucleus and containing three distinct common points, ${{\mathcal Q}}_0 \cap \Pi={{\mathcal Q}}_1 \cap \Pi$ by Lemma 2.1 of Luyckx [@L]. Thus, $\varphi(J)\subset \Pi$. By Theorem \[thm pencil\], ${{\mathcal H}}_1, {{\mathcal H}},\mu(\varphi(M)),\mu(\varphi(N))$ are quadrics in a same pencil. Thus, ${{\mathcal H}}_{1}$ and ${{\mathcal H}}$ meet $\mu(\varphi(N))$ in the same conjugate pair of points with respect to $\mathbb F_{q^2}$ [@BH Theorem 5.1]. Hence the bases of ${{\mathcal Q}}_0$ and ${{\mathcal Q}}_1$ pass through the conjugate pair of points in ${\mbox{\rm PG}}(4,q^2)$, and so do ${{\mathcal Q}}_0 \cap {{\mathcal Q}}_1$ and $\Pi$. Hence, the plane $\Pi$, where $\varphi(J)$ lies on, contains the line $\mu(\varphi(N)) \in {{\mathscr S}}$. We prove that a unital ${{\mathcal U}}$ is classical by studying the hyperbolic Buekenhout unital ${{\mathcal U}}'$ in $\overline{ \pi({{\mathscr S}})}$ defined by ${{\mathcal P}}$, where $\overline{ \pi({{\mathscr S}})}$ is the projective plane constructed by the Bruck-Bose construction [@BB1; @BB2] (see André [@An] for an alternative treatment). \[thm main\] ${{\mathcal U}}$ is classical. Consider the incidence structure $\pi({{\mathscr S}})$ whose points are the affine points of ${\mbox{\rm PG}}(4,q)$ and whose lines are the affine planes of ${\mbox{\rm PG}}(4,q)$ each containing a line of ${{\mathscr S}}$. The incidence of $\pi({{\mathscr S}})$ is the incidence of ${\mbox{\rm PG}}(4,q)$. By Bruck and Bose [@BB1], $\pi({{\mathscr S}})$ is an affine translation plane of order $q^2$. Complete $\pi({{\mathscr S}})$ to a projective plane $\overline{ \pi({{\mathscr S}})}$ by adding a line at infinity, $L_\infty$. Since ${{\mathscr S}}$ is regular, $\overline{ \pi({{\mathscr S}})}$ is Desarguesian [@BB2]. By Buekenhout [@Bkt], ${{\mathcal P}}$ corresponds to a hyperbolic Buekenhout unital ${{\mathcal U}}'$ in $\overline{ \pi({{\mathscr S}})}$. Let $a_i$ be the points on $L_\infty$ corresponding to the $\varphi(A_i)$’s in ${{\mathcal R}}_0$. Then the point set of ${{\mathcal U}}'$ is $({{\mathcal P}}\setminus \Sigma) \cup \{a_1,a_2,\cdots,a_{q + 1}\}$. Since ${{\mathscr S}}$ is regular by Theorem \[thm regular\], ${{\mathcal U}}'$ is classical by Barwick [@Bar]. Let $\varphi': {{\mathcal U}}\longrightarrow {{\mathcal U}}'$ be a function defined by $\varphi'(x_i) = a_i $ for $ i = 1,2,\cdots, n + 1$; $\varphi'(y) = \varphi(y) $ for any point $y$ of ${{\mathcal U}}$ not on $L$; $\varphi'(J) = \{\varphi(y) \mid y \in J \}$ for any line $J$ of ${{\mathcal U}}$ missing $L$; $\varphi'(K) = \{\varphi(y) \mid y \in K \setminus L \} \cup \{a_i\}$ for any line $K$ of ${{\mathcal U}}$ meeting $L$ at some point $x_i$; $\varphi'(L) = \{a_1,a_2,\cdots,a_{q + 1}\}$. Note that $\varphi'$ is a well-defined function. Indeed, for any line $J_1 \notin {\mathcal S^{}_{L} }$ of ${{\mathcal U}}$ missing $L$, its image $\varphi'(J_1)$ is on a secant plane on a line of ${{\mathscr S}}$ by Lemma \[lemma Jcoplanar\], and hence is a line of ${{\mathcal U}}'$. As for a line $J_2 \in {\mathcal S^{}_{L} }$ of ${{\mathcal U}}$ missing $L$, its image $\varphi'(J_2)$ is on the secant plane of ${{\mathcal P}}$ determined by the point $\bf N$ and the line $\mu(\varphi(J_2))\in {{\mathscr S}}$, and hence $\varphi'(J_2)$ is a line of ${{\mathcal U}}'$. As for a line $K$ meeting $L$ at some $x_i$, its image $\varphi'(K)$ consists of $a_i$ and the $n$ affine points of ${\mbox{\rm PG}}(4,n)$ on a line of ${{\mathcal P}}$ meeting $\varphi(A_i)$, and hence is a line of ${{\mathcal U}}'$. Since $\varphi$ is an isomorphism, $\varphi'$ preserves incidence. Clearly, $\varphi'$ is injective. Thus, $\varphi'$ is a design isomorphism and ${{\mathcal U}}$ is classical. By [@HW2 Theorem 1.1], since ${{\mathcal U}}$ satisfies ($p$), ${{\mathcal U}}$ can be embedded in a projective plane $\pi$ as a polar unital. The author does not know whether $\pi$ is Desarguesian or not. This article is part of the author’s Ph.D. thesis, written under the supervision of Philip P.W. Wong at the University of Hong Kong. The author acknowledges the suggestions of Philip Wong during the preparation of the paper. The author also wish to thank Tim Penttila and Yee Ka Tai for their suggestions of simplifying the proof of Lemma \[lemma Jcoplanar\]. [1]{} J. André, Über nicht-Desarguessche Ebenen mit transitiver Translationsgruppe, [Math. Z.]{} [60]{} (1954) 156–186. R.D. Baker, G.L. Ebert, G. Korchmáros, T. Szőnyi, Orthogonally divergent spreads of Hermitian curves, in: F.de Clerck, J. Hirschfeld (Eds.), [Finite Geometry and Combinatorics]{} in: London Math. Soc. Lecture Note Ser., vol.13, Cambridge Univ. Press, Cambridge, 1993, pp.17–30. S.G. Barwick, A characterization of the classical unital, [Geom. Dedicata]{} [52]{} (1994) 175–180. S.G. Barwick, G. Ebert, [Unitals in Projective Planes]{}, Springer Monogr. Math., Springer-Verlag, New York, NY, 2008. R.H. Bruck, Construction problems of finite projective planes, in R.C. Bose, T.A. Dowling (Eds.), [Combinatorial mathematics and its applications]{}, Chapter 27, Univ. of North Carolina Press, Chapel Hill, 1969, pp. 426–514. R.H. Bruck, R.C. Bose, The construction of translation planes from projective spaces, [J. Algebra]{} [1]{} (1964) 85–102. R.H. Bruck, R.C. Bose, Linear representations of projective planes in projective spaces, [J. Algebra]{} [4]{} (1966) 117–172. A.A. Bruen, J.W.P. Hirschfeld. Intersections in Projective Space II: Pencils of Quadrics. [European J. Combin.]{} [9]{} (1988) 255–270. F. Buekenhout, Existence of unitals in finite translation planes of order $q^2$ with a kernel of order $q$, [Geom. Dedicata]{} [5]{} (1976) 189–194. P.J. Cameron, N. Knarr, Tubes in ${\mbox{\rm PG}}(3, q)$, [Europ. J. Combin.]{} [27]{} (2006) 114–124. P. Dembowski, Inversive planes of even order, [Bull. Amer. Math. Soc.]{} [69]{} (1963) 850–854. P. Dembowski, [Finite Geometries]{}, Springer-Verlag, New York, NY, 1968. P. Dembowski, D.R. Hughes, On finite inversive planes, [J. London Math. Soc.]{} [40]{} (1965) 171–182. J. Dover, Subregular spreads of Hermitian unitals, [Des. Codes Cryptogr.]{} [39]{} (2006) 5–15. J.W.P. Hirschfeld, [Projective Geometries over Finite Fields]{}, second ed., Clarendon Press, Oxford, 1998. J.W.P. Hirschfeld, [Finite Projective Spaces of Three Dimensions]{}, Oxford Univ. Press, Oxford, 1985. D.R. Hughes, F.C. Piper, [Projective Planes]{}, GTM, vol. 6, Springer-Verlag, Berlin, 1973. A.M.W. Hui, P.P.W. Wong, On embedding a unitary block design as a polar unital and an intrinsic characterization of the classical unital, [J. Combin. Theory Ser. A]{} [122]{} (2014) 39–52. A.M.W. Hui, Extending some induced substructures of an inversive plane, submitted to [Design Codes Cryptogr.]{} D. Luyckx, A geometric construction of the hyperbolic fibrations associated with a flock, $q$ even, [Design Codes Cryptogr.]{} [39]{} (2006) 281–288. M.E. O’Nan, Automorphisms of unitary block designs, [J. Algebra]{} [20]{} (1972) 495–511. S.E. Payne, J.A. Thas, [Finite Generalized Quadrangles*]{}, second ed., EMS Ser. Lect. Math., European Mathematical Society (EMS), Zürich, 2009. F.C. Piper, Unitary block designs, in: R.J. Wilson (Ed.), Graph Theory and Combinatorics, in: Res. Notes Math., vol. 34, Pitman, Boston, 1981, pp. 98–105. J.A. Thas, Flocks of finite egglike inversive planes, in: A. Barlotti (Ed.), Finite Geometric Structures and their Applications. C.I.M.E., II Ciclo, Bressanone, Edizioni Cremonese, Rome, 1972, pp. 189–191. H. Wilbrink, A characterization of the classical unitals, in: N.L. Johnson, M.J. Kallahar, C.T. Long (Eds.), Finite Geometries, in: Lect. Notes Pure Appl. Math., vol. 82, Marcel Dekker, New York, 1983, pp. 445–454.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We provide conditions that classify cover times for sequences of random walks on random graphs into two types: One type (Type 1) is the class of cover times that are of the order of the maximal hitting times scaled by the logarithm of the size of vertex sets. The other type (Type 2) is the class of cover times that are of the order of the maximal hitting times. The conditions are described by some parameters determined by the underlying graphs: the volumes, the diameters with respect to the resistance metric, the coverings or packings by balls in the resistance metric. We apply the conditions to and classify a number of examples, such as supercritical Galton-Watson trees, the incipient infinite cluster of a critical Galton-Watson tree and the Sierpinski gasket graph.' author: - 'Yoshihiro Abe [^1]' title: Cover times for sequences of reversible Markov chains on random graphs --- [Keywords]{}: Cover time; Maximal hitting time; Random graph; Covering;\ Packing\ [2010 Mathematics Subject Classification]{}: Primary 60J10\ Secondary 05C80 Introduction and main results ============================= Introduction {#intro} ------------ Let $G = (V(G), E(G))$ be a finite, connected graph and $\tau_{\text{cov}} (G)$ be the first time at which the simple random walk on $G$ visits every vertex. The cover time for the simple random walk is defined by $$t_{\text{cov}} (G) := \max_{x \in V(G)} E^x (\tau_{\text{cov}} (G)).$$ Cover times depend deeply on structural properties of the underlying graphs. Erdős-Rényi random graphs in several regimes are good examples. It is well known that as the percolation probability changes from the supercritical regime to the critical regime, the structure of the Erdős-Rényi random graph (such as the volume, the diameter) evolves. Cooper and Frieze [@CF1] and Barlow, Ding, Nachmias and Peres [@BDNP] estimated the cover time for the simple random walk on the Erdős-Rényi random graph in the supercritical and critical cases, respectively and showed that the order of the cover time also evolves. We will investigate the relationship between cover times and structures of the underlying graphs in a more general setting. In order to introduce our general framework, we consider the maximal hitting time defined by $$t_{\text{hit}} (G) := \max_{x, y \in V(G)} E^x (\tau_y (G)),$$ where $\tau_x (G)$ is the hitting time of $x$ by the simple random walk on $G.$\ In general, the following inequality holds for any finite, connected graphs: $$t_{\text{hit}} (G) \le t_{\text{cov}} (G) \le 2t_{\text{hit}} (G) \cdot \log \|V(G)\|. \label{ineq-cov-hit}$$ The inequality on the right-hand side is often called Matthews bound (see Lemma \[matthews\]). In view of (\[ineq-cov-hit\]), it is useful to classify cover times into the following two extreme types (see Definition \[deftype1,2\] for the precise definition):\ \ (i) cover times that are of the order of $t_{\text{hit}} (G) \cdot \log\|V(G)\|$\ (we will call them Type 1),\ (ii) cover times that are of the order of $t_{\text{hit}} (G)$ (we will call them Type 2).\ Note that the maximal hitting time can be estimated via the volume and the diameter with respect to the resistance metric of the underlying graph (see Lemma \[commute time\] for the precise statement). In this paper, we will provide sufficient conditions that classify cover times for a sequence of random walks on random graphs into Type 1 and Type 2 in terms of the volume, the resistance diameter and the covering or packing number of the underlying graphs (see section \[sec:1\] for precise definitions of these parameters). We apply the conditions to many examples (see Table 1 below). Although details of some specific cover times are already known, the novelty of this paper is that we first unify separate methods of estimating cover times into one and add some new examples such as supercritical Galton-Watson trees and critical Galton-Watson trees conditioned to survive. We provide intuitions for the sufficient conditions. Roughly speaking, if one can find a packing consisting of a large number of big disjoint balls with respect to the effective resistance metric, then the cover times will be of Type 1 (Theorem \[type1\]). Many supercritical random graphs admit such packings. For example, we can take a family of large number of big trees as a packing for supercritical Galton-Watson family trees and supercritical Erdős-Rényi random graphs (see section \[sec:7\], \[sec:9\]). On the other hand, it can be shown that cover times will be of Type 2 if the number of balls required to cover the underlying graphs increases no more than (double) exponentially, as the radii of balls with respect to the resistance metric decrease exponentially (Theorem \[type2\]). A wide variety of critical random graphs and fractal graphs satisfy this property (see section \[sec:11\], \[sec:12\], \[sec:14\]). General bounds on cover times have been studied previously(see [@MCMT], [@BDNP], [@DLP]). The Matthews bound (see Lemma \[matthews\]) and the lower bound in terms of Gaussian free fields [@DLP] together with the Sudakov minoration (see Lemma \[sudakov\]) give very useful ingredients for obtaining the condition for Type 1. The upper bound via Gaussian free fields [@DLP] and the Dudley’s entropy bound (see Lemma \[dudley\]) are essential to the conditions for Type 2. In the next subsection, we give our main results. For a set $S$, we will write $\|S\|$ to denote the cardinality of $S$. Throughout this paper, we use $c, c^{\prime}, c_1, c_2, \dotsc $ to denote constants that does not depend on the size of $G$. Main results {#sec:1} ------------ To state our main results, we first prepare some definitions. Let $G^N = (V(G^N), E(G^N), \mu^N), N \in \mathbb{N}$ be a sequence of random weighted graphs, where $V(G^N)$ is the vertex set, $E(G^N)$ is the edge set and $\mu^N$ is a non-negative symmetric weight function on $V(G^N) \times V(G^N)$ which satisfies $\mu^N_{xy} > 0$ if and only if $\{x,y\} $ $\in E(G^N).$ We assume that these weighted graphs are defined on a common probability space with a probability measure $\textbf{P}$ and that $G^N$ is a finite, connected graph, $\textbf{P}$-a.s. In this paper, the following four parameters (volume, resistance diameter, packing number, covering number) play important roles in estimating cover times.\ The volume of $G^N$ is defined by $$\mu^N (G^N) := \sum_{x, y \in V(G^N)} \mu_{xy}^N.$$ The effective resistance is a powerful tool for studying random walks on weighted graphs (see Lemma \[commute time\]). For $x,y \in V(G^N), x \neq y,$ we define the effective resistance between $x$ and $y$ by $$R_{\text{eff}}^N (x,y)^{-1} := \inf \{\mathcal{E}^N(f,f) : f \in \mathbb{R}^{V(G^N)}, f(x) = 1, f(y) = 0 \},$$ where $\mathcal{E}^N (f,g) := \frac{1}{2} \displaystyle \sum_{\begin{subarray}{c} u,v \in V(G^N) \\ \{u,v\} \in E(G^N) \end{subarray} } \mu^N_{uv} (f(u) - f(v))(g(u) - g(v)), f,g \in \mathbb{R}^{V(G^N)}.$\ If we define $R_{\text{eff}}^N (x,x) = 0$ for all $x \in V(G^N),$ it is known that $R_{\text{eff}}^N (\cdot, \cdot)$ is a metric on $V(G^N).$ The resistance diameter is defined by $$\text{diam}_{R}(G^N) := \displaystyle \max_{x,y \in V(G^N)} R^N_{\text{eff}}(x,y).$$ We define the resistance ball with radius $r$ centered at $x \in V(G^N)$ by $$B_{\text{eff}}^N (x, r) := \{y \in V(G^N) : R_{\text{eff}}^N (x, y) \le r\}.$$ We call a family of resistance balls $\{B_{\text{eff}}^N (x_1, r_1), \cdots, B_{\text{eff}}^N (x_{m}, r_m) \}$ a packing for $G^N$ if these resistance balls are disjoint with each other.\ The packing number for $(G^N, r)$ is defined by $$\begin{aligned} n_{\text{pac}} (G^N, r) := \max \Big \{m &\ge 1 : ~\text{there exist} ~x_1, \cdots, x_m \in V(G^N) ~\text{such that} \\ &\{B_{\text{eff}}^N (x_1, r), \cdots, B_{\text{eff}}^N (x_m, r) \}~ \text{is a packing for}~ G^N \Big \}.\end{aligned}$$ We call a family of resistance balls $\{B_{\text{eff}}^N (x_1, r_1), \cdots, B_{\text{eff}}^N (x_{m}, r_m) \}$ a covering for $G^N$ if $$V(G^N) \subset \displaystyle \bigcup_{k = 1}^m B_{\text{eff}}^N (x_k, r_k).$$ The covering number for $(G^N, r)$ is defined by $$\begin{aligned} n_{\text{cov}} (G^N, r) := \min \Big \{m &\ge 1 : ~\text{there exist} ~x_1, \cdots, x_m \in V(G^N) ~\text{such that} \\ &\{B_{\text{eff}}^N (x_1, r), \cdots, B_{\text{eff}}^N (x_m, r) \}~ \text{is a covering for}~ G^N \Big \}.\end{aligned}$$ The discrete time random walk on $G^N$ is the Markov chain $((X_n)_{n \geqslant 0}, P^x, x \in V(G^N))$ with transition probabilities $(p(x,y))_{x,y \in V(G^N)}$ defined by $p(x,y)$ $:= \mu_{xy}^N/\mu_x^N$, where $\mu_x^N := \sum_{y \in V(G^N)} \mu_{xy}^N$. Let $\tau_{\text{cov}} (G^N)$ be the first time at which the random walk visits every vertex of $V(G^N)$. We define the cover time for the random walk on $G^N$ as follows: $$t_{\text{cov}}(G^N) := \max_{x \in V(G^N)} E^x(\tau_{\text{cov}}(G^N)).$$ We also define the maximal hitting time for the random walk on $G^N$ by $$t_{\text{hit}} (G^N) := \max_{x, y \in V(G^N)} E^x(\tau_y (G^N)),$$ where $\tau_x (G^N)$ is the hitting time of $x \in V(G^N)$ by the random walk on $G^N$. We give the precise definitions of types for a sequence of cover times. \[deftype1,2\] (1) A sequence of cover times $(t_{\text{cov}} (G^N) )_{N \in \mathbb{N} }$ is Type 1 if $$\lim_{\lambda \to \infty} \liminf_{N \to \infty} \textbf{P} \left( \lambda^{-1} \le \frac{t_{\text{cov}} (G^N)}{t_{\text{hit}} (G^N) \cdot \log \|V(G^N)\|} \le 2 \right) = 1. \label{def1}$$ (2) A sequence of cover times $(t_{\text{cov}} (G^N) )_{N \in \mathbb{N} }$ is Type 2 if $$\lim_{\lambda \to \infty} \liminf_{N \to \infty} \textbf{P} \left( 1 \le \frac{t_{\text{cov}} (G^N)}{t_{\text{hit}} (G^N)} \le \lambda \right) = 1. \label{def2}$$ \[remgeneralbound\] By (\[ineq-cov-hit\]), the upper bound of the event in (\[def1\]) and the lower bound of the event in (\[def2\]) always hold. We are now ready to state our main theorems. We first state the sufficient condition for cover times to be Type 1. We will say that a sequence of events $(B_N)_{N \ge 0} $ holds with high probability (abbreviated to w.h.p.) if $\lim_{N \to \infty} \textbf{P}(B_N) = 1$. \[type1\] (1) Suppose there exist $c_1, c_2> 0$ and functions $v,r : \mathbb{N} \to [0,\infty) $ with $ \lim_{N \to \infty} v(N) = \infty$ such that w.h.p., the following holds: $$\log \|V(G^N)\| \le c_1 \log v(N),~\text{diam}_R (G^N) \le c_2 r(N). \label{condition-of-type1-0}$$ Then there exists $c_3 >0$ such that w.h.p., $$t_{\text{cov}}(G^N)/\mu^N(G^N) \le c_3 r(N)\log v(N).$$ \(2) Suppose that there exist $c_4, c_5 > 0$ and functions $v,r : \mathbb{N} \to [0,\infty) $ with $ \lim_{N \to \infty} v(N) = \infty$ such that w.h.p., $$\log \{n_{\text{pac}} (G^N, c_4 r(N)) \} \ge c_5 \log v(N). \label{condition-of-type1}$$ Then there exists $c_6 > 0$ such that w.h.p., $$t_{\text{cov}}(G^N)/\mu^N(G^N) \ge c_6 r(N)\log v(N).$$ \(3) Under conditions (\[condition-of-type1-0\]) and (\[condition-of-type1\]), $(t_{\text{cov}} (G^N))_{N \in \mathbb{N}}$ is Type 1.\ We next state sufficient conditions for cover times to be Type 2. \[type2\] (1) Suppose that there exist functions $v,r : \mathbb{N} \to [0,\infty) $ with $ \lim_{N \to \infty} v(N) = \infty$ and a function $p : [1, \infty) \to [0, 1]$ with $\lim_{\lambda \to \infty} p(\lambda) = 0$ satisfying the following for all $\lambda \ge 1$ and sufficiently large $N \in \mathbb{N}:$ $$\textbf{P}(\mu^N(G^N) \le \lambda v(N)) \ge 1 - p(\lambda), \label{condition-of-type2-(1)-1}$$ and there exists a random non-increasing sequence $(\ell_k^N)_{k \ge 0}$ satisfying $\ell_0^N = \text{diam}_R(G^N), \ell_{k_0^N - 1}^N > 0$ and $\ell_{k_0^N}^N = 0$ for some $k_0^N \in \mathbb{N}$ such that $$\textbf{P}\Big (\sum_{k = 1}^{k_0^N} \sqrt{\ell_{k-1}^N \log \{ n_{\text{cov}} (G^N, \ell_k^N) \}} \le \lambda \sqrt{r(N)} \Big) \ge 1 - p(\lambda). \label{condition-of-type2-(1)-2}$$ Then there exists $c>0$ such that for all $\lambda \ge c$ and sufficiently large $N \in \mathbb{N}$, $$\label{cov-estimate-type2} \textbf{P}(t_{\text{cov} }(G^N) > \lambda v(N) r(N)) \le \inf_{0 < \theta < 1} \left \{ p((\lambda/c)^{\theta}) + p \left((\lambda/c)^{\frac{1 - \theta}{2} } \right) \right \}.$$ (2) Suppose that there exist functions $v,r : \mathbb{N} \to [0,\infty) $ with $ \lim_{N \to \infty} v(N) = \infty$ and a function $p : [1, \infty) \to [0, 1]$ with $\lim_{\lambda \to \infty} p(\lambda) = 0$ satisfying the following for all $\lambda \ge 1$ and sufficiently large $N \in \mathbb{N}:$ $$\textbf{P}(\mu^N(G^N) < {\lambda}^{-1} v(N)) \le p(\lambda), ~\textbf{P}(\text{diam}_R(G^N) < {\lambda}^{-1} r(N) ) \le p(\lambda). \label{condition-of-type2-(3)}$$ Then there exists $c > 0$ such that for all $\lambda \ge c$ and sufficiently large $N \in \mathbb{N}$, $$\textbf{P}(t_{\text{cov} }(G^N) < {\lambda}^{-1} v(N) r(N)) \le \inf_{0 < \theta < 1} \Big \{ p \Big (\Big (\frac{\lambda}{c} \Big)^{\theta} \Big) + p \Big ( \Big (\frac{\lambda}{c} \Big)^{1 - \theta} \Big) \Big \}.$$ (3) Under the conditions (\[condition-of-type2-(1)-1\]), (\[condition-of-type2-(1)-2\]) and (\[condition-of-type2-(3)\]), $(t_{\text{cov}} (G^N))_{N \in \mathbb{N}}$ is Type 2.\ \[type2rems\] (1) In general, we can not replace (\[cov-estimate-type2\]) by the statement that $t_{\text{cov} }(G^N) \le cv(N)r(N)$ w.h.p., for some $c > 0$ (see Proposition \[cgwprop2\]). We thus state Theorem \[type1\] and Theorem \[type2\] in a slightly different way.\ (2) If the conditions (\[condition-of-type1-0\]) and (\[condition-of-type1\]) in Theorem \[type1\] hold $\textbf{P}$-almost surely for sufficiently large $N \in \mathbb{N},$ the results of Theorem \[type1\] also hold $\textbf{P}$-almost surely for sufficiently large $N \in \mathbb{N}.$\ (3) If the events of (\[condition-of-type2-(1)-1\]), (\[condition-of-type2-(1)-2\]) and (\[condition-of-type2-(3)\]) in Theorem \[type2\] hold $\textbf{P}$-almost surely for sufficiently large $N \in \mathbb{N},$ the results of Theorem \[type2\] also hold $\textbf{P}$-almost surely for sufficiently large $N \in \mathbb{N}$ ($\lambda$ will be replaced by some constants).\ (4) On some class of planar graphs, the condition (\[condition-of-type1\]) always holds; Let $(G^N)_{N \ge 0}$ be a sequence of $\textbf{P}$-a.s. finite, planar connected random graphs with maximum degree $c > 0$ and $\mu_{xy}^N = 1$ for all $\{x, y \} \in E(G^N).$ Suppose that there exists $c_7 > 0$ and a function $v : \mathbb{N} \to [0,\infty) $ with $ \lim_{N \to \infty} v(N) = \infty$ such that w.h.p., $\log \|V(G^N)\| \ge c_7 \log v(N).$ Then by Lemma 3.1 of [@JS], (\[condition-of-type1\]) holds with the function $v$ and $r(N) = \log v(N).$\ (5) Typically, we take an exponentially decreasing sequence as $(\ell_k^N)_{k \ge 0}$ in (\[condition-of-type2-(1)-2\]) (for example, $\ell_k^N = \frac{\text{diam}_R (G^N)}{2^k}$). Applying these theorems, we will estimate and classify cover times for several specific random graphs. We give a list of the results in Table 1. We explain the notation in Table 1. The notation $m$ is the mean of the offspring distribution of the corresponding branching process. Supercritical Erdős-Rényi random graphs I, II have the percolation probability $c/N$, $f(N)/N$ respectively, where $c > 1$ is a constant and $\lim_{N \to \infty} f(N)/ \sqrt{N} = \lim_{N \to \infty} \log N/ f(N) = 0.$ ‘IIC’ is the abbreviation of ‘incipient infinite cluster’ and $p_N$ is the survival probability up to $N$ level (see subsection \[sec:11\]). Table 1: Orders of volumes and cover times for random graphs and types of cover times Random graph Volume Cover time Type ----------------------------------------------------- --------------- ----------------- ------ Supercritical Galton-Watson family trees $m^N$ $N^2 m^N$ 1 Supercritical Erdős-Rényi random graphs I $N$ $N (\log N)^2$ 1 Supercritical Erdős-Rényi random graphs II $N f(N)$ $N \log N$ 1 The IIC for critical Galton-Watson family tree $N p_N^{- 1}$ $N^2 p_N^{- 1}$ 2 Critical Erdős-Rényi random graphs $N^{2/3}$ $N$ 2 The range of random walk in $\mathbb{Z}^d, d \ge 5$ $N$ $N^2$ 2 Sierpinski gasket graphs $3^N$ $5^N$ 2 \ \ Concerning the IIC for critical Galton-Watson family trees, Aldous [@Al] and Barlow, Ding, Nachmias and Peres [@BDNP] have estimated the cover times for critical Galton-Watson family trees for finite variance offspring distributions. Our result extends these results to the case where the offspring distribution is in the domain of attraction of a stable law with index $\alpha \in (1, 2].$ Our result clarifies that the cover time for the IIC depends on the survival probability of the branching process up to some level. In addition to this example, we give new estimates on cover times for supercritical Galton-Watson family trees, the range of random walk in $\mathbb{Z}^d, d \ge 5$ and Sierpinski gasket graphs. Note that for supercritical Erdős-Rényi random graphs, better estimates are already known [@CF1; @J] and that for critical Erdős-Rényi random graphs, the correct order is already known [@BDNP]. We cite these examples to compare Type 1 and Type 2. In Section \[sec:8\], we will estimate the cover time for the largest supercritical percolation cluster inside a box in $\mathbb{Z}^d, d \ge 2.$ However, we are not able to obtain the correct order (see Remark \[remscpc\]). Note that there are graphs where the cover times can not be classified as either Type 1 or Type 2. For example, let $G^N$ be a deterministic graph with unit weights consisting of a complete graph with $N$ vertices and $a_N$ other vertices, each attached by a single edge to a distinct vertex of the complete graph, where $a_N$ is a positive number satisfying $2 \le a_N \le N$. One can show that $\text{diam}_R (G^N) = 2 + 2/N,$ $n_{\text{pac}} (G^N, \ell) \ge a_N$ for all $0 \le \ell \le 1$ and $n_{\text{cov}} (G^N, \text{diam}_R (G^N)/2^k) \le a_N + 1$ for all $1 \le k \le \lfloor \log_2 N \rfloor.$ By Theorem \[type1\] (2), Lemma \[commute time\] and Lemma \[barlow-ding-nachmias-peres\] below, we have for some $c, c^{\prime} > 0,$ $$c \cdot t_{\text{hit}} (G^N) \cdot \log a_N \le t_{\text{cov}} (G^N) \le c^{\prime} \cdot t_{\text{hit}} (G^N) \cdot \log a_N.$$ This implies that if $\displaystyle \lim_{N \to \infty} a_N = \infty$ and $\displaystyle \lim_{N \to \infty} \frac{\log a_N}{\log N} = 0$, then the sequence of cover times $(t_{\text{cov}} (G^N))_{N \in \mathbb{N}}$ is neither Type 1 nor Type 2. We give the outline of this paper. In Section \[sec:2\], we prove Theorem \[type1\] and Theorem \[type2\]. In Section \[sec:6\], using Theorem \[type1\] and Theorem \[type2\], we estimate and classify cover times for the examples in Table 1. Proof of Theorem \[type1\] and Theorem \[type2\] {#sec:2} ================================================ In this section, we prove Theorem \[type1\] and Theorem \[type2\]. Known results {#sec:3} ------------- We state some known results on cover times and Gaussian free fields that we will use in this paper. Throughout the following lemmas, $G = (V(G), E(G))$ will be a finite, connected graph and $\mu$ will be the weight function with $\mu (G) := \sum_{x, y \in V(G)} \mu_{xy}.$ Let $\{\eta_x \}_{x \in V(G)}$ be the Gaussian free field on $G$ defined on a probability space with a probability measure $\mathbb{P}.$\ Recently, Ding, Lee and Peres [@DLP] proved the following surprising result, which says that cover times have a close relationship with Gaussian free fields. ([@DLP], Theorem 1.9 and Theorem (MM)) \[ding-lee-peres\] There exist $c_1, c_2 > 0$ such that $$c_1 \cdot \mu(G) \cdot \Bigg(\mathbb{E} \max_{x \in V(G) } \eta_x \Bigg)^2 \le t_{\text{cov} }(G) \le c_2 \cdot \mu(G) \cdot \Bigg(\mathbb{E} \max_{x \in V(G) } \eta_x \Bigg)^2.$$ The following commute time identity is well-known and useful for estimating the maximal hitting time. See, for instance, Theorem 2.1 of [@CRRST] or Proposition 10.6 of [@MCMT]. \[commute time\] Let $\tau_x$ be the hitting time of $x \in V(G)$ by the random walk on $G$. For all $x, y \in V(G)$, $$E^x (\tau_y) + E^y (\tau_x) = \mu(G) R_{\text{eff}} (x, y).$$ In particular, $$\frac{1}{2} \mu (G) \text{diam}_R(G) \le t_{\text{hit}} (G) \le \mu (G) \text{diam}_R (G).$$ Fix $x, y \in V(G)$. $\Pi$ is an edge-cutset between $x$ and $y$ if $\Pi$ is a subset of $E(G)$ such that every path from $x$ to $y$ has an edge belonging to $\Pi$. The following Nash-Williams inequality is useful for obtaining lower bounds on effective resistances. See, for example, Proposition 9.15 of [@MCMT]. \[Nash-Williams\] Fix $x, y \in V(G)$. Let $(\Pi_k)_{k \ge 1}$ be a sequence of edge-cutsets between $x$ and $y$ with $\Pi_k \cap \Pi_{\ell} = \emptyset$ for all $k \neq \ell$. Then, $$R_{\text{eff}} (x, y) \ge \sum_{k \ge 1} \big(\sum_{\{u, v \} \in \Pi_k} \mu_{uv} \big)^{- 1}.$$ Proof of Theorem \[type1\] {#sec:4} -------------------------- We provide the proof of Theorem \[type1\]. The following lemma is known as the Matthews bound. See, for example, Theorem 11.2 of [@MCMT] (see also the original work of Matthews [@Ma]). \[matthews\] Let $(X_n)_{n \ge 0}$ be an irreducible Markov chain on a finite state space $V$ and $t_{\text{cov}}, t_{\text{hit}}$ be its cover time and maximal hitting time, respectively. Then, $$t_{\text{cov}} \le t_{hit} \cdot (\log \|V\| + 1).$$ We also use the next fact, called Sudakov minoration. See, for instance, Lemma 2.1.2 of [@TGC]. \[sudakov\] Let $\{\eta_x \}_{x \in V(G)}$ be a Gaussian free field on a weighted graph $G$. There exists $c > 0$ such that for all $V^{'} \subset V(G)$, $$\mathbb{E} \max_{x \in V^{'}} \eta_x \ge c \Big(\min_{\begin{subarray}{c} y, z \in V^{'} \\ y \neq z \end{subarray} } \sqrt{R_{\text{eff}} (y, z)} ~\Big) \sqrt{\log \|V^{'}\|}.$$ Proof of Theorem \[type1\]. We first prove (1). By Lemma \[commute time\] and (\[condition-of-type1-0\]), we get w.h.p., $$t_{\text{hit}} (G^N) \le \mu^N (G^N) \cdot \text{diam}_R (G^N) \le c_2 \mu^N(G^N) r(N). \label{ineq-of-pf-of-thm1.1-1}$$ So, using Lemma \[matthews\], (\[condition-of-type1-0\]) and (\[ineq-of-pf-of-thm1.1-1\]) , we have that w.h.p., $$\begin{aligned} t_{\text{cov}} (G^N) &\le t_{\text{hit}} (G^N) \cdot (\log \|V(G^N)\| + 1) \\ & \le 2 c_1 c_2 \mu^N (G^N) r(N) \log v(N). \end{aligned}$$ Next, we prove (2). Let $x_1, \cdots, x_{n_{\text{pac}} (G^N, c_4 r(N))}$ be vertices satisfying that the set of resistance balls $\{B_{\text{eff}}^N (x_k, c_4 r(N)) : 1 \le k \le n_{\text{pac}} (G^N, c_4 r(N)) \}$ is a packing for $G^N.$ Set $V^{'} := \{ x_1, \cdots, x_{n_{\text{pac}} (G^N, c_4 r(N))} \}.$ Using (\[condition-of-type1\]), Lemma \[ding-lee-peres\] and Lemma \[sudakov\], we have that there exist $c_7, c_8 > 0$ such that w.h.p., $$\begin{aligned} t_{\text{cov}} (G^N) &\ge c_7 \mu^N (G^N) \Big (c_8 \sqrt{c_4 r(N)} \sqrt{\log \{n_{\text{pac}} (G^N, c_4 r(N)) \}} \Big )^2 \notag \\ &\ge c_4 c_5 c_7 c_8^2 ~\mu^N (G^N) r(N) \log v(N). \label{ineq-of-pf-of-thm1.1-2} \end{aligned}$$ The inequalities (\[condition-of-type1-0\]), (\[ineq-of-pf-of-thm1.1-1\]) and (\[ineq-of-pf-of-thm1.1-2\]) imply the conclusion of (3). $\Box$ Proof of Theorem \[type2\] {#sec:5} -------------------------- We prove Theorem \[type2\]. The following fact is a minor extension of Theorem 1.1 of [@BDNP] and provides useful general upper bounds on cover times. \[barlow-ding-nachmias-peres\] Let $G = (V(G), E(G))$ be a graph and $\mu$ be the weight function with $\mu (G) := \sum_{x, y \in V(G)} \mu_{xy}.$\ Let $(\ell_k)_{k \ge 0}$ be a non-increasing sequence with $\ell_0 = \text{diam}_R (G), \ell_{k_0 - 1} > 0$ and $\ell_{k_0} = 0$ for some $k_0 \in \mathbb{N}.$\ There exists $c > 0$ such that $$t_{\text{cov} }(G) \le c \bigg(\sum_{k = 1}^{k_0} \sqrt{\ell_{k-1} \log \{n_{\text{cov}} (G, \ell_k) \} } \bigg)^2 \cdot \mu(G).$$ Lemma \[barlow-ding-nachmias-peres\] follows from the following result. See, for example, Theorem 11.17 of [@LT]. \[dudley\] Let $I$ be a finite set and $\{\eta_x \}_{x \in I}$ be a Gaussian process. Set $d (x,y) := \sqrt{\mathbb{E} (\eta_x - \eta_y)^2}$ and $$\begin{aligned} n(I, d, \ell) := \min \{m \ge 1 : ~&\text{there exist}~x_1, \cdots, x_m \in I \\ &\text{such that}~I \subset \bigcup_{k = 1}^m \{y \in I : d(x_k, y) \le \ell \} \}.\end{aligned}$$ Then there exists $c > 0$ such that $$\mathbb{E} \max_{x \in I} \eta_x \le c \int_0^{\infty} \sqrt{\log \{n (I, d, \ell) \} } d\ell.$$ Proof of Lemma \[barlow-ding-nachmias-peres\]. Let $\{\eta_x \}_{x \in V(G)}$ be a Gaussian free field on $G.$ Note that $d(x,y) = \sqrt{\mathbb{E} (\eta_x - \eta_y)^2} = \sqrt{R_{\text{eff}} (x, y)}.$\ In particular, $n(V(G), d, \ell) = n_{\text{cov}} (G, \ell^2).$\ Since $n_{\text{cov}} (G, \ell)$ is non-increasing with respect to $\ell,$ we have $$\begin{aligned} &~~~\int_0^{\infty} \sqrt{\log \{n (V(G), d, \ell) \} } d\ell \notag \\ &\le \int_0^{\infty} \sqrt{\log \{n_{\text{cov}} (G, \ell^2) \} } d\ell \notag \\ &\le \sum_{k = 1}^{k_0} \int_{\sqrt{\ell_k}}^{\sqrt{\ell_{k-1}}} \sqrt{\log \{n_{\text{cov}} (G, \ell^2) \} } d\ell \notag \\ &\le \sum_{k = 1}^{k_0} \sqrt{\ell_{k-1} \log \{n_{\text{cov}} (G, \ell_k) \} }. \label{dudley-bound}\end{aligned}$$ Lemma \[ding-lee-peres\], Lemma \[dudley\] and (\[dudley-bound\]) imply the conclusion. $\Box$ Proof of Theorem \[type2\]. First, we prove (1). Fix $\lambda \ge 1,$ sufficiently large $N \in \mathbb{N}$ and $\theta \in (0, 1).$ Set $$B := \Big \{ \sum_{k = 1}^{k_0^N} \sqrt{\ell_{k-1}^N \log \{ n_{\text{cov}} (G^N, \ell_k^N) \}} \le \lambda^{\frac{1-\theta}{2}} \sqrt{r(N)} \Big \}.$$ By (\[condition-of-type2-(1)-1\]), (\[condition-of-type2-(1)-2\]) and Lemma \[barlow-ding-nachmias-peres\], we have for some $c_1 > 0$ that $$\begin{aligned} &~~~\textbf{P} (t_{\text{cov}} (G^N) > c_1 \lambda v(N) r(N) ) \\ &\le \textbf{P} (\mu^N(G^N) > \lambda^{\theta} v(N) ) + \textbf{P} (B^c)\\ &\le p(\lambda^{\theta}) + p(\lambda^{\frac{1 - \theta}{2}}),\end{aligned}$$ which implies the conclusion of (1). Next, we prove (2). Fix $\lambda \ge 1,$ sufficiently large $N \in \mathbb{N},$ and $\theta \in (0, 1).$ By (\[condition-of-type2-(3)\]), Lemma \[commute time\] and the fact that $t_{\text{cov}} (G^N) \ge t_{\text{hit}} (G^N)$ $\textbf{P}$-a.s., we have that $$\begin{aligned} &~~~\textbf{P} \left(t_{\text{cov}} (G^N) < \frac{\lambda^{- 1}}{2} v(N) r(N) \right ) \\ &\le \textbf{P} (\mu^N (G^N) < \lambda^{- \theta} v(N) ) + \textbf{P} (\text{diam}_R(G^N) < \lambda^{- (1 - \theta )} r(N) ) \\ &\le p(\lambda^{\theta}) + p(\lambda^{1 - \theta}),\end{aligned}$$ which implies the conclusion of (2). Using Lemma \[commute time\] and the results of (1) and (2), we can easily obtain the conclusion (3). We omit the detail. $\Box$ Examples {#sec:6} ======== In this section, we estimate and classify cover times for a number of specific random graphs by using Theorem \[type1\] and Theorem \[type2\]. Given a graph $G$, we will write $d_G (x, y)$ to denote the graph distance between $x$ and $y$ in the graph $G$. From Subsection 3.1 to 3.7, we assume that $\mu_{xy}^N = 1$ for all $\{x, y \} \in E(G^N)$ and $N \in \mathbb{N}$ $\textbf{P}$-a.s. Supercritical Galton-Watson family trees {#sec:7} ---------------------------------------- Let $(Z_N)_{N \ge 0}$ be a Galton-Watson process defined on a probability space with probability measure $\mathbb{P}$ and $\mathcal{T}$ be its family tree. We assume that $m := \mathbb{E} (Z_1) \in (1, \infty).$ $\mathcal{T}_{\le N}$ and $\mathcal{T}_N$ are the first $N$ generations and the set of $N$-th generation of $\mathcal{T}$ respectively. In particular, $Z_N = \|\mathcal{T}_N\|.$ $\Tilde{\mathcal{T}}_N$ is a set of vertices among $N$-th generation that have infinite line of descent. We consider the conditional measure $\textbf{P} := \mathbb{P}(~ \cdot ~ \|~Z_n \neq 0 ~\text{for all}~ n \in \mathbb{N})$. We prove the following proposition. \[scgwprop\] There exist $c_1, c_2 > 0$ such that $\textbf{P}$-a.s., for sufficiently large $N \in \mathbb{N},$ $$c_1 N^2 \le t_{cov}(\mathcal{T}_{\le N}) / \|E(\mathcal{T}_{\le N})\| \le c_2 N^2,$$ and $(t_{cov}(\mathcal{T}_{\le N}))_{N \in \mathbb{N}}$ is Type 1. In the proof, we use the following well-known fact. See, for example, Theorem 1 (page 49), Theorem 3 (page 30) and Lemma 4 (page 31) of [@AN]. \[scgwlem\] Let $(Z_N)_{N \ge 0}$ be a Galton-Watson process with mean $m \in (1, \infty).$\ (1) Set $\tilde{Z}_N := \|\tilde{\mathcal{T}}_N\|.$ Under the probability measure $\mathbb{P} (\cdot \| Z_n \neq 0~ \text{for all}~ n \in \mathbb{N}),$ $(\tilde{Z}_N)_{N \ge 0}$ is a Galton-Watson process whose offspring distribution has generating function $$\tilde{f} (s) = \frac{f((1-q)s + q) - q}{1 - q},$$ where $f$ is the generating function of $Z_1$ and $q := \mathbb{P} (Z_n = 0~ \text{for some}~ n \in \mathbb{N}).$\ (2) There exist a sequence of constants $(C_N)_{N \in \mathbb{N}}$ with $\displaystyle \lim_{N \to \infty} C_N = \infty$ and $\displaystyle \lim_{N \to \infty} \frac{C_{N + 1}}{C_N} = m$ and a random variable $W$ such that $$\lim_{N \to \infty} \frac{Z_N}{C_N} = W ~\mathbb{P} \text{-a.s.}, ~\mathbb{P} (W < \infty) = 1 ~\text{and}~\mathbb{P} (W = 0) = q.$$ Proof of Proposition \[scgwprop\]. We check almost-sure versions of (\[condition-of-type1-0\]) and (\[condition-of-type1\]) in Theorem \[type1\] with $\log v(N) = r(N) = N.$\ By the Chebyshev inequality, we have for all $\alpha > m,$ $$\textbf{P} (\|\mathcal{T}_{\le N}\| > \alpha^N) \le \frac{\textbf{E} (\|\mathcal{T}_{\le N}\|)}{\alpha^N} \le \frac{1}{1-q} \cdot \frac{m}{m - 1} \cdot \Big (\frac{m}{\alpha} \Big )^N.$$ So, by the Borel-Cantelli lemma, $\|\mathcal{T}_{\le N}\| \le \alpha^N$ for sufficiently large $N \in \mathbb{N},$ $\textbf{P}$-a.s. Since $R_{\text{eff}}^N (x,y) = d_{\mathcal{T}_{\le N}} (x,y)$ for all $x,y \in \mathcal{T}_{\le N},$ we get $\text{diam}_R (\mathcal{T}_{\le N} ) \le 2N,$ $\textbf{P}$-a.s. We set $V^{'} := \{g_N(v) : v \in \Tilde{\mathcal{T}}_{\lfloor \frac{N}{2} \rfloor} \}$, where $g_N(v) \in \mathcal{T}_N$ is a fixed descendant of $v \in \Tilde{\mathcal{T}}^{~}_{\lfloor \frac{N}{2} \rfloor}.$ We also set $\tilde{Z}_N := \|\tilde{\mathcal{T}}_N\|.$ By Lemma \[scgwlem\] (1), $(\tilde{Z}_N)_{N \ge 0}$ is a Galton-Watson process with mean $m$ and zero extinction probability. By applying Lemma \[scgwlem\] (2) to $(\tilde{Z}_N)_{N \ge 0},$ we have $$\lim_{N \to \infty} \frac{\tilde{Z}_{N+1}}{\tilde{Z}_N} = m, \textbf{P} \text{-a.s., and so} \lim_{N \to \infty} (\tilde{Z}_N)^{1/N} = m, \textbf{P} \text{-a.s.}$$ In particular, we have $\|V^{'}\| = \tilde{Z}_{\lfloor \frac{N}{2} \rfloor} \ge {\alpha}^{\lfloor \frac{N}{2} \rfloor}$ for sufficiently large $N \in \mathbb{N},$ $\textbf{P}$-a.s., for all $1 < \alpha < m.$ We also know that $R_{\text{eff} }^N(x, y) > 2 \lfloor \frac{N}{2} \rfloor$ for all $x, y \in V^{'}, x \neq y,$ $\textbf{P}$-a.s. Therefore, $\{B_{\text{eff}}^N (x, \lfloor \frac{N}{2} \rfloor) : x \in V^{'} \}$ is a packing for $\mathcal{T}_{\le N}$ and $\log \{n_{\text{pac}} (\mathcal{T}_{\le N}, \lfloor \frac{N}{2} \rfloor) \} \ge \lfloor \frac{N}{2} \rfloor \log \alpha,$ for sufficiently large $N \in \mathbb{N},$ $\textbf{P}$-a.s., for all $1 < \alpha < m.$ By Remark \[type2rems\] (2), the conclusion holds. $\Box$\ The largest supercritical percolation cluster inside a box in $\mathbb{Z}^d$ {#sec:8} ---------------------------------------------------------------------------- We consider Bernoulli bond percolation model on $\mathbb{Z}^d$. In this model, each edge in $\mathbb{E}^d$ is open with probability $p$ and closed with probability $1 - p$ independently, where $\mathbb{E}^d := \{ \{x, y \} : x, y \in \mathbb{Z}^d, \sum_{i = 1}^d \|x_i - y_i\| = 1 \}$ and $x_i$ is the $i$ th coordinate of $x \in \mathbb{Z}^d$. We write the corresponding probability measure on $\{0, 1 \}^{\mathbb{E}^d}$ by $\textbf{P}_p$. A sequence $\Gamma = (x^0, \dotsc, x^n)$ is an open path in $S \subset \mathbb{Z}^d$ connecting $x$ and $y$ if $x^0 = x, x^n = y, x^i \in S$ for all $0 \le i \le n$ and $\{x^{i - 1}, x^i \}$ is an open edge for all $1 \le i \le n$. We define the cluster at $x$ in $S \subset \mathbb{Z}^d$ by $$\mathcal{C}^S (x) := \{y \in S :~\text{there exists an open path in $S$ connecting $x$ and $y$} \}.$$ The critical probability is defined by $$p_c(\mathbb{Z}^d ) := \inf \{p : ~\textbf{P}_p (\mathcal{C}^{\mathbb{Z}^d} (0) ~\text{is infinite} ) > 0 \}.$$ Let $\mathcal{C}_d(N)$ be the largest cluster in a box $[-N, N]^d$. We prove the following results. \[scpcprop\] (1) For $d = 2, p > p_c(\mathbb{Z}^2)$, there exist $c_1, c_2 > 0$ such that $$\displaystyle \lim_{N \to \infty} \textbf{P}_p (c_1 N^2 (\log N)^2 \le t_{cov}(\mathcal{C}_2(N)) \le c_2 N^2 (\log N)^3) = 1.$$ (2) For $d \ge 3, p > p_c(\mathbb{Z}^d)$, there exist $c_3, c_4 > 0$ such that $$\displaystyle \lim_{N \to \infty} \textbf{P}_p (c_3 N^d \log N \le t_{cov}(\mathcal{C}_d(N)) \le c_4 N^d (\log N)^{\frac{2d - 1}{d - 1} }) = 1.$$ \[remscpc\] Unfortunately, we are not able to obtain the correct order of the cover time. If $\text{diam}_R (\mathcal{C}_2 (N))$ is of order $\log N$ as stated in Corollary 3.1 of [@BK], we can obtain the correct order $(N^2 (\log N)^2)$ of the cover time for $\mathcal{C}_2 (N).$ However, from the proof of Corollary 3.1 of [@BK], we can only obtain that $\text{diam}_R (\mathcal{C}_2 (N))$ is of order $(\log N)^2.$ In particular, we can only state that $t_{\text{cov}} (\mathcal{C}_2 (N))$ is of order $N^2 (\log N)^3.$ We use the following lemmas. ([@BM], Proposition 1.2) \[scpclem0\] For $d \ge 2, p > p_c(\mathbb{Z}^d),$ there exists $c > 0$ such that w.h.p., $$\|\mathcal{C}_d(N)\| \ge c N^d.$$ Let $G = (V(G), E(G))$ be a finite graph. For $S \subset V(G),$ we define the external boundary of $S$ under the graph $G$ by $\partial_e S := \{x \in V(G) \backslash S : \text{there exists}~ y \in S ~\text{such that}~\{x, y \} \in E(G) \}.$ Set $L_x := \displaystyle \sum_{k = 1}^{\lfloor \log_2 \|V(G)\| \rfloor} \displaystyle \max \left(\frac{\|S\|}{\|\partial_e S\|^2} + \frac{1}{\|\partial_e S\|} \right),$ where the maximum is taken over all connected subsets $S$ of $V(G)$ satisfying $x \in S$ and $\|V(G)\| / 2^{k + 1} < \|S\| \le \|V(G)\| / 2^k.$ ([@BK], Theorem 2.1) \[scpclem1\] Let $G = (V(G), E(G))$ be a finite graph. There exists $c > 0$ such that for all $x, y \in V(G)$, $$R_{\text{eff}} (x, y) \le c (L_x + L_y).$$ ([@Pe], Corollary 1.4) \[scpclem2\] Fix $d \ge 2, p > p_c (\mathbb{Z}^d).$ There exist $c, c^{'} > 0$ such that $$\begin{aligned} \lim_{N \to \infty} \textbf{P}_p \Big(\|\partial_e S\| \ge c \|S\|^{1 - 1/d} ~ &\text{for all connected subsets} ~S \subset \mathcal{C}_d(N) \\ &\text{with} ~c^{'} (\log N)^{\frac{d}{d - 1}} \le \|S\| \le \frac{\|\mathcal{C}_d(N)\|}{2} \Big) = 1,\end{aligned}$$ where $\partial_e S$ is the external boundary of $S$ under the graph $\mathcal{C}_d(N).$ Proof of Proposition \[scpcprop\]. First, we prove the upper bounds by checking (\[condition-of-type1-0\]) in Theorem \[type1\] with $\log v(N) = \log N$ and $r(N) = (\log N)^{\frac{d}{d - 1}}$. It is clear that $\|\mathcal{C}_d(N)\| \le \|[-N, N]^d \cap \mathbb{Z}^d\| \le (2N + 1)^d,$ $\textbf{P}$-a.s. If $\|\partial_e S\| \ge c \|S\|^{1 - 1/d}$ for all connected subset $S \subset \mathcal{C}_d(N)$ with $c^{'} (\log N)^{\frac{d}{d - 1}} \le \|S\| \le \frac{\|\mathcal{C}_d(N)\|}{2},$ then we get for some $c_5 > 0,$ $$\begin{aligned} &\sum_{k = 1}^{\lfloor \log_2 \|\mathcal{C}_d(N)\| \rfloor} \max \Big \{ \frac{\|S\|}{\|\partial_e S\|^2} + \frac{1}{\|\partial_e S\|} : S~\text{is a connected subset of}~\mathcal{C}_d(N) \\ &~~~~~~~~~~~~~~~~~~~~~\text{satisfying}~x \in S ~\text{and}~ \|\mathcal{C}_d(N)\| / 2^{k + 1} < \|S\| \le \|\mathcal{C}_d(N)\| / 2^k \Big \} \\ &\le \sum_{k = 1}^{\lfloor \log_2 \{\|\mathcal{C}_d(N)\| / c^{'} (\log N)^{\frac{d}{d - 1}} \} \rfloor - 1} \Big (\frac{1}{c^2} + \frac{1}{c} \Big) \\ &+ \sum_{k = \lfloor \log_2 \{\|\mathcal{C}_d(N)\| / c^{'} (\log N)^{\frac{d}{d - 1}} \} \rfloor}^{\lfloor \log_2 \|\mathcal{C}_d(N)\| \rfloor} \Big (\frac{\|\mathcal{C}_d (N)\|}{2^k} + 1 \Big ) \\ &\le c_5 (\log N)^{\frac{d}{d - 1}} ~\text{for all}~x \in \mathcal{C}_d (N). \end{aligned}$$ Therefore, by Lemma \[scpclem1\] and Lemma \[scpclem2\], there exists $c_6 > 0$ such that w.h.p., $$\text{diam}_R(\mathcal{C}_d(N)) \le c_6 (\log N)^{\frac{d}{d - 1}}.$$ By Theorem \[type1\] (1), we obtain the upper bound. Next, we prove the lower bound for $d = 2$ by checking (\[condition-of-type1\]) in Theorem \[type1\] with $\log v(N) = \log N$ and $r(N) = \log N.$\ If $\|\mathcal{C}_2(N)\| \ge c_7 N^2$, there exist $c_8 > 0, x, y \in \mathcal{C}_2(N)$ such that $d_{\mathbb{Z}^2}(x,y) > c_8 N.$\ We define a square with side length $2k$ centered at $u$ and its internal boundary by $$Q(u, k) := \{v \in \mathbb{Z}^2 : v_i \in [u_i - k, u_i + k], i = 1, 2 \},$$ $$\partial_i Q(u, k ) := \{v \in Q(u, k) : \exists w \in \mathbb{Z}^2 \backslash Q(u, k) ~\text{such that}~ \{v, w \} \in \mathbb{E}^2 \}.$$ Since $y \notin Q(x, \lfloor \frac{c_8}{2} N \rfloor ),$ there exists $x^k \in \mathcal{C}_2(N)$ such that $x^k \in \partial_i Q(x, k\lfloor \sqrt{N} \rfloor )$ for all $0 \le k \le \frac{\lfloor \frac{c_8}{2} N \rfloor }{\lfloor \sqrt{N} \rfloor}.$ Fix $x^k, x^{\ell}, 0 \le k < \ell \le \frac{\lfloor \frac{c_8}{2} N \rfloor }{\lfloor \sqrt{N} \rfloor}.$ Since $d_{\mathbb{Z}^2 }(x^k, x^{\ell} ) \ge \lfloor \sqrt{N} \rfloor$, there exists a positive integer $a(N) \in [\lfloor \frac{\lfloor \sqrt{N} \rfloor}{2} \rfloor, \infty)$ such that $x^{\ell} \in \partial_i Q(x^k, a(N) ).$ We write $\Pi_j := \{\{u, v \} \in \mathbb{E}^2 : u \in \partial_i Q(x^k, j - 1) ~\text{and}~v \in \partial_i Q(x^k, j) \}, 1 \le j \le a(N).$ Under the induced graph $G_{cN}$ with vertex set $[- cN, cN]^2 \cap \mathbb{Z}^2$ for some sufficiently large constant $c > 0,$ $(\Pi_j)_{1 \le j \le a(N)}$ is a sequence of edge-cutsets between $x^k$ and $x^{\ell}$. So, we have by Lemma \[Nash-Williams\] that for some $ c_9 > 0$, $$\label{scpcineq} R_{\text{eff}}^N (x^k, x^{\ell}) \ge R_{\text{eff}}^{G_{cN}} (x^k, x^{\ell}) \ge c_9 \log N,$$ where $R_{\text{eff}}^{G_{cN}} (\cdot, \cdot)$ is the effective resistance in the graph $G_{cN}.$\ Set $V^{'} := \{ x^0, x^1, \dotsc, x^{\lfloor \frac{\lfloor \frac{c_8}{2} N \rfloor }{\lfloor \sqrt{N} \rfloor} \rfloor} \}.$ By (\[scpcineq\]), $\{B_{\text{eff}}^N (x, \frac{c_9}{4} \log N) : x \in V^{'} \}$ is a packing for $\mathcal{C}_2(N).$ So, there exists $c_{10} > 0$ such that w.h.p., $$\log \{n_{\text{pac}} (\mathcal{C}_2(N), \frac{c_9}{4} \log N) \} \ge c_{10} \log N.$$ Therefore, by Theorem \[type1\] (2) and Lemma \[scpclem0\], we get the lower bound for $d = 2.$ We next prove the lower bound for $d \ge 3$ by checking (\[condition-of-type1\]) in Theorem \[type1\] with $\log v(N) = \log N$ and $r(N) = 1.$ Fix $u, v \in \mathcal{C}_d (N), u \neq v.$ Set $\Pi := \{ \{u, x \} : \{u, x \} \in E(\mathcal{C}_d(N)) \}.$ $\Pi$ is an edge-cutset between $u$ and $v$ in the graph $\mathcal{C}_d (N).$ So, by Lemma \[Nash-Williams\], we have that $R_{\text{eff}}^N (u, v) \ge 1/\|\Pi\| \ge 1/2d.$ In particular, $\{B_{\text{eff}}^N (x, 1/8d) : x \in \mathcal{C}_d(N) \}$ is a packing for $\mathcal{C}_d(N).$ So, by Lemma \[scpclem0\], we have for some $c_{11} > 0,$ $$\log \{n_{\text{pac}} (\mathcal{C}_d (N), 1/8d) \} \ge c_{11} \log N ~\text{w.h.p.}$$ Therefore, by Theorem \[type1\] (2) and Lemma \[scpclem0\], we obtain the lower bound for $d \ge 3$. $\Box$\ Supercritical Erdős-Rényi random graph I {#sec:9} ---------------------------------------- Let $G(N, p)$ be the Erdős-Rényi random graph. This is obtained from the complete graph with $N$ vertices by retaining each edge with probability $p$ independently. We assume that $p = \frac{c}{N}$, where $c > 1$ is a positive constant. Let $\mathcal{C}^N$ be the largest connected component of $G(N, p)$. We revisit Theorem 2a of [@CF1]. Note that Cooper and Frieze [@CF1] has obtained a better estimate than the following Proposition \[scerprop\]. See Remark \[scerrem\] below. \[scerprop\] There exist $c_1, c_2 > 0$ such that $$\displaystyle \lim_{N \to \infty} \textbf{P} (c_1 N (\log N)^2 \le t_{cov}(\mathcal{C}^N) \le c_2 N (\log N)^2) = 1,$$ and $(t_{cov}(\mathcal{C}^N))_{N \in \mathbb{N}}$ is Type 1. Proof. We check (\[condition-of-type1-0\]) and (\[condition-of-type1\]) in Theorem \[type1\] with $v(N) = N$ and $r(N) = \log N.$ It is known that w.h.p., $(1 - \epsilon) \frac{cx(2-x)}{2} N \le \|E(\mathcal{C}^N)\| \le (1 + \epsilon) \frac{cx(2-x)}{2} N$ for any $\epsilon > 0,$ where $x$ is the solution of $x = 1 - e^{-cx} $ in $(0, 1)$ (see Section 3.1.3 of [@CF1]). By Theorem 6 of [@CL], there exists $c_3 > 0$ such that w.h.p., $$\text{diam}_R(\mathcal{C}^N) \le \text{diam}(\mathcal{C}^N) \le c_3 \log N.$$ The largest connected component $\mathcal{C}^N$ consists of a 2-core $C_2$ (the largest subgraph of $\mathcal{C}^N$ with minimum degree $2$) and a mantle $\textbf{M}$ (a collection of trees which are sprouting from different vertices of $C_2$). By Lemma 9 and P7a of [@CF1], w.h.p., there exists a subset $V^{'} \subset \mathcal{C}^N$ which satisfies the following:\ \ (i) Every $v \in V^{'}$ is a leaf of a tree $T_v$ in $\textbf{M},$\ (ii) Let $w(v)$ be the root of $T_v$. Then, $d_{\mathcal{C}^N} (v, w(v)) = \lceil \log N/(2(cx - \log c)) \rceil$,\ (iii) For any $\epsilon > 0, c_4 N^{1/2 - \epsilon} \le \|V^{'}\| \le c_5 N^{1/2 + \epsilon}, ~T_u \neq T_v ~\text{for all}~ u, v \in V^{'}, u \neq v,$ where $c_4, c_5 > 0$ are some constants.\ \ (Indeed, choose ‘special vertices’ in their terminology in Section 3.1.2 of [@CF1].)\ In particular, if $u, v \in V^{'}, u \neq v$, then every path from $u$ to $v$ contains a common path of length $\lceil \log N/(2(cx - \log c)) \rceil$. By Lemma \[Nash-Williams\], we have that $R_{\text{eff} }^N(u, v) > 2 \lceil \log N/(2(cx - \log c)) \rceil.$ In particular, $\{B_{\text{eff}}^N (v, \lceil \frac{\log N}{2(cx - \log c)} \rceil) : v \in V^{'} \}$ is a packing for $\mathcal{C}^N.$ So, we have for some $c_6 > 0,$ $$\log \Big \{n_{\text{pac}} \Big(\mathcal{C}^N, \Big \lceil \frac{\log N}{2(cx - \log c)} \Big \rceil \Big) \Big \} \ge \log \|V^{'}\| \ge c_6 \log N,~\text{w.h.p.}~\Box$$ \[scerrem\] In [@CF1], Cooper and Frieze proved that for any $\epsilon > 0,$ w.h.p., $$(1 - \epsilon) \frac{cx(2 - x)}{4(cx - \log c)} N(\log N)^2 \le t_{\text{cov}}(\mathcal{C}^N) \le (1 + \epsilon) \frac{cx(2 - x)}{4(cx - \log c)} N(\log N)^2.$$ Supercritical Erdős-Rényi random graph II {#sec:10} ----------------------------------------- We consider the Erdős-Rényi random graph $G(N, p)$ again. Here we assume that $p = \frac{f(N)}{N}$, where $\lim_{N \to \infty} \frac{\log N}{f(N)} = \lim_{N \to \infty} \frac{f(N)}{N^{1/2}} = 0.$ In this regime, $G(N, p)$ is connected w.h.p. We revisit Theorem 1.1(i) of [@J]. Note that Jonasson [@J] has obtained a better estimate than the following Proposition \[scer2prop\]. See Remark \[scer2rem\] below. \[scer2prop\] There exist $c_1, c_2 > 0$ such that $$\displaystyle \lim_{N \to \infty} \textbf{P} (c_1 N \log N \le t_{cov}(G(N, p)) \le c_2 N \log N ) = 1,$$ and $(t_{cov}(G(N, p)))_{N \in \mathbb{N}}$ is Type 1. In the proof, we use the following lemma. (Lemma 3.1 and Proposition 3.1 of [@J]) \[scer2lem\] Fix any $\epsilon > 0$. Then, w.h.p., $$(1 - \epsilon) f(N) \le \mu_x^N \le (1 + \epsilon) f(N), ~\text{for all}~x \in G(N, p),$$ $$\text{diam}_R (G(N, p)) \le \frac{2}{(1 - \epsilon) f(N)}.$$ Proof of Proposition \[scer2prop\]. We check (\[condition-of-type1-0\]) and (\[condition-of-type1\]) in Theorem \[type1\] with $v(N) = f(N) N$ and $r(N) = 1/f(N).$ By Lemma \[scer2lem\], there exist $c_3, c_4 > 0$ such that w.h.p., $$c_3 f(N) N \le \|E(G(N, p))\| \le c_4 f(N) N.$$ By this together with Lemma \[scer2lem\], (\[condition-of-type1-0\]) holds.\ By Lemma \[Nash-Williams\] and Lemma \[scer2lem\], there exists $c_5 > 0$ such that $\{B_{\text{eff}}^N (x, \frac{c_5}{f(N)}) : x \in G(N, p) \}$ is a packing for $G(N, p)$ w.h.p. So, we have for some $c_6 > 0,$ $$\log \{n_{\text{pac}} (G(N, p), c_5 / f(N) ) \} \ge c_6 \log \{ f(N)N \},~\text{w.h.p.}~\Box$$ \[scer2rem\] In [@J], Jonasson proved that if $\lim_{N \to \infty} \frac{\log N}{f(N)} = 0$, then for any $\epsilon > 0,$ w.h.p., $$(1 - \epsilon) N \log N \le t_{\text{cov}}(G(N, p)) \le (1 + \epsilon) N \log N.$$ The incipient infinite cluster for critical Galton-Watson family trees {#sec:11} ---------------------------------------------------------------------- Let $(Z_N)_{N \ge 0}$ be a critical Galton-Watson process with offspring distribution $Z$ in the domain of attraction of a stable law with index $\alpha \in (1, 2 ].$ That is, there exists a sequence $(a_N)_{N \ge 0}$ such that $\frac{Z[N] - N}{a_N} \stackrel{d}{\to} X$, where $\text{E}e^{- \lambda X} = e^{- \lambda^{\alpha}}$ and $Z[N]$ is the sum of $N$ i.i.d copies of $Z$. We write $\mathcal{T}$ to denote its family tree. We use the notation $\mathcal{T}_{\le N}, \mathcal{T}_N$ as in Subsection 3.1. We set $p_N := \text{P} (Z_N > 0)$. In [@Ke], Kesten considered the Galton-Watson tree conditioned to survive: ([@Ke], Lemma 1.14) For any family tree $T$ of $k$ generations, $$\lim_{N \to \infty} \text{P} (\mathcal{T}_{\le k} = T \| Z_N > 0) = \|T_k\| \text{P} (\mathcal{T}_{\le k} = T ).$$ We set $P_0 (T) = \|T_k\| P(\mathcal{T}_{\le k} = T)$. $P_0$ has a unique extension to a probability measure $\textbf{P}$ on the set of infinite family trees.\ By this lemma, we can take a family tree with the distribution $\textbf{P}$. We write this by $\mathcal{T}^{}$ and call it incipient infinite cluster. We set $Z_N^{} := \|\mathcal{T}_N^{}\|.$ \[critical GW\] There exist $c_1, c_2, c > 0$ such that for all $\lambda, N \ge c$, $$\textbf{P}(t_{\text{cov}}(\mathcal{T}_{\le N}^{}) \ge \lambda N^{\frac{2 \alpha - 1}{\alpha - 1}} \ell(N)^{- 1} ) \le c_1 \lambda^{- c_2},$$ $$\textbf{P}(t_{\text{cov}}(\mathcal{T}_{\le N}^{}) \le \lambda^{- 1} N^{\frac{2 \alpha - 1}{\alpha - 1}} \ell(N)^{- 1}) \le c_1 \lambda^{- c_2},$$ where $\ell (N)$ is a slowly varying function at infinity satisfying $p_N = N^{- \frac{1}{\alpha - 1}} \ell (N).$\ Furthermore, $(t_{\text{cov}}(\mathcal{T}_{\le N}^{}))_{N \in \mathbb{N}}$ is Type 2. Barlow, Ding, Nachmias and Peres [@BDNP] proved that in the case $\alpha = 2$, conditioned on the event $\{\|\mathcal{T}\| \in [N, 2N] \},$ $t_{\text{cov}} (\mathcal{T}) / N^{\frac{3}{2}}$ is tight. In the proof, we use the following facts. (Proposition 2.2, 2.5, 2.7 and Lemma 2.3 of [@CK]) \[cgwlem\]\ (1) There exists a slowly varying function at infinity $\ell(N)$ which satisfies that $p_N = N^{- \frac{1}{\alpha - 1}} \ell(N)$ and that for any $\epsilon > 0$, there exist $c_3, c_4 > 0$ such that $$c_3 \left(\frac{N}{N^{'}} \right)^{- \epsilon} \le \frac{\ell(N)}{\ell(N^{'})} \le c_4 \left(\frac{N}{N^{'}} \right)^{\epsilon}, ~\text{for all}~1 \le N^{'} \le N.$$ (2) Set $J(\lambda) := \{N \in \mathbb{N}: Z_N^{} \le \lambda p_N^{- 1}, \|E(\mathcal{T}_{\le N}^{} )\| \ge \lambda^{- 1} N p_N^{- 1}, \|\mathcal{T}_{\le N}^{} \| \le \lambda N p_N^{- 1} \}.$ Then there exist $c_5, c_6 > 0$ such that for all $N \in \mathbb{N}$ and $\lambda > 0$, $$\textbf{P} (N \in J(\lambda) ) \ge 1 - c_5 \lambda^{- c_6}.$$ Proof of Proposition \[critical GW\].\ By Lemma \[cgwlem\] (2) and the fact that $N \le \text{diam}_R (\mathcal{T}_{\le N}^{}) \le 2N$ $\textbf{P}$-a.s., the conditions (\[condition-of-type2-(1)-1\]) and (\[condition-of-type2-(3)\]) in Theorem \[type2\] hold for $v(N) = Np_N^{-1}$ and $r(N) = N.$ So, we only need to check (\[condition-of-type2-(1)-2\]) with $r(N) = N.$\ The idea of the following argument came from the proof of Theorem 3.1 of [@BDNP]. We write $\mathcal{T}^{, x}$ to denote the subtree rooted at $x \in \mathcal{T}^{}$. Set $r_{k,j}^N := \lfloor \frac{j}{2^{k+2}} N \rfloor, k \in \mathbb{N}, 0 \le j \le 2^{k+2}.$\ Fix $k \in \mathbb{N}$ and $0 \le j \le 2^{k+2} - 1.$ We say that $x \in \mathcal{T}_{r_{k,j}^N }^{} $ is $k$-good if $\mathcal{T}_{(r_{k, j + 1}^N - r_{k,j}^N ) }^{, x} \neq \emptyset.$ We assume $\lambda \ge c_7$, where $c_7$ is a sufficiently large positive constant. Set for all $0 \le j \le 2^{k + 2} - 1,$ $$A_{k, j}^N := \{x \in \mathcal{T}_{r_{k,j}^N}^{} : x~\text{is $k$-good} \}.$$ We define $$A_{k}^N := \begin{cases} \displaystyle \bigcup_{j = 0}^{2^{k+2} - 1} A_{k, j}^N & \text{if} ~0 \le k \le \lfloor \frac{\log N}{\log 2} \rfloor - 2, \\ \\ \mathcal{T}_{\le N}^{} & \text{otherwise}. \end{cases}$$ We define $\ell_k^N := \frac{\text{diam}_R (\mathcal{T}_{\le N}^{})}{2^k}$ for $0 \le k \le \lfloor \frac{\log N}{\log 2} \rfloor - 2$ and $\ell_k^N = 0$ otherwise.\ Since $\{B_{\text{eff}}^N (x, \ell_k^N) : x \in A_k^N \}$ is a covering for $\mathcal{T}_{\le N}^{}$ for all $k \ge 0,$ we get for all $k \ge 0,$ $$n_{\text{cov}} (\mathcal{T}_{\le N}^{}, \ell_k^N) \le \|A_k^N\|. \label{covering-number-ct}$$ Fix $0 \le k \le \lfloor \frac{\log N}{\log 2} \rfloor - 2$ and $1 \le j \le 2^{k+2} - 1.$ By Lemma 2.2 of [@Ke] (note that in [@Ke], Kesten assumed the variance of offspring distribution is finite, but the same result holds under our situation), for $\tilde{\lambda} > 0$, $$\begin{aligned} &~~~~\textbf{P}(\|A_{k, j}^N\| \ge \tilde{\lambda} \| \mathcal{T}_{\le r_{k,j}^N}^{} = T, H_{\le r_{k,j}^N} = (v_i)_{0 \le i \le r_{k,j}^N} ) \\ &= \textbf{P}(\|A_{k, j}^N \backslash \{ v_{r_{k,j}^N} \}\| \ge \tilde{\lambda} - 1 \| \mathcal{T}_{\le r_{k,j}^N}^{} = T, H_{\le r_{k,j}^N} = (v_i)_{0 \le i \le r_{k,j}^N} ) \\ &= \textbf{P}(\text{Bin}(\|T_{r_{k,j}^N}\| - 1, p_{(r_{k,j + 1}^N - r_{k,j}^N)} ) \ge \tilde{\lambda} - 1 ), \end{aligned}$$ where $T$ is a family tree of $r_{k,j}^N$ generations, $H_{\le r_{k,j}^N}$ is a backbone (the unique infinite line of descent of $\mathcal{T}^{}$) up to $r_{k,j}^N$ th level and $(v_i)_{0 \le i \le r_{k,j}^N}$ is a sequence of vertices such that $v_i \in T_i$ for all $0 \le i \le r_{k,j}^N.$ We also note that for all $0 \le m \le \Big \lfloor \frac{\tilde{\lambda}}{2p_{(r_{k,j + 1}^N - r_{k,j}^N )}} \Big \rfloor,$ $$\begin{aligned} &\textbf{P}(\text{Bin}(m, p_{(r_{k,j + 1}^N - r_{k,j}^N)} ) \ge \tilde{\lambda} - 1) \\ &\le \textbf{P}(\text{Bin}(\Big \lfloor \frac{\tilde{\lambda}}{2p_{(r_{k,j + 1}^N - r_{k,j}^N )} } \Big \rfloor, p_{(r_{k,j + 1}^N - r_{k,j}^N)} ) \ge \tilde{\lambda} - 1 ).\end{aligned}$$ Therefore, for $\tilde{\lambda} > 2$, $$\begin{aligned} &~~~~\textbf{P}(\|A_{k, j}^N\| \ge \tilde{\lambda} ) \\ &\le \textbf{P}(\text{Bin}(\Big \lfloor \frac{\tilde{\lambda}}{2p_{(r_{k,j + 1}^N - r_{k,j}^N)}} \Big \rfloor, p_{(r_{k,j + 1}^N - r_{k,j}^N)} ) \ge \tilde{\lambda} - 1 ) \\ &+ \textbf{P}(Z_{r_{k,j}^N}^{} > \Big \lfloor \frac{\tilde{\lambda}}{2p_{(r_{k,j + 1}^N - r_{k,j}^N)}} \Big \rfloor ).\end{aligned}$$ By the Chebyshev inequality, the first term is bounded by $\frac{2 \tilde{\lambda}}{(\tilde{\lambda} - 2 )^2}$. By Lemma \[cgwlem\] (1) (2), the second term is bounded by $c_8 j^{c_9} \tilde{\lambda}^{- c_{10}}$ for some $c_8, c_9, c_{10} > 0.$ So, we have that $$\begin{aligned} &~~~~\textbf{P}(\|A_k^N\| \ge \exp (\lambda 2^{k/2}) ) \\ &\le \textbf{P} \bigg(\displaystyle \bigcup_{j = 1}^{2^{k+2} - 1} \bigg \{ \|A_{k, j}^N\| \ge \frac{\exp (\lambda 2^{k/2}) - 1}{2^{k+2}} \bigg \} \bigg) \\ &\le \displaystyle \sum_{j = 1}^{2^{k+2} - 1} \Bigg \{\frac{2 \cdot \frac{\exp (\lambda 2^{k/2}) - 1}{2^{k+2}}}{(\frac{\exp (\lambda 2^{k/2}) - 1}{2^{k+2}} - 2)^2 } + c_8 j^{c_9} \bigg(\frac{\exp (\lambda 2^{k/2}) - 1}{2^{k+2}} \bigg)^{- c_{10} } \Bigg \} \\ &\le c_{11} 2^{-k} \lambda^{- c_{12}} ~~~~\text{for some}~ c_{11}, c_{12} >0.\end{aligned}$$ From this fact, we have that $$\textbf{P} \Bigg(\displaystyle \bigcup_{k = 0}^{\lfloor \frac{\log N}{\log 2} \rfloor - 2} \bigg \{ \|A_k^N\| \ge \exp (\lambda 2^{k/2}) \bigg \} \Bigg) \le 2c_{11} \lambda^{- c_{12}}. \label{covering-number-ct-2}$$ If $\|A_k^N\| \le \exp (\lambda 2^{k/2})$ for all $0 \le k \le \lfloor \frac{\log N}{\log 2} \rfloor - 2$ and $\|\mathcal{T}_{\le N}^{}\| \le \lambda N p_N^{-1},$ we have by (\[covering-number-ct\]), $$\sum_{k = 1}^{\lfloor \frac{\log N}{\log 2} \rfloor - 1} \sqrt{\ell_{k-1}^N \log \{n_{\text{cov}} (\mathcal{T}_{\le N}^{}, \ell_k^N) \}} \le c_{13} \sqrt{\lambda N}$$ for some $c_{13} > 0$.\ So, by (\[covering-number-ct-2\]) and Lemma \[cgwlem\] (2), (\[condition-of-type2-(1)-2\]) in Theorem \[type2\] holds with $r(N) = N.$ $\Box$\ We can also say that $t_{\text{cov}}(\mathcal{T}_{\le N}^{}) N^{- \frac{2 \alpha - 1}{\alpha - 1}} \ell (N)$ is not concentrated. \[cgwprop2\] For all $\lambda \ge 1,$ $$\liminf_{N \to \infty} \textbf{P}(t_{\text{cov}}(\mathcal{T}_{\le N}^{}) N^{- \frac{2 \alpha - 1}{\alpha - 1}} \ell (N) \ge \lambda ) > 0.$$ To prove this fact, we use the following result. ([@Pa], Theorem 4) \[pakes\] The random variable $Z_N^{} p_N$ converges in law to a random variable $Z^{}$ with $\mathbb{E} (e^{- \theta Z^{}}) = (1 + \theta^{\alpha - 1})^{- \frac{\alpha}{\alpha - 1}}$ for $\theta \ge 0.$ Proof of Proposition \[cgwprop2\].\ By the fact that $t_{\text{cov}} (\mathcal{T}_{\le N}^{}) \ge t_{\text{hit}} (\mathcal{T}_{\le N}^{}) \ge \frac{1}{2} N \|E(\mathcal{T}_{\le N}^{})\|$ (we have used Lemma \[commute time\]), for $\lambda >0,$ $$\textbf{P}(t_{\text{cov}}(\mathcal{T}_{\le N}^{}) N^{- \frac{2 \alpha - 1}{\alpha - 1}} \ell (N) \ge \lambda ) \ge \textbf{P}(\|E(\mathcal{T}_{\le N}^{})\| \ge 2 \lambda N p_{N}^{- 1} ).$$ Using the proof of Proposition 2.5 of [@CK] (in page 1429) when $\alpha \in (1, 2)$ and Lemma \[pakes\] when $\alpha = 2$, we have that for $\lambda \ge 1$ and some $c_{14}, c_{15} > 0,$ $$\liminf_{N \to \infty} \textbf{P} (\|E(\mathcal{T}_{\le N}^{})\| \ge \lambda N p_N^{- 1}) \ge c_{14} \liminf_{N \to \infty} \textbf{P} (Z_{N^{'}}^{} p_{N^{'}} > c_{15} \lambda) > 0,$$ where $N^{'} = \lfloor \frac{N}{3} \rfloor.$ This implies the conclusion. $\Box$ Critical percolation clusters {#sec:12} ----------------------------- Let $\Hat{G}^N$ be a graph with $N$ vertices and the maximum degree $d \in [3, N - 1]$. $\Hat{G}_{p}^N$ is obtained by retaining each edge of $\Hat{G}^N$ with probability $p \in (0, 1)$ independently. Let $\mathcal{C}^N$ be the largest connected component of $\Hat{G}_p^N$ and $\mathcal{C}(x)$ be the connected component of $\Hat{G}_p^N$ which contains $x \in V(\Hat{G}^N)$. We define balls and their boundaries as follows: $$B_p(x, r; \Hat{G}^N) := \{ y \in V(\Hat{G}_p^N) : d_{\Hat{G}_p^N} (x, y) \le r \},$$ $$\partial B_p (x, r; \Hat{G}^N) := \{y \in V(\Hat{G}_p^N) : d_{\Hat{G}_p^N} (x, y) = r \}.$$ We also set $$\Gamma_p(x, r; \Hat{G}^N ) := \sup_{G \subset \Hat{G}^N} \textbf{P}_{G} (H_p(x, r; G) ),$$ where $H_p (x, r; G ) := \{\partial B_p (x, r; G) \neq \emptyset \},$ the supremum is taken over all subgraphs of $\Hat{G}^N$ and $\textbf{P}_G$ is a percolation probability measure on $G$. In particular, we write $\textbf{P} := \textbf{P}_{\Hat{G}^N}$.\ We assume that $$p \le \frac{1 + A N^{- 1/3} }{d - 1} ~\text{for some}~ A \in \mathbb{R}, \label{ineqcr1}$$ and that there exist $c_1, c_2 > 0$ and $a : (0, \infty) \to (0, \infty)$ such that for sufficiently large $\lambda > 0$ and $N \ge a(\lambda),$ $$\textbf{P}(\|\mathcal{C}^N\| \le {\lambda}^{- 1} N^{\frac{2}{3} } ) \le c_1 {\lambda}^{- c_2}. \label{ineqcr2}$$ In the case that $\Hat{G}^N$ is the complete graph with $N$ vertices and $p = 1/N$, it is known that (\[ineqcr2\]) holds (see Theorem 2 of [@NP2]). We revisit Theorem 3.1 of [@BDNP]. \[crprop\] Under the assumption (\[ineqcr1\]) and (\[ineqcr2\]), there exist $c_3, c_4 > 0$ such that for sufficiently large $\lambda > 0$ and $N \ge \max \{\lambda^3, a(\lambda) \},$ $$\textbf{P}(t_\text{cov} (\mathcal{C}^N ) > \lambda N ) \le c_3 \lambda^{- c_4 }, ~\textbf{P}(t_\text{cov} (\mathcal{C}^N ) < {\lambda}^{- 1} N ) \le c_3 \lambda^{- c_4 },$$ and $(t_\text{cov} (\mathcal{C}^N) )_{N \in \mathbb{N}}$ is Type 2. \[crrem-0\] Barlow, Ding, Nachmias and Peres [@BDNP] have already considered the cover time for the critical random graphs. To prove this proposition, we use the following facts (most of them are proved in [@NP]). \[crlem\] (1) There exists $c_5 > 0$ such that for all subgraphs $G \subset \Hat{G}^N, x \in V(G), \lambda >0$ and sufficiently large $N \in \mathbb{N},$ $$\textbf{E}\|B_p(x, r; G )\| \le c_5 e^{A \lambda} r,~~~~~~~ \text{for all}~ r \le \lambda N^{\frac{1}{3} }, \label{ineqcr3}$$ $$\Gamma_p (x, r; \Hat{G}^N) \le c_5/r, ~~~~~~~~~\text{for all} ~r \le N^{\frac{1}{3} }, \label{ineqprobab}$$ where $A$ is the constant in (\[ineqcr1\]).\ (2) There exists $c_6 > 0$ such that for sufficiently large $N \in \mathbb{N}, \lambda > 0,$ $$\textbf{P} (\|E (\mathcal{C}^N)\| \ge \lambda N^{2/3} ) \le c_6 \lambda^{- 1}, \textbf{P} (\text{diam} (\mathcal{C}^N) \ge \lambda N^{1/3} ) \le c_6 \lambda^{- 1},$$ where $\text{diam} (\mathcal{C}^N) = \max_{x, y \in V(\mathcal{C}^N)} d_{\mathcal{C}^N} (x, y).$\ (3) There exists $c_7 > 0$ such that for sufficiently large $\lambda > 0$ and $N \ge \lambda^2,$ $$\textbf{P} (\exists x \in V(\Hat{G}^N), \|\mathcal{C} (x) \| > \lambda^{- 1/12} N^{2/3}~\text{and}~\text{diam}_R (\mathcal{C} (x)) < \lambda^{-1} N^{1/3} ) \le c_7 \lambda^{-1/6}.$$ (4) There exists $c_8 > 0$ such that for sufficiently large $\lambda > 0, N \ge \max \{\lambda^3, a(\lambda) \},$ $$\textbf{P} (\|E (\mathcal{C}^N)\| < \lambda^{-1} N^{2/3} ) \le c_8^{-1} \lambda^{- c_8}, \textbf{P} (\text{diam}_R (\mathcal{C}^N) < \lambda^{-1} N^{1/3} ) \le c_8^{-1} \lambda^{- c_8}.$$ To prove (3) of this lemma, we use Proposition 5.6 in [@NP]. So, we recall some terms in [@NP].\ Fix $x \in \Hat{G}^N, r, L \in \mathbb{N}, k < r.$ For $j < r,$ a lane for $(x, r, j)$ is an edge $\{u, v \}$ with $u \in \partial B_p (x, j - 1; \Hat{G}_p^N)$ and $v \in \partial B_p (x, j ; \Hat{G}_p^N)$ such that a path from $u$ to a vertex in $\partial B_p (x, r ; \Hat{G}_p^N)$ passes $\{u,v \}$ and does not intersect $\partial B_p (x, j - 1; \Hat{G}_p^N)$ except the starting vertex.\ We say $x$ is $L$-lane rich for $(k, r)$ if we have a subset $I \subset [\lfloor k/2 \rfloor, k] \cap \mathbb{Z}$ with $\|I\| > \lfloor \frac{1}{2} (k - \lfloor \frac{k}{2} \rfloor) \rfloor$ such that for any $j \in I$, there exist at least $L$ lanes for $(x, r, j).$\ ([@NP], Proposition 5.6) \[l-lane-rich\] Suppose that $x \in V(\Hat{G}^N), L \in \mathbb{N}, k \le r/2$ and $r < N^{1/3}.$ Then there exists $c_9 > 0$ such that $$\textbf{P} (x~\text{is $L$-lane rich for}~(k, r) ) \le c_9 L^{- 1} r^{- 1}.$$ Proof of Lemma \[crlem\]* By the proof of Theorem 1.2 and Theorem 1.3 of [@NP] in page 1281, (1) holds. The results of (2) are proved in [@NP] in page 1274 and 1283.\ The result of (4) follows from (3) and (\[ineqcr2\]). So, we only prove (3).\ We use Lemma \[l-lane-rich\] with $$k = \Big \lfloor \frac{1}{2} \Big \lfloor N^{1/3} \Big (\frac{\lambda}{32} \Big)^{-1/3} \Big \rfloor \Big \rfloor, r = \Big \lfloor N^{1/3} \Big (\frac{\lambda}{32} \Big)^{-1/3} \Big \rfloor, L = \Big \lfloor \Big (\frac{\lambda}{32} \Big)^{2/3} \Big \rfloor.$$ Suppose that $x$ is not $L$-lane rich for $(k, r)$ and $\text{diam} (\mathcal{C} (x)) \ge (\frac{\lambda}{32})^{-1/3} N^{1/3}.$\ Since $x$ is not $L$-lane rich for $(k, r),$ there exists a subset $I \subset [\lfloor k/2 \rfloor, k] \cap \mathbb{Z}$ with $\|I\| \ge \lfloor \frac{1}{2} (k - \lfloor \frac{k}{2} \rfloor) \rfloor$ such that for all $j \in I,$ the number of lanes for $(x, r, j)$ is less than $L.$\ For $j \in I,$ let $\Pi_j$ be a set of all lanes for $(x, r, j).$ Note that by the property of $I,$ we have $\|\Pi_j\| \le L.$\ Because $\text{diam} (\mathcal{C} (x)) \ge (\frac{\lambda}{32})^{-1/3} N^{1/3},$ there exists a vertex $x_0$ in $\partial B_p (x, r; \Hat{G}_p^N).$ Since $\Pi_j$ is an edge-cutset between $x$ and $x_0$ for all $j \in I$, we get by Lemma \[Nash-Williams\] for sufficiently large $\lambda > 0, N \ge \lambda^2,$ $$\text{diam}_R (\mathcal{C} (x)) \ge R_{\text{eff}}^N (x, x_0) \ge \sum_{j \in I} 1/\|\Pi_j\| \ge \|I\|/L \ge \lambda^{-1} N^{1/3}.$$ Therefore, we have $$\begin{aligned} &\textbf{P} (\|\mathcal{C} (x)\| > \lambda^{-1/12} N^{2/3}, \text{diam}_R ( \mathcal{C} (x)) < \lambda^{-1} N^{1/3} ) \\ &\le \textbf{P} (\|\mathcal{C} (x)\| > \lambda^{-1/12} N^{2/3}, \text{diam} ( \mathcal{C} (x)) < \Big (\frac{\lambda}{32} \Big)^{-1} N^{1/3} ) \\ &+ \textbf{P} (x~\text{is $L$-lane rich for}~(k, r)). \end{aligned}$$ By (\[ineqcr3\]), we get for some $c_{10} > 0,$ $$\begin{aligned} &\textbf{P} (\|\mathcal{C} (x)\| > \lambda^{-1/12} N^{2/3}, \text{diam} ( \mathcal{C} (x)) < \Big (\frac{\lambda}{32} \Big)^{-1} N^{1/3} ) \notag \\ &\le \textbf{P} (\|B_p (x, r; \Hat{G}^N)\| > \lambda^{-1/12} N^{2/3}) \notag \\ &\le \frac{\textbf{E} \|B_p (x, r; \Hat{G}^N)\|}{\lambda^{-1/12} N^{2/3}} \notag \\ &\le \frac{c_5 e^A r}{\lambda^{-1/12} N^{2/3}} \notag \\ &\le c_{10} \lambda^{- 1/4} N^{- 1/3}. \label{term1-2-cr}\end{aligned}$$ By Lemma \[l-lane-rich\] and (\[term1-2-cr\]), we have for some $c_{11} > 0$ and sufficiently large $\lambda > 0, N \ge \lambda^2,$ $$\textbf{P} (\|\mathcal{C} (x) \| > \lambda^{- 1/12} N^{2/3}~\text{and}~\text{diam}_R (\mathcal{C} (x)) < \lambda^{-1} N^{1/3}) \le c_{11} \lambda^{- 1/4} N^{- 1/3}. \label{crg}$$ Set $X := \|\{x \in V(\Hat{G}^N) : \|\mathcal{C} (x) \| > \lambda^{- 1/12} N^{2/3}~\text{and}~\text{diam}_R (\mathcal{C} (x)) < \lambda^{-1} N^{1/3} \}\|.$ Note that if $X > 0,$ then $X > \lambda^{- 1/12} N^{2/3}.$ So, by the Chebyshev inequality and (\[crg\]), we have $$\begin{aligned} &\textbf{P} (\exists x \in V(\Hat{G}^N), \|\mathcal{C} (x) \| > \lambda^{- 1/12} N^{2/3}~\text{and}~\text{diam}_R (\mathcal{C} (x)) < \lambda^{-1} N^{1/3}) \\ &\le \textbf{P} (X \ge \lambda^{- 1/12} N^{2/3} ) \\ &\le c_{11} \lambda^{-1/6}. ~\Box\end{aligned}$$ Proof of Proposition \[crprop\]. By Lemma \[crlem\] (2) (4), (\[condition-of-type2-(1)-1\]) and (\[condition-of-type2-(3)\]) in Theorem \[type2\] hold for $v(N) = N^{2/3}$ and $r(N) = N^{1/3}.$ So, we only need to check (\[condition-of-type2-(1)-2\]) with $r(N) = N^{1/3}.$ The condition (\[condition-of-type2-(1)-2\]) follows from Lemma \[crlem\] and a minor modification of the proof of Theorem 3.1 of [@BDNP]. To make the paper self-contained, we briefly recall the argument of [@BDNP]. Fix $x \in V(\Hat{G}^N), 0 \le k \le k_0^N : = 2 \lfloor \log_2 \log N \rfloor,$ sufficiently large $\lambda > 0,$ and $N \ge \max \{ \lambda^4, a(\lambda) \}.$ By (\[ineqcr3\]), we have a sequence $(r_{k, j}^N)_{j = 0}^{\lfloor 4 \lambda^2 2^k \rfloor}$ satisfying $r_{k, 0}^N = 0, \frac{(j - 1/2) N^{1/3}}{4 \lambda 2^k} \le r_{k, j}^N \le \frac{j N^{1/3}}{4 \lambda 2^k}$ and $\textbf{E} \|\partial B_p (x, r_{k, j}^N ; \Hat{G}^N ) \| \le 16 \lambda^2 c_5 e^{A \lambda} 2^k$ for all $1 \le j \le \lfloor 4 \lambda^2 2^k \rfloor.$ We say that $y \in \partial B_p (x, r_{k, j}^N ; \Hat{G}^N)$ is $k$-good if $y$ and a vertex in $\partial B_p (x, r_{k, j+1}^N ; \Hat{G}^N)$ are connected by a path which does not intersect $\partial B_p (x, r_{k, j}^N ; \Hat{G}^N)$ except $y.$ Set $$A_{k}^N (x) := \begin{cases} \displaystyle \bigcup_{j = 0}^{\lfloor 4 \lambda^2 2^k \rfloor} \{y \in \partial B_p (x, r_{k, j}^N ; \Hat{G}^N) : y~\text{is $k$-good} \} & \text{if} ~0 \le k \le k_0^N - 1, \\ \\ \mathcal{C} (x) & \text{if}~ k = k_0^N. \end{cases}$$ We define $\ell_k^N(x) : = \frac{\text{diam}_R (\mathcal{C}(x))}{2^k}$ for $0 \le k \le k_0^N - 1$ and $\ell_{k_0^N}^N (x) := 0.$ Under the events that $\text{diam}(\mathcal{C}(x)) \le \lambda N^{1/3}$ and $\text{diam}_R (\mathcal{C} (x)) \ge \lambda^{-1} N^{1/3},$ the set of resistance balls $\{ B_{\text{eff}}^N (y, \ell_k^N (x)) : y \in A_k^N (x) \}$ is a covering for $\mathcal{C}(x)$ and $n_{\text{cov}} (\mathcal{C} (x), \ell_k^N (x)) \le \|A_k^N (x)\|$ for all $0 \le k \le k_0^N.$ By (\[ineqprobab\]), we get for some $c_{12} > 0,$ $$\begin{aligned} &\textbf{P} \Big(\exists x \in V(\Hat{G}^N), 0\le \exists k \le k_0^N, \|\mathcal{C} (x)\| > \lambda^{-1/12} N^{2/3}, \|A_k^N (x)\| \ge e^{(\|A\| + 1) \lambda 2^{\frac{k}{2}}} \Big) \notag \\ &\le c_{12}^{-1} \lambda^{-c_{12}}. \label{ineq-covering}\end{aligned}$$ Set $\ell_k^N := \frac{\text{diam}_R (\mathcal{C}^N)}{2^k}$ for $0 \le k \le k_0^N - 1$ and $\ell_{k_0^N}^N := 0.$ By (\[ineq-covering\]) together with Lemma \[crlem\] (2), (3) and (\[ineqcr2\]), we have for some $c_{13}, c_{14} > 0,$ $$\begin{aligned} &\textbf{P} \Big(\sum_{k = 1}^{k_0^N} \sqrt{\ell_{k-1}^N \log \{ n_{\text{cov}} (\mathcal{C}^N, \ell_k^N ) \} } \ge c_{13} \lambda \sqrt{N^{1/3}}~ \Big) \\ &\le \textbf{P} \Big(\|\mathcal{C}^N\| > \lambda^{-1/12} N^{2/3}, \text{diam} (\mathcal{C}^N) < \lambda N^{1/3} \\ &~~~~~~~~~~~~~~~~~~~~~~~~~\text{and}~\sum_{k = 1}^{k_0^N} \sqrt{\ell_{k-1}^N \log \{ n_{\text{cov}} (\mathcal{C}^N, \ell_k^N ) \} } \ge c_{13} \lambda \sqrt{N^{1/3}}~ \Big) \\ &+ \textbf{P} (\|\mathcal{C}^N\| \le \lambda^{-1/12} N^{2/3}) + \textbf{P} (\text{diam} (\mathcal{C}^N) \ge \lambda N^{1/3}) \\ &\le \textbf{P} (\exists x \in V(\Hat{G}^N), \|\mathcal{C} (x) \| > \lambda^{- 1/12} N^{2/3}~\text{and}~\text{diam}_R (\mathcal{C} (x)) < \lambda^{-1} N^{1/3} ) \\ &+ \textbf{P} \Big(\exists x \in V(\Hat{G}^N), 0 \le \exists k \le k_0^N, \|\mathcal{C} (x)\| > \lambda^{-1/12} N^{2/3}, \|A_k^N (x)\| \ge e^{(\|A\| + 1) \lambda 2^{\frac{k}{2}}} \Big) \\ &+ \textbf{P} (\|\mathcal{C}^N\| \le \lambda^{-1/12} N^{2/3}) + \textbf{P} (\text{diam} (\mathcal{C}^N) \ge \lambda N^{1/3}) \\ &\le c_{14}^{-1} \lambda^{-c_{14}}. \Box\end{aligned}$$ The range of random walk in $\mathbb{Z}^{d}, d \ge 5$ {#sec:13} ----------------------------------------------------- Let $d \ge 5$. We write $(S_n)_{n \ge 0}$ to denote the simple random walk in $\mathbb{Z}^{d}$ started from 0 which is defined on a probability space with probability measure $\textbf{P}$. Let $G^N$ be a graph with vertex set $V(G^N) := \{S_n : 0 \le n \le N \}$ and edge set $E(G^N) := \{ \{S_{n - 1}, S_n \} : 1 \le n \le N \}$. We prove the following proposition. \[trprop\] There exist $c_1, c_2 > 0$ such that $\textbf{P}$-a.s., for sufficiently large $N \in \mathbb{N}$, $$c_1 N^2 \le t_{\text{cov}}(G^N) \le c_2 N^2,$$ and $(t_{\text{cov}}(G^N))_{N \in \mathbb{N}}$ is Type 2. Let $(S_{- n} )_{n \ge 0}$ be an independent copy of $(S_n)_{n \ge 0}$ and set $S = (S_n)_{n \in \mathbb{Z} }$. Let $\mathcal{T}$ be the set of cut-times, that is, $\mathcal{T} := \{n : S_{(- \infty, n] } \cap S_{[n + 1, \infty) } = \emptyset \}.$ We can write $\mathcal{T} \cap (0, \infty) = \{T_n : n \in \mathbb{N} \}$, where $0 < T_1 < T_2 < \dotsc $. Set cut-points $C_n := S_{T_n}$. We use the following fact. \[RRW\] ([@Cr], Lemma 2.2 (see also [@CHK], (5.6)) ) $$\lim_{n \to \infty} \frac{T_n}{n} = \tau(d) := \mathbf{E}(T_1 \| 0 \in \mathcal{T} ) \in [1, \infty), ~~\textbf{P}-a.s.$$ Proof of Proposition \[trprop\]. We check almost-sure versions of (\[condition-of-type2-(1)-1\]), (\[condition-of-type2-(1)-2\]) and (\[condition-of-type2-(3)\]) in Theorem \[type2\] with $v(N) = r(N) = N.$ For $N \in \mathbb{N}$, there exists $M = M(N) \in \mathbb{N}$ such that $T_M \le N < T_{M + 1}$. Because $d_{G^N}(0, C_M) \ge M,$ we have that $\|E(G^N)\| \ge M, ~~\textbf{P}$-a.s. By Lemma \[RRW\], there exist $c_3, c_4 > 0$ such that $c_3 N \le M \le c_4 N,$ for sufficiently large $N \in \mathbb{N}$, $\textbf{P}$-a.s. So, $\textbf{P}$-a.s., for sufficiently large $N \in \mathbb{N}$, $$\|E(G^N)\| \ge c_3 N.$$ Every path from $0$ to $C_M$ must pass edges $\{S_{T_n}, S_{T_n + 1} \}_{1 \le n \le M - 1}$. So, by Lemma \[Nash-Williams\], there exists $c_5 > 0$ such that $\textbf{P}$-a.s., for sufficiently large $N \in \mathbb{N}$, $$\text{diam}_R(G^N) \ge R_{\text{eff}}^N(0, C_M) \ge M - 1 \ge c_5 N. \label{ineqrr}$$ By definition, $$\|E(G^N)\| \le N, ~\text{and}~ \text{diam}_R(G^N) \le \text{diam}(G^N) \le N, ~ \textbf{P}-a.s.$$ Fix $1 \le k \le \lfloor \log_2 \log (c_5 N) \rfloor.$ We define $A_k^N$ as follows: $$A_k^N := \begin{cases} \{S_{\lfloor j \frac{c_5 N}{2^{k + 1}} \rfloor} : 0 \le j \le \lfloor \frac{2^{k + 1}}{c_5} \rfloor \}, & \text{if}~1 \le k \le \lfloor \log_2 \log (c_5 N) \rfloor - 1, \\ \{ S_j : 0 \le j \le N \} & \text{otherwise}. \end{cases}$$ It is not hard to check that $V(G^N) \subset \bigcup_{u \in A_k^N } B^N (u, \frac{c_5 N}{2^k} ),$ where $B^N (u, r ) = \{v \in V(G^N) : d_{G^N} (u, v) \le r \}.$ Set $k_0^N = \lfloor \log_2 \log (c_5 N) \rfloor.$ By (\[ineqrr\]), we have that $\textbf{P}$-a.s., for sufficiently large $N \in \mathbb{N}$, $$V(G^N) \subset \bigcup_{u \in A_k^N } B_{\text{eff} }^N (u, \ell_k^N ),$$ where $\ell_k^N = \frac{\text{diam}_R (G^N)}{2^k}$ for $1 \le k \le k_0^N - 1$ and $\ell_k^N = 0$ otherwise. Because $n_{\text{cov}} (G^N, \ell_k^N) \le \|A_k^N\| \le \lfloor \frac{2^{k + 1}}{c_5} \rfloor + 1 \le c_6 2^k$ for some $c_6 > 0$ and all $k < k_0^N,$ we have $\textbf{P}$-a.s., for sufficiently large $N \in \mathbb{N},$ $$\sum_{k=1}^{k_0^N} \sqrt{\ell_{k-1}^N \log \{n_{\text{cov}} (G^N, \ell_k^N) \}} \le c_7 \sqrt{N} ~\text{for some}~c_7 > 0.$$ By Remark \[type2rems\] (3), we complete the proof. $\Box$ Sierpinski gasket graphs {#sec:14} ------------------------ Let $p_1, p_2, p_3$ be vertices of an equilateral triangle in $\mathbb{R}^2$. We define three contraction maps $\psi_i : {\mathbb{R} }^2 \to {\mathbb{R} }^2, i = 1, 2, 3$ as follows: $$\psi_i (x) = p_i + \frac{x - p_i }{2}, ~~~i = 1, 2, 3, x \in {\mathbb{R} }^2.$$ $G^N$ is a graph with the following vertex and edge sets: $V(G^N) := \displaystyle \bigcup_{i_1 \dotsc i_N =1 }^{3} \psi_{i_1 \dotsc i_N } (V_0), $\ $E(G^N) := \{\{\psi_{i_1 \dotsc i_N } (x), \psi_{i_1 \dotsc i_N } (y) \} : x, y \in V_0, x \neq y, i_1, \dotsc, i_N \in \{1, 2, 3 \} \}$,\ where $V_0 := \{p_1, p_2, p_3 \}$ and $\psi_{i_1 \dotsc i_N } := \psi_{i_1} \circ \dotsc \circ \psi_{i_N}.$\ Random weights $(\mu_{xy}^N )_{\{x, y \} \in E(G^N) }$ are i.i.d. random variables with a common distribution which is supported on $[c_1, c_2]$, where $0< c_1 \le c_2 < \infty.$ We will establish the following estimate of the cover time for $G^N$: \[sgprop\] There exist $c_3, c_4 > 0$ such that for all $N \in \mathbb{N}$, $\textbf{P}$-a.s., $$c_3 5^N \le t_{\text{cov} } (G^N) \le c_4 5^N,$$ and $(t_{\text{cov}} (G^N))_{N \in \mathbb{N}}$ is Type 2. To prove this proposition, we prepare some notations. For $i_1, \dotsc, i_n \in \{1, 2, 3 \}$ and $n \le N,$ let $G_{i_1 \dotsc i_n}^N$ be the induced graphs with vertex set $V(G_{i_1 \dotsc i_n}^N)$ which is the intersection of $V(G^N)$ and an equilateral triangle with vertices $\psi_{i_1 \dotsc i_n}(p_i), i = 1, 2, 3.$ We use the following lemma. The resistance estimate is obtained, for example, from arguments in section 7 of [@Ba] or section 1.3 of [@St]. \[sglem\] There exist $c_5, c_6 > 0$ such that for all $N \in \mathbb{N},$ $$c_5 3^N \le \|\mu(G^N)\| \le c_6 3^N,~~~c_5 \left(\frac{5}{3} \right)^N \le \text{diam}_R(G^N) \le c_6 \left(\frac{5}{3} \right)^N ~\textbf{P}\text{-a.s.}$$ Proof of Proposition \[sgprop\]. By Lemma \[sglem\], almost-sure versions of (\[condition-of-type2-(1)-1\]) and (\[condition-of-type2-(3)\]) hold for $v(N) = 3^N$ and $r(N) = (\frac{5}{3})^N.$ We only need to check an almost-sure version of (\[condition-of-type2-(1)-2\]) with $r(N) = (\frac{5}{3})^N.$\ Set $\ell_k^N = c_6 (\frac{5}{3})^{N - k}$ for $0 \le k < N$ and $\ell_k^N = 0$ otherwise. Let $x_{i_1, \cdots, i_k}^N$ be a fixed vertex in $V(G_{i_1 \cdots i_k}^N).$ By Lemma \[sglem\], $\{B_{\text{eff}}^N (x_{i_1 \cdots i_k}^N, \ell_k^N ) : i_1, \cdots, i_k \in \{1, 2, 3 \} \}$ is a covering for $G^N$ $\textbf{P}\text{-a.s.}$ In particular, we get $$n_{\text{cov}} (G^N, \ell_k^N ) \le 3^k ~\textbf{P}\text{-a.s.}$$ Therefore, we have for some $c_7 > 0$ and all $N \in \mathbb{N},$ $$\sum_{k = 1}^{N} \sqrt{\ell_{k-1}^N \log \{ n_{\text{cov}} (G^N, \ell_k^N) \}} \le c_7 \sqrt{\Big (\frac{5}{3} \Big )^N} ~\textbf{P}\text{-a.s.}$$ By Remark \[type2rems\] (3), we complete the proof. $\Box$ It will be possible to estimate cover times for Sierpinski gasket graphs in higher dimensions and nested fractals by applying arguments similar to the above proof. [Acknowledgements.]{}\ We would like to thank Professor Takashi Kumagai for fruitful discussions and careful reading of an early version of this paper. We would like to thank Dr. Ryoki Fukushima for helpful comments and suggestion that the Seneta-Heyde theorem (Lemma \[scgwlem\] (2)) is useful for the proof of Proposition \[scgwprop\]. [99]{} D. J. Aldous. Random walk covering of some special trees. [J. Math. Anal. Appl.]{} [157]{} (1991), 271-283. K. B. Athreya and P. E. Ney. [Branching Processes.]{} Dover Publications, 2004. M. T. Barlow. Diffusions on fractals. Springer, Berlin, 1998. Lecture Notes in Mathematics [1690]{}, Ecole d’Eté de Probabilités de Saint-Flour XXV-1995. M. T. Barlow, J. Ding, A. Nachmias, and Y. Peres. The evolution of the cover time. [Combin, Probab. Comput.]{} [20]{} (2011), 331-345. I. Benjamini and G. Kozma. A resistance bound via an isoperimetric inequality. [Combinatorica]{} [25]{} (6) (2005), 645-650. I. Benjamini and E. Mossel. On the mixing time of a simple random walk on the super critical percolation cluster. [Probab. Theory Relat. Fields]{} [125]{} (2003), 408-420. A. K. Chandra, P. Raghavan, W. L. Ruzzo, R. Smolensky and P. Tiwari. The electrical resistance of a graph captures its commute and cover times. [Comput. Complexity]{} [6]{} (1996/1997), 312-340. F. Chung and L. Lu. The diameter of sparse random graphs. [Adv. in Appl. Math.]{} [26]{} (2001), 257-279. C. Cooper and A. Frieze. The cover time of the giant component of a random graph. [Random Struct. Alg.]{} [32]{} (2008), 401-439. D. A. Croydon. Random walk on the range of random walk. [J. Stat. Phys.]{} [136]{} (2009), 349-372. D. A. Croydon, B. M. Hambly and T. Kumagai. Convergence of mixing times for sequences of random walks on finite graphs. [Electron. J. Probab.]{} [17]{} (2012), 1-32. D. Croydon and T. Kumagai. Random walks on Galton-Watson trees with infinite variance offspring distribution conditioned to survive. [Electron. J. Probab.]{}, [13]{} (2008), 1419-1441. J. Ding, J. R. Lee, and Y. Peres. Cover times, blanket times, and majorizing measures. [Ann. of Math.]{} [175]{} (2012), 1409-1471. J. Jonasson. On the cover time for random walks on random graphs. [Combin, Probab. Comput.]{} [7]{} (1998), 265-279. J. Jonasson and O. Schramm. On the cover time of planar graphs. [Elect. Comm. in Probab.]{} [5]{} (2000), 85-90. H. Kesten. Subdiffusive behavior of random walk on a random cluster. [Ann. Inst. H. Poincaré Probab. Statist.]{} [22]{} (1986), 425-487. M. Ledoux and M. Talagrand. [Probability in Banach Spaces.]{} Springer-Verlag, New York, 1991. D. A. Levin, Y. Peres, and E. L. Wilmer. [Markov chains and mixing times.]{} American Mathematical Society, Providence, RI, 2009. With a chapter by James G. Propp and David B. Wilson. P. Matthews. Covering problems for Brownian motion on spheres. [Ann. Probab.]{} [16]{} (1988), 189-199. A. Nachmias and Y. Peres. Critical random graphs: diameter and mixing time. [Ann. Probab.]{} [36]{} (2008), 1267-1286. A. Nachmias and Y. Peres. The critical random graph, with martingales. [Israel J. Math.]{} [176]{} (2010), 29-41. A. G. Pakes. Some new limit theorems for the critical branching process allowing immigration. [Stochastic Processes Appl.]{} [3]{} (1975), 175-185. G. Pete. A note on percolation on $\mathbb{Z}^d$: Isoperimetric profile via exponential cluster repulsion. [Elect. Comm. in Probab.]{} [13]{} (2008), 377-392. R. S. Strichartz. [Differential equations on fractals : a tutorial.]{} Princeton University Press, 2006. M. Talagrand. [The generic chaining.]{} Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2005. Upper and lower bounds of stochastic processes. [^1]: Research Institute for Mathematical Sciences, Kyoto University, Kyoto 606-8502, Japan. E-mail: [email protected]	{ "pile_set_name": "ArXiv" }
--- abstract: 'We report the first experimental realization of perfect W-state in a superconducting qubit based system. In contrast to maximally entangled state, the perfect W state is different in weights and phases of the terms contained in the maximally entangled W-state. The prefect W state finds important applications in quantum information processing tasks such as perfect teleportation, superdense coding, secret sharing etc. The efficiency of generation is quantified by fidelity which is calculated by performing full quantum state tomography. To verify the presence of genuine nonlocality in the generated state, we experimentally perform Mermin’s inequality tests. Further, we have also demonstrated splitting and sharing of quantum information using the experimentally generated state.' author: - Manoranjan Swain - Vipin Devrari - Amit Rai - 'Bikash K. Behera' - 'Prasanta K. Panigrahi' date: 'Received: date / Accepted: date' title: 'Generation of perfect W-state and demonstration of its application to quantum information splitting ' --- [example.eps]{} gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore Introduction ============ Entanglement [@l1; @n1] has become the central part of quantum computation and quantum information processing [@a4; @a5] tasks such as quantum teleportation [@m1; @a3; @m2; @a2], dense coding [@m3; @a25], quantum cryptography [@m4; @m5; @a17; @a18; @a19] etc. The correlation found between entangled quantum particles is quite different from that of classical correlation and are practically impossible according to classical physics [@l1]. The entanglement of bipartite states has been extensively studied. However multipartite entanglement [@p1; @p2] which involves entanglement between more than two subsystems has a much more complicated structure. Multipartite entangled states find applications in quantum computing and quantum information processing tasks [@a16; @a15]. Moreover, they are also relevant in diverse area of physics such as condensed matter physics [@a1] and quantum gravity [@a114]. In the class of multi-partite entangled states, there exist the W-state [@pw1] which is well recognized due to its robustness against particle loss. The general form of W-state [@pnfl_1] is given as, $$\label{eq1} \begin{split} &\arrowvert W\rangle= \sum p_i \|1_i,\{0\}\rangle,\\ \hspace{1cm} &\sum{\|p_i\|^2}=1. \end{split}$$ If one chooses $p_i$=$\frac{1}{\sqrt{N}}$(where N is the number of qubits), the obtained state is known as the maximally entangled W-state, which is given as, $$\label{eq3} \arrowvert W\rangle= \frac{1}{\sqrt{N}}(\arrowvert100.....0\rangle+ \arrowvert010.....0\rangle+........... + \arrowvert 0.....001\rangle)$$ There are many proposals [@gsa; @pkp; @a30; @a31; @a33] as well as experimental realizations [@a32; @a34; @a35] of maximally entangled W-state. However, the usability of maximally entangled W-state in teleportation was questioned by Gorbachev $\emph{et al.}$ [@gorbachev] and was concluded to show non-unit fidelity in quantum teleportation by Joo $\emph{et al.}$ [@pw2]. As suggested by Agrawal and Pati [@pw3], there exist another type of W-state which is useful for various quantum information processing tasks such as perfect quantum teleportation [@pw3], quantum superdense coding [@pw3], quantum information sharing [@pw4], quantum information splitting and sharing [@pw5] etc. These states are known as perfect W-states. These are slightly different in weights and phases of the terms contained in the maximally entangled form of W-state. The generalized three-qubit perfect W-state is given as, $$\label{eq2} \arrowvert W_{p,s}\rangle= \frac{1}{\sqrt{2+2s}}(\arrowvert100\rangle+ \sqrt{s}e^{i\Phi_1}\arrowvert010\rangle+ \sqrt{s+1}e^{i\Phi_2}\arrowvert001\rangle)$$ where s is a real number, $\Phi_1$ and $\Phi_2$ are phases acquired by the state. This general form contains a wide class of W-type states. For simplicity, we choose s=1, $\Phi_1$=0 and $\Phi_2$=0 and the resultant state can be represented as, $$\label{pw} \arrowvert W_{p,1}\rangle= \frac{1}{2}(\arrowvert100\rangle+ \arrowvert010\rangle+ \sqrt{2}\arrowvert001\rangle)$$ The state represented in Eq. $ \ref{pw}$ is found to have important applications in prefect teleportation and superdense coding [@pw3]. So, the experimental generation of the state is essential for quantum computation and information processing tasks. Earlier various schemes were proposed to generate this state [@pkp; @pw6; @pw8; @pw9; @pw10; @pw12]. However, to the best of our knowledge, the experimental generation of such a state has not been reported yet. Here for the first time we present the experimental generation of the perfect W-state in a superconducting system based platform. IBM, by developing a cloud based quantum computing platform [@pw13] has created new opportunities for research in quantum physics. The building blocks of this platform are superconducting qubits which are maintained at a very low temperature to be in working condition. The coherence times of these qubits ranges from 5$\mu$s to 55$\mu$s. This system has been utilized to perform diverse experiments such as, experimental violation of inequalities in support of quantum entanglement [@qv_f3; @n1_Diego], quantum router [@bn1], quantum simulations [@qv_LiertaarXiv2018; @bn2], demonstration of no-hiding theorem [@l3], demonstration of quantum secret sharing [@l4], demonstration of quantum walks [@walk1; @walk2; @walk3] etc. In the present work, we use IBM’s one of the five qubit quantum computer to experimentally generate the three-qubit entangled perfect W-state represented in Eq. $\ref{pw}$. To show the quality of generation we have done full quantum state tomography of the obtained state and calculated the state fidelity. In addition to the first experimental generation, we have performed an application i.e. splitting of quantum information using the experimentally generated state. Splitting of quantum information is a robust way of sending information as it requires mutual collaboration between the recipients. One can retrieve the information only when other recipients agree by becoming controller of the process. In this paper, we experimentally implement the scheme proposed in Ref. [@pw5] by considering perfect-W state as the quantum channel. Our work in this paper is organized as follows. Section \[generation\] and \[ent\] give the details of experimental generation and entanglement verification of the state. In Section \[inf\], we have demonstrated splitting of quantum information. Finally we conclude in Section \[con\]. Experimental generation of perfect W- state {#generation} =========================================== In this section, we give a detailed analysis on experimental generation of the perfect W-state. To experimentally generate the state, we choose IBM’s one of the superconducting qubit based five-qubit quantum computer platform, ‘ibmq-vigo’. ![Circuit for generation of perfect W-state. X is the Pauli X-gate and U3($\theta$, $\phi$, $\lambda$) is the general unitary gate.[]{data-label="circuit"}](PerfectWcircuit.png){width="90.00000%" height="33.00000%"} ![Experimental outcome of the circuit given in Fig. \[circuit\] when run for 8192 shots in ibmq-vigo. Bar represents the standard deviation from mean probability. []{data-label="outcome"}](untitled.png){width="60.00000%" height="40.00000%"} The experimental result of the algorithm shown in Fig. \[circuit\] is presented in Fig. \[outcome\]. The result is to be followed by considering the first digit as the first qubit. For example in the term ‘100’, the qubits are to be read in sequential manner i.e. ‘q0, q1, q2 $\to$ 1, 0, 0’. To characterize the experimentally obtained state only $\|Z\rangle$ basis measurements are not sufficient. To collect all information about the state complementary measurements are to be taken on the same state. The process followed in this work to fully characterize the state is ‘Quantum state tomography’ (QST) [@pw31; @bn3]. The objective of QST is to reconstruct the density matrix of the state by taking series of measurements on the obtained state. This reconstruction of density matrix needs $4^N$-1(N is the number of qubits) different measurements on the state. After the full quantum state tomography we followed Stokes’s formula [@pw31] to construct the density matrix. The stokes formula to calculate density matrix for three-qubit state is given by, $$\label{a1} \hat{\rho} = \frac{1}{2^3}\sum_{i,j,k=0}^{3} (S_i \otimes S_j \otimes S_k) \hat{\sigma}_i \otimes \hat{\sigma}_j \otimes \hat{\sigma}_k$$ where $S_0$= $P_{\|0\rangle}$ + $P_{\|1\rangle}$, $S_1$= $P_{\|+\rangle_x}$ - $P_{\|-\rangle_x}$, $S_2$= $P_{\|+\rangle_y}$ - $P_{\|-\rangle_y}$ and $S_3$= $P_{\|0\rangle}$ - $P_{\|1\rangle}$. $\sigma_0$, $\sigma_1$, $\sigma_2$ and $\sigma_3$ are identity and Pauli matrices respectively. The theoretical and experimental density matrices are represented in Fig. \[density\]. To check the closeness of the obtained state with the theoretical one we have calculated fidelity [@pw32], which is given as: $$\label{fid} F(\rho^T,\rho^E)= \bigg\|Tr \bigg({\sqrt{\sqrt{\rho^T}\rho^E\sqrt{\rho^T}}}\bigg)\bigg\|^2$$ The fidelity of the state generation is calculated to be 0.75 $\pm$ 0.02. Non-locality of perfect W-state {#ent} =============================== The non-locality of perfect W-state is verified by using Mermin’s inequality [@mermin]. $$\label{ineq} \|M\|=\|E(ABC)-E(A B'C')-E(A'B'C)-E(A'BC')\|\leq 2$$ The three particle form of Mermin’s inequality [@merminform] represented in Eq. $ \ref{ineq}$. A, A$^\prime $, B, B$^\prime $, C, and C$^\prime $ are set of measurements on particles. Note that in quantum mechanics, these set of measurements can be considered as spin measurements and hence can be specified by linear combination of Pauli spin operators. In the present case for the calculation of quantum mechanical expectation values of the terms appearing in the R.H.S of Eq. $\ref{ineq}$ the choice of spin measurements has been restricted to X-Z plane only. Considering the measurement directions for unprimed ones to Z-direction and primed ones to X-direction, the value of Mermin’s polynomial ($\|M\|$) is calculated to be $\approx$ 2.9.\ To perform this task in IBMQ use choose ibmq-vigo. Following the process as has been done in [@mp] the measurements are taken. The outputs of the measurements are listed in Table \[qv\_TabI\]. [p[.7cm]{}\| p[1.7cm]{}\| p[1.7cm]{} \|p[1.7cm]{}\| p[1.7cm]{}]{}\ OC & P $\pm$ SD for $ABC$ & P $\pm$ SD for $A'B'C$ & P $\pm$ SD for $A'BC'$ &P $\pm$ SD for $AB'C'$\ 000 & 0.039$\pm$0.003 & 0.253$\pm$0.007 & 0.370$\pm$0.018 & 0.370$\pm$0.007\ 001 & 0.458$\pm$0.007 & 0.137$\pm$0.004 & 0.037$\pm$0.003 & 0.037$\pm$0.014\ 010 & 0.235$\pm$0.007 & 0.020$\pm$0.002 & 0.059$\pm$0.012 & 0.039$\pm$0.009\ 011 & 0.009$\pm$0.001 & 0.122$\pm$0.010 & 0.069$\pm$0.003 & 0.303$\pm$0.003\ 100 & 0.230$\pm$0.003 & 0.025$\pm$0.005 & 0.055$\pm$0.012 & 0.065$\pm$0.003\ 101 & 0.009$\pm$0.001 & 0.119$\pm$0.009 & 0.252$\pm$0.024 & 0.051$\pm$0.013\ 110 & 0.004$\pm$0.002 & 0.235$\pm$0.004 & 0.090$\pm$0.006 & 0.083$\pm$0.009\ 111 & 0.012$\pm$0.001 & 0.084$\pm$0.005 & 0.066$\pm$0.006 & 0.048$\pm$0.004\ \[qv\_TabI\] Using the data presented in Table \[qv\_TabI\], the value of the Mermin’s polynomial is calculated to be 2.516 $\pm$ 0.027. The result shows the violation of the inequality and hence verifies the non-locality of the obtained state. Splitting quantum information {#inf} ============================= We start by giving a brief theoretical description of the process. Let us consider Alice, Bob and Charlie share an entangled perfect W-state of the form given below, $$\label{pw2} \arrowvert W\rangle= \frac{1}{2}(\arrowvert0_1 0_2 1_3\rangle+ \arrowvert0_1 1_2 0_3\rangle+ \sqrt{2}\arrowvert1_1 0_2 0_3\rangle)$$ Alice also possess an information qubit which is of the form, $$\|\phi_0\rangle= \alpha\|0_0\rangle+\beta\|1_0\rangle$$ The combined state is given as, $$\label{pwc} \begin{aligned} \arrowvert \Psi \rangle= \frac{1}{2}\alpha(\arrowvert0_0 0_1 0_2 1_3\rangle+ \arrowvert0_0 0_1 1_2 0_3\rangle+ \sqrt{2}\arrowvert0_0 1_1 0_2 0_3\rangle) \\ + \frac{1}{2}\beta(\arrowvert1_0 0_1 0_2 1_3\rangle+ \arrowvert1_0 0_1 1_2 0_3\rangle+ \sqrt{2}\arrowvert1_0 1_1 0_2 0_3\rangle) \end{aligned}$$ Alice then performs Bell measurement on her two qubits leaving Bob and Charlie as receivers. Neither Bob nor charlie can retrieve the information alone. Hence one of them has to become the controller and the other can retrieve the message by performing X and Z measurements on his qubit. Experimental implementation - At initial stage, the qubits q1, q2, q3 are prepared in perfect W-state form. The qubit q0 which was in a state $\|0\rangle$, is projected on to an arbitrary state by sequence of single qubit gates i.e. U3(pi/3,0,0), $T^\dagger$, $S^\dagger$, and H. The circuit to generate the state and its simulation result in ibmq-vigo are shown in Fig. \[stckt\] and Fig. \[stateresult\](a) respectively. ![Arbitrary state generation circuit in ibmq-vigo.[]{data-label="stckt"}](ArbitraryState.png){width=".6\linewidth"} Now let us assume, Alice has control over q0 and q1 and Bob and Charlie control q2 and q3 respectively. Alice performs Bell basis measurement on the two qubits she possess. The message can be received by any one from Bob and Charlie but not without the help of other. Let’s suppose Charlie becomes the controller and Bob becomes the receiver in the present case. To achieve this process Bob has to perform two-particle unitary transformation on the qubits q2 and q3. The required unitary matrix is given as: The circuit to implement this unitary operation between q2 and q3 is shown in Fig. \[unitary\]. This unitary operator projects charlie’s qubit to state $\|0\rangle$ and the message signal is transferred to Bob. Finally Bob can retrieve the message by applying controlled Pauli X and Z gates on his qubit. The whole communication process is shown in Fig. \[controlledcommunication\]. ![Unitary operation between Bob and charlie which makes Bob the receiver.[]{data-label="unitary"}](Unitary_Operation.png){width="90.00000%" height="3.cm"} ![Quantum information splitting circuit. The first section is for generation of the entangled state. The second section is for the Bell basis measurement and the third section is for controlled operation and q2 measurement. A.S. is abbreviated for arbitrary state.[]{data-label="controlledcommunication"}](controlledcommunication.pdf){width="120.00000%" height="8cm"} The communication process circuit was simulated as well as experimentally run on ibmq-vigo device for 8192 shots. The simulation and experimental outcome of measurement on q2 is shown in Fig. \[stateresult\] (b) and (c) respectively. Following the process of quantum state tomography the experimental density matrix of the state was obtained. The theoretical and the experimental (one out of many sets of experimental data) density matrices of the state are represented by the following equations: $$\label{ext} \rho^T= \begin{pmatrix} 0.194 & 0.250 \\ 0.250 & 0.806 \end{pmatrix} + i \begin{pmatrix} 0 & - 0.306 \\ 0.306 & 0 \end{pmatrix}$$ $$\label{exd} \rho^E= \begin{pmatrix} 0.289 & 0.174 \\ 0.174 & 0.709 \end{pmatrix} + i \begin{pmatrix} 0 & - 0.118 \\ 0.118 & 0 \end{pmatrix}$$ The fidelity of the communication process is calculated to be 0.805 $\pm$ .006. Conclusion {#con} ========== In conclusion we have shown the first experimental generation of perfect W-state on a superconducting qubit based platform with a fidelity of 0.75 $\pm$ 0.02. The Mermin inequality tests on the state clearly verified the presence of genuine entanglement between the qubits. Besides the experimental generation we have also shown controlled communication using the experimentally generated state. The fidelity of the communication process was obtained to be 0.805 $\pm$ .006. Acknowledgments {#acknowledgments .unnumbered} =============== One of the authors (A.R.) gratefully acknowledges a research grant from Science and Engineering Research Board (SERB), Department of Science and Technology (DST), Government of India (Grant No. CRG/2019/005749) during this work. B.K.B. acknowledges the prestigious Prime Minister’s Research Fellowship awarded by DST, Govt. India. Authors acknowledge IBM team for providing free access to their cloud computing platform. Einstein A., Podolsky B., and Rosen N., Can Quantum-Mechanical Description of Physical Reality Be Considered Complete? Phys. Rev. 47, 777 (1935) Horodecki R., Horodecki P., Horodecki M., and Horodecki K., Quantum entanglement. Rev. Mod. Phys. 81, 865 (2009) Bennett C. H.& DiVincenzo D., Quantum information and computation. Nature 404, 247 (2000) Ekert A. and Jozsa R., Quantum computation and Shor’s factoring algorithm. Rev. Mod. Phys. 68, 733 (1996) Bennett C. H., Brassard G., Crépeau C., Jozsa R., Peres A., and Wootters W. K., Teleporting an unknown quantum state via dual classical and Einstein-Podolsky-Rosen channels. Phys. Rev. Lett. 70, 1895 (1993) Gottesman D.& Chuang I. L., Demonstrating the viability of universal quantum computation using teleportation and single-qubit operations. Nature 402, 390 (1999) Pirandola S., Eisert J, Weedbrook C., Furusawa A., Braunstein S. L. , Advances in quantum teleportation. Nat. Photonics 9, 641 (2015) Zeilinger A., Quantum teleportation, onwards and upwards. Nature Physics 14, 3 (2018) Guo Y., Liu B. H., Li C. F., and Guo G. C., Advances in Quantum Dense Coding. Adv. Quantum Technol. 2, 1900011 (2019) Muralidharan S. and Panigrahi Prasanta K., Perfect teleportation, quantum-state sharing, and superdense coding through a genuinely entangled five-qubit state. Phys. Rev. A 77, 032321 (2008) Ekert A. K., Quantum cryptography based on Bell’s theorem. Phys. Rev. Lett. 67, 661 (1991) Gisin N. et al, Quantum cryptography. Rev. Mod. Phys. 74, 145 (2002) Jennewein T. et al, Quantum Cryptography with Entangled Photons. Phys. Rev. Lett. 84, 4729 (2000) Naik D. S. et al, Entangled State Quantum Cryptography: Eavesdropping on the Ekert Protocol. Phys. Rev. Lett. 84, 4733 (2000) Tittel W. et al, Quantum Cryptography Using Entangled Photons in Energy-Time Bell States. Phys. Rev. Lett. 84, 4737 (2000) Bhaskara V. S., Panigrahi P. K.,Generalized concurrence measure for faithful quantification of multiparticle pure state entanglement using Lagrange’s identity and wedge product. Quantum Information Processing 16, 118 (2017) Banerjee S , Panigrahi P. K., Quantifying Parallelism of Vectors is the Quantification of Distributed n- party Entanglement. J. Phys. A: Math. Theor. 53, 095301 (2020) Kempe J, Multiparticle entanglement and its applications to cryptography. Phys. Rev. A 60, 910 (1999) Yeo Y. and Chua W. K., Teleportation and Dense Coding with Genuine Multipartite Entanglement. Phys. Rev. Lett. 96, 060502 (2006). Amico L. et al, Entanglement in many-body systems. Rev. Mod. Phys. 80, 517 (2008) Rangamani M. and Takayanagi T. Holographic Entanglement Entropy, Lect. Notes Phys. 931, 1 (2017). Dür, W., Vidal, G., and Cirac, J.I.: Three qubits can be entangled in two inequivalent ways. Phys. Rev. A 62, 062314 (2000) Rai A. and Agarwal G. S., Possibility of coherent phenomena such as Bloch oscillations with single photons via W states. Phys. Rev. A 79, 053849 (2009) Biswas A. and Agarwal G. S., Preparation of W, GHZ, and two-qutrit states using bimodal cavities. J. Mod. Opt. 51, 1627 (2004) Rao D. D. B., Ghosh S., and Panigrahi P. K., Generation of entangled channels for perfect teleportation channels using multi-electron quantum dots. Phys. Rev. A 78, 042328 (2008) Wang H. F., et al, Robust and scalable scheme to generate multipartite atom–photon and atom–atom entangled W states by interference, J. Opt. Soc. Am. B 29, 257 (2012) Balachandran V. and Gong J., Scalable engineering of multipartite W states in a spin chain. Phys. Rev. A 85, 062303 (2012) Perez-Leija A. et al, Generating photon-encoded W-states in multiport waveguide-array systems. Phys. Rev. A 87, 013842 (2013) Neeley M. et al, Generation of three-qubit entangled states using superconducting phase qubits. Nature (London) 467, 570 (2010). Gräfe M. et al, On-chip generation of high-order single-photon W-states. Nature Photon 8, 791 (2014) Kagalwala K. H. et al, Single-photon three-qubit quantum logic using spatial light modulators. Nat Commun 8, 739 (2017) Gorbachev V.N., Trubilko A.I., Rodichkina A.A., Can the states of the W-class be suitable for teleportation. Phys. Lett. A, 314, 267 (2003) Joo, J., Park, Y. J., Oh, S., and Kim, J.: Quantum teleportation via a W state. New J. Phys. 5, 136 (2003) Agrawal P. and Pati A. K., Perfect teleportation and superdense coding with W states. Phys. Rev. A. 74, 062320 (2006) Nie Y.-y., Li Y.-h., Liu J.-c., and Sang M.-h., Quantum information splitting of an arbitrary three-qubit state by using a genuinely entangled five-qubit state and a Bell-state. Opt. Commun. 284, 1457 (2011) Zheng S. B., Splitting quantum information via W states. Phys. Rev. A 74, 054303 (2006) Liu Y. and Ding P. -C., Scheme for generating W states via distant cavities. J. At. Mol. Sci. 4, 79 (2013) Raissi Z. and Karimipour V., Creating maximally entangled states by gluing. Quant. Inf. Process. 16, 81 (2017) Xin B., Xu P., Zheng Y.Z., A robust scheme for generating a peculiar W-class state with atoms trapped in separate cavities. Int. J. Quantum Inf. 9, 947 (2011) Wang, XW., Yang, GJ., Su, YH. et al. Simple schemes for quantum information processing with W-type entanglement. Quantum Inf Process 8, 431 (2009) Dong L. et al, Nearly deterministic preparation of the perfect W state with weak cross-Kerr nonlinearities. Phys. Rev. A 93, 012308 (2016) IBM Quantum Experience. research.ibm.com/ quantum Alsina, D., Latorre, J. I.: Experimental test of Mermin inequalities on a five-qubit quantum computer. Phys. Rev. A 94, 012314 (2016) García-Martín, D. and Sierra, G.: Five Experimental Tests on the 5-Qubit IBM Quantum Computer. Journal of Applied Mathematics and Physics, 6, 1460-1475 (2018) Behera, B.K., Reza, T., Gupta, and Panigrahi, P.K.: Designing quantum router in IBM quantum computer. Quantum Inf. Process. 18, 328 (2019) Cervera-Lierta, A.: Exact Ising model simulation on a Quantum Computer. Quantum 2, 114 (2018) Manabputra, Behera, B.K. and Panigrahi, P.K. A simulational model for witnessing quantum effects of gravity using IBM quantum computer. Quantum Inf. Process. 19, 119 (2020) Pokharel, B., Anand, N, Fortman, B., and Lidar, D.: Demonstration of fidelity improvement using dynamical decoupling with superconducting qubits. Phys. Rev. Lett. 121, 220502 (2018) Kalra, A.R., Gupta, N., Behera, B.K. et al, Demonstration of the no-hiding theorem on the 5-Qubit IBM quantum computer in a category-theoretic framework. Quantum Inf. Process. 18, 170 (2019) Joy D., M Sabir, Behera B. K., Panigrahi P. K.,Implementation of quantum secret sharing and quantum binary voting protocol in the IBM quantum computer, Quantum Inf. Process. 19, 33 (2020). Chatterjee Y., Devrari V, Behera B. K., and Panigrahi P. K., Experimental realization of quantum teleportation using coined quantum walks, Quantum Inf Process 19, 31 (2020). Adhikari A., Behera B. K. and Panigrahi P. K., Circuit Design for Continuous Time Quantum Walks on Cycle Graph and its Experimental Demonstration in IBM Quantum Computer, DOI: 10.13140/RG.2.2.21140.76168. Balu, R., Castillo, D. and Siopsis, G. Physical realization oftopological quantum walks on IBM-Q and beyond, Quantum Sci. Technol. 3, 035001(2018). J. B. Altepeter, D. F. V. James, and P. G. Kwiat, Quantum State Tomography. Vishnu, P.K., Joy, D., Behera, B.K, and Panigrahi P.K.: Experimental demonstration of non-local controlled-unitary quantum gates using a five-qubit quantum computer. Quantum Inf. Process. 17, 274 (2018) M. A. Nielsen and I. L. Chuang, Quantum computationand quantum information, Cambridge University Press,Cambridge, UK, 2010. Mermin, N.D.: Extreme quantum entanglement in a superposition of macroscopically distinct states. Phys. Rev. Lett. 65, 1838 (1990) Cereceda, J. L.: Three-particle entanglement versus three-particle nonlocality. Phys. Rev. A 66, 024102 (2002) Swain, M., Rai, A., Behera, B.K., and Panigrahi P.K. Experimental demonstration of the violations of Mermin’s and Svetlichny’s inequalities for W and GHZ states. Quantum Inf Process 18, 218 (2019).	{ "pile_set_name": "ArXiv" }
--- abstract: 'GitHub is the largest source code repository in the world. It provides a git-based source code management platform and also many features inspired by social networks. For example, GitHub users can show appreciation to projects by adding [stars]{} to them. Therefore, the number of stars of a repository is a direct measure of its popularity. In this paper, we use multiple linear regressions to predict the number of stars of GitHub repositories. These predictions are useful both to repository owners and clients, who usually want to know how their projects are performing in a competitive open source development market. In a large-scale analysis, we show that the proposed models start to provide accurate predictions after being trained with the number of stars received in the last six months. Furthermore, specific models—generated using data from repositories that share the same growth trends—are recommended for repositories with slow growth and/or for repositories with less stars. Finally, we evaluate the ability to predict not the number of stars of a repository but its rank among the GitHub repositories. We found a very strong correlation between predicted and real rankings (Spearman’s $\mathit{rho}$ greater than 0.95).' author: - \| \ \ \ bibliography: - 'bibfile.bib' title: Predicting the Popularity of GitHub Repositories --- Introduction {#sec:introduction} ============ GitHub is the largest software repository in the world. Despite git-based source code management services (init, clone, add, commit, push, etc), GitHub also supports social coding features. For example, developers can show appreciation to a repository by using the [star]{} button, which essentially plays the same role as the [like]{} button in other social networks. Therefore, the number of stars of a repository works like an easily accessible and reliable proxy to its popularity [@Borges2015; @Hudson2016]. In fact, the top-starred repositories in GitHub are widely known software projects, e.g., [jquery/jquery]{} (a library to HTML scripting), [torvalds/linux]{} (the Linux kernel), [rails/rails]{} (a web framework for Ruby), and [docker/docker]{} (an application container engine). Due to the relevance of GitHub in modern open source development, researchers started to study the popularity of GitHub repositories. For example, a study by Zho et al. shows that the adoption of standard folders (e.g., doc, test, examples) may have an impact on project code popularity [@Zhu2014]. In another study, Aggarwal et al.[@Aggarwal2014] show that popular projects tend to attract more documentation collaborators. Weber and Luo attempted to differentiate popular and unpopular Python projects on GitHub using machine learning techniques [@Weber2014]. Recently, we investigated the factors that impact the popularity of GitHub repositories, including programming language, application domain, repository owner (user or organization), age, and release frequency [@Hudson2016]. For this purpose, we collected historical data about the number of stars of 2,500 popular repositories. We also used this dataset to identify four patterns of popularity growth, which we called slow, moderate, fast, and viral. In this paper, we extend our previous work by investigating the use of multiple linear regressions to [predict the popularity]{} of GitHub repositories. Prediction models have been successfully used to infer the popularity of content in other social networks, such as the number of views of YouTube videos [@Roy2013; @Pinto2013; @Figueiredo2013] and the number of tweets associated to a given hashtag [@tsur2012; @ma2012; @ma2013]. However, to our knowledge, we are the first to attempt to predict the popularity—measured by the number of stars—of software projects hosted at GitHub. Specifically, we compute and investigate multiple linear regression models over two types of data: [generic]{} and [specific]{}. By generic, we refer to models produced from the complete dataset considered in this paper, which includes historical data about the number of stars of 4,248 popular GitHub repositories. By specific, we refer to models produced from repositories that share similar growth trends. These trends are inferred using the KSC algorithm [@Yang2011], which clusters time series with similar shapes. We address three major research questions in the paper: - [RQ \#1: What is the accuracy of the generic prediction models?]{} We report the Relative Squared Error (RSE) of the regression models computed using the time series of number of stars of all projects in our dataset. - [RQ \#2: What is the accuracy of the specific prediction models?]{} First, using the KSC clustering algorithm, we identify four major growth trends among the systems in our dataset. Then, we evaluate the accuracy of the regression models computed over the time series of each cluster. - [RQ \#3: What is the accuracy of the repositories rank as predicted using the generic and specific models?]{} In the previous RQs, our goal is to the predict the total number of stars, using generic and specific models. By contrast, in this final RQ, we evaluate the ability of both models to predict not the number of stars of a repository after a time, but its rank among the repositories in our dataset. The remainder of this paper is organized as follows. First, we present the dataset used in the paper (Section \[sec:dataset\]) and discuss the methodology followed in the study (Section \[sec:study-design\]). Section \[sec:results\] presents our results, by exploring and discussing answers for the three proposed research questions. Finally, Section \[sec:validity\] discusses threats to validity and Section \[sec:conclusion\] concludes the paper. Dataset {#sec:dataset} ======= The initial dataset used in this paper includes historical data about the top-5,000 public repositories with more stars in GitHub. All data was obtained using the GitHub API, which provides services to search public repositories and to retrieve specific information about them (e.g., stars). First, we collect basic data about the repositories (i.e., owner, stars, creation date, programming language, etc.). Next, for each repository, we collect historical data about the number of stars. For this purpose, we used a service from the API that returns all star events of a given repository. For each star, these events store the date and the user responsible to starring the repository. However, GitHub API returns at most 100 events by request (i.e., a page) and at most 400 pages. For this reason, it is not currently possible to retrieve all star events of systems with more than 40K stars, which is the case of seven repositories: [FreeCodeCamp]{} (112,397 stars), [twbs/bootstrap]{} (95,293 stars), [vhf/free-programming-books]{} (54,208 stars), [mbostock/d3]{} (49,173 stars), [angular/angular.js]{} (48,787 stars), [FortAwesome/Font-Awesome]{} (41,621 stars), [facebook/react]{} (41,037 stars). Moreover, 278 repositories have no main programming language identified. These repositories do not store source code, e.g., [jlevy/the-art-of-command-line]{} (26,298 stars) or are moved/removed repositories, e.g., [nodejs/node-v0.x-archive]{} (37,354 stars). Therefore, we also remove these repositories from the dataset. Additionally, we only consider the stars gained in the last 52 weeks of each repository. Thus, repositories with less than 52 weeks are also removed from the dataset (468 repositories). Figure \[fig:dataset-stars\] shows the number of stars of the 4,248 repositories in our dataset. This number ranges from 39,149 stars ([jquery/ jquery]{}) to 1,248 stars ([mikeflynn/egg.js]{}). As presented, the distribution is right skewed (quantiles 5% = 1,307 stars and 95% = 9,360 stars). The mean and median number of stars are 3,393 and 2,240, respectively. Table \[tab:dataset-top\] lists the top-10 repositories with more stars. These repositories have at least 30K stars and belong to four different domains (Web Frameworks and Libraries, Software Tools, Documentation, and System Software). ---------------------------- -------- -------- [jquery/jquery]{} Web 39,149 [robbyrussell/oh-my-zsh]{} Tools 36,373 [airbnb/javascript]{} Doc 34,064 [h5bp/html5-boilerplate]{} Web 33,704 [meteor/meteor]{} Web 33,594 [torvalds/linux]{} System 31,702 [daneden/animate.css]{} Web 31,549 [facebook/react-native]{} Web 31,217 [rails/rails]{} Web 30,779 [docker/docker]{} System 30,742 ---------------------------- -------- -------- : Top-10 repositories with more stars[]{data-label="tab:dataset-top"} Next, we built the stars time series of each repository from the stars events. These time series consist of the number of stars gained by week since the repository creation date up to April 25, 2016, when we collected our data. As an example, Figure \[fig:dataset-jquery-timeseries\] shows the time series retrieved for [jquery/jquery]{}, the most starred repository in our dataset. This repository has 369 weeks (x-axis) and the number of stars increased from 1,692 stars to 39,149 stars (y-axis). Study Design {#sec:study-design} ============ In this section, we detail the techniques and models used to predict the number of stars of GitHub repositories. We also discuss how we evaluate the accuracy of these models. [Prediction Technique:]{} We rely on multiple linear regression to predict the popularity of GitHub repositories. Multiple linear regression differs from simple regression by considering that all variables are not equally important [@freedman2009statistical]. The general form of a multiple linear regression is as follows: $$Y_{t} = b_{0} + b_{1}X_{t_{1}} + b_{2}X_{t_{2}} + ... + b_{r}X_{t_{r}}$$ where $Y_{t}$ is the dependent variable (number of stars at week $t$), $X_{t_{i}}$ are the independent variables (stars in weeks $i$, $1 \leq i \leq r \leq t$), and $b_{j}$ are the regression coefficients ($0 \leq j \leq r \leq t$). [Estimating the Errors:]{} To evaluate the accuracy of the models, we use the Relative Squared Error (RSE). Assume that $N(r, t)$ is the real number of stars of a repository $r$ in the week $t$. Moreover, assume that $\widehat{N}(r, t_{r}, t)$ is the number of stars predicted for $r$ at the week $t$ from the popularity data of the first $t_{r}$ weeks. The RSE for this prediction is given by[@Pinto2013]: $$RSE = \left ( \frac{\widehat{N}(r, t_{r}, t)}{N(r, t)} - 1 \right )^{2}$$ For a collection $\mathcal{R}$ of repositories, the mean Relative Squared Error (mRSE) is defined as the arithmetic mean of the RSE values of all repositories in $\mathcal{R}$, as given by: $$mRSE = \frac{1}{\left \|\mathcal{R} \right \|} * \sum_{r \in \mathcal{R}} \left ( \frac{\widehat{N}(r, t_{r}, t)}{N(r, t)} - 1 \right )^{2}$$ [Cross-Validation:]{} As ilustrated in Figure \[fig:cross-validation\], we perform cross-validation to assess the prediction models. We use 10 folds, i.e., the repositories are randomly partitioned in 10-folds and we use nine folds to build the prediction models (training set) and the remaining fold to evaluate their accuracy (validation set). [Generic and Specific Models:]{} We generate models for two datasets: [generic]{} and [specific]{}. By generic, we refer to models produced from the complete dataset, i.e., from the time series with the number of stars collected for 4,248 repositories. By specific, we refer to models produced from repositories that shared similar growth trends. As in our previous work [@Hudson2016], we rely on the KSC algorithm [@Yang2011] to identify growth trends in our dataset. This algorithm clusters time series with similar shapes using a metric that is invariant to scaling and shifting. In other words, each cluster groups time series that share similar growth trends. Particularly, to answer RQ \#2 we produce specific models considering only the time series in each cluster. We use the $\beta_{CV}$ heuristic [@Menasce2001] to define the best number $k$ of clusters. $\beta_{CV}$ is defined as the ratio of the coefficient of variation of the intracluster distances and the coefficient of variation of the intercluster distances. The smallest value of $k$ after which the $\beta_{CV}$ ratio remains roughly stable should be selected. In our dataset, the values of $\beta_{CV}$ stabilize for $k = 5$ (see Figure \[fig:bcv\]). Therefore, we configure KSC to produce five clusters. Figure \[fig:centroids\] shows the time series representing the clusters’ centroids of the five clusters. The trends presented by clusters C1, C2, and C3 suggest a linear growth in the number of stars. The trend presented by cluster C4 differs from the first three ones due variations in the number of stars over the time. Finally, cluster C5 suggests a viral growth. As presented in Table \[tab:clusters-description\], cluster C1 concentrates almost half of the repositories in our dataset (49.1%) while cluster C5 has the lowest concentration (1.2%). Table \[tab:clusters-description\] also presents the percentage of growth of each cluster, considering the centroids time series. This percentage ranges from 19.9% (cluster C1) to 1,659.1% (cluster C5). ---- --------------- --------- -- C1 2,087 (49.1%) 19.9 C2 1,456 (34.2%) 61.3 C3 521 (12.2%) 175.1 C4 131 (3.0%) 883.2 C5 53 (1.2%) 1,659.1 ---- --------------- --------- -- : Popularity Trends Description[]{data-label="tab:clusters-description"} When answering RQ \#2, we do not consider specific models for cluster C5 due to two main reasons: (a) it includes only 53 repositories (1.2%); (b) as presented in Figure \[fig:centroids\], the time series in this cluster do not have a linear shape.\ [Repositories Ranking:]{} To answer RQ \#3, we compute three rankings: (i) repositories sorted according to the predicted number of stars using a generic prediction model (configured with $t_{r} = 26$ weeks and $t = 52$); (ii) repositories sorted according to the predicted number of stars using specific prediction models (configured with $t_{r} = 26$ weeks and $t = 52$); (iii) repositories sorted according to their real number of stars, as provided by GitHub API, on April 25, 2016, i.e., the last week we consider to build the time series of stars. In the first two rankings, the rank positions range from 1 to 4,248 (which is the dataset size). However, the third ranking includes all repositories in the previous rankings plus 468 repositories that entered the list of the most popular repositories in the year before April 25, 2016. As examples, we have [apple/swift]{} and [Netflix/falcor]{}. [Netflix/falcor]{} is not among the top-5,000 most popular repositories on April 25, 2015, when we select the repositories used in the study, but it gained popularity to the point of being the 481st most popular repository one year later. [apple/swift]{} was created on October 10, 2015; despite this it is the 23rd most popular repository on April 25, 2016, when we define the real ranking. In this way, the investigation conducted to answer RQ \#3 includes the cases where a repository falls in the ranking not only due to a better performance of the repositories used to produce the prediction models, but also due to the performance of any other repository. [@lrrrl@]{} & & &\ & 6,160 & 5,369 & -12.84 & ------------------------------------------------------------------------ \ [robbyrussell/oh-my-zsh ]{} & 13,536 & 11,829 & -12.61 & ------------------------------------------------------------------------ \ [airbnb/javascript ]{} & 17,026 & 14,882 & -12.59 & ------------------------------------------------------------------------ \ [h5bp/html5-boilerplate ]{} & 4,896 & 4,691 & -4.19 & ------------------------------------------------------------------------ \ [meteor/meteor ]{} & 9,919 & 10,082 & +1.64 & ------------------------------------------------------------------------ \ [torvalds/linux ]{} & 10,566 & 9,682 & -8.37 & ------------------------------------------------------------------------ \ [daneden/animate.css ]{} & 10,492 & 9,452 & -9.91 & ------------------------------------------------------------------------ \ [facebook/react-native ]{} & 18,443 & 19,373 & +5.04 & ------------------------------------------------------------------------ \ [rails/rails ]{} & 5,701 & 5,128 & -10.05 & ------------------------------------------------------------------------ \ [docker/docker ]{} & 10,268 & 9,721 & -5.33 & ------------------------------------------------------------------------ \ & 213 & 298 & +39.91 & ------------------------------------------------------------------------ \ [infinitered/ProMotion ]{} & 119 & 238 & +100.00 & ------------------------------------------------------------------------ \ [nslocum/design-patterns-in-ruby ]{} & 640 & 731 & +14.22 & ------------------------------------------------------------------------ \ [jbt/markdown-editor ]{} & 621 & 744 & +19.81 & ------------------------------------------------------------------------ \ [mumble-voip/mumble ]{} & 565 & 667 & +18.05 & ------------------------------------------------------------------------ \ [Manabu-GT/ExpandableTextView ]{} & 676 & 623 & -7.84 & ------------------------------------------------------------------------ \ [apache/flink ]{} & 890 & 712 & -20.00 & ------------------------------------------------------------------------ \ [mafintosh/mongojs ]{} & 322 & 381 & +18.32 & ------------------------------------------------------------------------ \ [rofl0r/proxychains-ng ]{} & 813 & 790 & -2.83 & ------------------------------------------------------------------------ \ [mikeflynn/egg.js ]{} & 584 & 793 & +35.79 & ------------------------------------------------------------------------ \ Results {#sec:results} ======= *RQ \#1: What is the accuracy of the generic prediction models?\ In order to start answering this question, we produce [generic]{} prediction models and assess their accuracy using 10-fold cross validation for different values of $t_{r}$ (prediction data, see Figure \[fig:cross-validation\]). In all cases, we use the models to predict the number of stars at week 52 ($t = 52$, in Figure \[fig:cross-validation\]). In other words, we use the number of stars in the first $t_{r}$ weeks to predict the number of stars in the 52nd week (last week we considered when collecting the number of stars). Figure \[fig:generic-mrse\] reports the average error (mRSE) across all models. For small values of $t_{r}$ the models do not perform well, e.g., for $t_{r} = 10$ weeks mRSE $= 5.858 \pm 4.372$ (mean $\pm$ 95% confidence interval). However, as we increase the values of $t_{r}$, the results are more accurate. For example, mRSE $= 0.432 \pm 0.257$ for $t_{r} = 26$ weeks. This means that we can predict with a low error the number of stars six months ahead, using as training data the past six months of stars. Figure \[fig:generic-correlation\] shows a scatter plot that correlates the number of stars gained and the RSE for the generic models produced using $t_{r} = 26$ weeks. Each point in this figure represents a repository. We ran Spearman’s rank correlation test and the resulting correlation coefficient $\mathit{rho}$ is -0.50, with $\emph{p-value} < 0.001$. Therefore, the generic models are more accurate for the repositories that gained many stars in the period. Table \[tab:generic-example\] lists the prediction results for the top-10 and bottom-10 repositories with more stars in our dataset. The column [[“Stars”]{}]{} shows the real number of stars gained in 52 weeks and the column [[“Predicted”]{}]{} presents the number of stars predicted for the same period using a generic model ($t_{r} = 26$ weeks and $t = 52$). The difference between the real and the predicted values ranges from 1.64% to 12.84%, in absolute values, for the top-10 repositories and from 2.83% to 100% for the bottom-10 ones. [0.25]{} [0.25]{} [0.25]{} [0.25]{} Finally, we evaluate the accuracy of the generic prediction models for different values of the target week $t$. Figure \[fig:generic-mrse2\] shows the average error for $t = 26$ weeks (half year) and $t = 104$ weeks (two years). The figure also includes the average error for $t = 52$ weeks (already presented in Figure \[fig:generic-mrse\]). In all cases, we see a decreasing trend of the average error measure. However, for higher values of $t$, we need less prediction data to achieve similar average errors. For example, using $t = 52$ weeks and a fraction of time equals to 0.5 (i.e, 26 weeks) the average error (mRSE) is $0.432 \pm 0.257$. For $t = 104$ weeks, a similar average error (mRSE $= 0.460 \pm 0.182$) happens for a fraction of time of 0.36 (i.e., 38 weeks).\ [@lcrrrrl@]{} & & & & &\ & C1 & 6,160 & 5,578 & -9.45 & +3.39 & ------------------------------------------------------------------------ \ [robbyrussell/oh-my-zsh ]{} & C2 & 13,536 & 12,826 & -5.25 & +7.37 & ------------------------------------------------------------------------ \ [airbnb/javascript ]{} & C2 & 17,026 & 20,140 & +18.29 & -5.70 & ------------------------------------------------------------------------ \ [h5bp/html5-boilerplate ]{} & C1 & 4,896 & 4,690 & -4.21 & -0.02 & ------------------------------------------------------------------------ \ [meteor/meteor ]{} & C2 & 9,919 & 10,571 & +6.57 & -4.93 & ------------------------------------------------------------------------ \ [torvalds/linux ]{} & C2 & 10,566 & 10,498 & -0.64 & +7.72 & ------------------------------------------------------------------------ \ [daneden/animate.css ]{} & C2 & 10,492 & 10,045 & -4.26 & +5.65 & ------------------------------------------------------------------------ \ [facebook/react-native ]{} & C3 & 18,443 & 18,432 & -0.06 & +4.98 & ------------------------------------------------------------------------ \ [rails/rails ]{} & C1 & 5,701 & 5,386 & -5.53 & +4.53 & ------------------------------------------------------------------------ \ [docker/docker ]{} & C2 & 10,268 & 9,468 & -7.79 & -2.46 & ------------------------------------------------------------------------ \ & C1 & 213 & 209 & -1.88 & +38.03 & ------------------------------------------------------------------------ \ [infinitered/ProMotion ]{} & C1 & 119 & 155 & +30.25 & +69.75 & ------------------------------------------------------------------------ \ [nslocum/design-patterns-in-ruby ]{} & C3 & 640 & 922 & +44.06 & -29.84 & ------------------------------------------------------------------------ \ [jbt/markdown-editor ]{} & C2 & 621 & 667 & +7.41 & +12.40 & ------------------------------------------------------------------------ \ [mumble-voip/mumble ]{} & C2 & 565 & 583 & +3.19 & +82.35 & ------------------------------------------------------------------------ \ [Manabu-GT/ExpandableTextView ]{} & C3 & 676 & 729 & +7.84 & 0 & ------------------------------------------------------------------------ \ [apache/flink ]{} & C3 & 890 & 853 & -4.04 & +14.29 & ------------------------------------------------------------------------ \ [mafintosh/mongojs ]{} & C1 & 322 & 309 & -4.04 & +14.29 & ------------------------------------------------------------------------ \ [rofl0r/proxychains-ng ]{} & C3 & 813 & 886 & +8.98 & -6.15 & ------------------------------------------------------------------------ \ [mikeflynn/egg.js ]{} & C1 & 584 & 523 & -10.45 & +25.34 & ------------------------------------------------------------------------ \ RQ \#2: What is the accuracy of the specific prediction models?\ \[sec:results:rq2\] In this second research question, we generate [specific]{} prediction models for the repositories in each cluster (presented in Section \[sec:study-design\]) and assess their accuracy using 10-fold cross validation for different values of $t_{r}$. As in the first question, we predict the number of stars at week 52. Figure \[fig:specific-mrse\] reports the average error across all specific models. Cluster C1, which concentrates almost half of the repositories, presents a fast decreasing in the average error, e.g., mRSE $= 18.500 \pm 14.501$ for $t_{r} = 1$ week and $0.127 \pm 0.020$ for $t_{r} = 10$ weeks. This suggests that specific models for this cluster require very few data to provide accurate predictions. Cluster C2 also presents accurate results for any value of $t_{r}$. As we can observe, the accuracy of the models for cluster C2 is better than the accuracy for C1 when considering $t_{r} \leq 6$ weeks. However, for $t_{r} > 6$ weeks, the accuracy of C2 is slightly lower. For example, mRSE $= 0.030 \pm 0.009$ ($t_{r} = 26$ weeks) and mRSE $= 0.038 \pm 0.009$ ($t_{r} = 26$ weeks) for clusters C1 and C2, respectively. Cluster C3, which presents the fastest linear trend, shows an initial increasing in the average error, followed by a drastic reduction. This happens due to inaccurate results of two repositories: [tessalt/echo-chamber-js]{} (RSE = $284.29$) and [gilesbowkett/rewind]{} (RSE = $224.48$), which gained a high number of stars at weeks 8 and 13, respectively.[^1] Figure \[fig:specific-improvement-clusters\] shows boxplots with the improvements per cluster. The improvements are calculated from the gains achieved by specific models ($t_{r} = 26$ weeks). As we can observe, specific models improve the predictions in all clusters, considering the median values. The median improvements for each cluster are 15.72%, 1.08%, 2.00%, and 6.66%, respectively. The repositories in cluster C1 take more advantage of specific models (1st quartile = 2.43%). By contrast, clusters C3 and C4 have the highest percentage of repositories with a worst performance (1st quartile equal to -9.46% and -11.79%, respectively). Table \[tab:specific-example\] lists the specific prediction results for the top-10 and bottom-10 repositories with more stars in our dataset. The column [[“Stars”]{}]{} shows the real number of stars gained in 52 weeks and the column [[“Predicted”]{}]{} presents the number of stars predicted for the same period using specific models ($t_{r} = 26$ weeks). The difference between the real and predicted number of stars ranges from 0.06% to 18.29%, in absolute values, for the top-10 repositories. Column [[“[”]{}]{}% Improve]{} shows the gains achieved by specific models, when compared with the predictions provided by the generic models. As we can see, the specific models increase the accuracy of the predictions for six out of ten repositories, from 3.39% to 7.72%. For the bottom-10 repositories, the difference between the real and predicted values ranges from 1.88% to 44.06% (column [[“[”]{}]{}% Diff]{}). In this case, the specific models produced more accurate results for seven out of ten repositories (column [[“[”]{}]{}% Improve]{}). RQ \#3: What is the accuracy of the repositories rank as predicted using the generic and specific models?\ Figure \[fig:prediction-points\] shows scatter plots correlating the real rank and predicted rankings using generic and specific models. The red line represents the identity function, i.e., a perfect match between the real and predicted ranks. Points above this line are repositories where the predicted rank is higher than the real one (we refer to this kind of error as an underestimation; e.g., a repository is predicted at the 10th position, but in fact it is in position 5th). By contrast, points below the identity line have a predicted rank lower than the real one (we refer to this error as an overestimation; e.g., a repository is predicted at the 5th position, but in fact it in in position 10th). Initially, we can observe that both models tend to overestimate many predictions, i.e., we usually have more points below the identity line. This happened because 468 repositories were created and/or quickly became more popular than the ones in our dataset. These repositories are called newcomers, in the context of this research question. Suppose for example that a newcomer appears at rank $i$; in this case it increases the rankings of all repositories with a predicted rank greater than $i$. This shift in the rankings is not detected by the prediction models we investigate, since they do not have information about new systems appearing in the rankings. However, we decide to consider newcomers in this first part of RQ \#3 to simulate a situation that will appear in the practice. [0.5]{} [0.5]{} To compare the real and the predicted rankings, we use the Spearman’s correlation test. Since the test requires as input two vectors with the same size, we removed the newcomers from the real rankings. For the generic model, we found a strong correlation between the ranks ($\mathit{rho} = 0.9534$ and p-value* $< 0.001$). For the specific models, the correlation is slightly better ($\mathit{rho} = 0.9777$ and p-value $< 0.001$). Figure \[fig:generic-spearman-rho\] shows the Spearman’s coefficient for different groups of top-repositories. For the top-16 repositories, the correlation using the generic model is lower than the one using the specific models ($\mathit{rho} = 0.9321$ and $\mathit{rho} = 0.9821$, respectively). For the other top-values values, this difference decreases. However, the rankings as predicted by the specific models present slightly better results in all cases. -------------------------------------------- ------- -- ------- ------- -- -------- ------ [jquery/jquery ]{} 1 1 1 0 0 [robbyrussell/oh-my-zsh ]{} 2 2 2 0 0 [airbnb/javascript ]{} 3 5 3 2 0 [h5bp/html5-boilerplate ]{} 4 4 5 0 1 [meteor/meteor ]{} 5 3 4 -2 -1 [torvalds/linux ]{} 6 8 6 2 0 [daneden/animate.css ]{} 7 9 8 2 1 [facebook/react-native ]{} 8 6 7 -2 -1 [rails/rails ]{} 9 10 10 1 1 [docker/docker ]{} 10 11 11 1 1 [jbt/markdown-editor ]{} 4,707 3,908 4,001 -799 -706 [apache/flink ]{} 4,708 4,180 4,167 -528 -541 [google/ion ]{} 4,709 — — — — [Manabu-GT/ExpandableTextView ]{} 4,710 4,149 4,016 -561 -694 [iPaulPro/Android-ItemTouchHelper-Demo ]{} 4,711 — — — — [mumble-voip/mumble ]{} 4,712 4,011 4,092 -701 -620 [mafintosh/mongojs ]{} 4,713 4,080 4,164 -633 -549 [mikeflynn/egg.js ]{} 4,714 3,569 4,194 -1,145 -520 [wequick/Small ]{} 4,715 — — — — [rofl0r/proxychains-ng ]{} 4,716 4,136 3,824 -580 -892 -------------------------------------------- ------- -- ------- ------- -- -------- ------ Table \[tab:prediction-example\] shows the predicted rank for the top-10 and bottom-10 repositories in our dataset. The column [[“Real”]{}]{} represents the real rank in GitHub when our dataset was collected. The column [[“Predicted”]{}]{} shows the predicted rank using the generic (column [[“Generic”]{}]{}) and the specific models (column [[“Specific”]{}]{}). Repositories that were created or became popular after the date we start collecting the time series are marked with [[“—”]{}]{}. Finally, the column [[“Diff”]{}]{} shows the difference between the predicted and the real rankings. As mentioned, both predictions are more accurate for the top repositories and tend to overrate the rank of the bottom repositories. For the top-10 repositories, the difference in absolute values between the ranks ranges from 0 to 2 (generic models) and from 0 to 1 (specific models). For the bottom-10 repositories, the difference between the ranks ranges from 528 to 1,145 (generic models) and from 520 to 892 (specific models). However, it is important to notice that the distribution of the number of stars per repository is heavy-tailed [@Borges2015]. Therefore, minor differences in the predicted number of stars can represent a movement of hundreds of positions in the relative order of a repository. For example, an error of 175 stars in the number of stars predicted to [Apache/Flink]{} is responsible to generate a error of 528 positions in its ranking (from position 4,708 in the real ranking; to position 4,180 in the ranking predicted using a generic model). Threats to Validity {#sec:validity} =================== [Measuring popularity using the number of stars.]{} In the investigation reported in this paper, we measure popularity using the number of stars of the GitHub repositories, as in other studies [@Weber2014; @Aggarwal2014]. However, we highlight that developers can star a repository for other reasons, for example, to create bookmarks.\ [Repositories selection.]{} GitHub has 17,136,765 public repositories, including forks (in June 15, 2016). For this study, we started with the top-5,000 repositories with more stars and after a cleaning step, we analyze 4,248 repositories. However, we stress that our goal is to predict popularity of most starred repositories. For example, 12,462,551 repositories (73%) have no stars and probably will never receive one in their lifetime. In other words, it is probably easier (and less useful) to make predictions for a dataset with all public repositories in GitHub.\ [Growth trends]{}. The selection of the number of clusters is a key parameter in clustering algorithms like KSC. To mitigate this threat, we use the $\beta_{CV}$ heuristic [@Menasce2001] to define the best number $k$ of clusters. We also discard cluster C5 from our evaluation of specific models (RQ \#2), since the repositories in this cluster do not follow a linear growth. Notice, however, that cluster C5 includes only 53 repositories (1.2%). Related Work {#sec:related_work} ============ Our work was inspired by the vast literature on defect prediction. For example, a systematic literature review listed 208 defect prediction studies [@hall12], which differ regarding the software metrics used for prediction, the modeling technique, the granularity of the independent variable, and the validation technique. As independent variables, the studies use source code metrics (size, cohesion, coupling, etc), change metrics, process metrics, code smells instances, etc. The modeling techniques vary with respect to linear regression, logistic regression, naive bayes, neural networks, etc. In this paper, instead of predicting the future number of defects of a system, we rely on multiple linear regressions to predict the number of stars of GitHub repositories. In a previous paper, we studied the popularity of GitHub repositories aiming to answer four research questions [@Hudson2016]. We concluded that the three most common domains on GitHub are web libraries and frameworks, non-web libraries and frameworks, and software tools. Additionally, we found that repositories owned by organizations are more popular than the ones owned by individuals (RQ \#1). We also found a strong correlation between stars and forks and a weak correlation both between stars and commits, and between stars and contributors (RQ \#2). We concluded that repositories have a tendency to receive more stars right after their first release (RQ \#3). We also showed that there is an acceleration in the number of stars gained after releases (RQ \#4). In a previous note, we started an investigation about the popularity of GitHub repositories [@Borges2015]. We showed that the number of stars follows a highly skewed distribution. Jian et al. explore [why]{} and [how]{} developers fork [what]{} from [whom]{} in GitHub [@jiang2016]. They report that some repository owners are popular, and attract many forks; other owners are unpopular and rarely attract forks. They also present that attractive owners have higher percentage of organizations, more followers and earlier registration in GitHub. Since forks are relevant operations in GitHub, future work may investigate prediction models for number of forks. Martin et al. [@Martin2016FSE] record time-series information about popular Google Play apps and investigate how release frequency can affect an app’s performance, as measured by rating, popularity and number of user reviews. They label as “impactful releases” the ones that caused a significant change on the app’s popularity, as inferred by Causal Impact Analysis (a form of causal inference). They report that more mentions of features and fewer mentions of bug fixing increase the chance for a release to be impactful. Couto et al. [@jss2014] follow a similar approach but to identify causal relationships between changes in internal measures of software quality (coupling, cohesion, complexity, etc) and the number of defects reported for a system. Several other studies examine the relationship between popularity of mobile apps and their code properties [@Datta2013; @Fu2013; @LinaresVasquez2013; @Ruiz2014; @Tian2015; @Guerrouj2015; @Palomba2015; @Corral2015]. Yuan et al. investigate 28 factors along eight dimensions to understand how high-rated Android applications are different from low-rated ones [@Tian2015]. Their result shows that external factors, like number of promotional images, are the most influential factors. Ruiz et al. examine the relationship between the number of ad libraries and app’s user ratings [@Ruiz2014]. They show that there is no relationship between the number of ad libraries in an app and its rating. Guerrouj et al. analyse changes of Android API elements between releases and report that high app churn leads to lower user ratings [@Guerrouj2015]. Linares-V[á]{}squez et al. investigate how the fault- and change-proneness of Android API elements relate to applications’ lack of success [@LinaresVasquez2013]. They state that making heavy use of fault- and change-prone APIs can negatively impact the success of apps. Popularity prediction in other social networks is the target of several studies. In Twitter, Ma et al. predict hashtag popularity to identify fast emerging topics attracting collective attention [@ma2013]. Their results reveal that context features (e.g., number of users that tweeted the hashtag) are relatively more effective than content feature (e.g., number of tweets with the hashtag). Tsur and Rappoport used a hybrid approach based on linear regressions to predict the spread of ideas in Twitter and found that a combination of content features with temporal and topological features minimizes prediction error [@tsur2012]. In YouTube, Szabo and Huberman found a strong linear correlation between the logarithmically transformed popularity of videos at early and later times. Based on this finding, they present a model to predict future popularity [@Szabo2010]. Pinto et al. propose two prediction models based on multivariate linear regression that incorporate information about historical patterns [@Pinto2013]. Finally, Roy et al. propose a framework called [SocialTransfer]{} that utilizes knowledge from social streams (e.g., Twitter) to discover sudden popularity bursts in videos. They show that social trends have a ripple effect as they spread from the Twitter domain to the video domain [@Roy2013]. To our knowledge, we are the first to target popularity prediction—measured by the number of stars—of software projects in the GitHub social coding network. Conclusion {#sec:conclusion} ========== In this paper, we use multiple linear regressions to predict the popularity of GitHub repositories. We found that general models, i.e., models produced using the top GitHub repositories, start to provide accurate predictions when they are trained with data from six months and used to predict the number of stars six months ahead (RQ \#1). We also found that specific models, i.e., models produced using repositories that share the same growth trend, can reduce the average prediction error and produce reliable predictions using less data. For the most common growth trend in our dataset, which includes almost half of the repositories, specific models improved significantly the accuracy of the predictions (RQ \#2). Finally, we report that prediction models tend to overestimate the repositories ranks. However, when newcomers are not considered, there is a very strong correlation between predicted and real rankings (RQ \#3). As future work, we plan to extend the specific prediction models to consider different programming languages. We plan to investigate different approaches to predict popularity, for example, epidemic models and machine learning models. We also plan to investigate predictions for other measures provided by GitHub, such as forks, watchers, and contributors. Acknowledgments =============== This research is supported by FAPEMIG and CNPq. [^1]: Because cluster C5 does not follow a linear trend it is not included in our analysis.	{ "pile_set_name": "ArXiv" }
CALT-68-2184\ hep-th/9807135 = 20pt John H. Schwarz\ California Institute of Technology, Pasadena, CA 91125, USA Abstract > In the strong coupling limit type IIA superstring theory develops an eleventh dimension that is not apparent in perturbation theory. This suggests the existence of a consistent 11d quantum theory, called M theory, which is approximated by 11d supergravity at low energies. In this review we describe some of the evidence for this picture and some of its implications. To be published in a special issue of Physics Reports in memory of Richard Slansky Introduction ============ Superstring theory is currently undergoing a period of rapid development in which important advances in understanding are being achieved. The purpose of this review is to describe a portion of this story to physicists who are not already experts in this field.[^1] The focus will be on explaining why there can be an eleven-dimensional vacuum, even though there are only ten dimensions in perturbative superstring theory. The nonperturbative extension of superstring theory that allows for an eleventh dimension has been named [M theory]{}. The letter M is intended to be flexible in its interpretation. It could stand for [magic,]{} [mystery,]{} or [meta]{} to reflect our current state of incomplete understanding. Those who think that two-dimensional supermembranes (the M2-brane) are fundamental may regard M as standing for [membrane.]{} An approach called [Matrix theory]{} is another possibility. And, of course, some view M theory as the [mother]{} of all theories. Superstring theory first achieved widespread acceptance during the [first superstring revolution]{} in 1984-85. There were three main developments at this time. The first was the discovery of an anomaly cancellation mechanism [@green84], which showed that supersymmetric gauge theories can be consistent in ten dimensions provided they are coupled to supergravity (as in type I superstring theory) and the gauge group is either SO(32) or $E_8 \times E_8$.[^2] Any other group necessarily would give uncancelled gauge anomalies and hence inconsistency at the quantum level. The second development was the discovery of two new superstring theories—called [heterotic]{} string theories—with precisely these gauge groups [@gross84]. The third development was the realization that the $E_8 \times E_8$ heterotic string theory admits solutions in which six of the space dimensions form a Calabi–Yau space, and that this results in a 4d effective theory at low energies with many qualitatively realistic features [@candelas85]. Unfortunately, there are very many Calabi–Yau spaces and a whole range of additional choices that can be made (orbifolds, Wilson loops, etc.). Thus there is an enormous variety of possibilities, none of which stands out as particularly special. In any case, after the first superstring revolution subsided, we had five distinct superstring theories with consistent weak coupling perturbation expansions, each in ten dimensions. Three of them, the [type I]{} theory and the two heterotic theories, have ${\cal N} = 1$ supersymmetry in the ten-dimensional sense. Since the minimal 10d spinor is simultaneously Majorana and Weyl, this corresponds to 16 conserved supercharges. The other two theories, called [type IIA]{} and [type IIB]{}, have ${\cal N} = 2$ supersymmetry (32 supercharges) [@green82]. In the IIA case the two spinors have opposite handedness so that the spectrum is left-right symmetric (nonchiral). In the IIB case the two spinors have the same handedness and the spectrum is chiral. The understanding of these five superstring theories was developed in the ensuing years. In each case it became clear, and was largely proved, that there are consistent perturbation expansions of on-shell scattering amplitudes. In four of the five cases (heterotic and type II) the fundamental strings are oriented and unbreakable. As a result, these theories have particularly simple perturbation expansions. Specifically, there is a unique Feynman diagram at each order of the loop expansion. The Feynman diagrams depict string world sheets, and therefore they are two-dimensional surfaces. For these four theories the unique $L$-loop diagram is a closed orientable genus-$L$ Riemann surface, which can be visualized as a sphere with $L$ handles. External (incoming or outgoing) particles are represented by $N$ points (or “punctures”) on the Riemann surface. A given diagram represents a well-defined integral of dimension $6L + 2N - 6$. This integral has no ultraviolet divergences, even though the spectrum contains states of arbitrarily high spin (including a massless graviton). From the viewpoint of point-particle contributions, string and supersymmetry properties are responsible for incredible cancellations. Type I superstrings are unoriented and breakable. As a result, the perturbation expansion is more complicated for this theory, and the various world-sheet diagrams at a given order (determined by the Euler number) have to be combined properly to cancel divergences and anomalies [@green85]. An important discovery that was made between the two superstring revolutions is called [T duality]{} [@giveon94]. This is a property of string theories that can be understood within the context of perturbation theory. (The discoveries associated with the [second superstring revolution]{} are mostly nonperturbative.) T duality shows that spacetime geometry, as probed by strings, has some surprising properties (sometimes referred to as [quantum geometry]{}). The basic idea can be illustrated by the simplest example. This entails considering one spatial dimension to form a circle (denoted $S^1$). Then the ten-dimensional geometry is $R^9 \times S^1$. T duality identifies this string compactification with one of a second string theory also on $R^9 \times S^1$. However, if the radii of the circles in the two cases are denoted $R_1$ and $R_2$, then $$R_1 R_2 = \alpha'. \label{Tdual}$$ Here $\alpha' = \ell_s^2$ is the universal Regge slope parameter, and $\ell_s$ is the fundamental string length scale (for both string theories). The tension of a fundamental string is given by $$T = 2\pi m_s^2 = \frac{1}{2\pi\alpha'},$$ where we have introduced a fundamental string mass scale $$m_s = (2\pi\ell_s)^{-1}.$$ Note that T duality implies that shrinking the circle to zero in one theory corresponds to decompactification of the dual theory. Compactification on a circle of radius $R$ implies that momenta in that direction are quantized, $p = n/R$. (These are called [Kaluza–Klein excitations]{}.) These momenta appear as masses for states that are massless from the higher-dimensional viewpoint. String theories also have a second class of excitations, called [winding modes]{}. Namely, a string wound $m$ times around the circle has energy $E = 2\pi R \cdot m \cdot T = mR/\alpha'$. Equation (\[Tdual\]) shows that the winding modes and Kaluza–Klein excitations are interchanged under T duality. What does T duality imply for our five superstring theories? The IIA and IIB theories are T dual [@ginsparg87]. So compactifying the nonchiral IIA theory on a circle of radius $R$ and letting $R \rightarrow 0$ gives the chiral IIB theory in ten dimensions! This means, in particular, that they should not be regarded as distinct theories. The radius $R$ is actually a $vev$ of a scalar field, which arises as an internal component of the 10d metric tensor. Thus the type IIA and type IIB theories in 10d are two limiting points in a continuous moduli space of quantum vacua. The two heterotic theories are also T dual, though there are technical details involving Wilson loops, which we will not explain here. T duality applied to the type I theory gives a dual description, which is sometimes called I’. The names IA and IB have also been introduced by some authors. For the remainder of this paper, we will restrict attention to theories with maximal supersymmetry (32 conserved supercharges). This is sufficient to describe the basic ideas of M theory. Of course, it suppresses many fascinating and important issues and discoveries. In this way we will keep the presentation from becoming too long or too technical. The main focus will be to ask what happens when we go beyond perturbation theory and allow the coupling strength to become large in the type II theories. The answer in the IIA case, as we will see, is that another spatial dimension appears. M Theory ======== In the 1970s and 1980s various supersymmetry and supergravity theories were constructed. (See [@salam], for example.) In particular, supersymmetry representation theory showed that ten is the largest spacetime dimension in which there can be a matter theory (with spins $\leq 1$) in which supersymmetry is realized linearly. A realization of this is 10d super Yang–Mills theory, which has 16 supercharges [@brink77]. This is a pretty ([i.e.]{}, very symmetrical) classical field theory, but at the quantum level it is both nonrenormalizable and anomalous for any nonabelian gauge group. However, as we indicated earlier, both problems can be overcome for suitable gauge groups (SO(32) or $E_8 \times E_8$) when the Yang–Mills theory is embedded in a type I or heterotic string theory. The largest possible spacetime dimension for a supergravity theory (with spins $\leq 2$), on the other hand, is eleven. Eleven-dimensional supergravity, which has 32 conserved supercharges, was constructed 20 years ago [@cremmer78a]. It has three kinds of fields—the graviton field (with 44 polarizations), the gravitino field (with 128 polarizations), and a three-index gauge field $C_{\mu\nu\rho}$ (with 84 polarizations). These massless particles are referred to collectively as the [supergraviton]{}. 11d supergravity is also a pretty classical field theory, which has attracted a lot of attention over the years. It is not chiral, and therefore not subject to anomaly problems.[^3] It is also nonrenormalizable, and thus it cannot be a fundamental theory. Though it is difficult to demonstrate explicitly that it is not finite as a result of “miraculous” cancellations, we now know that this is not the case. However, we now believe that it is a low-energy effective description of M theory, which is a well-defined quantum theory [@witten95a]. This means, in particular, that higher dimension terms in the effective action for the supergravity fields have uniquely determined coefficients within the M theory setting, even though they are formally infinite (and hence undetermined) within the supergravity context. Intriguing connections between type IIA string theory and 11d supergravity have been known for a long time. If one carries out [dimensional reduction]{} of 11d supergravity to 10d, one gets type IIA supergravity [@campbell84]. Dimensional reduction can be viewed as a compactification on circle in which one drops all the Kaluza–Klein excitations. It is easy to show that this does not break any of the supersymmetries. The field equations of 11d supergravity admit a solution that describes a supermembrane. In other words, this solution has the property that the energy density is concentrated on a two-dimensional surface. A 3d world-volume description of the dynamics of this supermembrane, quite analogous to the 2d world volume actions of superstrings, has been constructed [@bergshoeff87]. The authors suggested that a consistent 11d quantum theory might be defined in terms of this membrane, in analogy to string theories in ten dimensions.[^4] Another striking result was the discovery of double dimensional reduction [@duff87]. This is a dimensional reduction in which one compactifies on a circle, wraps one dimension of the membrane around the circle and drops all Kaluza–Klein excitations for both the spacetime theory and the world-volume theory. The remarkable fact is that this gives the (previously known) type IIA superstring world-volume action [@green84b]. For many years these facts remained unexplained curiosities until they were reconsidered by Townsend [@townsend95a] and by Witten [@witten95a]. The conclusion is that type IIA superstring theory really does have a circular 11th dimension in addition to the previously known ten spacetime dimensions. This fact was not recognized earlier because the appearance of the 11th dimension is a nonperturbative phenomenon, not visible in perturbation theory. To explain the relation between M theory and type IIA string theory, a good approach is to identify the parameters that characterize each of them and to explain how they are related. Eleven-dimensional supergravity (and hence M theory, too) has no dimensionless parameters. As we have seen, there are no massless scalar fields, whose vevs could give parameters. The only parameter is the 11d Newton constant, which raised to a suitable power ($-1/9$), gives the 11d Planck mass $m_p$. When M theory is compactified on a circle (so that the spacetime geometry is $R^{10} \times S^1$) another parameter is the radius $R$ of the circle. Now consider the parameters of type IIA superstring theory. They are the string mass scale $m_s$, introduced earlier, and the dimensionless string coupling constant $g_s$. An important fact about all five superstring theories is that the coupling constant is not an arbitrary parameter. Rather, it is a dynamically determined vev of a scalar field, the [dilaton,]{} which is a supersymmetry partner of the graviton. With the usual conventions, one has $g_s = \langle e^\phi\rangle$. We can identify compactified M theory with type IIA superstring theory by making the following correspondences: $$\label{M1} m_s^2 = 2\pi R m_p^3$$ $$\label{M2} g_s = 2\pi Rm_s.$$ Using these one can derive other equivalent relations, such as $$g_s = (2\pi Rm_p)^{3/2}$$ $$m_s = g_s^{1/3} m_p.$$ The latter implies that the 11d Planck length is shorter than the string length scale at weak coupling by a factor of $(g_s)^{1/3}$. Conventional string perturbation theory is an expansion in powers of $g_s$ at fixed $m_s$. Equation (\[M2\]) shows that this is equivalent to an expansion about $R=0$. In particular, the strong coupling limit of type IIA superstring theory corresponds to decompactification of the eleventh dimension, so in a sense M theory is type IIA string theory at infinite coupling.[^5] This explains why the eleventh dimension was not discovered in studies of string perturbation theory. These relations encode some interesting facts. The fact relevant to eq. (\[M1\]) concerns the interpretation of the fundamental type IIA string. Earlier we discussed the old notion of double dimensional reduction, which allowed one to derive the IIA superstring world-sheet action from the 11d supermembrane (or M2-brane) world-volume action. Now we can make a stronger statement: The fundamental IIA string actually [is]{} an M2-brane of M theory with one of its dimensions wrapped around the circular spatial dimension. No truncation to zero modes is required. Denoting the string and membrane tensions (energy per unit volume) by $T_{F1}$ and $T_{M2}$, one deduces that $$T_{F1} = 2\pi R \, T_{M2}.$$ However, $T_{F1} = 2\pi m_s^2$ and $T_{M2} = 2\pi m_p^3$. Combining these relations gives eq. (\[M1\]). Type II superstring theories contain a variety of $p$-brane solutions that preserve half of the 32 supersymmetries. These are solutions in which the energy is concentrated on a $p$-dimensional spatial hypersurface. (The world volume has $p+1$ dimensions.) The corresponding solutions of supergravity theories were constructed by Horowitz and Strominger [@horowitz91]. A large class of these $p$-brane excitations are called [D-branes]{} (or D$p$-branes when we want to specify the dimension), whose tensions are given by [@polchinski95] $$T_{Dp} = 2\pi {m_s^{p+1}}/{g_s}.$$ This dependence on the coupling constant is one of the characteristic features of a D-brane. It is to be contrasted with the more familiar $g^{-2}$ dependence of soliton masses (e.g., the ’t Hooft–Polyakov monopole). Another characteristic feature of D-branes is that they carry a charge that couples to a gauge field in the RR sector of the theory. (Such fields can be described as bispinors.) The particular RR gauge fields that occur imply that even values of $p$ occur in the IIA theory and odd values in the IIB theory. In particular, the D2-brane of the type IIA theory corresponds to our friend the supermembrane of M theory, but now in a background geometry in which one of the transverse dimensions is a circle. The tensions check, because (using eqs. (\[M1\]) and (\[M2\])) $$T_{D2} = 2\pi {m_s^3}/{g_s} = 2\pi m_p^3 = T_{M2}.$$ The mass of the first Kaluza–Klein excitation of the 11d supergraviton is $1/R$. Using eq. (\[M2\]), we see that this can be identified with the D0-brane. More identifications of this type arise when we consider the magnetic dual of the M theory supermembrane. This turns out to be a five-brane, called the M5-brane.[^6] Its tension is $T_{M5} = 2\pi m_p^6$. Wrapping one of its dimensions around the circle gives the D4-brane, with tension $$T_{D4} = 2\pi R \,T_{M5} = 2\pi m_s^5/g_s.$$ If, on the other hand, the M5-frame is not wrapped around the circle, one obtains the NS5-brane of the IIA theory with tension $$T_{NS5} = T_{M5} = 2\pi m_s^6/g_s^2.$$ This 5-brane, which is the magnetic dual of the fundamental IIA string, exhibits the conventional $g^{-2}$ solitonic dependence. To summarize, type IIA superstring theory is M theory compactified on a circle of radius $R=g_s \ell_s$. M theory is believed to be a well-defined quantum theory in 11d, which is approximated at low energy by 11d supergravity. Its excitations are the massless supergraviton, the M2-brane, and the M5-brane. These account both for the (perturbative) fundamental string of the IIA theory and for many of its nonperturbative excitations. The identities that we have presented here are exact, because they are protected by supersymmetry. Type IIB Superstring Theory =========================== In the previous section we discussed type IIA superstring theory and its relationship to eleven-dimensional M theory. In this section we consider type IIB superstring theory, which is the other maximally supersymmetric string theory with 32 conserved supercharges. It is also 10-dimensional, but unlike the IIA theory its two supercharges have the same handedness. Since the spectrum contains massless chiral fields, one should check whether there are anomalies that break the gauge invariances—general coordinate invariance, local Lorentz invariance, and local supersymmetry. In fact, the UV finiteness of the string theory Feynman diagrams (and associated [modular invariance]{}) ensures that all anomalies must cancel. This was verified also from a field theory viewpoint [@alvarez83]. The low-energy effective theory that approximates type IIB superstring theory is type IIB supergravity [@green82; @schwarz83], just as 11d supergravity approximates M theory. In each case the supergravity theory is only well-defined as a classical field theory, but still it can teach us a lot. For example, it can be used to construct $p$-brane solutions and compute their tensions. Even though such solutions themselves are only approximate, supersymmetry considerations ensure that their tensions, which are related to the kinds of charges they carry, are exact. Another significant fact about type IIB supergravity is that it possesses a global $SL(2,R)$ symmetry. It is instructive to consider the bosonic spectrum and its $SL(2,R)$ transformation properties. There are two scalar fields—the dilation $\phi$ and an [axion]{} $\chi$, which are conveniently combined in a complex field $$\rho = \chi + ie^{-\phi}.$$ The $SL(2,R)$ symmetry transforms this field nonlinearly: $$\rho \rightarrow \frac{a\rho + b}{c\rho + d},$$ where $a,b,c,d$ are real numbers satisfying $ad - bc = 1$. However, in the quantum string theory this symmetry is broken to the discrete subgroup $SL(2,Z)$ [@hull94], which means that $a,b,c,d$ are restricted to be integers. Defining the vev of the $\rho$ field to be $$\langle \rho \rangle = \frac{\theta}{2\pi} + \frac{i}{g_s},$$ the $SL(2,Z)$ symmetry transformation $\rho \rightarrow \rho + 1$ implies that $\theta$ is an angular coordinate. More significantly, in the special case $\theta = 0$, the symmetry transformation $\rho \rightarrow - 1/\rho$ takes $g_s \rightarrow 1/g_s$. This symmetry, called [S duality]{}, implies that the theory with coupling constant $g_s$ is equivalent to coupling constant $1/g_s$, so that the weak coupling expansion and the strong coupling expansion are identical! The bosonic spectrum also contains a pair of two-form potentials $B_{\mu\nu}^{(1)}$ and $B_{\mu\nu}^{(2)}$, which transform as a doublet under $SL(2,R)$ or $SL(2,Z)$. In particular, the S duality transformation $\rho \rightarrow - 1/\rho$ interchanges them. The remaining bosonic fields are the graviton and a four-form potential $C_{\mu\nu\rho\lambda}$, with a self-dual field strength. They are invariant under $SL(2)$. In the introductory section we indicated that the type IIA and type IIB superstring theories are T dual, meaning that if they are compactified on circles of radii $R_A$ and $R_B$ one obtains equivalent theories for the identification $R_AR_B = \ell_s^2$. Moreover, in sect. 2 we saw that the type IIA theory is actually M theory compactified on a circle. The latter fact encodes nonperturbative information. It turns out to be very useful to combine these two facts and to consider the duality between M theory compactified on a torus $(R^9 \times T^2)$ and type IIB superstring theory compactified on a circle $(R^9 \times S^1)$. Recall that a torus can be described as the complex plane modded out by the equivalence relations $z \sim z + w_1$ and $z \sim z + w_2$. Up to conformal equivalence, the periods can be taken to be $1$ and $\tau$, with Im $\tau > 0$. However, in this characterization $\tau$ and $\tau' = (a\tau + b)/(c\tau + d)$, where $a,b,c,d$ are integers satisfying $ad - bc = 1$, describe equivalent tori. Thus a torus is characterized by a modular parameter $\tau$ and an $SL(2,Z)$ modular group. The natural, and correct, conjecture at this point is that one should identify the modular parameter $\tau$ of the M theory torus with the parameter $\rho$ that characterizes the type IIB vacuum [@schwarz95a; @aspinwall95a]! Then the duality gives a geometrical explanation of the nonperturbative S duality symmetry of the IIB theory: the transformation $\rho \rightarrow - 1/\rho$, which sends $g_s \rightarrow 1/g_s$ in the IIB theory, corresponds to interchanging the two cycles of the torus in the M theory description. To complete the story, we should relate the area of the M theory torus $(A_M)$ to the radius of the IIB theory circle $(R_B)$. This is a simple consequence of formulas given above $$m_p^3 A_M = (2 \pi R_B)^{-1}.$$ Thus the limit $R_B \rightarrow 0$, at fixed $\rho$, corresponds to decompactification of the M theory torus, while preserving its shape. Conversely, the limit $A_M \rightarrow 0$ corresponds to decompactification of the IIB theory circle. The duality can be explored further by matching the various $p$-branes in 9 dimensions that can be obtained from either the M theory or the IIB theory viewpoints [@schwarz95b]. When this is done, one finds that everything matches nicely and that one deduces various relations among tensions, such as $$T_{M5} = \frac{1}{2\pi} (T_{M2})^2.$$ This relation was used earlier when we asserted that $T_{M2} = 2\pi m_p^3$ and $T_{M5} = 2\pi m_p^6$. Even more interesting is the fact that the IIB theory contains an infinite family of strings labelled by a pair of relatively prime integers $(p,q)$ [@schwarz95a]. These integers correspond to string charges that are sources of the gauge fields $B_{\mu\nu}^{(1)}$ and $B_{\mu\nu}^{(2)}$. The $(1,0)$ string can be identified as the fundamental IIB string, while the $(0,1)$ string is the D-string. From this viewpoint, a $(p,q)$ string can be regarded as a bound state of $p$ fundamental strings and $q$ D-strings [@witten95b]. These strings have a very simple interpretation in the dual M theory description. They correspond to an M2-brane with one of its cycles wrapped around a $(p,q)$ cycle of the torus. The minimal length of such a cycle is proportional to $\|p+q \tau\|$, and thus (using $\tau = \rho$) one finds that the tension of a $(p,q)$ string is given by $$T_{p,q} = 2\pi\|p + q\rho\| m_s^2. \label{pqtension}$$ The normalization has been chosen to give $T_{1,0} = 2\pi m_s^2$. Then (for $\theta = 0$) $T_{0,1} = 2\pi m_s^2/g_s$, as expected. Note that decay is kinematically forbidden by charge conservation when $p$ and $q$ are relatively prime. When they have a common division $n$, the tension is the same as that of an $n$-string system. Whether or not there are threshold bound states is a nontrivial dynamical question, which has different answers in different settings. In this case there are no such bound states, which is why $p$ and $q$ should be relatively prime. Imagine that you lived in the 9-dimensional world that is described equivalently as M theory compactified on a torus or as the type IIB superstring theory compactified on a circle. Suppose, moreover, you had very high energy accelerators with which you were going to determine the “true” dimension of spacetime. Would you conclude that 10 or 11 is the correct answer? If either $A_M$ or $R_B$ was very large in Planck units there would be a natural choice, of course. But how could you decide otherwise? The answer is that either viewpoint is equally valid. What determines which choice you make is which of the massless fields you regard as “internal” components of the metric tensor and which ones you regards as matter fields. Fields that are metric components in one description correspond to matter fields in the dual one. U Dualities =========== Maximal supergravity theories (ones with 32 conserved supercharges) typically have a noncompact global symmetry group $G$. For example, in the case of type IIB supergravity in ten dimensions the group is $SL(2,R)$. When one does dimensional reduction one finds larger groups in lower dimensions. For example, ${\cal N} = 8$ supergravity in four dimensions has a noncompact $E_7$ symmetry [@cremmer78b]. More generally, for $D = 11-d$, $3\leq d \leq 8$, one finds a maximally noncompact form of $E_d$, denoted $E_{d,d}$. These are statements about classical field theory. The corresponding statement about superstring theory/M theory is that if we toroidally compactify M theory on $R^D \times T^d$ or type IIB superstring theory on $R^D \times T^{d-1}$, the resulting moduli space of theories is invariant under an infinite discrete [U duality]{} group. The group, denoted $E_d (Z)$, is a maximal discrete subgroup of the noncompact $E_{d,d}$ symmetry group of the corresponding supergravity theory [@hull94]. An example that we will focus on below is $$E_3 (Z) = SL(3,Z) \times SL(2,Z).$$ The U duality groups are generated by the Weyl subgroup of $E_{d,d}$ plus discrete shifts of axion-like fields. The subgroup $SL(d,Z) \subset E_d (Z)$ can be understood as the geometric duality (modular group) of $T^d$ in the M theory picture. This generalizes the $SL(2,Z)$ discussed in the preceding section. The subgroup $SO (d-1, d-1; Z) \subset E_d (Z)$ is the T duality group of type IIB superstring theory compactified on $T^{d-1}$. These two subgroups intertwine nontrivially to generate the entire $E_d (Z)$ U duality group. Suppose we wish to focus on M theory and disregard type IIB superstring theory. Then we have a geometric understanding of the $SL(d,Z)$ subgroup of $E_d (Z)$ from considering M theory on $R^{11-d} \times T^d$. But what does the rest of $E_d(Z)$ imply? To address this question it will suffice to consider the first nontrivial case to which it applies, which is $d=3$. In this case the U duality group is $SL(3,Z) \times SL(2,Z)$. The first factor is geometric from the M theory viewpoint and nongeometric from the IIB viewpoint, whereas the second factor is geometric from the IIB viewpoint and nongeometric from the M theory viewpoint. So the question boils down to understanding the implication of the $SL(2,Z)$ duality in the M theory construction. Specifically, we want to understand the nontrivial $\tau \rightarrow - 1/\tau$ transformation. To keep the story as simple as possible, we will take the $T^3$ to be rectilinear with radii $R_1, R_2, R_3$ ([i.e.]{}, $g_{ij} \sim R_i^2 \delta_{ij}$) and assume that $C_{123} = 0$. Let us suppose that $R_3$ corresponds to the “eleventh” dimension that takes us to the IIA theory. Then we have IIA theory on a torus with radii $R_1$ and $R_2$. The nongeometric duality of M theory is T duality of IIA theory. T duality gives a mapping to an equivalent point in the moduli space for which $$R_i \rightarrow R'_i = \frac{\ell_s^2}{R_i} = \frac{\ell_p^3}{R_3 R_i}\quad i = 1,2,$$ with $\ell_s$ unchanged. Note that we have used eq. (\[M1\]), reexpressed as $\ell_p^3 = R_3 \ell_s^2$. Under a T duality the string coupling constant also transforms. The rule is that the coupling of the effective theory (8d in this case) is invariant: $$\frac{1}{g_8^2} = 4\pi^2 \frac{R_1 R_2}{g_s^2} = 4\pi^2 \frac{R'_1 R'_2}{(g'_s)^2}.$$ Thus $$g'_s = \frac{g_s \ell_s^2}{R_1 R_2}.$$ What does this imply for the radius of the eleventh dimension $R_3$? Using eq. (\[M2\]), $$R_3 = g_s\ell_s \rightarrow R'_3 = g'_s \ell_s.$$ Thus $$R'_3 = \frac{g_s\ell_s^3}{R_1R_2} = \frac{\ell_p^3}{R_1 R_2}.$$ However, the 11d Planck length also transforms, because $$\ell_p^3 = g_s \ell_s^3 \rightarrow (\ell'_p)^3 = g'_s \ell_s^3$$ implies that $$(\ell'_p)^3 = \frac{g_s\ell_s^5}{R_1 R_2} = \frac{\ell_p^6}{R_1R_2 R_3}.$$ The perturbative IIA description is only applicable for $R_3 \ll R_1, R_2$. However, even though T duality was originally discovered in perturbation theory, it is supposed to be an exact nonperturbative property. Therefore this duality mapping should be valid as an exact symmetry of M theory without any restriction on the radii. Another duality is an interchange of circles, such as $R_3 \leftrightarrow R_1$. This corresponds to the nonperturbative S duality of the IIB theory, as we discussed earlier. Combining these dualities we obtain the desired nongeometric duality of M theory on $T^3$ [@blau97]. It is given by $$\label{a} R_1 \rightarrow \frac{\ell_p^3}{R_2R_3},$$ and cyclic permutations, accompanied by $$\label{b} \ell_p^3 \rightarrow \frac{\ell_p^6}{R_1 R_2 R_3}.$$ Equations (\[a\]) and (\[b\]) have a nice interpretation. Equation (\[a\]) implies that $${1 \over R_1} \rightarrow (2\pi R_2) (2\pi R_3) T_{M2} .$$ Thus it interchanges Kaluza–Klein excitations with wrapped supermembrane excitations. It follows that these six 0-branes belong to the ([3]{}, [2]{}) representation of the U-duality group. Equation (\[b\]) implies that $$T_{M2} \rightarrow (2\pi R_1) (2\pi R_2) (2\pi R_3) T_{M5}.$$ Therefore it interchanges an unwrapped M2-brane with an M5-brane wrapped on the $T^3$. Thus these two 2-branes belong to the ([1]{}, [2]{}) representation of the U-duality group. This basic nongeometric duality of M theory, combined with the geometric ones, generates the entire U duality group in every dimension. It is a property of quantum M theory that goes beyond what can be understood from the effective 11d supergravity, which is geometrical. This analysis has been extended to allow $C_{123} \not= 0$ [@obers]. In this case there are indications that the torus should be considered to be [noncommutative]{} [@douglas97]. The D3-Brane and ${\cal N}=4$ Gauge Theory ========================================== D-branes have a number of special properties, which make them especially interesting. By definition, they are branes on which strings can end—D stands for [Dirichlet]{} boundary conditions. The end of a string carries a charge, and the D-brane world-volume theory contains a $U(1)$ gauge field that carries the associated flux. When $n$ D$p$-branes are coincident, or parallel and nearly coincident, the associated $(p + 1)$-dimensional world-volume theory is a $U(n)$ gauge theory. The $n^2$ gauge bosons $A_\mu^{ij}$ and their supersymmetry partners arise as the ground states of oriented strings running from the $i$th D$p$-brane to the $j$th D$p$-brane. The diagonal elements, belonging to the Cartan subalgebra, are massless. The field $A_\mu^{ij}$ with $i \not= j$ has a mass proportional to the separation of the $i$th and $j$th branes. This separation is described by the vev of a corresponding scalar field in the world-volume theory. The $U(n)$ gauge theory associated with a stack of $n$ D$p$-branes has maximal supersymmetry (16 supercharges). The low-energy effective theory, when the brane separations are small compared to the string scale, is supersymmetric Yang–Mills theory. These theories can be constructed by dimensional reduction of 10d supersymmetric $U(n)$ gauge theory to $p+1$ dimensions. In fact, that is how they originally were constructed [@brink77]. For $p\leq 3$, the low-energy effective theory is renormalizable and defines a consistent quantum theory. For $ p = 4,5$ there is good evidence for the existence nongravitational quantum theories that reduce to the gauge theory in the infrared. For $p\geq 6$, it appears that there is no decoupled nongravitational quantum theory [@sen97]. A case of particular interest, which we shall now focus on, is $p = 3$. A stack of $n$ D3-branes in type IIB superstring theory has a decoupled ${\cal N} = 4, $ $d = 4$ $U(n)$ gauge theory associated to it. This gauge theory has a number of special features. For one thing, due to boson–fermion cancellations, there are no $UV$ divergences at any order of perturbation theory. The beta function $\beta(g)$ is identically zero, which implies that the theory is scale invariant (aside from scales introduced by vevs of the scalar fields). In fact, ${\cal N}=4, $ $d=4$ gauge theories are conformally invariant. The conformal invariance combines with the supersymmetry to give a superconformal symmetry, which contains 32 fermionic generators. Half are the ordinary linearly realized supersymmetrics, and half are nonlinearly realized ones associated to the conformal symmetry. The name of the superconformal group in this case is $SU(4\|4)$. Another important property of ${\cal N}=4$, $ d=4$ gauge theories is electric-magnetic duality [@montonen77]. This extends to an $SL(2,Z)$ group of dualities. To understand these it is necessary to include a vacuum angle $\theta_{YM}$ and define a complex coupling $$\tau = \frac{\theta_{YM}}{2\pi} + i \frac{4\pi}{g_{YM}^2}.$$ Under $SL(2,Z)$ transformations this coupling transforms in the usual nonlinear fashion $\left(\tau \rightarrow \frac{a\tau+b}{c\tau+d}\right)$ and the electric and magnetic fields transform as a doublet. Note that the conformal invariance ensures that $\tau$ is a meaningful scale-independent constant. Now consider the ${\cal N}=4 $ $U(n)$ gauge theory associated to a stack of $n$ D3-branes in type IIB superstring theory. There is an obvious identification, that turns out to be correct. Namely, the $SL(2,Z)$ duality of the gauge theory is induced from that of the ambient type IIB superstring theory. In particular, the $\tau$ parameter of the gauge theory is the value of the complex scalar field $\rho$ of the string theory. This makes sense because $\rho$ is constant in the field configuration associated to a stack of D3-branes. The D3-branes themselves are invariant under $SL(2,Z)$ transformations. Only the parameter $\tau = \rho$ changes, but it is transformed to an equivalent value. All other fields, such as $B_{\mu\nu}^{(i)}$, which are not invariant, vanish in this case. As we have said, a fundamental $(1,0)$ string can end on a D3-brane. But by applying a suitable $SL(2,Z)$ transformation, this configuration is transformed to one in which a $(p,q)$ string—with $p$ and $q$ relatively prime—ends on the D3-brane. The charge on the end of this string describes a dyon with electric charge $p$ and magnetic $q$, with respect to the appropriate gauge field. More generally, for a stack of $n$ D3-branes, any pair can be connected by a $(p,q)$ string. The mass is proportional to the length of the string times its tension, which we saw is proportional to $\|p + q\rho\|$. In this way one sees that the electrically charged particles, described by fundamental fields, belong to infinite $SL(2,Z)$ multiplets. The other states are nonperturbative excitations of the gauge theory. The field configurations that describe them preserve half of the supersymmetry. As a result their masses saturate a BPS bound and are given exactly by the considerations described above. An interesting question, whose answer was unknown until recently, is whether ${\cal N}=4 $ gauge theories in four dimensions also admit nonperturbative excitations that preserve 1/4 of the supersymmetry. To explain the answer, it is necessary to first make a digression to consider three-string junctions. As we have seen, type IIB superstring theory contains an infinite multiplet of strings labelled by a pair of relatively prime integers $(p,q)$. Three strings, with charges $(p_i, q_i), $ $i = 1,2,3,$ can meet at a point provided that charge is conserved [@aharony96; @schwarz96]. This means that $$\label{x} \sum p_i = \sum q_i = 0,$$ if the three strings are all oriented inwards. (This is like momentum conservation in an ordinary Feynman diagram.) Such a configuration is stable, and preserves 1/4 of the ambient supersymmetry provided that the tensions balance. It is easy to see how this can be achieved. If one regards the plane of the junction as a complex plane and orients the direction of a $(p,q)$ string by the phase of $p + q\tau$, then eqs. (\[pqtension\]) and (\[x\]) ensure a force balance. The three-string junction has an interesting dual M theory interpretation. If one of the directions perpendicular to the plane of the junction is taken to be a circle, then we have a string junction in nine dimensions. This must have a dual interpretation in terms of M theory compactified on a torus. We have already seen that a $(p,q)$ string corresponds to an M2-brane with one of its cycles wrapped on a $(p,q)$ cycle of the torus. So now we join three such cylindrical membranes together. Altogether we have a single smooth M2-brane forming a $Y$, like a junction of pipes. The three arms are wrapped on $(p_i, q_i)$ cycles of the torus. This is only possible topologically when eq. (\[x\]) is satisfied. We can now describe a pretty construction of 1/4 BPS states in ${\cal N}=4$ gauge theory, due to Bergman [@bergman97]. Such a state is described by a 3-string junction, with the three prongs terminating on three different D3-branes. This is only possible for $n \geq 3$, which is a necessary condition for 1/4 BPS states. The mass of such a state is given by summing the lengths of each string segment weighted by its tension. This gives a result in agreement with the BPS formula. Clearly this is just the beginning of a long story, since the simple picture we have described can be generalized to arbitrarily complicated string webs. So long as the web is in a plane, charges are conserved at the junctions, and all string segments are oriented in the way we have described, the configuration will be 1/4 BPS. Remarkably, arbitrarily high spins can occur. There are simple rules for determining them [@bergman98]. When the web is nonplanar, supersymmetry is completely broken, and reliable mass calculations become difficult. However, one should still be able to achieve a reliable qualitative understanding of such excitations. In general, there are regions of moduli space in which such nonsupersymmetric states are stable. Conclusion ========== In this brief review we have described some of the interesting advances in understanding superstring theory that have taken place in the past few years. Many others, such as studies of black hole entropy, have not even been mentioned. The emphasis has been on the nonperturbative appearance of an eleventh dimension in type IIA superstring theory, as well as its implications when combined with superstring T dualities. In particular, we argued that there should be a consistent quantum vacuum, whose low-energy effective description is given by 11d supergravity. The relevant quantum theory – called M theory – has important features, such as the nongeometric U duality described in section 4, that go beyond what can be understood within ordinary (nonrenormalizable) 11d supergravity. What we have described makes a convincing self-consistent picture, but it does not constitute a complete formulation of M theory. In the past two years there have been some major advances in that direction, which we will briefly mention here. The first, which goes by the name of [Matrix Theory]{} [@banks96], bases a formulation of M theory in flat 11d spacetime in terms of the supersymmetric quantum mechanics of N D0-branes in the large N limit. This proposal has been generalized to include an interpretation for finite N. In that case Susskind has proposed an identification with [discrete light-cone quantization]{} of M theory, in which there are N units of momentum along a null compact direction [@susskind97]. Both versions of Matrix Theory have passed all tests that have been carried out, some of which are very nontrivial. At times there appeared to be discrepancies, but these were all the result of subtle errors that have now been tracked down. The construction has a nice generalization to describe compactification of M theory on a torus $T^n$ [@taylor96]. However, it does not seem to be useful for $n > 5$ [@sen97], and other compactification manifolds are (at best) awkward to handle. Another shortcoming of this approach is that it treats the eleventh dimension differently from the other ones. Another proposal relating superstring and M theory backgrounds to large N limits of certain field theories has been put forward recently by Maldacena [@maldacena97] and made more precise by others [@gubser98]. In this approach, there is a conjectured duality ([i.e.]{}, equivalence) between a conformally invariant field theory (CFT) in $n$ dimensions and type IIB superstring theory or M theory on an Anti-de-Sitter space (AdS) in $n+1$ dimensions. The remaining $9-n$ or $10-n$ dimensions form a compact space, the simplest cases being spheres. The three examples with unbroken supersymmetry are $AdS_5 \times S^5$, $AdS_4 \times S^7$, and $AdS_7 \times S^4$. This approach is sometimes referred to as [AdS/CFT duality]{}. This is an extremely active and very promising subject. It has already taught us a great deal about the large N behavior of various gauge theories. As usual, the easiest theories to study are ones with a lot of supersymmetry, but it appears that in this approach supersymmetry breaking is more accessible than in previous ones. For example, it might someday be possible to construct the QCD string in terms of a dual AdS gravity theory, and use it to carry out numerical calculations of the hadron spectrum. Indeed, there have already been some preliminary steps in this direction [@ooguri98]. Despite all of the successes that have been achieved in advancing our understanding of superstring theory and M theory, there clearly is still a long way to go. In particular, despite much effort and several imaginative proposals, we still do not have a convincing mechanism for ensuring the vanishing (or extreme smallness) of the cosmological constant for nonsupersymmetric vacua. Superstring theory is a field with very ambitious goals. The remarkable fact is that they still seem to be realistic. However, it may take a few more revolutions before they are attained. [999]{} A. Sen, hep-th/9802051. M.B. Green and J.H. Schwarz, [Phys. Lett.]{} [149B]{} (1984) 117. D.J. Gross, J.A. Harvey, E. Martinec, and R. Rohm, [Phys. Rev. Lett.]{} [54]{} (1985) 502. P. Candelas, G.T. Horowitz, A. Strominger, and E. Witten, [Nucl. Phys.]{} [B258]{} (1985) 46. M.B. Green and J.H. Schwarz, [Phys. Lett.]{} [109B]{} (1982) 444. M.B. Green and J.H. Schwarz, [Phys. Lett.]{} [151B]{} (1985) 21; [Nucl. Phys.]{} [B255]{} (1985) 93. For a review see A. Giveon, M. Porrati, and E. Rabinovici, [Phys. Rept.]{} [244]{} (1994) 77, hep-th/9401139. P. Ginsparg, [Phys. Rev.]{} [D35]{} (1987) 648; M. Dine, P. Huet, and N. Seiberg, [Nucl. Phys.]{} [B322]{} (1989) 301. A. Salam and E. Sezgin, eds., [Supergravities in Diverse Dimensions]{}, reprints in 2 vols., World Scientific (1989). L. Brink, J.H. Schwarz, and J. Scherk, [Nucl. Phys.]{} [B121]{} (1977) 77; F. Gliozzi, J. Scherk, and D. Olive, [Nucl. Phys.]{} [B122]{} (1977) 253. E. Cremmer, B. Julia, and J. Scherk, [Phys. Lett.]{} [76B]{} (1978) 409. P. Hořava and E. Witten, [Nucl. Phys.]{} [B460]{} (1996) 506, hep-th/9510209. E. Witten, [Nucl. Phys.]{} [B443]{} (1995) 85, hep-th/9503124. C. Campbell and P. West, [Nucl. Phys.]{} [B243]{} (1984) 112; M. Huq and M. Namazie, [Class. Quantum Grav.]{} [2]{} (1985) 293; F. Giani and M. Pernici, [Phys. Rev.]{} [D30]{} (1984) 325. E. Bergshoeff, E. Sezgin, and P.K. Townsend, [Phys. Lett.]{} [B189]{} (1987) 75. M.J. Duff, P.S. Howe, T. Inami, and K.S. Stelle, [Phys. Lett.]{} [B191]{} (1987) 70. M.B. Green and J.H. Schwarz, [Phys. Lett.]{} [136B]{} (1984) 367. P.K. Townsend, [Phys. Lett.]{} [B350]{} (1995) 184, hep-th/9501068. G.T. Horowitz and A. Strominger, [Nucl. Phys.]{} [B360]{} (1991) 197. J. Polchinski, [Phys. Rev. Lett.]{} [75]{} (1995) 4724, hep-th/9510017; p. 293 in [Fields, Strings, and Duality]{} (TASI 96), eds. C. Efthimiou and B. Greene, World Scientific 1997, hep-th/9611050. L. Alvarez-Gaumé and E. Witten, [Nucl. Phys.]{} [B234]{} (1983) 269. J.H. Schwarz and P.C. West, [Phys. Lett.]{} [126B]{} (1983) 301; J.H. Schwarz, [Nucl. Phys.]{} [B226]{} (1983) 269; P. Howe and P. West, [Nucl. Phys.]{} [B238]{} (1984) 181. C. Hull and P. Townsend, [Nucl. Phys.]{} [B438]{} (1995) 109, hep-th/9410167. J.H. Schwarz, [Phys. Lett.]{} [B360]{} (1995) 13, Erratum: [Phys. Lett.]{} [B364]{} (1995) 252, hep-th/9508143. P.S. Aspinwall, [Nucl. Phys. Proc. Suppl.]{} [46]{} (1996) 30, hep-th/9508154. J.H. Schwarz, [Phys. Lett.]{} [B367]{} (1996) 97, hep-th/9510086. E. Witten, [Nucl. Phys.]{} [B460]{} (1996) 335, hep-th/9510135. E. Cremmer and B. Julia, [Phys. Lett.]{} [80B]{} (1978) 48; [Nucl. Phys.]{} [B159]{} (1979) 141. S. Elitzur, A. Giveon, D. Kutasov, and E. Rabinovici, [Nucl. Phys.]{} [B509]{} (1998) 122, hep-th/9707217; M. Blau and M. O’Loughlin, hep-th/9712047. N.A. Obers, B. Pioline, and E. Rabinovici, hep-th/9712084. A. Connes, M. Douglas, and A. Schwarz, [J. High Energy Phys.]{} [02]{} (1998) 003, hep-th/9711162; M. Douglas and C. Hull, [J. High Energy Phys.]{} [02]{} (1998) 008, hep-th/9711165. A. Sen, hep-th/9709220; N. Seiberg, [Phys. Rev. Lett.]{} [79]{} (1997) 3577, hep-th/9710009. C. Montonen and D. Olive, [Phys. Lett.]{} [B72]{} (1977) 117; P. Goddard, J. Nuyts, and D. Olive, [Nucl. Phys.]{} [B125]{} (1977) 1; H. Osborne, [Phys. Lett.]{} [B83]{} (1979) 321. O. Aharony, J. Sonnenschein, and S. Yankielowicz, [Nucl. Phys.]{} [B474]{} (1996) 309, hep-th/9603009. J.H. Schwarz, p. 359 in [Fields, Strings, and Duality]{} (TASI 96), eds. C. Efthimiou and B. Greene, World Scientific 1997, hep-th/9607201. O. Bergman, hep-th/9712211. O. Bergman and B. Kol, hep-th/9804160. T. Banks, W. Fischler, S. Shenker, and L. Susskind, [Phys. Rev.]{} [D55]{} (1997) 112, hep-th/9610043; For a review see T. Banks, hep-th/9710231. L. Susskind, hep-th/9704080. W. Taylor, [Phys. Lett.]{} [B394]{} (1997) 283, hep-th/9611042. J.M. Maldacena, hep-th/9711200. S.S. Gubser, I.R. Klebanov, and A.M. Polyakov, [Phys. Lett.]{} [B428]{} (1998) 105, hep-th/9802109; E. Witten, hep-th/9802150. C. Csaki, H. Ooguri, Y. Oz, and J. Terning, hep-th/9806021; R. de Mello Koch, A. Jevicki, M. Mihailescu, J.P. Nunes, hep-th/9806125. [^1]: For a more detailed review see ref. [@sen98]. [^2]: A discussion with Richard Slansky helped to convince us that $E_8 \times E_8$ would work. [^3]: Unless the spacetime has boundaries. The anomaly associated to a 10d boundary can be cancelled by introducing $E_8$ supersymmetric gauge theory on the boundary [@horava95]. [^4]: Most experts now believe that M theory cannot be defined as a supermembrane theory. [^5]: The $E_8 \times E_8$ heterotic string theory is also eleven-dimensional at strong coupling [@horava95]. [^6]: In general, the magnetic dual of a $p$-brane in $d$ dimensions is a $(d - p - 4)$-brane.	{ "pile_set_name": "ArXiv" }
--- abstract: \| By proposing device-independent protocols, S. Pironio et al. \[Nature 464, 1021-1024 (2010)\] and R. Colbeck et al. \[Nature Physics 8, 450-453 (2012)\] proved that new randomness can be generated by using perfectly free random sources or partially free ones as seed. Subsequently, Li et al. \[Phys. Rev. A 84, 034301 (2011)\] studied this topic in the framework of semi-device-independent and proved that new randomness can be obtained from perfectly free random sources. Here we discuss whether and how partially free random sources bring us new randomness in semi-device-independent scenario. We propose a semi-device-independent randomness expansion protocol with partially free random sources, and obtain the condition that the partially free random sources should satisfy to generate new randomness. In the process of analysis, we acquire a new 2-dimensional quantum witness. Furthermore, we get the analytic relationship between the generated randomness and the 2-dimensional quantum witness violation. PACS numbers : 03.67.Ac, 05.40.-a author: - 'Yu-Qian Zhou$^{1}$' - 'Hong-Wei Li$^{2}$' - 'Yu-Kun Wang$^{1}$' - 'Dan-Dan Li$^{1}$' - 'Fei Gao$^{1}$' - 'Qiao-Yan Wen$^{1}$' bibliography: - 'apssamp.bib' nocite: '[@]' title: 'Semi-device-independent randomness expansion with partially free random sources' --- \[sec:level1\]Introduction ========================== Perfectly free random bits have both theoretical and practical significance. In the aspect of theory, perfectly free random bits are beneficial for the foundation of physical theory to establish symmetries \[1\]. In practical applications, perfectly free random bits could be used in many important fields, especially in cryptography. Almost all the security of cryptographic protocols depends on perfectly free random bits. For example, in the well known BB84 protocol \[2\], the security will be seriously limited once an eavesdropper uses partially free random bits to replace the perfectly free ones \[3\]. Recently, the studies of device-independent (DI) and semi-device-independent (SDI) protocols have attracted a lot of attention. Here, DI means that no assumption is made on the devices used to perform protocols \[4\]. Subsequently, M. Paw[ł]{}owski introduced the concept of SDI meaning that the devices in protocols are noncharacterized except the tight bound of the dimension of the potential required systems \[5\]. Randomness expansion is the protocol in which random sources are used as seed to produce new randomness. Recently, R. Colbeck proposed a DI randomness expansion protocol based on the tripartite GHZ-type entangled states \[6\] and S. Pironio et al.* proposed the protocol based on Bell inequality violation \[7\]. These results demonstrated that perfectly free sources can be expanded in the framework of DI. In 2012, R. Colbeck et al. showed that new randomness can also be obtained by using partially free bits as seed in the framework of DI (more precisely, the partially free bits can be amplified to make perfectly free ones and this process also is a DI randomness amplification protocol) \[1\]. Subsequently, Li et al. studied this interesting topic in the framework of SDI and proved that new randomness can be produced from perfectly free sources by presenting SDI randomness expansion protocols \[8, 9\]. Therefore, whether and how partially free sources bring us new randomness in the framework of SDI is a problem about which people may be curious. Here, we demonstrate that new randomness can be generated from partially free sources in the SDI scenario by proposing a SDI randomness expansion protocol with partially free sources. Different from the assumption that $\varepsilon_{1}=\varepsilon_{2}$ in the Ref. \[1\], we consider a more general case,where $\varepsilon_{1}=\varepsilon_{2}$ is not strictly required in our protocol, and obtain the condition that $\varepsilon_{1}, \varepsilon_{2}$ should fulfill to generate new randomness (the choices of states and measurements are derived from $\varepsilon_{1}$-$free$ source and $\varepsilon_{2}$-$free$ source, respectively). A new 2-dimensional quantum witness is gained in the process of randomness certification. Furthermore, the analytic relationship between the generated randomness and the 2-dimensional quantum witness violation is acquired. This paper is structured as follows. In Sec. , we recall the definition of partially free random sources and introduce a SDI randomness expansion protocol with partially free sources. In Sec. , the condition which partially free sources should satisfy to generate new randomness, and certification parameters are obtained. In Sec. , the analytic relationship between the generated randomness and 2-dimensional quantum witness violation is concluded. In Sec. , we summarize our results. \[sec:level1\]model description =============================== In order to better explain our theory, first of all, we give a detailed definition of partially free random sources in this section. Let $X$ be a variable, considering its causal structure in relativistic space time, we call a variable $\lambda$ cannot be caused by $X$ if $\lambda$ are not in the future lightcone of X. Denote the parameter $ \Lambda$ as the set of variables which cannot be caused by $X$ and are interested in our devices. The variables in $ \Lambda$ may be provided by an eavesdropper or a higher theory \[1\]. A variable bit $X$ is called $\varepsilon$-$free$ bit, $\varepsilon<\frac{1}{2}$, if it satisfies $\|P(0\|\Lambda=\lambda)-\frac{1}{2}\|\leq\varepsilon$ for all $\lambda\in \Lambda$. Particularly, $X$ is called perfectly free bit as $\varepsilon=0$. In this paper, we say that bits are picked according to $\varepsilon$-$free$ source if each bit is $\varepsilon$-$free$ and independents of other bits. Secondly, we introduce a SDI randomness expansion protocol with partially free sources based on $n\rightarrow1$ quantum random access codes (QRACs) (see Fig 1). Based on the typical causal structure of our protocol \[1\], we assume that $\lambda$ may be correlated with two sources of weak randomness, the states prepared by Alice and the measurements performed by Bob. A detailed description of our scenario is described as follows: Alice picks $n$ bits $a=a_{0}a_{1}\cdot\cdot\cdot a_{n-1}$ according to $\varepsilon_{1}$-$free$ source $S_{1}$ and encoded to 1 qubit $\rho_{a,\lambda}$, then Alice sends it to Bob via quantum channel. Bob performs two dimensional measurement $\{M_{y,\lambda}^{b},b=0,1\}$ decided by $y=0,1,\cdot\cdot\cdot, n-1$ which is picked according to $\varepsilon_{2}$-$free$ source $S_{2}$, and emits the measurement outcome $b$. In particular, there is not entanglement in the devices. ![SDI randomness expansion with partially free sources. The dashed line represent that the hidden variable $\lambda$ may be correlated with these parts. Our protocol consists of two black box, which do not contain entanglement, in safe area.](6.eps){width="50.00000%"} In this paper, we construct the 2-dimensional quantum witness using the expected success probability which is different from the Ref. \[11\] and draws better conclusions. The expected success probability for the scenario is $$E\equiv\sum_{a,y}P(a,y)P(b=a_{y}\|a,y)=\sum_{\lambda}P(\lambda)E_{\lambda},$$ where $E_{\lambda}=\sum_{a,y}P(a,y\|\lambda)P(b=a_{y}\|a,y,\lambda)$ and $P(b\|a,y,\lambda)=$tr$(\rho_{a,\lambda}M_{y,\lambda}^{b})$. Probability distribution of $P(a,y,b)$ can be estimate by repeating the procedure many times, the value of $E$ can then be estimated. Thirdly, we introduce the definition of the min-entropy function: $$H_{\infty}(B\|A,Y,\Lambda)\equiv -\log_{2}\max_{a,y,b,\lambda} \sum_{\lambda\in\Lambda}P(\lambda)P(b\|a,y,\lambda)$$ to quantify the randomness of the measurement outcome for the scenario with the set $\Lambda$. Here the SDI randomness expansion with partially free sources based on $2\rightarrow 1$ QRAC is primarily discussed. The feasible region and the randomness certification of our protocol are explored in next section. Feasible Region and randomness certification ============================================ In the DI randomness amplification proposed by R. Colbeck et al. \[1\], only one case of $\varepsilon_{1}=\varepsilon_{2}$ is discussed, and the relationship between quantum dimension witness and the min-entropy bound cannot be given as there are infinite parameters needed to be considered. In this section, we relax the assumption of $\varepsilon_{1}=\varepsilon_{2}$. Namely, the random resources of Alice is actually not required to be same as Bob’s, and obtain the feasible region. The good partially free sources are quite precious resource, our setting benefits to allocate partially free sources more reasonably and effectively. On the other aspect, the figure of the relationship between 2-dimensional quantum witness violation and the min-entropy bound will be obtained through an optimization process. If there exists a protocol about SDI randomness expansion with partially free sources where Alice and Bob have the $\varepsilon_{i}$-$free$ source $S_{i},i=0,1$, respectively, and new randomness is certified, the pair $(\varepsilon_{1},\varepsilon_{2})$ is called a feasible pair. The Feasible Region $R$ of SDI randomness expansion with partially free sources is the set of all feasible pairs $(\varepsilon_{1},\varepsilon_{2})$. It is evident for any $\lambda\in\Lambda$ that the randomness extracted from the outcome $b$ will reduce to 0 with the increase of the distance between the probability distribution of $P(a,y\|\lambda)$ and the uniform distribution on $a,y$. We assume that the eavesdropper attacks our devices in order to make our protocol get the least randomness and attempt not to led us finding that the random sources has been changed. To achieve his targets, the eavesdropper has to let $$\begin{aligned} \|P(a_{i}=0\|\lambda)-\frac{1}{2}\|&=&\varepsilon_{1},i=0,1.\nonumber\\ \|P(y=0\|\lambda)-\frac{1}{2}\|&=&\varepsilon_{2},\end{aligned}$$ for any $\lambda\in\Lambda$ and $$P(a,y)=\sum_{\lambda}P(\lambda)P(a,y\|\lambda)=\frac{1}{8}$$ for any $a\in\{00,01,10,11\}, y\in\{0,1\}.$ Without loss of generality, we can assume that there are only 8 hidden variables $\lambda_{k}, k=0,1,\cdot\cdot\cdot,7$ corresponding to 8 cases in Eq. (3). For the sake of convenience, let $$\begin{aligned} P(a_{i}=0\|\lambda_{k})&=&\frac{1}{2}+(-1)^{k_{i}}\varepsilon_{1},i=0,1.\nonumber\\ P(y=0\|\lambda)&=&\mbox{}\frac{1}{2}+(-1)^{k_{2}}\varepsilon_{2}.\end{aligned}$$ where $k_{0}k_{1}k_{2}$ is the binary notation of $k$. It is easy to see that the eavesdropper can achieve Eq. (3) and Eq. (4) at the same time, see appendix A for the proof. Under the above attack of the eavesdropper, if for a pair $(\varepsilon_{1},\varepsilon_{2})$, 2-dimensional quantum witness violation still exist and the min-entropy is larger than $0$ as 2-dimensional quantum witness violation reach its maximum, then we can say that the pair $(\varepsilon_{1},\varepsilon_{2})$ belongs to the feasible region $R$. Denote $E_{\lambda_{k}, c}$ as the expected success probability with parameter $\lambda_{k}$ through a classical process. For any $k$, the maximum value of $E_{\lambda_{k}, c}$ can be obtained using the encoding map $a_{0}a_{1}\rightarrow a_{k_{2}}$, the bit $a_{k_{2}}$ decoding map $0\rightarrow0, 1\rightarrow1$ and the bit $a_{(1-k_{2})}$ decoding map $0,1\rightarrow k_{(1-k_{2})}$, and $E_{\lambda_{k}, c}$ reach the same maximum value for any $k$. Obviously, the maximum value of $E$ through a classical process is $$\begin{aligned} E_{c}=\frac{3}{4}+\frac{1}{2}(\varepsilon_{1}+\varepsilon_{2})-\varepsilon_{1}\varepsilon_{2}.\end{aligned}$$ Denote $E_{\lambda_{k}, q}$ as the expected success probability with parameter $\lambda_{k}$ through a quantum process, $E_{\lambda_{k}}$ apparently meet the linear relationship. For any $k$, to reach the maximum value of $E_{\lambda_{k}, q}$, as a general rule, every quantum state $\rho_{a,\lambda_{k}}$ and positive operator valued measure (POVM) $\{M_{y,\lambda}^{0},M_{y,\lambda}^{1}\}$ performed in the 2-dimensional space should be considered. Here, each mix state can be written as a convex combination of pure states. On the other hand, any POVM can be described as a convex combination of projective measurements, which include projective measurements with rank 1, measurements $ \{I, 0\}$ and $\{0, I\}$ \[12\]. Different from the Ref. \[8, 9\], we pinpoint that measurements $ \{I, 0\}$ and $\{0, I\}$ also need to be considered in our protocol. Nevertheless, we can prove that once measurements $ \{I, 0\}$ or $\{0, I\}$ are chosen, the relationship $E_{\lambda_{k}, q}\leq E_{c}$ will be satisfied and is tight. In conclusion, here only pure states and projective measurements with rank 1 need to considered. To visualize the pure states and projective measurements with rank 1, we consider Bloch sphere representation. Without loss of generality, set the Bloch sphere representation of measurement $\{M_{y,\lambda_{k}}^{0},M_{y,\lambda_{k}}^{1}\}$ as $\{\textbf{\emph{v}}_{y,\lambda_{k}},-\textbf{\emph{v}}_{y,\lambda_{k}}\}$ and $\textbf{\emph{v}}_{0,\lambda_{k}}=(1,0,0)$ for any $k$. Let the Bloch vector $\textbf{\emph{r}}_{a,\lambda_{k}}$ be the Bloch sphere representation of the pure state $\rho_{a,\lambda_{k}}$. For any pair $(\varepsilon_{1},\varepsilon_{2})$, define $$t \equiv\frac{8\varepsilon_{1}^{2}(1+4\varepsilon_{2}^{2})}{1+16\varepsilon_{1}^{4}-4\varepsilon_{2}^{2}-64\varepsilon_{1}^{4}\varepsilon_{2}^{2}}\geq0.$$ $$\textbf{\emph{v}}_{a,\lambda_{k}}\equiv\sum_{i=0,1}(-1)^{a_{i}}(\frac{1}{2}+(-1)^{k_{2}}\varepsilon_{2})\textbf{\emph{v}}_{i,\lambda_{k}}.$$ If $t>1$, the optimal encoding-decoding strategy for any $k$: $\textbf{\emph{v}}_{1,\lambda_{k}}=(1, 0, 0)$ and $\textbf{\emph{r}}_{a,\lambda_{k}}=\textbf{\emph{v}}_{a,\lambda_{k}}/\\|\textbf{\emph{v}}_{a,\lambda_{k}}\\|$. Hence the maximum value of $E$ is $$E=\frac{3}{4}+\frac{1}{2}\varepsilon_{2}+\varepsilon_{1}^{2}(1-2\varepsilon_{2})\leq E_{c}.$$ If $t\leq1$, the optimal encoding-decoding strategy for any $k$: $\textbf{\emph{v}}_{1,\lambda_{k}}=((-1)^{k_{0}+k_{1}}t,\sqrt{1-t^{2}}, 0)$ and $\textbf{\emph{r}}_{a,\lambda_{k}}=\textbf{\emph{v}}_{a,\lambda_{k}}/\\|\textbf{\emph{v}}_{a,\lambda_{k}}\\|$. After the simple analysis and calculation, $E_{\lambda_{k},q}$ will reach the same maximum value and the maximum value of $E$ through a quantum process is $$\begin{aligned} E_{q}=\frac{1}{2}+\frac{1}{2}\sqrt{\frac{1}{2}+8\varepsilon_{1}^{4}+2\varepsilon_{2}^{2}+32\varepsilon_{1}^{4}\varepsilon_{2}^{2}}.\end{aligned}$$ For $E=E_{q}$, the min-entropy is $$\begin{aligned} H_{\infty}(B\|A,Y,\Lambda)=1-\log_{2}(1+\frac{t+\delta}{\sqrt{\delta^{2}+2t\delta+1}}),\end{aligned}$$ where $\delta=(1+2\varepsilon_{2})/(1-2\varepsilon_{2}).$ This implies that a pair $(\varepsilon_{1},\varepsilon_{2})$ belongs to the feasible region $R$ once it satisfies $t\leq1$ , $ E_{ c}<E_{q}$ and $H_{\infty}(B\|A,Y,\Lambda)>0$. Then the feasible region is obtained and demonstrated in Fig 2. Moreover, a new tight bound for 2-dimensional classical and quantum systems are given as $E_{c}$ and $E_{q}$, respectively. Namely, a new 2-dimensional quantum witness is presented. See appendix B for a detailed calculation of Eq. (9)-(11). ![The feasible region $R$ of SDI randomness expansion with partially free random sources is the yellow region under the green line but does not include the green line. Alice pick $a$ according to $\varepsilon_{1}$-$free$ source $S_{1}$ and Bob pick $y$ according to $\varepsilon_{2}$-$free$ source $S_{2}$, respectively.](4.eps){width="50.00000%"} Next, for arbitrary pair $(\varepsilon_{1},\varepsilon_{2})\in$ $R$, we begin to discuss the min-entropy bound for a given expected success probability $E$, which can be resolved by the following optimization problem: $$\begin{aligned} & &\min_{a,y,b,\lambda}H(B\|A,Y,\Lambda)\nonumber\\ & &\mbox{}\text{subject to}: E=\sum_{k=0}^{7}P(\lambda_{k})E_{\lambda_{k}},\nonumber\\ & &\mbox{}E_{\lambda_{k}}=\sum_{a,y}P(a,y\|\lambda_{k})P(b=a_{y}\|a,y,\lambda_{k}),\end{aligned}$$ the optimization is carried out by quantum states $\rho_{a,\lambda}$ and POVMs $\{M_{y,\lambda}^{0},M_{y,\lambda}^{1}\}$ chosen in 2-dimensional Hilbert space for $a\in\{00,01,10,11\}$ and $y\in\{0,1\}$. After that we can estimate the min-entropy bound. Then true random numbers can be produced by a randomness extractor \[13\]. In fact, it plays an important role on many aspects that the min-entropy bound can be estimated as the analytic function of 2-dimensional quantum witness violation, such as security analysis of SDI randomness expansion \[14\]. analytic function ================= In the SDI randomness expansion proposed by Li et al., the figure of the relationship between 2-dimensional quantum witness and the min-entropy is given, but the analytic relationship is not discussed. In this section, for arbitrary pair $(\varepsilon_{1},\varepsilon_{2})\in$ $R$, we explore the analytic relationship between 2-dimensional quantum witness $$E=\sum_{k=0}^{7}P(\lambda_{k})E_{\lambda_{k}}$$ and the min-entropy bound $H(B\|A,Y,\Lambda)=-\log_{2}p$, where $$1/2+(t+\delta)/(2\sqrt{\delta^{2}+2t\delta+1})\leq p\leq 1\nonumber$$ deduced from Eq. (11). We might take $\lambda_{0}$ as example. To depict a encoding-decoding strategy influenced by the hidden variable $\lambda_{0}$, we extract two parameters $(E_{\lambda_{0}},\max_{a,y,b}P(b\|a,y,\lambda_{0}))$, which can be regarded as points in the 2-dimensional coordinate system. For a given encoding-decoding strategy, $E_{\lambda_{0}}$ can be said as the convex combination of success conditional expected success probabilities obtained by pure states and projective measurements, and $\max_{a,y,b}P(b\|a,y,\lambda_{0})$ is not more than the convex combination of the maximal guess probability obtained by the same pure states and projective measurements. That is, for a given value of $E_{\lambda_{0}}$, the convex set composed of the realizable points achieved by pure states and projective measurements will provide a upper concave bound for $\max_{a,y,b}P(b\|a,y,\lambda_{0})$, denote the upper bound as $p_{\lambda_{0}}$. We only discuss this upper concave bound in the following. Apparently, $p_{\lambda_{0}}$ can be viewed as a concave function of $E_{\lambda_{0}}$. Denote $p_{\lambda_{0}}=C(E_{\lambda_{0}})$, $C$ is a concave function. On the other hands, with the increase of $E_{\lambda_{0}}$, the randomness generated by the quantum process will also be monotone increasing, as consequences $C$ is a continuous and decreasing function. Fortunately, this discussion also applies to other hidden variables $\lambda_{k}, k\neq0.$ For any realizable point $(E_{\lambda_{0}},\max_{a,y,b}P(b\|a,y,\lambda_{0}))$, other hidden variables can realize through a single code. It is to say that other hidden variables will reach the same bound $p_{\lambda_{k}}$ as $p_{\lambda_{0}}$ for $E_{\lambda_{k}}=E_{\lambda_{0}},$ i.e., $p_{\lambda_{k}}=C(E_{\lambda_{k}})$. For the given $E$ as indicated in Eq. (13), the lower bound of the min-entropy is $$\begin{aligned} H(B\|A,Y,\Lambda)&=&-\log_{2}\sum_{k=0}^{7}P(\lambda_{k})\max_{a,y,b,\lambda_{k} }P(b\|a,y,\lambda_{k})\nonumber\\ &=&\mbox{}-\log_{2}\sum_{k=0}^{7}P(\lambda_{k})P_{\lambda_{k}}.\end{aligned}$$ Using the Jensen’s inequality, it is natural that if and only if $E_{\lambda_{k}}=E_{\lambda_{k'}}, k\neq k'$ , the lower bound of min-entropy will be reached. Without loss of generality, we might take $E=E_{\lambda_{0}}$, then $p=p_{\lambda_{0}}$. The lower bound of min-entropy can be described as $$H(B\|A,Y,\Lambda)=-\log_{2}C(E).$$ The next work mainly describe the function $C$. Denote $E_{l}$ as $$E_{l}=\max\{E_{\lambda_{0}}:C(E_{\lambda_{0}})=1\}.$$ Obviously, we can deduce that $C(E)=1$ as $E\leq E_{l}$ according to the monotonicity of the function $C$. Therefore, the function $C$ can be completely depicted once the one in the closed interval $[ E_{l}, E_{q}]$ is obtained and the value of $E_{l}$ is determined, which only are discussed in the following. Based on the Refs. \[8-10\], the function $C$ is monotropic as $E\in [ E_{l}, E_{q}]$. Then the function $C$ has a inverse function denoted as $C^{-1}$, i.e., $E=C^{-1}(p)$. The function $C^{-1}$ can be obtained by the following optimization: $$\begin{aligned} & &\max_{a,y, b, \lambda_{0}}E=\sum_{a,y}P(a,y\|\lambda_{0})P(b=a_{y}\|a,y,\lambda_{0})\nonumber\\ & &\mbox{}\text{subject to}: \max_{a,y,b,\lambda_{0}}P(b\|a,y,\lambda_{0})=p,\end{aligned}$$ where $1/2+(t+\delta)/(2\sqrt{\delta^{2}+2t\delta+1})\leq p\leq 1$, the optimization is carried out pure quantum states $\rho_{a,\lambda}$ and projective measurements chosen in 2-dimensional Hilbert space for $a\in\{00,01,10,11\}, y\in\{0,1\}$ and $\lambda\in\Lambda$. Firstly, we focus on the value of $E$ achieved by arbitrary pure states and projective measurements with rank 1 in the optimization (17). Fortunately, $E$ can be viewed as a function $G(\varepsilon_{1},\varepsilon_{2},p)$ determined by $\varepsilon_{1},\varepsilon_{2}$ and $p$ (see the appendix C). Secondly, consider the optimization (17) achieved by arbitrary pure states and measurements $\{I, 0\}$ , $\{0, I\}$ in optimization (17), only $E=E_{c}$ as $p=1$ is obtained. If $E_{\varepsilon_{1},\varepsilon_{2}}\geq E_{ c}$ is established. The strategy with measurements $\{I, 0\}$ or $\{0, I\}$ can be simulated by one with pure states and projective measurements with rank 1, where $E_{\varepsilon_{1},\varepsilon_{2}}=G(\varepsilon_{1},\varepsilon_{2},1)$. Here $E=G(\varepsilon_{1},\varepsilon_{2},p)$. On the contrary, if $E_{\varepsilon_{1},\varepsilon_{2}}< E_{c}$, then we have to discuss the convex set which is composed of points $(G(\varepsilon_{1},\varepsilon_{2},p),p)$ and $(E_{c},1)$ to obtain the upper bound of $E$ denoted as $F(\varepsilon_{1},\varepsilon_{2},p)$, which is a function of $\varepsilon_{1},\varepsilon_{2}$ and $p$. In fact, the points $(F(\varepsilon_{1},\varepsilon_{2},p), p)$ also provide a lower bound of min-entropy. However, whether the bound is tight or not cannot be determined. Based on the above analysis, we have $E_{l}=\max\{E_{c}, E_{\varepsilon_{1},\varepsilon_{2}}\}$ and for $E_{l}\leq E\leq E_{q}$, $$E=C^{-1}(p)=\left\{ \begin{array}{lcl} G(\varepsilon_{1},\varepsilon_{2},p), & &if \ E_{\varepsilon_{1},\varepsilon_{2}}\geq E_{c},\\ F(\varepsilon_{1},\varepsilon_{2},p)& &if \ E_{\varepsilon_{1},\varepsilon_{2}}< E_{c}, \end{array} \right.$$ where $1/2+(t+\delta)/(2\sqrt{\delta^{2}+2t\delta+1})\leq p\leq 1$. In fact, the function $G$ and $F$ can describe the relationship between the min-entropy bound and the 2-dimensional quantum witness violation in detail. Denote $\beta=\arccos(2p-1)$, we have $$\begin{aligned} G(\varepsilon_{1},\varepsilon_{2},p)=\max_{\alpha\in[0,\pi-4\beta], i=1,2}\{G_{i}(\varepsilon_{1},\varepsilon_{2},p,\alpha)\}.\nonumber\end{aligned}$$ The analytic functions $G_{1},G_{2}$ are $$\begin{aligned} G_{1}(\varepsilon_{1},\varepsilon_{2},p,\alpha)&=&\frac{1}{2}+\frac{1}{2}(\frac{1}{2}-\varepsilon_{1})^{2}(\frac{1}{2}-\varepsilon_{2})[\delta \cos\beta\nonumber\\ & &\mbox{}+\cos(\beta+\alpha)+f(\varepsilon_{1},\varepsilon_{2},p,\alpha)].\end{aligned}$$ $$\begin{aligned} G_{2}(\varepsilon_{1},\varepsilon_{2},p,\alpha)&=&\frac{1}{2}+\frac{1}{2}(\frac{1}{2}-\varepsilon_{1})^{2}(\frac{1}{2}-\varepsilon_{2})[\delta \sigma \cos\beta\nonumber\\ & &\mbox{}+\sigma \cos(\beta+\alpha)+g(\varepsilon_{1},\varepsilon_{2},p,\alpha)],\nonumber\end{aligned}$$ where $\sigma=(1+2\varepsilon_{1})/(1-2\varepsilon_{1})$ ($f,g$ see Eq. (C12), Eq. (C13), respectively). The function $F$ can be depicted by the function $G$ and $$\begin{aligned} F(\varepsilon_{1},\varepsilon_{2},p)&=&(G(\varepsilon_{1},\varepsilon_{2},p_{0})-E_{c})(1-p)/(1-p_{0})\nonumber\\ & &\mbox{}+E_{c}\end{aligned}$$ as $p\geq p_{0}$, $F(\varepsilon_{1},\varepsilon_{2},p)=G(\varepsilon_{1},\varepsilon_{2},p)$ as $p<p_{0}$, where $p_{0}$ satisfies $$\frac{G(\varepsilon_{1},\varepsilon_{2},p_{0})-E_{c}}{p_{0}-1}=\min_{p}\{\frac{G(\varepsilon_{1},\varepsilon_{2},p)-E_{c}}{p-1}\}.\nonumber$$ In particular, for the cases of $\varepsilon_{1}=\varepsilon_{2}<0.1358$; $\varepsilon_{1}<0.2203, \varepsilon_{2}=0$ and $\varepsilon_{1}=0, \varepsilon_{2}<0.5$, we have $E_{\varepsilon_{1},\varepsilon_{2}}=\max_{\alpha\in[0,\pi-4\beta]}\{G_{1}(\varepsilon_{1},\varepsilon_{2},1,\alpha)\}.$ In the Ref. \[8\], the analytic relationship between the min-entropy bound and the 2-dimensional quantum witness violation is $E=G(0,0,2^{-H_{\infty}(B\|A,Y,\Lambda)})$. ![The relationship between the min-entropy bound and the 2-dimensional quantum witness for $0\leq\varepsilon<0.1358$, where the choices of states and measurements are derived from $\varepsilon$-$free$ sources $S_{1}$ and $S_{2}$, respectively.](5.eps){width="50.00000%"} Many works only consider projective measurements with rank 1 just because they happen to satisfy $E_{\varepsilon_{1},\varepsilon_{2}}\geq E_{c}$, such as the Refs. \[8, 9\]. We might take $\varepsilon_{1}=\varepsilon_{2}=\varepsilon$ as example. In fact, $E_{\varepsilon_{1},\varepsilon_{2}}\geq E_{c}$ as $\varepsilon\leq 0.12348$. But $E_{\varepsilon_{1},\varepsilon_{2}}< E_{c}$ as $0.12348<\varepsilon <0.1358$. This shows that the situation of $E_{\varepsilon_{1},\varepsilon_{2}}<E_{c}$ will occur with the increase of $\varepsilon_{1},\varepsilon_{2}$ and the measurements $\{I,0\}$ and $\{0.I\}$ must be taken into consideration. Furthermore, the relationship between the min-entropy bound and the 2-dimensional quantum witness for $\varepsilon_{1}=\varepsilon_{2}$ is demonstrated as the Fig.3. Conclusion ========== We proved that partially free sources can bring us new randomness and proposed a SDI randomness expansion protocol with partially free sources based on $2\rightarrow1$ QRAC. In our protocol, the condition that the partially free sources should satisfy to generate new randomness was gained without strictly requiring $\varepsilon_{1}=\varepsilon_{2}$ (the choices of states and measurements are derived from $\varepsilon_{1}$-$free$ source and $\varepsilon_{2}$-$free$ source, respectively). Furthermore, a new 2-dimensional quantum witness and the analytic relationship between the generated randomness and the 2-dimensional quantum witness violation were obtained. In addition, the advantage of no containing entanglement which is introduced in the Ref. \[8, 9\] also apply to our protocol. We conjecture that it can get better results in the SDI randomness expansion with partially free sources based on $n\rightarrow1$ QRACs for $n\geq3$. The authors would like to thank Z. Q. Yin for many valuable suggestions and Q. N. Zhou for Fig.3. This work is supported by NSFC (Grant Nos. 61272057, 61170270,U1304604), Beijing Higher Education Young Elite Teacher Project (Grant Nos. YETP0475, YETP0477). The assumption in Eq. (5) apparently satisfy Eq. (3). In addition, Eq. (4) is equivalent to $$\begin{aligned} P(a_{i}=0)&=&\sum_{k=0}^{7}P(\lambda_{k})P(a_{0}=0\|\lambda_{k})=\frac{1}{2},i=0,1.\nonumber\\ P(y=0)&=&\sum_{k=0}^{7}P(\lambda_{k})P(y=0\|\lambda_{k})=\frac{1}{2}.\end{aligned}$$ The Eq. (A1) can be written as $$\begin{aligned} \sum_{k=0,1,2,3}P(\lambda_{k})&-&\sum_{k=4,5,6,7}P(\lambda_{k})=0.\nonumber\\ \sum_{k=0,1,4,5}P(\lambda_{k})&-&\sum_{k=2,3,6,7}P(\lambda_{k})=0.\nonumber\\ \sum_{k=0,2,4,6}P(\lambda_{k})&-&\sum_{k=1,3,5,7}P(\lambda_{k})=0.\end{aligned}$$ They are the linear equations of $\lambda_{k}, k=0,1,\cdot\cdot\cdot,7$. It is easy to see there must be many solutions to Eq. (A2) as there are three equations but eight variables. The expected success probability with variable $\lambda_{k}$ is $$E_{\lambda_{k}}=\sum_{a,y}P(a,y\|\lambda_{k})P(b=a_{y}\|a,y,\lambda_{k})$$ for $k=0,1,\cdot\cdot\cdot,7.$ We might take the $\lambda_{0}$ as example. Set the Bloch sphere representation of the pure state $\rho_{a,\lambda_{0}}$ , projective measure $\{M_{y,\lambda_{0}}^{0},M_{y,\lambda_{0}}^{1}\}$ as the Bloch vector $\textbf{\emph{r}}_{a,\lambda_{0}}$, $\{\textbf{\emph{v}}_{y,\lambda_{0}},-\textbf{\emph{v}}_{y,\lambda_{0}}\}$ for $a\in\{00,01,10,11\}$ and $y\in\{0,1\}$, respectively. Without loss of generality, let $\textbf{\emph{v}}_{0,\lambda_{0}}=(1,0,0).$ By the Ref. \[10\], we can know $$\begin{aligned} P(b\|a,y,\lambda_{0})&=&tr(\rho_{a,\lambda_{0}}M_{y,\lambda_{0}}^{b})\nonumber\\ &=&\mbox{}\frac{1}{2}(1+\textbf{\emph{r}}_{a,\lambda_{0}}\cdot(-1)^{b}\textbf{\emph{v}}_{y,\lambda_{0}}),\end{aligned}$$ where $``\cdot"$ denotes the inner product. With a small amount of calculation, we get $$\begin{aligned} E_{\lambda_{0}}&=&\frac{1}{2}+\frac{1}{2}\sum_{a,y}P(a,y\|\lambda_{0})\textbf{\emph{r}}_{a,\lambda_{0}}\cdot(-1)^{b}\textbf{\emph{v}}_{y,\lambda_{0}}\nonumber\\ &=&\mbox{}\frac{1}{2}+\frac{1}{2}\sum_{a}P(a\|\lambda_{0})\textbf{\emph{r}}_{a,\lambda_{0}}\cdot\textbf{\emph{v}}_{a,\lambda_{0}}\nonumber\\ &\leq&\mbox{}\frac{1}{2}+\frac{1}{2}\sum_{a}P(a\|\lambda_{0})\\|\textbf{\emph{v}}_{a,\lambda_{0}}\\|\nonumber\\\end{aligned}$$ where $\textbf{\emph{v}}_{a,\lambda_{0}}$ defined as Eq. (8). Only the case of $\parallel\textbf{\emph{v}}_{a,\lambda_{0}}\parallel\neq 0$ is discussed in the following. If and only if $\textbf{\emph{r}}_{a,\lambda_{0}}=\textbf{\emph{v}}_{a,\lambda_{0}}/\\|\textbf{\emph{v}}_{a,\lambda_{0}}\\|$, the Eq. (B3) can achieve the maximum value. Furthermore, set $\theta$ is the angle between $\textbf{\emph{v}}_{0,\lambda_{0}}$ and $\textbf{\emph{v}}_{1,\lambda_{0}}$, then $$\begin{aligned} & &\\|\textbf{\emph{v}}_{00,\lambda_{0}}\\|^{2}+\\|\textbf{\emph{v}}_{01,\lambda_{0}}\\|^{2}=1+4\varepsilon_{2}^{2}.\nonumber\\ & &\mbox{}P(00\|\lambda_{0})+P(11\|\lambda_{0})=\frac{1}{2}+2\varepsilon_{1}^{2}.\nonumber\\ & &\mbox{}P(01\|\lambda_{0})+P(10\|\lambda_{0})=\frac{1}{2}-2\varepsilon_{1}^{2},\end{aligned}$$ where Alice and Bob have the $\varepsilon_{i}$-$free$ source $S_{i},i=0,1$ to pick $a,y$, respectively. With the knowledge of Eq. (B4), we have $$\begin{aligned} \sum_{a}P(a\|\lambda_{0})\\|\textbf{\emph{v}}_{a,\lambda_{0}}\\|&=&(\frac{1}{2}+2\varepsilon_{1}^{2})\\|\textbf{\emph{v}}_{00,\lambda_{0}}\\|\nonumber\\ & &\mbox+(\frac{1}{2}-2\varepsilon_{1}^{2})\\|\textbf{\emph{v}}_{01,\lambda_{0}}\\|\nonumber\\ &\leq&\mbox{}\sqrt{\frac{1}{2}+8\varepsilon_{1}^{4}}\sqrt{1+4\varepsilon_{2}^{2}}.\end{aligned}$$ If and only if $\\|\textbf{\emph{v}}_{00,\lambda_{0}}\\|/\\|\textbf{\emph{v}}_{01,\lambda_{0}}\\|=(1+4\varepsilon_{1}^{2})/(1-4\varepsilon_{1}^{2})$, that is, the angle $\theta$ must satisfies $\cos\theta=t$, $t$ is defined as Eq. (7), the Eq. (B5) can achieve the maximum value. If $t\leq1$ for a pair $(\varepsilon_{1},\varepsilon_{2})$. Fortunately, let $\theta=\arccos t$, we will reach the maximum value of $E_{\lambda_{0}}$ denoted as $$\begin{aligned} E_{\lambda_{0}, q}^{ max}=\frac{1}{2}+\frac{1}{2}\sqrt{\frac{1}{2}+8\varepsilon_{1}^{4}+2\varepsilon_{2}^{2}+32\varepsilon_{1}^{4}\varepsilon_{2}^{2}}.\end{aligned}$$ At the same time, the maximum success probability with variable $\lambda_{0}$ can be obtained: $$\begin{aligned} P(b=0\|00,0,\lambda_{0})=\frac{1}{2}(1+\frac{t+\delta}{\sqrt{\delta^{2}+2t\delta +1}}).\end{aligned}$$ If $t>1$ for a pair $(\varepsilon_{1},\varepsilon_{2})$, we have to set $\cos\theta=1$, i.e., $\theta=0$ to obtain the maximum value $$\begin{aligned} E_{\lambda_{0}, q}^{ max}=\frac{3}{4}+\frac{1}{2}\varepsilon_{2}+\varepsilon_{1}^{2}(1-2\varepsilon_{2}).\end{aligned}$$ For the sake of simplicity, suppose $E=E_{\lambda_{0}}$ and $p=p_{\lambda_{0}}$. To reach the maximum value of $E$ satisfying a condition that the probability distribution satisfies $\max_{a,y,b}P(b\|a,y,\lambda_{0})=p=\frac{1}{2}(1+\cos\beta)$, we consider the question that which one of 16 probability $P(b\|a,y,\lambda_{0})$ should reach $p$, where $$(t+\delta)/\sqrt{\delta^{2}+2t\delta +1}\leq\cos\beta\leq 1.$$ Apparently, it must be a guessing success probability and need to satisfy the following conditions: ($\romannumeral1$) It must be the probability $P(b=a_{0}\|a,y=0,\lambda_{0})$ for $a\in\{00,01,10,11\}$. Since $P(b=a_{0}\|a,y=0,\lambda_{0})\geq P(b=a_{1}\|a,y=1,\lambda_{0})$ can achieve a larger value of $E$ concluded from $P(y=0\|\lambda_{0})\geq P(y=1\|\lambda_{0})$. ($\romannumeral2$) The bolch vectors $\textbf{\emph{r}}_{a,\lambda_{0}}$, $\textbf{\emph{v}}_{y,\lambda_{0}}$ for all $a,y$ are in a plane, $\textbf{\emph{r}}_{a,\lambda_{0}}$ fall on the area between $(-1)^{a_{0}}\textbf{\emph{v}}_{0,\lambda_{0}}$ and $(-1)^{a_{1}}\textbf{\emph{v}}_{1,\lambda_{0}}$. Then only four cases of $P(b=a_{0}\|a,y=0,\lambda_{0})=p$ for $a\in\{00,01,10,11\}$ need to be discussed on the strict precondition ($\romannumeral2$). It is noteworthy that $$P(b=a_{1}\|a,y=1,\lambda_{0})\leq P(b=a_{0}\|a,y=0,\lambda_{0})$$ always be established no matter which one of $P(b=a_{0}\|a,y=0,\lambda_{0})$ reach $p$. Case 1: Let $$\begin{aligned} P(b=1\|11,y=0,\lambda_{0})=p, \end{aligned}$$ that is, the angle between $-\textbf{\emph{v}}_{0,\lambda_{0}}$ and $\textbf{\emph{r}}_{11,\lambda_{0}}$ is $\beta$, suppose the angle between $\textbf{\emph{r}}_{11,\lambda_{0}}$ and $-\textbf{\emph{v}}_{1,\lambda_{0}}$ is $\beta+\alpha.$ Since $ P(b=1\|11,y=1,\lambda_{0})\leq p$, let $ \alpha\geq 0$ . Fortunately, for any $ \alpha\geq 0$, there is at least one choice can satisfy $P(b=0\|00,y,\lambda_{0})\leq p$ which let $\textbf{\emph{r}}_{00,\lambda_{0}}$ along the same direction as $\textbf{\emph{v}}_{0,\lambda_{0}}+\textbf{\emph{v}}_{1,\lambda_{0}}$. In order to ensure $$\begin{aligned} P(b=0\|01,y=0,\lambda_{0})&\leq&p,\nonumber\\ P(b=1\|10,y=0,\lambda_{0})&\leq&p,\end{aligned}$$ then $\alpha\leq \pi -4\beta$ concluded from $[\pi-(2\beta+\alpha)]/2\geq\beta.$ We confirm that the range of $\alpha$ is $[0, \pi -4\beta]$. If the value of $\alpha$ is determinated, the angle between $\textbf{\emph{v}}_{0,\lambda_{0}}$ and $\textbf{\emph{v}}_{1,\lambda_{0}}$ is determined, then the vector $\textbf{\emph{v}}_{1,\lambda_{0}}$ is determined since we have set $\textbf{\emph{v}}_{0,\lambda_{0}}=(1,0,0).$ Next only need to consider how to place $\textbf{\emph{r}}_{a,\lambda_{0}}$ for the purposes of reaching the largest $E$. Firstly, $\textbf{\emph{r}}_{00,\lambda_{0}}$ has been determined. Secondly, denote $\varphi$ as the angle between $\textbf{\emph{r}}_{01,\lambda_{0}}$ and $\textbf{\emph{v}}_{0,\lambda_{0}}$ . $\varphi\leq\pi/2$ is obtained derived from the Eq. (C2). To get a lager value of $E$, we want to set $$\begin{aligned} \textbf{\emph{r}}_{01,\lambda_{0}}=\textbf{\emph{v}}_{01,\lambda_{0}}/\\|\textbf{\emph{v}}_{01\lambda_{0}}\\|, \end{aligned}$$ but must guarantee $\varphi \geq \beta$ for the sake of that Eq. (C4) will not be established, i.e., if Eq. (C5) want to be established, we must guarantee that $$\begin{aligned} \tan\varphi =\frac{\sin(2\beta+\alpha)}{\delta-\cos(2\beta+\alpha)}\geq \tan\beta. \end{aligned}$$ Using the knowledge of trigonometric functions, the Eq. (C5) is equivalent to $\sin(3\beta+\alpha)\geq \delta\sin\beta.$ Combining the condition $\alpha\in[0, \pi -4\beta]$ and $\sin(3\beta+\alpha)\geq \delta\sin\beta$ yields that the Eq. (C6) can be established only for $\alpha\in[a_{1}, a_{2}]$, where $a_{1}=\max \{0, \arcsin(\delta\sin\beta)-3\beta\}, a_{2}=\pi-3\beta-\arcsin(\delta\sin\beta)$. If $\alpha\in[0, a_{1})\cup(a_{2}, \pi-4\beta],$ we have to set $\varphi=\beta$ to get a lager value of $E$. At last, by the similar way, suppose $\textbf{\emph{r}}_{10,\lambda_{0}}=\textbf{\emph{v}}_{10,\lambda_{0}}/\\|\textbf{\emph{v}}_{10\lambda_{0}}\\|$ for $\alpha\in[a_{1}, a_{2}]$. If $\alpha\in[0, a_{1})\cup(a_{2}, \pi-4\beta],$ suppose the angle between $\textbf{\emph{r}}_{10,\lambda_{0}}$ and $-\textbf{\emph{v}}_{0,\lambda_{0}}$ is $\beta$, Assume $\textbf{\emph{a}}_{00,\lambda_{0}}=\textbf{\emph{v}}_{00,\lambda_{0}}/\\|\textbf{\emph{v}}_{00,\lambda_{0}}\\|$ for $\alpha\in[b_{1}, b_{2}]$. In the case of $\alpha\in[0, b_{1})\cup(b_{2}, \pi-4\beta]$, let $\beta$ the angle between $\textbf{\emph{a}}_{00,\lambda_{0}}$ and $\textbf{\emph{v}}_{0,\lambda_{0}}$ as $\beta$, where $b_{1}=\arcsin(\delta\sin\beta)-\beta, b_{2}=\min\{\pi-4\beta, \pi-\arcsin(\delta\sin\beta)-\beta\}$. It’s worth noting that $b_{1}\leq a_{2}$ for any pair of the feasible region. Based on the analysis of the above, we can get the analytic function of $E$ in the case 1 and denote it as $G_{1}(\varepsilon_{1},\varepsilon_{2},p,\alpha)$ shown as Eq. (20). With some calculation, we can obtain $$\begin{aligned} G_{1}(\varepsilon_{1},\varepsilon_{2},p,\alpha)&=&\sum_{a,y}P(a,y\|\lambda_{0})P(b=a_{y}\|a,y,\lambda_{0})\nonumber\\ &=&\mbox{}\frac{1}{2}+\frac{1}{2}(\frac{1}{2}-\varepsilon_{1})^{2}(\frac{1}{2}-\varepsilon_{2})[\delta \cos\beta\nonumber\\ & &\mbox{}+\cos(\beta+\alpha)+f(\varepsilon_{1},\varepsilon_{2},p,\alpha)]\end{aligned}$$ where $f(\varepsilon_{1},\varepsilon_{2},p,\alpha)$ is displayed as Eq. (C12). $$f(\varepsilon_{1},\varepsilon_{2},p,\alpha)=\left\{ \begin{aligned} & (2\delta\sigma+\sigma^{2}\delta)\cos\beta +\sigma^{2}\cos(\beta+\alpha)-2\sigma\cos(3\beta+\alpha)& & if\ \alpha\in[0,a_{1})\cup(b_{2},\pi-4\beta],\\ & \sigma^{2}\delta\cos\beta+\sigma^{2}\cos(\beta+\alpha)+2\sigma\sqrt{\delta^{2}+1-2\delta\cos(2\beta+\alpha)}& & if\ \alpha\in[a_{1}, b_{1}),\\ & \sigma^{2}\sqrt{\delta^{2}+1+2\delta\cos(2\beta+\alpha)}+2\sigma\sqrt{\delta^{2}+1-2\delta\cos(2\beta+\alpha)}& & if\ \alpha\in [b_{1}, a_{2}),\\ & \sigma^{2}\sqrt{\delta^{2}+1+2\delta\cos(2\beta+\alpha)}+2\delta\sigma\cos\beta-2\sigma\cos(3\beta+\alpha)& & if\ \alpha\in [a_{2}, b_{2}].\\ \end{aligned} \right.$$ $$g(\varepsilon_{1},\varepsilon_{2},p,\alpha)=\left\{ \begin{aligned} & \delta\sigma\cos\beta+\sigma\cos(\beta+\alpha)+(\sigma^{2}\delta+\delta)\cos\beta-(\sigma^{2}+1)\cos(3\beta+\alpha)& &if \ \alpha\in[0,a_{1})\cup(b_{2},\pi-4\beta],\\ & \delta\sigma\cos\beta+\sigma\cos(\beta+\alpha)+(\sigma^{2}+1)\sqrt{\delta^{2}+1-2\delta\cos(2\beta+\alpha)}& &if \ \alpha\in[a_{1}, b_{1}),\\ & \sigma\sqrt{\delta^{2}+1+2\delta\cos(2\beta+\alpha)}+(\sigma^{2}+1)\sqrt{\delta^{2}+1-2\delta\cos(2\beta+\alpha)}& &if \ \alpha\in [b_{1}, a_{2}),\\ & \sigma\sqrt{\delta^{2}+1+2\delta\cos(2\beta+\alpha)}+(\sigma^{2}\delta+\delta)\cos\beta-(\sigma^{2}+1)\cos(3\beta+\alpha)& &if \ \alpha\in [a_{2}, b_{2}]. \end{aligned} \right.$$ Case 2: Let $$\begin{aligned} P(b=0\|01,y=0,\lambda_{0})=p,\nonumber \end{aligned}$$ that is, the angle between $-\textbf{\emph{v}}_{0,\lambda_{0}}$ and $\textbf{\emph{r}}_{01,\lambda_{0}}$ is $\beta$, set the angle between $\textbf{\emph{r}}_{01,\lambda_{0}}$ and $-\textbf{\emph{v}}_{1,\lambda_{0}}$ is $\beta+\alpha.$ Obviously, the range of $\alpha$ remains $[0,\pi-4\beta].$ By a similar way, denote $E$ as $G_{2}(\varepsilon_{1},\varepsilon_{2},p,\alpha)$ in this case. we have $$\begin{aligned} G_{2}(\varepsilon_{1},\varepsilon_{2},p,\alpha)&=&\frac{1}{2}+\frac{1}{2}(\frac{1}{2}-\varepsilon_{1})^{2}(\frac{1}{2}-\varepsilon_{2})[\delta \sigma \cos\beta\nonumber\\ & &\mbox{}+\sigma \cos(\beta+\alpha)+g(\varepsilon_{1},\varepsilon_{2},p,\alpha)]\end{aligned}$$ The detailed description of $g$ is displayed in Eq. (C13). Case 3: Let $$\begin{aligned} P(b=1\|10,y=0,\lambda_{0})=p.\nonumber \end{aligned}$$ The angle between $-\textbf{\emph{v}}_{0,\lambda_{0}}$ and $\textbf{\emph{r}}_{10,\lambda_{0}}$ is $\beta$, set the angle between $\textbf{\emph{r}}_{01,\lambda_{0}}$ and $-\textbf{\emph{v}}_{1,\lambda_{0}}$ is $\beta+\alpha.$ The analytic function of $E$ is equal to $G_{2}(\varepsilon_{1},\varepsilon_{2},p,\alpha)$ concluded from $P(a=10)=P(a=01).$ Case 4: Let $$\begin{aligned} P(b=0\|00,y=0,\lambda_{0})=p.\nonumber \end{aligned}$$ Set the angle between $\textbf{\emph{r}}_{00,\lambda_{0}}$ and $\textbf{\emph{v}}_{1,\lambda_{0}}$ is $\beta+\alpha.$ Denote $E$ as $G_{3}(\varepsilon_{1},\varepsilon_{2},p,\alpha)$ in this case and $\alpha\in[0,\pi-4\beta].$ By the same analysis as the case 1, we get $$G_{3}(\varepsilon_{1},\varepsilon_{2},p,\alpha)=G_{1}(\varepsilon_{1},\varepsilon_{2},p,\alpha)$$ for $\alpha\in[0,b_{1}]\cup[b_{2},\pi-4\beta].$ For $\alpha\in[b_{1}, b_{2}]$, we have $$\begin{aligned} G_{3}(\varepsilon_{1},\varepsilon_{2},p,\alpha)&=&\frac{1}{2}+\frac{1}{2}(\frac{1}{2}-\varepsilon_{1})^{2}(\frac{1}{2}-\varepsilon_{2})[\sigma^{2}\delta\cos\beta\nonumber\\ & &\mbox{}+\sqrt{\delta^{2}+1+2\delta\cos(2\beta+\alpha)}+\nonumber\\ & &\mbox{}\sigma^{2}\cos(\beta+\alpha)+k(\varepsilon_{1},\varepsilon_{2},p,\alpha)]\nonumber\end{aligned}$$ where $k(\varepsilon_{1},\varepsilon_{2},p,\alpha)=$ $$\left\{ \begin{aligned} & 2\sigma\sqrt{\delta^{2}+1-2\delta\cos(2\beta+\alpha)}& & if\ \alpha\in[b_{1}, a_{2})\\ & 2\delta\sigma\cos\beta-2\sigma\cos(3\beta+\alpha)& & if\ \alpha\in [a_{2}, b_{2}].\\ \end{aligned} \right.$$ Therefore, for $\alpha\in[b_{1}, b_{2}]$, $$\begin{aligned} G_{1}(\varepsilon_{1},\varepsilon_{2},p,\alpha)\geq G_{3}(\varepsilon_{1},\varepsilon_{2},p,\alpha)\end{aligned}$$ always is established. We have $$G(\varepsilon_{1},\varepsilon_{2},p)=\max_{\alpha\in[0,\pi-4\beta];j=1,2}\{G_{i}(\varepsilon_{1},\varepsilon_{2},p,\alpha)\}.\nonumber$$ concluded from Eq. (C9) and Eq. (C11), where $G_{1}$ and $G_{2}$ are shown as Eq. (C7) and Eq. (C8). In particular, $\arcsin(\delta\sin\beta)\leq0$ and $\pi-\arcsin(\delta\sin\beta)-\beta\geq\pi-4\beta$ can be obtained by the tool of Matlab as $\varepsilon_{1}=\varepsilon_{2}\leq 0.1358$. Then $a_{1}=0, b_{2}=\pi-4\beta$, and the set of $[0,a_{1})\cup(b_{2},\pi-4\beta]$ does not exist. The calculation process will becomes much more simple. \[sec:TeXbooks\] R. Colbeck, and R. Renner, Nature Physics 8, pp. 450-453 (2012). C. H. Bennett, and G. Brassard, in Proceedings of the IEEE International Conference on Computers, Systems and Signal Processing, Bangalore, India (IEEE, New York, 1984), pp. 175-179. J. Bouda, M. Pivoluska, M. Plesch, and C. Wilmott1, Phys. Rev. A 86, 062308 (2012). R. Gallego, N. Brunner, C. Hadley, and A. Acin, Phys. Rev. Lett. 105, 230501 (2010). M. Paw[ł]{}owski and N. Brunner, Phys. Rev. A 84, 010302(R)(2011). R. Colbeck and A. Kent, J. Phys. A: Math. Theor. 44, 095305 (2011). S. Pironio et al., Nature (London) 464, 1021-1024 (2010). H-W. Li, Z-Q. Yin, Y-C. Wu, X-B. Zou, S. Wang, W. Chen, G-C. Guo, and Z-F. Han, Phys. Rev. A 84, 034301 (2011). H-W. Li, M. Pawlowski, Z-Q. Yin, G-C. Guo , and Z-F. Han, Phys. Rev. A 85, 052308 (2012). A. Ambainis, D. Leung, L. Mancinska, and M. Ozols, e-print arXiv:0810.2937. Y-K. Wang, S-J. Qin, T-T. Song, F-Z. Guo, W. Huang, and H-J. Zuo, Phys. Rev. A 89, 032312 (2014). L. Masanes, e-print arXiv:0512100. N. Nisan, and A.Ta-Shma, J. Comput. Syst. Sci. 58, pp. 148-173 (1999). S. Fehr, R. Gelles, and C. Schaffner, Phys. Rev. A 87, 012335 (2013). J. F. Dynes, Z. L. Yuan, A. W. Sharpe, and A. J. Shields, Appl. Phys. Lett. 93, 031109 (2008). M. Ren, E. Wu, Y. Liang, Y. Jian, G. Wu, and H. Zeng, Phys. Rev. A 83, 023820 (2011).	{ "pile_set_name": "ArXiv" }
--- author: - \| Fabrizio Catanese\ Universität Bayreuth date: 'October 29, 2002' title: 'From Abel’s Heritage: transcendental objects in algebraic geometry and their algebrization.' --- [^1] Welcome the Abel Prize and long live the memory of Abel, long live mathematics! [^2] \ Introduction ============ $${\bf CONTENTS}$$ - 2\. Abel, the algebraist ? - 3\. The [geometrization]{} of Abel’s methods. - 4\. Algebraization of the Geometry. - 5\. Further[^3] links to the Italian School. - 6\. More new results and open problems. The Encyclopedic Dictionary of Mathematics (edited in Japan) is not a book devoted to the history of mathematics, but tries instead to briefly introduce the reader to the current topics of mathematical research. By non only lexicographical coincidence it starts with “Abels, Niels Henrik” as topic 1. It contains a succint biography of Abel “ Niels Henrik Abel (august 5, 1802 - april 6 , 1829) ... In 1822, he entered the University of Christiania \[today’s Oslo\] .. died at twenty-six of tuberculosis. His best known works are: the result that algebraic equations of order five or above cannot in general be solved algebraically; the result that [^4]\Abelian equations \[i.e., with Abelian Galois group\] can be solved algebraically; the theory of \binomial series and of \elliptic functions; and the introduction of \Abelian functions. His work in both algebra and analysis, written in a style conducive to easy comprehension, reached the highest level of attainment of his time.” Talking about Abel’s heritage entails thus talking about a great part of modern mathematics, as it is shown by the ubiquity of concepts such as Abelian groups, Abelian integrals and functions, Abelian Varieties, and of relatives of theirs as anabelian geometry, nonabelian Hodge theory... Writing is certainly a more difficult task than talking, when the time limits force us to plan our way on secure and direct tracks: for this reason we decided that the present text, with the exception of a couple of protracted mathematical discussions, should essentially be the text of our oral exposition at the Abel Bicentennial Conference. Thus its aim is just to lead the reader along quite personal views on history and development of mathematics, and on certain topics in the still very alive subject of transcendental algebraic geometry. One could declare its Leitmotiv to be G.B. Vico’s theory of cycles in the history of mankind, adapted to the analysis of mathematical evolutions and revolutions: Geometry in ancient Greece, Algebra by the Arabs and in early Renaissance, Geometry again by B. Cavalieri and his indivisibles, Analysis and Physics by the Bernoulli’s,.. and later on an intricate succession of points of view and methods, often alternative to each other, or striving in directions opposite one to the other, which all together enriched our knowledge and understanding of the mathematical reality. Therefore, if we conceive algebra, geometry .. more as methodologies than as domains of knowledge, comes out naturally the difficult question: which way of doing mathematics is the one we are considering ? This question, probably a sterile question when considering the history of mathematics, is however a very important one when we are making choices for future directions of mathematical research: to purport this assertion it will suffice only to cite the (for me, even exaggerated) enthusiasm of nowadays algebraic geometers for the new insights coming to their field by physical theories, concepts and problems. In any case, in our formerly bourgeois world, idle questions with provocative answers to be defended at tea time at home or in a Cafe’, have often motivated interesting discussions, and it is just my hope to be able to do the same thing here. Abel, the algebraist? ===================== So I will start, as due, by citing Hermann Weyl’s point of view (cf. [@Weyl31],[@Weyl] and also [@yag], pp. 26 and 151, for comments), expressed in an address directed towards mathematics teachers, and later published in the Journal : Unterrichtsblätter für Mathematik und Naturwissenschaften, Band 38 (1933), s. 177-188. There are two Classes of mathematicians: - ALGEBRAISTS: as Leibniz, Weierstrass - GEOMETERS-PHYSICISTS: as Newton, Riemann, Klein and people belonging to different classes may tend to be in conflict with each other. The tools of the algebraists are : logical argumentation, formulae and their clever manipulation, algorithms. The other class relies more on intuition, and graphical and visual impressions. For them it is more important to find a new truth than an elegant new proof. The concept of rigour is the battlefield where the opposite parties confront themselves, and the conflicts which hence derived were sometimes harsh and longlasting. The first well known example is the priority conflict between Leibniz and Newton concerning the invention of the Calculus (which however was invented independently by the two scientists, as it is currently agreed upon). The inputs which the two scientists provided did indeed integrate themselves perfectly. On one hand the pure algebraic differential quotient $dy/dx$ would be a very dry concept (algebraists nevertheless are still nowadays very keen on inflicting on us the abstract theory of derivations!) without the intuition of velocities and curve tangents; on the other hand, in the analysis of several phenomena, a physical interpretation of Leibniz’s rule can turn out to be amazingly complicated . More closely related with Abelian integrals and their periods was Weierstrass’ constructive criticism of the “Principles” by Riemann and Dirichlet. As also pointed out in the contribution[^5] by Schappacher, this conflict soon became the Berlin-Göttingen conflict, and almost deflagrated between Weierstrass and Felix Klein (who continued on the way started as a student of Clebsch)[^6] . Klein’s antipathy for Weierstrass was more intellectual than personal: Klein put a special enphasys on geometrical and physical intuition, which he managed to develop in the students by letting them construct solid (plaster, or metal) models of curves and surfaces in ordinary 3-space, or letting them draw very broad (1 x 2 meters) paper tables of cubic plane curves with an explicit plotting of their ${\ensuremath{\mathbb{Q}}}$-rational points. A concrete witness to this tradition is the exhibition of plaster models of surfaces which are to be found still nowadays in the Halls of the mathematical Institute in Göttingen. These models were then produced by the publishing company L. Brill in Darmstadt, later by the Schilling publishing company, and sold around the world: I have personally seen many of those in most of the older Deparments I have visited (cf. the 2 Volumes edited by G.Fischer on “Mathematical Models” [@fisch86] [^7] ). A similar trick, with computer experiments replacing the construction of models, is still very much applied nowadays in the case where professors have to supervise too many more students’ theses than they can really handle. Weierstrass had a victory, in the sense that not only the theory of calculus, but also the theory of elliptic functions is still nowadays taught almost in the same way as it was done in his Berlin lectures. But Klein’s “defeat” (made harder by the long term competition with Poincare’ about the proof of the uniformization theorem, see later) was however a very fertile humus for the later big growth of the Göttingen influence, and certainly Weyl’s meditations which we mentioned above were reflecting this historically important controversy. Where does then Abel stay in this classification? I already took position, with my choice of the title of this section: Abel is for me an algebraist and I was glad [^8] to hear Christian Houzel stressing in his contribution the role of Abel’s high sense of rigour. Abel’s articles on the binomial coefficients, on the summation of series, and on the solution of algebraic equations testify his deep concern for the need of clear and satisfactory proofs [^9]. Of course, like many colours are really a mixture of pure colours, the same occurs for mathematicians, and by saying that he was deep down an “algebraist” I do not mean to deny that he possessed a solid geometrical intuition, as we shall later point out. In fact, Abel himself was proud to introduce himself during his travels as ’Professor of Geometry’ [^10]. The best illustration of his synthetical point of view is shown by the words (here translated from French) with which he begins his Memory (XII-2 “Mémoire sur une propriété générale d’une classe tres étendue de fonctions transcendantes.”): [^11] “ The transcendental functions considered until nowadays by the geometers are a very small number. Almost all the theory of transcendental functions is reduced to the one of logarithmic, exponential and circular functions, which are essentially one only kind. Only in recent times one has begun to consider some other functions. Among those, the elliptic transcendentals, about which M. Legendre developed so many elegant and remarkable properties, stay in the first rank.” The statement “are essentially one only kind” is the one we want now to comment upon. Algebraists like indeed short formulae, and these are in this case available. It suffices to consider the single formula : $exp (x) := \Sigma_{n=0}^{ \infty} \frac{ x^n} {n!} $, and then, by considering $ exp(ix)$ and the inverse functions of the ones we can construct by easy algebraic manipulations, we get easily ahold of the wild proliferation of functions which for instance occupy the stage of the U.S.A. Calculus courses (sin, cos, sec, cosec, tg, cotg, and their hyperbolic analogues). No doubt, synthesis is a peculiarity of pure mathematics, and applied or taught mathematics may perhaps need so many different names and functions: but, pretty sure, Abel stood by the side of synthesis and conciseness. His well known saying, that he was able to learn so rapidly because he had been: “ studying the Masters and not their pupils”, is quite valid nowadays. Today there is certainly an inflation of books and divulgations, many are second, third hand or even further. Abel’s point of view should be seriously considered by some pedagogists who want to strictly regulate children’s learning, forcing them to study n-th hand knowledge. Perhaps this is a strictly democratic principle, by which one wants to prevent some children from becoming precociously wise (as Abel did), and possibly try to stop their intellectual growth (this might be part of a more general ambitious program, sponsored by Television Networks owners). In any case, Abel had read the masters, and he knew many functions: he still belongs to the mathematical era where functions are just concretely defined objects and not subsets of a Cartesian product satisfying a geometrical condition. One of Abel’s main contributions was to consider not only ample classes of functions defined by integrals of algebraic functions, but to study their inverse functions and their periodicity (the so called elliptic functions being among the latter). At his time did not yet exist the concept of “Willkürliche Funktionen” (= “arbitrary functions”), quite central e.g. for Weierstrass, Dini, Peano and Hilbert (cf. [@dini], [@weier], [@g-p] ) and which motivated much of the developments in the theory of sets leading to the construction of several pathological situations (as Lebesgue’s non constant function with derivative almost everywhere zero, [@leb] ). Most of the functions he considered were in fact written as $\int_{x_0}^x y(t) dt $, where $y(x)$ is an [algebraic function]{} of $x$, which simply means that the function is defined on some interval in ${\ensuremath{\mathbb{R}}}$ and that there exists a polynomial $ P(x,y) \in {\ensuremath{\mathbb{C}}}[x,y]$ [^12] such that $ P(x,y(x)) \equiv 0.$ The functions given by these integrals, or by sums of several of these, are nowadays called [Abelian functions]{}. The above statement is by and large true, with however a single important exception, concerning Abel’s treatment of functional equations: there he considers quite generally the functions which occur as solution of certain [functional equations]{}. As example, we take the content of an article also considered in Houzel’s talk, VI-Crelle I (1826). This article is a real gem: it anticipates S.Lie’s treatment of Lie group germs, and yields actually a stronger result (although, under the assumption of commutativity): Let $f : U {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{R}}}$ be a germ of function defined on a neighbourhood of the origin, $ {\ensuremath{\mathbb{R}}}^2 \supset U \ni 0 $, such that $$f (z, f(x,y))$$ is symmetric in $x,z,y$ (i.e., in today’s terminology, we have an Abelian Lie group germ in 1-variable). Then there exists a germ of change of variable $ \psi: ({\ensuremath{\mathbb{R}}},0) {\ensuremath{\rightarrow}}({\ensuremath{\mathbb{R}}},0) $ such that our Lie group becomes $({\ensuremath{\mathbb{R}}}, +)$, or, more concretely, such that $$\psi (f(x,y)) = \psi(x) + \psi(y).$$ Among the Masters’ work which Abel studied was certainly, as already mentioned, M. Legendre and his theory of elliptic integrals. For these, already considered by Euler and Lagrange, Legendre devised a [normal form]{} (here $R(x)$ is a rational function of $x$): $$\int_{x_0}^t \frac{R(x)} {\surd \overline{ (1 - x^2) ( 1- k^2 x^2) } } dx.$$ In the remarkable paper XVI-1 (published on Crelle, Bd. 2,3 (1827, 1828)), entitled “Recherches sur les fonctions elliptiques”, Abel, as we already mentioned , writes clearly, after observing that the study of these elliptic integrals can be reduced to the study of integrals of the first, second and third kind [^13] $$\int \frac{ d \theta} {\surd \overline{ 1 - c^2 sin^2 \theta } }; \int d \theta \surd \overline{ 1 - c^2 sin^2 \theta } , \int \frac{ d \theta } { ( 1 + n sin^2 \theta) \surd \overline{ 1 - c^2 sin^2 \theta } },$$ " These three functions are the ones that M. Legendre has considered, especially the first, which enjoys the most remarkable and the simplest properties. I am proposing myself, in this Memoir, to consider the inverse function, i.e., the function $\phi (a)$ , determined by the equations $$a = \int \frac{ d \theta} {\surd \overline{ 1 - c^2 sin^2 \theta } }$$ $$sin \theta = \phi (a) = x ."$$ In the following pages , where he manages to give a simple proof of the double periodicity of the given function $\phi$, Abel shows the clarity of his geometric intuition. He simply observes that in Legendre’s normal form one should win the natural resistance to consider non real roots, and actually it is much better to consider the case where $ k^2 < 0$, and thus he considers (his notation) the integral $$\int_{x_0}^t \frac{1} {\surd \overline{ (1 - c^2 x^2) ( 1+ e^2 x^2) } } dx,$$ where $c,e$ are strictly positive real numbers. The roots of the radical are the points $ \pm 1/c , \pm (1/e) \surd \overline{-1}$, we have a rectangular symmetry around the origin and we have two periods $ \omega , \tilde{ \omega }$ obtained by integrating on the two closed paths lying over the segments joining pairs of opposite roots: $$\omega = 4 \int_{0}^{1/c} \frac{1} {\surd \overline{ (1 - c^2 x^2) ( 1+ e^2 x^2) } } dx ,\ \tilde{ \omega } = 4 \int_{0}^{i/e} \frac{1} {\surd \overline{ (1 - c^2 x^2) ( 1+ e^2 x^2) } } dx.$$ It is straightforward to observe that the two periods (of the [real curve]{}) are such that $\omega \in {\ensuremath{\mathbb{R}}}$, respectively $\tilde{ \omega } \in i {\ensuremath{\mathbb{R}}}$, thus they are linearly independent over the real numbers. The conclusion is that the inverse function is [doubly periodic]{} with one real and one imaginary period, so that its fundamental domain is a rectangle with sides parallel to the real, resp. imaginary axis. Since every elliptic integral of the first kind can be reduced to this form, their inverse functions are all doubly periodic (unlike the circular functions, which possess only one period). Abel does not bother to highlight the geometry underlining his argument, it is clear that he has developed a good geometrical intuition, but his style is extremely terse and concise. This conciseness becomes almost abrupt in the other article XIII-Vol. 2 [^14], “Theorie des transcendantes elliptiques”. This long memoir starts nulla interposita more (it was probably unfinished): “ For more simplicity I denote the radical by $\surd \overline{R}$, whence we have to consider the integral $$\int \frac{P dx} {\surd \overline{R} },$$ $P$ denoting a rational function of $x$.” It is divided into three chapters , the first devoted to the reduction of elliptic integrals by means of algebraic functions, the second to the reduction of elliptic integrals by means of logarithmic functions, and finally the third is entitled " A remarkable relation which exists among several integrals of the form $$\int \frac{dx} {\surd \overline{R} },\int \frac{x dx} {\surd \overline{R} }, \int \frac{x^2 dx} {\surd \overline{R} }, \int \frac{dx} { (x-a)\surd \overline{R} }.$$ The memoir contains the explicit discussions of several concrete problems concerning such reductions, and contains for many of those problems explicit references to Legendre. It looks to me a rather early work, because we directly see the influence of the study of Legendre, but rather important for two reasons. The first reason, already clear from the title of chapter III, is that Abel here for the first time considers the question of the relations holding among sums of elliptic integrals. This problem will be considered more generally for all algebraic integrals of arbitrary genus $g$ in his fundamental Memoir XII-1, entitled “Mémoire sur une propriété générale d’une classe tres -étendue de fonctions transcendantes”, presented on october 10 1826 to the Académie des Sciences de Paris, and published only in 1841. The main theorem of the latter was so formulated: “ If we have several functions whose derivatives can be roots of the same algebraic equation \[if $y(x)$ is an algebraic function of $x$, i.e., there is a polynomial $P$ such that $P(x,y(x)) \equiv 0$, then for each rational function $f(x,y)$ there is a polynomial $F(x,y)$ such that $F(x, f(x,y(x))) \equiv 0$\][^15], with coefficients rational functions of one variable \[x\], one can always express the sum of an arbitrary number of such functions by means of an algebraic and a logarithmic function, provided that one can establish among the variables of these functions a certain number of algebraic relations”. The first theorem is then given through formula (12): $$\int f(x_1,y_1) dx_1 + \int f(x_2,y_2) dx_2 + \dots \dots + \int f(x_{\mu},y_{\mu}) dx_{\mu} = v [(t_1, .. t_k)] \ :$$ here $f(x,y)$ is a rational function, we take the $\mu$ points which form the complete intersection of \[$P(x,y) = 0$\] and \[$G_t(x,y) = 0$\] where $G_t$ depends rationally upon the parameter $ t = (t_1, .. t_k)$ , and the conclusion is, as we said, that $v$ is the sum of a rational and of a logarithmic function. Abel also explains clearly in the latter memoir that the number of these relations is a number, which later on was called the genus of the curve $C$ birational to the plane curve of equation $P(x,y) = 0$. The way we understand the hypothesis of the theorem nowadays is through the geometric condition: if the Abel sum of these points is constant in the Jacobian variety of $C$. I will come back to the geometric interpretations in the next section, let me now return to the second reason of importance of the cited Memoir XIII, 2. For instance, in Chapter I, Abel gives very explicit formulae, e.g. for the reduction of integrals of the form $$\int \frac{x^m dx} {\surd \overline{R} },$$ where $R(x)$ is a polynomial of degree $3,4$, to the integrals $$\int \frac{dx} {\surd \overline{R} },\int \frac{x dx} {\surd \overline{R} }, \int \frac{x^2 dx} {\surd \overline{R} }.$$ The above integral is calculated by recursions, starting from the equation $$d (Q \surd\overline{R}) = S \frac{ dx} {\surd \overline{R} } ,$$ and writing explicitly $$S = \phi(0) + \phi(1) x + \dots + \phi(m) x^m$$ $$Q = f(0) + f(1) x + \dots + f(m-3) x^{m-3} .$$ This is an example of Abel’s mastery in the field of [Differential Algebra]{}. Although the modern reader, as well as Sylow and Lie, may underscore the impact of these very direct calculations, it seems to me that there has been a resurgence of this area of mathematics, especially in connection with the development of computer algorithms and programs which either provide an explicit integration of a given function by elementary functions [^16], or decide that the given function does not admit an integration by elementary functions ( this problem was solved by Risch, and later concrete decision procedures and algorithms were given by J Davenport and Trager [@risch], [@dav79-1], [@dav79-2],[@dav81], [@trag79]). Differential algebra is also the main tool in the article (XVII-2) “ Mémoire sur les fonctions transcendantes de la forme $\int y \ dx$ ou $y$ est une fonction algébrique de $x$.” This paper looks very interesting and somehow gave me the impression (or at least I liked to see it in this way) of being a forerunner of the applications of Abelian integrals to questions of transcendence theory. This time I will state the main theorem by slightly altering Abel’s original notation Assume that $\phi$ is a \[non trivial\] polynomial $$\phi (r_1, \dots , r_{\mu}, w_1, \dots w_p)$$ and that $\phi \equiv 0$ if we set $$r_i = \int y_i(x)\ dx, \ w_j = u_j (x),$$ where $y_i, u_j$ are algebraic functions. Then there is a \[non trivial\] linear relation $ \Sigma_i c_i\int y_i(x)\ dx = P(x)$, with constant coefficients $c_i$ and with $P$ an algebraic function. Let $y_i(x)\ dx$ be linearly independent Differentials of the I Kind on a Riemann surface. Then the respective integrals are algebraically independent. Nowadays, we would use the periodicity of these functions on the universal cover of the algebraic curve, or the order of growth of the volume of a periodic hypersurface (this argument was the one used later by Cousin, cf. [@cous02]) to infer that the polynomial must be linear (vanishing on the complex linear span of the periods). Abel’s argument is instead completely algebraic in nature, he chooses in fact the polynomial $\phi$ to be of minimal degree with respect to $r_{\mu}$, and applies $d / dx$ to the relation in order to obtain (Abel’s notation)[^17] $$\Sigma_j \phi ' (r_j)\ y_j + \phi ' (x) = 0.$$ Abel shows then that $\phi$ has degree $1$ in $r_{\mu}$, since writing $$R = r_{\mu}^k + P_0 r_{\mu}^{k-1} + P_1 r_{\mu}^{k-2} + \dots = 0$$ he gets: $$0 = \frac{d}{dx} (R) = r_{\mu}^{k-1} ( k y_{\mu} + P_0') + \{ \dots \} r_{\mu}^{k-2} + \dots = 0$$ whence, by the minimality of the degree $k$, this polynomial in $r_{\mu}$ has all coefficients identically zero, in particular $ ( k y_{\mu} + P_0') \equiv 0 $, therefore $$r_{\mu} = \int y_{\mu} dx = - (1/k) P_0$$ is the desired degree one relation. By induction Abel derives the full statement that $\phi$ is linear in $r_1, \dots , r_{\mu}$. The geometrization of Abel’s methods. ====================================== The process of geometrization and of a deeper understanding of Abel’s discoveries went a long way, with alternate phases, for over 150 years. We believe that [^18] a fundamental role for the geometrization was played by the italian school of algebraic geometry, which then paved the way for some of the more abstract developments in algebraic geometry. Although it was very depressing for Abel that his fundamental Memoir XII-1 was not read by A. Cauchy (this is the reason why it took more than 15 years before it was published), still in Berlin Abel found the enthusiastic support of L. Crelle, who launched his new Journal by publishing the articles of Abel and Jacobi. Recall that finally fame and recognition were reaching Abel through the offer of a professorship in Berlin, which, crowning the joint efforts of Crelle and Jacobi, arrived however a few days after Abel had died. Especially inspired by the papers of Abel was Jacobi, in Berlin, who also wrote a revolutionary article on elliptic function theory, entitled “Fundamenta nova theoriae functionum ellipticarum”. It was Jacobi who introduced the words ’Abelian integrals’, ’Abelian functions’: Jacobi ’s competing point of view (few competitions however were so positive and constructive in the history of mathematics) started soon to prevail. Jacobi introduced the so called [elliptic theta functions]{}, denoted $\theta_{00}, \theta_{01},\theta_{10},\theta_{11},$ (cf. e.g. [@tric] or [@mum3]) and expressed the elliptic functions, like Abel’s $\phi$, inverse of the elliptic integral of the first kind, as a ratio of theta functions. A much more general definition of [theta-series]{} (the expression was later coined by Rosenhain and Göpel, followers of Riemann) was given by B. Riemann, who defined his [Riemann Theta function]{} as the following series of exponentials $$\theta (z, \tau) : = \Sigma_{n \in {\ensuremath{\mathbb{Z}}}^g} \ exp ( 2 \pi \ i [ \frac{1}{2} \ ^t n \tau n + ^t n z ])$$ where $ z \in {\ensuremath{\mathbb{C}}}^g$, $\tau \in \mathcal H_g = \{ \tau \in Mat (g,g, {\ensuremath{\mathbb{C}}}) \| \tau = ^t \tau, Im (\tau) \ {\rm is \ positive \ definite} \}.$ The theta function converges because of the condition that $Im (\tau) $ is positive definite, it admits ${\ensuremath{\mathbb{Z}}}^g$ as group of periods, being a Fourier series, and it has moreover a $\tau {\ensuremath{\mathbb{Z}}}^g$-quasi-periodicity which turns out to be the clue for constructing $2g$-periodic meromorphic functions as quotients of theta series. With a small variation (cf. [@mum3]) one defines the theta-functions with characteristics $$\theta [a, b](z, \tau) : = \Sigma_{n \in {\ensuremath{\mathbb{Z}}}^g} \ exp ( 2 \pi \ i [ \frac{1}{2} \ ^t (n+a) \tau (n+a) + ^t (n+a) (z+b) ]),$$ and the Jacobi functions $ \theta_{a, b}(z, \tau)$ are essentially the functions $ \theta [a/2, b/2](2 z, \tau).$ Beyond the very explicit and beautiful formulae, what lies beyond this apparently very analytic approach is the [pioneeristic principle]{} that any [meromorphic function f]{} on a complex manifold $X$ can be written as $$f = \frac{\sigma_1}{ \sigma_2}$$ of two relatively prime sections of a unique [Line Bundle L ]{} on $X$. This formulation came quite long after Jacobi, but Jacobi’s work had soon a very profound impact. For instance, one of the main contributions of Jacobi was the solution of the [inversion problem]{} explicitly for genus $g = 2$. Concretely, Jacobi considered a polynomial $R(x)$ of degree $6$, and then, given the two Abelian integrals $$u_1 (x_1, x_2) := \int_{x_0}^{x_1} \frac{dx} { \surd \overline{R}} + \int_{x_0}^{x_2} \frac{dx} { \surd \overline{R}},$$ $$u_2 (x_1, x_2) := \int_{x_0}^{x_1} \frac{x dx} { \surd \overline{R}} + \int_{x_0}^{x_2} \frac{x dx} { \surd \overline{R}},$$ he found that the two symmetric functions $$s_1 := x_1 + x_2 , s_2 := x_1 x_2 ,$$ are $4$-tuple periodic functions of $u_1, u_2$. Under the name [Jacobi inversion problem]{} went the generalization of this result for all genera $g$, and the solution to the Jacobi inversion problem was one of the celebrated successes of Riemann. Nowadays the result is formulated as follows: given a compact Riemann surface $C$ of genus $g$, let $\omega_1, \dots \omega_g$ be a basis of the space $H^0(\Omega^1_C)$ of holomorphic differentials on $C$ [adapted]{} to a [symplectic]{} basis $\alpha_1, \dots \alpha_g, \beta_1, \dots \beta_g$ for the Abelian group of closed paths $H_1(C, {\ensuremath{\mathbb{Z}}})$: this means that $$( \int_{\alpha_i} \omega_j )= (\delta_{i,j}) = I_g , (\int_{\beta_i} \omega_j ) = (\tau_{i,j}),$$ where $I_g$ is the Identity $(g \times g)$ Matrix and $\tau \in \mathcal H_g$, and that the intersection matrices satisfy $(\alpha_i , \alpha_j) = 0, (\beta_i , \beta_j) = 0, (\alpha_i , \beta_j) = I_g.$ Then the [Abel-Jacobi]{} map $ C^g \rightarrow {\ensuremath{\mathbb{C}}}^g / ({\ensuremath{\mathbb{Z}}}^g + \tau {\ensuremath{\mathbb{Z}}}^g) : = Jac (C)$, associating to the $g$-tuple $P_1, \dots P_g$ of points of $C$ the sum of integrals (taken modulo $({\ensuremath{\mathbb{Z}}}^g + \tau {\ensuremath{\mathbb{Z}}}^g) $) $$a_g(P_1, \dots P_g) := \int_{P_0}^{P_1} (\omega) + \dots \int_{P_0}^{P_g} (\omega) ,$$ ($\omega$ being the vector with $i$-th component $\omega_i$), is [surjective]{} and yields a [birational map]{} of the [symmetric product]{} $$C^{(g)} := Sym^g (C) := C^g / {\ensuremath{\mathcal{S}}}_g,$$ ${\ensuremath{\mathcal{S}}}_g$ being the symmetric group of permutation of $g$ elements, onto the [Jacobian Variety of C]{} $$\ \ Jac (C) := {\ensuremath{\mathbb{C}}}^g / ({\ensuremath{\mathbb{Z}}}^g + \tau {\ensuremath{\mathbb{Z}}}^g).$$ We see the occurrence of a matrix $\tau$ in the so called Siegel upper half space $\mathcal H_g$ of symmetric matrices with positive definite imaginary part: this positive definiteness ensures the convergence of Riemann’s theta series, and indeed Riemann used explicitly his theta function to express explicitly the symmetric functions of the coordinates of a $g$-tuple $P_1, \dots P_g$ of points of $C$ as rational functions of theta factors. Today, we tend to forget about these explicit formulae, and we focus our attention to the geometric description of the Abel-Jacobi maps for any $n$-tuple of points of $C$ in order to grasp the power of the discoveries of Abel, Jacobi and Riemann. The [Abel-Jacobi]{} maps $ C^n \rightarrow {\ensuremath{\mathbb{C}}}^g / ({\ensuremath{\mathbb{Z}}}^g + \tau {\ensuremath{\mathbb{Z}}}^g) : = Jac (C)$, given by $$a_n(P_1, \dots P_n) := \int_{P_0}^{P_1} (\omega) + \dots \int_{P_0}^{P_n} (\omega) ,$$ and which naturally factor through the symmetric product $C^{(n)}$, enjoy the following properties (cf. the nice and concise lecture notes by D. Mumford [@mum74] for more on this topic, and also [@corn] for a clear modern presentation) - \(1) [Abel’ s Theorem:]{} The fibres are projective spaces corresponding to [linearly equivalent]{} divisors, i.e. $ a_n(P_1, \dots P_n) = a_n(Q_1, \dots Q_n)$ if and only if there is a rational function $f$ on $C$ with polar divisor $ P_1 + \dots + P_n$ and divisor of zeros $ Q_1 + \dots + Q_n.$ - \(2) For $n = g-1$ we have that the image of $a_{g-1}$ equals, up to translation, the hypersurface $\Theta \subset Jac (C)$ whose inverse image in ${\ensuremath{\mathbb{C}}}^g$ is given by the vanishing of the Riemann theta function $\Theta= \{ z\| \theta (z, \tau) = 0 \}.$ - \(3) For $n = g$ we have that $a_{g}$ is onto and birational. - \(4) For $n \geq 2g-1$ we have that $a_{n}$ is a fibre bundle with fibres projective spaces ${\ensuremath{\mathbb{P}}}^{n-g}$. - \(5) For n=1 we have an embedding of the curve $C$ inside $Jac(C)$, and in such a way that $C$ meets a general translate of the theta divisor $\Theta$ in exactly $g$ points. And in this way one gets an explicit geometrical description of the inverse to the Abel Jacobi map $a_g$. Indeed, to a point $ y \in Jac (C)$ we associate the $g$-tuple of points of $C$ given by the intersection $C \cap (\Theta + y)$ (here we think of $C$ $\subset Jac (C)$ under the embedding $a_1$). Many of the above properties are very special and lead often to a characterization of curves among algebraic varieties. The above formulations are moreover the fruit of a very long process of maturation whose evolution is not easy to trace. For instance, when did the concept of the [Jacobian variety]{} of a curve ( Riemann surface) $C$ make its first appearance? This notion certainly appears in the title of the papers by R. Torelli in 1913 ([@tor]) but apparently [^19] the name was first used by F. Klein and became very soon extremely popular. Observe however that in the classical treatise by Appell and Goursat ([@appgour]), dedicated to analytical functions on Riemann surfaces, and appeared first in 1895, although Jacobi’s inversion theorem is amply discussed, no Jacobian or whatsoever variety is mentioned. The French school of Humbert, Picard, Appell and Poincaré was very interested about the study of the so called “hyperelliptic varieties”, generalization of the elliptic curves in the sense that they were defined as algebraic varieties $X$ of dimension $n$ admitting a surjective entire holomorphic map [^20] $ u: {\ensuremath{\mathbb{C}}}^n {\ensuremath{\rightarrow}}X$ . Among those are the so called [Abelian Varieties]{} [^21], which are the projective varieties which have a structure of an algebraic group. In particular, Picard proved a very nice result in dimension $d= 2$, which was observed by Ciliberto (cf. his article in the present volume, also for related historical references) to hold quite generally. We want to give here a simple proof of this result, which we found during the Conference, and which makes clear one basic aspect in which the higher dimensional geometry has a different flavour than the theory of curves [^22]: namely there is no ramification in passing from Cartesian to symmetric products. Let $X$ be an algebraic variety of dimension $ d \geq 2$ and assume that there is a natural number $n$ such that the $n$-th symmetric product $X^{(n)}$ is birational to an Abelian variety $A$. Then $\ n=1$ (whence, $X$ is birational to an Abelian variety). Moreover, this result ilustrates another main difference between the geometry of curves and the one of higher dimensional varieties $X$: the latter has a quite different flavour, because only seldom one can resort to the help of subsidiary Abelian varieties for the investigation of a higher dimensional variety $X$. [[Proof. ]{}]{}W.l.o.g. we may assume that $X$ be smooth. Let us consider the projection $\pi$ of the Cartesian product onto the symmetric product, and observe that $\pi : X^n {\ensuremath{\rightarrow}}X^{(n)}$ is unramified in codimension $1$. We get a rational map $f : X^n {\ensuremath{\rightarrow}}A$ by composing $\pi$ with the given birational isomorphism, and then we observe that every rational map to an Abelian variety is a morphism. It is moreover clear from the construction that $f$ is not branched in codimension $1$: in particular, it follows that $X^n$ is birational to an unramified covering of $A$, whence $X^n$ is birational to an Abelian variety. Let us introduce now the following notation: for a smooth projective variety $Y$ we consider the algebra of global holomorphic forms $H^0(\Omega_Y^): = \oplus_{i=0, \dots dim (Y)} H^0(\Omega_Y^i) $ ([holomorphic algebra]{}, for short). This graded algebra is a birational invariant, and for an Abelian variety $A$ it is the free exterior algebra over $ H^0(\Omega_A^1) $. Now, the holomorphic algebra $H^0(\Omega_{X^n}^)$ of a Cartesian product $X^n$ is the tensor product of $n$ copies of the holomorphic algebra $H^0(\Omega_{X}^)$ of $X$. Denote $H^0(\Omega_{X}^)$ by $B$: then we reached the conclusion that $B^{\otimes n}$ is a free exterior algebra over its part of degree $1$, $B_1^1 \oplus B_2^1 \dots \oplus B_n^1$. It follows that also $B$ is a free exterior algebra (i.e., $X$ enjoys the property that its holomorphic algebra $H^0(\Omega_{X}^)$ is a free exterior algebra over $H^0(\Omega_{X}^1)$). Moreover, the holomorphic algebra of the symmetric product is the invariant part of this tensor product (for the natural action of the symmetric group in $n$-letters ${\ensuremath{\mathcal{S}}}_n$), and by our assumption $C:= (B^{\otimes n})^{{\ensuremath{\mathcal{S}}}_n} $ is also a free exterior algebra. However, $C^1 = B^1$, and $r:= dim (B^1)$ is also the highest degree $i$ such that $ B^i \neq 0$. But then $C^{nr} \neq 0$, contradicting the property that $C$ is a free exterior algebra with $ dim \ (C^1) = r$, if $n \neq 1$. Thus $n=1$, and $X$ is birational to an Abelian variety. In the case of dimension $d=1$, the same algebraic arguments easily yield that the n-th symmetric product of a curve $C$ is not birational to an Abelian variety if $n \neq g$, $ g:= dim \ (H^0(\Omega^1_C))$. Once more, the algebra of differential forms, as in Abel’s work, has played the pivotal role. The importance of this algebra was observed also by Mumford ([@mum68]) who used it to show that on an algebraic surface $X$ with $H^0(\Omega^2_X) \neq 0$, the group of $0$-cycles ( Sums $\Sigma _im_i P_i$ of points $P_i \in X$ with integer multiplicities $m_i \in {\ensuremath{\mathbb{Z}}}$) modulo rational equivalence is not finite dimensional, contrary to the hope of Severi, ( D. Mumford sarcastically wrote: “One must admit that in this case the [technique]{} of the italians was superior to their vaunted intuition” [^23]) who unfortunately was basing his proposed theory on a wrong article ([@sev32], where not by chance the error was an error of ramification). It must be again said that the italian school, and especially Castelnuovo, gave a remarkable impetus to the geometrization of the theory of Abelian varieties. This approach, especially through the work of Severi, influenced Andre’ Weil who understood the fundamental role of Abelian varieties for many questions of algebraic number theory. Weil used these ideas to construct ([@weilva]) the Jacobian variety of a curve as a quotient of the symmetric product $C^{(g)}$, and then, for a $d$-dimensional variety with $d \geq 2$, the Albanese variety $Alb(X)$ as a quotient (in the category of Abelian varieties) of the Jacobian $J(C)$ of a sufficiently general linear section $ C = X \cap H_1 \cap H_2 \dots \cap H_{d-1}$. It must be however said that also the later geometric constructions were deeply influenced by the bilinear relations which Riemann, through a convenient dissection of his Riemann surface $C$, showed to hold for the periods of the Abelian integrals of the first kind of $C$. Nowadays, the usual formulation is (according to Auslander and Tolimieri, [@a-t] pages 267 and 274, the first formulation is essentually due to Gaetano Scorza in [@scor16], while the second is essentially due to Hermann Weyl in [@Weyl34], [@Weyl36], with refinements from A. Weil’s book [@weilVK]) Let $\Gamma$ be a discrete subgroup of a complex vector space $V$, such that the quotient $V/\Gamma$ is compact (equivalently, $\Gamma \otimes {\ensuremath{\mathbb{R}}}\cong V$): then we say that the complex torus $V/\Gamma$ satisfies the two Riemann bilinear relations if - I\) There exists an alternating form $ A : \Gamma \times \Gamma {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{Z}}}$ such that $A$ is the imaginary part of an Hermitian form $H$ on $V$ - II\) $H$ is positive definite. Or, alternatively, a complex structure on $\Gamma \otimes {\ensuremath{\mathbb{C}}}$, i.e., a decomposition $\Gamma \otimes {\ensuremath{\mathbb{C}}}= V \oplus \bar{V}$ and an element $ A \in \Lambda^2 (\Gamma)^{\vee}$ yield a polarized Abelian variety if the component of $A$ in $\Lambda^2 (V)^{\vee} \subset \Lambda^2 (\Gamma \otimes {\ensuremath{\mathbb{C}}})^{\vee}$ is zero and then its component in $(V)^{\vee} \otimes (\bar{V})^{\vee}$ is a positive definite Hermitian form. [^24] The basic theorem characterizing complex Abelian varieties is however due to Henri Poincaré ([@poi84], [@poi02]) who proved the linearization of the system of exponents, i.e., the more difficult necessary condition in the theorem, by an averaging procedure (integrating the ambient Hermitian metric of $ X {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{P}}}^N$ with respect to the translation invariant measure of $X = V/\Gamma$, he obtained a translation invariant Hermitian metric). A complex torus $X = V/\Gamma$ is an algebraic variety if and only if the two Riemann bilinear relations hold true for $V/\Gamma$. Both conditions are equivalent to the existence of a meromorphic function $f$ on the complex vector space $V$ whose group of periods is exactly $\Gamma$ (i.e., $\Gamma = \{ v \in V\| f(z + v) \equiv f (z) \} ). $ Poincaré had an extensive letter exchange with Klein (cf. Klein’s Collected Works, where pages 587 to 621 of Vol. III are devoted to the “Briefwechsel” between the two, concerning the problem of uniformization, and their early attempts, which were based on a ’principle of continuity’ which was not so easy to justify [^25]), especially related to the study of discontinuous groups, acting not only on ${\ensuremath{\mathbb{C}}}^n$ as in the case of tori, but also on the hyperbolic upperhalf plane ${\ensuremath{\mathbb{H}}}$. The main result, whose complete proof was obtained in 1907 by a student of Klein, Koebe, and by Poincaré independently, was the famous uniformization theorem that again we state in its modern formulation for the sake of brevity. If a Riemann surface is not the projective line ${\ensuremath{\mathbb{P}}}^1_{{\ensuremath{\mathbb{C}}}}$, the complex plane ${\ensuremath{\mathbb{C}}}$, nor ${\ensuremath{\mathbb{C}}}^$ or an elliptic curve, then its universal covering is the (Poincaré) upper half plane ${\ensuremath{\mathbb{H}}}$. The reason why the upper half plane “belongs” to Poincaré is that Klein prefered to work with the biholomorphically equivalent model given by the unit disk ${\ensuremath{\mathbb{D}}}:= \{ z \in {\ensuremath{\mathbb{C}}}\| \|z\| < 1 \}.$ In this way Klein was capable of making us the gift of beautiful symmetries given by tesselations of the disk by fundamental domains for the action of very explicit Fuchsian groups (discrete subgroups $\Gamma$ of $PSU(1,1,{\ensuremath{\mathbb{C}}})$) with compact quotient ${\ensuremath{\mathbb{D}}}/ \Gamma$. To summarize the highlights of the turn of the century, when geometry was a very central topic, one should say that several new geometries came to birth at that time: but the new developments were based on new powerful analytic tools, which were the bricks of the new building. However, although the birth of differential geometry lead to new geometrical theories based on infinite processes where metric notions played a fundamental role, algebraic geometry went on with alternating balance between geometrical versus algebraic methods. Algebraization of the geometry ============================== At this moment, a witty reader, tired of the distinction “algebra”-“non algebra”, might also remind us that the popular expression “This is algebra for me” simply means: ’I do not understand a single word of this’. There is a serious point to it: the concept algebra is slightly ambiguous, and a very short formula could be not very inspiring without a thoroughful explanation of its meaning(s), and of all the possible consequences and applications. One of the best ways to understand a formula is for instance to relate it to a picture, to see it thus related to a geometrical or dynamical process. Needless to say, the best example of such an association is the Weierstrass equation of a plane cubic curve $$C^1_3 = \{ (x,y,z) \in {\ensuremath{\mathbb{P}}}^2_{{\ensuremath{\mathbb{C}}}} \| y^2 z = 4 x^3 - g_2 x z^2 - g_3 z^3 \}.$$ To this equation we immediately associate the picture yielding the group law of $C=C^1_3$, i.e., a line $L$ intersecting $C$ in the three points $P, Q$ and $T= - (P+Q)$. Where have we seen this picture first?[^26] Well, in my case, I (almost) saw it first in the book by Walker on Algebraic curves, exactly in the last paragraph, in the section 9.1 entitled “Additions of points on a cubic”. The book was written in 1949, and if we look at books on algebraic curves written long before, the group law is not mentioned there. For instance, Coolidge’s book “A treatise on algebraic plane curves” has a paragraph entitled “elliptic curves” , pages 302-304, and the main theorems are first that an elliptic curve is birational to a plane cubic, and then the [Cross-Ratio Theorem]{} asserting that if $P\in C^1_3 \subset {\ensuremath{\mathbb{P}}}^2$ is any point, through $P$ pass exactly $4$ tangent lines, and their cross ratio is independent of the choice of $P \in C$. The Weierstrass equation, and the explanation that the cubic curve is uniformized through the triple $(1,{\ensuremath{\mathcal{P}}}, {\ensuremath{\mathcal{P}}}')$, where ${\ensuremath{\mathcal{P}}}$ is the Weierstrass function, comes later, as due after the Riemann Roch theorem [^27], on pages 363-367 in the paragraph “Curves of genus 1”. Going to important textbooks of the Italian tradition, like Enriques and Chisini’s 4 Volumes on the “Lezioni sulla teoria geometrica delle equazioni e delle funzioni algebriche”[^28] we see that Volume IV contains the Book 6, devoted to “Funzioni ellittiche ed Abeliane”. Here Abel’s theorem is fully explained, and on page 77 we see Abel’s theorem for elliptic curves, on page 81 the addition theorem for the ${\ensuremath{\mathcal{P}}}$ function of Weierstrass: the geometry of the situation is fully explained, i.e., that three collinear points sum to zero in the group law given by the sum of Abelian integrals of the first kind $$u_1 + u_2 + u_3 = \int_{x_0}^{x_1} \frac{dx}{y} + \int_{x_0}^{x_2} \frac{dx}{y} +\int_{x_0}^{x_3} \frac{dx}{y} = 0.$$ It is also observed that the inverse of the point $P = (1,x,y)$ is the point $P = (1,x,-y)$, as a consequence of the fact that ${\ensuremath{\mathcal{P}}}$ is an even function. This Book 6 is clearly influenced by Bianchi’s Lecture Notes “ Lezioni sulla teoria delle Funzioni di variabile complessa”[^29] whose Part 2 is entirely devoted to “Teoria delle funzioni ellittiche”, and in the pages 315-322 the addition theorem for ${\ensuremath{\mathcal{P}}}$ is clearly explained, moreover “Alcune applicazioni geometriche” are given in the later pages 415-418. So, the picture is there, is however missing the wording: a plane cubic is an Abelian group through the sum obtained via linear equivalence of divisors, namely, the sum of three points $P,Q,T$ is zero if and only if the divisor $P+Q+T$ is linearly equivalent to a fixed divisor $D$ of degree 3 (in the Weierstrass model, $D$ is the divisor $3O$, $O$ being the flex point at infinity). As it was explained to me by Norbert Schappacher (cf. [@schap90]), the works of Mordell and Weil in the 1920’s are responsible of this new wording and perspective. In fact, these authors considered a cubic curve $C$ whose equations has coefficients in a field $K$, and noticed that the set $C(K)$ of $K$-rational points, i.e., the points whose coordinates are in $K$, do indeed form a subgroup. For this they did not need that one flex point should be $K$-rational, since essentially, once we have a $K$-rational point $O$, we can reembed the elliptic curve $C$ by the linear system $\|3O\|$, and then obtain a new cubic $C'$ whose $K$ rational points are exactly those of $C$. Through these works started an exciting new development, namely the geometrization of arithmetic, which was one of the central developments in the 20-th century mathematics. For instance, the theory of elliptic curves over fields of finite characteristic was (cf[@weil29]) built by Weil and then Tate who (cf. the quite late appearing in print [@tate]), starting from the Weierstrass equation, slightly modified into $$y^2 z = x^3 - p_2 x z^2 - p_3 z^3$$ started to construct analogues of the theory of periods.[^30] Going back to Bianchi, it is Klein’s and Bianchi’s merit to have popularized the geometric picture of elliptic curves, and actually Bianchi went all the way through in some of his papers to describe the beautiful geometry related to the embeddings of elliptic curves as non degenerate curves of degree $n$ in ${\ensuremath{\mathbb{P}}}^{n-1}$ ($\forall n \geq 3$). It took however quite long till a purely algebraic interpretation of Weierstrass’ equation made its way through. Nowadays we would associate to an elliptic curve $C$ and to a point $O \in C$ the ${\ensuremath{\mathbb{N}}}$-graded ring $$\bigoplus_{m=0}^{\infty} H^0 (C, {\ensuremath{\mathcal{O}}}_C(mO)): = {\ensuremath{\mathcal{R}}}(C, {\ensuremath{\mathcal{L}}})$$ where $C$ is defined over a (non algebraically closed) field $K$ of characteristic $\neq 2$, $O$ is a $K$-rational point and ${\ensuremath{\mathcal{L}}}:={\ensuremath{\mathcal{O}}}_C(O)$. As a consequence of the Riemann Roch theorem we obtain the following $${\ensuremath{\mathcal{R}}}(C, {\ensuremath{\mathcal{L}}}) \cong K [u,\xi, \eta ] / ( \eta^2 - \xi^3 + p_2 \xi u^4 + p_3 u^6 ),$$ where $deg(u)=1$, $deg(\xi)=2$,$deg(\eta)=3$, $div(u) = O$. To go back to the original Weierstrass equations it suffices to observe that $${\ensuremath{\mathcal{P}}}= \xi / u^2, {\ensuremath{\mathcal{P}}}' = \eta/u^3 ,$$ and that $ x := \xi u, y := \eta , z := u^3 $ are a basis of the vector space $H^0 (C, {\ensuremath{\mathcal{O}}}_C(3O))$. One sees also clearly how the Laurent expansion at $O$ of ${\ensuremath{\mathcal{P}}}$ is determined by and determines $p_2, p_3$. Surprisingly, the following general problem is still almost completely open, in spite of a lot of research in this or similar directions Describe the graded ring ${\ensuremath{\mathcal{R}}}(A, {\ensuremath{\mathcal{L}}})$ for ${\ensuremath{\mathcal{L}}}$ an ample divisor on an Abelian variety, for instance in the case where ${\ensuremath{\mathcal{L}}}$ yields a principal polarization. Before explaining the status of the question, I would first like to explain its importance. Take for instance the case of an elliptic curve $C$ whose ring is completely described (the ring does not depend upon the choice of $O$ because we have a transitive group of automorphisms provided by translations for the group law of $C$). We want for instance to describe the geometry of the embedding of $C$ as a curve of degree $4$ in ${\ensuremath{\mathbb{P}}}^3$. We observe that, at least in the case where $K$ is algebraically closed, any such embedding is given by the linear system $\|4 O\|$, for a suitable choice of $O$. The coordinates of the map are given by a basis of the vector space $H^0(C, {\ensuremath{\mathcal{O}}}_C(4O))$, i.e., by 4 independent homogeneous elements of degree $4$ in our graded ring ${\ensuremath{\mathcal{R}}}(C, {\ensuremath{\mathcal{L}}})$. These are easily found to be equal to $s_0 := u^4, s_1 := u^2 \xi, s_2 :=\xi^2, s_3 :=u \eta$. Then we obviously have the two equations $$s_0 s_2 = s_1 ^2 , \ \ s_3^2 = s_1 s_2 + p_2 s_0 s_1 + p_3 s_0^2 ,$$ holding for the image of $C$ (the second is obtained by the “Weierstrass” equation once we multiply by $u^2$). These are all the equations, essentially by Bezout’s theorem, since $C$ maps to a curve of degree $4$ and $ 4 = 2 \times 2$. From an algebraic point of view, what we have shown is the process of determining a subring of a given ring, and the nowadays computer algebra programs like “Macaulay” have standard commands for this operation (even if sometimes the computational complexity of the process may become too large if one does not use appropriate tricks). Classically, a lot of attention was devoted to the geometric study of the maps associated to the linear systems $\| m \Theta\|$, where $\Theta$ is the divisor yielding a principal polarization of the given Abelian variety. For instance, the $4$-ic Kummer surface is the image in ${\ensuremath{\mathbb{P}}}^3$ of a principally polarized Abelian surface under the linear system $\| 2 \Theta\|$, yielding a $ 2 : 1$ morphism which identifies a point $v$ to $-v$, and blows up the $16$ $2$-torsion points to the $16$ nodal singularities of the image surface. There is a wealth of similar results, which can be found for instance in the books of Krazer [^31], Krazer-Wirtinger, and Coble ([@kraz], [@k-w], [@coble]), written in the period 1890-1926. The first books are directly influenced by the Riemann quadratic relations, i.e., linear relations between degree two monomials in theta functions with characteristics (and their coefficients being also products of “Thetanullwerte”, i.e., values in $0$ of such thetas with characteristics), and show an attempt to use geometrical methods starting from analytic identities. Coble’s book is entitled “Algebraic geometry and theta functions”, and is already influenced by the breakthrough made by Lefschetz in his important Memoir ([@lef]). Lefschetz used systematically the group law to show that if $s_1(v), \dots s_r(v) \in H^0(m \Theta)$ and we choose points $a_1, \dots a_r $ such that $ \Sigma_i a_i = 0$, then the product $s_1(v+a_1) \dots s_r(v + a_r) \in H^0(r m \Theta)$. Then he chooses sufficiently many and sufficiently general points $a_1, \dots a_r$ so that these sections separate points and tangent vectors. Thus Lefschetz proves in particular $\| m \Theta\|$ yields a morphism for $ m\geq 2$ and an embedding for $m \geq 3$. The direction started by Lefschetz was continued by many authors, notably Igusa, Mumford, Koizumi, Kempf, who proved several results concerning the equations of the image of an Abelian variety (e.g., that the image of $\| m \Theta\|$ is an intersection of quadrics for $m \geq 5$). I will later return on some new ideas related to these developments, I would like now to focus on the status of the problem I mentioned. The case where $A$ has dimension $g=1$ being essentially solved by Weierstrass, the next question is whether the answer is known for $g=2$. This is the case, since the description of the graded ring was obtained by A. Canonaco in 2001 ([@can02]); an abridged version of his result is as follows Let $A$ be an Abelian surface and $\Theta$ be an effective divisor yielding a principal polarization: then the graded ring ${\ensuremath{\mathcal{R}}}(A, {\ensuremath{\mathcal{O}}}_A(\Theta))$ has a presentation with $11$ generators, in degrees $ (1,2^3,3^5,4^2)$, and $37$ relations, in degrees $ (4,5^6,6^{17},7^{10}, 8^3)$. 1\) Canonaco gives indeed explicitly the $37$ relations, whose shape however is not always the same. One obtains in this way an interesting stratification of the Moduli space of p.p. Abelian surfaces. Does this stratification have a simple geometrical meaning in terms of invariant theory? The proof uses at a certain point some computer algebra aid, since the equations are rather complicated. Nevertheless, we would like to sketch the simple geometric ideas underlying the algebraic calculations, since, as one can easily surmise, they are related to the aforementioned Riemann’s developments of Abel’s investigations. [[Proof. ]{}]{}The key point is thus that, in the case where $\Theta$ is an irreducible divisor, $\Theta$ is isomorphic to a smooth curve $C$ of genus $2$ (the other case where $\Theta$ is reducible is easier, since then $A$ is a product of elliptic curves, and $\Theta$ is the union of a vertical and of a horizontal curve). One uses first of all the exact sequence (for $n\geq 2$, since for $n=1$ the right arrow is no longer surjective) $$0 {\ensuremath{\rightarrow}}H^0(A,{\ensuremath{\mathcal{O}}}_A( (n-1)\Theta)) {\ensuremath{\rightarrow}}H^0(A,{\ensuremath{\mathcal{O}}}_A( n\Theta)) {\ensuremath{\rightarrow}}H^0(\Theta,{\ensuremath{\mathcal{O}}}_{\Theta}( n\Theta)) {\ensuremath{\rightarrow}}0$$ and of the isomorphism $$H^0(\Theta,{\ensuremath{\mathcal{O}}}_{\Theta}( n\Theta)) \cong H^0(C,{\ensuremath{\mathcal{O}}}_C( n K_C)).$$ One relates thus our graded ring to the canonical ring of the curve $C$, which is well known, the canonical map of $C$ yielding a double covering of ${\ensuremath{\mathbb{P}}}^1$ branched on $6$ points. In more algebraic terms, there is a homogeneous polynomial $R(y_0,y_1)$ of degree $6$ such that $${\ensuremath{\mathcal{R}}}(C,{\ensuremath{\mathcal{O}}}_C( K_C) ) \cong K [y_0,y_1,z] / (z^2 - R(y_0,y_1)).$$ One can summarize the situation by observing that, if $ {\ensuremath{\mathcal{R}}}_A := {\ensuremath{\mathcal{R}}}(A, {\ensuremath{\mathcal{O}}}_A(\Theta))$, ${\ensuremath{\mathcal{R}}}_C := {\ensuremath{\mathcal{R}}}(C,{\ensuremath{\mathcal{O}}}_C( K_C) )$, then ${\ensuremath{\mathcal{R}}}_A$ surjects onto the subring ${\ensuremath{\mathcal{R}}}'$ of ${\ensuremath{\mathcal{R}}}_C$ defined by ${\ensuremath{\mathcal{R}}}' : = \bigoplus_{n\geq 2} H^0(C,{\ensuremath{\mathcal{O}}}_C( n K_C))$. To lift the ring structure of ${\ensuremath{\mathcal{R}}}'$, which is not difficult to obtain, to the ring structure of ${\ensuremath{\mathcal{R}}}_A$ we use again Abel’s theorem, i.e. the sequence of maps $$C^2 {\ensuremath{\rightarrow}}C^{(2)} {\ensuremath{\rightarrow}}A$$ where the last is a birational morphism contracting to a point the divisor $E$ consisting of the set of pairs $\{ (P, i(P))\| P \in C \}$ in the canonical system of $C$ (here $i$ is the canonical involution of the curve $C$). Letting $D_i$ be the pull back of a fixed Weierstrass point of $C$ under the $i$-th projection of $C \times C$ onto $C$, we obtain that $$H^0(A,{\ensuremath{\mathcal{O}}}_A( n\Theta)) \cong H^0(C^2,{\ensuremath{\mathcal{O}}}_{C^2}( n (D_1 + D_2 +E)))^+,$$ where the superscript $^+$ denotes the $+1$-eigenspace for the involution of $C^2$ given by the permutation exchange of coordinates. It also helps to consider that the $\Theta$ divisor of $A$ is the image of a vertical divisor $ \{P\} \times C$. Instead, the diagonal $\Delta_C$ of $C^2$ enters also in the picture because the pull back (under $h= \phi_K \times \phi_K : C \times C {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{P}}}^1 \times {\ensuremath{\mathbb{P}}}^1$, $\phi_K$ being the canonical map of $C$) of the diagonal of ${\ensuremath{\mathbb{P}}}^1 \times {\ensuremath{\mathbb{P}}}^1$ is exactly the divisor $\Delta_C + E$, whence $\Delta_C + E$ is linearly equivalent to $2 D_1 + 2 D_2$. We omit the more delicate parts of the proof, which are however based on the above linear equivalences and on the action of the dihedral group $ D_{4\times 2}$ on $C \times C$ (this is a lift, via the $({\ensuremath{\mathbb{Z}}}/2)^2$ Galois cover $h: C \times C {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{P}}}^1 \times {\ensuremath{\mathbb{P}}}^1$ of the permutation exchange of coordinates on $ {\ensuremath{\mathbb{P}}}^1 \times {\ensuremath{\mathbb{P}}}^1$). 1\) This approach should work in principle for the more general case of the Jacobian variety of a hyperelliptic curve. In fact, for each Jacobian variety we have the sequence of maps $$C^g {\ensuremath{\rightarrow}}C^{(g)} {\ensuremath{\rightarrow}}A$$ and the $\Theta$ divisor is the image of a big vertical divisor $ \{P\} \times C^{g-1}$. Again we have a $({\ensuremath{\mathbb{Z}}}/2)^g$ Galois cover $h : C^g {\ensuremath{\rightarrow}}({\ensuremath{\mathbb{P}}}^1)^g$ and a semidirect product of the Galois group with the symmetric group in $g$ letters ${\ensuremath{\mathcal{S}}}_g$ (the group of the $g$-dimensional cube). 2\) Another question is whether there does exist a more elegant, or just shorter presentation for the ring. [THE ALGEBRA OF THETA FUNCTIONS AND THE ALGEBRA OF REPRESENTATION THEORY]{} A more conceptual understanding of the several identities of general theta functions came through the work of Mumford ([@mum66-7] , cf. also [@igu] and [@weil64]). In Mumford’s articles and in Igusa’s treatise one finds a clear path set by choosing representation theory as a guide line, especially as developed by Weyl, Heisenberg and von Neumann. The basic idea is shortly said: let $G$ be a compact topological group, endowed henceforth with the (translation invariant) Haar measure $d \mu_G$. Consider then the vector space $V= L^2(G, {\ensuremath{\mathbb{C}}})$: then we have an action $\tau$ of the group $G$ on $V$ by translations, $\tau_{\gamma} (f )(g) := f (g \gamma ^{-1})$. Defining the group of characters $G^$ as $\ G^ := Hom (G, {\ensuremath{\mathbb{C}}}^)$ we have an action of $G^$ on $V$ given by multiplication $ \chi f (g) := \chi(g) f(g)$. The two actions fail to commute, but by very little, since $$\chi [\tau_{\gamma} (f )] (g) = \chi(g)f (g \gamma ^{-1})$$ $$\tau_{\gamma} ( \chi f ) (g) = \chi(g \gamma ^{-1})f (g \gamma ^{-1}) = \chi( \gamma) ^{-1} \chi (g)f (g \gamma ^{-1})$$ thus commutation fails just up to multiplication with the constant function $\chi( \gamma) ^{-1}$. Together, the action of $G$ and of $G^$ generate a subgroup of the Heisenberg group, a central extension $$1 {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{C}}}^ {\ensuremath{\rightarrow}}Heis (G) {\ensuremath{\rightarrow}}G \times G^* {\ensuremath{\rightarrow}}1.$$ It turns out that the algebra of theta functions is deeply related to the representation theory of the Heisenberg group of the Abelian variety $A$ (we see $A$ as the given group $G$). But, as Mumford pointed out, we have a more precise relation which takes into account a given line bundle ${\ensuremath{\mathcal{L}}}$. In the case of an Abelian variety $A$, the group of characters is endowed with a complex structure viewing it as the [Picard variety]{} $Pic^0(A)$, the connected component of $0$ in$H^1(A,{\ensuremath{\mathcal{O}}}_A^)$. $Pic^0(A)$ is also called the dual Abelian variety, and a non degenerate line bundle ${\ensuremath{\mathcal{L}}}$ is one for which the homomorphism $\phi_{{\ensuremath{\mathcal{L}}}} : A {\ensuremath{\rightarrow}}Pic^0(A)$, defined by $\phi_{{\ensuremath{\mathcal{L}}}} (x) = T_x^({\ensuremath{\mathcal{L}}}) \otimes {\ensuremath{\mathcal{L}}}^{-1}$ ($T_x$ denoting translation by $x$), is surjective (hence with finite kernel $K({\ensuremath{\mathcal{L}}})$). Mumford introduces the finite Heisenberg group associated to ${\ensuremath{\mathcal{L}}}$ via the so called Thetagroup of ${\ensuremath{\mathcal{L}}}$, defined as ${\rm Theta}({\ensuremath{\mathcal{L}}}) := \{ (x, \psi)\| \psi: {\ensuremath{\mathcal{L}}}\cong T_x^* ({\ensuremath{\mathcal{L}}}) \}$. ${\rm Theta}({\ensuremath{\mathcal{L}}})$ is a central extension of $K({\ensuremath{\mathcal{L}}})$ by ${\ensuremath{\mathbb{C}}}^$, but since $K({\ensuremath{\mathcal{L}}})$ is finite, if $n$ is the exponent of $K({\ensuremath{\mathcal{L}}})$, the central extension is induced (through extension of scalars) by another central extension $$1 {\ensuremath{\rightarrow}}\mu_n {\ensuremath{\rightarrow}}\Theta({\ensuremath{\mathcal{L}}}) {\ensuremath{\rightarrow}}K({\ensuremath{\mathcal{L}}}) {\ensuremath{\rightarrow}}1,$$ where $\mu_n$ is the group of $n$-th roots of unity. Moreover, the alternating form $\alpha : \Gamma \times \Gamma {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{Z}}}$ given by the Chern class of ${\ensuremath{\mathcal{L}}}$ gives a non degenerate symplectic form on $K({\ensuremath{\mathcal{L}}})$ with values in $\mu_n$, thus allowing to easily obtain from $\Theta({\ensuremath{\mathcal{L}}})$ the Heisenberg group of a finite group $G$. The geometry of the situation is that the group $K({\ensuremath{\mathcal{L}}})$ acts on the projective space associated to the vector space $H^0(A, {\ensuremath{\mathcal{L}}})$: but if we want a linear representation on the vector space $H^0(A, {\ensuremath{\mathcal{L}}})$ we must see this vector space as a representation of the finite Heisenberg group $\Theta({\ensuremath{\mathcal{L}}})$ (thus we have a link with Schur’s theory of multipliers of a projective representation). In the case of ${\ensuremath{\mathcal{L}}}= {\ensuremath{\mathcal{O}}}_A (n \Theta)$, $K({\ensuremath{\mathcal{L}}})$ consists of the subgroup $A_n$ of n-torsion points, and another central idea, when we have to deal with a field of positive characteristic, is to replace the vector space $\Gamma \otimes {\ensuremath{\mathbb{R}}}$ with the inverse limit of the subgroups $A_n$. The story is too long and too recent to be further told here: Mumford used this idea in order to study the Moduli space of Abelian varieties over fields of positive characteristics, and in turn this was used to take the reduction modulo primes of Abelian varieties defined over number fields. These results were crucial for arithmetic applications, especially Faltings’ solution ([@falt83]) of the [Mordell conjecture. Let a curve $C$ of genus $g\geq 2$ be defined over a number field $K$: then the set $C(K)$ of its rational points is finite.]{} It must be furthermore said that the algebraic calculations allowed by the study of the characters of representations of the finite Heisenberg groups has lead also to a better concrete understanding of equations and geometry of Abelian varieties. Surprisingly enough, even in the case of elliptic curves this has led, together with Atiyah’s study of vector bundles on elliptic curves (cf.[@ati57]) to a deeper understanding of the geometry of symmetric products of elliptic curves and their maps to projective spaces (cf. [@ca-ci]). The recent literature is so vast that we have chosen to mention just a single but quite beautiful example, due to Manolache and Schreyer (cf. [@man-schr]). The authors give several equivalent descriptions of the moduli space $X(1,7)$ of Abelian surfaces $S$ with a polarization $L$ whose elementary divisors are $(1,7)$. Their main result is that this moduli space is birational to the Fano 3-fold $V_{22}$ of polar hexagons to the Klein plane quartic curve $C$ ( of equation $ x^3 y + y^3 z + z^3 x = 0 $) which is a compactification of the moduli space $X(7)$ of elliptic curves $E$ with a level $7$ structure, i.e., elliptic curves given with an additional isomorphism of the group $E_7$ of torsion points with $({\ensuremath{\mathbb{Z}}}/ 7)^2$. The Klein quartic is rightly famous because it admits then as group of automorphisms the group ${\ensuremath{\mathbb{P}}}SL (2,{\ensuremath{\mathbb{Z}}}/ 7) = SL (2,{\ensuremath{\mathbb{Z}}}/ 7) / \{ \pm I \}$, a group of cardinality $168 $, and as it is well known this makes the Klein quartic the curve of genus $3$ with the maximal number of automorphisms (cf. [@acc]). Now, it is easy to suspect some connection between the pairs $(S,L)$ and the pairs $(E, {\ensuremath{\mathcal{L}}}: = {\ensuremath{\mathcal{O}}}_E(7 O))$ once we have learnt of the finite Heisenberg group: in fact, the respective groups for $L$ and for ${\ensuremath{\mathcal{L}}}$ are isomorphic, and the respective complete linear systems $\|L\|$ and $\|{\ensuremath{\mathcal{L}}}\|$ yield embeddings for $E$, respectively for the general $S$, into ${\ensuremath{\mathbb{P}}}^6$. That is, we view both $E$ and $S$ as Heisenberg invariant subvarieties of the same ${\ensuremath{\mathbb{P}}}^6$ with an action of $({\ensuremath{\mathbb{Z}}}/7)^2$ provided by the projectivization of the standard representation of the Heisenberg group on ${\ensuremath{\mathbb{C}}}^{{\ensuremath{\mathbb{Z}}}/7}$. The geometry of the situation tells us that $E$ is an intersection of $28 - 14 = 14$ independent quadrics, while we expect $S$ to be contained in $28- 28= 0$ quadrics, so we seem to be stuck without a new idea. The central idea of the authors is to think completely in algebraic terms, looking at a self dual locally free Hilbert resolution of the ideal of $S$, which has length $5$ instead of $4$ (because of $H^1(S, {\ensuremath{\mathcal{O}}}_S) \neq 0$). It turns out that the middle matrix, because of Heisenberg symmetry, boils down to a $3 \times 2$ matrix of linear forms on a certain ${\ensuremath{\mathbb{P}}}^3$. Then, the $3$ determinants of the $2 \times 2$-minors yield three quadric surfaces whose intersection is a twisted cubic curve $\Gamma_S$ which is shown to completely determine $S$. In this way one realizes the moduli space as a certain subvariety of the Grassmann variety of $3$-dimensional vector subspaces of the vector space $H^0({\ensuremath{\mathbb{P}}}^3, {\ensuremath{\mathcal{O}}}_{ {\ensuremath{\mathbb{P}}}^3 } (2))$ of quadrics in the given ${\ensuremath{\mathbb{P}}}^3$ (one takes the locus of subspaces where a certain antisymmetric bilinear map of vector bundles restricts to zero). This subvariety is the Fano $3$-fold $V_{22}$ mentioned above, which is a rational variety. At this point we don’t want to deprive the reader of the pleasure of learning the intricate details from the original sources ([@man-schr], [@schr]): but we need at least to explain what is a polar hexagon of a plane quartic curve $C$ with equation $ f(x) = 0$. Observe that the polynomial $f(x)$ depends upon exactly $15$ coefficients, while, if we take $6$ linear forms $l_i(x)$, they depend on $18$ affine parameters, and we expect therefore to have a $3$-dimensional variety $Hex(C)$ parametrizing the $6$-tuples of such linear forms such that $f(x) = \Sigma_{i} l_i(x)^4.$ $Hex(C)$ is called the variety of polar hexagons, and it is indeed a $3$-fold in the case of the Klein quartic. Finally, these constructions allow to find the Klein quartic as the discriminant of the net of quadrics in ${\ensuremath{\mathbb{P}}}^3$ associated to the Hilbert resolution of $S$. Also the identification of the Fano $3$-fold with $Hex(C)$ is based on the study of higher syzygies, but beyond this many other beautiful classical results are used, which are due to Klein, Scorza, and Mukai (cf. [@klein78], [@scor99], [@muk92]). Especially nice is the old theorem of Scorza ([@scor99]), proved in 1899, that the variety of plane quartic curves is birational to the variety of pairs of a plane quartic curve $D$ given together with an even theta-characteristic (this amounts to writing the equation of $D$ as the discriminant of a net of quadrics in ${\ensuremath{\mathbb{P}}}^3$). This theorem is a clear example of the geometrization of ideas coming from the theory of theta functions (which, as we saw, are certain Fourier series, and therefore, seemingly, purely analytical objects). [ABELIAN VARIETIES AND MULTILINEAR ALGEBRA]{} The roots of these developments, which historically go under the name “The problem of Riemann matrices”, and occupied an important role for the birth of the theory of rings, modules and algebras, are readily explained by the following basic Given two tori $T = V/ \Gamma, T' = V'/ \Gamma'$, any holomorphic map $f : T {\ensuremath{\rightarrow}}T'$ between them is induced by a complex linear map $ F: V {\ensuremath{\rightarrow}}V'$ such that $ F (\Gamma) \subset \Gamma'$. Whence, for a $g$-dimensional torus, the ring $$End(T) := \{ f : T {\ensuremath{\rightarrow}}T \| f \ {\rm is \ holomorphic }, f(0) = 0\}$$ is the subring of the ring of matrices $ Mat (2g, 2g, {\ensuremath{\mathbb{Z}}}) $ given by $$End(T) :\{ B \in End ( \Gamma) \| B \in (V^{\vee} \otimes V ) \oplus (\bar{V}^{\vee} \otimes \bar{V}) \subset (\Gamma \otimes {\ensuremath{\mathbb{C}}})^{\vee} \otimes (\Gamma \otimes {\ensuremath{\mathbb{C}}})\}$$ since then the restriction of $B $ to $V \subset (\Gamma \otimes {\ensuremath{\mathbb{C}}})$ is complex linear. In general, the study of endomorphism rings of complex tori is not completely achieved. The main tool which makes the case of Abelian Varieties easier is the famous [Poincaré’s complete reducibility theorem]{} (cf. [@poi84]) Let $A'$ be a subabelian variety of an Abelian variety $A$: then there exists another Abelian variety $A"$ and an [isogeny]{} (a surjective homomorphism with finite kernel) $ A' \times A" {\ensuremath{\rightarrow}}A$. [[Proof.* ]{}]{}The datum of $A'$ amounts to the datum of a sublattice $\Gamma' \subset \Gamma $ which is saturated ( $ \Gamma /\Gamma'$ is torsion free) and complex (i.e., there is a complex subspace $W \subset V$ with $ W \oplus \overline{W } = \Gamma' \otimes {\ensuremath{\mathbb{R}}}\subset \Gamma \otimes {\ensuremath{\mathbb{R}}}= V \oplus \overline{V } $ ). Now, given the alternating form $A$, its orthogonal in $\Gamma$ yields a sub-lattice $\Gamma "$ spanning the complex subspace $U$ orthogonal to $W$ for the Hermitian bilinear product associated to $H$: since $H$ is positive definite, we obtain an orthogonal direct sum $ V = W \oplus U$, and we define $ A" : = U /\Gamma"$. We sketched the above proof just with the purpose of showing how the language of modern multilinear algebra is indeed very appropriate for these types of questions. The meaning of the reducibility theorem is that, while for general tori a subtorus $ T' \subset T$ only yields a quotient torus $ T / T'$, here we get a direct sum if we consider an equivalence relation which identifies two isogenous Abelian vareties. Algebraically, the winning trick was thus to classify first Endomorphism Rings tensored with the rational integers, because If $T = V / \Gamma, T' = V / \Gamma'$ are isogenous tori, then $End(T) \otimes {\ensuremath{\mathbb{Q}}}\equiv End(T') \otimes {\ensuremath{\mathbb{Q}}}$. And then the study is restricted to the one of [Simple Abelian Varieties]{}, i.e., of the ones which do not admit any Abelian subvariety whatsoever (naturally, this concept was very much inspired by the analogous concept of curves which do not admit a surjective and not bijective mapping onto a curve of positive genus). The classification of Endomorphism rings of Abelian varieties was achieved through a long series of works by Scorza, Rosati, Lefschetz and Albert (cf. e.g. [@scor16], [@lef], [@albe], [@ros]) and today one can find an exposition in Chapters 5 and 9 of the book by Lange and Birkenhake [@l-b], cf. also, for an historical account, the article by Auslander and Tolimieri [@a-t]. Although the methods of Scorza and Rosati were more geometrrical, certainly more than the later ones by Albert, who essentially worked in the new direction set up by Emmy Noether, i.e., of the abstract algebra, a central role is played by a notion due to Rosati, the so-called [Rosati involution]{}. Given an endomorphism with integral matrix $B$, the Rosati involution associates to it ( $A$ being a Riemann integral matrix as in (3.3), (3.4)) the matrix $B' : A^{-1} \ ^{t}B A$. The Rosati involution is positive in the sense that the symmetric bilinear form $(B_1, B_2) := Tr (B_1' B_2 + B_1 B_2')$ yields a positive definite scalar product. It turns out that the classification of Riemann matrices is very close to the study of rational Algebras with a positive involution, and abstract arguments imply that these simple algebras are skew fields ${\ensuremath{\mathbb{F}}}$ of finite dimension over ${\ensuremath{\mathbb{Q}}}$ of two types - \(I) The centre $K$ of ${\ensuremath{\mathbb{F}}}$ is a totally real number field and, if $K \neq {\ensuremath{\mathbb{F}}}$, then ${\ensuremath{\mathbb{F}}}$ is a quaternion algebra over $K$. Moreover, for every embedding $\sigma: K {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{R}}}$, ${\ensuremath{\mathbb{F}}}\otimes {\ensuremath{\mathbb{R}}}$ is always definite (${\ensuremath{\mathbb{F}}}\otimes {\ensuremath{\mathbb{R}}}\cong {\ensuremath{\mathbb{H}}}$), or always indefinite ($ {\ensuremath{\mathbb{F}}}\otimes {\ensuremath{\mathbb{R}}}\cong M(2, {\ensuremath{\mathbb{R}}})$). - \(II) The centre $K$ of ${\ensuremath{\mathbb{F}}}$ is a totally complex quadratic extension of a totally real number field $K_0$, and then ${\ensuremath{\mathbb{F}}}\otimes {\ensuremath{\mathbb{C}}}$ is a matrix algebra $M(r,{\ensuremath{\mathbb{C}}})$ such that the positive involution extends to the standard involution $ C {\ensuremath{\rightarrow}}\ \overline{ ^tC }$. The analytic moduli theory of Abelian varieties owes much to the work of Siegel and to his ’Symplectic geometry’ ([@sieg43]): today the space of matrices ${\ensuremath{\mathcal{H}}}_g : = \{\tau \in Mat(g,g,{\ensuremath{\mathbb{C}}})\| \tau = ^t\tau , \ Im (\tau) > 0 \} $ is called the Siegel upper half space, and it is a natural parameter space for Abelian varieties, since, depending on the polarization, there is a subroup $\Gamma$ of $Sp(2g, {\ensuremath{\mathbb{Q}}})$ such that the moduli space is, analytically, the quotient $ {\ensuremath{\mathcal{H}}}_g / \Gamma$. The moduli theory of Abelian varieties with a certain polarization and endomorphism structure was pursued relatively recently by Shimura (cf. [@shim]), and it is a currently very active field of research for the arithmetic applications of the theory of such Shimura varieties. I do not need to cite for instance the (recently proved) so called Shimura-Taniyama-Weil conjecture about the modularity of elliptic curves defined over ${\ensuremath{\mathbb{Q}}}$: I can simply refer to the talk by Wiles. In this direction, however, the current tendency is to develop also much the geometry, since one has to look at the reduction of these modular varieties modulo primes. The hope is that this study will play a primary role for the pursuing of the so called Langlands program, which is a vast generalization of the previously cited conjecture, proposing to relate modular forms arising in different contexts ( cf. [@langl70], and [@langl76], [@langl79], [@del79] for early accounts of the story). I hope that some more competent author than me will report about this development in the present volume. I want instead to end this section by pointing out (cf. [@zap]) how important the role of Scorza was for the development of the field of abstract algebra in Italy: his path started with correspondences between curves, but, as we contended here, his researches centered about Riemann matrices made him realize about the relevance of the powerful new algebraic concepts. Further links to the italian school =================================== We mentioned in the previous section how the research of Rosati and Scorza was very much influenced by the new geometric methods of the italian school of algebraic geometry. As we said, a crucial role was played by Castelnuovo: concerning Abel’s theorem, in the article [@cast93], entitled “The 1-1 correspondences between groups of $p$ points on a curve of genus $p$”, he explained how one could e.g. formulate the fundamental theorem about the inversion of Abelian integrals as a consequence of the theory of linear series on a curve (a development starting with the geometric interpretation of the Riemann Roch theorem). It is interesting to observe that, when he wrote a final note in the edition appearing in his collected works, he points out that the results can be formulated in a simpler way if one introduces the concept of the Jacobian variety of the curve. These notes added around 1935 are rather interesting: for instance, in the note to the paper entitled “On simple integrals belonging to an irregular surface” ([@cast05]) he pointed out that exactly in this Memoir he introduced the concept of the so called [Picard variety]{}, applyimg this concept to the study of algebraic surfaces. In fact, the theorem of Picard to which Castelnuovo refers, proved by Picard in [@pic], and with precisions by Painlevé in [@painl], is indeed the characterization of the Abelian varieties, (we add to it a slight rewording in modern language) Let $V^p$ be a $p$-dimensional algebraic variety admitting a transitive $p$-dimensional abelian group of birational transformations: then the points $\xi \epsilon V$ are uniformized by entire $2p$-periodic functions on ${\ensuremath{\mathbb{C}}}^p$, $$\xi_k = \phi_k (u_1, \dots ,.. u_p)$$ (i.e., $V$ is birational to a complex torus of dimension $p$). The first main result of Castelnuovo in [@cast05] is the construction of the so called Albanese variety and Albanese map of an algebraic surface $X$. Recall that, in modern language, the Albanese variety of a projective variety $X$ is the Abelian variety $ ( H^0 (\Omega^1_X ))^{\vee} / H_1(X)$ where $ H_1(X)$ is the lattice, in the dual vector space of $ H^0 (\Omega^1_X )$, given by integration along closed paths. The Albanese map, defined up to translation, as a result of the choice of a base point $x_0$, associates to a point $x$ the linear functional $\alpha(x):= \int_{x_0}^x ( {\rm mod } \ H_1(X) )$.[^32] The second result, obtained independently by F. Severi in [@sev05], concerns the equality of the irregularity of an algebraic surface and the dimension of the space of holomorphic one forms. Both proofs were relying on a shaky proof given by Enriques one year before, in [@enr05], claiming the existence a continuous system $\Sigma$ of dimension $q: = p_g - p_a$ of “inequivalent curves”(i.e., such that for a generic curve $C \in \Sigma$, the set of curves in $\Sigma$ which are linearly equivalent to $C$ has dimension zero). Fortunately, a corrrect analytical proof was later found by Poincaré in [@poi10]. Enriques and Severi tried for a long time to repair the flaw in Enriques’ geometrical arguments, although in the end it started to become clear the need for higher order differential elements (i.e., higher order terms in the Taylor expansion of the curve variation). The fruit of the researches carried on much later in the 50’s was to show that indeed, for varieties defined over algebraically closed fields of positive characteristic, the [arithmetic irregularity]{} $q: = p_g - p_a$ was in general larger than the [geometric irregularity]{} defined as the maximal dimension of a continuous system of “inequivalent curves”. We refer the reader to Mumford’s classical book [@mumlc], relating this question to the non reducedness of the [Picard scheme ]{} $H^1( {\ensuremath{\mathcal{O}}}_X^{})$. Castelnuovo was instead more interested in the applications of the previous theorem, the most important one being the theorem [@cast05-2] that an algebraic surface with arithmetic genus $p_a$ smaller than $-1$ is birationally ruled. This theorem is indeed one of the key theorems of the classification of algebraic surfaces, since it also implies the well known [Castelnuovo rationality criterion: a surface is rational if and only if the bigenus $P_2$ and the irregularity $q$ vanish]{}. Without opening a new story, I would like to observe that the so called “Enriques classification” of algebraic surfaces, done by Castelnuovo and Enriques, was one of the most interesting cooperations in the history of mathematics, which took place in the years from 1892 to 1914 (and especially intense in the period 1892 to 1906). Besides the published papers, one can consult today the book entitled “Riposte armonie” [@RA] (“Hidden harmonies”, as are the ones governing algebraic surfaces), which contains around 670 letters (or postcards) written from Federigo Enriques to Guido Castelnuovo. Naturally, also Castelnuovo wrote quite many letters, but apparently Enriques did not bother to keep them. This already shows where Castelnuovo and Enriques respectively belong, in the rough distinction made by H. Weyl which we quoted above, although both of them were obviously geometers. In fact, Enriques used to discuss mathematics with his assistents during long walks in gardens or parks, and would only sometime stop to write something with his stock on the gravel. Moreover, as Guido Castelnuovo wrote of him[^33]with affection , he was a “mediocre reader, who saw in a page not what was written, but what he wanted to see”; certainly his brain was always active like a vulcano. After the first world war the collaboration of the two broke up, more on the side of Castelnuovo. As his daughter Emma Castelnuovo writes, Enriques would regularly visit his sister (Castelnuovo’s wife) at their house, and after dinner the two mathematicians went to a separate sitting room, where Enriques wanted to discuss his many new ideas, while Castelnuovo had prudently instructed his wife to come after some time and interrupt their conversation with some excuse. Castelnuovo was 6 years older than Enriques, was always calm and mature, and, after the appearance of his ground breaking two notes over algebraic surfaces ([@cast91]) he was a natural reference for the brilliant student Enriques, who graduated in Pisa in 1891 (just at the time when Guido became a full professor). Enriques wanted first to perfection his studies under the guidance of Segre in Torino, but instead got a fellowship in Rome by Cremona, and there, in 1892, started the intense mathematical interchange with Castelnuovo. In his first Memoir ([@enr93]) Enriques, after an interesting historical introduction, sketches the main tools to be used for the birational study of algebraic surfaces, namely: the theory of linear systems of curves, the canonical divisor and the operation of adjunction. Some results, as the claim that two birational quartic surfaces are necessarily projectively equivalent, are today known not to hold. It took a long time to make things work properly, and it is commonly agreed that the joint paper [@c-e14] marks the achievement of the classification theory. For the later steps, (main ones, as well known to anyone who understands the structure of the classification theorem) a very important role played the [(IP) De Franchis’ theorem on irrational pencils* ]{}([@df05], [@cast05-2]) and the [(HS) Classification of hyperelliptic surfaces ]{}. [(IP)]{} De Franchis’ theorem, obtained independently also by Castelnuovo and Enriques, asserts that if on a surface $S$ there are several linearly independent one forms $\omega_i, i=1, \dots r$ which are pointwise proportional, then there is a mapping $f : S {\ensuremath{\rightarrow}}C$ to an algebraic curve $C$ such that these forms are pull backs of holomorphic one forms on $C$. It was used by Castelnuovo to show that under the inequality $ p_g \leq 2q-4$ there is a mapping to a curve of positive genus. This theorem leads to a typical example of the algebraization of the geometry: although the Hodge theory of Kähler varieties was established with the use of hard analytic tools which underlie Hodge’s theorem on harmonic integrals ([@hod], [@kod]), what turned out to be very fertile was the simple algebraic formulation in terms of the cohomology algebra of a projective variety. Using this, Z. Ran, M. Green and myself ([@ran], [@cat91]) were independently able to extend the result of de Franchis to the case of higher dimensional varieties and higher dimensional targets. In this way the ideas of the italian school came back to intense life, and became an important tool for the investigation of the fundamental groups of algebraic varieties (for instance N. Mok [@mok] tried to extend this result to infinite dimensional representations, with the hope of using such a result for the solution of the so called Shafarevich conjecture about the universal coverings of algebraic varieties). [(HS)]{} The second main work of de Franchis [^34], together with Bagnera, was the classification of hyperelliptic surfaces, i.e. of surfaces whose universal covering space is biholomorphic to ${\ensuremath{\mathbb{C}}}^2$ (cf. [@ba-df07], [@ba-df07-P] ,[@ba-df07], [@e-s07], [@e-s08],[@e-s09],[@e-s10]). This classification was also obtained by Enriques and Severi, and the Bordin Prize was awarded to Enriques and Severi in 1907, and to the sicilian couple in 1909 [^35]. Strange as it may seem that two couples get two prices for the same theorem, instead of sharing one, this story is even more complicated, since the first version of the paper by Enriques and Severi was withdrawn after a conversation of Severi with de Franchis, and soon corrected. Bagnera and de Franchis were only a little later, since they had to admit a restriction (a posteriori useless, since no curve on an Abelian surface is contractible, cf. [@df36]): their proof was however simpler, and further simplified by de Franchis much later ([@df36-2]). Another beautiful result, and this one even more related to Abelian integrals, is the famous [Torelli’ s Theorem: Let $C$, $C'$ be two algebraic curves whose Jacobian varieties are isomorphic as polarized Abelian varieties (equivalently, admitting the same matrix of periods for Abelian integrals of the I kind): then $C$ and $C'$ are birationally isomorphic, [@tor]]{}. Torelli was born in 1884 and was a student of Bertini, in Pisa, where he attended also Bianchi’s lectures: he was for short time assistant of Severi, and died prematurely in the first world war, in 1915. These years at the beginning of the 20-th century in Italy were thus quite exciting. In the book by J. Dieudonné, [@dieu] vol.1, Chapter VI, entitled “Developpement et chaos”, contains a paragraph dedicated to “ L’ école italienne et la théorie des systemes linéaires”, namely, devoted to the second generation of the italian school. As third generation of the italian school, we became very interested, also because of this criticism, to become fully acquainted with the results of these precursors. What I found as a very interesting peculiarity reading the book by Enriques on “Algebraic surfaces” ([@enr49]), was the mixture of theorems, proofs, speculations, and history of the genesis of the mathematical ideas. For instance, chapter IX, entitled “Irregular surfaces and continuous systems of disequivalent curves”, has a section 6 on “History of the theory of continuous systems” (pages 339-347). One can read there that in 1902 Francesco Severi, who had just graduated in Torino, following the advice of his master C. Segre, accepted a position in Bologna as assistant of Severi. Under the influence of Enriques, Severi started the investigation of the particular surfaces which occur as the symmetric square of a curve (cf. [@sev03]). According to Enriques, this research lead him to consider the [Problem of the base]{} for the class of divisors modulo numerical equivalence. We may also observe that these surfaces have a very special geometry, and, although a general characterization has not yet been described in general, we have some quite recent (cf. [@ccm], [@h-p01], [@pi01]) results The symmetric squares $C^{(2)}$ of a curve of genus $3$ are the only irregular surfaces of general type with $p_g \geq 3$ presenting the [non-standard case ]{} for the bicanonical map, i.e., such that - Their bicanonical map $\phi_{2K}$ is not birational onto its image - $S$ does not contain any continuous system of curves of genus $2$. Moreover, any algebraic surface with $p_g = q = 3$ is either such a symmetric product (iff $K^2=6$ for its minimal model), or has $K^2=8$ and is the quotient of a product $C_1 \times C_2$ of two curves of respective genera $2$ and $3$ by an involution $i = i_1 \times i_2$ where $C_1 / i_1$ has genus $1$, while $i_2$ operates freely. The symmetric square $C^{(2)}$ of a curve of genus $2$ occurs in another characterization of the [non-standard case ]{} given by Ciliberto and Mendes-Lopes ([@c-m1], [@c-m2]). The double covers $S$ of a principal polarized Abelian surface $A$, branched on a divisor algebraically equivalent to $ 2 \Theta$, are the only irregular surfaces of general type with $p_g =2$ presenting the [non-standard case ]{} for the bicanonical map. In these theorems plays an essential role the continuous system of [paracanonical curves ]{}, i.e., of those curves which are algebraically equivalent to a canonical divisor. To this system is devoted section 8 of the cited Chapter IX of [@enr49], and there Enriques, after mentioning false attempts by Severi and himself to determine the dimension of the paracanonical system $\{ K\}$, analyses the concrete case of a surface $C^{(2)}$, with $C$ of genus $3$, in order to conjecture that $ dim \{ K\} = p_g$. The assumptions conjectured by Enriques were not yet the correct ones, but, under the assumptions that the surface does not contain any irrational pencil of genus $\geq 2$, the conjecture of Enriques was proved by Green and Lazarsfeld, via the so called “Generic vanishing theorems” (cf. [@g-l87], [@g-l91]). I will not dwell further on this very interesting topic , referring the reader to the survey, resp. historical, articles [@cat91-B] [@cil91-B]. I should however point out that further developments are taking place in this direction, following a seminal paper by Mukai ([@muk81]) who extended the concept of Fourier transforms (don’t forget that theta functions are particular Fourier series!) to obtain an isomorphism between the derived category of coherent sheaves on an Abelian variety $X$ and the one of its dual Abelian variety $\hat{X} := Pic^0(X)$. One specimen is the combination of Mukai’s technique with the theory of generic vanishing theorems by Green and Lazarsfeld to obtain limitations on the singularities of divisors on an Abelian variety (cf. [@hac00], where one can also find references to previous work by Kollar, Ein and Lazarsfeld). Speaking about links with the italian school I should not forget the beautiful lectures I heard in Pisa from Aldo Andreotti on complex manifolds and on complex tori. Through his work with F. Gherardelli ([@a-g76]), I got in touch with a problem of transcendental nature which occupies a central place in Severi’s treatise on Quasi Abelian Varieties. [Quasi Abelian varieties ]{}, in Severi’ s terminology, are the Abelian complex Lie groups which sit as Zariski open sets in a projective variety. Whence, they are quotients $ {\ensuremath{\mathbb{C}}}^n / \Gamma$ where $\Gamma$ is a discrete subgroup of ${\ensuremath{\mathbb{C}}}^n$, thus of rank $r \leq 2n$, and the above algebraicity property leads again to the two Riemann bilinear relations: - I\) There exists an alternating form $ A : \Gamma \times \Gamma {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{Z}}}$ such that $A$ is the imaginary part of an Hermitian form $H$ on $V$ - II\) there exists such an $H$ which is positive definite (in this case $H$ is not uniquely determined by $A$). Andreotti and Gherardelli conjectured that [The Riemann bilinear relations hold if and only if there is a meromorphic function of ${\ensuremath{\mathbb{C}}}^n$ with group of periods equal to $\Gamma$ ]{}. This conjecture was the first Ph.D. problem I gave, and after some joint efforts, Capocasa and I were able to prove it in [@c-c]. [HOMOLOGICAL ALGEBRA for Abelian and irregular Varieties]{} As already mentioned, David Mumford’s ground breaking articles ([@mum66-7]) set up the scope of laying out a completely algebraic theory of theta functions. His attempt was not the only one, for instance Barsotti (cf. [@bar1],[@bar2],[@bar3],[@bar4], and [@langav] for references about his early work) had another approach to Abelian varieties, based on power series, Witt vectors and generalizations of them (Witt covectors cf. [@bar3]), and the so called “Prostapheresis formula” (cf. [@bar4]). Discussing here the respective merits of both approaches would be difficult, but at least I can say that, while Barsotti’s work is mainly devoted to Abelian varieties in positive characteristic, the theory of theta groups, as already mentioned, is also a very useful tool in characteristic zero. The title of Mumford’s series of articles is “On the equations defining abelian varieties”, which has a different meaning than “The equations defining abelian varieties” [^36]. Thus, he set up a program which has been successfully carried out in the case of several types of Abelian varieties. More generally, one can set as a general target the one of studying the equations of irregular varieties, i.e., of those which admit a non trivial Albanese map. Since otherwise the problem is set in too high a generality, let me give a concrete example (for many topics I will treat now, consult also the survey paper [@cat97], which covers developments until 1996). Chapter VIII of the book by F. Enriques (finished by his assistants Pompilj and Franchetta after the death of Enriques) is devoted to the attempt to find explicitly the canonical surfaces $S$ of low canonical degree in ${\ensuremath{\mathbb{P}}}^3$: i.e., one considers surfaces with $p_g=4$ and with birational canonical map $\phi:= \phi_K : S {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{P}}}^3$. If the canonical system has no base points, then we will have a surface of degree $d = K^2 \geq 5$. The cases $K^2 = 5,6$ are easy to describe, and for $K^2 \geq 7$ Enriques made some proposals to construct some regular surfaces (with a different method, Ciliberto [@cil81] was able to construct these for $ 7 \leq K^2 \leq 10$, and to sketch a classification program, later developed in [@cat84], based essentially on Hilbert ’s syzygy theorem). It was possible to treat the irregular case (cf. [@c-s02]) using a new approach based on Beilinson’s theorem ([@bei78]) for coherent sheaves ${\ensuremath{\mathcal{F}}}$ on ${\ensuremath{\mathbb{P}}}^n$, which allows to write every such sheaf as the cohomology of a [monad]{} (a complex with cohomology concentrated at only one point) functorially associated with ${\ensuremath{\mathcal{F}}}$. The natural environment for rings not necessarily generated in degree $1$ is however the [weighted projective space]{}, which is the projective spectrum $Proj({\ensuremath{\mathcal{A}}})$ of a polynomial ring ${\ensuremath{\mathcal{A}}}:= {\ensuremath{\mathbb{C}}}[x_0, \dots, x_n]$ graded in a non standard way, so that the indeterminates $x_i$ have respective degrees $m_i$ which are positive integers, possibly distinct. Canonaco ([@can00]) was first able to extend Beilinson’s theorem to the weighted case under some restriction on the characteristic of the base field, and later ([@can02]) not only removed this restriction, but succeeded to construct a unique functorial Beilinson type complex, making use of a new theory, of so called [graded schemes]{}. A concrete application given was to determine the canonical ring of surfaces with $K^2=4, p_g = q = 2$ (for these surfaces there is no good canonical map to the ordinary 3-space ${\ensuremath{\mathbb{P}}}^3$). Although a general theory appears to be very complicated, it thus turns out that Abelian varieties (and for instance related symmetric products of curves) offer crucial examples (admitting sometime a geometrical characterization) for the study of irregular varieties. Some of them were already discussed before, we want to give a new one which is particularly interesting, and yields (cf. [@c-s02] for more details) an easy counterexample to an old “conjecture” by Babbage (cf. [@cat81] and [@bea79] for references and the first counterexample). Let $J = J(C)$ be the Jacobian of a curve $C$ of genus $3$, and let $A {\ensuremath{\rightarrow}}J$ be an isogeny of degree $2$. The inverse image $S$ of the theta divisor $ \Theta \cong_{bir} C^{(2)}$ provides $A$ with a polarization of type $(1,1,2)$. The canonical map of $S$ factors through an involution $\iota$ with $32$ isolated fixed points, and the canonical map of the quotient surface $\Sigma: = S / \iota$, whose image is a surface $Y$ of degree $6$ in ${\ensuremath{\mathbb{P}}}^3$ having $32$ nodal isolated singularities, and a plane cubic as double curve. Moreover, $\Sigma$ is the normalization of $Y$. It is now difficult for me to explain and to foresee exactly what principles should these examples illustrate, let me however try. In Theorem 3.1 I tried to give an explicit example of the “algebraization” of the geometry, showing how the question of the birationality to Abelian varieties of symmetric products of varieties can be reduced to pure exterior algebra arguments. Abelian varieties are just, so to say, the complex incarnation of exterior algebras. On the other hand, a companion article by Bernstein-Gelfand-Gelfand ([@bgg78]) appeared next to the cited article [@bei78] by Beilinson. In abstract setting, it shows the equivalence of the derived category of coherent sheaves on ${\ensuremath{\mathbb{P}}}^n$ and the category of finite modules over the exterior algebras. While it is not yet completely clear how to extend this result to the weighted case, quite recently in [@efs01] it was shown how the BGG method allows to write functorially not only the sheaves, but also the homomorphisms in the Beilinson monad. Note that the exterior algebra of BGG is the exterior algebra over the indeterminates of a polynomial ring, and is therefore apparently geometrically unrelated for the moment to the exterior algebra of an Abelian variety. Given however a morphism $f : X {\ensuremath{\rightarrow}}A$, we attach to it the induced homomorphism between the respective holomorphic algebras of $A$ and $X$, $f^* : H^0(\Omega^_A) {\ensuremath{\rightarrow}}H^0(\Omega^_X)$. Whence, we obtain a module over the exterior algebra, and we associate to it a Beilinson monad. This procedere shows that to $f$ we associate some geometric objects related to the Gauss maps corresponding to $f$. It is just the converse which looks more problematic, is it possible to associate, to a map $f : X {\ensuremath{\rightarrow}}{\ensuremath{\mathbb{P}}}^n$ to a projective space, a geometric map to an Abelian variety giving a realization of the module $f_* {\ensuremath{\mathcal{O}}}_X$? In general, progress on the question of canonical rings or equations of irregular varieties might require at least further combinations of the several existing techniques which we have mentioned. More new results and open problems ================================== [6.1. The Torelli problem]{} The Torelli theorem, mentioned in the previous section, was again at the centre of attention in the 50’s, when several new proofs were found, by Weil, Matsusaka, Andreotti ( and many others afterwards). Particularly geometrical was the proof given by A. Andreotti ([@andr58]), who showed that, given a Jacobian variety $J(C)$ with a theta divisor $ \Theta \cong_{bir} C^{(g-1)}$, the Gauss map of $\Theta$ is ramified on the dual variety $C^{\vee} \subset {\ensuremath{\mathbb{P}}}^{g-1}$ of the canonical image of the curve $C$. Using projective duality, one sees therefore that $(J(C), \Theta)$ determines $C$ (Andreotti had then to treat the hyperelliptic case separately). Andreotti and Mayer ([@a-m67]) pushed the study of the geometry of canonical curves, especially of the quadrics of rank $4$ containing them, to obtain some equations valid for the period matrices of curves inside the Siegel upper half plane. This paper, written at a time when the fashion was oriented in quite different directions, had a great impact on the revival of classical researches about algebraic curves, Abelian varieties (cf. [@acgh84] and [@l-b] for references).[^37] In the same years Philip Griffiths (cf. [@grif68], [@grif70], [@grif84] and references therein) greatly extended the theory of the periods of Abelian integrals, proposing to use the Hodge structures of varieties, i.e. the isomorphism class of the datum $$H^(X, {\ensuremath{\mathbb{Z}}}) {\ensuremath{\rightarrow}}H^(X, {\ensuremath{\mathbb{Z}}}) \otimes {\ensuremath{\mathbb{C}}}= \oplus_{p,q} H^{p,q}(X)$$ to study their moduli (note that a modern formulation of Torelli’s theorem is that the birational isomorphism class of an algebraic curve is determined by the Hodge structure on its cohomology algebra). A prominent role played in his program the attempt to find a reasonable generalization of Torelli’s theorem, and indeed (cf. [@grif84]) a lot of Torelli type theorems were proved for very many classes of special varieties. Since then, a basic question has been the one of finding sufficient conditions for the validity of an infinitesimal Torelli theorem for the period map of holomorphic $n$-forms (i.e., for the Hodge structure on $ H^n(X, {\ensuremath{\mathbb{Z}}}) \otimes {\ensuremath{\mathbb{C}}}$) for a variety $X$ of dimension $n$. The question is then, roughly speaking, whether the period map is a local embedding of the local moduli space. Observe moreover that the $n$-forms are the only forms which surely exist on simple cyclic ramified coverings $Y_D$ of a variety $X$ of general type, branched on pluricanonical divisors $D$ (divisors $ D \in \| mK_X \|$), and in this context the Torelli problem is a quantitative question about how large $m$ has to be in order that the variation of Hodge structure distinguishes the $Y_D$’s (cf. the article [@mig95] which gave a very interesting application of these ideas to families of higher dimensional varieties, opening a new direction of research). The validity of such an infinitesimal Torelli theorem can be formulated in purely algebraic terms as follows: is the cup product $$H^1(X, T_X) \times H^0 (X, \Omega^n_X) {\ensuremath{\rightarrow}}H^1 (X, \Omega^{n-1}_X)$$ non degenerate in the first factor? Classical examples by Godeaux and Campedelli and modern ones (cf.[@bpv84] for references) produce surfaces of general type with $p_g = q=0$ (thus with a trivial Hodge structure), yet with moduli. In view of Andreotti’s interpretation of Torelli’s theorem, one suspects then that a good condition might be the very ampleness of the canonical divisor, i.e., the condition that the canonical map be an embedding. This might unfortunately not be the case, as shown in a joint paper with I. Bauer [@b-c02]: There are surfaces of general type with onjective canonical morphism and such that the infinitesimal Torelli theorem for holomorphic $2$-forms does not hold for each surface in the moduli space. Examples of such behaviour are quotients $(C_1 \times C_2) / ({\ensuremath{\mathbb{Z}}}/3)$, where $C_1 {\ensuremath{\rightarrow}}C_1 / ({\ensuremath{\mathbb{Z}}}/3) \cong {\ensuremath{\mathbb{P}}}^1$ is branched on $3k+2$ points, and $({\ensuremath{\mathbb{Z}}}/3)$ acts freely on $C_2$ with a genus $3$ quotient. It would be interesting to establish stronger geometrical properties of the canonical map which guarantee the validity of the infinitesimal Torelli theorem for the holomorphic $n$-forms. [6.2. The Q.E.D. problem]{} If higher dimensional varieties were products of curves, life would be much simpler. It obviously cannot be so, since there are plenty of varieties which are simply connected (e.g., smooth hypersurfaces in ${\ensuremath{\mathbb{P}}}^n$ with $n \geq 3$), without being rational. Can life be simpler ? It is a general fact of life that, in order to make the study of algebraic varieties possible, one must introduce some equivalence relation. The most classical one is the so called birational equivalence, which allows in particular not to distinguish between the different projective embeddings of the same variety. Let $X$ and $Y$ be projective varieties defined over the field $K$: then they are said to be birational if their fields of rational functions are isomorphic : $K(X) \cong K(Y)$. Moreover, one must allow algebraic varieties to depend on parameters, for instance the complex hypersurfaces of degree $d$ in ${\ensuremath{\mathbb{P}}}^n$ depend on the coefficients of their equations: but if these are complex numbers, we can have uncountably many birational classes of algebraic varieties. To overcome this difficulty, Kodaira and Spencer introduced the notion of [deformation equivalence]{} for complex manifolds: they ([@k-s58]) defined two complex manifolds $X'$, $X$ to be [directly deformation equivalent]{} if there is a proper holomorphic submersion $ \pi : \Xi {\ensuremath{\rightarrow}}\Delta$ of a complex manifold $\Xi$ to the unit disk in the complex plane, such that $X, X'$ occur as fibres of $\pi$. If we take the equivalence relation generated by direct deformation equivalence, we obtain the relation of deformation equivalence, and we say that $X$ is a deformation of $X'$ in the large if $X, X'$ are deformation equivalent. These two notions extend to the case of compact complex manifolds the classical notions of irreducible, resp. connected, components of moduli spaces. It was recently shown however( [@man01], [@k-k01],[@cat01], ) that it is not possible to give effective conditions in order to guarantee the deformation equivalence of algebraic varieties, as soon as the complex dimension becomes $\geq 2$. Thus in [@catqed] I introduce the following relation Let $n$ be a positive integer, and consider, for complex algebraic varieties $X, Y$ of dimension $n$, the equivalence relation generated by - \(1) Birational equivalence - \(2) Flat deformations with connected base and with fibres having only at most canonical singularities - \(3) Quasi étale maps, i.e., morphisms which are unramified in codimension $1$. This equivalence will be denoted by $X \cong_{QED} Y$ (QED standing for: quasi-étale-deformation). - Singularities play here an essential role. Note first of all that, without the restriction on these given in (2), we obtain the trivial equivalence relation (since every variety is birational to a hypersurface). - Assume that a variety $X$ is rigid, smooth with trivial algebraic fundamental group: then $X$ has no deformations, and there is no non trivial quasi-étale map $Y {\ensuremath{\rightarrow}}X$. In this case the only possibility, to avoid that $X$ be isolated in its QED-equivalence class, is that there exists a quasi-étale map $f : X {\ensuremath{\rightarrow}}Y$. If $f$ is not birational, however, the Galois closure of $f$ yields another quasi-étale map $\phi : Z {\ensuremath{\rightarrow}}X$, thus it follows that $f$ is Galois and we have a contradiction if $Aut(X) = \{1\}$. It does not look so easy to construct such a variety $X$. Are there invariants for this equivalence? A recent theorem of Siu ([@siu02]) shows that the Kodaira dimension is invariant by QED equivalence. It is an interesting question to determine the QED equivalence classes inside the class of varieties with fixed dimension $n$, and with Kodaira dimension $k$. For curves and special surfaces, there turns out to be only one class ([@catqed]): In the case $ n \leq 2, k \leq 1$ the following conditions are equivalent - \(i) $X \cong_{QED} Y$ - \(ii) $dim X= dim Y= n$, $Kod(X) = Kod(Y)= k$ The previous result uses heavily the Enriques classification of algebraic surfaces. We can paraphrase the problem, for the open case of surfaces of general type, using Enriques’ words ([@enr49]): “We used to say in the beginning that, while curves have been created by God, surfaces are the work of the devil. It appears instead that God wanted to create for surfaces a finer order of hidden harmonies..” Are here then new hidden harmonies to be found? [Acknowledgements.]{} I wish to thank Arnfinn Laudal for convincing me to try to learn about Abel’s life and to look at his papers, and for allowing me to do it comfortably at home. I would also like to thank Ragni Piene for some useful comments. [Bibliographical remark.]{} We are not in the position to even mention the most important references. However, some of the references we cite here contain a vast bibliography, for instance [@sieg], pages 193-240, and [@zar], pages 248-268. [99]{} N. H. Abel, “[Oeuvres complétes, Tomes I et II ]{}”, L. Sylow and S.Lie ed., [Im. Grondahl Son , Christiania ]{} (1881), 621 pp., 341 pp. . R.D.M. Accola, “[Topics in the theory of Riemann surfaces]{}”, [Springer Lecture Notes in Mathematics. 1595]{}, (1994), ix, 105 p.. G. Albanese, “[Corrispondenze algebriche fra i punti di due superficie algebriche. ]{}”, [Ann. R. Scuola Normale Sup. Pisa, (II), 3 ]{} (1934), 1-26, 149-182 . G. Albanese, “[Collected papers. ]{}”, with an historical note by C.Ciliberto and E. Sernesi, [Queen’s papers in pure and applied mathematics, vol. 103 ]{} (1996), 408 pp. A.A. Albert, “[Structure of algebras. ]{}” [ Amer. Math. Soc. Coll. Publ. 24, New York]{}, (1939). A. Andreotti, “[On a theorem of Torelli. ]{}” [ Am. J. Math. 80]{}, (1958), 801-828 . A. Andreotti, A. Mayer, “[On period relations for abelian integrals on algebraic curves]{}” [ Ann. Scuola Normale Sup. Pisa 21]{}, (1967), 189-238. A. Andreotti, F. Gherardelli, “[Some remarks on quasi-abelian manifolds]{}”, in ’Global analysis and its applications’, [ I.A.E.A. Vienna]{}, (1976), 203-206. P. Appell, E. Goursat “[Théorie des fonctions algébriques et de leurs intégrales. ]{}” [ Gauthier-Villars, Paris]{},(1895), 530 pp.. E. Arbarello, M. Cornalba, P. Griffiths, J. Harris, “[Geometry of algebraic curves I. ]{}” [ Grundlehren der math. Wiss. 267, Springer]{},(1984), 386 pp.. M. F. Atiyah, “[Vector bundles over an elliptic curve. ]{}” [ Proc. Lond. Math. Soc., III. Ser. 7]{}, 414-452 (1957). L. Auslander, R. Tolimieri “[A matrix free treatment of the problem of Riemann matrices. ]{}” [ Bull. Amer. Math. Soc. 5, 3 ]{}, (1981), 263-312. L. Bagnera, M. de Franchis “[Sur les surfaces hyperelliptiques. ]{}” [ Comptes Rendus Acad. Sci. Paris, 145 ]{}, (1907), 747-749. L. Bagnera, M. de Franchis “[Sopra le superficie algebriche che hanno le coordinate del punto generico esprimibili con funzioni meromorfe quadruplamente periodiche di due parametri, I, II. ]{}” [Rend. Accad. Lincei, 16 ]{}, (1907), 492-498, 596-603. L. Bagnera, M. de Franchis “[Le superficie algebriche le quali ammettono una rappresentazione parametrica mediante funzioni iperellittiche di due argomenti. ]{}” [Memorie Soc. Italiana delle Scienze, detta Accad. dei XL, 15 ]{}, (1908), 251-343. I. Barsotti, “[Medoti analitici per varietá abeliane in caratteristica positiva. Cap. 1- 2 , Cap. 3-4 , Cap. 5, 6, 7 ]{}” [Ann. Sc. Norm. Super. Pisa, Sci. Fis. Mat., III. ]{}[Ser. 18, ]{} (1964), 1-25 , [Ser. 19, ]{} (1965), 277-330, (1965), 481- 512 ; [Ser. 20, ]{}(1966), 101-137, 331-365 . I. Barsotti, “[Considerazioni sulle funzioni theta ,]{}” [Sympos. Math., Roma 3, Probl. Evolut. Sist. solare, Nov. 1968 e Geometria, Febb. 1969]{}, (1970), 247-277 . I. Barsotti, “[Bivettori,]{}” in ’Algebraic geometry, int. Symp. Centen. Birth F. Severi’, Roma 1979, [Symp. Math. 24]{}, (1981), 23-63 . I. Barsotti, “[Differential equations of theta functions, Continuation.]{}” [Rend. Accad. Naz. Sci. Detta XL, V. Ser., Mem. Mat. 7, No.1.]{} , (1983), 227-276 and [Mem. Mat. 9]{}, (1985), 215-236 . W. Barth, C. Peters and A. Van de Ven, “[Compact Complex Surfaces]{}”, [Ergebnisse der Mathematik und ihrer Grenzgebiete, Folge 3, B.4, Springer-Verlag]{}, (1984), 304 pp.. I. Bauer, F. Catanese, “[Symmetry and variation of Hodge structure]{}”, preprint 2002. A. Beauville, “[L’application canonique pour les surfaces de type general]{}”,[Inv. Math. 55 ]{},(1979), 121-140. A. Beilinson, “[Coherent sheaves on ${\ensuremath{\mathbb{P}}}^n$ and problems of linear algebra]{}”, [ Funkt. Anal. i Pril. 12 (3)]{}, (1978), 68-69, translated in [Funct. Anal. Appl. 12 ]{}, 214-216. I.N. Bernstein, I.M. Gelfand, S.I. Gelfand, “[Algebraic bundles on ${\ensuremath{\mathbb{P}}}^n$ and problems of linear algebra]{}”, [ Funkt. Anal. i Pril. 12 (3)]{}, (1978), 66-68, translated in [Funct. Anal. Appl. 12 ]{}, 212-214. L. Bianchi, “[Lezioni sulla teoria delle funzioni di variabile complessa]{}”, [ Spoerri, Pisa]{}, (1916). U. Bottazzini, “[“Algebraic truths” vs “geometric fantasies”: Weierstrass’ response to Riemann. ]{}” Li, Ta Tsien (ed.) et al., Proceedings of the international congress of mathematicians, ICM 2002, Beijing, China, August 20-28, 2002. Vol. III: Invited lectures.[Beijing: Higher Education Press]{}. 923-934 (2002). A.Canonaco, “["A Beilinson-type theorem for coherent sheaves on weighted projective spaces ]{}”, [ Jour. of Algebra 225]{}, (2000), 28-46. A.Canonaco, “[“The Beilinson complex and canonical rings of irregular surfaces”]{}”, [ Tesi di Perfezionamento, Scuola Normale Superiore, Pisa]{}, (2002). F. Capocasa - F. Catanese, “[Periodic meromorphic functions.]{}” [Acta Math. 166]{} (1991) 27-68. G. Castelnuovo, “[Osservazioni intorno alla geometria sopra una superficie. Note I, II]{}” [Rendiconti del R. Ist. Lombardo, s.2, vol. 24]{} (1891) pages 246-254 and 255-265 of the Memorie Scelte. G. Castelnuovo, “[Le correspondenze univoche tra gruppi di $p$ punti sopra una curva di genere $p$.]{}” [Rendiconti del R. Ist. Lombardo, s.2, vol. 25]{} (1893) pages 79-94 of the Memorie Scelte. G. Castelnuovo, “[Gli integrali semplici appartenenti ad una superficie irregolare.]{}” [Rendiconti della R. Accad. dei Lincei, s.V, vol. XIV]{} (1905) pages 473-500 of the Memorie Scelte. G. Castelnuovo, “[Sulle superficie aventi il genere aritmetico negativo.]{}” [Rendiconti del Circ. Mat. di Palermo, t. XX ]{} (1905) pages 501-506 of the Memorie Scelte. G. Castelnuovo, “[Memorie scelte.]{}” [Zanichelli, Bologna]{} (1937). G. Castelnuovo, F. Enriques “[Die algebraischen Flächen vom Gesichtspunkte der birationalen Transformationen aus.]{}” [Enzycl. der math. Wissensch. III , 2, 6-6b]{} (1914), 674-768. F. Catanese, “[Babbage’s conjecture, contact of surfaces, symmetric determinantal varieties and applications.]{}”, [Inv. Math. 63]{} (1981), 433-465. F. Catanese, “[Commutative algebra methods and the equations of regular surfaces.]{}” in ’Algebraic Geometry Bucharest 1982’, [Springer Lect. Notes in Math. 1056]{} (1981), 68-111. F. Catanese, “[Moduli and classification of irregular Kähler manifolds (and algebraic varieties) with Albanese general type fibrations. Appendix by Arnaud Beauville.]{}” [Inv. Math. 104]{} (1991) 263-289; Appendix 289 . F. Catanese, “[Recent results on irregular surfaces and irregular Kähler manifolds.]{}” in ’Geometry and complex variables’ [Lecture Notes in pure appl. Math. 132, Marcel Dekker, New York]{} (1991), 59-88. F. Catanese, “[Homological algebra and algebraic surfaces]{}”, in ’Algebraic Geometry’, J.Kollár. et al. ed. [Proc. Symp. Pure Math. 62, Amer. Math. Soc., Providence R.I. ]{} (1997) , 3 - 56. F. Catanese, “[Deformation types of real and complex manifolds]{}”, in the Proc. of the Chen-Chow Memorial Conference, Advanced Topics in Algebraic Geometry and Algebraic Topology, [Nankai Tracts in Math. 5, World Scientific ]{} (2002) , 193-236. F. Catanese, “[Moduli Spaces of Surfaces and Real Structures]{}”, [Ann. Math. 158 ]{} (2003), n.2, 539-554. F. Catanese, “[Deformation in the large of some complex manifolds, I]{}”, preprint 2002, to appear in a Volume in Memory of Fabio Bardelli, Ann. Mat. pura e appl. F. Catanese, [Q.E.D. for algebraic varieties]{}, Preprint (2002). F. Catanese, C. Ciliberto, “[Symmetric products of elliptic curves and surfaces of general type with $p_g = q=1$.]{}” [ J. Algebr. Geom. 2, No.3]{}, 389-411 (1993). F. Catanese, C. Ciliberto, Mendez-Lopez, “[On the classification of irregular surfaces of general type with nonbirational bicanonical map,]{}” [ Trans. Am. Math. Soc. 350, No.1]{}, (1998), 275-308. F. Catanese, F.O. Schreyer [Canonical projections of irregular algebraoc surfaces]{}, in ’Algebraic Geometry. A volume in memory of Paolo Francia, [De Gruyter, Berlin, New York ]{}(2002), 79-116. A. Cayley, “[Desiderata and suggestions n.1. The theory of groups]{}”, [Amer. Jour. Math, Vol. 1, 1]{} (1878), 50-52. A. Cayley, “[Desiderata and suggestions n. 2. The theory of groups: graphical representation]{}”, [Amer. Jour. Math, Vol. 1, 2]{} (1878), 174-176. C. Chabauty, “[Limite d’ensembles et gŽomŽtrie des nombres]{}”,[Bull. Soc. Math. France 78]{}, (1950), 143–151. S.S. Chern, “[Complex manifolds]{}”, [Publ. Mat. Univ. Recife]{} (1958). W.L. Chow, “[Abstract theory of the Picard and Albanese varieties]{}”, [Ann. of Math. ]{}, (1959), . C. Ciliberto, “[Canonical surfaces with $p_g=p_a=4$ and $K^2 = 5,\dots,10$,]{}”, [Duke Math. J. 48]{}, (1981), 121-157. C.Ciliberto, “[A few comments on some aspects of the mathematical work of F. Enriques.]{}” in ’Geometry and complex variables’ [Lecture Notes in pure appl. Math. 132, Marcel Dekker, New York]{} (1991), 89-109. C. Ciliberto, “[M. de Franchis and the theory of hyperelliptic surfaces]{}”, in ’Studies in the history of modern mathematics’, [Supplemento Rend. Circ. Mat. di Palermo, S.II 55 ]{}(1998), 45-73. C. Ciliberto, M.Mendes Lopes, “[On surfaces with $p_g = q =2$ and non birational bicanonical map]{}”, [Adv. in Geom. 2]{} (2002), n. 3, 281-300. C. Ciliberto, M.Mendes Lopes, “[On surfaces with $p_g = 2, q =1$ and non-birational bicanonical map]{}”, in ’Algebraic Geometry. A volume in memory of Paolo Francia, [De Gruyter, Berlin, New York ]{}(2002), 117-126. C. Ciliberto, E. Sernesi, “[Some aspects of the scientific activity of Michele de Franchis]{}”, in ’Opere di de Franchis’, [Supplemento Rend. Circ. Mat. di Palermo, S.II 27 ]{}(1991), 3-36. Coble, Arthur B. “[Algebraic geometry and theta functions]{}”. Revised printing. [American Mathematical Society Colloquium Publication, vol. X American Mathematical Society, Providence, R.I]{}, (1929), (II edition 1961) vii+282 pp . M. Cornalba, “[Complex tori and Jacobians]{}”. [Complex Anal. Appl., int. Summer Course Trieste 1975, Vol. II]{}, (1976) 39-100 . P. Cousin, “[Sur les fonctions periodiques]{}” [Ann. Sci. École Norm. Sup. 19]{} (1902), 9-61. Davenport, J.H. ”[Algorithms for the integration of algebraic functions. ]{}” Symbolic and algebraic computation, EUROSAM ’79, int. Symp., Marseille 1979,[Lect. Notes Comput. Sci. 72]{}, 415-425 (1979). Davenport, J.H. ”[The computerisation of algebraic geometry. ]{}” Symbolic and algebraic computation, EUROSAM ’79, int. Symp., Marseille 1979,[Lect. Notes Comput. Sci. 72]{}, 119-133 (1979). Davenport, James Harold ”[On the integration of algebraic functions. ]{}” [Lecture Notes in Computer Science, 102. Berlin Heidelberg New York: Springer-Verlag]{} 197 p. (1981). U. Dini, ”[Fondamenti per la teorica delle funzioni di variabile reale. ]{}” [Spoerri, Pisa]{}, (1878). M. de Franchis, ”[Sulle superficie algebriche le quali contengono un fascio irrazionale di curve ]{}”, [Rendiconti Circ. Mat. di Palermo, XX]{}(1905), 49-54. M. de Franchis, ”[Dimostrazione del teorema fondamentale sulle superficie iperellittiche ]{}”, [Rendiconti Accad. Lincei, 24]{}(1936), 3-6. M. de Franchis, ”[Sulla classificazione delle superficie iperellittiche ]{}”, [Scritti matematice in onore di Berzolari]{}(1936), 613-615. J. Dieudonné, ”[Cours de géométrie algébrique, vol. 1,2 ]{}”, [Collection SUP 10,11, Presses Universitaires de France ]{}(1974), 234 pp. 222 pp., P. Deligne, ”[Variétés de Shimura : interprétation modulaire, et techniques de construction de modéles canoniques. ]{}”, in ’Automorphic forms, representations, and L-functions’, [Proc.Symp. Pure Math. 33, vol. 2, A.M.S.]{}(1979), 247-290. R. Donagi, ”[The tetragonal construction]{}”, [Bull. Am. Math. Soc., New Ser. 4]{}, (1981) 181-185 . D. Eisenbud, G. Fløystad, F.-O. Schreyer, ”[Sheaf cohomology and free fesolutions over exterior algebras]{}”, preprint, math.AG/0104203. “[Encyclopedic dictionary of mathematics. Vol. I. Abel to multivariate analysis. ]{}” Translated from the second Japanese edition. Edited by Sh™kichi Iyanaga and Yukiyoshi Kawada. [MIT Press, Cambridge, Mass.-London]{}, (1977). xv+883 pp. “[Encyclopedic dictionary of mathematics. Vol. II. Networks to zeta functions.]{}” Translated from the second Japanese edition. Edited by Sh™kichi Iyanaga and Yukiyosi Kawada. [MIT Press, Cambridge, Mass.-London]{}, 1977. pp. 885–1750. F. Enriques, ”[Ricerche di geometria sulle superficie algebriche.]{}” [Memorie Accad. Torino, s.2 vol. 44]{}(1893), 171-232. F. Enriques, ”[Sulla proprieta’ caratteristica delle superficie irregolari.]{}” [Rend. Accad. Sci. Bologna, nuova Serie vol. 9]{}(1905), 5-13. F. Enriques, ”[Le superficie algebriche.]{}” [Zanichelli, Bologna]{}, (1949), 464 pp. (Translation by F.Catanese, C. Ciliberto, R. Pardini, to be published by Canbridge University Press). F. Enriques, ”[Memorie scelte di geometria, vol. I, II, III.]{}” [Zanichelli, Bologna]{}, (1956), 541 pp.,(1959), 527 pp.,(1966), 456 pp. . F. Enriques, F. Severi, ”[Intorno alle superficie iperellittiche.]{}” [Rend. Accad. Lincei, s. V vol. 16]{}(1907), 443-453. F. Enriques, F. Severi, ”[Intorno alle superficie iperellittiche irregolari.]{}” [Rend. Accad. Lincei, s. V vol. 17]{}(1908), 4-9. F. Enriques, F. Severi, ”[Mémoire sur les surfaces hyperelliptiques, I.]{}” [Acta Math. vol. 32]{}(1909), 283-392. F. Enriques, F. Severi, ”[Mémoire sur les surfaces hyperelliptiques, II.]{}” [Acta Math. vol. 33]{}(1910), 223-403. G. Faltings, ”[EndlichkeitssŠtze fŸr abelsche VarietŠten Ÿber Zahlkšrpern.]{}” [Invent. Math. 73 ,no. 3]{}(1983), 349–366. ”[Erratum: "Finiteness theorems for abelian varieties over number fields]{}”. [Invent. Math. 75, no. 2]{} (1984), 381. G. Faltings, C.L. Chai, ”[Degeneration of abelian varieties.]{}” [Ergebnisse der Mathematik und ihrer Grenzgebiete (3) , 22. Springer-Verlag, Berlin]{}, ( 1990). xii+316 pp. G. Fischer, (ed.) ”[Mathematische Modelle.]{}” Aus den Sammlungen von Universitäten und Museen. Mit 132 Fotografien. (Mathematical models. From the collections of universities and museums. With 132 photographs). Bildband und Kommentarband. [Braunschweig/Wiesbaden: Friedr. Vieweg & Sohn]{}, (1986), XII, 129 pp.; VIII, 89 pp.. A. Genocchi, ”[Calcolo differenziale]{}”, con aggiunte del prof. G. Peano [Torino]{}, (1884). M. Green, R. Lazarsfeld ”[Deformation theory, generic vanishing theorems, and some conjectures of Enriques, Catanese and Beauville]{}”. [Inv. Math. 90]{}, (1987), 416-440. M. Green, R. Lazarsfeld ”[Higher obstructions to deforming cohomology groups of line bundles]{}”. [Jour. Amer. Math. Soc. 4]{}, (1991), 87-103. P. Griffiths, ”[Periods of integrals on algebraic manifolds I,II]{}”. [Amer. Jour. Math. 90]{}, (1968), 568-626, 805-865. P. Griffiths, ”[Periods of integrals on algebraic manifolds III]{}”. [Publ. Math. I.H.E.S. 38]{}, (1970), 125-180. P. Griffiths, (ed) ”[Topics in transcendental algebraic geometry]{}”. [Annals of Math. Studies 106, Princeton Univ. Press]{}, (1984), 316 pp.. C.D. Hacon, ”[Fourier transforms, generic vanishing theorems and polarizations of abelian varieties]{}”. [Math. Zeit. 235]{}, (2000), 717-726. C.D. Hacon, R. Pardini, ”[Surfaces with $p_g = q =3$]{}”, [Trans. Amer. Math. Soc. 354]{} (2002), n.7, 2631-2638. W.V.D. Hodge, ”[Theory and application of the harmonic integrals]{}”, [ Cambridge Univ. Press]{} (1952). J.I. Igusa, ”[Theta functions ]{}”. [Die Grundlehren der mathematischen Wissenschaften, Band 194. Springer-Verlag, New York-Heidelberg]{}, (1972) x+232 pp. G.Kempf, ”[On the geometry of a theorem of Riemann]{}” . [ Ann. of Math. (2) 98 ]{}(1973), 178–185. G. Kempf, ”[ Projective coordinate rings of abelian varieties.]{}” [ Algebraic analysis, geometry, and number theory (Baltimore, MD, 1988) Johns Hopkins Univ. Press, Baltimore, M]{}, (1989), 225–235. V. Kharlamov and V. Kulikov,”[ On real structures of real surfaces . ]{}” [math.AG/0101098]{}. F. Klein, ”[ Über die Transformationen siebenter Ordnung der elliptischen Funktionen]{}”, [Math.Ann. 14 ]{}, (1878) . F. Klein, “[Gesammelte mathematische Abhandlungen , Bände I, II, III]{}”. Band I: Liniengeometrie, Grundlegung der Geometrie, zum Erlanger Programm. Edited by R. Fricke und A. Ostrowski (with additions by Klein), ii+xii+612 pp. (1 plate). Band II: Anschauliche Geometrie, Substitutionsgruppen und Gleichungstheorie, zur mathematischen Physik. Edited by R. Fricke und H. Vermeil (with additions by Klein), ii+vii+713 pp. Band III : Elliptische Funktionen, insbesondere Modulfunktionen, hyperelliptische und Abelsche Funktionen, Riemannsche Funktionentheorie und automorphe Funktionen, Anhang verschiedene Verzeichnisse. Edited by R. Fricke, H. Vermeil und E. Bessel-Hagen (with additions by F. Klein), ii+ix+774+36 pp. Reprint of the first edition [Julius Springer Verlag, Berlin]{}, (1923), [Springer-Verlag, Berlin-New York]{}, 1973. K. Kodaira ,“[The theory of harmonic integrals and their applications to algebraic geometry]{}” (1952), in “Kunihiko Kodaira: Collected Works”, [Princeton Univ. Press and Iwanami Shoten]{}, (1975), 488-582 . K. Kodaira, J. Morrow ,“[Complex manifolds]{}” [Holt, Rinehart and Winston]{}, New York-Montreal, Que.-London (1971). K. Kodaira , D. Spencer ”[On deformations of complex analytic structures I-II]{}”, [Ann. of Math. 67]{} (1958), 328-466 . A. Krazer, “[Lehrbuch der Thetafunktionen]{}” [Teubner, Leipzig]{}, (1903). A. Krazer, W. Wirtinger “[Abelsche Funktionen und allgemeine Thetafunktionen]{}” [Enzykl. Math. Wiss. II B 7]{}, (1920), 604-873. S. Lang, C. ”[Abelian varieties. ]{}” [Interscience Publ. Inc. ]{}, (1959), reprinted by Springer Verlag (1983). H. Lange, C. Birkenhake, ”[Complex Abelian varieties. ]{}” [Grundlehren der mathematischen Wissenschaften 302, Springer Verlag ]{}, (1992). R.P. Langlands, ”[Problems in the theory of automorphic forms. ]{}”, in ’Lectures in modern analysis and applications’, [Springer Lecture Notes in Math. 170]{}(1970), 18-86. R.P. Langlands, ”[Some contemporary problems with origins in the Jugendtraum. ]{}”, in ’Mathematical developments arising from Hilbert problems’, [Proc.Symp. Pure Math. 28, vol. 2, A.M.S.]{}(1976), 401-418. R.P. Langlands, ”[Automorphic representations, Shimura varieties, and motives, ein Märchen. ]{}”, in ’Automorphic forms, representations, and L-functions’, [Proc.Symp. Pure Math. 33, vol. 2, A.M.S.]{}(1979), 205-246. H. Lebesgue, ”[Intégrale, longuer, aire. ]{}”, [ Annali di Mat. pura ed appl. 7]{},(1902), 231-358. S. Lefschetz, ”[On certain numerical invariants of algebraic varieties with application to abelian varieties ]{}”, [ Trans. Amer. Math. Soc. 22 ]{} no. 3,(1921), 327–406 and no. 4, (1921),407–482. M. Manetti, ”[On the Moduli Space of diffeomorphic algebraic surfaces]{}”, [Inv. Math. 143]{} (2001), 29-76. N. Manolache, F.O. Schreyer, ”[Moduli of $(1,7)$-polarized abelian surfaces via syzygies. ]{}” [ Math. Nachr. 226]{}, 177-203 (2001). T. Matsusaka, ”[On a theorem of Torelli. ]{}” [ Am. J. Math. 80]{}, 784-800 (1958). L. Migliorini, ”[A smooth family of minimal surfaces of general type over a curve of genus at most one is trivial. ]{}” [ Jour. Alg. Geom. 4]{}, (1995), 668-684. N. Mok, ”[The generalized theorem of Castelnuovo-de Franchis for unitary representations.]{}” in ’ Geometry from the Pacific Rim. Proceedings of the Pacific Rim geometry conference, National University of Singapore, Republic of Singapore, December 12–17, 1994,’ Berrick, et al.,ed., [ Walter de Gruyter, New York]{} (1997), 261-284. S. Mukai, ”[Duality between $D(X) and D(\hat{X})$ with its applications to Picard sheaves]{}”, [Nagoya Math. Jour. 81]{}, (1981), 153-175. S. Mukai, ”[Fano 3-folds]{}”, D. Mumford, ”[Lectures on curves on an algebraic surface]{}”. [Annals of Math. Studies, n. 59, Princeton Univ. Press]{}, (1966). D. Mumford, ”[On the equations defining abelian varieties.I, II, III]{}”, [ Invent. Math. ]{}1 (1966 )287–354, 3 (1967) , 75–135 , 3 (1967), 215–244. D. Mumford, ”[Rational equivalence of $0$-cycles on surfaces]{}”. [J. Math. Kyoto Univ. 9]{}, (1968) 195–204. D. Mumford, ”[Abelian varieties]{}” [Tata Institute of Fundamental Research Studies in Mathematics, Oxford University Press. VIII]{} (1970). D. Mumford, ”[Curves and their Jacobians]{}” [The University of Michigan Press, Ann. Arbor]{}, (1975) 104 p. . D. Mumford, ”[Tata lectures on theta. I]{}” With the assistance of C. Musili, M. Nori, E. Previato and M. Stillman. [Progress in Mathematics, 28. BirkhŠuser Boston, Inc., Boston, MA]{}, (1983). xiii+235 pp. ”[Tata lectures on theta. II. Jacobian theta functions and differential equations]{}”. With the collaboration of C. Musili, M. Nori, E. Previato, M. Stillman and H. Umemura.[Progress in Mathematics, 43. BirkhŠuser Boston, Inc., Boston, MA]{}, (1984) xiv+272 pp.. ”[Tata lectures on theta. III]{}” With the collaboration of Madhav Nori and Peter Norman. [Progress in Mathematics, 97. BirkhŠuser Boston, Inc., Boston, MA]{} (1991), viii+202 pp. s Norman, A.C.; Davenport, J.H. “[Symbolic integration - the dust settles? ]{}” Symbolic and algebraic computation, EUROSAM ’79, int. Symp., Marseille 1979, [Lect. Notes Comput. Sci. 72]{}, 398-407 (1979). P.Painlevé, “[Sur les fonctions qui admettent un théoreme d’ addition]{}”, [Acta Math., 27 ]{}, (1903). Patterson, S.J. “[Abel’s theorem. ]{}” Bottazzini, U. (ed.), Studies in the history of modern mathematics. IV. Palermo: [Circolo Matemˆtico di Palermo, Suppl. Rend. Circ. Mat. Palermo, II. Ser. 61]{}, 9-48 (1999) É. Picard, “[Sur la théorie des groupes et des surfaces algébriques. ]{}” [Rendiconti del Circ. Mat. di Palermo, IX ]{} (1895). P. Pirola, “[Algebraic surfaces with $p_g-q-3$ and no irrational pencils ]{}”, [Manuscripta Math. 108 ]{} (2002), n.2, 163-170. H. Poincaré, “[Sur la réduction des intégrales abéliennes. ]{}” [Bull. Soc. Math. France 12 ]{} (1884), 124-143. H. Poincaré, “[Sur les fonctions abéliennes. ]{}” [Acta Math. 26 ]{} (1902), 43- 98. H. Poincaré, “[Sur les courbes tracées sur les surfaces algebriques. ]{}” [Annales de l’ École normale Sup.,III, 27 ]{} (1910), 43- 98. R.H. Risch, “[On subvarieties of abelian varieties.]{}” [Inv. Math. 62]{}, (1981), 459-479. U. Bottazzini, A. Conte, P. Gario editors, F. Enriques , author, “[Riposte Armonie.]{}”, [Bollati Boringhieri, Torino]{} (1996), 722 pp. . S. Recillas, “[Jacobians of curves with $g^1_4 $’s are the Prym’s of trigonal curves.]{}” [Bol. Soc. Mat. Mex., II. Ser. 19]{}, (1974) 9-13 . R.H. Risch, “[The solution of the problem of integration in finite terms.]{}” [Bull. Amer. Math. Soc. 76]{}, (1970), 605–608. C. Rosati, “[Sulle matrici di Riemann. ]{}” [Rend. Circ. Mat. Palermo, 53 ]{}, (1929), 79-134. N. Schappacher, “[Développement de la loi de groupe sur une cubique. \[Development of the law of composition on a cubic curve\] ]{}” [Séminaire de ThŽorie des Nombres, Paris 1988–1989, Progr. Math., 91, BirkhŠuser Boston, Boston, MA]{},( 1990), 159–184. N. Schappacher, R. Schoof,“[Beppo Levi and the arithmetic of elliptic curves. ]{}” [Math. Intelligencer 18 ]{}(1996), no. 1, 57–69. F.O.Schreyer, “[Geometry and algebra of prime Fano 3-folds of genus $22$]{}”, [Comp. Math. ]{}, (2002) . G. Scorza, “[Sopra la teoria delle figure polari delle curve piane del $4^o$ ordine]{}”, [Ann.di Mat.(3) 2 ]{}, (1899), 155-202. G. Scorza, “[Intorno alla teoria generale delle matrici di Riemann]{}”, [Rend. Circ. Mat. Palermo 41 ]{}, (1916), 263-379. F. Severi, “[Sulle superficie che rappresentano le coppie di punti di una curva algebrica.]{}” [Atti della R. Accad. delle Scienze di Torino, 38]{}, (1903), 185-200. F. Severi, “[Il teorema d’ Abel sulle superficie algebriche.]{}” [Annali di Mat. s.III, t.XII]{}, (1905), 55-79. F. Severi, “[La serie canonica e la teoria delle serie principali di gruppi di punti sopra una superficie algebrica.]{}” [Comment. Math. Helv. 4]{}, (1932), 268-326. F. Severi, “[ Funzioni quasi-abeliane. ]{}” [Pontificiae Academiae Scientiarum Scripta Varia 4. Roma: Pontificia Academia Scientiarum. ]{}327 p. (1947). F. Severi, “[ Memorie scelte. Vol. I. ]{}” [Cesare Zuffi Editore. XX, Bologna]{}, 458 p. (1950). G. Shimura, ”[On analytic families of polarized abelian varieties and automorphic functions]{}”, [Ann. of Math. 78 ]{}, (1963), 149-193. C.L. Siegel, ”[Symplectic geometry]{}” [Amer. Jour. Math., n. 65]{} (1943), 1-86, reprinted by Academic Press, New York and London, (1964). C.L. Siegel, ”[Topics in complex function theory, Vol. III]{}” [Interscience Tracts in pure and appl. Math., n. 25, Wiley, N.Y.]{} (1973). J. T. Siu, ”[Extension of twisted pluricanonical sections with plurisubharmonic weight and invariance of semipositively twisted plurigenera for manifolds not necessarily of general type. ]{}” in ’Complex Geometry, Collections of papers dedicated to Hans Grauert, I. Bauer et al ed. [Springer Verlag]{}, 223-277 (2002). J. T. Tate, ”[The arithmetic of elliptic curves. ]{}” [Invent. Math. 23]{}, 179-206 (1974). R. Torelli, ”[Sulle varieta’ di Jacobi, I, II. ]{}” [Rendiconti R. Accad. dei Lincei 22-2]{}, (1913), 98-103, 437-441. B.M. Trager, ”[Integration of simple radical extensions. ]{}” Symbolic and algebraic computation (EUROSAM ’79, Internat. Sympos., Marseille, 1979), pp. 408–414, [Lecture Notes in Comput. Sci., 72]{}, Springer, Berlin-New York, (1979). F. Tricomi, ”[Funzioni ellittiche. ]{}” Second edition, [Nicola Zanichelli Editore, Bologna]{}, (1951), ix+343 pp. K. Ueno, “[Classification theory of algebraic varieties and compact complex spaces.]{}” [Lecture Notes in Mathematics 439,]{} Springer-Verlag, XIX, 278 p.(1975). K. Weierstraß, “[Einleitung in die Theorie der analytischen Funktionen. ]{}” , Vorlesung Berlin 1878, Lecture notes taken and with supplementary material by Adolf Hurwitz. Preface by R. Remmert, ed. Peter Ullrich. [Dokumente zur Geschichte der Mathematik 4. Deutsche Mathematiker Vereinigung, Freiburg; Friedr. Vieweg Sohn, Braunschweig]{}, (1988) xxx+184 pp.. A. Weil, “[ L’ arithmétique sur les courbes algébriques. ]{}” [ Acta Math. 52]{}, (1929), 281-315. A. Weil, “[ Sur les courbes algébriques et les variŽtŽs qui s’en deduisent. ]{}” [Actualités scientifiques et industrielles. 1041. Paris: Hermann & Cie. IV]{}, 85 p. (1948); Publ. Inst. Math. Univ. Strasbourg 7 (1945). A. Weil, “[Variétés abeliennes et courbes algébriques. ]{}” [Paris: Hermann & Cie.]{} 163 p. (1948). A. Weil, “[ Zum Beweis des Torellischen Satzes. ]{}” [ Nachr. Akad. Wiss. Gšttingen, Math.-Phys. Kl., IIa 1957]{}, 32-53 (1957). A. Weil, “[ Sur certains groupes d’opérateurs unitaires. ]{}” [Acta Math. 111]{}, (1964), 143-211 . A. Weil, “[Introduction a l’ étude des variétés kählériennes. ]{}” [Paris: Hermann & Cie.]{}, (1958). H.Weyl, “[Topologie und abstrakte Algebra als zwei Wege mathematischen Verständnisses]{}” [Unterrichtsblätter für Mathematik und Naturwissenschaften, band 38 ]{}, (1933), s. 177-188, also in H.Weyl, “[Gesammelte Abhandlungen. Bände I, II, III, IV.]{}” Edited by K. Chandrasekharan [Springer-Verlag, Berlin-New York ]{} (1968) ,Band I: vi+698 pp., Band II: iv+647 pp., Band III: iv+791 pp., Band IV: ix+694 pp.. H.Weyl, “[On generalized Riemann matrices]{}”, [Ann.Math. 35 ]{}, (1934), 714-729. H.Weyl, “[Generalized Riemann matrices and factor sets]{}”, [Ann.Math. 37 ]{}, (1936), 709-745. I. M. Yaglom,“[Felix Klein and Sophus Lie. Evolution of the idea of symmetry in the nineteenth century.]{}” Translated from the Russian by Sergei Sossinsky. Translation edited by Hardy Grant and Abe Shenitzer.[BirkhŠuser Boston, Inc., Boston, MA]{} (1988). xii+237 pp.. G. Zappa, “[The papers of Gaetano Scorza on group theory. ]{}” [ Atti Accad. Naz. Lincei, Cl. Sci. Fis. Mat. Nat., IX. Ser., Rend. Lincei, Mat. Appl. 2, No.2]{}, (1991), 95-101 . O. Zariski, “[Algebraic surfaces]{}”, [Ergebnisse der Math. und ihrer Grenzgebiete, Band III Heft 5 , Springer Verlag]{} (1935), second supplemented edition in [Ergebnisse 61]{}, (1970), 270 pp. [Author’s address:]{} Prof. Fabrizio Catanese\ Lehrstuhl Mathematik VIII\ Universität Bayreuth\ D-95440, BAYREUTH, Germany e-mail: [email protected] [^1]: The present research, an attempt to treat history and sociology of mathematics and mathematics all at the same time, took place in the framework of the Schwerpunkt “Globale Methode in der komplexen Geometrie”, and of the EC Project EAGER. [^2]: Was the greatest enemy of mathematics Alexander the Great, who cut the Gordian knot instead of peacefully writing a book about it? [^3]: This topic has in fact been treated quite extensively in the contribution by Ciliberto. [^4]: A star next to a theme denotes that a section of the dictionary is devoted to the discussion of the topic [^5]: Here and after, I will refer to the oral contributions given at the Abel Bicentennial Conference, and not to the articles published in this Volume. [^6]: We have noticed that another article devoted to this topic has appeared after we gave the talk, namely [@bott02] by U. Bottazzini. [^7]: Writes G. Fischer : “ There were certainly other reasons than economic for the waning interest in models. ...... More and more general and abstract viewpoints came to the forefront of mathematics....... Finally Nicolas Bourbaki totally banned pictures from his books.” [^8]: For many years on, until after the first world war, there used to be a course in Italian Universities entitled “Lezioni di analisi algebrica ed infinitesimale”. This shows that the birth of Analysis as a new separated branch, trying to appear more on the side of applied mathematics, is a relative novelty which ends the reconciliation made for the dualism Leibniz-Newton. [^9]: cf. the article by: J J O’Connor and E F Robertson in http://www-history.mcs.st-andrews.ac.uk/References/Abel.html, citing his letter to Holmboe from Berlin. Here we can read : “ In other words, the most important parts of mathematics stand without foundation”. [^10]: To be perfectly honest, all we know is that in 1826 he signed himself in at the “Goldenes Schiff” in Predazzo as ‘Abel, professore della geometria.’ [^11]: This important article was lying at the centre of the contribution by Griffiths, and is also amply commented upon in Kleiman’s contribution. [^12]: Observe that we wrote ${\ensuremath{\mathbb{C}}}[x,y]$ instead of ${\ensuremath{\mathbb{R}}}[x,y]$. It is commonly agreed that, if we want to summarize in two words which the greatest contribution of Abel and Jacobi was, then it was to consider the elliptic integrals not just as functions of a real variable, but also as functions of a complex variable. So, we owe to them the birth of the theory of holomorphic functions. [^13]: Since an elliptic integral is the one where we consider a square root $\surd\overline{ P(x) } $ where $P$ is a polynomial of degree $3$ or $4$, one can reduce it, after applying a projective transformation of the line ${\ensuremath{\mathbb{P}}}^1$, to a square root $\surd \overline{ Q(y^2) }$, where $Q$ is quadratic, and then we can view it as an integral on the unit circle, whose projection to the $x$-line ${\ensuremath{\mathbb{P}}}^1$ yields a double cover. [^14]: The second volume of the edition [@abel81] by Sylow and Lie, Christiania 1881, contains the unpublished papers of Abel, with a few exceptions. As the editors remark, this edition, posterior by more than 30 years to the edition of 1839 edited by the friend and colleague of Abel, Holmboe, was financed by the Norwegian Parliament after the great demand for Abel’s works (Holmboe’s edition went rapidly out of print) of which especially the French Mathematical Society made itself interpreter. The two editors decided to omit in the second volume three articles which were partly based on an erroneous memoir of his youth, written in Norwegian, where Abel thought he could prove that the general equation of degree n can be solved by radicals. This is the only published article which is not appearing in the Holmboe edition, nor in Volume I of the edition by Sylow and Lie. [^15]: Here and elsewhere, \[..\] stands for an addition of the present author [^16]: I.e., by rational functions or by logarithms (more generally, one can consider algebraic functions and logarithms of these). [^17]: Why was not Abel using the notation $\partial / \partial r_j$ ? [^18]: As amply illustrated by Ciliberto in his contribution. [^19]: We heard this claim from S.J. Patterson in Göttingen, soon after he had written the article [@pat99]. [^20]: The letter $u$ clearly stands for “uniformization map”. [^21]: Which however, at the time of Torelli’s paper, 1913, and also afterwards, were called Picard Varieties . With the Prize winning Memoir by Lefschetz, [@lef], the terminology Abelian varieties became the only one in use. [^22]: One could argue whether the beauty of 1-dimensional geometry bears similarities to the surprising isomorphisms of classical groups of small order. These also, by the way, are related to Abel’s heritage. For instance, the isomorphism ${\ensuremath{\mathcal{S}}}_4 \cong A(2, ({\ensuremath{\mathbb{Z}}}/2 ))$, whence ${\ensuremath{\mathcal{S}}}_3 \cong A(2, ({\ensuremath{\mathbb{Z}}}/2 ))/({\ensuremath{\mathbb{Z}}}/2 )^2 \cong PGL (2,({\ensuremath{\mathbb{Z}}}/2 )) \cong Aut ({\ensuremath{\mathbb{H}}})$. Here, ${\ensuremath{\mathbb{H}}}$ is the group of order $8$ of unit integral quaternions, and the last isomorphism is related to some later development in the theory of algebraic curves, namely to Recilla’s tetragonal construction, see [@rec], [@don] [^23]: However, as well known, there are Italians with techniques and ideas, and others who are not perfect. In particular, while it is not difficuult to find errors or wrong assertions in Enriques and Severi, it is rather hard to do this with Castelnuovo. [^24]: The second characterization is very useful for the study of fibre bundles of Abelian varieties, as we had opportunity to experience ourselves, cf. [@cat02-2] [^25]: In his unpublished Fermi Lectures held in Pisa in 1976, D. Mumford explained how this approach was working, using clarifications due to Chabauty, [@chab] [^26]: Not always mathematical concepts have a birthday. But sometimes it happens , as for another Legacy by Abel: the concept of an abstract group was born in 1878, with a “Desiderata and suggestions” by A. Cayley [@cay1]. In this note appears first the multipication table of a group. Immediately afterwards, however, in [@cay2], Cayley realized that it is much better to work with what is now called the “Cayley graph” of a group endowed with a set of generators. [^27]: “With proof communicated to the author by prof. Osgood verbally in Nov. 1927”. [^28]: Edited by Zanichelli in the respective years 1915, 1918, 1924, 1934. [^29]: Spoerri, Pisa, 1916 [^30]: De hoc satis, because the talks by Faltings and Wiles were exactly dealing with these aspects of Abel’s legacy. [^31]: "Theorie der Thetafunktionen, Teubner, 1894. [^32]: Giacomo Albanese emigrated from Italy in 1936 to Sao Paulo, Brasil, where, soon after the war, he became closely acquainted with André Weil, who taught there, as well as Zariski. Weil is responsible for the name “Albanese variety”, but Ciliberto and Sernesi in [@albcp] write: “Whilst the attribution of this concept to Albanese is dubious, ...” . Indeed the basis for this is an article, [@alb34], where Albanese studies correspondences beween algebraic surfaces through the consideration of the induced action on the space of holomorphic $1$-forms. The coupled names “Albanese” and “Picard” appear in the title of the article by W.L. Chow “ Abstract theory of the Picard and Albanese varieties”,[@chow]. Indeed, in the 50’s, one main purpose was to distinguish among the two dual varieties, which are only isogenous, and not in general isomorphic. According to the historical note of Lang on page 52 of [@langav], Matsusaka was the first to give a construction of the Albanese variety using the generic curve. [^33]: In the commemoration opening vol. I of his selected papers in geometry, [@enrMS], IX - XXII. [^34]: Michele de Franchis, born in Palermo, 1875, was very much influenced by the teaching of Scorza. He was also the Director of the Rendiconti Circolo Matematico di Palermo in the years 1914-1946, as the successor of the founder G.B. Guccia [^35]: Lefschetz was instead recipient of the Bordin Prize in 1919 for the cited Memoir [@lef]. [^36]: Computers however have not only helped us to do some explicit calculations, but, according to F.O. Schreyer, they have also made us wiser: by showing us explicit equations which need several pages to be written down, they make us wonder whether having explicit equations means any better understanding. [^37]: I came to read this beautiful paper under the instigation of Francesco Gherardelli. He explained to me that the citation from “Il teatro alla moda” by Benedetto Marcello: “As a first duty shall the modern poet ignore all about the ancient Roman and Greek poets, because these last too ignored everything about modern poetry” was motivated by the extreme difficulties that Aldo had when he wanted to lecture in Paris about this result without first explaining the excellence of the rings of coefficients he was using, or use the notation $g^1_r$ without explaining the representability of such functors.	{ "pile_set_name": "ArXiv" }
[Remarks on DSR and Gravity]{} [F. Hinterleitner]{} [Institute of Theoretical Physics and Astrophysics,\ Faculty of Science of the Masaryk University,\ Kotlářská 2, 611 37 Brno, Czech republic]{} Modifications of Special Relativity by the introduction of an invariant energy and/or momentum level (so-called Doubly Special Relativity theories, DSR) or by an energy-momentum dependence of the Planck constant (Generalized Uncertainty Principle, GUP) are compared with classical gravitational effects in an interaction process. For the low energy limit of the usual formulations of DSR to be equivalent to Newtonian gravity, a restrictive condition is found. GUP yields an effective repulsion, in analogy to gravitational repulsion in loop quantum cosmology. Introduction ============ Tentative quantum theories of gravity – string theory as well as loop quantum gravity or non-commutative geometry – indicate the existence of an invariant length scale of the order of magnitude of the Planck length, which is in obvious contradiction with standard Lorentz symmetry, when taken seriously to arbitrary small scales. For this reason in the last years various attempts were made to modify Special Relativity (SR) [@GAC; @MS] in such a way that one (or more) invariant quantities in addition to the speed of light would be reconciled with the relativity principle. Theories of this kind are called Doubly or Deformed Special Relativity (DSR). Early examples were formulated in momentum space by the application of nonlinear representations of the Lorentz group to the energy and momentum of physical objects, such that there is an invariant value of energy and/or momentum of the order of the Planck energy and the Planck momentum. Technically this may be achieved by splitting energy/momentum variables into “physical" ones, usually denoted by $E$ and $\vec p$, and “pseudo"-variables (sometimes called platonic variables) $\varepsilon$ and $\vec\pi$, with both kinds of variables related by an invertible nonlinear transformation. Pseudo-variables satisfy the usual linear relations of SR, in consequence, the physical ones are acted upon by the (boost sector of the) Poincaré group in a deformed, nonlinear way. Note that the denomination ‘deformed’, common in the present context, denotes merely the action of the Lorentz group and has nothing to do with deformations of the Poincaré Lie algebra. Pseudo-variables, although being mere auxiliary quantities in the construction of modified Lorentz transformations, have the following formal significance. Providing the linear representation of the Poincaré group, they carry the usual vector space structure of SR momentum space, whereas the space of $E$ and $\vec p$ becomes curved. Therefore, when subsystems are composed to a whole, it is the pseudo-variables which must be additive in order to preserve the underlying Lorentz group structure. For this reason in the calculations of reaction thresholds or cross sections of particle interactions in the framework of DSR, conservation rules are formulated in terms of them [@JV; @LN]. This leads to a violation of ordinary energy/momentum conservation; particularly the total energy/momentum of a composite system never exceeds the invariant values, as long as the energy/momentum of its components are below them. (This is the so-called soccer-ball problem, see, e.g. [@Sta; @soc].) Possible physical consequence are anomalies of reaction thresholds and an energy dependence of the speed of light in some versions of DSR. Even if the effect is tiny (of the order [photon energy/Planck energy]{}), it might become measurable when photons run over cosmic distances. A recently observed slight energy dependence of the time of arrival of photons from a $\gamma$ ray flare might be interpreted in this sense, if we knew the mechanism of emission, see [@magic]. More advanced versions of DSR are completed by modified space-time Lorentz transformations, associated to the transformations in momentum space in different ways. Some of these attempts assume momentum space to be a de Sitter space [@sitt], other ones make use of Hopf algebra techniques [@hopf]. These approaches lead to non-commutative space-time, denoted by $\kappa$-Minkowski space [@kappa]. In this framework the Poincaré lie algebra itself is deformed. Other methods lead to energy-momentum dependent space-time metrics, called “rainbow metrics" [@rain], recently investigated with the formalism of Finsler geometries [@fisler]. Nevertheless, for the purpose of the present paper we are going to make some simple physical considerations only in momentum space. DSR, at least in its original guise, is formally independent of gravity, even if the corrections it makes to SR are interpreted as effective description of the imprints of quantum geometry in form of some texture of flat space, present even in the limit of vanishing gravitational field. Thereby gravity is mainly needed as an explanation for the departure of physical energy-momentum conservation in DSR in such a way that the gravitational field is thought as a reservoir for the non-conserved energy and momentum, without specification of its properties at extremely small distances and of the way it interacts with matter. A more concrete relation to gravity exists in the Hopf algebra approach, as Hopf algebra methods appear as a branch of quantum gravity research in their own right [@maj]. Concerning the relation of DSR to Loop Quantum Gravity, in [@abl] there is a rather heuristic derivation of modified energy-momentum relations, inherent to DSR, from spatial discreteness, but it is also explained that a violation of Lorentz invariance can neither be derived nor excluded from the present form of Loop Quantum Gravity. Although DSR is supposed to reproduce gravitational effects in the quantum gravity regime, it is an open issue, how it compares to classical gravity. One may ask whether its low energy limit should be in accordance with the effects of a classical gravitational field, the low energy limit of quantum gravity. In the literature there are essentially two different points of view, relating DRS to different partial aspects of full General Relativity (GR): - No relation to classical gravity is proposed in [@top], where DSR represents the topological degrees of freedom of the gravitational field, i.e. the remnant, when the local degrees of freedom of the gravitational field is removed. This approach is supported by the successful formulation of 2+1 gravity, well known to be a topological theory, as kind of DSR [@2+1]. - On the other hand, in [@rain] a “Correspondence Principle" is formulated in the form that in the low-energy limit of DSR classical GR should be recovered. A logically different approach to the Lorentz invariance problem is to separate between a particle’s energy and momentum on the one hand and the frequency and wave vector of the associated quantum wave function on the other hand, with the advantage of an immediate connection between the formulations in momentum and in position space and a clear physical meaning of all variables in the game [@SH]. An important consequence is an energy dependent Planck constant, leading to modified uncertainty relations (Generalized Uncertainty Principle, GUP) and possibly, but not necessarily, an energy dependent speed of light. The above-mentioned correspondence principle and the interpretation of DSR as topological gravity being not equivalent, it is worthwhile to compare low-energy approximations of DSR and GUP to classical gravity in its simplest, i.e. Newtonian form. The program of the present paper is to perform an elementary test, namely to apply these approximations to a scattering process, as possible physical effects always arise in connection with interactions between moving objects. In the current DSR philosophy pseudo-variables must be associated to interaction processes, so it is sufficient and logically convenient to define them only in interaction regions, as it was done in [@SH], whereas the asymptotic variables are the physical ones. The inclusion of interacting objects opens a door to the introduction of gravity as a further interaction, leading to a small perturbation, and not as a quantum property of space. The only further ingredients, used beside Newtonian gravity in the next sections, are the mass-energy relation and the de Broglie wavelength of particles. In detail we will use the following approximations: - Newtonian gravity, understood as lowest-order approximation of general relativity, in other words, as a simplified substitute for a curved background. - Quantum field theory in first-order perturbative approximation. - A general lowest-order ansatz for DSR-like corrections of SR. We are going to compare only unspecified interactions in the absence and in the presence of classical gravity, so the considerations are independent of specific high-energy quantum effects, like varying coupling constants. In the next two sections, $c$ and $\hbar$ are set equal to one, they will have to be restored in section 4. Gravity in two-particle interactions ==================================== Central collision ----------------- We consider the scattering of two identical particles with repulsive interaction in the centre-of-momentum reference frame. In perturbative quantum field theory the free particles approach each other, exchange virtual interaction particles and then move away freely. During the free motion the gravitational interaction of particles does not play a role, but we will take it into account in the interaction process. If we assume first a central collision and restrict ourselves to first-order Feynman diagrams, we can describe the situation as follows. At the interaction vertices, when the particles reach a certain minimal distance, they stop and their kinetic energy materializes as a virtual exchange particle. Provided the asymptotic kinetic energy is high enough, the gravitational field of the virtual particle furnishes a significant amount of additional energy for the interaction process in comparison with the gravitation-less interaction, and the particles come closer to each other, as if they had a higher asymptotic kinetic energy. In the following we are going to formulate these considerations up to first order in the gravitational constant $G$. We assume two particles with masses $m$ and (absolute values of) asymptotic momenta $p$. In the absence of gravity, at the interaction vertices, with the particles at their minimal distance denoted by $2r_0$, the asymptotic kinetic energy of both of them transforms into the energy of the interaction particle, $$\label{Ek} {\cal E}=2\left(\sqrt{p^2+m^2}-m\right).$$ When Newtonian gravity is added to the system and the particles are assumed to be massive, there are two effects (the mutual attraction of the two rest masses is considered as negligible): Due to gravitational attraction each particle has a potential energy $$\Delta E_1=-\frac{G{\cal E}m}{r_0}$$ in the moment when it stops at a distance $r_0$ from the scattering centre. For a rough estimate of the minimal distance in terms of the asymptotic kinetic energy we take the de Broglie wavelength $\lambda$ of the exchanged particle, whose mass is assumed to be negligible in comparison with its total energy, so that the transmitted momentum is approximately equal to $\cal E$, $$\label{l} 2r_0\approx\lambda\approx\frac{1}{\cal E},$$ and the gravitational potential energy of each of the scattered particles becomes $$\label{E1} \Delta E_1\approx-2\,\frac{m{\cal E}^2}{m_P^2}=-8\,\frac{m\left(\sqrt{p^2+m^2}-m\right)^2}{m_P^2},$$ where we have introduced the Planck mass $m_P=1/\sqrt{G}$, which, in our units, stands also for the Planck energy and the Planck momentum. The second effect, which is independent of the mass of the particles, is the self-energy $\Delta\cal E$ of the exchange particle, whose order of magnitude is estimated by modeling it ad hoc as a homogenous sphere of radius $r_0$, $$\label{cale} \Delta{\cal E}\approx-\frac{3}{5}\frac{G{\cal E}^2}{r_0}=-\frac{6}{5}\frac{{\cal E}^3}{m_P^2}.$$ One half of $\Delta\cal E$ is associated to each of the scattered particles to give rise to an energy difference $$\Delta E_2\approx-\frac{24}{5}\frac{\left(\sqrt{p^2+m^2}-m\right)^3}{m_P^2}.$$ Of course, in view of our rough approximations and the homogenous sphere being rather an indication of ignorance than a seriously-meant model, the factors 6/5 and 24/5 appear ridiculous and will be absorbed into order-of-magnitude factors later. For a more exact description of the scattering of high-energy particles, whose masses do not play a role, an Aichelburg-Sexl metric [@Aic] would be convenient, for our considerations the above simple estimate may be sufficient. The interesting fact is that $\Delta E_1$ goes as $m{\cal E}^2$ and $\Delta E_2$ as ${\cal E}^3$. As we are looking for gravitational effects for realistic particles, we always have ${\cal E}\gg m$, so that $\Delta E_1$, containing the rest mass $m$, will be normally subdominant in comparison with $\Delta E_2$. While the virtual particle, and in connection with it the gravitational potential, come into being, the collision partners are attracted and come closer to each other than they would in absence of gravity. During this process the total energy is constant, the kinetic energy increases and compensates the negative potential energy. As it is only the kinetic energy, which plays a role in the interaction process, we can replace gravity by an effective growth of energy and momentum. On the other hand, after the collision the particles are slightly slowed down by gravity and their asymptotic outgoing energy is smaller than the energy immediately after the interaction, so that asymptotic energy conservation is guaranteed. (We do not assume graviton production, so that gravity is conservative.) So, instead of speaking about gravity, it is possible to ascribe an effective energy $E_{\rm eff}$ to the particles, enlarged by $-\Delta E_1$ and $-\Delta E_2$ in comparison with the asymptotic values, $$\label{eps} E_{\rm eff}=E-\Delta E_1-\Delta E_2=E\left(1+\alpha\,\frac{mE}{m_P^2}+\beta\,\frac{E^2}{m_P^2}\right)$$ with factors $\alpha$ and $\beta$ of the order of magnitude around 1 to 10. Having ascribed an effective energy to the incoming particles, we can also ascribe an effective momentum to them, simply by using the free high-energy-momentum relation $E\approx p$, $$\label{pp} p_{\rm eff}=p\left(1+\alpha\,\frac{mp}{m_P^2}+\beta\,\frac{p^2}{m_P^2}\right).$$ In some analogy to DSR, gravity is now hidden in effective variables. (One might wonder whether a calculation involving a Newtonian potential can be applicable to relativistic particles. Relations (\[eps\]) and (\[pp\]) are justified by the fact that the Newtonian potential is used only close to the turning points of the particles, when they slow down to nonrelativistic velocities.) Note that we have considered interactions in first order of a perturbative expansion. In higher order, when one takes into account more vertices, the interaction process becomes smoother, it is divided into more steps and sets in earlier, i.e. at larger distances, than in first order. In consequence, in higher order diagrams the influence of gravity will become weaker, so the above first-order estimates are rather an upper bound for gravitational modifications. To summarize, classical gravity influences the in- and outgoing particles when they are close to their vertices, if the energy is sufficiently high. This is described in two kinds of variables, both of which have an immediate physical meaning: The effective ones, $E_{\rm eff}$ and $p_{\rm eff}$, appearing at the vertices and entering cross section calculations, and the asymptotic ones, denoted by $E$ and $p$, playing the role of “bare" variables in connection with classical gravity. Non-central collision --------------------- In the central collisions considered above the minimal distance of colliding particles and the energy of a virtual particle have been modified, quantities that are hardly accessible to direct measuring, whereas the actual asymptotic energy/momentum are unaffected, so the discussion is physically rather meaningless so far. The situation improves in the case of non-central collisions of two particles with impact parameter $b$. In first order perturbation theory this is described in the following way: A particle moves straight ahead to the point of minimal distance $r_0$ from the scattering centre, its interaction vertex. There its radial momentum reverses by the exchange of a virtual particle and it flies away along a straight line at a scattering angle $\vartheta_0$ from its ingoing direction. (100,65)(10,-15) (10,10)[(1,0)[80]{}]{} (30,10)[(-1,0)[20]{}]{} (30,10) (30,10)[(-3,5)[20]{}]{} (30,10)[(-3,-5)[10]{}]{} (10,10)[(5,-3)[15]{}]{} (20,-6) (10,-6)[(1,0)[80]{}]{} (30,10)[(0,-1)[16]{}]{} (31,11)[V]{} (10,11)[A]{} (20,-1)[B]{}(19,-10)[C]{}(30,-10)[D]{}(15,40)[outgoing particle]{}(64,12)[incoming particle]{} (70,0)[(0,1)[10]{}]{} (70,0)[(0,-1)[6]{}]{}(72,0)[$b$]{}(24,11)[$\vartheta_0$]{} In the figure the particle comes from the right, the scattering centre is denoted by C and the vertex by V. A, B, and D are auxiliary points. We may read off the following relations. The triangles VCD and AVB are similar with the angles at $\angle$CVD and $\angle$BAV being equal to $\vartheta_0/2$. The radial component $\overrightarrow{\rm VB}$ of the momentum $\vec p=\overrightarrow{\rm VA}$ at V is $$p_r=p\sin\frac{\vartheta_0}{2},$$ the relation between the particle’s minimal distance $r_0=\overline{\rm VC}$ from C and the impact parameter is $$b=r_0\cos\frac{\vartheta_0}{2}.$$ At the vertex $p_r$ becomes zero for a moment, so that during the interaction process the kinetic energy is given only by the component orthogonal to it, namely $\overline{\rm BA}=p\cos\frac{\vartheta_0}{2}$. The energy, contributed from both ingoing particles to the virtual exchange particle is therefore equal to $${\cal E}=2\left(\sqrt{p^2+m^2}-\sqrt{p^2\cos^2\frac{\vartheta_0}{2}+m^2}\,\right).$$ Under the assumption $p\gg m$ the exchanged energy $\cal E$ can be expanded in two different ways, according to the scattering angle $\vartheta_0$. If $\vartheta_0$ is small, ${\cal E}\approx2p(1-\cos\frac{\vartheta_0}{2})$ is small, too. The particles do not slow down much and remain relativistic and the considerations of the foregoing subsection, involving a Newtonian potential, become inappropriate. In the other case, when the collision is almost central and $\vartheta_0$ is close enough to $180^\circ$, so that $\cos\frac{\vartheta_0}{2}\ll\frac{m}{p}$, the particles slow down to nonrelativistic speed and the calculations with Newtonian gravity are more reliable. Now the energy transfer $$\label{et} {\cal E}\approx2p\left[1-\frac{m}{p}\left(1+\frac{1}{2}\frac{p^2}{m^2}\cos^2\frac{\vartheta}{2} \right)\right]$$ is large and for the wavelength associated with the exchange particle we can again use the relativistic relation (\[l\]), $\lambda_0=1/\cal E,$ giving an estimate for the minimal distance $2r_0$ of the particles. An expansion of the potential energy $\Delta E_1$ of the rest masses of the particles in the gravitational field of the virtual particle, and the gravitational self-energy $\Delta\cal E$ of the latter one yields $$\Delta E_1\approx-2\,\frac{m\,{\cal E}^2}{m_P^2}\approx -8\,\frac{mp^2}{m_P^2}\left[1-2\frac{m}{p}\left(1+\frac{1}{2}\frac{p^2}{m^2} \cos^2\frac{\vartheta_0}{2}\right)\right]$$ and $$2\Delta E_2=\Delta{\cal E}\approx-\frac{6}{5}\frac{{\cal E}^3}{m_P^2}\approx-\frac{48}{5}\frac{p^3}{m_P^2}\left[1-3\frac{m}{p}\left(1+\frac{1}{2} \frac{p^2}{m^2}\cos^2\frac{\vartheta_0}{2}\right)\right].$$ With the leading contributions of these corrections the effective energy of the virtual particle, ${\cal E}_{\rm eff}={\cal E}-2\Delta E_1-\Delta\cal E$, becomes $$\label{aff} {\cal E}_{\rm eff}\approx 2p\left[1+\frac{24}{5}\frac{p^2}{m_P^2}-\frac{4}{5}\frac{mp}{m_P^2}\left( 4+9\frac{p^2}{m^2}\cos^2\frac{\vartheta_0}{2}\right)\right].$$ The last two terms in parenthesis are of the same order, because $\cos^2\frac{\vartheta_0}{2}$ is of order $m^2/p^2$. Importantly, there is a leading order correction, quadratic in $p/m_P$, independent from the scattering angle, and a smaller one, of order $mp/m_P^2$, depending on $\vartheta_0$. When gravity is again replaced by introducing the effective energy of the exchange particle, the wavelength of the latter one becomes in leading order (coming from $\Delta E_2$) $$\lambda\approx\frac{1}{{\cal E}_{\rm eff}}\approx\frac{1}{2p\left(1+\beta\frac{p^2}{m_P^2}\right)},$$ where, for convenience, the fancy numerical factor $24/5$ is again replaced by $\beta$. In the figure this means that the particle comes closer to the centre C, the vertex V is shifted a small distance to the left, so that the distance $\overline{\rm CV}$ becomes $\lambda/2$ instead of $\lambda_0/2$ and the scattering angle becomes modified from $\vartheta_0$ to $\vartheta$. From the relation $$\label{imp} 2b=\lambda_0\cos\frac{\vartheta_0}{2}=\lambda\cos\frac{\vartheta}{2}$$ we obtain the modification of the scattering angle $$\label{teta} \cos\frac{\vartheta}{2}\approx\left(1+\beta\,\frac{p^2}{m_P^2}\right) \cos\frac{\vartheta_0}{2}.$$ Due to the universality of gravity the “bare" scattering angle $\vartheta_0$ is unobservable, but it is possible to compare (\[teta\]) to the analogous result from DSR, obtained in the next section. Comparison with DSR =================== Now we are in a position to compare the two sets of variables constructed in the foregoing section with the “physical" and the “pseudo"-variables in DSR. Once the deformed, nonlinear relations for the physical variables are derived and modified kinematic relations are established with the aid of the linear pseudo-variables, they can in principle be forgotten, and all the consequences, the deformed dispersion relations between energy and momentum, the ensuing violations of conservation laws, etc. are ascribed to gravity. Here we go the opposite way by asking the question whether a gravity-motivated deformation of SR is in its first approximation compatible with DSR. In the foregoing section we have seen that gravity influences particle scattering in the same way as if the particles had a slightly higher effective kinetic energy. In the following considerations this enhanced effective energy-momentum is set into relation with the DSR pseudo-variables and the actual asymptotic kinetic energy is related to the physical variables, as usual [@ww]. Also in view of the desired parallel between DSR and gravity this association of variables is plausible in the following extrapolation: The unbounded pseudo-variables describe the situation with a repulsive potential, that goes to infinity at zero distance, in the absence of gravity: To reach smaller and smaller distances from each other, the particles must have arbitrarily high asymptotic energies. In most cases the same is true in the presence of Newtonian gravity, but the asymptotic energy necessary to bring particles close together, is lower. This actual energy is described by the physical variables, which are smaller. Moreover, being bounded, they predict distance zero at a finite asymptotic energy, thus mimicking a gravitational collapse, when the exchange particle’s energy reaches the Planck region. By this fact DSR is closer to GR than to Newtonian theory. Here, of course, we are going to compare only the leading corrections stemming from the inclusion of classical gravity, as well as from DSR, both based on the ratio $p/m_P$ ($=E/m_P$ in our assumption). Whereas in DSR the power of these ratio is a matter of an ad hoc definition, classical gravity in three space dimensions fixes the lowest order to be two, due to the simple fact that $G=1/m_P^2$. This is in contrast to linear DSR corrections, considered in [@KS], for example. For the comparison of classical gravity and DSR in non-central scattering, considered in subsection (2.2), we assume a typical lowest-order DSR relation (for different kinds of such approximations, see [@exp]) between $p$ and $\pi$ $$\label{beta} \pi\approx p\left[1+\kappa\left(\frac{p}{m_P}\right)^n\right]$$ with a constant $\kappa$ of order not too far from unity and some positive integer power $n$. This kind of relation is in good accordance with the leading term $\propto p^2$ in (\[pp\]), derived in connection with central collisions, if $n=2$. Considering an almost central collision from a DSR point of view, we replace $p$ by $\pi$, the variable related to interaction proceses, in the wavelength of the virtual particle, so that $\lambda\approx1/2\pi$. Then from (\[imp\]) we obtain the modified scattering angle, $$\label{thet} \cos\frac{\vartheta}{2}\approx\left[1+\kappa\left(\frac{p}{m_P}\right)^n\right] \cos\frac{\vartheta_0}{2}$$ and from comparison of (\[thet\]) with (\[teta\]) it follows that (at least in the considered scattering example) the lowest order correction term of DSR can agree with Newtonian gravity, when it is quadratic in the ratio $p/m_P$. Then only the constants $\beta$ and $\kappa$ must be matched. The result is also a first order approximation in the scattering angle around $180^\circ$. To consider further $\vartheta$-dependent terms does not make much sense in the scope of the Newtonian framework, because for faster scattering processes there would be significant general relativistic corrections. As Newtonian gravity is the lowest order correction of SR coming from GR, we have obtained a condition for DSR theories to satisfy the correspondence principle in its full meaning, namely, the lowest-order effects of DSR must be quadratic. DSR 2 for example, proposed in [@MS], with linear corrections would be at odds with it. One important difference of the present approach to “full" DSR is the use of the free energy-momentum relations for both kinds of variables, rather than of modified ones for $E$ and $p$. This important aspect of DSR does not show up in the present calculations, because in the considered approximations the mass term does not play a role, and $E\approx p$ as well as $\varepsilon\approx\pi$, the calculations were essentially carried out for momenta alone. The above considerations can easily be applied to higher dimensions. In $d>3$ space dimensions the Planck mass $m_P^{(d)}$ is by orders of magnitude smaller, on the other hand, the Newtonian potential goes as $r^{2-d}$. See for example [@G]. In consequence, the lowest order correction of the scattering angle behaves as $$\frac{p^{d-1}}{\left(m_P^{(d)}\right)^{d-1}},$$ if there are compactified dimensions, large enough for classical gravity to be a reasonable approximation when the minimal distance is as small as the magnitude of these dimensions. In consequence, in these cases the lowest-order corrections of DSR must be of order $d-1$, when compatibility with classical gravity is desired. Comparison with GUP =================== This approach is characterized by making a principal distinction between the energy and momentum $(E,\vec p\,)$ of a particle and its associated frequency and wave vector $(\omega,\vec k\,)$. Their relation is most generally written as $$\label{Efp} (\omega,\vec k\,)=\left(E\cdot f(E,\vec p\,),\vec p\cdot g(E,\vec p\,)\right).$$ Energy and momentum are assumed to be unbounded, whereas $\omega$ and $\vec k$ are bounded by orders of magnitude 1/[Planck time]{} and 1/[Planck length]{}, respectively. The functions $f$ and $g$ can be chosen analogously to the functions relating physical and pseudo-variables in arbitrary versions of DSR. Nevertheless, the interpretation is different: There are no merely auxiliary variables, both $(E,\vec p\,)$ and $(\omega,\vec k\,)$ have a clear physical meaning and there is a natural relation between momentum and position space from the beginning. Comparing (\[Efp\]) with the standard relation $$(E,\vec p\,)=\hbar(\omega,\vec k\,),$$ one finds energy-momentum dependent constants $$\tilde{\hbar}(E,\vec p\,)=\frac{1}{f(E,\vec p\,)} \hspace{2cm}\mbox{and}\hspace{2cm} \tilde{c}(E,\vec p\,)=\frac{\omega}{\|\vec k\|}=c\,\frac{f(E,\vec p\,)}{g(E,\vec p\,)}.$$ For high energies $\tilde\hbar$ increases, increasing the quantum-mechanical uncertainties. We shall restrict ourselves to the case $\tilde c=c$, i.e. $f=g$. In analogy to (\[beta\]) we assume a lowest-order relation ($p=\|\vec p\|$ and $k=\|\vec k\|$) $$p=\tilde{\hbar}k\approx\hbar k\left[1+\kappa\left(\frac{p}{m_P}\right)^n\right],$$ leading to the wavelength $$\lambda=\frac{1}{k}\approx\frac{\tilde\hbar}{2p}.$$ (Note that $k$ is the wave vector of the exchange particle, $p$ is the momentum of one incoming particle.) As before, from (\[imp\]) one obtains the correction of the scattering angle, $$\cos\frac{\vartheta}{2}\approx\left[1-\kappa\left(\frac{p}{m_P}\right)^n\right]\cos\frac{\vartheta_0}{2}.$$ With the same choice of transformation functions between the different sets of variables in first approximation, GUP has yielded just the opposite sign of the DSR correction in (\[thet\]). Conclusion ========== One main result of the considerations of this paper is the condition that DSR corrections to SR must be quadratic in $p/m_P$ in lowest order to fulfill the correspondence principle in the given example. It is not shown that this condition is sufficient in every situation and the calculations do not show how DSR differs from GR, when higher particle energies are involved. A relation of the present result with the interpretation of DSR in [@top] as the topological part of GR would depend on the properties of particle trajectories in topological 3+1 gravity. The second result concerns GUP, where the situation is quite different. DSR produces an effective attractive force, GUP, on the other hand, results in a repulsive force, which is not a big surprise, as it lays lower bounds to space and time intervals. In contrast to DSR, rather than competing with Newtonian gravity (and thus GR), GUP counteracts it, thus qualifying as a description of pure quantum gravity effects, which has nothing at all to do with classical gravity. In order not to collide with GR, the lowest-order term in GUP must be of a higher power than 2. There is an interesting parallel to loop quantum cosmology [@LQC], where a repulsive behaviour of gravity at short distances, which helps to avoid singularities, is observed. In the considered example the effects of DSR and those of GUP would be equivalent for some $n>2$, if the roles of $(E,\vec p\,)$ and $(\varepsilon,\vec\pi)$ were interchanged. $\varepsilon$ and $\vec\pi$ would be energy and momentum ascribed to free particles, which can be boosted to arbitrary values with respect to a certain reference frame, as long as no interaction takes place. The physical energy $E$ and momentum $\vec p$, on the other hand, play a role in interactions, which would be in accordance with their interpretation as measurable quantities, as measurements always go along with some interactions. Due to the exchanged roles of $(E,\vec p\,)$ and $(\varepsilon,\vec\pi)$ in relation to the common DSR interpretation, reaction thresholds anomalies would be equally small as in usual DSR, but in the opposite direction. For example, when conventional DSR predicts an insignificant lowering of the GZK cutoff [@SM], the reversed interpretation would lead to an (equally insignificant) raising. (Recent observations do not seem to confirm a shift of the GZK cutoff at all [@GZK].)\ [Acknowledgement]{}. Supported by the Ministry of Education of the Czech republic, contract number MSM 0021622409. The author thanks the Perimeter Institute for warm hospitality and support and S. Hossenfelder and K. Bering for helpful discussions. [99]{} G. Amelino-Camelia, Relativity in spacetimes with short-distance structure governed by an observer-independent (Planckian) length, Int. J. Mod. Phys. D 11 (2002) 35, gr-qc/0012051 J. Magueijo, L. Smolin J. Magueijo and L. Smolin, Lorentz invariance with with an invariant energy scale, Phys. Rev. Lett. 88 (2002) 190403, hep-th/0112090 S. Judes and M. Visser Conservation Laws in “Doubly Special Relativity", Phys. Rev. D 68 (3003) 045001, gr-qc/0205067 J. Lukierski and A. Nowicki, Doubly special relativity versus $\kappa$-deformation of relativistic kinematics, Int. J. Mod. Phys. A 18 (2003) 7, hep-th/0203065 G. Amelino-Camelia, Status of Relativity with observer-independent length and velocity scales, Proceedings of the 37th Karpacz Winter School of Theoretical Physics, AIP Conference Proceedings 589, gr-qc/0106004 S. Hossenfelder, Multi-Particle States in Deformed Special Relativity, hep-th/0702016 J. Ellis, N. Mavromatos, D. Nanopoulos, A. Sakharov, E. Sarkisyan and the MAGIC Collaboration, Probing Quantum Gravity using Photons from a Mkn 501 Flare Observed by MAGIC, astro-ph/0708.2889 J. Kowalski-Glikman, De Sitter space as an arena for Doubly Special Relativity, Phys. Lett. B 547 (2002) 291-296, hep-th/0207279 A. Agostini, G. Amelino-Camelia and F. D’Andrea, Hopf-algebra description of noncommutative-spacetime symmetries, Int. J. Mod. Phys. A 19 (2004) 5187-5220, hep-th/0306013 L. Freidel and J. Kowalski-Glikman, $\kappa$-Minkowski space, scalar field, and the issue of Lorentz invariance, hep-th/0710.2886 J. Magueijo and L. Smolin, Gravity’s Rainbow, Class. Quantum Grav. 21 (2004) 1725-1736, gr-qc/0305055 F. Girelli, S. Liberati and L. Sindoni, Planck-Scale Modified Dispersion Relations and Finsler Geometry, gr-qc/0611024 S. Majid, Hopf algebras for physics at the Planck scale, Class. Quantum Grav. 5 (1988) 1587-1606 M. Bojowald, H. Morales-Técotl and H. Sahlmann, On Loop Quantum Gravity Phenomenology and the Issue of Lorentz Invariance, Phys. Rev. D 71 (2005) 2717-2742, gr-qc/0411101 J. Kowalski-Glikman, Doubly Special Relativity: facts and prospects, to appear in “Towards Quantum Gravity - different approaches to a new understanding of the space, time, and matter", D. Oriti ed. Cambridge University Press, gr-qc/0603022 L. Freidel, J. Kowalski-Glikman and L. Smolin, 2+1 gravity and doubly special relativity, Phys. Rev. D 69 (2004) 044001, hep-th/0307085 S. Hossenfelder, Interpretation of Quantum Field Theories with a minimal Length Scale, Phys. Rev. D73 (2006) 105013, hep-th/0603032 P. Aichelburg and R. Sexl, On the Gravitational Field of a Massless Particle, Gen. Relativity and Grav. 2 (1971) 303 G. Amelino-Camelia, Kinematic solution of the UHE-cosmic-ray puzzle without a preferred class of inertial observers, Int. J. Mod. Phys. D12 (2003) 1211-1226, astro-ph/0209232 T. Konopka and S. Major, Observational limits on quantum geometry effects, New J. Phys. 4 (2002) 57, hep-ph/0201184 G. Amelino-Camelia, Are we at the dawn of quantum-gravity phenomenology? Lect. Notes Phys. 541 (2000) 1, gr-qc/9910089 M. Bleicher, How to Create Black Holes on Earth? physics/0703062 D. Hayman, F. Hinterleitner and S. Major, On reaction thresholds in doubly special relativity, Phys. Rev. D69 15006 (2004), gr-qc/0312089 M. Bojowald, Loop quantum cosmology, Liv. Rev. Rel 8, 11 (2005), gr-qc/0601085 Observation of the GZK cutoff by the HiRes Experiment, HiRes Collaboration, arXiv:astro-ph/0703099	{ "pile_set_name": "ArXiv" }
--- abstract: 'In this article, we introduce a $2$-parameter family of affine connections and derive the Ricci curvature. We first establish an integral Bochner technique. On one hand, this technique yields a new proof to our recent work in [@LX] for substatic manifolds. On the other hand, this technique leads to various geometric inequalities and eigenvalue estimates under a much more general Ricci curvature conditions. The new Ricci curvature condition interpolates between static Ricci tensor and $1$-Bakry-Émery Ricci, and also includes the conformal Ricci as an intermediate case.' address: - \| Department of Mathematics\ University of Alabama at Birmingham\ Birmingham, AL 35294 - \| School of Mathematical Sciences\ Xiamen University\ 361005, Xiamen, P.R. China author: - Junfang Li and Chao Xia title: An integral formula for affine connections --- [^1] Introduction ============ The classical Reilly formula is actually an integral Bochner formula for gradient vector fields on manifolds with boundary in references. It has been proven to be a quite useful tool in differential geometry. Motivated by a work of Qiu and the second author [@QX], we have established a generalized Reilly type formula in previous work [@LX]. Such a generalization enabled us to prove a Heintze-Karcher-Ros-Brendle type inequality under a sub-static condition. Such kind of inequality, which could lead to an Alexandrov type rigidity theorem, has been proved first by Brendle [@Br]. See also recent work by Wang-Wang [@WW]. Moreover, the general formula has been used to prove several geometric inequalities in [@LX] and also applied by Chen-Wang-Wang-Yau [@CWWY] to prove the stability of quasi-local energy with respect to a static spacetime. The formula was proved via very basic integration-by-parts with respect to the Levi-Civita connection, although the computation is complicated and tedious. The key point is that we introduced a “weight” function $V$, which was motivated by Brendle and Brende-Hung-Wang [@Br; @BHW]. In this article, we adapt a new point of view to recover the formula in [@LX]. We find that the formula in [@LX] is indeed an integral Bochner formula for some special vector fields with respect to a special [torsion-free affine ]{}connection instead of the Levi-Civita connection. Moreover, this turns out to be a general phenomenon that a wide class of [torsion-free affine ]{}connections give rise to a class of Reilly type formulas. Let $(M^n, \bar g)$ be an $n$-dimensional smooth Riemannian manifold and $\nb$ be the Levi-Civita connection of $\bar g$. Let $V=e^u$ be a positive smooth function on $M$, where $u$ is a smooth function on $M$. We call $(M, \bar g, V)$ a Riemannian triple. For $\a, \g\in \mathbb{R}$, we define a $2$-parameter family of affine connections: given two vector fields $X, Y$ on $M$, let $$\begin{aligned} D^{\a,\g}_XY=%&\nb_XY+\a V^{-1}dV(X)Y+\a V^{-1}dV(Y)X +\g \bar g(X,Y)V^{-1}\nb V\nonumber \nb_XY+\a du(X)Y+\a du(Y)X +\g \bar g(X,Y)\nb u. \label{defi D}\end{aligned}$$ For simplicity, we often omit the superscript $\a,\g$ when it is clear in the context. The Ricci curvature under $D^{\a,\g}$ is (see Proposition \[thm Ricci\] below) $$\begin{aligned} \label{Ricci} Ric^{D}:=&Ric-[(n-1)\a+\g]\bar\nabla^2 u+\big[(n-1)\a^2-\g^2\big]du\otimes du\nonumber\\ &+\left[\g \bar\De u +(\g^2+(n-1)\a\g)\|\nb u\|^2\right]\bar g. \end{aligned}$$ There are two trivial cases among all $D^{\a,\g}$. One is the Levi-Civita connection for $\bar g$ when $\a=\g=0$, while the other is the Levi-Civita connection for the conformal metric $e^{2\a u} \bar g$ when $\a=-\g$. For other choices of $\a$ and $\g$, $D^{\a,\g}$ may not be a Levi-Civita connection for any Riemannian metric. For the case $\a=0, \g=1$, one sees from that $$\begin{aligned} Ric^{D^{0,1}}= Ric-\frac{\bar \nabla^2 V}{V}+\frac{\bar \Delta V}{V}\bar g,\end{aligned}$$ where $Ric$ is the usual Ricci curvature for $\bar g$. We call $Ric^{D^{0,1}}$ [static Ricci tensor]{}. A Riemannian triple $(M, \bar g, V)$ satisfying $VRic^{D^{0,1}}=0$ is referred to a static manifold in the literature, see e.g. [@Cor]. For the case $\a=\frac{1}{n-1}, \g=0$, one sees from that $$\begin{aligned} Ric^{D^{\frac{1}{n-1},0}}= Ric-\mathrm{\bar \nabla^2} u +\frac{1}{n-1}d u\otimes d u.\end{aligned}$$ This is in fact the [$1$-Bakry-Émery Ricci tensor]{} in the literature which was introduced by Bakry and Émery [@BaE]. The fact that this affine connection gives rise to the $1$-Bakry-Émery Ricci tensor has also been observed by Wylie-Yeroshkin [@WY] in their recent studies of manifolds with density. The main result of this paper is the following Reilly type integral formula. \[theo RF0\] Let $(M^n, \bar g, V=e^u)$ be an $n$-dimensional Riemannian triple and $\a,\g\in \mathbb{R}$. Let $D=D^{\a,\g}$ be the affine connection defined as in (\[defi D\]) and $\tau=(n+1)\a+\g$. Let $\phi$ be a smooth function on a bounded domain $\O\subset M$ with smooth boundary $\S$. Then the following integral formula holds: $$\begin{aligned} \label{formula RF 1} &\di\int_{\O}V^\tau \left[\left\|\bd^D \phi\right\|^2-\left\|\bn^{2,D} \phi\right\|_{\bar g}^2 \right] -V^\tau Ric^{D}( \nb^D \phi, \nb^D \phi) d\O \\ =&\di\int_{\S}V^{\tau}\left[H^D\<\bn^D \phi,\nu\>^2+\left(h-\g u_\nu g\right)(\n^D\phi, \n^D \phi)-2V^{-\g}\<\n^D \phi, \n^D(V^\g \phi_\nu)\>\right]dA.\nonumber \end{aligned}$$ The notations $\nb^D$, $\bn^{2,D}$ and $\bd^D$ play the role of gradient, Hessian and Laplacian with respect to $D$, the exact definition will be given in Definition \[notation\]. $H^D:=H+(n-1)\a u_\nu$ is the affine mean curvature, where $H$ is the usual mean curvature.\ Formula reduces to Reilly’s original formula for $\bar g$ in the case $\a=\g=0$ and for $e^{2\a u}\bar g$ in the case $\a=-\g$. Moreover, when $\a=0, \g=1$, it reduces to the following Let $(M^n, \bar g, V=e^u)$ be an $n$-dimensional Riemannian triple. Let $\phi$ be a smooth function on a bounded domain $\O\subset M$ with smooth boundary $\S$. We have $$\begin{aligned} \label{theo RF sub} &\di\int_{\O}V^3\Big[(\bar\De \phi+2\bar\n u\bar\n \phi)^2-\|\bar\n_i\bar\n_j\phi+\bar\n_i u\bar\n\phi_j+\bar\n_j u\bar\n_i\phi\|^2\Big]d\O\\ =&\di\int_{\S}V^{3}\Big(H\phi_\nu^2+\left(h-u_\nu g\right)(\n\phi, \n\phi)-2 V^{-1}\<\n \phi, \n (V\phi_\nu)\>\Big)dA\nonumber\\ &+\int_{\O}V^{3}(Ric-\frac{\nb^2 V}{V}+ \frac{\bar\De V}{V}\bar g)(\nb\phi,\nb\phi)d\O.\nonumber \end{aligned}$$ If we let $\phi=\frac{f}{V}$ in , then we recover Theorem 1.1 in [@LX] by a different method. Let us illustrate the basic idea of the proof of Theorem \[theo RF0\]. It is well known that a local Bochner formula holds for general vector fields under an affine connection. Since the connection is in general not metric compatible, we have only a divergent form instead of the Laplacian of some function in the Bochner identity, see Proposition \[Bochner formula\]. Nevertheless, we are able integrate this local Bochner formula to get an integral formula. To achieve an effective Reilly type formula, there are two innovative ingredients with this method. First, we choose a right volume form, which is a “weight”, to get the divergent-free property. Second, we choose a suitable vector field $X$ which satisfies $DX$ is symmetric. It turns out that we choose $X=\bn^D \phi$ and the volume form $V^\tau d\O$, see Lemmata \[sym vf\] and \[weight\]. With the integral formula in Theorem \[theo RF0\] at hand, we are able to prove Heintze-Karcher type, Minkowski type, and Lichnerowicz type inequalities. \[thm ineq\] Let $(M^n, \bar g, V=e^u)$ be an $n$-dimensional Riemannian triple and $\a,\g\in \mathbb{R}$. Let $D=D^{\a,\g}$ be the affine connection defined as in (\[defi D\]) and $\tau=(n+1)\a+\g$. Then we have the following results. - [Heintze-Karcher type inequality:]{} if $Ric^{D}\ge 0$ and $H^D>0$, then $$n\di\int_{\O}V^\tau d\O\le\di (n-1)\int_{\S}\frac{V^\tau}{H^D} dA. \label{Heintze-Karcher}$$ Equality in holds only if $\S$ is umbilical. - [Minkowski inequality:]{} If $Ric^{D}\ge 0$ and $h-\g u_\nu g> 0$, then $$\begin{aligned} \label{rm2} \left(\int_\S V^{\tau-\a} dA\right)^2\geq \frac{n}{n-1}\int_\Omega V^\tau d\O \int_\S H^DV^{\tau-2\a} dA, \end{aligned}$$ Equality in holds only if $\S$ is umbilical. - [Lichnerowicz inequality:]{} If $Ric^{D}\ge (n-1)V^{\a-\g} \bar g$ and\ [a) ]{}$\S=\emptyset$, then $\lambda_1(\bd^D)\ge n;$\ [b) ]{}$\S \neq\emptyset$ and $\S$ satisfies $H^D\ge 0$, then $\lambda_1^{Dir}(\bd^D)\ge n;$\ [c) ]{}$\S\neq\emptyset$ and $h-\g u_\nu g\ge 0$, then $\lambda_1^{Neu}(\bd^D)\ge n.$ Here $\lambda_1, \lambda_1^{Dir}$ and $\lambda_1^{Neu}$ indicate the closed, the Dirichlet and the Neumann first (nonzero) eigenvalue of the affine Laplacian $\bd^D$, i.e., there exists some non-trivial $\phi$ such that $\bd^D \phi=-\l_1\phi$ with Dirichlet boundary condition $\phi=0$ or Neumann boundary condition $\phi_\nu=0$. - Theorem \[thm ineq\] reduces to Heintze-Karcher, Minkowski, and Lichnerowicz inequalities for $\bar g$ in the case $\a=\g=0$ or $e^{2\a u} \bar g$ in the case $\a=-\g$. See Section 2.1 for an overview. - In the case $\a=0, \g=1$, the Heintze-Karcher type inequalities were first proved by Brendle [@Br], then by Wang-Wang [@WW] for non-homologous static manifolds without warped product structure by using the same method and by the authors for general cases using Reilly type formulas in [@LX]. The Minkowski type inequalities have been proved in our previous work [@LX]. - Theorem \[thm ineq\] gives new geometric inequalities under the condition of nonnegative $1$-Bakry-Émery Ricci, which is the case $\a=\frac{1}{n-1}, \g=0$. To illustrate the idea, we only list the example of the Heintze-Karcher type inequality and the others hold true similarly. [Corollary.]{} [Let $(M^n, \bar g, e^ud\O)$ be a smooth weighted Riemannian manifold and $\O$ be a bounded domain in $M$. If the $1$-Bakry-Émery Ricci is nonnegative, namely, $$\begin{aligned} &Ric-\bar\nabla^2 u+\frac{1}{n-1}du\otimes du\nonumber\ge 0, \label{Ric nonnegative tilde 1} \end{aligned}$$ and the weighted mean curvature $H+\<\bar\nabla u,\nu\>>0$, then the following inequality holds: $$n\di\int_{\O}e^{\tau u}d\O\le\di (n-1)\int_{\S}\frac{e^{\tau u}}{H+\<\bar\nabla u,\nu\>} dA, \label{Heintze-Karcher coro 1}$$ where $\tau=\frac{n+1}{n-1}$. Moreover, if equality in holds, then $\S$ is umbilical. ]{} We remark that the weight in is $e^{\tau u}$ instead of $e^u$. The new weight volume form $e^{\tau u}d\O$ has a property that it is parallel under the affine connection $D^{\a,\g}$. In the special case of $1$-Bakry-Émery Ricci curvature, this parallel property has been observed by Wylie-Yeroshkin [@WY] recently. In particular, for a Riemannian triple $(M, g, V)$ whose static Ricci tensor has a positive lower bound, we get the first eigenvalue estimate for the operator $V\bar \De - \bd V\cdot$. \[coro\] Let $(M^n, \bar g, V)$ be an $n$-dimensional Riemannian triple. Let $\O$ be a bounded domain in $M$ with smooth boundary $\S$. Assume the static Ricci tensor satisfies $$V Ric- \mathrm{\bar \nabla^2} V+ \bar \De V \bar g \geq (n-1) \bar g.$$ Then we have\ [a) ]{} if $\S=\emptyset$, then $\lambda_1(V\bar \De - \bd V\cdot)\ge n;$\ [b) ]{} if $\S \neq\emptyset$ and $\S$ satisfies $H^D\ge 0$, then $\lambda_1^{Dir}(V\bar \De - \bd V\cdot)\ge n;$\ [c) ]{} if $\S\neq\emptyset$ and $h-\g u_\nu g\ge 0$, then $\lambda_1^{Neu}(V\bar \De - \bd V\cdot)\ge n.$ The rest of the paper is organized as follows. In section \[sec pre\], we recall classical results, introduce our affine connections, fix the notations, and give the Ricci curvature under affine connections. In section \[sec Bochner\], we establish the Bochner formula and prove the main theorem, Theorem \[theo RF0\]. In section \[sec inequalities\], we prove the Heintze-Karcher type and the Minkowski type inequalities of Theorem \[thm ineq\]. In section \[sec Lich\], we prove the Poincare type and the Lichnerowicz type inequalities. In the [Appendix]{}, we prove Proposition \[thm Ricci\]. Preliminaries and notations {#sec pre} =========================== Classical results. ------------------ Let us first recall the classical Reilly’s formula [@Re1]. For a bounded domain $\O$ with boundary $\S$ in an $n$-dimensional smooth Riemannian manifold $(M, \bar g)$ and $\phi\in C^\infty(\bar \O)$, the classical Reilly’s formula reads as $$\begin{aligned} &&\int_\O (\bd \phi)^2-\|\bn^2 \phi\|^2 -Ric(\bn \phi, \bn \phi) d\O=\int_\S H\phi_\nu^2+h(\n \phi, \n \phi)+ 2\phi_\nu\De \phi dA.\end{aligned}$$ Here and throughout this paper, $\bn$ and $\bd$ denote the gradient and the Laplacian on $\O$ respectively, $\n$ and $\De$ denote the gradient and the Laplacian on $\S$ respectively with respect to the induced metric from $\bar g$. $d\O$ and $dA$ are the Riemannian volume form of $\bar g$ and the induced area element from $\bar g$ respectively. $\nu$ is the normal vector field of $\S$ and $\phi_\nu=\nb_\nu\phi$ is the normal derivative of $\phi$. $h(X,Y)=\<\nb_X \nu, Y\>$ is the classical second fundamental form of $\S$ and $H= \tr_{\bar g} h$ is the usual mean curvature of $\S$. With the Reilly formula, some classical geometric inequalities can be readily proved. [Heintze-Karcher inequality:]{} If $Ric\ge 0$ and $\S$ is mean convex, i.e., $H>0$, then $$\begin{aligned} \label{HK0} \int_\S \frac{1}{H}dA\ge \frac{n}{n-1} Vol(\O).\end{aligned}$$ [Minkowski inequality:]{} If $Ric\ge 0$ and $\S$ is convex, i.e., $h \ge 0$, then $$\begin{aligned} \label{M0} Area(\S)^2 \ge \frac{n}{n-1} Vol(\O)\int_\S H dA.\end{aligned}$$ [Lichnerowicz inequality:]{} If $Ric\ge (n-1)\bar g$ and $\S$ is empty, then $$\begin{aligned} \label{L0} \l_1(\bd)\ge n.\end{aligned}$$ Similar inequalities like were first derived by Heintze-Karcher [@HK] using the classical approach of Jacobian fields from Riemanian geometry. Ros [@Ros] proved the current form of this inequality using Reilly’s formula. Inequality was first derived by Minkowski [@Min] in the Euclidean case as a consequence of the famous Brunn-Minkowski theorem in convex geometry. Reilly [@Re3] proved this inequality under the condition of nonnegative Ricci by using his formula. Recently, Wang-Zhang [@WZ] gave an alternative proof of Minkowski inequality (\[M0\]) using ABP method. Inequality was proved by Lichnerowicz [@Lic] using the classical Bochner technique. Notations under affine connections {#subsec Ricci} ---------------------------------- As in the introduction, a two parameter family of [torsion free]{} affine connections $D^{\a, \g}$ is defined on $M$ for $\a, \g\in \mathbb{R}$: $$\begin{aligned} D^{\a,\g}_XY=%&\nb_XY+\a V^{-1}dV(X)Y+\a V^{-1}dV(Y)X +\g \bar g(X,Y)V^{-1}\nb V\nonumber \nb_XY+\a du(X)Y+\a du(Y)X +\g \bar g(X,Y)\nb u.\end{aligned}$$ One checks directly that $D^{\a,\g}$ is torsion-free. For a general affine connection, we adapt the following convention of Ricci curvature. \[defi Ricci\] Given an affine connection $D$, for any vector fields $X,Y$, we define the Ricci curvature as $$\begin{aligned} Ric^D(X,Y)=\o^i\left(R^D(e_i,X)Y\right), \end{aligned}$$ where $\{e_i\}$ is a local frame of the tangent bundle, $\{\o^i\}$ is the dual $1$-form of $\{e_i\}$ and the Riemann curvature operator $R^D$ is defined as $$\begin{aligned} R^D(X,Y)Z= D_{X}D_YZ-D_YD_{X}Z-D_{[X,Y]}Z. \end{aligned}$$ In the case we have a Riemannian metric $\bar g$, the Ricci curvature $Ric^D$ of $D$ can also be interpreted as $$\begin{aligned} Ric^D(X,Y)=\<R^D(e_i,X)Y, e_i\>, \end{aligned}$$ where $\{e_i\}$ is an orthonormal frame of the tangent bundle. By direct computation, we derive the following representation of $Ric^D$ for $D=D^{\a,\g}$ in terms of Levi-Civita connection $\bar\n$. \[thm Ricci\] The Ricci curvature $Ric^D$ of $D=D^{\a,\g}$ satisfies the following identity: $$\begin{aligned} Ric^D=&Ric-[(n-1)\a+\g]\bar\nabla^2 u+[(n-1)\a^2-\g^2]du\otimes du\nonumber\\ &+\left\{\g \bar\De u +\g[(n-1)\a+\g]\|\nb u\|^2\right\}\bar g. \end{aligned}$$ We will prove Proposition \[thm Ricci\] in [Appendix]{}. As already mentioned in the introduction, the Ricci curvature $Ric^{D}$ of the new affine connection $D^{\a,\g}$ not only yields new Ricci curvature tensors, but also recovers some of known examples in the literature, which includes conformal Ricci tensor from conformal geometry, and static Ricci tensor raised from General Relativity and the $1$-Bakry-Éemery Ricci tensor from manifolds with density. Next we explain the notations in . \[notation\] - The $D$-gradient on $\O$ and $\S$ are defined respectively by $$\bn^D \phi:= V^{\g-\a}\bn \phi, \quad \n^D \phi:= V^{\g-\a}\n \phi$$ - The $D$-Hessian $\bn^{2,D} \phi$ and $D$-Laplacian $\bd^{D} \phi$ on $\O$ are defined respectively by $$\begin{aligned} \bn^{2,D} \phi&:=& D(V^{\g-\a} \nb \phi)\\&=& V^{\g-\a} \left[\bn^2 \phi+\g du\otimes d\phi+\g d\phi \otimes du+ \a \<\bn u, \bn \phi\>\bar g\right], \nonumber\end{aligned}$$ and $$\begin{aligned} \bd^{D} \phi:= \tr_{\bar g} (\bn^{2,D} \phi)= V^{\g-\a} \left[\bd \phi+(2\g+ n\a)\<\bn u, \bn \phi\>\right]. \nonumber\end{aligned}$$ We note that in the case $\a=1, \g=-1$, the $D$-gradient, the $D$-Hessian and the $D$-Laplacian are in consistence with the classical ones for conformal metric $e^{2u}\bar g$. By virtue of this, we believe Definition \[notation\] is natural for $D^{\a,\g}$. Bochner technique for general affine connections {#sec Bochner} ================================================ In this section, we review a Bochner identity for general affine connection and prove Theorem \[theo RF0\]. It is well known that under Levi-Civita connection $\nb$, the following Bochner formula holds: for a smooth vector field $X$ on $M$ with the property that $\nb X$ is symmetric, $$\begin{aligned} \bar \De \frac12\|X\|^2=\|\bar \n X\|^2+ \bar \nabla_X (\div_{\bar g} X)+Ric(X, X),\end{aligned}$$ see e.g. Petersen [@Petersen] Proposition 33, page 207. Under an affine connection, the following Ricci identity holds. [(Ricci identity)]{}\[Ricci iden\] Under local coordinates $\{\P_i\}$, for any smooth vector field $X$, we have $$\begin{aligned} D_iD_jX^i=D_jD_iX^i+R^{D}_{jk}X^k. \label{Ricci identity} \end{aligned}$$ A vector filed $X$ can be viewed as a $(1,0)$-tensor field and we have $$\begin{aligned} D^2X(V,W)=&D_W(D_VX)-D_{D_WV}X,\nn D^2X(W,V)=&D_V(D_WX)-D_{D_VW}X. \label{Ricci identity proof equ 1} \end{aligned}$$ It follows that $$\begin{aligned} D^2X(W,V)-D^2X(V,W)=&R^{D}(V,W)X. \label{Ricci identity proof equ 2} \end{aligned}$$ Let $\{e_i\}$ be an local frame of the tangent bundle, $\{\o^i\}$ is the dual $1$-form of $\{e_i\}$. Then we have from $$\begin{aligned} \o^i\left(D^2X(e_j,e_i)\right)- \o^i\left(D^2X(e_i,e_j)\right)=&Ric^{D}(e_j, X). \end{aligned}$$ We finish the proof. \[Bochner formula\] [(Bochner formula)]{} Let $D$ be an affine connection on $M$ and $Ric^D$ be the Ricci curvature of $D$. Let $X$ be a smooth vector field on $M$. Then we have $$\begin{aligned} \div^D (D_XX)=(DX)^t \cdot DX+d(\div^D X)(X)+Ric^{D}(X, X). \label{Bochner}\end{aligned}$$ where $div^D$ is the divergence operator w.r.t. $D$, $\div^D Y=D_i Y^i$ for a vector field $Y$, $(DX)^t$ is the transpose of $D X$. Under local coordinates $\{\P_i\}$, reads as $$\begin{aligned} D_i(X^jD_jX^i)=&D_iX^j D_jX^i+X^jD_jD_iX^i+R^{D}_{ji}X^iX^j. \label{Bochner1}\end{aligned}$$ Moreover, $$\begin{aligned} D_i(X^jD_jX^i-X^iD_jX^j) =&D_iX^j D_jX^i-(D_jX^j)^2+R^{D}_{ji}X^iX^j. \label{Bochner equ 1}\end{aligned}$$ Using tensor calculus, we have $$\begin{aligned} D_i(X^jD_jX^i)=&D_iX^j D_jX^i+X^jD_iD_jX^i. \label{tensor calc}\end{aligned}$$ Combining with the Ricci identity we get . We also have by using tensor calculus that $$\begin{aligned} X^jD_jD_iX^i=& D_j(X^jD_iX^i)- (D_j X^j)(D_i X^i) \nonumber\\=&D_i(X^iD_jX^j)- (D_i X^i)^2. \label{tensor calc 2}\end{aligned}$$ Inserting into , we get . Our aim is to derive an integral formula from the local Bochner formula . From now on, let $(M^n, \bar g, V)$ be an $n$-dimensional Riemannian triple and $\a,\g\in \mathbb{R}$. Let $D=D^{\a,\g}$ be the affine connection defined as in (\[defi D\]). Note that $Ric^{D}$ is symmetric. In order to obtain a useful integral formula, we need the following two important ingredients: 1. $DX$ is symmetric, i.e., $$\begin{aligned} D_iX^j=D_jX^i;\end{aligned}$$ 2. The left hand side of (\[Bochner equ 1\]) needs to be a “divergent form" with respect to some choice of volume form. In the following two lemmata, we will find an appropriate vector field $X$ and also a compatible volume form. \[sym vf\] Let $\phi$ be a smooth function on $M^n$. Let $$\begin{aligned} X=\bn^D\phi=V^{\g-\a}\nb \phi.\label{vf}\end{aligned}$$ Then $$DX= V^{\g-\a} \left[\bn^2 \phi+\g du\otimes d\phi+\g d\phi \otimes du+ \a \bar g\left(\bn u, \bn \phi\right)\bar g\right]=\bn^{2,D}\phi$$ is symmetric. Moreover, $$\div^D X:=D_i X^i=V^{\g-\a} \left(\bar\De \phi+(2\g+n\a)\<\nb u, \nb \phi\>\right)=\bd^D \phi.$$ Recall under local coordinates, $$\begin{aligned} % \nb_i\nb^j\phi=&\<\nb_i\nb \phi,\P_j\>\\ D_iX^j=&\<D_i X,\bar g^{jk}\P_k\>. \label{hessian}\end{aligned}$$ Under normal coordinates, we have $$\begin{aligned} D_iX_j=&\<D_i(V^{\g-\a}\bar\nabla\phi), \P_j\>\nonumber\\ =&\nb_i(V^{\g-\a}\nb_j\phi)+\a V^{\g-\a}u_i\phi_j+\a V^{\g-\a}\nb u\nb \phi \delta_{ij}+\g V^{\g-\a}\phi_iu_j\nonumber\\ =&V^{\g-\a}\left(\nb_i\nb_j\phi+\g u_i\phi_j+\g \phi_iu_j+\a\<\nb u, \nb \phi\> \delta_{ij}\right). \label{Bochner equ 2}\end{aligned}$$ Clearly, $DX$ is symmetric. \[weight\] Let $W$ be any smooth vector field on $M$. Then $$\begin{aligned} V^\tau D_iW^i=\nb_i(V^\tau W^i) \label{div} \end{aligned}$$ where $\tau=(n+1)\a+\g$, is a divergent form with respect to the Riemannian volume form $d\O$. ($d\O$ denotes the volume form induced by Riemannian metric $\bar g$ throughout this paper) By definition of $D$, we have $$\begin{aligned} D_iW^i=&\nb_iW^i+\a u_i W^i+\a u_k W^k \delta^i_i+\g W^i u_i\nonumber\\ =&\nb_iW^i+[(n+1)\a+\g] u_i W^i. \label{lemm div equ 1} \end{aligned}$$ Thus, $$\begin{aligned} V^\tau D_iW^i=& V^\tau\nb_iW^i+[(n+1)\a+\g] V^{\tau-1}V_i W^i\nonumber\\ = & V^\tau\nb_iW^i+\tau V^{\tau-1}V_i W^i\nonumber\\ = & \nb_i(V^\tau W^i). \label{lemm div equ 2} \end{aligned}$$ As an immediate corollary, we can show that the volume form $V^\tau d\O$ is parallel under the new affine connection $D^{\a,\g}$. We thank the referee for pointing out this fact to us. The special case of $\a =\frac{1}{n-1}$ and $\g=0$ was proved in [@WY]. We have $D_X (V^\tau d\O) =0$ for any smooth vector field $X$. Given an arbitrary vector field $W$ with compact support in $\O$. From Lemma \[weight\], $$\int_\O W^i D_i (V^\tau d\O)= -\int_\O V^\tau D_iW^i d\O =-\int_\O \nb_i(V^\tau W^i) d\O=0.$$ Since $W$ is arbitrary, we conclude that $D_i (V^\tau d\O)=0$ for any $i$. Choosing $X=\nb^D \phi$ in , we obtain $$\begin{aligned} & D_i\Big((\nb^D \phi)^j D_j((\nb^D \phi)^i)-(\nb^D \phi)^i D_j((\nb^D \phi)^j)\Big)\nonumber \\=&\|\bn^{2,D}\phi \|_{\bar g}^2-\|\bd^D \phi\|^2+Ric^{D}(\bar\nabla^D\phi,\bar\nabla^D\phi). \label{Bochner equ 1'}\end{aligned}$$ Multiplying with $V^\tau$ and integrating over a bounded domain $\O\subset M$, we have $$\begin{aligned} &\di\int_{\O}V^\tau D_i\Big((\nb^D \phi)^j D_j((\nb^D \phi)^i)-(\nb^D \phi)^i D_j((\nb^D \phi)^j)\Big) d\O\nonumber\\ =&\di\int_{\O}V^\tau \left[\|\bn^{2,D}\phi \|_{\bar g}^2-\|\bd^D \phi\|^2\right]+ V^\tau Ric^{D}(\bar\nabla^D\phi,\bar\nabla^D\phi) d\O. \label{integral formula} \end{aligned}$$ Applying (\[div\]) and the Stokes’ theorem on , we obtain the Reilly type integral formula. \[theo RF\] Let $(M^n, \bar g, V=e^u)$ be an $n$-dimensional Riemannian triple and $\a,\g\in \mathbb{R}$. Let $D=D^{\a,\g}$ be the affine connection defined as in (\[defi D\]) and $\tau=(n+1)\a+\g$. Let $\phi$ be a smooth function on a bounded domain $\O\subset M$ with smooth boundary $\S$. Then the following integral formula holds: $$\begin{aligned} &\di\int_{\O}V^\tau \left[\left\|\bd^D \phi\right\|^2-\left\|\bn^{2,D} \phi\right\|_{\bar g}^2 \right] -V^\tau Ric^{D}( \nb^D \phi, \nb^D \phi) d\O \\ =&\di\int_{\S}V^{\tau}\left[H^D\<\bn^D \phi,\nu\>^2+\left(h-\g u_\nu g\right)(\n^D\phi, \n^D \phi)-2V^{-\g}\<\n^D \phi, \n^D(V^\g \phi_\nu)\>\right]dA.\nonumber \end{aligned}$$ Using (\[div\]), we have $$\begin{aligned} &\di\int_{\O}V^\tau D_i\Big((\nb^D \phi)^i D_j((\nb^D \phi)^j)-(\nb^D \phi)^j D_j((\nb^D \phi)^i)\Big) d\O\nonumber\\ =&\di\int_{\O}\bar\nabla_i\Big[V^\tau\Big((\nb^D \phi)^i D_j((\nb^D \phi)^j)-(\nb^D \phi)^j D_j((\nb^D \phi)^i)\Big)\Big] d\O\nonumber\\ =&\di\int_{\S}V^{\tau} \Big(\<\bn^D\phi,\nu\>D_j((\nb^D \phi)^j)-V^{\g-\alpha}\phi^j\<D_j(\nb^D \phi),\nu\>\Big) dA. \end{aligned}$$ Thus we only need to simplify the boundary term. At any fixed point $P\in \S$, we choose normal coordinates with respect to $\bar g$ such that the indices $a=1,\cdots, n-1$ represents coordinates on $\S$ and $j=1,\cdots,n-1, \nu$ for coordinates on $\O$. For simplicity, we will not distinguish upper and lower indexes and we denote $\phi_{ij}$ as the Hessian $\phi$ with respect to the Levi-Civita connection. Using (\[Bochner equ 2\]), we obtain $$\begin{aligned} & \<\bn^D\phi,\nu\>D_j((\nb^D \phi)^j) =V^{2(\g-\a)}\big(\bar\De\phi\phi_\nu+(2\g+n\a)\<\bn u, \bn \phi\>\phi_\nu\big),\label{theo RF equ 2} \end{aligned}$$ $$\begin{aligned} &V^{\g-\a}\phi^j\<D_j((\nb^D \phi)),\nu\>=V^{2(\g-\a)}\big(\phi_{j\nu}\phi_j+\g u_\nu\phi_j^2+(\a+\g) \<\bn u, \bn \phi\> \phi_\nu\big). \label{theo RF equ 3} \end{aligned}$$ Combining (\[theo RF equ 2\]) and (\[theo RF equ 3\]), we have $$\begin{aligned} \label{theo RF equ 5} &\<\bn^D\phi,\nu\>D_j((\nb^D \phi)^j)-V^{\g-\a}\phi^j\<D_j(\nb^D \phi),\nu\>\\ =&V^{2(\g-\a)}\Big(\phi_\nu\phi_{aa}-\phi_a\phi_{a\nu}+(n-1)\a u_\nu\phi_\nu^2+(\g+(n-1)\a) u_a\phi_a\phi_\nu-\g\phi_a^2u_\nu\Big)\nonumber\\ =&V^{2(\g-\a)}\Big(H\phi_\nu^2+\phi_ah_{ab}\phi_b+\phi_\nu\De\phi-\phi_a\n_a\phi_\nu\nonumber\\ &+(n-1)\a u_\nu\phi_\nu^2+(\g+(n-1)\a) u_a\phi_a\phi_\nu-\g\phi_a^2u_\nu\Big),\nonumber \end{aligned}$$ where we used in the last equality the Gauss-Weigarten formula $$\begin{aligned} \phi_{aa}=\De\phi+H\phi_\nu,\qquad \phi_{a\nu}=\n_a\phi_\nu-h_{ab}\phi_b.\end{aligned}$$ Using $H^D=H+(n-1)\a u_\nu$ in , we get $$\begin{aligned} \label{theo RF equ 6} &\di\int_{\S}V^{\tau} \Big(\<\bn^D\phi,\nu\>D_j((\nb^D \phi)^j)-\phi^j\<D_j(\nb^D \phi),\nu\>\Big) dA\\ =&\di\int_{\S}V^{\tau+2(\g-\a)}\Big(H^D\phi_\nu^2+(h-\g u_\nu g)(\n\phi, \n \phi)\Big)\nonumber\\&+\int_\S V^{\tau+2(\g-\a)}\Big(\phi_\nu\De\phi -\<\n\phi,\n\phi_\nu\>+ (\g+(n-1)\a)\< \n u,\n \phi\>\phi_\nu\Big)dA. \nonumber \end{aligned}$$ Integrating by parts for the last line of and noting $-[\tau+2(\g-\a)]+(\g+(n-1)\a)=-2\g$, we get the assertion. Heintz-Karcher type and Minkowski type inequalities {#sec inequalities} =================================================== In this section, we will give proofs to the Heintz-Karcher type and Minkowski type inequalities stated in Theorem \[thm ineq\].\ [Proof of Theorem \[thm ineq\] (i).]{}Recall that $\bd^D \phi=V^{\g-\a} \left[\bd \phi+(2\g+ n\a) \bar g\left(\bn u, \bn \phi\right)\right].$ We know from the standard elliptic PDE theory that the following Dirichlet boundary value problem $$\label{Dirichlet3} \left\{ \begin{array}[]{rlll} \di \bd^D \phi&=&1&\mathrm{ in }\ \O,\\ \phi&=& 0 &\mathrm{ on }\ \S,\\ \end{array} \right.$$ admits a unique smooth solution $\phi\in C^{\infty}(\overline{\O})$. We will use the solution $\phi$ of the Dirichlet problem (\[Dirichlet3\]) in . For $\tau=(n+1)\a+\g$, we have $$\begin{aligned} \label{case2-1} \frac{n-1}{n}\int_\O V^\tau=&\frac{n-1}{n}\int_\O V^{\tau}(\bd^D \phi)^2\nonumber\\ \geq& \int_\O V^{\tau}\Big[\|\bd^D \phi\|^2-\|\bn^{2,D}\phi\|^2\Big]\nonumber\\ \ge &\int_{\S} V^{\tau}H^D \<\bn^D \phi,\nu\>^2, \end{aligned}$$ where in the first inequality, we have used the Cauchy-Schwarcz inequality, in the second inequality we have used integral formula (\[formula RF 1\]), the nonnegativity of $Ric^{D}$, and the Dirichlet boundary condition. On the other hand, using equation , divergent structure , and integration by parts,we have $$\begin{aligned} \label{case2-2} \int_\O V^\tau &=&\int_{\O}V^{\tau} \bd^D \phi\\ &=&\int_{\S} V^{\tau}\<\bn^D \phi,\nu\>. \nonumber\end{aligned}$$ Combining , and using Hölder’s inequality, we obtain $$\begin{aligned} \label{equa1} \left(\int_\O V^\tau d\O\right)^2&=& \left(\int_\S V^{\tau}\<\bn^D \phi,\nu\> dA\right)^2\\ &\leq& \int_\S V^{\tau}H^D\<\bn^D \phi,\nu\>^2 dA\int_\S \frac{V^\tau}{H^D}dA\nonumber \\&\leq &\frac{n-1}{n}\int_\O V^\tau d\O\int_\S \frac{V^\tau}{H^D}dA.\nonumber\end{aligned}$$ The assertion for the inequality follows. If the equality holds, we have $$\begin{aligned} \label{equa2}\bn^{2,D}\phi=\frac1n \bar g.\end{aligned}$$ Restricting on $\S$, using $\phi=0$ on $\S$ and Gauss formula, we conclude that $h_{\a\b}=\l g_{\a\b}$ for some smooth function $\l$, i.e., $\S$ is umbilic. [Proof of Theorem \[thm ineq\] (ii).]{} Consider the Neumann boundary value problem $$\left\{ \begin{array}[]{rlll} \di \bd^D \phi&=&1&\mathrm{ in }\ \O,\\ V^\g \phi_\nu&=& c &\mathrm{ on }\ \S,\\ \end{array} \right. \label{Neumann}$$ where $c=\frac{\int_{\O}V^\tau}{\int_{\S}V^{\tau-\a}}$. The existence and uniqueness follows from the Fredholm alternative as in standard elliptic PDE theory. We will apply the solution $\phi$ of (\[Neumann\]) to the integral formula (\[formula RF 1\]). By using the Cauchy-Schwarz inequality, the equation and boundary condition in (\[Neumann\]) and the curvature assumptions, we get $$\begin{aligned} \frac{n-1}{n}\int_\O V^\tau d\O&\geq &\int_\S V^\tau H^D \<\bn^D\phi,\nu\>^2dA \\&=& \int_\S V^\tau H^D V^{2\g-2\a}\phi_\nu^2dA \\&=& c^2\int_\S V^{\tau-2\a} H^D dA.\end{aligned}$$ Inserting the value of $c$ we get the assertion. If the equality holds, we have $$\begin{aligned} \label{equa3}\bn^{2,D}\phi=\frac1n \bar g,\end{aligned}$$ and $$\begin{aligned} \label{equa4}(h-\g u_\nu g)(\nabla \phi,\nabla \phi)=0.\end{aligned}$$ Since by assumption $(h_{\a\b}-\g\frac{V_{\nu}}{V}g_{\a\b})>0$ on $\S$, it follows from that $\phi=0$ on $\S$. Restricting on $\S$, we see that $\S$ is umbilical. Poincaré type and Lichnerowicz type inequalities {#sec Lich} ================================================ Along the same line of the above results, we now prove a Poincare type inequality. \[Poincare\] Let $(M^n, \bar g, V=e^u)$ be an $n$-dimensional Riemannian triple and $\a,\g\in \mathbb{R}$. Let $\O$ be a bounded domain in $M$ with smooth boundary $\S$. Assume $Ric^{D}$ of $D=D^{\a, \g}$ is positive definite. For any $f\in C^\infty(M)$ and $\tau=(n+1)\a+\g$, if one of the following alternatives holds, - $\S=\emptyset$ and $\di\int_{\O}f V^\tau d\O=0$; - $\S\neq\emptyset$, $f\equiv 0$ on $\S$ and $H^D\ge0$; - $\S\neq \emptyset$, $\di\int_{\O}f V^\tau d\O=0$ and $\S$ satisfies $h-\g u_\nu g\ge 0$. Then we have $$\begin{aligned} \frac{n}{n-1}\di\int_{\O}f^2V^\tau d\O\le \di\int_{\O}\left<{(\mathrm{Ric}^{D})}^{-1}\nb f,\nb f\right>V^\tau d\O. \label{Poincare case 1} \end{aligned}$$ The proof is similar as in [@Mi] while we use $Ric^D$ in this paper. In case (i), we solve PDE $$\begin{aligned} \label{closed Poincare} \bd^D \phi=f \hbox{ in }\O.\end{aligned}$$ In case (ii), we solve the Dirichlet boundary value problem below, $$\label{Dirichlet Poincare} \left\{ \begin{array}[]{rlll} \di \bd^D \phi&=&f&\mathrm{ in }\ \O,\\ \phi&=& 0 &\mathrm{ on }\ \S.\\ \end{array} \right.$$ In case (iii), we solve the Neumann boundary value problem $$\label{Neumann Poincare} \left\{ \begin{array}[]{rlll} \di \bd^D \phi&=&f&\mathrm{ in }\ \O,\\ \phi_{\nu}&=& 0 &\mathrm{ on }\ \S.\\ \end{array} \right.$$ Problems and are solvable since $\di\int_{\O}fV^\tau d\O=0$. In all these three cases, we apply the integral formula (\[formula RF 1\]) to the solutions of the PDEs, i.e. $\phi$ satisfying $\bd ^D \phi =f$. By Cauchy-Schwarz inequality, we have $$\begin{array}[]{rll} & \di\frac{n-1}{n}\int_{\O}f^2V^\tau - \int_{\O}V^\tau Ric^{D}(\bn^D \phi, \bn^D \phi),\\ \ge&\di\int_{\S}V^{\tau}\left[H^D\<\bn^D \phi,\nu\>^2+\left(h-\g u_\nu g\right)(\n^D\phi, \n^D \phi)-2V^{-\g}\<\n^D \phi, \n^D(V^\g \phi_\nu)\>\right]dA.\label{equ extra 1} \end{array}$$ where the right hand side contains only boundary integrations. Next, we will show that in all these cases, the right hand side boundary integrations are nonnegative. In case (i), the boundary $\S=\emptyset$ and the result follows immediately. In case (ii), the first boundary integral in (\[equ extra 1\]) is nonnegative since $H^D\ge0$. Recall from Definition \[notation\] (i), $\n^D \phi=V^{\gamma -\alpha}\n \phi$ where $\n $ is the covariant derivative with respect to the induced metric of the boundary $\S$. Thus we have $\n^D\phi \equiv 0$, since $\phi\equiv 0$ on $\S$. In case (iii), $\phi_\nu\equiv 0$ on $\S$. We observe that $\left<\bar\n^D\phi, \nu\right>=V^{\g-\alpha}\phi_\nu\equiv 0$ and $\n^D(V^\g\phi_\nu)=V^{\g-\alpha}\n(V^\g\phi_{\nu})\equiv0$. Thus the first and last boundary terms in (\[equ extra 1\]) are all zero. Under the condition that $h-\g u_\nu g\ge0$, we conclude that the right hand side boundary terms are all nonnegative. Equivalently, we have shown, in all the three cases, the following inequality holds, $$\begin{aligned} \di\frac{n-1}{n}\int_{\O}f^2V^\tau \ge \int_{\O}V^\tau Ric^{D}(\bn^D \phi, \bn^D \phi). \label{Neumann Poincare equ 1}\end{aligned}$$ On the other hand, by integration by parts and recall Definition \[notation\], also noting the divergence property , we have $$\begin{aligned} \begin{array}[]{rll} \di\int_{\O}f^2V^\tau =&\di\int_{\O}fV^\tau \bd^D \phi=\int_{\O} f\bar\n(V^{\tau}\bar\n^D\phi )\\ \\ =&\displaystyle\int_{\S} fV^{\tau}\bar\n^D_\nu\phi -\int_{\O}V^{\tau}\<\bn^D \phi, \bar \n f\>\\ \\ =&\displaystyle-\int_{\O}V^{\tau}\<\bn^D \phi, \bar \n f\>,\\ \end{array} \label{Neumann Poincare equ 1.2}\end{aligned}$$ where the last identity holds due to the following observations: case(i), $\S=\emptyset$; case (ii), $f\equiv 0$ on $\S$; case (iii), $\phi_\nu\equiv 0$ on $\S$. Applying Hölder’s inequality to (\[Neumann Poincare equ 1.2\]), we obtain $$\begin{aligned} \label{Neumann Poincare equ 2} \di&&\left(\int_{\O}f^2V^\tau \right)^2 \le \left(\di\int_{\O}V^\tau Ric^{D}(\bn^D \phi,\bn^D\phi)\right)\left(\int_{\O}V^\tau \left<{(Ric^{D})}^{-1}\nb f,\nb f\right>\right).\end{aligned}$$ Combining (\[Neumann Poincare equ 1\]) and (\[Neumann Poincare equ 2\]), we proved . As a consequence, we get the Lichnerowicz type inequality for the first eigenvalue of $D$-Laplacian $\bd^D$. [Proof of Theorem \[thm ineq\] (iii).]{} By using the divergence property , One sees directly that the first eigenvalues have the following variational representation $$\begin{aligned} \lambda_1^{Dir} =\inf_{\{f\in C^1(\overline \O); f\not\equiv 0, f\|_{\S}=0\}}\di\frac{\int_\O\|\bar \n f\|^2V^{\tau+\g-\a} d\O}{\int_M f^2V^\tau d\O}, \label{eigenvalue} \end{aligned}$$ $$\begin{aligned} \lambda_1^{Neu} =\inf_{\{f\in C^1(\overline \O); f\not\equiv 0, \int_\O f V^\tau=0\}}\di\frac{\int_M\|\bar \n f\|^2V^{\tau+\g-\a} d\O}{\int_M f^2V^\tau d\O}. \label{eigenvalue Neumann} \end{aligned}$$ The assertion follows immediately from Theorem \[Poincare\] above by using the curvature condition $Ric^D\ge (n-1)V^{\a-\g}\bar g$. [Proof of Corollary \[coro\].]{} Let $\a=0, \g=1$. It is sufficient to observe the following fact: the Dirichlet boundary problem for PDEs $$\left\{ \begin{array}[]{rlll} \di \bd^D\phi &=&-\lambda_1^{Dir} \phi&\mathrm{ in }\ \O,\\ \phi&=& 0 &\mathrm{ on }\ \S,\\ \end{array} \right.$$ and $$\left\{ \begin{array}[]{rlll} \di V\bar \De f-\bar \De Vf&=&-\lambda_1^{Dir} f&\mathrm{ in }\ \O,\\ f&=& 0 &\mathrm{ on }\ \S,\\ \end{array} \right.$$ are equivalent under the correspondence $\phi=\frac{f}{V}$. The same phenomenon holds for Neumann boundary problems. Then Corollary \[coro\] follows from Theorem \[thm ineq\] (iii). Appendix: Proof of Proposition \[thm Ricci\] {#appenA} ============================================ In this appendix we prove Proposition \[thm Ricci\]. We only need to prove it at any fixed point $P$ under normal coordinates system $\{\P_i\}_{i=1}^n$ such that $\bar g_{ij}(P)=\delta_{ij}$, $\P_k\bar g_{ij}(P)=0$, and $\nb_{\P_i}\P_j(P)=0$. Throughout the proof we will use these properties implicitly. We use $u_i:=du(\P_i)$ and let $u_{ij}$ be the Hessian of $u$ with respect to $\nb$. By definition, we first observe that $$\begin{aligned} D_i\P_j=\nb_i\P_j+\a u_i\P_j+\a u_j\P_i+\g \d_{ij}\nb u, \label{thm Ricci equ1} \end{aligned}$$ and $$\begin{aligned} D_k(D_i\P_j)=&\nb_k(D_i\P_j)+\a u_k(D_i\P_j)+\a du(D_i\P_j)\P_k+\g \bar g(\P_k,D_i\P_j)\nb u. \label{thm Ricci equ2} \end{aligned}$$ Next, we compute each term on the right of (\[thm Ricci equ2\]). It turns out $$\begin{aligned} \nb_k(D_i\P_j)=&\nb_k(\nb_i\P_j)+\a u_{ik}\P_j+\a u_{jk}\P_i+\g \d_{ij}\nb_k\nb u,\nonumber\\ \a u_k (D_i\P_j)=&\a^2u_ku_i\P_j+\a^2 u_ku_j\P_i+\a\g \d_{ij}u_k\nb u,\nonumber\\ \a du(D_i\P_j)\P_k=& 2\a^2u_iu_j\P_k+\a\g \d_{ij}\|\nb u\|^2\P_k,\nonumber\\ \g \bar g(\P_k,D_i\P_j)\nb u=&(\a\g u_i \d_{jk}+\a\g u_j\d_{ik})\nb u +\g^2\d_{ij}u_k\nb u. \label{thm Ricci equ3} \end{aligned}$$ Adding up these four terms into , we have $$\begin{aligned} D_k(D_i\P_j)=&\nb_k(\nb_i\P_j)+\a u_{ik}\P_j+\a u_{jk}\P_i+\g \d_{ij}\nb_k\nb u\nonumber\\ &+\a^2u_ku_i\P_j+\a^2 u_ku_j\P_i+\a\g \d_{ij}u_k\nb u\nonumber\\ &+ 2\a^2u_iu_j\P_k+\a\g \d_{ij}\|\nb u\|^2\P_k\nonumber\\ &+(\a\g u_i \d_{jk}+\a\g u_j\d_{ik})\nb u +\g^2\d_{ij}u_k\nb u. \label{thm Ricci equ4}\end{aligned}$$ Using the metric tensor $\bar g$, we have $$\begin{aligned} \left< D_k(D_i\P_j),\P_k\right>_{\bar g}=&\left< \nb_k(\nb_i\P_j),\P_k\right>_{\bar g}+2\a u_{ij}+\g \bar\De u \d_{ij}\nonumber\\ &+\big[\a^2+(2n+1)\a^2+2\a\g\big]u_iu_j\nonumber\\ &+\big[(n+1)\a\g+\g^2\big]\|\nb u\|^2 \d_{ij}. \label{thm Ricci equ5}\end{aligned}$$ By swapping $k$ and $i$ in (\[thm Ricci equ4\]), we obtain $$\begin{aligned} D_i(D_k\P_j)=&\nb_i(\nb_k\P_j)+\a u_{ik}\P_j+\a u_{ji}\P_k+\g \d_{kj}\nb_i\nb u\nonumber\\ &+\a^2u_ku_i\P_j+\a^2 u_iu_j\P_k+\a\g \d_{kj}u_i\nb u\nonumber\\ &+ 2\a^2u_ku_j\P_i+\a\g \d_{kj}\|\nb u\|^2\P_i\nonumber\\ &+(\a\g u_k \d_{ij}+\a\g u_j\d_{ik})\nb u +\g^2\d_{kj}u_i\nb u. \label{thm Ricci equ4.1}\end{aligned}$$ Similarly, we have $$\begin{aligned} \left< D_i(D_k\P_j),\P_k\right>_{\bar g} =& \left< \nb_i(\nb_k\P_j),\P_k\right>_{\bar g}+\big[(n+1)\a+\g\big]u_{ij}+2\a\g\|\nb u\|^2\d_{ij}\nonumber\\ &+\big[(n+3)\a^2+2\a\g+\g^2\big]u_iu_j. \label{thm Ricci equ5.1}\end{aligned}$$ Combining (\[thm Ricci equ5\]) and (\[thm Ricci equ5.1\]), using Definition \[defi Ricci\] we have $$\begin{aligned} R^D_{ij}= &\left< R^D(\P_k,\P_i)\P_j,\P_k\right>_{\bar g} \nonumber\\ =& \left< D_k(D_i\P_j),\P_k\right>_{\bar g} - \left< D_i(D_k\P_j),\P_k\right>_{\bar g} \nonumber\\ =&R_{ij}-\big[\a(n-1)+\g\big]u_{ij}+\big[(n-1)\a^2-\g^2\big]u_iu_j\nonumber\\ &+\g \bar\De u\d_{ij}+\big[(n-1)\a\g+\g^2\big]\|\nb u\|^2\d_{ij}. \label{thm Ricci equ 6}\end{aligned}$$ This finishes the proof for $Ric^D$.\ Part of the work was done during both authors’€˜ visit at McGill University in Fall, 2015. We would like to thank the department for the hospitality and professor Pengfei Guan for constant support. The authors also would like to thank the referee for his critical reading and valuable suggestions. [99]{} Bakry, D.; Émery, Michel. [Diffusions hypercontractives.]{} Séminaire de Probabilités, XIX, 1983/84, 177-€“206, Lecture Notes in Math., 1123, Springer, Berlin, 1985. Brendle, S., [Constant mean curvature surfaces in warped product manifolds,]{} Publ. Math. Inst. Hautes Études Sci. 117 (2013), 247-269. Brendle, S.; Hung, P.K. ; Wang, M.T., [A Minkowski inequality for hypersurfaces in the Anti-de Sitter-Schwarzschild manifold,]{} Comm. Pure Applied Math. 69 (2016) Issue 1, 124-144. Po-Ning Chen, Mu-Tao Wang, Ye-Kai Wang, Shing-Tung Yau, [Quasi-local energy with respect to a static spacetime]{}, arXiv:1604.02983. Corvino, J., [Scalar curvature deformation and a gluing construction for the Einstein constraint equations.]{} Comm. Math. Phys. 214 (2000), no. 1, 137-189. E. Heintze and H. Karcher, [A general comparison theorem with applications to volume estimates for submanifolds,]{} Ann. Sci. École Norm. Sup. 11, 451-470 (1978). Lichnerowicz, A.,Géométrie des groupes de transformations. Travaux et recherches mathématiques, Vol. III, Dunod, Parsi, 1958. Kolesnikov, Alexander V.; Milman Emanuel. [Poincaré and Brunn-Minkowski inequalities on weighted Riemannian manifolds with boundary]{}. arXiv:1310.2526v5 Li, Junfang; Xia, Chao. [An integral formula and its applications on sub-static manifolds,]{} arXiv:1603.02201, 2016 Minkowski, H., Volumen und Oberfläche. Math. Ann. 57 (1903) 447-495. Petersen, Peter. [Riemannian Geometry, Second Edition.]{} New York: Springer-Verlag, GTM 171, 2006. Qiu, G.; Xia, C., [A Generalization of Reilly’s Formula and its Applications to a New HeintzeÃ¢Â€Â“Karcher Type Inequality.]{} Int. Math. Res. Not. IMRN 2015, no. 17, 7608-7619. Reilly, R. C., [On the Hessian of a function and the curvatures of its graph.]{} Michigan Math. J. 20 (1973), 373-383. Reilly, R. C., [Applications of the Hessian operator in a Riemannian manifold.]{} Indiana Univ. Math. J. 26 (1977), no. 3, 459-472. Reilly, R. C., Geometric applications of the solvability of Neumann problems on a Riemannian manifold. Arch. Rational Mech. Anal. 75 (1980), no. 1, 23-29. Ros, A., [Compact hypersurfaces with constant higher order mean curvatures,]{} Revista Mathmática Iberoamericana, 3 (1987) 447-453. Wang, Xiaodong; Wang, Y.-K. [Brendle’s inequality on static manifolds]{}. arXiv:1603.00379 Wang, M.-T.; Wang, Y.-K.; Zhang, X., [Minkowski formulae and Alexandrov theorems in spacetime,]{} J. Differential Geom. to appear. arXiv: 1409.2190. Wylie, William; Yeroshkin, Dmytro. [On the geometry of Riemannian manifolds with density.]{} arXiv:1602.08000. Wang, Yu; Zhang, Xiangwen. [An Alexandroff-Bakelman-Pucci estimate on Riemannian manifolds.]{} Adv. Math. 232 (2013), 499â-512. Xia, C., [A Minkowski type inequality in space forms,]{} Calc. Var. Partial Differential Equations 55 (2016), no. 4, 55:96. [^1]: Research of CX is supported in part by NSFC (Grant No. 11501480) and the Fundamental Research Funds for the Central Universities (Grant No. 20720150012).	{ "pile_set_name": "ArXiv" }
ULM-TP/02-5\ July 2002 [Torus quantization for spinning particles]{}\ [Stefan Keppeler]{}\ [Abteilung Theoretische Physik, Universität Ulm, Albert-Einstein-Alle 11, D-89069 Ulm, Germany]{}\ [[email protected]]{}\ PACS numbers: 03.65.Sq, 03.65.Pm, 31.15.Gy [2]{} Semiclassical methods for multi-component wave equations have been a topic of constant interest over the last decade, both for their physical applications and the mathematical structures behind them [@LitFly91a; @LitFly91b; @EmmWei96; @BolKep98]. In a seminal article Littlejohn and Flynn [@LitFly91b] summarized some of the previous efforts in this direction, stressed the importance of geometric or Berry phases in this context and developed a general quantization scheme. Their method, however, does not cover situations in which the so-called principal Weyl symbol of the Hamiltonian has (globally) degenerate eigenvalues. But this problem shows up for the Dirac equation, as we will explain below. It was emphasized by Emmrich and Weinstein [@EmmWei96] that in such a situation integrability of the so-called ray Hamiltonians (which in our case will be given by $H^+$ and $H^-$ defined in eq. (\[H+-\]) below) is not a sufficient condition that allows for an explicit semiclassical quantization. We discuss this problem for the particular case of the Dirac equation, but our method also translates to more general situations. The semiclassical analysis of the Dirac equation was started by Pauli [@Pau32] who showed that the rapidly oscillating phase of a WKB-like ansatz has to solve a relativistic Hamilton-Jacobi equation. Later Rubinow and Keller [@RubKel63] related the amplitude of the semiclassical solution to classical spin precession (i.e. Thomas precession [@Tho27]). So far, however, all these efforts did not result in general semiclassical quantization conditions as they were put forward by Keller for the Schrödinger equation [@Kel58]. In this article we present the main steps in the derivation of such conditions, finally leading to eq. (\[quant\_cond\]) below. To this end it will be necessary to extend the notion of integrability, see [@Arn78], from Hamiltonian systems to a certain skew product flow which arises naturally in the semiclassical treatment of the Dirac equation. We also illustrate our method for an example, namely the relativistic Kepler problem, which yields Sommerfeld’s fine structure formula. We first briefly summarize the determination of semiclassical wave functions for the Dirac equation. Details can be found in [@RubKel63; @BolKep98; @BolKep99a]. Consider the stationary Dirac equation $\op{H}_{\rm D} \Psi = E \Psi$ with Hamiltonian $$\op{H}_{\rm D} = c\vecalph\left(\frac{\hbar}{\ui} \nabla - \frac{e}{c} \vecA(\vecx) \right) + \beta mc^2 + e\phi(\vecx)$$ defined on a suitable domain in $L^2(\R^3) \otimes \C^4$. It describes the motion of a particle with mass $m$, charge $e$ and spin $\frac{1}{2}$ in electro-magnetic potentials $\phi$ and $\vecA$. The Dirac algebra is realized by the $4\times4$ matrices $$\vecalph = \begin{pmatrix} 0 & \vecsig \\ \vecsig & 0 \end{pmatrix} \quad \text{and} \quad \beta = \begin{pmatrix} \eins_2 & 0 \\ 0 & -\eins_2 \end{pmatrix} \, ,$$ where $\vecsig$ is the vector of Pauli matrices and $\eins_2$ denotes the $2\times2$ unit matrix. We make a semiclassical ansatz of the form $$\label{scAnsatz} \Psi(\vecx) = \left( \sum_{k\geq0} \left(\frac{\hbar}{\ui}\right)^k a_k(\vecx) \right) \ue^{\frac{\ui}{\hbar}S(\vecx) }$$ with a scalar phase function $S$ and spinor-valued amplitudes $a_k$. Inserting this ansatz into the Dirac equation and sorting by orders of $\hbar$ in leading order one finds $$\label{eq:leading_order} \left[ H_{\rm D}(\nabla S, \vecx) - E \right] a_0 = 0$$ with the matrix-valued function $$H_{\rm D}(\vecp,\vecx) = c\vecalph\left( \vecp - \frac{e}{c} \vecA(\vecx) \right) + \beta mc^2 + e\phi(\vecx) \, ,$$ on classical phase space. The system (\[eq:leading\_order\]) of linear equations only has a solution with non-trivial $a_0$ if the expression in square brackets has an eigenvalue zero, i.e. if $S$ solves one of the two Hamilton-Jacobi equations $H^\pm(\nabla S,\vecx) = E$ with classical Hamiltonians $$\label{H+-} H^\pm(\vecp,\vecx) = e\phi \pm \sqrt{ c^2 \left(\vecp - \frac{e}{c} \vecA \right)^2 + m^2c^4} \,$$ for particles with positive and negative kinetic energy, respectively. From standard Hamilton-Jacobi theory, see e.g. [@Arn78], we conclude that the rapidly oscillating phase of the wave function (\[scAnsatz\]) can be determined by integration along solutions $(\vecP_\pm(t),\vecX_\pm(t))$ of Hamilton’s equations of motion generated by the Hamiltonians (\[H+-\]). Locally we have $\vecP_\pm(t) = \nabla S^\pm(\vecX_\pm(t))$, and thus $$\label{S_als_Wegintegral} S^\pm(\vecx) = S^\pm(\vecy) + \int_{\vecy}^{\vecx} \vecP_\pm \, \ud \vecX_\pm$$ where we denote by $\vecy = \vecX_\pm(0)$ the (arbitrarily chosen) starting point of integration. If we set $\vecxi := \vecP_\pm(0)$ we can also write $(\vecP_\pm(t),\vecX_\pm(t)) = \phi_{H^\pm}^t(\vecxi,\vecy)$ with the Hamiltonian flows $\phi_{H^\pm}^t$. The eigenspaces corresponding to the eigenvalues $H^\pm(\vecp,\vecx)$ of $H_D(\vecp,\vecx)$ have dimension two and we denote by $V_\pm(\vecp,\vecx)$ the $4\times2$ matrices of orthonormal eigenvectors, i.e. $V_+^\dag V_+ = \eins_2 = V_-^\dag V_-$, $V_+^\dag V_- = 0 = V_-^\dag V_+$ and $V_+ V_+^\dag + V_- V_-^\dag = \eins_4$, see [@BolKep99a] for details. For concreteness we now seek a semiclassical wave function corresponding to the classical dynamics with $H^+$, and in order to simplify notation drop the index “$+$”. Since eq. (\[eq:leading\_order\]) is a matrix equation it does not only require $S$ to solve the Hamilton-Jacobi equation, but also $a_0$ to be of the form $a_0(\vecx) = V(\nabla S,\vecx) \, b(\vecx)$ with a $\C^2$-valued $b$. An equation for $b$ can be derived from the next-to-leading order equation, obtained when inserting the semiclassical ansatz (\[scAnsatz\]) into the Dirac equation, by multiplication with $V_+^\dag$ from the left, cf. [@RubKel63; @BolKep98; @BolKep99a], $$\begin{aligned} \begin{split}\label{transportb} (\nab{p}H) \nab{x} b &+ \frac{\ui}{2} \vecsig \vcB(\nab{x} S,\vecx) b\\ &+ \frac{1}{2} \nab{x}[ \nab{p}H(\nab{x} S,\vecx)] b= 0 \, , \end{split}\\ \vcB(\vecp,\vecx) :&= \frac{ec^2}{\varepsilon(\varepsilon+mc^2)} \left( \vecp-\frac{e}{c}\vecA \right) \times \vecE - \frac{ec}{\varepsilon} \vecB \, . \end{aligned}$$ Here we used the abbreviation $\varepsilon := \sqrt{(c\vecp-e\vecA)^2+m^2c^4}$, and $\vecE(\vecx) = -\nabla\phi(\vecx)$ and $\vecB(\vecx)=\nabla\times\vecA(\vecx)$ denote the electric and magnetic fields, respectively. Viewed as an equation along the orbit $\phi_H^t(\vecxi,\vecy)$, the first term in (\[transportb\]) constitutes a time derivative along the classical translational dynamics which we shall denote by a dot. The solution of (\[transportb\]) with vanishing $\vcB$ is known to be given by $\sqrt{\det\frac{\partial\vecy}{\partial\vecx}}$, see e.g. [@Kel58], and thus the ansatz $b = \sqrt{\det\frac{\partial\vecy}{\partial\vecx}} \, u$ leaves us with the spin transport equation $$\label{spintransport} \dot{u} + \frac{\ui}{2} \vecsig \vcB(\phi_H^t(\vecxi,\vecy)) \, u = 0 \, .$$ The solution of (\[spintransport\]) can be written as $u(t) = d(\vecxi,\vecy,t) u(0)$ with an $\SU(2)$-matrix $d(\vecxi,\vecy,t)$. We explicitly indicate the dependence on the initial point $(\vecxi,\vecy)$ of the classical trajectory along which we integrate until time $t$. Through the covering map $\varphi: \SU(2) \to \SO(3)$ we can associate with the spin transporter $d$ a rotation matrix $R(\vecxi,\vecy,t)$, and one easily verifies that $\vecs(t) := R(\vecxi,\vecy,t) \, \vecs(0)$ solves the spin precession equation $$\label{spinprec} \dot{\vecs} = \vcB(\phi_H^t(\vecxi,\vecy)) \times \vecs$$ on the two-sphere $S^2$ (i.e. $\vecs \in \R^3$, $\|\vecs\|=1$). This is the equation of Thomas precession [@Tho27] thus emerging from a semiclassical analysis of the Dirac equation. It turns out that all properties of the semiclassical wave function $\Psi \sim a_0 \exp(\frac{\ui}{\hbar}S)$ can be determined from the solution $\phi_H^t(\vecxi,\vecy)$ of Hamilton’s equations of motion and the solution $\vecs(t)$ of eq. (\[spinprec\]). Thus the skew product $$\label{Ycl} Y_{\rm cl}^t (\vecxi,\vecy,\vecs(0)) := \left( \phi_{H}^t(\vecxi,\vecy), R(\vecxi,\vecy,t) \vecs(0) \right) ,$$ which defines a flow on the extended classical phase space $\R^{2d} \times S^2$, should be considered as the classical dynamical system corresponding to the Dirac equation, cf. [@BolKep99b; @BolGlaKep01]. The key question in semiclassical quantization is now whether it is possible to find a single-valued wave function $\Psi \sim a_0 \exp(\frac{\ui}{\hbar}S)$ which solves the above equations. Let us briefly recall the procedure in the spinless case [@Kel58]. In standard semiclassics for the Schrödinger equation one invokes integrability of the classical flow $\phi_H^t$: Besides the classical Hamiltonian $H=:A_1$ there are $d-1$ further conserved quantities, $A_2, \hdots, A_d$ (for a system with $d$ degrees of freedom; we only specialize to $d=3$ later) with mutually vanishing Poisson brackets, $\{A_j,A_k\}=0$. Then the Theorem of Liouville and Arnold, see [@Arn78 chapter 10], guarantees that a (compact and connected) invariant level set $\{(\vecp,\vecx) \, \| \, \vecA = const. \}$ has the topology of a $d$-torus $\T^d$ on which the flows $\phi_{A_1}^t, \hdots, \phi_{A_d}^t$ generated by $A_1,\hdots,A_d$ commute. By integration along the flow lines of $\phi_{A_2}^t, \hdots, \phi_{A_d}^t$ – analogous to the integration along $\phi_H^t$ in (\[S\_als\_Wegintegral\]) – this allows for a definition of the phase function $S$ which is unique up to the contributions of non-contractible loops. Demanding single-valuedness of the semiclassical wave function $\Psi \sim a_0 \ue^{\frac{\ui}{\hbar}S}$ yields the Einstein-Brillouin-Keller (EBK) quantization conditions $$\label{EBK} \oint_{\mathcal{C}_j} \vecp \, \ud\vecx = 2\pi\hbar \left( n_j + \frac{\mu_j}{4} \right) \, , \quad n_j \in \Z \, ,$$ where $\{ \mathcal{C}_j \, \| \, j=1,\hdots,d \}$ denotes a basis of non-contractible loops on the torus characterized by the action variables $I_j = \frac{1}{2\pi} \oint_{\mathcal{C}_j} \vecp \, \ud\vecx$. The number $\mu_j \in \{ 1,2,3,4 \}$ is the Maslov index, see [@MasFed81], of the cycle $\mathcal{C}_j$ which, roughly speaking, counts the number of points along $\mathcal{C}_j$ at which the pre-factor $\sqrt{\det\frac{\partial\vecy}{\partial\vecx}}$ becomes singular. All these terms also appear in the situation with non-zero spin, and we now have to examine how the spin contribution modifies this picture. When we include the spin contribution $d(\vecxi,\vecy,t)$ the situation becomes more complicated and integrability of $\phi_H^t$ will, in general, not be a sufficient condition to allow for an explicit semiclassical quantization. This can be seen as follows: Transporting the spinor-valued amplitude $u$ along a closed path $\mathcal{C}_j$ on a Liouville-Arnold torus the initial and final value, $u_{\rm i}$ and $u_{\rm f}$, respectively, differ not only by a phase but are related by an $\SU(2)$-transformation, $u_{\rm f} = d_j u_{\rm i}$, $d_j \in \SU(2)$. Mathematically speaking, we are considering a connection in a $\C^2$-bundle with $\SU(2)$-holonomy. If there was only one such loop, as in a system with one translational degree of freedom, we could choose $u_{\rm i}$ to be an eigenvector of $d_j$, thus reducing the $\SU(2)$-holonomy to a simple phase factor. However, for $d \geq 2$ degrees of freedom this is impossible since the holonomy factors for different loops are, in general, given by non-commuting elements of the holonomy group $\SU(2)$. This is a general problem in semiclassics for multi-component wave equations with globally degenerate eigenvalues of the principal symbol, as was emphasized in a general setting by Emmrich and Weinstein [@EmmWei96]. In our situation of semiclassics for spinning particles we will solve this problem by imposing additional conditions on the “field” $\vcB$, which generates the classical spin precession (\[spinprec\]). From a physical point of view it is not surprising that we need a stronger condition than just integrability of the translational dynamics $\phi_H^t$; since we identified the skew product (\[Ycl\]) as the classical dynamics corresponding to the Dirac equation, we should also say under which circumstances we want to call the spin dynamics (or rather the combination of translational and spin dynamics) integrable. We do this by the following definition.\ [Definition]{} [The skew product $Y_{\rm cl}^t$ is called integrable, if (i) the underlying Hamiltonian flow $\phi_H^t$ is integrable in the sense of Liouville and Arnold and (ii) the flows $\phi_{A_2}^t,\hdots,\phi_{A_d}^t$ can also be extended to skew products $\Ycl_j^t$ on $\R^{2d} \times S^2$ ($Y_{\rm cl}^t \equiv \Ycl_1^t$) with fields $\vcB_j$ fulfilling $$\label{spin_involution} \{A_j,\vcB_k\} + \{\vcB_j,A_k\} - \vcB_j \times \vcB_k = 0 \, .$$ ]{}Condition (\[spin\_involution\]) plays the same role as the condition $\{A_j,A_k\}=0$ does in the scalar case; it guarantees that all skew products $\Ycl_j^t$ commute [@Kep_prep]. Under these conditions we are able to prove the following theorem.\ [Theorem]{} [If the skew product flow $Y_{\rm cl}^t$ is integrable, the combined phase space $\R^{2d} \times S^2$ can be decomposed into invariant bundles $\mathcal{T}_\theta \stackrel{\pi}{\longrightarrow} \T^d$ over Liouville-Arnold tori $\T^d$ with fiber $S^1$. The bundles can be embedded in $\T^d \times S^2$ such that the fibers are characterized by the latitude with respect to a local direction $\vecn(\vecp,\vecx)$, i.e. $$\label{T_theta} \mathcal{T}_\theta = \{ (\vecp,\vecx,\vecs) \in \T^d \times S^2 \, \| \, \sphericalangle (\vecs,\vecn(\vecp,\vecx)) = \theta \} \, .$$ ]{}The proof of this theorem will be given elsewhere [@Kep_prep]. The geometry of the the invariant sets $\mathcal{T}_\theta$ is illustrated in figure $\ref{fig:torus_mit_sphaere}$: a Liouville-Arnold torus is sketched as a $2$-torus; at two different points we show the attached sphere together with the local axes $\vecn$ and a corresponding parallel of latitude. If the skew product flow $Y_{\rm cl}^t$ is integrable, the theorem allows us to construct semiclassical wave functions which imply generalized quantization conditions involving the spin degree of freedom. We briefly sketch the construction and then state the quantization conditions. As in the case without spin we define the semiclassical wave function by integration along the flow lines of $\phi_{A_1}^t, \hdots, \phi_{A_d}^t$. In addition we choose the $\C^2$-valued part $u$ such that it is an eigenvector of $\vecsig\vecn(\vecp,\vecx)$ at each point of the Liouville-Arnold torus $\T^d$. (This is only possible if the skew product $Y_{\rm cl}^t$, and not just the Hamiltonian flow $\phi_H^t$, is integrable .) Then the semiclassical wave function is unique up to the contribution of non-contractible loops on $\T^d$. Transporting a classical spin vector along such a loop $\mathcal{C}_j$ by a combination of the (commuting) skew products $\Ycl_1^t, \hdots, \Ycl_d^t$, one finds that it is rotated by an angle $\alpha_j$, while integrability of $Y_{\rm cl}^t$ ensures that it stays on the same parallel of latitude. Consequently, the semiclassical wave function is multiplied by a phase factor $\ue^{\mp\ui\alpha_j/2}$, the sign depending on whether we have chosen $u$ to be an eigenvector of $\vecsig\vecn$ with eigenvalue $+1$ or $-1$. Demanding single-valuedness of the wave function, the total phase change when moving along a loop $\mathcal{C}_j$ has to be an integer multiple of $2\pi$, yielding the quantization conditions $$\label{quant_cond} \oint_{\mathcal{C}_j} \vecp \, \ud \vecx = 2\pi \hbar \left( n_j + \frac{\mu_j}{4} + m_s \frac{\alpha_j}{2\pi} \right) \, ,$$ where in addition to the terms in (\[EBK\]) the spin contribution with the spin quantum number $m_s = \pm \frac{1}{2}$ enters. We remark that analogous quantization conditions can be derived for the Pauli equation [@Kep_prep]. There we can also choose to describe particles with arbitrary spin $s\in\frac{1}{2}\N_0$ by replacing the Pauli matrices $\vecsig$ with a higher dimensional irreducible representation of $\su(2)$. This changes neither the corresponding classical system (which is always given by a skew product on $\R^{2d} \times S^2$) nor the construction of the semiclassical solutions; only in the quantization conditions (\[quant\_cond\]) the spin quantum number $m_s$ then takes the values $-s, -s+1, \hdots, s$. We conclude by illustrating these new quantization conditions for a famous example, namely Sommerfeld’s fine structure formula [@Som16]. To this end we have to quantize the relativistic Kepler problem with classical Hamiltonian $$H(\vecp,\vecx) = -\frac{e^2}{\|\vecx\|} + \sqrt{c^2\vecp^2 + m^2c^4} \, .$$ The problem can be transformed to action and angle variables, see e.g. [@Som16], and the new Hamiltonian depends only on the two action variables $I_r$ and $L$. Here $I_r$ denotes the action variable corresponding to a radial cycle (e.g. from perihelion to aphelion and back), and $L$ is the modulus of angular momentum $\vecL = \vecx \times \vecp$. In 1916 Sommerfeld quantized this system using the old quantum theory, since quantum mechanics was still to be invented, not to think about spin or the Dirac equation. Accordingly, he chose the quantization conditions $$I_r = \hbar n_r \quad \text{and} \quad L = \hbar l$$ with integers $n_r \in \N_0$ and $l \in \N$. More than ten years later it was confirmed that the energy levels resulting from these conditions are exactly the same as one finds by solving the corresponding Dirac equation [@Gor28; @Dar28]. This is insofar surprising as the Dirac equation not only includes relativistic effects, but also takes into account spin-orbit coupling, which Sommerfeld could not know about. Quantizing the problem with the new conditions (\[quant\_cond\]) yields $$I_r = \hbar \left( n_r + \frac{1}{2} \pm \frac{\alpha_r}{2\pi} \right) \quad \text{and} \quad L = \hbar \left( l + \frac{1}{2} \pm \frac{\alpha_L}{2\pi} \right)$$ with integers $n_r$ and $l$ and a Maslov contribution of $\frac{1}{2}$ for both variables. For the spin rotation angle $\alpha_L$ we find $\alpha_L = 2\pi$ for any spherically symmetric system [@Kep_prep]. Intriguingly, for the relativistic Kepler problem $\alpha_r$ is also given by $2\pi$! Therefore, the conditions (\[quant\_cond\]) and Sommerfeld’s method yield the same values for $I_r$ and $L$, thus leading to the same energy levels. A careful analysis of the values that $n_r$ and $l$ can assume (one finds $n_r \geq 0$ and $l\geq \frac{1}{2} \mp \frac{1}{2}$) shows that with the semiclassical quantization scheme developed here one also obtains the correct multiplicities, which Sommerfeld was unable to extract with his method. Summarizing, we can say that, by a freak of nature, Sommerfeld was able to obtain the correct energy levels of the Dirac hydrogen atom because, roughly speaking, the corrections due to wave mechanics (the Maslov term $\frac{1}{2}$) and those due to the spin $\frac{1}{2}$ of the electron cancel for this particular problem. I would like to thank Jens Bolte for helpful discussions and I gratefully acknowledge financial support from the Deutsche Forschungsgemeinschaft (DFG) under contract no. Ste 241/10-2. [10]{} R. G. Littlejohn and W. G. Flynn: [[G]{}eometric [P]{}hases in the [B]{}ohr-[S]{}ommerfeld [Q]{}uantization of [M]{}ulticomponent [W]{}ave [F]{}ields]{}, Phys. Rev. Lett. [66]{} (1991) 2839–2842. R. G. Littlejohn and W. G. Flynn: [[G]{}eometric phases in the asymptotic theory of coupled wave equations]{}, Phys. Rev. A [44]{} (1991) 5239–5256. C. Emmrich and A. Weinstein: [[G]{}eometry of the transport equation in multicomponent [WKB]{} approximations]{}, Commun. Math. Phys. [176]{} (1996) 701–711. J. [B]{}olte and S. [K]{}eppeler: [[S]{}emiclassical Time Evolution and Trace Formula for Relativistic Spin-1/2 Particles]{}, Phys. Rev. Lett. [81]{} (1998) 1987–1991. W. [P]{}auli: [[D]{}iracs [W]{}ellengleichung des [E]{}lektrons und geometrische [O]{}ptik]{}, [H]{}elv. [P]{}hys. [A]{}cta [5]{} (1932) 179–199. S. I. Rubinow and J. B. Keller: [[A]{}symptotic [S]{}olution of the [D]{}irac [E]{}quation]{}, [P]{}hys. [R]{}ev. [131]{} (1963) 2789–2796. L. H. Thomas: [The Kinematics of an Electron with an Axis]{}, Philos. Mag. [3]{} (1927) 1–22. J. B. Keller: [[C]{}orrected [B]{}ohr-[S]{}ommerfeld [Q]{}uantum [C]{}onditions for [N]{}onseparable [S]{}ystems]{}, [A]{}nn. [P]{}hys. ([NY]{}) [4]{} (1958) 180–185. V. I. Arnold: [Mathematical Methods of Classical Mechanics]{}, [S]{}pringer-[V]{}erlag, [N]{}ew [Y]{}ork, (1978). J. [B]{}olte and S. [K]{}eppeler: [A semiclassical approach to the [D]{}irac equation]{}, Ann. Phys. (NY) [274]{} (1999) 125–162. J. [B]{}olte and S. [K]{}eppeler: [[S]{}emiclassical form factor for chaotic systems with spin 1/2]{}, J. Phys. A [32]{} (1999) 8863–8880. J. Bolte, R. Glaser and S. Keppeler: [Quantum and classical ergodicity of spinning particles]{}, Ann. Phys. (NY) [293]{} (2001) 1–14. V. P. Maslov and M. V. Fedoriuk: [[S]{}emi-[C]{}lassical [A]{}pproximation in [Q]{}uantum [M]{}echanics]{}, [D]{}. [R]{}eidel, [D]{}odrecht, (1981). S. Keppeler: (in preparation). A. Sommerfeld: [[Z]{}ur [Q]{}uantentheorie der [S]{}pektrallinien]{}, Ann. Phys. (Leipzig) [51]{} (1916) 1–94, 125–167. W. Gordon: [[D]{}ie [E]{}nergieniveaus des [W]{}asserstoffatoms nach der [D]{}iracschen [Q]{}uantentheorie des [E]{}lektrons]{}, Z. Phys. [48]{} (1928) 11–14. C. G. Darwin: [[T]{}he [W]{}ave [E]{}quations of the [E]{}lectron]{}, Proc. R. Soc. London Ser. A [118]{} (1928) 654–680.	{ "pile_set_name": "ArXiv" }
--- abstract: 'Content-based retrieval supports a radiologist decision making process by presenting the doctor the most similar cases from the database containing both historical diagnosis and further disease development history. We present a deep learning system that transforms a 3D image of a pulmonary nodule from a CT scan into a low-dimensional embedding vector. We demonstrate that such a vector representation preserves semantic information about the nodule and offers a viable approach for content-based image retrieval (CBIR). We discuss the theoretical limitations of the available datasets and overcome them by applying transfer learning of the state-of-the-art lung nodule detection model. We evaluate the system using the LIDC-IDRI dataset of thoracic CT scans. We devise a similarity score and show that it can be utilized to measure similarity 1) between annotations of the same nodule by different radiologists and 2) between the query nodule and the top four CBIR results. A comparison between doctors and algorithm scores suggests that the benefit provided by the system to the radiologist end-user is comparable to obtaining a second radiologist’s opinion.' author: - 'Ilia Kravets[^1]' - 'Tal Heletz[ ^fnsymbol[1]{}^]{}' - Hayit Greenspan title: ' Nodule2vec: a 3D Deep Learning System for Pulmonary Nodule Retrieval Using Semantic Representation' --- Introduction ============ Motivation and Background ------------------------- Lung cancer is a leading cause of cancer mortality in both men and women, accounting for nearly 25% of all cancer deaths[@Siegel2020]. The chances of treating lung cancer successfully are much higher if the treatment starts at an early stage. Currently, Low-Dose Computed Tomography (LDCT) screening is the most effective way for pulmonary nodules detection and diagnosis, and its usage has increased dramatically over the last two decades. However, scan examination and diagnosis is a very time-consuming task that requires a lot of invaluable radiologist time. To assist radiologists to quickly and effectively diagnose tumors, it is important to present them with similar historical cases. Examination of similar cases can be beneficial in two aspects. First, the radiologist can have access to the labeling information that other doctors gave in similar cases and thus can deduce the status of the current case. Second, the radiologist can examine the related case development history past the similarly looking LDCT scan, as if peeking at a possible future of the current case and infer a more accurate prognosis. For example, a similar case biopsy outcome can suggest whether it is advisable to perform a biopsy in the current case. Many contemporary works that apply machine learning techniques to medical imaging aim to replace the doctors and directly produce a diagnosis. The usual downsides of this approach are the lack of the output interpretability and uncertain robustness guarantees in the light of potential bugs, input variation, or even adversarial attacks. Moreover, they are usually lacking in the amount of diagnostic data they can provide, often limited by a handful of bits of information, like benign-vs-malignant binary classification. The solution we describe is a content-based image retrieval (CBIR) system. Given a pulmonary nodule, our system retrieves several similar nodules from the historical database, potentially enriched with the relevant clinical records, to aid the doctor in the diagnostic process. Our main contributions include: - We develop a system to provide the radiologist semantically-meaningful decision support in nodule analysis, in contrast to providing an automated nodule diagnosis. - We identify architectural constraints to our deep learning system and provide theoretical justification for utilizing transfer learning. - We define a proxy task to learn the transformation of a 3D nodule to a latent vector based on semantic features defined by medical experts. - We study semantic information preservation by our system, devise an evaluation technique that considers the lack of consensus between the doctors, and show that our method retrieves highly relevant results. Related Work ------------ Most of the previous research uses classic computer vision methods for feature extraction, using a 2D representation of the nodules. This approach fails to capture the full spatial nodule information. A significant challenge in CBIR is the definition of the distance between two entities so that they will be considered semantically similar in addition to being visually similar. Often, the evaluation of the models is hindered by the lack of consensus between human annotators. A few works in the field include the following: Lam et al.[@Lam2007] performed CBIR on 2D slices of the 3D nodules using classical image descriptors. Dhara et al.[@Dhara2016] extended this approach with manually defined volumetric features. Pan et al.[@Pan2016] used spectral clustering to transfer a 3D nodule to hash code used to retrieve similar nodules. Wei et al.[@Wei2017] proposed a learned distance metric. Finally, Loyman & Greenspan[@Loyman2019] study included LIDC-IDRI rating regression from 2D slices to obtain embeddings using deep learning. Methods ======= Data ---- ### LIDC-IDRI. The Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) image collection consists of diagnostic and lung cancer screening thoracic computed tomography (CT) scans with marked-up annotated lesions[@lidc_data; @lidc_paper]. A panel of four experienced radiologists performed independent segmentation and initial categorization. Lesions categorized as nodules larger than 3mm in diameter were further assessed for nine subjective characteristics: subtlety, internal structure, calcification, sphericity, margin, lobulation, spiculation, radiographic solidity, and malignancy[@McNittGray2007]. Each characteristic consists of either a discrete category set or an integer rating on a five-point scale. We use pylidc software [@Hancock2016] to access radiologist annotations. Analysis of nodule characteristic distribution revealed four characteristics with very low variability between nodules, which we decided to omit from further processing. Adopting the LUNA16 [@Setio2017] approach we also limited the analysis to nodules accepted by at least three out of four radiologists. This resulted in 1186 nodules, each with three or four sets of segmentation and five-dimensional rating characteristic (subtlety, sphericity, margin, lobulation, and malignancy). We normalize all rating values to $[0,1]$ range to aid the implementation. Methodology ----------- ### CBIR Using Embeddings. Our approach to CBIR is partially inspired by the natural language processing technique called word2vec[@Mikolov2013]. We learn a function $f:V\to S$ from a high dimensional space of CT voxels $V$ to a much lower dimensional space $S$, which we call “a semantic space". A desirable property of $S$ is to capture nodule similarity as perceived by radiology experts, that is, for some distance function $d:S^2\to\mathbb R$ we expect two vectors $s_1, s_2 \in S$ to be relatively close (have small $d\left(s_1, s_2\right)$) if the corresponding nodules would be considered similar by the doctors and vice versa. We call vectors in $S$ “embeddings". Not necessarily interpretable per se, we would like an embedding to incorporate both characteristical information as defined by radiologists as well as some visual information about the nodule. A traditional approach to learn embeddings is to define a proxy task, such that training a deep learning model to solve this task would produce the embeddings as a byproduct. ### Theoretical Architecture Constraints. According to PAC theory (e.g. see [@Mohri2018 ch. 2]) bound of the generalization error over hypothesis space $\mathcal H$ and number of training samples $N$ is: $$R(h) \le \hat R_S(h) + O\left(\sqrt{\frac{\log\left\|\mathcal H\right\|}{N}}\right)$$ Our dataset has a very limited number of nodules, while the deep learning network capable of extracting the information from the 3D space of voxels can have many millions of parameters, that is, $N\ll\log{\|\mathcal H\|}$. While some researchers question the tightness of such theoretical generalization error bounds, we still wary of the model overfitting in our setting. Therefore, we decide that our proxy task may not be sufficient to train a robust feature extraction. ### Feature Extraction. Fortunately, thanks to LUNA16[@Setio2017] and Kaggle Data Science Bowl (DSB) 2017[@dsb2017] competitions there are a lot of previous works tackling pulmonary nodule analysis. In this work we apply a transfer learning technique to the winning DSB solution[@Liao2019] which is based on the U-net architecture[@Ronneberger2015]. It has initially been trained on a dataset including lots of scans not found in LIDC-IDRI and optimized for different tasks, namely, pulmonary nodule detection and binary whole-scan classification. For our feature extraction, we reuse the pre-trained U-net backbone of the nodule detector and drop the region proposal head comprising of the last two convolutional layers. ### Rating Regression. Here we define a proxy task as a radiologist rating regression. We extend the feature extraction network with three fully connected layers with the output being a five-dimensional vector (). A target is defined as a mean of radiologist ratings applied to the nodule at hand. We use MSE loss, training only the head while keeping the backbone weights fixed. A ten-dimensional vector from the second-to-last layer is used as an embedding. ![ Regression network for embeddings learning (only feature map is shown). The trainable head comprises of the last three fully connected layers. See [@Liao2019] for a detailed explanation about the backbone. []{data-label="fig:regression"}](regression-architecture.png){height="0.3\textheight"} Results and Discussion ====================== Semantic Information Preservation --------------------------------- In this section we study whether the embeddings produced by our method preserve semantic meaning, that is, nodules with similar characteristics produce similar embeddings. ### t-SNE. We run t-SNE[@Maaten2008] dimensionality reduction over the random sample of the embeddings space (). For simplicity, the coloring here only reflects malignancy: samples with high malignancy (mean rating $>3$ out of 5) are colored red and others blue. The separation of red and blue is not ideal because embedding vectors preserve more information than just a malignancy level. However, we observe that the malignant nodules are clustered together, unlike the benign nodules. The more detailed analysis of malignant embeddings reveals that the distance from the center of mass of the red cluster roughly corresponds to the inverse malignancy rating of the nodule (not shown). ### Hierarchical Clustering. We also run a hierarchical clustering of a random subset of embeddings using Ward’s minimum variance method[@Ward1963] (). We analyze the top three splits and observe that the mean malignancy rating of nodules in the leftmost group (green) is much higher than in other groups, that is, the first split happens to cluster malignant vs benign nodules. Similarly, the second split isolates low-margin benign nodules (red) from other benign ones, and finally, the third split differentiates based on subtlety rating (among benign nodules with high margin score). [0.49]{} ![Semantic information preservation in embedding space[]{data-label="fig:seminfo"}](t-sne.png "fig:"){width="\textwidth"} [0.49]{} ![Semantic information preservation in embedding space[]{data-label="fig:seminfo"}](ward-clustering.png "fig:"){width="\textwidth"} Usability: Top-$k$ Evaluation ----------------------------- The purpose of our system is to aid a radiologist by presenting top-$k$ cases similar to the query nodule. A radiologist can then inspect each presented case, assessing morphological similarity and reviewing historical diagnosis or even broader clinical records of the patients with presumably similar cases. Aiming for interface simplicity and maximum user productivity we prefer a minimum $k$ which still provides enough information to support a doctor’s decision. We study the semantic space distance between the retrieved samples and a query and decide that beyond $k=4$ CBIR results provide diminishing returns. Qualitative Assessment: Example Output {#sec:examples} -------------------------------------- shows example results of our system. The query nodule is presented on the left side (gray border). Only a single CT slice is shown, with doctors segmentation marks superimposed. A 3D mesh of consensus segmentation is presented below the slice. Query ratings (before normalization) together with the mean are provided on the left. The Top-4 CBIR results are shown on the right (green border) together with their mean ratings. Notice the shape and rating similarity of the CBIR results to that of the query. A non-match nodule is displayed (red border) for comparison. We remind the reader that in contrast to visualization the CBIR query consists of a 3D CT patch only. ![CBIR evaluation examples. See text for the description.[]{data-label="fig:examples"}](example1.jpg){height="3.9cm"} [0.49]{} ![CBIR evaluation examples. See text for the description.[]{data-label="fig:examples"}](example2.jpg "fig:"){width="\textwidth"} [0.49]{} ![CBIR evaluation examples. See text for the description.[]{data-label="fig:examples"}](example3.jpg "fig:"){width="\textwidth"} Quantitative Assessment ----------------------- To quantify the CBIR performance we conduct two evaluations: First, we compare the CBIR-based algorithm to medical experts. Then we compare to a recent work in the field [@Loyman2019]. ### Comparison to Human Experts. We define a nodule rating consensus as a mean vector of ratings assigned to the same nodule by several radiologists. A dissent score of the specific radiologist ratings is defined as an RMSE between it and the consensus among other radiologists. Similarly, a dissent score of an algorithm is the RMSE between ratings predicted by the algorithm and the consensus of all radiologists (using normalized rating range $[0,1]$). We use a rating consensus of $k$ top CBIR results as a naïve algorithm prediction. We also provide, for comparison, a dissent score of a random algorithm that predicts one of the existing ratings randomly regardless of the input. shows a distribution of dissent scores, while reports their mean and standard deviation (measured over five-fold cross-validation). We can see that the naïve CBIR-based algorithm for $k=4$ improves on the doctors’ diagnosis on average as much as the latter improves on a random guess (1.5 times lower mean dissent score). That is, a distance between the query and CBIR results is at least comparable to the ranking uncertainty of the query itself, which is a strong indicator of the embedding space quality. ![ PDF and CDF of the dissent score of random predictions, doctor ratings, and naïve CBIR-based algorithm with $k\in \{1, 2, 4, 8\}$. For clarity, all graphs show MLE fit of the log-normal distribution. The lower dissent score is better. []{data-label="fig:alg_vs_doc"}](alg-vs-doc.png){width="\textwidth"} ---------- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ -------------------------------------------------------------------------------------------------------------------------- Method random 0.35 0.13 1.43 1.13 1.27 1.44 1.47 2.81 1.59 2.64 3.08 2.86 doctors 0.23 0.09 0.72 0.69 0.74 0.73 0.75 0.83 0.64 0.98 0.95 0.85 CBIR k=1 0.19 0.08 0.89 0.79 0.84 0.94 0.87 0.59 0.48 0.58 0.61 0.56 0.81\ CBIR k=2 & 0.17 & 0.07 & 0.76 & 0.70 & 0.74 & 0.81 & 0.73 & 0.49 & 0.41 & 0.48 & 0.52 & 0.47 & 0.80\ CBIR k=4 & 0.15 & 0.06 & 0.70 & 0.65 & 0.68 & 0.74 & 0.68 & 0.45 & 0.38 & 0.44 & 0.46 & 0.44 & 0.80\ CBIR k=8 & 0.14 & 0.06 & 0.66 & 0.63 & 0.64 & 0.70 & 0.64 & 0.43 & 0.37 & 0.42 & 0.44 & 0.41 & 0.79\ CBIR mean & & & & & & & & & & & & & 0.79\ Loyman et al. & & & 0.93 & 0.83 & 0.89 & 0.94 & 0.68 & 0.84 & 0.47 & 0.27 & 0.37 & 0.84 & 0.75\ ************************** ---------- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ -------------------------------------------------------------------------------------------------------------------------- : Quantitative comparison of methods: Random guess, human expert, presented CBIR solution (“CBIR"), and Loyman et al [@Loyman2019]. For the RMSE-based metrics, as defined here and in [@Loyman2019], lower is better. For Retrieval precision, higher is better. The best results are in bold. []{data-label="tab:perf"} ### Comparison to Recent Results. We extend with a comparison to [@Loyman2019]. Following the methodology defined in [@Loyman2019], we compute a prediction RMSE and STD over the dataset for each rating component separately (using rating range $[1,5]$) and compare to the rating regression results from [@Loyman2019]. We also compute a CBIR malignancy retrieval precision, that is: what portion of the retrieved $k$ results has a correct binary malignancy class (benign vs malignant). Since [@Loyman2019] only presents mean precision for $k\in\{1, 3, 5, 7, 9, 11, 13, 15\}$ we compute it as well for a meaningful comparison (“CBIR mean” in the Table). From the results shown in the Table, we can see that our CBIR-based rating prediction improves over [@Loyman2019] by 0.20 for RMSE and by 0.14 for STD on average. It also compares favorably to human experts. The retrieval precision is improved by 4%. Conclusion ========== In this work we prototype a CBIR system for pulmonary nodules. We develop a methodology to learn low-dimensional embeddings and present a theoretical justification for the architectural design selected. Our methodology facilitates learning of high-quality embeddings preserving both spatial and semantic information, all this despite very high data dimensionality and sample scarcity. We determine optimal usability settings and perform a qualitative analysis of CBIR results. Finally, we conduct a quantitative study demonstrating state-of-the-art results. While we believe that in reality, the diagnostic process is more complex than a 5-tuple rating can convey, we conclude that the CBIR output is highly relevant to the query, such that the benefit provided by the system to the radiologist end-user is comparable to obtaining a second radiologist’s opinion. ### Acknowledgments. Part of the work presented in this paper was done by the first two authors in the course of Y-Data program by Yandex School of Data Analysis. The authors would like to thank Kostya Kilimnik and Shlomo Kashani for the organization and support of the project initiation. [10]{} \[1\][`#1`]{} \[1\][https://doi.org/\#1]{} Armato, S.G., McLennan, G., Bidaut, L., McNitt-Gray, M.F., Meyer, C.R., Reeves, A.P., Zhao, B., Aberle, D.R., Henschke, C.I., Hoffman, E.A., Kazerooni, E.A., MacMahon, H., van Beek, E.J.R., Yankelevitz, D., Biancardi, A.M., Bland, P.H., Brown, M.S., Engelmann, R.M., Laderach, G.E., Max, D., Pais, R.C., Qing, D.P.Y., Roberts, R.Y., Smith, A.R., Starkey, A., Batra, P., Caligiuri, P., Farooqi, A., Gladish, G.W., Jude, C.M., Munden, R.F., Petkovska, I., Quint, L.E., Schwartz, L.H., Sundaram, B., Dodd, L.E., Fenimore, C., Gur, D., Petrick, N., Freymann, J., Kirby, J., Hughes, B., Casteele, A.V., Gupte, S., Sallam, M., Heath, M.D., Kuhn, M.H., Dharaiya, E., Burns, R., Fryd, D.S., Salganicoff, M., Anand, V., Shreter, U., Vastagh, S., Croft, B.Y., Clarke, L.P.: The lung image database consortium ([LIDC]{}) and image database resource initiative ([IDRI]{}): A completed reference database of lung nodules on [CT]{} scans. Medical Physics 38(2), 915–931 (Jan 2011). , McLennan, G., Bidaut, L., McNitt-Gray, M.F., Meyer, C.R., Reeves, A.P., Zhao, B., Aberle, D.R., Henschke, C.I., Hoffman, E.A., Kazerooni, E.A., MacMahon, H., Van Beek, E.J., Yankelevitz, D., Biancardi, A.M., Bland, P.H., Brown, M.S., Engelmann, R.M., Laderach, G.E., Max, D., Pais, R.C., Qing, D.P., Roberts, R.Y., Smith, A.R., Starkey, A., Batra, P., Caligiuri, P., Farooqi, A., Gladish, G.W., Jude, C.M., Munden, R.F., Petkovska, I., Quint, L.E., Schwartz, L.H., Sundaram, B., Dodd, L.E., Fenimore, C., Gur, D., Petrick, N., Freymann, J., Kirby, J., Hughes, B., Casteele, A.V., Gupte, S., Sallam, M., Heath, M.D., Kuhn, M.H., Dharaiya, E., Burns, R., Fryd, D.S., Salganicoff, M., Anand, V., Shreter, U., Vastagh, S., Croft, B.Y., Clarke, L.P.: Data from lidc-idri (2015). , <https://wiki.cancerimagingarchive.net/x/rgAe> Dhara, A.K., Mukhopadhyay, S., Dutta, A., Garg, M., Khandelwal, N.: Content-based image retrieval system for pulmonary nodules: Assisting radiologists in self-learning and diagnosis of lung cancer. Journal of Digital Imaging 30(1), 63–77 (Sep 2016). Hancock, M.C., Magnan, J.F.: Lung nodule malignancy classification using only radiologist-quantified image features as inputs to statistical learning algorithms: probing the lung image database consortium dataset with two statistical learning methods. Journal of Medical Imaging 3(4), 044504 (Dec 2016). Kaggle data science bowl 2017 (2017), <https://www.kaggle.com/c/data-science-bowl-2017>, accessed: Jan 2020 Lam, M.O., Disney, T., Raicu, D.S., Furst, J., Channin, D.S.: [BRISC]{}an open source pulmonary nodule image retrieval framework. Journal of Digital Imaging 20(S1), 63–71 (Aug 2007). Liao, F., Liang, M., Li, Z., Hu, X., Song, S.: Evaluate the malignancy of pulmonary nodules using the 3-d deep leaky noisy-[OR]{} network. [IEEE]{} Transactions on Neural Networks and Learning Systems 30(11), 3484–3495 (Nov 2019). Loyman, M., Greenspan, H.: Lung nodule retrieval using semantic similarity estimates. In: Hahn, H.K., Mori, K. (eds.) Medical Imaging 2019: Computer-Aided Diagnosis. [SPIE]{} (Mar 2019). Maaten, L.v.d., Hinton, G.: Visualizing data using t-sne. Journal of machine learning research 9(Nov), 2579–2605 (2008) McNitt-Gray, M.F., Armato, S.G., Meyer, C.R., Reeves, A.P., McLennan, G., Pais, R.C., Freymann, J., Brown, M.S., Engelmann, R.M., Bland, P.H., Laderach, G.E., Piker, C., Guo, J., Towfic, Z., Qing, D.P.Y., Yankelevitz, D.F., Aberle, D.R., van Beek, E.J., MacMahon, H., Kazerooni, E.A., Croft, B.Y., Clarke, L.P.: The lung image database consortium ([LIDC]{}) data collection process for nodule detection and annotation. Academic Radiology 14(12), 1464–1474 (Dec 2007). Mikolov, T., Chen, K., Corrado, G., Dean, J.: Efficient estimation of word representations in vector space (2013) Mohri, M., Rostamizadeh, A., Talwalkar, A.: Foundations of Machine Learning. The MIT Press, 2nd edn. (dec 2018) Pan, L., Qiang, Y., Yuan, J., Wu, L.: Rapid retrieval of lung nodule [CT]{} images based on hashing and pruning methods. [BioMed]{} Research International 2016, 1–10 (2016). Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, pp. 234–241. Springer International Publishing (2015). Setio, A.A.A., Traverso, A., de Bel, T., Berens, M.S., van den Bogaard, C., Cerello, P., Chen, H., Dou, Q., Fantacci, M.E., Geurts, B., van der Gugten, R., Heng, P.A., Jansen, B., de Kaste, M.M., Kotov, V., Lin, J.Y.H., Manders, J.T., S[ó]{}[ñ]{}ora-Mengana, A., Garc[í]{}a-Naranjo, J.C., Papavasileiou, E., Prokop, M., Saletta, M., Schaefer-Prokop, C.M., Scholten, E.T., Scholten, L., Snoeren, M.M., Torres, E.L., Vandemeulebroucke, J., Walasek, N., Zuidhof, G.C., van Ginneken, B., Jacobs, C.: Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: The [LUNA]{}16 challenge. Medical Image Analysis 42, 1–13 (Dec 2017). Siegel, R.L., Miller, K.D., Jemal, A.: Cancer statistics, 2020. [CA]{}: A Cancer Journal for Clinicians 70(1), 7–30 (Jan 2020). Ward, J.H.: Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association 58**(301), 236–244 (Mar 1963). Wei, G., Ma, H., Qian, W., Jiang, H., Zhao, X.: Content-based retrieval for lung nodule diagnosis using learned distance metric. In: 2017 39th Annual International Conference of the [IEEE]{} Engineering in Medicine and Biology Society ([EMBC]{}). [IEEE]{} (Jul 2017). [^1]: I. Kravets, T. Heletz – equal contribution	{ "pile_set_name": "ArXiv" }
--- abstract: \| For a fixed countably infinite structure $\Gamma$ with finite relational signature $\tau$, we study the following computational problem: input are quantifier-free $\tau$-formulas $\phi_0,\phi_1,\dots,\phi_n$ that define relations $R_0,R_1,\dots,R_n$ over $\Gamma$. The question is whether the relation $R_0$ is primitive positive definable from $R_1,\ldots,R_n$, i.e., definable by a first-order formula that uses only relation symbols for $R_1, \dots, R_n$, equality, conjunctions, and existential quantification (disjunction, negation, and universal quantification are forbidden). We show decidability of this problem for all structures $\Gamma$ that have a first-order definition in an ordered homogeneous structure $\Delta$ with a finite relational signature whose age is a Ramsey class and determined by finitely many forbidden substructures. Examples of structures $\Gamma$ with this property are the order of the rationals, the random graph, the homogeneous universal poset, the random tournament, all homogeneous universal $C$-relations, and many more. We also obtain decidability of the problem when we replace primitive positive definability by existential positive, or existential definability. Our proof makes use of universal algebraic and model theoretic concepts, Ramsey theory, and a recent characterization of Ramsey classes in topological dynamics. address: - \| Laboratoire d’Informatique (LIX), CNRS UMR 7161\ École Polytechnique\ 91128 Palaiseau\ France - \| Équipe de Logique Mathématique\ Université Diderot – Paris 7\ UFR de Mathématiques\ 75205 Paris Cedex 13, France - \| Équipe de Logique Mathématique\ Université Diderot – Paris 7\ UFR de Mathématiques\ 75205 Paris Cedex 13, France author: - Manuel Bodirsky - Michael Pinsker - Todor Tsankov bibliography: - 'local.bib' date: 'Version 5 – March 4, 2012' title: Decidability of definability --- Motivation and the Main Result {#sect:motivationresult} ============================== When studying a countably infinite relational structure $\Theta$, we often wish to know what $\Theta$ can express by its relations; for example, which other structures it interprets or defines. Concentrating on the latter, it would be pleasant to have an oracle which, given two structures $\Theta_1, \Theta_2$ on the same domain, tells us whether they define one another. If all structures we are interested in have finite signature, this is the same as having an oracle which, given a structure $\Theta$ and a relation $R$ on the same domain, tells us whether $R$ can be defined from $\Theta$. In this context, different notions of definability can be considered. The first notion that comes to mind is probably first-order definability: an $n$-ary relation $R$ is first-order definable in $\Theta$ iff there is a first-order formula $\phi(x_1,\ldots,x_n)$ over the language for $\Theta$ such that for all $n$-tuples $a$ of elements in $\Theta$ we have $a\in R$ iff $\phi(a)$ holds. In some applications, however, other notions of definability, in particular syntactic restrictions of first-order definability, are useful. We will be concerned here with primitive positive definability: a first-order formula is called primitive positive iff it is of the form $\exists y_1\ldots \exists y_m.\; \psi$, where $\psi$ is a conjunction of atomic formulas; and an $n$-ary relation $R$ is primitive positive definable over $\Theta$ iff it is first-order definable in $\Theta$ by means of a primitive positive formula $\phi(x_1,\ldots,x_n)$. Primitive positive definability is of importance in the study of the constraint satisfaction problem for $\Theta$, denoted by $\operatorname{CSP}(\Theta)$, in theoretical computer science. In such a problem, the input consists of a primitive positive sentence $\psi$ (that is, a primitive positive formula without free variables), and the question is whether $\psi$ is true in $\Theta$. Primitive positive definability of relations in $\Theta$ is important in the study of $\operatorname{CSP}(\Theta)$ because the CSP for an expansion of $\Theta$ by relations that are primitive positive definable in $\Theta$ can be reduced (in linear time) to $\operatorname{CSP}(\Theta)$. We will present here conditions under which the oracle which is to tell us whether a relation $R$ has a primitive positive definition from a finite language structure $\Theta$ can be a computer, i.e., under which the problem is decidable. In order to make the problem suitable for an algorithm, we need a finite representation of the input of the problem, that is, the relation $R$ and the structure $\Theta$. Our approach is to fix a base structure $\Gamma$ with finite relational language, and to assume that both $R$ and $\Theta$ have a quantifier-free definition in $\Gamma$. We then represent $R$ and $\Theta$ as quantifier-free formulas over $\Gamma$. Therefore, the input of our problem are quantifier-free formulas $\phi_0,\ldots,\phi_n$ in the language of $\Gamma$, of which $\phi_0$ defines the relation $R$, and $\phi_1,\ldots,\phi_n$ define the relations $R_1,\ldots,R_n$ of $\Theta$; the question is whether there is a primitive positive definition of $\phi_0$ that uses only relation symbols for $R_1,\dots,R_n$. We denote this computational problem by ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$. An algorithm for primitive positive definability has theoretical and practical consequences in the study of the computational complexity of CPSs. On the practical side, it turns out that hardness of $\operatorname{CSP}(\Theta)$ can usually be shown by presenting primitive positive definitions of relations for which it is known that the CSP is hard. Therefore, a procedure that decides primitive positive definability of a given relation is a useful tool to determine the computational complexity of CSPs. For the simplest of countable structures, namely the structure $(X;=)$ having no relations but equality, the decidability of ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ has been stated as an open problem in [@BodChenPinsker]. We will show here decidability of ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ for a large class of structures $\Gamma$ which we will now define. Let $\tau$ be a finite relational signature. The age of a $\tau$-structure $\Delta$ is the class of all finite $\tau$-structures that embed into $\Delta$. We say that a class $\mathcal C$ of finite $\tau$-structures, and similarly a structure with age $\mathcal C$, is - finitely bounded (in the terminology of [@MacphersonSurvey]) iff there exists a finite set of finite $\tau$-structures $\mathcal F$ such that for all finite $\tau$-structures $A$ we have $A \in \mathcal C$ iff no structure from $\mathcal F$ embeds into $A$; - Ramsey iff for all $k \geq 1$ and for all $H,P \in \mathcal C$ there exists $S\in\mathcal C$ such that $S \rightarrow (H)^P_k$, i.e., for all colorings of the copies of $P$ in $S$ with $k$ colors there exists a copy of $H$ in $S$ on which the coloring is constant (for background in Ramsey theory see [@GrahamRothschildSpencer]); - ordered iff the signature $\tau$ contains a binary relation that denotes a total order in every $A \in \mathcal C$. A structure is called homogeneous iff all isomorphisms between finite induced substructures[^1] extend to automorphisms of the whole structure. A structure $\Gamma$ is called a reduct of a structure $\Delta$ with the same domain iff all relations in $\Gamma$ are first-order definable in $\Delta$. We will prove the following. \[thm:main:pp\] Let $\Delta$ be a structure which is ordered, homogeneous, Ramsey, finitely bounded, and has a finite relational signature. Then for any reduct $\Gamma$ of $\Delta$ with finite relational signature the problem ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ is decidable. We remark that for finite structures $\Gamma$ the problem ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ is in co-NEXPTIME (and in particular decidable). For the variant where the finite structure $\Gamma$ is part of the input, the problem has recently shown to be also co-NEXPTIME-hard [@Willard-cp10]. Note that since $\Delta$ is homogeneous, it has quantifier elimination, i.e., every relation which is first-order definable in $\Delta$ can be defined by a quantifier-free formula. Hence, choosing $\Gamma=\Delta$, we see that our requirement for the relations in ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ to be given by quantifier-free formulas does not restrict the range of relations under consideration. Examples of structures $\Delta$ that satisfy the assumptions of Theorem \[thm:main:pp\] are $({\mathbb Q};<)$, the Fraïssé limit of ordered finite graphs (or tournaments [@RamseyClasses]), the Fraïssé limit of finite partial orders with a linear extension [@RamseyClasses], and the homogeneous universal ‘naturally ordered’ $C$-relations. (For definition and basic properties of $C$-relations, see [@MR1388893], in particular Theorem 14.7. The fact that the homogeneous universal naturally ordered $C$-relations have the Ramsey property follows from Theorem 4.3 in [@Mil79]; an explicit and elementary verification of the Ramsey property for the binary branching case can be found in [@BodirskyPiguet].) CSPs of reducts of such structures are abundant in particular for qualitative reasoning calculi in Artificial Intelligence. For instance, our result shows that it is decidable whether a given relation from Allen’s Interval Algebra [@Allen; @RandomReducts] is primitive positive definable in a given fragment of Allen’s Interval Algebra. As mentioned above, for $\Gamma=(X; =)$, the decidability of ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ has been posed as an open problem in [@BodChenPinsker]. Our results solve this problem, since $(X;=)$ is definable in $\Delta:=(\mathbb Q;<)$, which is ordered, homogeneous, Ramsey, and finitely bounded: the Ramsey property for this structure follows from the classical Ramsey theorem, and the other properties are easily verified. Using similar methods, decidability of the analogous problem for other syntactic restrictions of first-order logic can be shown in the same context. A formula is called existential iff it is of the form $\exists y_1\ldots \exists y_m.\; \psi$, where $\psi$ is quantifier-free. It is called existential positive iff it is existential and does not contain any negations. For a $\tau$-structure $\Gamma$, we denote by ${\operatorname{Expr}_{\operatorname{ex}}}(\Gamma)$ (${\operatorname{Expr}_{\operatorname{ep}}}(\Gamma)$) the problem of deciding whether a given quantifier-free $\tau$-formula $\phi_0$ has an existential (existential positive) definition over the structure with the relations defined by given quantifier-free $\tau$-formulas $\phi_1,\ldots,\phi_n$ in $\Gamma$. \[thm:main:exep\] Let $\Delta$ be a structure which is ordered, homogeneous, Ramsey, finitely bounded, and has a finite relational signature. Then for any reduct $\Gamma$ of $\Delta$ with finite relational signature the problems ${\operatorname{Expr}_{\operatorname{ex}}}(\Gamma)$ and ${\operatorname{Expr}_{\operatorname{ep}}}(\Gamma)$ are decidable. The assumptions on $\Delta$ in our theorems fall into two classes: the conditions of being ordered, homogeneous, Ramsey, and having finite relational signature imposed on $\Delta$ generally allow for a relatively good understanding (in a non-algorithmic sense) of the reducts of $\Delta$. The recent survey paper [@BP-reductsRamsey] summarizes what we know about reducts of such structures – their exciting feature is that many branches of mathematics, including model theory, combinatorics, universal algebra, and even topological dynamics are employed in their study, and indirectly also in our algorithm. The additional condition of being finitely bounded is needed to represent $\Delta$ algorithmically. This paper is organized as follows. In Section \[sect:henson\] we show that the assumption of $\Delta$ being finitely bounded is necessary for our decidability result. We then turn to the proof of Theorems \[thm:main:pp\] and \[thm:main:exep\]: in Section \[sect:preservation\] we cite preservation theorems of the form “$R$ is definable from $\Theta$ (in some syntactically restricted form of first-order logic) if and only if certain functions on the domain of $\Theta$ (which functions depends on the syntactic restriction) preserve $R$”. Section \[sect:canonical\] is devoted to the use of Ramsey theory in order to standardize functions that do not preserve $R$ – if such functions exist. Our decision procedure, presented in Section \[sect:algorithm\], then uses this standardization of functions and the preservation theorems to check whether or not $R$ is definable from $\Theta$. The paper ends with two sections containing further discussion and open problems. Undecidability of Definability {#sect:henson} ============================== This section demonstrates that the assumption in Theorem \[thm:main:exep\] of $\Delta$ being finitely bounded is necessary. We use a class of homogeneous digraphs introduced by Henson [@Henson]. A tournament is a directed graph without self-loops such that for all pairs $x,y$ of distinct vertices exactly one of the pairs $(x,y)$, $(y,x)$ is an arc in the graph. For a set of finite directed graphs $\mathcal N$, we write $\text{Forb}(\mathcal N)$ for the class of all finite directed graphs that do not embed one of the structures from $\mathcal N$. For all sets $\mathcal N$ of finite tournaments there exists a countably infinite homogeneous directed graph $\Gamma$ with age $\text{Forb}(\mathcal N)$ (this can be shown by amalgamation, see [@Hodges]). Moreover, those properties characterize $\Gamma$ up to isomorphism. Henson specified an infinite set $\mathcal T$ of finite tournaments $\Lambda_1,\Lambda_2,\dots$ with the property that $\Lambda_i$ does not embed into $\Lambda_j$ if $i \neq j$; the exact definition of this set is not important in what follows. But note that for two distinct subsets ${\mathcal N}_1$ and ${\mathcal N}_2$ of $\mathcal T$ the two sets $\text{Forb}({\mathcal N}_1)$ and $\text{Forb}({\mathcal N}_2)$ are distinct as well, and so are the respective homogeneous digraphs with age $\text{Forb}({\mathcal N}_1)$ and $\text{Forb}({\mathcal N}_2)$. Since there are $2^\omega$ many subsets of the infinite set $\mathcal T$, there are also that many distinct homogeneous directed graphs; they are often referred to as Henson digraphs. There exists a ordered directed graph $\Delta$ which is homogeneous and Ramsey such that ${\operatorname{Expr}_{\operatorname{pp}}}(\Delta)$ and ${\operatorname{Expr}_{\operatorname{ep}}}(\Delta)$ are undecidable. For any Henson digraph $\Gamma$, the class ${\mathcal C}$ of all expansions of the structures in the age of $\Gamma$ by a linear order is a Ramsey class; this can been shown by the partite method [@NesetrilRoedlPartite]. Moreover, there exists a homogeneous ordered digraph $\Delta$ with age ${\mathcal C}$ (again by amalgamation, see [@Hodges]), and $\Gamma$ is a reduct of $\Delta$. We show that non-isomorphic Henson digraphs $\Gamma_1$ and $\Gamma_2$ have distinct ${\operatorname{Expr}_{\operatorname{pp}}}$ problems. In the following, let $E$ denote a binary relation symbol that we use to denote the edge relation in graphs. In fact, we show the existence of a first-order formula $\phi_1$ over digraphs such that the input $\phi_0 := E(x,y)$ and $\phi_1$ is a yes-instance of ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma_1)$ and a no-instance of ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma_2)$, or vice-versa. Since there are uncountably many Henson digraphs, but only countably many algorithms, this clearly shows the existence of Henson digraphs $\Gamma$ such that ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ is undecidable. This finishes the proof since $\Gamma$ is a reduct of an ordered homogeneous Ramsey structure $\Delta$, as we have seen above, and ${\operatorname{Expr}_{\operatorname{pp}}}(\Delta)$ must be undecidable as well. The same argument shows undecidability of ${\operatorname{Expr}_{\operatorname{ep}}}(\Delta)$. By the definition of $\Gamma_1$ and $\Gamma_2$, there exists a finite digraph $\Omega$ which embeds into $\Gamma_1$ but not into $\Gamma_2$, or that embeds into $\Gamma_2$ but not into $\Gamma_1$. Assume without loss of generality the former. Let $s$ be the number of elements of $\Omega$, and denote its elements by $a_1,\dots,a_s$. Let $\psi$ be the formula with variables $x_1,\dots,x_s$ that has for distinct $i,j \leq s$ a conjunct $E(x_i,x_j)$ if $E(a_i,a_j)$ holds in $\Omega$, and a conjunct $\neg E(x_i,x_j) \wedge x_i \neq x_j$ otherwise. Let $\phi_1$ be the formula $\psi \wedge E(x_{s+1},x_{s+2})$. Let $D_1$ be the domain of $\Gamma_1$, and consider the relation $R_1 \subseteq (D_1)^{s+2}$ defined by $\phi_1$ in $\Gamma_1$. Let $R$ be a relational symbol of arity $s+2$. Let $\Theta$ be the structure with signature $\{R\}$, domain $D_1$, and where $R$ denotes the relation $R_1$. It is clear that $\exists x_1,\dots,x_s. \, R(x_1,\dots,x_s,x,y)$ is a primitive positive definition of $E(x,y)$ in $\Theta$. Now consider the relation $R_2$ defined by $\phi_1$ in $\Gamma_2$ over the domain $D_2$ of $\Gamma_2$. Since $\Omega$ does not embed into $\Gamma_2$, the precondition of $\phi_1$ is never satisfied, and the relation $R_2$ is empty. Hence, the relation $E(x,y)$ is certainly not first-order (and in particular not primitive positive) definable in $(D_2;R_2)$. Preservation Theorems {#sect:preservation} ===================== Let $\Gamma$ be a reduct of a homogeneous finitely bounded Ramsey structure $\Delta$ with finite relational signature. Our algorithm for ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ is based on the fact that if $R$ is not definable from $\Theta$, then there exists a certain kind of function which violates $R$; in order to decide whether or not $R$ is definable, the algorithm thus searches for such a function. In this section, we shall formulate this fact in more detail. A structure is called $\omega$-categorical iff its first-order theory has exactly one countable model up to isomorphism. For an $n$-tuple $a$ of elements of a structure $\Delta$, the type of $a$ is the set of all first-order formulas with $n$ free variables $x_1,\ldots,x_n$ that are satisfied by $a$. By a theorem of Ryll-Nardzewski (see for example the textbook [@Hodges]), a structure is $\omega$-categorical iff it has only finitely many different types of $n$-tuples (called $n$-types), for each $n\geq 1$. From this characterization it is straightforward to see that structures which are homogeneous and have a finite relational signature are $\omega$-categorical; in particular, this is true for the structure $\Delta$ of Theorems \[thm:main:pp\] and \[thm:main:exep\]. For an $n$-tuple $a$ of elements of a structure $\Delta$, the orbit of $a$ is the set $\{\alpha(a):\alpha\in\operatorname{Aut}(\Delta)\}$, where $\operatorname{Aut}(\Delta)$ denotes the automorphism group of $\Delta$. It is well-known that a structure is $\omega$-categorical iff it has for every $n\geq 1$ only finitely many orbits of $n$-tuples (called $n$-orbits). Moreover, in $\omega$-categorical structures two $n$-tuples have the same type iff they have the same orbit (see again [@Hodges]). In particular, every $n$-ary relation definable over an $\omega$-categorical structure is a finite union of orbits of $n$-tuples. Clearly, when $\Theta$ is a reduct of a structure $\Delta$, then $\operatorname{Aut}(\Theta)\supseteq\operatorname{Aut}(\Delta)$. Hence, if $\Delta$ is $\omega$-categorical, then so is $\Theta$; therefore, all structures that appear in this paper are $\omega$-categorical. If $R$ is an $m$-ary relation on a set $D$, and $f \colon D^n{\rightarrow}D$ is a finitary operation on $D$, then we say that $f$ preserves $R$ iff $f(r_1,\ldots,r_n)$ (calculated componentwise) is in $R$ for all $m$-tuples $r_1,\ldots,r_n\in R$. In other words, when $r_i = (r_i^1,\dots,r_i^m) \in R$ for all $i \leq n$, we require that $(f(r_1^1,\dots,r_n^1),\dots,f(r_1^m,\dots,r_n^m)) \in R$. Otherwise, we say that $f$ violates $R$. Observe that a permutation $\alpha$ acting on the domain of a structure $\Theta$ is an automorphism iff both $\alpha$ and its inverse preserve all relations of $\Theta$. An endomorphism of a structure $\Theta$ with domain $D$ is a unary operation $f \colon D{\rightarrow}D$ which preserves all relations of $\Theta$. A self-embedding of $\Theta$ is an injective unary operation $f \colon D{\rightarrow}D$ which preserves all relations of $\Theta$ and all complements of relations in $\Theta$. A polymorphism of $\Theta$ is a finitary operation $f \colon D^n{\rightarrow}D$ which preserves all relations of $\Theta$. We can now state the preservation theorem used by our algorithm. Statement (1) is well-known in model theory and follows from the standard proof of the theorem of Ryll-Nardzewski. Items (2) and (3) are consequences of the Theorem of [Ł]{}os–Tarski and the Homomorphism Preservation Theorem; for these theorems, see [@Hodges], for the (straightforward) proofs of statements (2) and (3) see [@RandomMinOps]. Item (4) is due to Bodirsky and Nešetřil [@BodirskyNesetrilJLC]. \[thm:preservation\] Let $\Theta$ be an $\omega$-categorical structure, and let $R$ be a relation on its domain. - $R$ has a first-order definition in $\Theta$ iff $R$ is preserved by all automorphisms of $\Theta$. - $R$ has an existential definition in $\Theta$ iff $R$ is preserved by all self-embeddings of $\Theta$. - $R$ has an existential positive definition in $\Theta$ iff $R$ is preserved by all endomorphisms of $\Theta$. - $R$ has an primitive positive definition in $\Theta$ iff $R$ is preserved by all polymorphisms of $\Theta$. Standardizing Functions {#sect:canonical} ======================= Theorem \[thm:preservation\] tells us that if a relation $R$ is not definable in an $\omega$-categorical structure $\Theta$, then this is witnessed by a some finitary function on the domain of $\Theta$; the kind of function depends on the notion of definability. In this section, we show that in the context of Theorems \[thm:main:pp\] and \[thm:main:exep\], this is even witnessed by a function which shows a certain regular behavior, making the search for such an (infinite!) function accessible to algorithms. We start by defining what we mean by regular behavior. Canonicity ---------- For a structure $\Delta$ and $n\geq 1$, we write $S_n^\Delta$ for the set of all $n$-types in $\Delta$. The cardinality of $S_n^\Delta$ is denoted by $o^\Delta(n)$. We write $S^\Delta:=\bigcup_{n\geq 1} S_n^\Delta$. For an $n$-tuple $a\in\Delta$, we write $\operatorname{tp}^\Delta(a)$ for the element of $S_n^\Delta$ corresponding to $a$. We drop the reference to the structure in this notation when the structure is clear from the context. A type condition between two structures $\Xi$ and $\Omega$ is a pair $(s,t)$, where $s\in S_n^{\Xi}$ and $t\in S_n^{\Omega}$ for the same $n\geq 1$. A function $f \colon \Xi{\rightarrow}\Omega$ satisfies a type condition $(s,t)$ iff for all $n$-tuples $a=(a_1,\ldots,a_n)$ in $\Xi$ of type $s$, the $n$-tuple $f(a)=(f(a_1),\ldots,f(a_n))$ in $\Omega$ is of type $t$. A behavior is a set of type conditions between two structures. A function has behavior $B$ iff it satisfies all the type conditions of the behavior $B$. For $n\geq 1$, a behavior $B$ is called $n$-complete iff for all types $s\in S^{\Xi}_n$ there is a type $t\in S^{\Omega}_n$ such that $(s,t)\in B$. It is called complete iff it is $n$-complete for all $n\geq 1$. A function $f \colon \Xi {\rightarrow}\Omega$ is canonical ($n$-canonical) iff it has a complete ($n$-complete) behavior. For $F\subseteq \Xi$ we say that $f$ satisfies a type condition $(s,t)$ on $F$ iff for all $n$-tuples $a=(a_1,\ldots,a_n)$ in $F$ of type $s$ (in $\Xi$, not in the substructure induced by $F$), the $n$-tuple $f(a)=(f(a_1),\ldots,f(a_n))$ in $\Omega$ is of type $t$. The notions of having a behavior on $F$ and of being canonical on $F$ are then defined naturally. Observe that a complete behavior is just a function from $S^{\Xi}$ to $S^{\Omega}$ which respects the sorts, i.e., $n$-types are sent to $n$-types. We remark that not every such function is necessarily the behavior of a canonical function from $\Xi$ to $\Omega$, but every canonical function from $\Xi$ to $\Omega$ does define a function from $S^{\Xi}$ to $S^{\Omega}$. A behavior is just a partial function from $S^{\Xi}$ to $S^{\Omega}$ respecting the sorts. For a relational structure $\Delta$, we write $n(\Delta)$ for the supremum of the arities of the relations of $\Delta$. Suppose that $n(\Xi)$ is finite and that $\Xi$ has quantifier elimination, i.e., every first-order formula in the language of $\Xi$ is equivalent to a quantifier-free formula over $\Xi$; this is in particular the case for the structure $\Delta$ of Theorems \[thm:main:pp\] and \[thm:main:exep\], since homogeneity implies quantifier elimination. Then the type of any tuple in $\Xi$ is determined by the types of its subtuples of length ${n(\Xi)}$. If moreover the same condition holds for $\Omega$ (in particular, if $\Omega=\Xi$), and we set $n$ to be the maximum of $n(\Xi)$ and $n(\Omega$), then a total function from $S^{\Xi}_{n}$ to $S^{\Omega}_{n}$ automatically defines a total function from $S^{\Xi}_{}$ to $S^{\Omega}_{}$. In other words, a function $f \colon \Xi{\rightarrow}\Omega$ is canonical iff it is $n$-canonical. Note also that $S^{\Xi}_k$ is finite for every $k\geq 1$ since $\Xi$ is $\omega$-categorical (this follows if $\Xi$ has quantifier elimination and finite relational signature, cf. [@Hodges]). Therefore, canonical functions can be represented by finite objects, namely by functions from $S^{\Xi}_{n}$ to $S^{\Omega}_{n}$. Since $\Omega$ is $\omega$-categorical as well, there are only finitely many functions from $S^{\Xi}_{n}$ to $S^{\Omega}_{n}$, and hence there exist only finitely many complete behaviors between $\Xi$ and $\Omega$, allowing to check all of them in an algorithm. Roughly, our goal in the following is to prove that functions witnessing that a relation $R$ is not definable in $\Theta$ can be assumed to be canonical; it will turn out that this is almost true. Calling Ramsey -------------- \[lem:canonicalOnArbitrarilyLargeFinite\] Let $\Xi$ be ordered Ramsey, let $\Omega$ be $\omega$-categorical, and let $f \colon \Xi{\rightarrow}\Omega$ be a function. Then for all finite substructures $F\subseteq \Xi$ there is a copy of $F$ in $\Xi$ on which $f$ is canonical. Set $n:=n(\Xi)$, and let $m:=o^{\Omega}(n)$. Now $f$ defines a coloring of the $n$-tuples in $\Xi$ by $m$ colors: the color of a tuple $a$ is just the type of $f(a)$ in $\Omega$. Note that if $P$, $S$ are ordered structures, then coloring copies of $P$ in $S$ is the same as coloring tuples of type $\operatorname{tp}(p)$, where $p$ is any tuple which enumerates $P$ – this is because every copy of $P$ in $S$ contains precisely one tuple of type $\operatorname{tp}(p)$, and every tuple of type $\operatorname{tp}(p)$ in $S$ induces precisely one copy of $P$ in $S$. Given any finite substructure $F$ of $\Xi$, enumerate all types of $n$-tuples that occur in $F$ by $t_1,\ldots,t_k$. There is a substructure $S_1$ of $\Xi$ such that whenever all tuples of type $t_1$ in $S_1$ are colored with $m$ colors, then there exists a substructure $H_1$ of $S_1$ isomorphic to $F$ on which the coloring is constant. Further, there is a substructure $S_2$ of $\Xi$ such that whenever all tuples of type $t_2$ in $S_2$ are colored with $m$ colors, then there exists a substructure $H_2$ of $S_2$ isomorphic to $S_1$ on which the coloring is constant. We iterate this $k$ times, arriving at a structure $S_k$. Now going back the argument, we find that $S_k$ contains a copy of $F$ on which all colorings are constant. That means that $f$ is canonical on this copy. We remark that this lemma would be false if one dropped the order assumption. We will now use Lemma \[lem:canonicalOnArbitrarilyLargeFinite\] in order to show that for ordered homogeneous Ramsey structures $\Delta$ with finite relational signature, arbitrary functions from $\Delta$ to $\Delta$ generate canonical functions from $\Delta$ to $\Delta$. To introduce this notion, we make the following observation. The set $\operatorname{End}(\Delta)$ of endomorphisms of a structure $\Delta$ forms a transformation monoid, i.e., it is closed under composition $f\circ g$ and contains the identity function $\operatorname{id}$. Moreover, it is closed (also called locally closed or local) in the topological sense, i.e., it is a closed subset of the space $D^D$, where $D$ is the domain of $\Delta$ equipped with the discrete topology. This implies that if a set ${\mathcal F}$ of functions from $D$ to $D$ preserves a set of given relations, then so does the smallest closed monoid containing ${\mathcal F}$. This motivates the following definition. \[defn:generatesUnary\] Let $D$ be a set, $g: D{\rightarrow}D$, and let ${\mathcal F}$ be a set of functions from $D$ to $D$. We say that ${\mathcal F}$ generates $g$ iff $g$ is contained in the smallest closed monoid containing ${\mathcal F}$. For a structure $\Delta$ with domain $D$ and a function $f \colon D{\rightarrow}D$, we say that $f$ generates $g$ over $\Delta$ iff $\{f\}\cup\operatorname{Aut}(\Delta)$ generates $g$. Equivalently, for every finite subset $F$ of $\Delta$, there exists a term $\alpha_0\circ (f\circ\alpha_1\circ\cdots\circ f\circ \alpha_n)$, where $n\geq 0$ and $\alpha_i\in\operatorname{Aut}(\Delta)$ for $0\leq i\leq n$, which agrees with $g$ on $F$. Note that every operation $f \colon D {\rightarrow}D$ generates an operation $g$ over $\Delta$ that is canonical as a function from $\Delta$ to $\Delta$, namely the identity operation. What we really want is that $f$ generates over $\Delta$ a canonical function $g$ which represents $f$ in a certain sense – it should be possible to retain specific properties of $f$ when passing to the canonical function. For example, when $f$ violates a given relation $R$, then we would like to have a canonical $g$ which also violates $R$ – this is clearly not the case for the identity function. Unfortunately, $f$ might be such that it violates a relation $R$ without generating any function that is canonical as a function from $\Delta$ to $\Delta$ and that violates $R$. We therefore have to refine our method: we would like to fix constants $c_1,\ldots,c_n\in\Delta$ which witness that $f$ violates $R$ and then have canonical behavior relative to these constants, i.e., on the structure $(\Delta,c_1,\ldots,c_n)$ which is $\Delta$ enriched by the constants $c_1,\ldots,c_n$. In order to do this, we must assure that $(\Delta,c_1,\ldots,c_n)$ still has the Ramsey property. This leads us into topological dynamics. An escapade in topological dynamics ----------------------------------- The goal of this subsection is to show the following proposition by using a recent characterization of the Ramsey property in topological dynamics. \[prop:addingConstantsPreservesRamsey\] Let $\Delta$ be ordered homogeneous Ramsey, and let $c_1,\ldots,c_n\in \Delta$. Then $(\Delta,c_1,\ldots,c_n)$ is ordered homogeneous Ramsey as well. We remark that it is easy to see that the expansion of any homogeneous structure by finitely many constants is again homogeneous, and that the nontrivial part of the proposition concerns the Ramsey property. We do not know if the same proposition holds if one does not assume $\Delta$ to be ordered. To prove the proposition, we use a theorem from [@Topo-Dynamics]. A topological group is a group $(G;\cdot)$ together with a topology on $G$ such that $(x,y) \mapsto xy^{-1}$ is continuous from $G^2$ to $G$. A group action of $G$ on a topological space $X$ is continuous iff it is continuous as a function from $G \times X$ into $X$. A topological group is extremely amenable iff any continuous action of the group on a compact Hausdorff space has a fixed point. \[thm:KPT\] An ordered homogeneous structure is Ramsey iff its automorphism group is extremely amenable. Thus the automorphism group of the structure $\Delta$ in Proposition \[prop:addingConstantsPreservesRamsey\] is extremely amenable. Note that the automorphism group of $(\Delta,c_1,\ldots,c_n)$ is an open subgroup of $\operatorname{Aut}(\Delta)$. The proposition thus follows from the following fact. Let $G$ be an extremely amenable group, and let $H$ be an open subgroup of $G$. Then $H$ is extremely amenable. Let $H$ act continuously on a compact space $X$; we will show that this action has a fixed point. Denote by $H \backslash G$ the set of right cosets of $H$ in $G$, i.e. $H \backslash G = \{Hg : g \in G\}$. Denote by $\pi \colon G \to H \backslash G$ the quotient map and let $s \colon H \backslash G \to G$ be a section for $\pi$ (i.e., a mapping satisfying $\pi \circ s = \mathrm{id}$) such that $s(H) = 1$. Let $\alpha$ be the map from $H \backslash G \times G \to H$ defined by $$\alpha(w, g) = s(w) g s(wg)^{-1} \; .$$ For $w \in H \backslash G$ and $g \in G$, note that $s(w)g$ and $s(wg)$ lie in the same right coset of $H$, namely $wg$, and hence the image of $\alpha$ is $H$. The map $\alpha$ satisfies[^2] $$\begin{aligned} \alpha(w,g_1g_2) & = s(w)g_1 g_2 (s(w g_1 g_2))^{-1} \\ & = s(w) g_1 s(w g_1) s(w g_1)^{-1} g_2 (s(w g_1 g_2))^{-1} \\ & = \alpha(w,g_1) \alpha(wg_1,g_2) \; .\end{aligned}$$ As $H$ is open, $H \backslash G$ is discrete. Hence, $s$ is continuous, and therefore $\alpha$ is continuous as a composition of continuous maps. The co-induced action $G \curvearrowright X^{H \backslash G}$ of $G$ on the product space $X^{H \backslash G}$ is defined by $$(g \cdot \xi)(w) = \alpha(w, g) \cdot \xi(wg).$$ To check that this action is continuous, it suffices to see that the map $(g, \xi) \mapsto (g \cdot \xi)(w)$ is continuous $G \times X^{H \backslash G} \to X$ for every fixed $w \in H \backslash G$. We already know that $\alpha$ is continuous and that the action $H \curvearrowright X$ is continuous. To see that $(g, \xi) \mapsto \xi (wg)$ is continuous, suppose that $(g_n, \xi_n) \to (g, \xi)$. Let $w=Hk$. As $g_n \to g$ and $k^{-1}Hk$ is open, we will have that eventually $g_ng^{-1} \in k^{-1}Hk$, giving that $kg_n (kg)^{-1} \in H$, or, which is the same, $Hkg_n = Hkg$. We obtain that for sufficiently large $n$, $wg_n = wg$. Therefore $\xi_n(w g_n) \to \xi(wg)$. By the extreme amenability of $G$, this action has a fixed point $ \xi_0$. Now we check that $\xi_0(H) \in X$ is a fixed point of the action $H \curvearrowright X$. Indeed, for any $h \in H$, $h \cdot \xi_0 = \xi_0$ and we have $$\xi_0(H) = (h \cdot \xi_0)(H) = \alpha(H, h) \cdot \xi_0(Hh) = h \cdot \xi_0(H),$$ finishing the proof. Minimal unary functions ----------------------- Using Proposition \[prop:addingConstantsPreservesRamsey\], we can now prove a ‘canonisation lemma’ that will be central in what follows. \[lem:generatesCanonicalWithConstants\] Let $\Delta$ be ordered homogeneous Ramsey with finite relational signature, $f \colon \Delta {\rightarrow}\Delta$, and let $c_1,\ldots,c_n\in \Delta$. Then $f$ generates over $\Delta$ a function which agrees with $f$ on $\{c_1,\ldots,c_n\}$ and which is canonical as a function from $(\Delta,c_1,\ldots,c_n)$ to $\Delta$. Let $(F_i)_{i\in \omega}$ be an increasing sequence of finite substructures of $(\Delta,c_1,\ldots,c_n)$ such that $\bigcup_{i\in\omega} F_i=(\Delta,c_1,\ldots,c_n)$. By Lemma \[lem:canonicalOnArbitrarilyLargeFinite\], for each $i\in\omega$ we find a copy $F_i'$ of $F_i$ in $(\Delta,c_1,\ldots,c_n)$ on which $f$ is canonical. By the homogeneity of $(\Delta,c_1,\ldots,c_n)$, there exist automorphisms $\alpha_i$ of $(\Delta,c_1,\ldots,c_n)$ sending $F_i$ to $F_i'$, for all $i\in\omega$. Since there are only finitely type conditions for $n((\Delta,c_1,\ldots,c_n))$-tuples, we may assume that if $f$ satisfies a type condition on $F_i'$, then it satisfies the same type condition on $F_{i+1}$. Then we can inductively pick automorphisms $\beta_i$ of $(\Delta,c_1,\ldots,c_n)$ such that $\beta_{i+1}\circ f\circ\alpha_{i+1}$ agrees with $\beta_i\circ f\circ\alpha_i$ on $F_i$, for all $i\in\omega$. The union over the functions $\beta_i\circ f\circ \alpha_i \colon F_i{\rightarrow}\Delta$ is a canonical function from $(\Delta,c_1,\ldots,c_n)$ to $\Delta$. The set of all closed transformation monoids on a fixed domain $D$ forms a complete lattice with respect to inclusion; it is the lattice of all endomorphism monoids of structures with domain $D$. Lemma \[lem:generatesCanonicalWithConstants\] has the following interesting consequence for this lattice. Let ${\mathcal N}, {\mathcal M}$ be closed monoids over the same domain. We say that ${\mathcal N}$ is minimal above ${\mathcal M}$ iff ${\mathcal M}\subsetneq {\mathcal N}$ and ${\mathcal M}\subsetneq {\mathcal R}\subseteq {\mathcal N}$ implies ${\mathcal R}={\mathcal N}$ for all closed monoids ${\mathcal R}$. Clearly, every minimal monoid above ${\mathcal M}$ is generated by a single function together with ${\mathcal M}$; such functions are called minimal as well (cf. [@RandomMinOps]). \[lem:minimalMonoidGeneratedByCanonical\] Let $\Theta$ be a structure with a finite relational signature which is a reduct of an ordered homogeneous Ramsey structure $\Delta$ in a finite relational signature, and let ${\mathcal N}$ be a minimal closed monoid above $\operatorname{End}(\Theta)$. Then there exist constants $c_1,\ldots,c_{n(\Theta)}\in \Delta$ and a function $f$ which is canonical as a function from $(\Delta,c_1,\ldots,c_{n(\Theta)})$ to $\Delta$ such that ${\mathcal N}$ is generated by $\operatorname{End}(\Theta)$ and $f$. Pick any $g\in {\mathcal N}\setminus\operatorname{End}(\Theta)$. Since $g{\notin}\operatorname{End}(\Theta)$, there exist a relation $R$ of $\Theta$ and a tuple $c:=(c_1,\ldots,c_{n(\Theta)})$ such that $R$ is violated on this tuple. By Lemma \[lem:generatesCanonicalWithConstants\], $g$ generates a function $f$ over $\Delta$ which is canonical as a function from $(\Delta,c_1,\ldots,c_{n(\Theta)})$ to $\Delta$ and which is identical with $g$ on $\{c_1,\ldots,c_{n(\Theta)}\}$. Then $f$ and $\operatorname{End}(\Theta)$ generate ${\mathcal N}$. \[prop:finiteMinimalReducts\] Let $\Theta$ be a finite relational signature reduct of an ordered homogeneous finite relational signature Ramsey structure $\Delta$. Then there are finitely many minimal closed monoids above $\operatorname{End}(\Theta)$, and every closed monoid containing $\operatorname{End}(\Theta)$ contains a minimal one. Observe that if $c,d$ are tuples of the same type in $\Delta$, and $f, g$ are canonical functions from $(\Delta,c)$ and $(\Delta,d)$ to $\Delta$, respectively, and their (complete) behaviors are identical, then $f$ and $g$ generate one another over $\Delta$. Thus, there are only finitely many inequivalent (in the sense of ‘do not generate one another’) functions generating minimal monoids. The upper bound for minimal monoids is the following: set $j:=o^\Delta(n(\Theta))$ (there are that many inequivalent choices for the tuple of constants of length $n(\Theta)$ in $\Delta$). For every type of an $n(\Theta)$-tuple $c$ in $\Delta$, set $r_c:=o^{(\Delta,c)}(n(\Delta))$. Set $r$ to be the maximum of the $r_c$. Define moreover $s:=o^{\Delta}(n(\Delta))$. Then a bound for the number of inequivalent minimal functions over $\operatorname{End}(\Theta)$ is $j\cdot s^r$. Minimal higher arity functions ------------------------------ Since primitive positive definability is characterized by finitary functions rather than unary functions (recall Theorem \[thm:preservation\]), we have to generalize our method to higher arities. Let $\Xi_1,\ldots,\Xi_m$ be a structures. For a tuple $x$ in the product $\Xi_1{\times}\cdots{\times}\Xi_m$ and $1\leq i\leq m$, we write $x_i$ for the $i$-th coordinate of $x$. The type of a sequence of tuples $a^1,\ldots,a^n\in \Xi_1{\times}\cdots{\times}\Xi_m$, denoted by $\operatorname{tp}(a^1,\ldots,a^n)$, is the $m$-tuple containing the types of $(a^1_i,\ldots,a^n_i)$ in $\Xi_i$ for each $1\leq i\leq m$. With this definition, also the notions of type condition, behavior, ($n$-)complete behavior, and ($n$-)canonical generalize in complete analogy from functions $f \colon \Xi{\rightarrow}\Omega$, where $\Xi$ is a “normal” structure, to functions $f \colon \Xi_1{\times}\cdots{\times}\Xi_m{\rightarrow}\Omega$ whose domain is a product. It is folklore that the Ramsey property is not lost when going to products; for the reader’s convenience, we provide a proof here. \[lem:ORPL\] Let $\Xi_1,\ldots,\Xi_m$ be ordered and Ramsey, and set $\Xi:=\Xi_1{\times}\cdots{\times}\Xi_m$. Let moreover a number $k\geq 1$, an $n$-tuple $(a^1,\ldots,a^n) \in\Xi$, and finite $F_i\subseteq\Xi_i$ be given. Then there exist finite $S_i\subseteq\Xi_i$ with the property that whenever the $n$-tuples in $S:=S_1{\times}\cdots{\times}S_m$ of type $\operatorname{tp}(a^1,\ldots,a^n)$ are colored with $k$ colors, then there is a copy of $F:=F_1{\times}\cdots{\times}F_m$ in $S$ on which the coloring is constant. We use induction over $m$. The induction beginning $m=1$ is trivial, so assume $m>1$ and that the lemma holds for $m-1$. For all $1\leq i\leq n$, set $c^i:=(a^i_1,\ldots,a^i_{m-1})$. By the induction hypothesis, there exist finite $S_i\subseteq\Xi_i$ for all $1\leq i\leq m-1$ such that whenever its $n$-tuples of type $\operatorname{tp}(c^1,\ldots,c^n)$ are colored with $k$ colors, then there is a copy of $F_1{\times}\cdots{\times}F_{m-1}$ in $S_1{\times}\cdots{\times}S_{m-1}$ on which the coloring is constant. Let $p$ be the number of $n$-tuples of this type in $S_1{\times}\cdots{\times}S_{m-1}$. Also by induction hypothesis, there exists a finite $S_{m,1}\subseteq\Xi_m$ with the property that whenever its $n$-tuples of type $\operatorname{tp}(a^1_m,\ldots,a^n_m)$ are colored with $k$ colors, then it contains a monochromatic copy of $F_m$. Further, there is a finite $S_{m,2}\subseteq \Xi_m$ with the property that whenever its subsets of this type are colored with $k$ colors, then it contains a monochromatic copy of $S_{m,1}$. Continue constructing finite substructures of $\Xi_m$ like that, arriving at $S_m:=S_{m,p}$. We claim that $S:=S_1{\times}\cdots{\times}S_m$ has the desired property. To see this, let a coloring $\chi$ of the $n$-tuples in $S$ of type $\operatorname{tp}(a^1,\ldots,a^n)$ be given. Let $b(1),\ldots,b(p)$ be an enumeration of all the $n$-tuples in $S_1{\times}\cdots{\times}S_{m-1}$ which have type $\operatorname{tp}(c^1,\ldots,c^n)$. For $1\leq i\leq p$ and $1\leq j\leq n$, we write $b(i)^j$ for the $j$-th component of $b(i)$ (note that this component is an $(m-1)$-tuple in $S_1{\times}\cdots{\times}S_{m-1}$). Now for all $1\leq i\leq p$, define a coloring $\chi^i$ of the $n$-tuples $t=(t^1,\ldots,t^n)$ in $S_m$ of type $\operatorname{tp}(a^1_m,\ldots,a^n_m)$ by setting $\chi^i(t):=\chi(b(i)^1t^1,\ldots,b(i)^nt^n)$, where $rs$ denotes the concatenation of two tuples $r,s$. By thinning out $S_m$ $p$ times, we obtain a copy $F_m'$ of $F_m$ in $S_m$ on which each coloring $\chi^i$ is constant with color $q^i$. Now by that construction, all $n$-tuples $b(i)$ have been assigned a color $q^i$, the assignment thus being a coloring of all the $n$-tuples of type $\operatorname{tp}(c^1,\ldots,c^n)$ in $S_1{\times}\cdots{\times}S_{m-1}$. By the choice of that product, there is a copy $F'_1{\times}\cdots{\times}F'_{m-1}$ of $F_1{\times}\cdots{\times}F_{m-1}$ in $S_1{\times}\cdots{\times}S_{m-1}$ on which that coloring is constant, say with value $q$. But that means that if a tuple $(d^1,\ldots,d^n) \in F_1'{\times}\cdots{\times}F_m'$ has type $\operatorname{tp}(a^1,\ldots,a^n)$, then $\chi(d^1,\ldots,d^n)=q$, proving our statement. We now generalize the notion of a transformation monoid to higher arities. Denote the set of all polymorphisms of $\Delta$ by $\operatorname{Pol}(\Delta)$. Irrespectively of the structure $\Delta$, this set contains all finitary projections and is closed under composition. Sets of finitary functions with these two properties are referred to as clones* – for a survey of clones on infinite sets, see [@GoldsternPinsker]. In addition, the clone $\operatorname{Pol}(\Delta)$ is a closed subset of the sum space of the spaces $D^{D^n}$, where $D$ is again taken to be discrete; such clones are called closed, local, or locally closed (cf. the corresponding terminology for monoids before). This means that if a set ${\mathcal F}$ of finitary functions on a domain $D$ preserves a set of given relations, then so does the smallest closed clone containing ${\mathcal F}$, motivating the following extension of Definition \[defn:generatesUnary\]. \[defn:generatesFinitary\] Let $D$ be a set, $g: D^m{\rightarrow}D$, and let ${\mathcal F}$ be a set of finitary operations on $D$. We say that ${\mathcal F}$ generates $g$ iff $g$ is contained in the smallest closed clone containing ${\mathcal F}$. For a structure $\Delta$ with domain $D$ and a function $f \colon D^n{\rightarrow}D$, we say that $f$ generates $g$ over $\Delta$ iff $\{f\}\cup\operatorname{Aut}(\Delta)$ generates $g$. Equivalently, for every finite subset $F$ of $\Delta^m$, there exists an $m$-ary term built from $f$, $\operatorname{Aut}(\Delta)$, and projections, which agrees with $g$ on $F$. As before, finitary functions on ordered homogeneous Ramsey structures generate canonical functions, and we can add constants to the language. \[lem:canonicalConstantsHigherArity\] Let $\Delta$ be ordered homogeneous Ramsey with finite relational signature, and let $f \colon \Delta^m{\rightarrow}\Delta$. Let moreover finite tuples $c_1=(c_1^1,\ldots,c_1^{n_1}),\ldots,c_m=(c_m^1,\ldots,c_m^{n_m})$ of constants in $\Delta$ be given. Then $f$ generates over $\Delta$ an $m$-ary operation $g$ on $\Delta$ which is canonical as a function from $(\Delta,c_1){\times}\cdots{\times}(\Delta,c_m)$ to $\Delta$ and which agrees with $f$ on all tuples $(c_1^{j_1},\ldots,c_m^{j_m})$. We recommend combining Lemma \[lem:ORPL\] with the methods of the preceding section in order to prove this. The set of all closed clones on a fixed domain $D$ forms a complete lattice with respect to inclusion; it is the lattice of all polymorphism clones of structures with domain $D$. This lattice has been investigated in universal algebra (see [@Pin-morelocal]). \[defn:minimalClone\] For closed clones ${\mathcal C}, {\mathcal D}$ on the same set, we say that ${\mathcal D}$ is minimal above ${\mathcal C}$ iff ${\mathcal C}\subsetneq {\mathcal D}$ and ${\mathcal C}\subsetneq {\mathcal E}\subseteq {\mathcal D}$ implies ${\mathcal E}={\mathcal D}$ for all closed clones ${\mathcal E}$. Every minimal closed clone above ${\mathcal C}$ is generated by ${\mathcal C}$ plus a single function $f$ outside ${\mathcal C}$; we call such a function $f$ minimal above ${\mathcal C}$ if there is no function of smaller arity which generates (together with ${\mathcal C}$) the same closed clone as $f$. Lemma \[lem:canonicalConstantsHigherArity\] allows us to find the minimal clones above a closed clone on an ordered homogeneous Ramsey structure. The main difference here compared with monoids is that the arities of minimal canonical functions are not bounded a priori, which means that there could be infinitely many minimal clones. The following lemma, which has been observed in [@tcsps-journal], yields a bound on the arities of minimal functions. \[lem:arityReduction\] Let $\Theta$ be a structure, $m \geq 1$, and let $R\subseteq \Theta^n$ be a relation which intersects precisely $m$ $n$-orbits of $\Theta$. If a function $f \colon \Theta^p {\rightarrow}\Theta$ violates $R$, then $f$ generates over $\Theta$ a function of arity $m$ which violates $R$, too. Let $O_1,\ldots,O_m$ be the orbits of $\Theta$ that are intersect $R$, and fix arbitrary tuples $s_i\in O_i$. Since $f$ violates $R$, there exist $r_1,\ldots,r_p\in R$ such that $f(r_1,\ldots,r_p){\notin}R$. Say that $b_i\in O_{j_i}$, for all $1\leq i\leq p$, and choose for all $1\leq i\leq p$ an automorphism $\alpha_i$ of $\Theta$ sending $s_{j_i}$ to $r_{i}$. The function $g(x_1,\ldots,x_m):=f(\alpha_{1}(x_{i_1}),\ldots,\alpha_p(x_{i_p}))$ has arity $m$ and violates $R$ since $g(s_1,\ldots,s_m)=f(r_1,\ldots,r_p)$ is not in $R$. \[prop:finiteMinimalClones\] Let $\Theta$ be a finite relational signature reduct of an ordered homogeneous Ramsey structure $\Delta$ with finite relational signature. Then there are finitely many minimal closed clones above $\operatorname{Pol}(\Theta)$, and every closed clone containing $\operatorname{Pol}(\Theta)$ contains a minimal one. Let $R_1,\ldots,R_n$ be the relations of $\Theta$. If $f$ is a minimal operation above $\operatorname{Pol}(\Theta)$, then it violates a relation $R_i$. By Lemma \[lem:arityReduction\], it generates over $\Theta$ a function of arity $o^\Theta(k_i)$, where $k_i$ is the arity of $R_i$, which still violates $R_i$. Setting $m$ to be the maximum of the $o^\Theta(k_i)$ where $1\leq i\leq n$, we get that every minimal clone above $\operatorname{Pol}(\Theta)$ is generated by a function of arity at most $m$. By Lemma \[lem:canonicalConstantsHigherArity\], such functions can be made canonical – the rest of the proof is just like the proof of Proposition \[prop:finiteMinimalReducts\]. If one wishes to determine the minimal clones above the endomorphism monoid of a structure $\Theta$, then there is a bound on the arities of minimal functions which only depends of the number of $2$-orbits of the structure $\Theta$, rather than the number of orbits of possibly longer tuples as in the preceding proof. Let $D$ be a set, and let $f \colon D^m {\rightarrow}D$ be an operation on $D$. Then $f$ is called essentially unary iff there exist $1\leq i\leq m$ and $F \colon D{\rightarrow}D$ such that $f(x_1,\ldots,x_m)=F(x_i)$. Conversely, $f$ is called essential iff it is not essentially unary. \[prop:finiteMinimalClonesAboveEnd\] Let $\Theta$ be any relational structure for which $o^\Theta(2)$ is finite. Then every minimal closed clone above $\operatorname{End}(\Theta)$ is generated by a function of arity at most $2\cdot o^\Theta(2)-1$ together with $\operatorname{End}(\Theta)$. Let ${\mathcal D}$ be a minimal closed clone above $\operatorname{End}(\Theta)$. If all the functions in ${\mathcal D}$ are essentially unary, then ${\mathcal D}$ is generated by a unary operation together with $\operatorname{End}(\Theta)$ and we are done. Otherwise, let $f$ be an essential operation in ${\mathcal D}$. Then one can verify that $f$ violates the $3$-ary relation $P_3$ defined by the formula $(x=y) \vee (y=z)$. The assertion then follows from Lemma \[lem:arityReduction\]: the $3$-ary subrelation of $P_3$ defined by the formula $x=y$ clearly consists of $o^\Theta(2)$ orbits in $\Theta$; similarly, the $3$-ary subrelation defined by $y=z$ consists of the same number of orbits. Since $P_3$ is the union of these two subrelations, and since the intersection of the two subrelations consists of exactly one orbit (namely, the triples with three equal entries), we obtain $2\cdot o^\Theta(2)-1$ different orbits for tuples in $P_3$. Observe that in Proposition \[prop:finiteMinimalClonesAboveEnd\], if $\Theta$ is a reduct of a structure $\Delta$, we can also write $2\cdot o^\Delta(2)-1$ for the arity bound if we wish to have a bound which is independent of $\Theta$, since $\Delta$ has at least as many $2$-orbits as $\Theta$. The Algorithm {#sect:algorithm} ============= We now present the algorithm proving Theorem \[thm:main:pp\]; the proof of the two statements of Theorem \[thm:main:exep\] is a subset. So we are given formulas $\phi_0,\ldots,\phi_n$ over $\Gamma$ which define relations $R_0,\ldots,R_n$ on the domain $D$ of $\Gamma$. Set $\Theta$ to be the reduct $(D;R_1,\ldots,R_n)$ of $\Gamma$, and write $R:=R_0$. We will decide whether there is a primitive positive definition of $R$ in $\Theta$. Operationalization ------------------ If there is no such definition, then since $\Theta$ is $\omega$-categorical, by Theorem \[thm:preservation\] there is a polymorphism $f$ of $\Theta$ which violates $R$; we call $f$ a witness. Our algorithm will now try to build a witness. If it fails to do so, then $R$ is primitive positive definable in $\Theta$; otherwise, it is not. Arity reduction --------------- Let $k$ be the arity of $R$. By Lemma \[lem:arityReduction\], if there exists a witness, then there exists also a witness of arity equal to the number of those $k$-orbits in $\Theta$ that intersect $R$. This number is not larger than $o^\Theta(k)$, which is not larger than $o^\Gamma(k)$ since $\operatorname{Aut}(\Gamma)\subseteq\operatorname{Aut}(\Theta)$. Set $m:=o^\Gamma(k)$; the algorithm now tries to detect a witness of arity $m$. Ramseyfication -------------- If $f$ is a witness of arity $m$, then there are $k$-tuples $c_1,\ldots,c_m\in R$ such that $f(c_1,\ldots,c_m){\notin}R$. By assumption, $\Gamma$ has a first-order definition in an ordered homogeneous structure $\Delta$ that is finitely bounded, Ramsey, and has finite relational signature. By Lemma \[lem:canonicalConstantsHigherArity\], $f$ generates over $\Delta$ an $m$-ary function $g$ which is canonical as a function from $(\Delta,c_1){\times}\cdots{\times}(\Delta,c_m)$ to $\Delta$ and which agrees with $f$ on all $m$-tuples whose $i$-th component is taken from the $k$-tuple $c_i$ for all $1\leq i\leq m$. In particular, $g$ still violates $R$ and preserves $\Theta$, and hence is a witness, too. Our algorithm thus tries to find a witness of this form. Finite representation --------------------- Let $n:=\max(s,n(\Delta),3)$, where $s$ is the maximal size of the finitely many finite forbidden substructures of $\Delta$. Since $n \geq n(\Delta)$, a function from $(\Delta,c_1){\times}\cdots{\times}(\Delta,c_m)$ to $\Delta$ is canonical iff it is $n$-canonical. Such functions can thus be represented as functions from $S^{(\Delta,c_1)}_n{\times}\cdots{\times}S^{(\Delta,c_m)}_n$ to $S^\Delta_n$. Note that the type space $S^{(\Delta,c_i)}_n$ only depends on the type of $c_i$ in $\Delta$. In other words, if we replace the tuple $c_i$ by a tuple $d_i$ of the same type in $\Delta$, we obtain the same possibilities of complete behavior. Since $o^\Delta(k)$ is finite, there are only finitely many choices of types for each $c_i$ – our algorithm tries all such choices (since $\Delta$ has a finite relational signature, and is homogeneous, those choices can be made effectively). For each choice for the types of the $c_i$, and for each function $\sigma$ from $S^{(\Delta,c_1)}_n{\times}\cdots{\times}S^{(\Delta,c_m)}_n$ to $S^\Delta_n$, the algorithm checks whether $\sigma$ is the behavior of a witness. Verification ------------ Given $\sigma$, we verify the following. - (Compatibility.) If $\sigma$ is a behavior of a canonical operation, then for all $1\leq k\leq n$ it must also be extendible to a function from $S^{(\Delta,c_1)}_k{\times}\cdots{\times}S^{(\Delta,c_m)}_k$ to $S^\Delta_k$. This is possible in the following situation: if $s$ is an $n$-type, then it has certain $k$-subtypes $t$, i.e., projections of tuples of type $s$ onto $k$ coordinates satisfy $t$. Now products of $k$-subtypes are automatically sent to a $k$-subtype under $\sigma$: if $s_1,\ldots,s_m$ are $n$-types and $I\subseteq \{1,\ldots,n\}$ is a set of size $k$ inducing $k$-subtypes $t_i$ of $s_i$, then $I$ induces a $k$-subtype of $\sigma(s_1,\ldots,s_m)$. Our algorithm checks for $n$-types $p_1,q_1,\ldots,p_m,q_m$ and all $I, J\subseteq \{1,\ldots,n\}$ that if $I$ and $J$ induce identical $k$-subtypes in $p_i$ and $q_i$, respectively, then they induce identical $k$-subtypes in $\sigma(p_1,\ldots,p_m)$ and $\sigma(q_1,\ldots,q_m)$ – otherwise, $\sigma$ is rejected as a candidate. If on the other hand $\sigma$ satisfies this condition, then it naturally extends to a function from $S^{(\Delta,c_1)}{\times}\cdots{\times}S^{(\Delta,c_m)}$ to $S^\Delta$ respecting arities, and we can compute the value of this function for every argument. In the following, we write $\sigma$ for this extended function. - (Violation.) Since $R$ has a first-order definition in $\Delta$, and automorphisms of $\Delta$ preserve first-order formulas, it follows that $R$ is a union of orbits, i.e., if $a,b$ are of the same type, then $a\in R$ iff $b\in R$. Set $t:=\sigma(\operatorname{tp}^{(\Delta,c_1)}(c_1),\ldots,\operatorname{tp}^{(\Delta,c_m)}(c_m))$. Our algorithm checks that $t$ is not a type in $R$, since we only want to accept $\sigma$ if it is the behavior of an operation which violates $R$ on $c_1,\ldots,c_m$. - (Preservation.) For every relation $R_i$ from $\Theta$, we check that $\sigma$ “preserves” $\Theta$ as follows: write $p$ for the arity of $R_i$. For all $p$-types $t_1,\ldots,t_m$ of tuples in $R_i$, we verify that $\sigma(t_1,\ldots,t_m)$ is the type of a tuple in $R_i$; otherwise we reject $\sigma$. We now argue that the algorithm finds a $\sigma$ satisfying our three conditions if and only if there is an $m$-ary polymorphism of $\Theta$ that violates $R$. It is clear that the type function of a witness will satisfy all the conditions, so one direction is straightforward. For the opposite direction, suppose that $\sigma$ is accepted by our algorithm. We build a canonical operation from $(\Delta,c_1){\times}\cdots{\times}(\Delta,c_m)$ to $\Delta$ in three steps. Let $\tau$ be the signature of $\Delta$. We first construct an infinite structure $\Pi$ with domain $D^m$ and signature $\tau \cup \{\sim\}$, where $\sim$ is a new binary relation symbol, as follows. This relation is for the proper treatment of equality of function values when realizing the behavior $\sigma$. For all $(a_1,b_1),\dots,(a_m,b_m) \in D^2$ with types $t_1,\dots,t_m$ in $(\Delta,c_1),\dots,(\Delta,c_m)$, respectively, if the 2-type $\sigma(t_1,\dots,t_m)$ contains $x_1 = x_2$ then we set $(a_1,\dots,a_m) \sim (b_1,\dots,b_m)$. Note that since $n \geq 3$ and because of the compatibility constraints and transitivity of equality, $\sim$ then denotes an equivalence relation on $D^m$. The other relations of $\Pi$ are defined as follows. Let $R$ be a $k$-ary relation from $\tau$. We add the $k$-tuple $((a^1_1,\dots,a^1_m),\dots,(a^k_1,\dots,a^k_m))$ to the relation $R$ of $\Pi$ if and only if $R(x_1,\dots,x_k)$ is contained in $\sigma(t_1,\dots,t_m)$, where $t_i$ is the type of the tuple $(a^1_i,\dots,a^k_i) \in D^k$ in $(\Delta,c_i)$. Since $n \geq n(\Delta) \geq k$, this is well-defined by the compatibility item of our algorithm. The quotient structure $\Pi /_{\sim}$ is defined to be the $\tau$-structure whose domain is the set $D/_{\sim}$ of all equivalence classes of $\sim$, and where $R(E_1,\dots,E_p)$ holds for a $p$-ary $R \in \tau$ and $E_1,\dots,E_p \in D/_{\sim}$ if and only if there are $b_1 \in E_1,\dots,b_p \in E_p$ such that $R(b_1,\dots,b_p)$ holds in $\Pi$. The final step is to show that there exists an embedding $f$ of $\Pi/_{\sim}$ into $\Delta$. By $\omega$-categoricity of $\Delta$ and a standard compactness argument (see, e.g., Lemma 2 in [@BodDalJournal]), it suffices to show every finite substructure $\Omega$ of $\Pi/_{\sim}$ embeds into $\Delta$. This follows from the fact that none of the forbidden substructures embeds into $\Omega$, since $n \geq s$, where $s$ is the size of the largest obstruction. Finally, observe that the mapping $g$ from $D^m$ to $D$ that maps every $u$ in $D^m$ to $f({u}/_{\sim})$ (where ${u}/_{\sim}$ denotes the $\sim$-equivalence class of $u$ in $\Pi$) is a polymorphism of $\Theta$ by the preservation item of the algorithm, and that $g$ violates $R$ by the violation item of the algorithm. Decidability of Polymorphism Conditions ======================================= In all known cases of structures $\Gamma$ with a finite relational signature and a first-order definition in a finitely bounded ordered homogeneous Ramsey structure, $\operatorname{CSP}(\Gamma)$ is tractable if and only if there exists a $4$-ary polymorphism $f$ of $\Gamma$ and an automorphism $\alpha$ of $\Gamma$ such that for all elements $x,y,z$ of $\Gamma$ $$\begin{aligned} f(x,y,z,z) & = \alpha(f(y,z,x,y)) & ()\end{aligned}$$ One can show that condition $()$ describes indeed the frontier between tractability and NP-hardness for reducts of $({\mathbb Q};<)$ and the random graph. It has also been conjectured to be the tractability frontier of $\operatorname{CSP}(\Gamma)$ for structures $\Gamma$ with a finite domain [@JBK; @Siggers]. When $\Gamma$ is given by defining quantifier-free formulas over $\Delta$, and $\Delta$ is given by its forbidden induced substructures, then the existence of $f,\alpha$ satisfying condition $()$ can be tested by an algorithm, by the techniques developed here. A $4$-ary operation $f$ satisfies this condition if and only if the type function $\sigma$ of $f$ satisfies $\sigma(t_1,t_2,t_3,t_3)=\sigma(t_2,t_3,t_1,t_2)$ for all $n$-types $t_1,t_2,t_3$ of $\Delta$, where $n \geq \max(n(\Delta),3,s)$ and $s$ is the maximal obstruction size of $\Delta$. Discussion and Open Problems ============================ We presented an algorithm that decides primitive positive definability in finite relational signature reducts $\Gamma$ of structures that are ordered, Ramsey, homogeneous, finitely bounded, and with finite relational signature. All of those structures $\Gamma$ are $\omega$-categorical. While the condition for $\Gamma$ might appear rather restrictive at first sight, it is actually are quite general: we want to point out that we do not require that $\Gamma$ is Ramsey, we only require that $\Gamma$ is definable in a Ramsey structure. We do not know of a single homogeneous structure $\Gamma$ with finite relational signature which is not* the reduct of an ordered homogeneous Ramsey structure with finite relational signature. Does every structure which is homogeneous in a finite relational signature have a homogeneous expansion by finitely many relations such that the resulting structure is Ramsey? A variant of this problem is the following. Does every $\omega$-categorical structure have an $\omega$-categorical expansion which is Ramsey? Note that our method is non-constructive: the algorithm does not produce a primitive positive definition in case that there is one. It is an interesting open problem to come up with bounds on the number of existential variables that suffice for a primitive positive definition of $R$ in $\Theta$. For many structures $\Gamma$ of practical interest, such as $({\mathbb Q};<)$ or the random graph, our algorithm can certainly be tuned so that ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$ becomes feasible for reasonable input size; in particular, the gigantic Ramsey constants involved in the proofs of our results do not affect the running time of our procedure. Another important open problem is whether the method can be extended to show decidability of our computational problem for first-order definability instead of primitive positive, existential positive, and existential definability; we denote this computational problem by ${\operatorname{Expr}_{\operatorname{pp}}}(\Gamma)$. By the theorem of Ryll-Nardzewski, first-order definability is characterized by preservation under automorphisms, i.e., surjective self-embeddings. But the requirement of surjectivity is difficult to deal with in our approach. Let $\Delta$ be a structure which is ordered, homogeneous, Ramsey, finitely bounded, and has a finite relational signature, and let $\Gamma$ be a reduct of $\Delta$ with finite relational signature. Is the problem ${\operatorname{Expr}_{\operatorname{fo}}}(\Gamma)$ decidable? [^1]: In this article, substructures are always meant to be induced; see [@Hodges]. [^2]: Such maps are called cocycles, and the given identity is called the cocycle identity.	{ "pile_set_name": "ArXiv" }
--- abstract: 'Since the seminal work of Ambrosetti and Prodi, the study of global folds was enriched by geometric concepts and extensions accomodating new examples. We present the advantages of considering fibers, a construction dating to Berger and Podolak’s view of the original theorem. A description of folds in terms of properties of fibers gives new perspective to the usual hypotheses in the subject. The text is intended as a guide, outlining arguments and stating results which will be detailed elsewhere.' author: - 'Marta Calanchi, Carlos Tomei and André Zaccur' title: Fibers and global geometry of functions --- Dedicated to Djairo, an example to follow in many directions [Keywords:]{} Dolph-Hammerstein theorem, Semilinear elliptic equations, Ambrosetti-Prodi theorem, folds. [MSC-class:]{} 35B32, 35J91, 65N30. Introduction {#sec:intro} ============ When we teach the first courses in calculus and complex or real analysis, a great emphasis is given to geometric issues: we plot graphs, enumerate conformal mappings among special regions, identify homeomorphisms. Alas, this is far from being enough: mappings become too complicated soon. Still, the geometric approach, especially combined with numerical arguments, is very fruitful in some nonlinear contexts. It is rather surprising that some infinite dimensional maps can be studied in a similar fashion — one may even think about their graphs! The examples which are amenable to such approach are very few, and they elicit the same sense of wonder that (the equally rare) completely integrable systems do: one is left with a feeling of deep understanding. This text is dedicated to some such examples. The interested reader could hardly do better than going through the review papers by Church and Timourian ([@CT1], [@CT2]), which cover extremely well the material up to the mid nineties. Their approach is strongly influenced by the original Ambrosetti-Prodi view of the problem, which we describe in Section \[globalfolds\]. In a nutshell, the global geometry of a proper function $F$ is studied through certain properties of its critical set $C$ together with its image $F(C)$, along with the stratification of $C$ in terms of singularities. This much less ambitious text is mainly an enumeration of techniques and of some recent developments, some of which have not been published. We mostly take the Berger-Podolak route ([@BP]) which has been extended by Podolak in [@P] and, we believe, still allows for improvement. Instead of the critical set, we concentrate on the restriction of $F$ to appropriate low dimensional manifolds (one dimensional, in the Ambrosetti-Prodi case), the so called fibers. Essentially, fibers are appropriate in the presence of [finite spectral interaction]{}, which roughly states that the function $F: X \to Y$ splits into a sum of linear and nonlinear terms, $F = L - N$ and $N$ deforms $L$ substantially only along a few eigenvectors spanning a subspace $V \subset X$. The domain splits into orthogonal subspaces, $X = H \oplus V$ and the hypotheses on the nonlinearities are naturally anisotropic. Different requests on $H$ and $V$ yield a global Lyapunov-Schmidt decomposition of $F$: on affine subspaces obtained by translating $H$, $F$ is a homeomorphism and complications due to the nonlinear term manifest on fibers, which are graphs of functions from $V$ to $H$. Fibers are also convenient for the verification of properness of $F$. In particular, one may search for folds in nonlinear maps defined on functions with unbounded domains, which are natural in physical situations. Fibers also provide the conceptual starting point for algorithms that solve a class of partial differential equations, an idea originally suggested by Smiley ([@S], [@SC]) and later implemented for finite spectral interaction of the Dirichlet Laplacian on rectangles in [@CT]. An abstract setup in the spirit of the characterization of folds as in [@CT1], or like the one we present in Section \[sec:adapted\], provide a better understanding of the role of the hypotheses in the fundamental example of Ambrosetti and Prodi. Elliptic theory seems to be less relevant than one might think, it is just that it provides a context in which the required hypotheses are satisfied. In Section 2, we present the seminal examples — the Dolph-Hammerstein homeomorphisms and the Ambrosetti-Prodi fold — in a manner appropriate for our arguments. Fibers and sheets are defined and constructed in Section 3. A global change of coordinates in Section 4 gives rise to adapted coordinates, in which the description of critical points is especially simple. A characterization of the critical points strictly in terms of spectral properties of the Jacobian $DF$ is given. Also, the three natural steps to identify global folds become easy to identify. Further study of how to implement each step is the content of Sections 5, 6 and 7. The last section is dedicated to some examples. The text is written as a guide: we try to convey the merits of a set of techniques, without providing details. Complete proofs will be presented elsewhere ([@CTZ1], [@CTZ2]). Alas, we stop at folds. There are scattered results in which local or global cusps were identified: again, the excellent survey [@CT2] covers the material up to the mid nineties. So far, the description of cusps seems rather ad hoc. There are characterizations ([@CT2]), but they are hard to verify and new ideas are needed. On the other hand, checking that maps are not global folds is rather simple, a matter of showing for example that some points in the image have more than two preimages. A numerical example is exhibited in Section \[subsec:numerics\]. The first examples in infinite dimension ======================================== Among the simplest continuous maps between Hilbert spaces are homeomorphisms, in particular linear isomorphisms. A second class of examples are folds. Homeomorphisms: Dolph and Hammerstein ------------------------------------- Dolph and Hammerstein ([@D], [@H]) obtained a simple condition under which nonlinear perturbation of linear isomorphisms are still homeomorphisms. A version of their results is the following. Start with a real Hilbert space $Y$ and a self-adjoint operator $L : X \subset Y \to Y$ for a dense subspace $X$ of $Y$. Let $\sigma(L)$ be the spectrum of $L$. \[theo:DH\] Let $[-c,c] \cap \sigma(L) = \emptyset$ and suppose $N: Y \to Y$ is a Lipschitz map with Lipschitz constant $n < c$. Equip $X$ with the graph topology, $\\| x\\|_X = \\| x \\|_Y + \\| Lx\\|_Y$. Then the map $F = L - N: X \to Y$ is a Lipschitz homeomorphism. Indeed, to solve $F(x) = y$, search for a fixed point of $$C_y : Y \to Y, \quad C_y(z) = N ( L^{-1}(z)) + y$$ which is a contraction because the operator $L^{-1}: Y \to Y$ has norm less than $1/c$ by standard spectral theory and then the map $N \circ L^{-1}$ is Lipschitz with constant less than $n/c< 1$ . As usual, the fixed point varies continuously with $y$. Clearly, $F$ is Lipschitz. To show the same for $F^{-1}$, keep track of the Banach iteration. Notice that the statement allows for differential operators between Sobolev spaces. Very little is required from the spectrum of $L$. Clearly, for symmetric bounded operators one should take $X= Y$. Breaking the barrier: the Ambrosetti-Prodi theorem {#globalfolds} -------------------------------------------------- What about more complicated functions? Ambrosetti and Prodi ([@AP]) obtained an exquisite example. After refinements by Micheletti and Manes ([@MM]), Berger and Podolak ([@BP]) and Berger and Church ([@BC]), the result may be stated as follows. Let $\Omega\subset{{\mathbb{R}}}^n$ be a connected, open, bounded set with smooth boundary (for nonsmooth boundaries, see [@TZ]). Let $H^2(\Omega)$ and $H^1_0(\Omega)$ be the usual Sobolev spaces and set $X = H^2(\Omega) \cap H^1_0(\Omega)$ and $Y = H^0(\Omega) = L^2(\Omega)$. The eigenvalues of the Dirichlet Laplacian $-\Delta: X \subset Y \to Y$ are $$\sigma(- \Delta) =\{0<\lambda_1<\lambda_2\le\ldots\rightarrow\infty\}.$$ Denote by $\phi_1$ the ($L^2$-normalized, positive) eigenvector associated to $\lambda_1$ and split $X = \ H_X \oplus V_X , Y = \ H_Y \oplus V_Y $ in [horizontal]{} and [vertical]{} orthogonal subspaces, where $V_X = V_Y = \langle \phi_1 \rangle$, the one dimensional (real) vector space spanned by $\phi_1$. \[theo:AP\] Let $F: X \to Y$ be $F= L-N$, where $L = - \Delta$, $N(u) = f(u)$, for a smooth, strictly convex function $f: {{\mathbb{R}}}\to {{\mathbb{R}}}$ satisfying $${{\operatorname{Ran}}}\ f' = (a,b) \, , \quad a < \lambda_1 < b <\lambda_2 \,.$$ Then there are global homeomorphisms $\zeta:X \to H_Y \oplus {{\mathbb{R}}}$ and $\xi:Y \to H_Y \oplus{{\mathbb{R}}}$ for which $\tilde{F}(z,t) = \xi \circ F \circ \zeta^{-1} (z,t) = (z,-t^2)$. Said differently, the following diagram commutes. $$\begin{array}{ccc} X& \stackrel{{\scriptstyle F}}{\longrightarrow}&Y\\ {\scriptstyle \zeta}\downarrow & & \downarrow{\scriptstyle \xi}\\ H_Y \oplus {{\mathbb{R}}}&\stackrel{{\scriptstyle (z, -t^2)}}{\longrightarrow}& H_Y \oplus {{\mathbb{R}}}\\ \end{array}$$ Functions which admit such dramatic simplification are called [global folds]{}. The vertical arrows in the diagram above are (global) changes of variables and sometimes will be $C^1$ maps, but we will not emphasize this point. The original approach by Ambrosetti and Prodi is very geometric ([@AP]). In a nutshell, they show that $F$ is a proper map whose critical set $C$ (in the standard sense of differential geometry, the set of points $u \in X$ for which the derivative $DF(u)$ is not invertible) is topologically a hyperplane, together with its image $F(C)$. They then show that $F$ is proper, its restriction to $C$ is injective and $F^{-1}(F(C)) = C$. Finally, they prove that both connected components of $X - C$ are taken injectively to the same component of $Y - F(C)$. Their final result is a counting theorem: the number of preimages under $F$ can only be $0$, $1$ or $2$. Berger and Podolak ([@BP]), on the other hand, construct a global Lyapunov-Schmidt decomposition for $F$. For $V_X= V_Y = \langle \phi_1 \rangle$, consider [affine horizontal (resp. vertical)]{} subspaces of $X$ (resp. $Y$), i.e., sets of the form $ H_X + t \phi_1 $, for a fixed $t \in {{\mathbb{R}}}$ (resp. $y + V_Y$, for $y \in H_Y$). Let $P: Y \to H_Y$ be the orthogonal projection. The map $P F_t: H_X \to H_Y, P F_t(w) = P F(w + t \phi_1)$, is a bi-Lipschitz homeomorphism, as we shall see below. Thus, the inverse under $F$ of vertical lines $y + V_Y$, for $y \in H_Y$ are curves $\alpha_y : {{\mathbb{R}}}\sim V_X \subset X \to H_X$, which we call [fibers]{}. Fibers stratify the domain $X$. Thus, to show that $F$ is a global fold, it suffices to verify that each restriction $F:\alpha_y \to V_Y \sim {{\mathbb{R}}}$, essentially a map from ${{\mathbb{R}}}$ to ${{\mathbb{R}}}$, is a fold. After such a remarkable example, one is tempted to push forward. This is not that simple: if the (generic) nonlinearity $f$ is not convex, there are points in $Y$ with four preimages ([@CTZ1]), so the associated map $F:X \to Y$ cannot be a global fold (for a numerical example, see Section \[subsec:numerics\]). Fibers and height functions {#sec:fibers} =========================== Fibers come up in [@BP] and [@SC] for $C^1$ maps associated to second order differential operators and in [@MST2] in the context of first order periodic ordinary differential equations. Due to the lack of self-adjointness, the construction in [@MST2] is of a very different nature. We follow [@P] and [@TZ], which handle Lipschitz maps, allowing the use of piecewise linear functions in the Ambrosetti-Prodi scenario, namely $f$ given by ${f'}(x) = a \hbox{ or } b$, depending if $x <0$ or $x>0$ ([@CFS], [@LM2]). Let $X$ and $Y$ be Hilbert spaces, $X$ densely included in $Y$. Let $L: X \subset Y \to Y$ be a self-adjoint operator with a simple, isolated, eigenvalue $\lambda_p$, with eigenvector $\phi_p \in X$ with $\\|\phi_p\\|_Y = 1$. Notice that $\lambda_p$ may be located anywhere in the spectrum $\sigma(L)$ of $L$. As before, consider horizontal and vertical orthogonal subspaces, $$X = \ H_X \oplus V_X \ , \ Y = \ H_Y \oplus V_Y \ , \ \hbox{ for } \ V_X = V_Y = \langle \phi_p \rangle$$ and the projection $P:Y \to H_Y$. Let $P F_t: H_X \to H_Y$ be the projection on $H_Y$ of the restriction of $F$ to the affine subspace $H_X + t \phi_p$, $P F_t(w) = P F(w+t \phi_p)$. In the same fashion, the nonlinearity $N : Y \to Y$ gives rise to maps $PN_t: H_Y \to H_Y$, which we require to be Lipschitz with constant $n$ independent of $t \in {{\mathbb{R}}}$ so that $$[-n,n] \cap \sigma(L) = \{ \lambda_p \} \ . \eqno{(H)}$$ The standard Ambrosetti-Prodi map fits these hypotheses. In this case, $X \subset Y$ are Sobolev spaces and the derivative $f ' : {{\mathbb{R}}}\to {{\mathbb{R}}}$ is bounded by $a$ and $b$. Set $$\gamma = (a + b) /2, \quad L = - \Delta - \gamma, \quad N(u) = f(u) - \gamma u$$ and $\lambda_p = \lambda_1$, the smallest eigenvalue of $-\Delta$. Then the Lipschitz constant $n$ of the maps $PN_t$ satisfies $n < \gamma - a = b - \gamma < \lambda_2 - \gamma$, so that $\lambda_1 - \gamma \le n$. \[theo:Fvv\] Let $F: X \to Y$ satisfy $(H)$ above. Then for each $t \in {{\mathbb{R}}}$, the map $P F_t$ is a bi-Lipschitz homeomorphism, and a $C^k$ diffeomorphism if $F$ is $C^k$. The Lipschitz constants for $P F_t$ and $(P F_t)^{-1}$ are independent of $t$. The proof follows Theorem \[theo:DH\] once the potentially nasty eigenvalue $\lambda_p$ is ruled out. Let $c$ be the absolute value of the point in $\sigma(L) \setminus \{ \lambda_p\}$ closest to $0$, so that $0 \le n < c$. The operator $L: X \to Y$ restricts to $L: H_X \to H_Y$, which is invertible self-adjoint, and again $L^{-1}: H_Y \to H_Y$ with $\\| L^{-1} \\| \le 1/c$. The solutions $w \in H_X$ of $P F_t(w) = g \in H_Y$ solve $PLu - PN(u) = Lw - PN_t(w) =g$ for $u = w + t \phi_p$. The solutions $w$ correspond to the fixed points of $C_g : H_Y \to H_Y$, where $$\quad C_g(z) = PN_t(L^{-1} z) + g, \quad \hbox{ for } \ Lw = z \in H_Y.$$ The map $C_g$ is a contraction with constant bounded by $ n/c < 1$ (independent of $t$). Now follow the proof of Theorem \[theo:DH\]. The attentive reader may have noticed that the effect of the nonlinearity $N$ along the vertical direction is irrelevant for the construction of fibers. The same construction applies when the interval $[-n,n]$ defined by the Lipschitz constant $n$ of $PN_t: H_X \to H_Y$ interacts with an isolated subset $I$ of $\sigma(L)$ — more precisely, $I = [-n,n] \cap \sigma(L)$ and there is an open neighborhood $U$ of $I \subset {{\mathbb{R}}}$ for which $I = U \cap \sigma(L)$. In this case $P$ is the orthogonal projection on $I$, which takes into account possible multiplicities. In the special situation when $I$ consists of a finite number of eigenvalues (accounting multiplicity), we refer to [finite spectral interaction]{} between $L$ and $N$. We concentrate on the case when $I = \{ \lambda_p \}$ consists of a simple eigenvalue. A more careful inspection of the constants in the Banach iteration in the proof above yields the following result ([@CT], [@TZ]). The image under $F$ of horizontal affine subspaces of $X$ are [sheets]{}. The inverse under $F$ of vertical lines of $Y$ are [fibers]{}. \[prop:fibers\] If $F$ is $C^1$, sheets are graphs of $C^1$ maps from $H_Y$ to $\langle \phi_p \rangle$ and fibers are graphs of $C^1$ maps from $\langle \phi_p \rangle$ to $H_X$. Sheets are essentially flat, fibers are essentially steep. We define what we mean by essential flatness and steepness. Let $\nu(y)$ be the normal at a point $y \in Y$ of (the tangent space of) a sheet, and $\tau(u)$ be the tangent vector at $ u \in X$ of a fiber. Then there is a constant $\epsilon \in (0 , \pi/2)$ such that $\phi_p$ makes an angle less than $\epsilon $ (or greater than $\pi - \epsilon$, due to orientation) with both vectors. Adapted coordinates and a plan {#sec:adapted} ============================== Suppose $L$ and $N$ interact at a simple eigenvalue $\lambda_p$. Write $$F(u) = PF (u) + \langle F(u), \phi_p \rangle \phi_p = PF (u) + h(u) \phi_p$$ where the map $h: X \to {{\mathbb{R}}}$ is called the [height function]{}. In the diagram below, invertible maps are bi-Lipschitz ([@TZ]) or $C^k$ diffeomorphisms, depending if $PF_t$ is Lipschitz or $C^k$. The smoothness of $h$ and $h^a = h \circ \Phi$ follow accordingly. $$\begin{array}{ccl} {X = H_X \oplus V_X}& \stackrel{{\scriptstyle F}}{\longrightarrow}& {Y = H_Y \oplus V_Y} \\ {\scriptstyle \Phi^{-1}=( P F_t, Id)}\searrow & & \nearrow{\scriptstyle F^a=F \circ \Phi=(Id, h^a)}\\ &{Y} & \\ \end{array}$$ The map $F$ has been put in [adapted coordinates]{} by the change of variables $\Phi$: $$F^a: Y \to Y \ , \quad (z,t) \mapsto (z, h^a(z,t)) \ .$$ Notice that fibers of $F$ are taken to vertical lines in the domain of $F^a = F \circ \Phi$. Explicitly, the vertical lines $\{ (z_0, t) : t \in {{\mathbb{R}}}\}$ parameterized by $z_0 \in H_Y$ correspond to fibers $u(z_0,t) = (P F_t)^{-1}(z_0) + t \phi_p = w(z_0,t) + t \phi_p $. Thus $F^a$ is just a rank one nonlinear perturbation: $$F^a(z,t) = (z, h^a(z,t)) \sim z + h^a(z+ t \phi_p) \phi_p \ .$$ In a very strict sense, this is also true of $F$. In order to make $F$ similar to an Ambrosetti-Prodi map, define $G = F^a \circ (- \Delta): X \to Y$: $$u \stackrel{- \Delta}{\longmapsto} z + t \phi_1 \stackrel{{\scriptstyle F^a}}{\longmapsto} z + t\phi_1 + (h^a(z+ t \phi_1)-t) \phi_1 = - \Delta u + \psi(u) \phi_1,$$ for some nonlinear functional $\psi$. We generalize slightly. \[prop:rankone\] Let $N$ be a $C^1$ map. Say $L$ and $N$ interact at a simple eigenvalue $\lambda_p$ and $L$ is invertible. Then, after a $C^1$ change of variables, the $C^1$ function $F= L-N: X \to Y$ becomes $G : X \to Y$, $G = L + \psi(u) \phi_p$, for some $\psi: X \to {{\mathbb{R}}}$. For Ambrosetti-Prodi operators $F(u) = - \Delta u - f(u)$, the nonlinear perturbation is given by a Nemitskii map $u \mapsto f(u)$. It is not surprising that once we enlarge the set of nonlinearities new global folds arise. For a map $F$ given in adapted coordinates by $F^a(z,t ) = (z , h^a(z,t))$, appropriate choices of the [adapted height function]{} $h^a$ yields all sorts of behavior. The critical set of $F:X \to Y$ is compatible with fibers as follows ([@BP], [@CTZ2]). \[prop:Cc\] Suppose the $C^1$ map $F:X \to Y$ admits fibers. Then $u_0$ is a critical point of $F$ if and only if it is a critical point of the height function $h$ along its fiber, or equivalently of the adapted height function $h^a$. Isolated local extrema have to alternate between maxima and minima. In particular, given the appropriate behavior at infinity at each fiber and the fact that all critical points are of the same type, we learn from a continuity argument that the full critical set $C$ is connected, with a single point on each fiber ([@CDT]). The study of a function $F: X \to Y$ reduces to three steps: 1. Stratify $X$ into fibers. 2. Verify the asymptotic behavior of $F$ along fibers. 3. Classify the critical points of the restriction of $F$ along fibers. The following result is natural from this point of view ([@CTZ2]). Let $F:X \to Y$ satisfies $(H)$ of Section \[sec:fibers\], so that, by Proposition \[prop:fibers\], $X$ stratifies in one dimensional fibers $\{u (z, t ) : t \in {{\mathbb{R}}}\}$, one for each $z \in H_Y$. \[prop:folds\] Suppose that, on each fiber, $$\lim_{t \to \pm \infty} \langle F( u (z, t )), \phi_p \rangle = \lim_{t \to \pm \infty} h(u (z, t )) = - \infty \ .$$ Suppose also that each critical point of $h$ restricted to each fiber is an isolated local maximum. Then $F: X \to Y$ is a global fold, in the sense that there are homeomorphisms on domain and image that give rise to a diagram as in Theorem \[theo:AP\]. To verify that such limits exist, one might check hypotheses $(V\pm)$ in Section \[subsec:asy\], but there are alternatives. Similarly, there are ways of obtaining fibers which do not fit the construction presented in Section \[sec:fibers\] (this is the case for perturbations of non-self-adjoint operators, Section \[subsec:nonself\]). The upshot is that there is some loss in formulating the three step recipe into a clear cut theorem. As trivial examples, $h^a(z,t)= - t^2$ is a global fold, whereas $h^a(z,t) = t^3 - t$ has a critical set consisting of two connected components having only (local) folds (from Section \[subsec:Morin\]). More complicated singularities require the dependence on $z$: not every fiber of $F$ (equivalently, vertical line in the domain of $F^a$) has the same number of critical points close to a cusp, for example. The reader is invited to check that $(z,t) \mapsto (z, t^3 - \langle z, \tilde{\phi} \rangle t)$ is a global cusp, for $\tilde{\phi}$ any fixed vector in $H_Y$. Higher order Morin singularities, considered in Section \[sec:singularities\], are obtained in a similar fashion. From the Proposition \[prop:rankone\], changes of variables on such maps yield nonlinear rank one perturbations of the Laplacian which are globally diffeomorphic to the standard normal forms of Morin singularities. We consider the standard Ambrosetti-Prodi scenario in the light of this strategy. For the function $F(u) = - \Delta u - f(u)$ defined in Theorem \[theo:AP\], elliptic theory yields all sort of benefits — the smallest eigenvalue of the Jacobian $DF(u)$ is always simple, the ground state may be taken to be a positive function in $X$. The hypotheses required for the construction of fibers in Theorem \[theo:Fvv\] do not imply the simplicity of the relevant eigenvalue: there are examples for which there is no naturally defined $C^1$ functional $\lambda_p: X \to {{\mathbb{R}}}$ because two eigenvalues collide. One might circumvent this difficulty by forcing the nonlinearity $N$ to be smaller, but it turns out that this is not necessary. The hypotheses instead imply the simplicity of $\lambda_p$ in an open neighborhood of the critical set $C$ of $F$, and this is all we need, as we shall see in Section \[spectralsing\]. The positivity of the ground state and the convexity of the nonlinearity $f$ are used in a combined fashion in the Ambrosetti-Prodi theorem to prove that along fibers the height function only has local maxima. Clearly, this is a property only of critical points. On the other hand, the nonlinearity $N(u) = f(u)$ is so rigid that the standard hypothesis of convexity of $f$ is essentially necessary, as shown in [@CTZ1]. More general nonlinearities require a better understanding of the singularities. We now provide more technical details on each of the three steps. Obtaining fibers in other contexts {#sec:fifibers} ================================== For starters, what if $L$ is not self-adjoint, or $X$ is not Hilbert? Podolak’s approach {#subsec:podolak} ------------------ Suppose momentarily that $X$ and $Y$ are Banach spaces. Let $L: X \to Y$ be a Fredholm operator of index zero with kernel generated by a vector $\phi_X$ and let $\phi_Y$ be a vector not in ${\operatorname{Ran}}L$. Podolak ([@P]) considered the following scenario, for which she obtained a lower bound on the number of preimages for a region of $Y$ of vectors with very negative component along $\phi_Y$. Split $X = H_X \oplus V_X$ where $V_X = \langle \phi_X \rangle$ and $H_X$ is any complement. Also, split $Y = H_Y \oplus V_Y$ where $H_Y= {\operatorname{Ran}}L$ and $V_Y = \langle \phi_Y \rangle$. In particular $L: H_X \to H_Y$ is an isomorphism. Also, define the associated projection $P: Y \to H_Y$. Write $u = w + t \phi_X, \ y = g + s \phi_Y$ for $w \in H_X$. The equation $F(u) = Lu - N(u) = y$ becomes $$L(w + t \phi_X) - N(w + t \phi_X) = L w - N(w + t \phi_X) = g + s \phi_Y,$$ and, as in Theorem \[theo:Fvv\], we are reduced to solving the map $$C_g : H_Y \to H_Y, \quad C_g(z) = PN_t(L^{-1} z) + g \ , \quad \hbox{ for } \ Lw = z \in H_Y.$$ Her hypotheses imply that such maps are contractions. Transplanting fibers -------------------- The estimates arising from spectral theorem in the Hilbert context are easy to obtain and possibly more effective. Podolak’s hypotheses are harder to verify. There is a possibility: getting fibers in Hilbert spaces and transplanting them to Banach spaces. This happens for example when moving from the Ambrosetti-Prodi example as a map between Sobolev spaces ([@BP]) to a map between Hölder spaces ([@AP]). The classification of singularities is simpler with additional smoothness (Section \[sec:singularities\]). Let $F= L - N: X \to Y$ satisfy hypothesis $(H)$ of Section \[sec:fibers\]. Consider the densely included Banach spaces $A \subset X$ and $B \subset Y$ allowing for the $C^1$ restriction $F: A \to B$ for which $V_X = V_Y \subset A$. Suppose that $DF(a): A \to B$ is a Fredholm operator of index zero for each $a \in A$. Then fibers of $F: X \to Y$ either belong to $A$ or do not intersect $A$. Said differently, if a point $u \in X$ belongs to $A$ then the whole fiber does. In the Ambrosetti-Prodi scenario, this proposition seems to be a consequence of elliptic regularity, which may be used to prove it. Regularity of eigenfunctions is irrelevant: fibers are the orbits of the vector field of their tangent vectors, which are inverses of the vertical vector under $DF(u)$, and necessarily lie in $A$ ([@CTZ2]). Tangent vectors are indeed eigenfunctions $\phi_p(u)$ of $DF(u)$ at critical points $u$. The fact that sheets and fibers are uniformly flat and steep (Proposition \[prop:fibers\]) allows one to modify vertical spaces ever slightly and still obtain space decompositions for which the Lyapunov-Schmidt decomposition, and hence the construction of fibers in Theorem \[theo:Fvv\], apply. In particular, transplants may be performed even when the eigenvector $\phi_p$ originally used to define the vertical spaces $V_X= V_Y$ do not have regularity, i.e., do not belong to $A \subset X$. We only have to require that $A$ is dense in $X$, so that $\phi_p$ can be well approximated by a new vertical direction. Fibers and Numerics {#subsec:numerics} ------------------- Finite spectral interaction is a very convenient context for numerics. Any question related to solving $F(u)= g$ for some fixed $g \in Y$ reduces to a finite dimensional problem in situations of finite spectral interaction, irrespective of additional hypotheses. If the interaction involves a simple eigenvalue $\lambda_p$, one simply has to look at the restriction of $F$ to the (one dimensional) fiber associated to the affine vertical line through $g$. Smiley and Chun realized the implications of this fact for numerical analysis ([@S], [@SC]). An implementation for functions $F(u) = - \Delta u - f(u)$ defined on rectangles $\Omega \subset {{\mathbb{R}}}^2$ was presented in [@CT]. In the forecoming sections, we will require more stringent hypotheses with the scope of obtaining very well behaved functions $F$ — we will mostly be interested in global folds. Such additional restrictions might improve on computations, but so far this has not seen to lead to substantial improvements on the available algorithms. $\begin{array}{cc} \epsfig{height=45mm,file=flinha.eps} & \epsfig{height=45mm,file=altura.eps} \end{array}$ \[fig:wil\] We present an example obtained from programs by José Cal Neto ([@CT]) and Otavio Kaminski. For $\Omega = [0,1] \times [0,2]$, $\lambda_1 \sim 12.337$ and $\lambda_2 \sim 19.739$. Consider $$- u_{xx} - u_{yy} - f(u) = g \ , \quad (x,y) \in \Omega \ , \quad u = 0 \ \hbox{ in } \ \partial \Omega \ ,$$ $$f'(x) = \frac{\lambda_2 - \lambda_1}{\pi} \ \big( \arctan (\frac{x}{10}) - \frac{2}{5}\ x \ e^{- (x/10)^2} \big) + \lambda_1 \ , \quad f(0) \sim 47.12$$ $$g(x,y) = - 100 \big( x(x-1)y^2(y-2) \big) - 35 \sin(\pi x) \sin(\frac{\pi y}{2}) \ .$$ On the left, we show the graphs of $f'$, which interacts only with $\lambda_1$. On the right, the height function $h$ associated to the fiber obtained by inverting the vertical line through $g$. The height value $-12.3$ is reached by four preimages, displayed below. Notice the cameo appearance of the maximum principle: the four graphs sit one on top of the other as one goes up along the fiber (this is very specific of interactions with $\lambda_1$ of the Laplacian with Dirichlet conditions). $\begin{array}{cc} \epsfig{height=45mm,file=respA.eps} & \epsfig{height=45mm,file=respB.eps} \\ \epsfig{height=45mm,file=respC.eps} & \epsfig{height=45mm,file=respD.eps} \end{array}$ \[fig:wil\] Asymptotics of $F$ on fibers and vertical lines =============================================== We stick to one dimensional fibers and consider two issues. 1. How does $F$ behave at infinity along fibers? 2. How do fibers look like at infinity ? The first question, to say the very least, is tantamount to characterizing the image of $F$. The second is not relevant for the theoretical study of the global geometry of $F$, since a (global) coordinate system leading to a normal form (like $(z,t) \mapsto (z, - t^2)) $ is insensitive to the shape of fibers. On the other hand, for numerical purposes, a uniform behavior at infinity of the fibers is informative. $F$ along fibers {#subsec:asy} ----------------- The inverse of a vertical line $z_0 + V_Y, z_0 \in H_Y$ is the fiber $u(z_0,t) = w(z_0,t) + t \phi_p$: $$F(u(z_0,t)) = z_0 + h^a(z_0,t)\, \phi_p \ . \eqno{(\ast)}$$ For a fixed $z_0 \in H_Y$, the $C^1$ map $ t\mapsto h^a (z_0,t)$ is the [adapted height function]{} of the fiber associated to $z_0$. Clearly, $$h^a(z_0, t) = \langle F(u(z_0, t)), \phi_p \rangle = \langle L(w(z_0,t) + t\phi_p) - N(u(z_0,t)) , \phi_p\rangle$$ so that $$h^a(z_0, t) = \lambda_p t - \langle N(u(z_0,t)), \phi_p \rangle.$$ In order to have $$\lim_{t \to \pm \infty} \langle F( u (z_0, t )), \phi_p \rangle = \lim_{t \to \pm \infty} h^a(z_0, t ) = - \infty$$ and some uniformity convenient to obtain properness as discussed in Section \[subsec:proper\], we require an extra hypothesis: For each $z_0 \in X$, there is a ball $U(z_0) \subset X$ and $\epsilon, T > 0, c_\pm$ such that, for $z \in U(z_0)$, $$\langle N(u(z,t)), \phi_p \rangle > (\lambda_p + \epsilon) t + c_+\ , \quad \hbox{ for } t > \,\ T \ , \eqno{(V+)}$$ $$\langle N(u(z,t)), \phi_p \rangle > (\lambda_p - \epsilon) t + c_- \ , \quad \hbox{ for } t < - T \, . \eqno{(V-)}$$ Notice that the asymptotic behavior on each fiber is the same. Asymptotic geometry of fibers ----------------------------- Again, parameterize fibers as $u(z,t)=w(z,t) + t\phi_p$. Under mild hypotheses, the vectors $w(z,t)/t$ have a limit for $t \to \pm \infty$, which is independent of $z$. A version of this result was originally obtained by Podolak ([@P]). Suppose that $F: X \to Y$, $F = L - N$ satisfies hypothesis $(H)$ of Section \[sec:fibers\]. Suppose also that, for every $u\in X$, $$\lim_{t\to +\infty}\frac{PN(tu)}{t} = N_\infty(u) \in Y.$$ Then there exist $w_{+}, w_{-} \in H_X$ such that, for every fiber $u(z,t)=w(z,t) + t\phi_p$, $$\lim_{t\to +\infty}\\|\frac{w(z,t)}{t} - w_+\\|_X = 0 \ , \quad \lim_{t\to -\infty}\\|\frac{w(z,t)}{t} - w_-\\|_X = 0$$ which are respectively the unique solutions of the equations $$Lw - PN_\infty(w + \phi_p) = 0\ , \quad Lw + PN_\infty(-w - \phi_p) = 0.$$ It turns out that $N_\infty = PN_\infty$ satisfies the same Lipschitz bound that the functions $PN_t$ in Theorem \[theo:Fvv\], which is why both equations are (uniquely) solvable. Fibers are asymptotically vertical if and only if $\lim_{\|t\| \to \infty} w(z,t)/t = 0$, or equivalently, $PN_\infty(\pm \phi_p) =0$. Indeed, in this case, $w=0$ is the unique solution of both equations. This is what happens in the Ambrosetti-Prodi scenario, where $PN_\infty(u) = (b-\gamma) Pu^+ - (a - \gamma) Pu^-$ (recall $u = u^+ - u^-$), since $\phi_p=\phi_1 >0$. Comparing $F$ on fibers and on vertical lines --------------------------------------------- One might wish to relate the heights of $F$ along fibers and vertical lines, which are easier to handle. In [@P] Podolak presented a scenario in which this is possible. We state a version of her result for the case $t \to + \infty$. Let $X\subset Y$ be Hilbert spaces with $X$ dense in $Y$. Let $L:X \to Y$ be a self-adjoint operator with $0 \in \sigma(L)$, a simple, isolated eigenvalue, associated to the normalized kernel vector $\phi_p$. Set $H_Y = \langle \phi_p \rangle^{\perp}$. Take $N:Y \to Y$ and $F= L-N: X \to Y$ so that 1. $\\|N(u) - N(u_0)\\|_Y \leq \epsilon \\|u-u_0\\|_Y \ , \quad \lim_{t\to +\infty}N(tu)/t = N_\infty(u)$ 2. $\langle N_\infty(\phi_p),\phi_p \rangle = - \lim_{t\to +\infty} \langle F(t\phi_p),\phi_p\rangle/ t >0$ 3. $\epsilon \ \\|{ \big( L\|_{H_Y}} \big)^{-1}\\|< 1/2 \ , \quad \epsilon^2 \ \\|{ \big( L\|_{H_Y}} \big)^{-1}\\| < 1/2 \ \langle N_\infty(\phi_p),\phi_p \rangle $ . Then, for each fiber $(z_0,t)$ in adapted coordinates, $$\big\| \ \lim_{t\to +\infty}\frac{h^a(z_0,t)}{t} - \langle N_\infty(\phi_p),\phi_p \rangle \big\| \ < \langle N_\infty(\phi_p),\phi_p \rangle.$$ The number $\langle N_\infty(\phi_p),\phi_p \rangle$ gives the asymptotic behaviour of the height of $F$ along the vertical line through the origin. The theorem implies that $F$ along the upper part of each fiber converges to the same infinity that $F$ along $\{ t \phi_p, \ t \ge 0 \}$. A context in which these hypotheses apply is the Ambrosetti-Prodi operator with a piecewise nonlinearity $f(u) = (\lambda_p + c)u^+ - (\lambda_p - c)u^-$ for a sufficiently small number $c>0$. However, for pairs $(\lambda_p - c_1, \lambda_p + c_2), p \ne 1$ in the Fučik spectrum of the (Dirichlet) negative second derivative, for which necessarily $c_1 \ne c_2$ (near $\lambda_p$), the condition involving $\epsilon^2$ does not hold and indeed the thesis is not true. Fibers and the properness of $F$ {#subsec:proper} -------------------------------- From a more theoretical point of view, fibers circumvent the fundamental issue of deciding if $F$ is proper. For example ([@MST2]), the map $$F: C^1({{\mathbb{S}}}^1) \to C^0 ({{\mathbb{S}}}^1), \quad u \mapsto u ' + \arctan (u)$$ is a diffeomorphism from the domain to the open region between two parallel planes, $$\big\{ \ y \in C^0 ({{\mathbb{S}}}^1) \ , \ - \pi^2 < \int_0^{2\pi} y(\theta) d \theta < \pi^2 \ \big\} \ .$$ Indeed, fibers in this case are simply lines parallel to the vertical line of constant functions, and each is taken to such region. Perhaps, it would be more appropriate to think of fibers as a tool to show properness ([@CTZ2]). As far as we know, for the Ambrosetti-Prodi map $F: X \to Y$ in unbounded domains, the properness has been proved only by making use of fibers (see Section \[sec:examples\]). The map $F: X \to Y$ satisfying hypotheses $(H)$ of Section \[sec:fibers\] and $(V\pm)$ above is proper if and only if the restriction of $F$ to each fiber is proper. Points in the Fučik spectrum of the (Dirichlet) second derivative give rise to maps $F$ which take the half-fiber $\{ u(0,t), t \ge 0\}$ to a single point $0$ ([@TZ]), which shows that $F$ is not proper, although the image of every vertical line has its vertical component taken to infinity. A possible definition of a topological degree for $F$ becomes innocuous — the relevant information is essentially the asymptotic behavior of $F$ along each fiber. Singularities {#sec:singularities} ============= Generic singularities both of $F$ and of each height function are very special — they are Morin singularities. Morin classified generic singularities of functions from ${{\mathbb{R}}}^n$ to ${{\mathbb{R}}}^n$ whose derivative at the singularity has one dimensional kernel ([@M]). This is sufficient for the study of critical points of height functions on one dimensional fibers, by Proposition \[prop:Cc\]. In order to do the same for the critical points of the whole function $F:X \to Y$, we need an equivalent classification for singularities of functions between infinite-dimensional spaces, which is very similar ([@CDT], [@MST2], [@R]) — this is how we proceed next. Morin theory in adapted coordinates {#subsec:Morin} ------------------------------------ The first step in Morin’s proof makes use of the implicit function theorem to write such a singularity at a point $(z_0, t_0)$ in adapted coordinates, as in Section \[sec:adapted\]: $$F^a: Y = H_Y \oplus V_Y \to Y = H_Y \oplus V_Y, \quad (z,t) \mapsto (z, h^a(z,t)).$$ Say $F^a$ is $C^{k+1}$. The point $(z_0, t_0)$ is a [Morin singularity of order]{} $k$ if and only if 1. $D_t h^a(z_0, t_0) = \cdots = D_t^k h^a(z_0, t_0) = 0$, $D_t^{k+1} h^a(z_0, t_0) \ne 0$. 2. The Jacobian $D(h^a, D_t h^a, \ldots, D_t^{k-1} h^a)(z_0, t_0)$ has maximum rank. Then, in a neighborhood of $(z_0, t_0)$ there is an additional change of variables which converts $F^a$ to the normal form $$(\tilde z, x, t) \mapsto (\tilde z, x, t^{k+1} + x_1 t^{k-1} + \cdots + x_{k-1} t ) \ .$$ Here the coordinates $(\tilde z, x)$ correspond to an appropriate splitting of $Y = \tilde Y \oplus {{\mathbb{R}}}^{k-1}$. Morin singularities of order 1, 2, 3 and 4 are called, respectively, folds, cusps, swallowtails and butterflies. Thus, the classification of critical points of $F$ boils down to the study of a family of one dimensional maps, the height functions restricted on fibers. The first requirement is specific to each fiber (i.e., one checks it for every fixed $z$ near $z_0$), whereas the second relates nearby fibers, i.e., one has to change $z$. Folds are structurally simpler than deeper singularities: the behavior along fibers near a fold point is always the same — essentially like $t \mapsto -t^2$, whereas this is not the case for cusps, where close to $t \mapsto t^3$ one finds $t \mapsto t^3 \pm \epsilon t$. There is something unsatisfying in the fact that the relevant properties of the critical points of $F$ requires knowledge of some version of the height function. This is circumvented by the next result ([@CTZ2]). \[spectralsing\] Suppose $F: X \to Y$ is $C^{k+1}$ and admits one dimensional fibers. Then there is an open neighborhood $U$ of the critical set $C$ with the properties below. 1. There is a unique $C^k$ map $\lambda_p: U \to {{\mathbb{R}}}$ for which $\lambda_p = 0$ on $C$ and is an eigenvalue of $DF$ elsewhere. 2. There is a strictly positive $C^k$ function $p : U \to {{\mathbb{R}}}^+$ such that $$\lambda_p (u(z, t)) = p(u(z,t)) \ D_t h(u(z, t)) \ , \quad u(z,t) \in U \ .$$ A point $u_0 = u(z_0,t_0)$ is a Morin singularity of order $k$ of $F$ if and only if 1. $ \lambda_p(u_0) = \cdots = D_t^{k-1} \lambda_p(u_0) = 0$ , $D_t^{k} \lambda_p(u_0) \ne 0$ , 2. The image of $D(\lambda_p, \ldots, D_t^{k-2} \lambda_p)(u_0)$ together with $D_t \lambda_p(u_0)$ span ${{\mathbb{R}}}^n$. There is an analogous characterization in adapted coordinates. Critical points of the height function -------------------------------------- Consider a critical point $u_0 \in C \subset X$ and the fiber $u(z_0, t)$ through it, $u(z_0, t_0 ) = u_0$. From Proposition \[spectralsing\], $u_0$ is a (topological) fold of the height function $h$ restricted to the fiber if and only if $u_0$ is a topologically simple root of $\lambda_p(u)$ along the fiber, i.e., $\lambda_p$ is strictly negative on one side of $u_0$ and strictly positive on the other. Once we reduce the issue to checking an eigenvalue along a fiber, [derivatives are irrelevant]{}: just study the quadratic form of the Jacobian. Clearly, this only handles topological equivalence between the function and a fold. More explicitly, in standard Ambrosetti-Prodi contexts, $\lambda_1(u_0)$ is the minimum value of the quadratic form $\langle DF(u_0) v, v \rangle$. The derivative $D_t u(z_0,t_0)$ of the ($C^1$) fiber is the eigenfunction $\phi_1(u_0) >0$, and it is easy to check that $\lambda_1$ increases with $t$ by the convexity of the nonlinearity $f$. This should be compared with differentiability arguments, which require some estimate on $\phi_1(u_0)$ (say, boundedness). The fact that all critical points are local maxima for height functions on fibers, as required in Proposition \[prop:folds\], suggest hypotheses to be checked only on the critical set of $F$. This is not the case in the original Ambrosetti-Prodi theorem: the statement of the theorem has the merit that it makes no reference to the critical set at all, an object which in principle is hard to identify. The convexity of the nonlinearity handles the difficulty and, rather surprisingly, is essentially necessary ([@CTZ1]). Further examples yielding local maximality are somewhat contrived. Some examples {#sec:examples} ============= The non-autonomous case ----------------------- The geometric formulation $F = L - N$ is not sufficient to accomodate situations of the form $F(u(x)) = - \Delta u(x) + f(x, u(x))$, the so called non- autonomous case. Hammerstein ([@H]) had already considered homeomorphisms of that form. A possibility is requiring that $X$ and $Y$ are function spaces defined on a domain $\Omega$, so that the variable $x$ makes sense. The formalism above carries over to this scenario without surprises. More precisely, as usual $X$ and $Y$ are Hilbert spaces, $X$ dense in $Y$. The linear operator $L: X \subset Y \to Y$ is self-adjoint with a simple eigenvalue $\lambda_p$ associated to a normalized eigenvector $\phi_p$. Let $P: Y \to H_Y = \langle \phi_p \rangle^\perp $ be the orthogonal projection. From the nonlinear term $N: \Omega \times Y \to Y$, define as before $PN_t: H_Y \to H_Y, t \in {{\mathbb{R}}}$ and require a Lipschitz estimate, $$\\| PN_t(x, w_1) - PN_t(x,w_0) \\|_Y \le n \\|w_1 - w_0\\|_Y, \quad \hbox{ for } w_0, w_1 \in H_Y ,$$ so that $[-n,n] \cap \sigma(L) = \{ \lambda_p \}$, which is the same hypothesis $(H)$ in Section \[sec:fibers\]. This obtains fibers for $F: X \to Y$ as in Theorem \[theo:Fvv\], which satisfy the same properties as those in the autonomous case, in particular, Proposition \[prop:fibers\]. The hypothesis which obtain appropriate asymptotic behavior of $F$ along fibers are the obvious counterparts of $(V+)$ and $(V-)$ in Section \[subsec:asy\]. For the classification of critical points, we simply do not distinguish between the autonomous and non-autonomous case: the subject has become a geometric issue. Schrödinger operators on ${{\mathbb{R}}}^n$ {#subsec:physics} ------------------------------------------- As was surely known by Ambrosetti and Prodi (and [@B] is an interesting example), the Laplacian with Dirichlet conditions might be replaced by more general self-adjoint operators. The approach in this text is flexible enough to handle nonlinear perturbations of Schrödinger operators on unbounded domains yielding global folds. In our knowledge there are no similar results in the literature. Tehrani ([@Te]) obtained counting results for Schrödinger operators in ${{\mathbb{R}}}^n$ in the spirit of those obtained by Podolak ([@P]), indicated in Section \[subsec:podolak\] . We state the by now natural hypotheses. Here $Y = L^2({{\mathbb{R}}}^n)$. 1. The free operator $ T = -\Delta + v(x): X \subset Y \to Y$ is self-adjoint, with simple, isolated, smallest eigenvalue $\lambda_1$ and positive ground state $\phi_1$. 2. $F: X \subset Y \to Y, F(u) = Tu - f(u)$ is a $C^1$ map. 3. The function $f \in C^2({{\mathbb{R}}})$ satisfies $f(0)=0$, $M \geq f''>0$, $f'({{\mathbb{R}}})=(a,b)$ and $ a < \lambda_1 < b < \min\{ \sigma(T) \setminus \{ \lambda_1\} \}$. 4. The Jacobians $DF(u): X \to Y$ are self-adjoint operators with eigenpair $(\lambda_1(u),\phi_1(u))$ sharing the properties of $(\lambda_1,\phi_1)$. \[theo:APBS\] Under these hypotheses, the map $F:X \to Y$ is a global fold. Such hypotheses are satisfied for $v(x) = x^2/2$, the one dimensional quantum harmonic oscillator, as well as for the hydrogen atom in ${{\mathbb{R}}}^3$, for which $v(x) = -1/\|x\|$. Hypotheses on the potential of a Schrödinger operator in order to obtain such properties are commonly studied in mathematical physics. The interested reader might consider [@BS], [@LL], [@RS]. More about this in [@CTZ2]. Perturbations of compact operators {#subsec:compact} ---------------------------------- We recall Mandhyan’s second example of a global fold ([@M2]), or better, a special case of the extension given by Church and Timourian ([@CT1]). For $\Omega \subset {{\mathbb{R}}}^n$ a compact subset, let $X= C^0(\Omega)$ and define the compact operator $$K: X \to X, \quad K(u)(x) = \int_\Omega k(x,y) u(y) dy$$ where the kernel $k \in C^0(\Omega \times \Omega)$ is symmetric and positive. Let $\mu_1 > \mu_2$ be the largest eigenvalues of $K$. Now let $f: {{\mathbb{R}}}\to {{\mathbb{R}}}$ be a strictly convex $C^2$ function satisfying $$0 < \lim_{x \to - \infty} f' (x) < 1/\mu_1 < \lim_{x \to \infty} f' (x) < 1/\|\mu_2\| \ .$$ Under these hypotheses for $K$ and $f$, the map $$G: X \to X, \quad G(u)(x)= u(x) - K f(u(y))$$ is a global fold. This is the kind of nonlinear map obtained if one started from the Ambrosetti-Prodi original operator $F(u) = - \Delta u - f(u)$ and inverted the Laplacian. Actually, one could take another track: instead of inverting the linear part, one might consider the inversion of the nonlinear map $u \mapsto f(u)$, since $f' $ is bounded away from zero. For maps $G(u) = Ku - f(u)$ obtained this way, we handle the case when $K$ is a general compact symmetric operator $K$. More precisely, let $\Omega \subset {{\mathbb{R}}}^n$, $B= C^0(\Omega)$ and $ Y = L^2( \Omega)$. Let $K:B \to B$ and $K:Y\to Y$ be compact operators which preserve the cone of positive functions. Also, $K:Y\to Y$ has simple largest eigenvalue $\lambda_p = \\|K \\|$ and second largest eigenvalue $\lambda_s$ . Let $f: {{\mathbb{R}}}\to {{\mathbb{R}}}$ be a strictly convex $C^2$ function, with $f(0)=0$ if $\Omega$ is unbounded. Suppose $$\lambda_s < a = \lim_{t \to - \infty} f'(t) < \lambda_p < b = \lim_{t \to \infty} f'(t) \ .$$ \[theo:APC\] The map $ F: B \to B \ , F(u) = Ku - f(u) $ is a global fold. The reader should notice that $F$ is Lipschitz but not differentiable as a map from $L^2(\Omega)$ to itself. Still, the direct construction of fibers in $C^0(\Omega)$ is not a simple matter, because properness of $F$ is not immediate. Transplanting fibers in this example is convenient, and was also used in Mandhyan’s context. Folds as perturbations of non-self-adjoint operators {#subsec:nonself} ---------------------------------------------------- McKean and Scovel ([@McKS], [@CT1]) studied the Riccati-like map on functions $$u \in L^2([0,1]) \mapsto u + (D_2)^{-1} f(u) \in L^2([0,1]), \quad f(x) = x^2 / 2 ,$$ where $(D_2)^{-1}$ is the inverse of the second derivative acting on $W^{1,2}([0,1])$ and showed that the critical set consists of a countable union of (topological) hyperplanes. Church and Timourian ([@CT]) showed that the restriction of such map to a neighborhood of one specific critical component is (after global homeomorphic change of variables) a fold. The techniques employed are in the spirit of the original Ambrosetti-Prodi paper. Fibers were relevant in ([@MST2]), where perturbations of first order differential equations (clearly, non-self-adjoint operators) were shown to be global folds. An example is the map on periodic functions with (generic) convex nonlinearities $f$, $$F: C^1 ({{\mathbb{S}}}^1) \to C^0({{\mathbb{S}}}^1), \quad u \mapsto u ' + f(u) \ .$$ McKean and Scovel ([@McKS]) and Kappeler and Topalov ([@KT]) considered the same map among Sobolev spaces, the celebrated [Miura map]{}, used as a change of variables between the Korteweg-deVries equation and its so called modified version. More recently, a perturbation of a non-self-adjoint elliptic operator (as in [@BNV], but with Lipschitz boundary) has been shown to yield a global fold ([@STZ]). Acknowledgements ---------------- The first author thanks the Departamento de Matemática, PUC-Rio, for its warm hospitality. The second and third authors gratefully acknowledge support from CAPES, CNPq and FAPERJ. We thank José Cal Neto and Otavio Kaminski for the numerical examples. [\[10\]]{} Marta Calanchi, Dipartimento di Matematica, Università di Milano, Via Saldini 50, 20133 Milano, Italia Carlos Tomei and André Zaccur, Departamento de Matemática, PUC-Rio, R. Mq. de S. Vicente 225, Rio de Janeiro, RJ 22453-900, Brazil [email protected] [email protected] [email protected]	{ "pile_set_name": "ArXiv" }
--- abstract: 'We review the unification of early-time inflation with late-time acceleration in several local modified gravity models which pass Solar System and cosmological tests. It is also demonstrated that account of non-local gravitational corrections to the action does not destroy the possibility of such unification. Dark matter effect is caused by composite graviton degree of freedom in such models.' author: - 'Shin’ichi Nojiri' - 'Sergei D. Odintsov' title: 'Modified gravity as realistic candidate for dark energy, inflation and dark matter[^1]' --- [ address=[Department of Physics, Nagoya University, Nagoya 464-8602, Japan]{}, ]{} [ address=[Instituciò Catalana de Recerca i Estudis Avançats (ICREA) and Institut de Ciencies de l’Espai (IEEC-CSIC), Campus UAB, Facultat de Ciencies, Torre C5-Par-2a pl, E-08193 Bellaterra (Barcelona), Spain]{}, ]{} Modified gravity (for general review of different models, see [@review]) is known to be very interesting candidate for dark energy. From another side, there exist modified gravity theories which may naturally unify the early-time inflation and late-time acceleration [@prd]. Recently, there were suggested realistic theories [@prd1; @prd2] which pass the local as well as cosmological tests and unify inflation with late-time acceleration (with the presence of radiation/matter dominance epoch at intermediate universe). We will review such models which represent $F(R)$ gravity or its combination with non-local gravity below. We will start from the theory with the action: \[Uf0\] S= d\^4 x . Here $f(R)$ is an appropriate function of the scalar curvature $R$ and $L_m$ is the Lagrangian density of matter. In order to generate the inflation, one may require \[Uf1\] \_[R]{} f (R) = - \_i . Here $\Lambda_i$ is an effective cosmological constant at the early universe and therefore it is natural to assume $\Lambda_i \gg \left(10^{-33}{\rm eV}\right)^2$. For instance, it could be $\Lambda_i\sim 10^{20\sim38}\left( {\rm eV}\right)^2$. In order that the current cosmic acceleration could be generated, let us consider that currently $f(R)$ is a small constant, that is, \[Uf3\] f(R\_0)= - 2R\_0 ,f’(R\_0)\~0 . Here $R_0$ is the current curvature $R_0\sim \left(10^{-33}{\rm eV}\right)^2$. Note that $R_0> \tilde R_0$ due to the contribution from matter. In fact, if we can regard $f(R_0)$ as an effective cosmological constant, the effective Einstein equation gives $R_0=\tilde R_0 - \kappa^2 T_{\rm matter}$. Here $T_{\rm matter}$ is the trace of the matter energy-momentum tensor. We should note that $f'(R_0)$ need not vanish exactly. Since we are considering the time scale of one-ten billion years, we only require $\left\| f'(R_0) \right\| \ll \left(10^{-33}\,{\rm eV}\right)^4$. The last condition is \[Uf4\] \_[R0]{} f(R) = 0 , which means that there is a flat spacetime solution. Instead of the model corresponding to (\[Uf1\]), we may consider a model which satisfies \[UU2\] \_[R]{} f (R) = R\^m , with a positive integer $m>1$ and a constant $\alpha$. In order to avoid the anti-gravity $f'(R)>-1$, we find $\alpha>0$ and therefore $f(R)$ should be positive at the early universe. On the other hand, Eq.(\[Uf3\]) shows that $f(R)$ is negative at the present universe. Therefore $f(R)$ should cross zero in the past. At the early universe, $f(R)$-term in (\[UU2\]) dominates the Einstein-Hilbert term. Now let assume that there exists matter with an equation of state parameter $w$. For spatially flat FRW universe, the scale factor $a(t)$ behaves as \[UUU2\] a(t) t\^[h\_0]{} ,h\_0 . Then the effective equation of the state parameter $w_{\rm eff}$, which is defined by \[UUU3\] w\_[eff]{}= -1 + , can be less than $-1/3$ and the accelerating expansion could occur if $m$ is large enough although $w_{\rm eff} > -1$, which is quintessence type. Then the inflation could occur due to the $R^m$ behavior of $f(R)$ in (\[UU2\]). We may consider a model [@prd1] \[UU2d\] f(R)= . Here $\alpha$, $\beta$, and $\gamma$ are positive constants and $n$ is a positive integer. (\[UU2d\]) gives [@prd1] \[UUU7\] R\_0={ () (1+ )}\^[1/n]{} , and therefore \[UU6\] f(R\_0) \~-2 R\_0 = ( 1 + ) . Then we find \[UU9\] \~2 R\_0 R\_0\^[-2n]{} ,\~4 [R\_0]{}\^2 R\_0\^[-2n]{} R\_I\^[n-1]{} ,\~2 R\_0 R\_0\^[-2n]{} R\_I\^[n-1]{} . In general $F(R)$ gravity contains graviton and scalar as physical modes. The propagation of the scalar field might give a large correction to the Newton law. In the model (\[UU2d\]), the correction to the Newton law could be small since the mass $m_\sigma$ of the extra scalar filed is large and given by $m_\sigma^2 \sim 10^{-160 + 109 n}\,{\rm eV}^2$ and in the air on the earth, $m_\sigma^2 \sim 10^{-144 + 98 n}\,{\rm eV}^2$ [@prd1]. In both cases, the mass $m_\sigma$ is very large if $n\geq 2$. As a model corresponding to (\[Uf1\]), we may consider[@prd2] \[tan7\] f(R) &=& -\_0 ( () + ()) && -\_I ( () + ()) . We now assume \[tan8\] R\_IR\_0 ,\_I \_0 ,b\_I b\_0 , and \[tan8b\] b\_I R\_I 1 . When $R\to 0$ or $R\ll R_0,\, R_I$, $f(R)$ behaves as \[tan9\] f(R) - ( + )R . and $f(0)=0$ again. When $R\gg R_I$, it follows \[tan10\] f(R) - 2\_I && -\_0 ( 1 + ()) -\_I ( 1 + ()) &\~& -\_I ( 1 + ()) . On the other hand, when $R_0\ll R \ll R_I$, one gets \[tan11\] f(R) && -\_0 - &\~& -2\_0 -\_0 . Here we have assumed (\[tan8b\]). We also find \[tan12\] f’(R)= - - , which has two valleys when $R\sim R_0$ or $R\sim R_I$. When $R= R_0$, we obtain \[tan13\] f’(R\_0)= - \_0 b\_0 - > - \_I b\_I - \_0 b\_0 . On the other hand, when $R=R_I$, we get \[tan14\] f’(R\_I)= - \_I b\_I - > - \_I b\_I - \_0 b\_0 . Then, in order to avoid the anti-gravity period, one obtains \[tan15\] \_I b\_I + \_0 b\_0 < 2 . We now investigate the correction to the Newton law and the matter instability issue. In the solar system domain, on or inside the earth, where $R\gg R_0$, $f(R)$ can be approximated by \[tan16\] f(R) \~-2 \_[eff]{} + 2\^[-b(R-R\_0)]{} . On the other hand, since $R_0\ll R \ll R_I$, by assuming Eq. (\[tan8b\]), $f(R)$ in (\[tan7\]) could be also approximated by \[tan17\] f(R) \~-2 \_0 + 2\^[-b\_0(R-R\_0)]{} , which has the same expression, after having identified $\Lambda_0 = \Lambda_{\rm eff}$ and $b_0=b$. Then, we may check the case of (\[tan16\]) only. We find that the effective mass has the following form \[tan18\] m\_\^2 \~ , which could be very large, that is, $m_\sigma^2 \sim 10^{1,000}\,{\rm eV}^2$ in the solar system and $m_\sigma^2 \sim 10^{10,000,000,000}\,{\rm eV}^2$ in the air surrounding the earth, and the correction to Newton’s law can be made negligible. Hence, both models pass local tests and may unify the inflation with dark energy. Moreover, the arguments presented in ref.[@dm] show that such models may also describe dark matter consistently. The reason is caused by composite graviton degree of freedom. Thus, not only modification of gravitational potential occurs (this gravitational correction explains the galaxy rotation curves) but also such gravitational dark matter shows the particles-like properties as predicted by observational data. As third model we consider the non-local gravity [@non1; @non2; @koivisto; @non3; @non4], whose action is given by \[nl1\] S=d\^4 x { R(1 + f((\^2)\^[-1]{}R)) + [L]{}\_[matter]{} } . Here $f$ is some function and $\nabla^2$ is the d’Almbertian for scalar field. The above action can be rewritten by introducing two scalar fields $\phi$ and $\xi$ in the following form [@non2]: \[nl2\] S=d\^4 x . By the variation over $\xi$, we obtain $\nabla^2\phi=R$ or $\phi=\left(\nabla^2\right)^{-1}R$. Substituting the above equation into (\[nl2\]), one re-obtains (\[nl1\]). Especially in case \[NLdS2\] f()=f\_0 \^[b]{}= f\_0 \^[-2H\_0 ]{} , we obtain a de Sitter solution: \[NLdS1\] H=H\_0 ,= - 4H\_0 t ,= - \^[-4bH\_0 t]{} + \^[-3H\_0 t]{} +1 . We may discuss the accelerating early-time and late-time cosmology in the non-local gravity where $F(R)$-term is added (the appearance of local and non-local corrections is typical for string theory low-energy effective action). The starting action is: \[nl30\] S=d\^4 x { R(1 + f((\^2)\^[-1]{}R)) + F(R) + [L]{}\_[matter]{} } . Here $F(R)$ is some function of $R$. We may propose several scenarios. One is that the inflation at the early universe is generated mainly by $F(R)$ part but the current acceleration is defined mainly by $f\left(\left(\nabla^2\right)^{-1}R\right)$ part. One may consider the inverse, that is, the inflation is generated by $f\left(\left(\nabla^2\right)^{-1}R\right)$ part but the late-time acceleration by $F(R)$. For instance, for the first scenario one can take: $F(R)= \beta R^2$. Here $\beta$ is a constant. We choose $f(\left(\nabla^2\right)^{-1}R)$ part as in (\[NLdS2\]) but $f_0$ is taken to be very small and $\phi$ starts with $\phi=0$. Hence, at the early universe $f\left(\left(\nabla^2\right)^{-1}R\right)$ is very small and could be neglected. Then due to the $F(R)$-term, there occurs (slightly modified) $R^2$-inflation. After the end of the inflation, there occurs the radiation/matter dominance era. In this phase, $f(\phi)$ becomes large as time goes by and finally this term dominates. As a result, deSitter expansion occurs at the present universe. For the second scenario, the early-time inflation is generated by $f\left(\left(\nabla^2\right)^{-1}R\right)$ part but the cosmic acceleration is generated by $F(R)$. As an $F(R)$-term, one can take the model [@HS]: \[HS1\] F\_[HS]{}(R)=- , The estimation of ref.[@HS] suggests that $R/m^2$ is not so small but rather large even at the present universe and $R/m^2\sim 41$. Hence, $F_{HS}(R)\sim - \frac{m^2 c_1}{c_2} + \frac{m^2 c_1}{c_2^2} \left(\frac{R}{m^2}\right)^{-n}$, which gives an “effective” cosmological constant $-m^2 c_1/c_2$ and generates the late-time accelerating expansion. One can show that \[HSbb1\] H\^2 \~ \~(70 [km/spc]{})\^2 \~(10\^[-33]{}[eV]{})\^2 . At the intermediate epoch, where the matter density $\rho$ is larger than the effective cosmological constant, $\rho > \frac{m^2 c_1}{c_2}$, there appears the matter dominated phase and the universe expands with deceleration. Hence, above model describes the effective $\Lambda$CDM cosmology. As a $f\left(\left(\nabla^2\right)^{-1}R\right)$ part, we consider theory (\[NLdS2\]) with $b=1/2$, again. It is assumed $f_0$ is large and $f\left(\left(\nabla^2\right)^{-1}R\right)$ term could be dominant at the early universe. Hence, following (\[NLdS1\]), $\phi$ becomes negative and large as time goes by and therefore $f(\phi)$ becomes small and could be neglected at late universe. Then there appears naturally the radiation/matter dominated phase. After that due to $F_{HS}(R)$-term (\[HS1\]), the late-time acceleration occurs. Thus, the possibility of consistent description of the universe expansion history from the early-time inflation till late-time acceleration in modified gravity (without/with non-local term) is demonstrated. As related gravitational phenomenon the appearance of dark matter is naturally explained in such models. The work by S.D.O. was supported in part by MEC (Spain) projects FIS2006-02842 and PIE2007-50I023, RFBR grant 06-01-00609 and LRSS project N.2553.2008.2. The work by S.N. is supported in part by the Ministry of Education, Science, Sports and Culture of Japan under grant no.18549001 and Global COE Program of Nagoya University provided by the Japan Society for the Promotion of Science (G07). [9]{} S. Nojiri and S. D. Odintsov, arXiv:hep-th/0601213. S. Nojiri and S. D. Odintsov, Phys. Rev. D [68]{}, 123512 (2003) \[arXiv:hep-th 0307288\]. S. Nojiri and S. D. Odintsov, Phys. Rev. D [77]{}, 026007 (2008) \[arXiv:0710.1738\[hep-th\]\]. G. Cognola, E. Elizalde, S. Nojiri, S. D. Odintsov, L. Sebastiani and S. Zerbini, Phys. Rev. D [77]{}, 046009 (2008) \[arXiv:0712.4017\[hep-th\]\]. S. Nojiri and S. D. Odintsov, arXiv:0807.0685\[hep-th\]; arXiv:0801.4843\[astro-ph\]. W. Hu and I. Sawicki, arXiv:0705.1158; Y. Song, H. Peiris and W. Hu, arXiv:0706.2399; S. Capozziello, M. De Laurentis, S. Nojiri and S. D. Odintsov, arXiv:0808.1335\[hep-th\]. S. Deser and R. Woodard, Phys. Rev. Lett. [99]{}, 111301 (2007). S. Nojiri and S. D. Odintsov, Phys. Lett. B [659]{}, 821 (2008) \[arXiv:0708.0924 \[hep-th\]\]. T. Koivisto, Phys. Rev. D [77]{}, 123513 (2008); \[arXiv:0807.3778\[gr-qc\]\]. S. Jhingan, S. Nojiri, S. D. Odintsov, M. Sami, I. Thongkool and S. Zerbini, Phys. Lett. B [663]{}, 424 (2008) \[arXiv:0803.2613 \[hep-th\]\]. S. Capozziello, E. Elizalde, S. Nojiri and S.D. Odintsov, \[arXiv:0809.1535\[hep-th\]\]. [^1]: Based on the talk given at DSU08 conference, June 2008.	{ "pile_set_name": "ArXiv" }
--- abstract: 'Based on the theoretical framework of generalized eikonal approximation we study the two-nucleon emission reactions in high $Q^2$ electro-disintegration of $^3He$. Main aim is to investigate those features of the reaction which can be unambiguously identified with the short range properties of the ground state nuclear wave function. To evaluate the differential cross section we work in the formalism of the decay function which characterizes the property of the ground state wave function as well as the decay properties of final two nucleon spectator system. Our main motivation here is to explore the accessibility of two– and three–nucleon short range correlations in $^3$He as well as to isolate unambiguously single and double rescattering processes in the reaction dynamics. Our analysis allowed us also to identify new approaches for investigating the role of the practically unknown three-nucleon forces in the ground state wave function of $^3He$.' address: - 'Department of Physics, Florida International University, Miami, Florida 33199' - ' Department of Physics, Pennsylvania State University, University Park, PA ' - 'Department of Nuclear Physics, Tel Aviv University, Tel Aviv, Israel' author: - 'M.M. Sargsian, T.V. Abrahamyan' - 'M.I. Strikman' - 'L.L. Frankfurt' title: \| Exclusive Electro-Disintegration of [$^3$]{}He at high [$\boldmath Q^2$:\ II. Decay Function Formalism]{} --- Introduction {#I} ============ In this work we study high $Q^2$ ($4\gtrsim Q^2\gtrsim 1$ GeV$^2$) exclusive $^3He(e,e'NN)N$ reactions in which one nucleon in the final state can be clearly identified as a knocked-out nucleon which carries practically all of the momentum of the virtual photon. In the part-I of the present work[@SASF1] we calculated the scattering amplitude of this type of reactions within the generalized eikonal approximation (GEA), in which one expresses the scattering amplitude through the sum of the diagrams corresponding to the $n$’th order rescattering of the knocked-out nucleon with the residual nucleons in the nucleus. In Ref.[@SASF1] we evaluated each diagram based on the effective Feynman diagram rules derived within the GEA[@ggraphs; @treview]. The manifestly covariant nature of Feynman diagrams allowed us to preserve both the relativistic dynamics and the kinematics of the rescattering while identifying the low momentum nuclear part of the amplitude with the nonrelativistic nuclear wave function. Such an approach allows us to account for the internal motion of residual target nucleons in the rescattering amplitude as well as finite excitation energies of residual nuclear system. These features of the GEA are crucial in describing electro-production reactions aimed at the study of short-range nuclear properties since these configurations are characterized by non-negligible values of bound nucleon momenta and excitation energies. The study of short-range nucleon correlations is the main goals of the present work. With short-range correlations we identify those interactions between bound nucleons that generate nucleon momenta exceeding the characteristic Fermi momentum of the nuclear system, $k_F\approx 250$ MeV/c. This encompasses interactions due to the short-distance repulsive core and the short to medium-distance tensor interactions of NN system as well as possible three-nucleon interactions which can have both short and medium distance terms. Working in the virtual nucleon framework[@SASF1] one describes the reaction in the Lab frame relating all non-nucleonic degrees of freedom effectively to the off-shellness of the knocked–out (virtual) nucleon in the nucleus. For recent analysis of similar reaction with two proton emission in the final state see Ref.[@CK]. In calculation of the differential cross section of the reaction we utilize the kinematics of high $Q^2$ quasielastic scattering with knocked-out nucleon identified unambiguously in the final state of the reaction. This allows us to employ the decay function formalism in which the $eA$ cross section of the reaction can be represented through the convolution of the off-shell electron–bound nucleon cross section and the decay function which characterizes the ground state properties of the nucleus as well as the decay of the spectator state in the final state of the reaction. The paper is organized as follows: In Sec. \[II\] we summarize briefly the reaction, specifics of the kinematics and general form of the differential cross section. In Sec.\[III\] we elaborate the decay function formalism and derive the expression for the decay function both for plane wave and distorted wave impulse approximations. For the latter case one obtains the expression based on GEA calculation of Ref.[@SASF1]. In Sec. \[IV\] the numerical analysis of both semi-inclusive $(e,e'N)$ and exclusive $(e,e'NN)$ reactions are presented. In these calculations as an input we use Bochum group’s calculation of ground state $^3He$ wave function[@Bochum], SAID group’s parameterization of low-to-intermediate energy NN scattering amplitudes[@SAID] as well as our updated parameterization of the high energy small angle NN scattering amplitude[@tabra]. The main focus in the numerical analysis is the study of two- and three- nucleon short range correlations (SRCs) and isolation of the effects associated with single and double rescattering of the knocked-out nucleon off residual nucleons in the nucleus. We observe significant sensitivity of the decay function to the configurations characteristic to the short-range two- and three-nucleon correlations in the nucleus. The $(e,e'NN)$ reaction provides also an ideal testing ground for single and double rescattering processes, which could play a crucial role for studies of color coherence phenomena in hard exclusive nuclear reactions. Our calculations allowed us also to identify kinematics which could provide unprecedented access to the three nucleon forces in $^3He$. Section \[V\] summarizes our results. Reaction, Kinematics and Cross Section {#II} ====================================== We are studying the reaction: $$e + ^3He \rightarrow e' + N_f + N_{r2}+ N_{r3} \label{reaction}$$ where $e$ and $e'$ are the initial and scattered electrons with four-momenta $k_{e}$ and $k^\prime_{e}$ respectively. The $^3$He nucleus has a four-momentum $P_{A}$. $N_f$, $N_{r2}$ and $N_{r3}$ correspond to knocked-out and two recoil nucleons with four-momenta $p_f$, $p_{r2}$ and $p_{r3}$ respectively. We define also the four-momentum of the virtual photon $q=(q_0,{\bf q},0_\perp)\equiv k_{e}-k_{e'}$ with $Q^2=-q^2$. The $z$ direction is chosen parallel to $\bf q$ and the scattering plane is the plane of the $\bf q$ and ${\bf k_{e'}}$ vectors. The missing momentum of the reaction is defined as: ${\bf p_m} = {\bf p_f} - {\bf q}$. In numerical estimations we will focus on the kinematic region: $$\begin{aligned} \mbox{(a)} \ 4 \ge Q^2\ge 1 \mbox{GeV}^2; \ \ \ \ \mbox{(b)} \ {\bf p_{f}}\approx {\bf q}, \ \ \ \ %\mbox{(c)} \ \|{\bf p_m}\|,\|{\bf p_{r2}}\|, \|{\bf p_{r3}}\| \le 400-500 \mbox{MeV/c}; \ \ \ \ \label{kin} \end{aligned}$$ which allows us to identify $N_f$ as a knocked-out nucleon. Furthermore, we will do numerical investigations for recoil, ${\bf p_{r2}}$, ${\bf p_{r3}}$ and missing ${\bf p_m}$ momenta covering the region of up to $700-800$ MeV/c (and $1~GeV/c$ for the kinematics dominated by the double rescattering). This exceeds the kinematic domain of $\le 400-500~MeV/c$, within which the GEA and virtual nucleon approximation can be reliably applied[@SASF1]. However, the expectation of the smooth transition to the relativistic region for the main properties of the decay function (see e.g. [@FS88]) gives some validity to our exploratory studies in the region of $\ge 500$ MeV/c. The properties we consider are the correlation relations between missing momentum/energy and momenta of the recoil nucleons, relative strength of 2N and 3N correlations as well as signatures identified with the single and double rescattering of knocked-out nucleon off spectator nucleons in the target. Cross Section {#IIb} ------------- In numerical calculations we restrict ourselves by consideration of unpolarized three- body electro-disintegration reaction of Eq.(\[reaction\]). The differential cross section of reaction (\[reaction\]) in which no polarizations are fixed is given by $$\begin{aligned} d^{12}\sigma&=&\frac{1}{4j_A}(2\pi)^{4}\delta^{4}(k_{e}+P_{A}-k_{e} ^{\prime}-p_{f}-p_{r2}-p_{r3})\frac{1}{4} A\cdot \sum\limits_{nucleons} \sum\limits_{spins} \|M_{fi}\|^2 \nonumber\\ &&\frac{d^3 k_{e}^{\prime}}{(2 \pi)^3 2E_{e}^{\prime}} \frac{d^3 p_{f}}{(2 \pi)^3 2E_{f}} \frac{d^3 p_{r2}}{(2 \pi)^3 2E_{r2}} \frac{d^3 p_{r3}}{(2 \pi)^3 2E_{r3}} \ , \label{cs1}\end{aligned}$$ where $j_A = \sqrt{(k_eP_A)^2-m_e^2M_A^2}$. Here we sum over final and average over initial spins. The factor 1/4 comes from the averaging over the initial polarizations of the electron and the nucleus. Since one of the recoil nucleons is not observed, one eliminates this degree of freedom by integrating over $d^3 p_{r3}$. Thus integrated differential cross section is $$\begin{aligned} d^{9}\sigma&=&\frac{1}{4j_A}(2\pi)^{4} \delta(E_{e}+M_{A}-E_{e}^{\prime}-E_{f}-E_{r2}-E_{r3})\frac{1}{4} A\cdot \sum\limits_{nucleons}\sum_{s_{e},s_{A}} \sum_{s_{e^{\prime}},s_{f},s_{r2},s_{r3}}\|M_{fi}\|^2 \nonumber\\ &&\frac{d^3 k_{e}^{\prime}}{(2 \pi)^3 2E_{e}^{\prime}} \frac{d^3 p_{f}}{(2 \pi)^3 2E_{f}} \frac{d^3 p_{r2}}{(2 \pi)^3 2E_{r2}} \frac{1}{(2 \pi)^3 2E_{r3}} \ , \label{cs2}\end{aligned}$$ where ${\bf p_{r3}}={\bf k_{e}}-{\bf k_{e}^{\prime}}-{\bf p_{f}}-{\bf p_{r2}}$. In Eqs. (\[cs1\]) and (\[cs2\]) the transition matrix, $M_{fi}$, represents the convolution of the electron and nuclear currents, in which the nuclear current within GEA[@SASF1] represents the sum of the IA, single and double rescattering amplitudes, $$M_{fi} = -4\pi\alpha {1\over q^2}j^e_{\mu} \cdot\left(A_{0}^{\mu} + A_{1}^{\mu} + A_{2}^{\mu}\right), \label{m}$$ where $A_{0}$, $A_{1}$ and $A_{2}$ are derived in Ref.[@SASF1] and represent impulse approximation, single and double rescattering amplitudes respectively. Their final expressions are given in Eqs.(11), (21) and (27) of Ref.[@SASF1] respectively. Note that IA corresponds to the impulse approximation contribution in which the pair distortion effects due to interaction of two outgoing recoil nucleons are taken into account. Nuclear Decay Function Formalism {#III} ================================ In description of semi-exclusive $(e,e'N)$ reactions it is conventional to introduce a spectral function, $S^N(E_m,{\bf p_m})$, which in the impulse approximation picture describes the probability of finding a struck-nucleon in the nucleus initially having missing momentum ${\bf p_m}$ and missing energy $E_m= q_0 - T_f - {p_m^2\over 2(A-1)m}$. The missing energy characterizes the excitation energy of the final $A-1$ system. We generalize this approach for the case of the exclusive reaction in (\[reaction\]), as well as for any semi-exclusive reactions involving two-nucleon emission by introducing a nuclear decay function which can be formally defined as[@FS88]: $$D^N({\bf p_m},E_m,{\bf p_{r2}}) = \sum\limits_{f}\left\| \langle A-1\| a^\dagger({\bf p_{r2}})a({\bf p_m}) \|\Psi_{A}\rangle \right\|^2\delta(E_m-(\epsilon^f_{A-2}+T_{r2}-\epsilon_A- {p_m^2\over 2(A-1)m})), \label{dfunction}$$ where one sums over the states, $f$, of $A-2$ residual nucleus and $\epsilon^f_{A-2} = E^f_{A-2}-(A-2)m$, where $E^f_{A-2}$ is the total energy of $A-2$ state. It follows from this definition that the decay function characterizes the joint probability to find a struck nucleon in the nucleus with momentum ${\bf p_m}$, missing energy $E_{m}$ and the recoil nucleon with momentum ${\bf p_{r2}}$ in the decay product of the residual $A-1$ nucleus. Note that for $A\ge 4$, Eq.(\[dfunction\]) assumes the sum over all possible (bound and continuum) $(A-2)$ substates, provided the total excitation energy of $A-1$ final state is $E_m$. The general properties of the decay function are discussed in [@FS88] where the qualitative features are studied within the two-nucleon correlation model. The following sum rule and normalization condition follow from Eq.(\[dfunction\]): $$\begin{aligned} & & \int d^3 p_{r2} D^N({\bf p_m},E_m,{\bf p_{r2}}) = S^N(E_m,{\bf p_m}); \nonumber \\ & & \sum\limits_{N}\int dE_m d^3p_{m} S^N(E_m,{\bf p_m}) = 1. \label{srule}\end{aligned}$$ Within the impulse approximation the cross section of the reaction with two-nucleon emission can be expressed through the decay function as follows: $${d\sigma \over dE'_ed\Omega'_e d^3p_f d^3p_{r2}} = {j_N\over j_A}\cdot A\sum\limits_{N} \sigma_{eN}(p_f,k_e,k'_e)\cdot D^N({\bf p_m},E_m,{\bf p_{r2}}), \label{crsIA}$$ where $E_m = T_{r2}+T_{r3}+\|\epsilon_A\| - T_{A-1}$, with $T_{A-1}$ being the kinetic energy of the center mass of the residual $A-1$ system. $j_N$ is the flux calculated for the moving nucleon with momentum ${\bf p_m}$, and $\sigma_{eN}$ represents the cross section of electron - off-shell nucleon scattering[^1]. Using the derivations from Sec.IIIa of Ref.[@SASF1] it is straightforward to calculate the decay function of reaction (\[reaction\]). According to Eq.(11) of Ref.[@SASF1] and Eq.(\[cs2\]) one obtains: $$\begin{aligned} & & D^N({\bf p_m},E_m,{\bf p_{r2}},t_{r2}) = {1\over 2}\sum\limits_{s_A,s_{m},s_{r2},s_{r3}} \mbox{\huge $\vert$} \sum_{s_2,s_3}\sum\limits_{t_2,t_3}\int d^3 p_{23} \Psi_{NN}^{\dagger p_{r23},s_{r2},t_{r2};s_{r3},t_{r3}}(p_{23},s_2,t_2;s_3,t_3) \nonumber \\ & & \times\Psi^{s_A}_{A}(p_{m},s_m,t_m;p_{2},s_2,t_2;p_{3},s_3,t_3)\mbox{\huge $\vert$}^2 \cdot\delta(E_m - T_{r2}-T_{r3}-\|\epsilon_{A}\|+{p_m^2\over 4m}). \label{Dia}\end{aligned}$$ Note that hereafter we use the ground state wave $\Psi^{s_A}$ normalized to A. To understand the role of the pair distortion in the final state of the residual nucleus we will compare $D^N$ with the PWIA version of the decay function ($D^N_{PWIA}$), which corresponds to retaining the plane wave part of the residual $NN$ system’s wave function, $\Psi_{NN}$ in Eq.(12) of Ref.[@SASF1]. We also can generalize the definition of the decay function to include the single and double rescatterings of the knock-out nucleon with residual nucleons. Such a generalization usually is meaningful within an approximation in which electromagnetic current of knocked-out nucleon is factorized from rescattering integrals of Eqs.(21) and (27) of Ref.[@SASF1]. Two conditions should be met in order for this factorization to be justified: - First, one should be able to neglect the charge-exchange part of the amplitude for high energy small angle $NN$ rescatterings. This follows, first, from the fact that at sufficiently high energies the energy dependence of the small angle hadronic scattering amplitude scales as $\sim s_{NN}^{J-1}$, where $J$ corresponds to the spin of the exchanged particle [@Azimov]. Since, at energies considered, charge-exchange requires predominantly spin-0 exchange in the $t$-channel as compared to spin-one exchange in the diagonal channel, one expects strong $s$ suppression of the charge-exchange amplitude as compared to the elastic $NN$ amplitude. Analysis of the existing data on NN scattering demonstrates (see e.g. [@GibbsLo]) that starting at $Q^2\ge 2$ GeV$^2$ the charge-exchange part amounts only a third of the forward scattering $NN$ amplitude and decreases linearly with an increase of $Q^2$. Moreover, this contribution is further suppressed due to the fact that the charge-exchange amplitude is predominantly real and does not interfere with the dominant part of the FSI amplitudes which are predominantly imaginary. Taking into account also the larger slope factor of the $t$ dependence of charge-exchange amplitude as compared to the forward scattering, one estimates only a few percent overall contribution to the $NN$ rescattering amplitude at $Q^2\ge 2$ GeV$^2$. This contribution decreases sharply with an increase of $Q^2$. - Secondly, the transfered momentum, $q$, should be large enough that one can neglect the $k$, $k_2$ and $k_3$ dependencies in the electromagnetic current in Eqs.(21) and (27) of Ref.[@SASF1]. Such an approximation is justified by the kinematic condition of Eq.(\[kin\]) and by the fact that the characteristic average momenta transferred during $NN$ rescattering are restricted by the slope of the exponent in Eq.(A2) of Ref.[@SASF1], i.e. $\langle k^2\rangle, \langle k_2^2\rangle, \langle k_3^2\rangle \le {2\over B}\sim 250$ (MeV/c)$^2$. Keeping only the diagonal part in the $NN$ rescattering amplitude and factorizing the electromagnetic current of knocked-out nucleon, we arrive at an expression similar to Eq.(\[crsIA\]) in which we refer $D$ as a decay function calculated within the distorted wave impulse approximation (DWIA), $D_{DWIA}$. Based on Eqs.(11,21,27) of Ref.[@SASF1], for $D_{DWIA}$ one obtains: $$\begin{aligned} & & D^N_{DWIA}(Q^2, {\bf p_m},E_m,{\bf p_{r2}},t_{r2}) = {1\over 2}\sum\limits_{s_A,s_{m},s_{r2},s_{r3}} \nonumber \\ & & \mbox{\huge $\vert$} % IA term \sum\limits_{s_2,s_3}\sum\limits_{t_2,t_3}\left\{ \int d^3 p_{23} \Psi_{NN}^{\dagger p_{r23},s_{r2},t_{r2};s_{r3},t_{r3}}(p_{23},s_2,t_2;s_3,t_3) \Psi^{s_A}_{A}(p_{m},s_m,t_f;p_{2},s_2,t_2;p_{3},s_3,t_3) \right. \nonumber \\ % FSI1 term & & -{1\over 2} \int {d^3 k d^3 p_{23} \over (2 \pi)^3} \Psi_{NN}^{\dagger p_{r23},s_{r2},t_{r2};s_{r3},t_{r3}}(p_{23},s_{2},t_2;s_3,t_3) {1\over \Delta^0-k_z+i\varepsilon} \nonumber \\ & & \times \left[ \chi_1(s^{NN}_{2})f^{t_f,t_{2}\|t_f,t_2}_{NN}(k_{\perp}) \cdot \Psi^{s_A}_{A}(p_{m}+k,s_m,t_f;-{p_m\over 2} + p_{23}-k,s_2,t_2; -{p_m\over 2} - p_{23},s_3,t_3) \right. \nonumber \\ & & \left. \ \ + \chi_1(s^{NN}_{3})f^{t_f,t_{3}\|t_f,t_3}_{NN}(k_{\perp})\cdot \Psi^{s_A}_{A}(p_{m}+k,s_m,t_f;-{p_m\over 2} + p_{23},s_2,t_2; -{p_m\over 2} - p_{23}-k,s_3,t_3) \right] \nonumber \\ % FSI2 term & & + {1\over 4}\int d^3 p'_{23} {d^3k_3\over (2\pi)^3}{d^3 k_2 \over (2 \pi)^3} \Psi_{NN}^{\dagger p_{r23},s_{r2},t_{r2};s_{r3},t_{r3}}(p'_{23},s_{2},t_2;s_{3},t_3) \cdot f^{t_3,t_{f}\|t_3,t_f}_{NN}(k_{3\perp})f^{t_2,t_{f}\|t_2,t_f}_{NN}(k_{2\perp})\nonumber \\ & & \times \left[ {\chi_2(s^{NN}_{b3})\over \Delta_3 - k_{3z} + i\varepsilon} {\chi_1(s^{NN}_{a2})\over \Delta^0 -k_{2z}-k_{3z}+i\varepsilon} + {\chi_2(s^{NN}_{b2})\over \Delta_2 - k_{2z} + i\varepsilon} {\chi_1(s^{NN}_{a3})\over \Delta^0 -k_{2z}-k_{3z}+i\varepsilon} \right] \nonumber \\ & & \times \left. \Psi^{s_A}_{A}(p_{m}+k_3+k_2,s_m,t_f;-{p_m\over 2}-k_2+p'_{23},s_2,t_2; -{p_m\over 2}-k_3 - p'_{23},s_3,t_3) \right \} % \mbox{\huge $\vert$} ^2 \nonumber \\ % delta function & & \times\delta(E_m - T_{r2}-T_{r3}-\|\epsilon_{A}\|+{p_m^2\over 4m}). \label{Ddwia}\end{aligned}$$ Further in the text we will refer $D_{DWIA}$ also as $D_{IA+FSI}$ [^2]. Numerical Results {#IV} ================= In our numerical analysis we use the following inputs: - [Ground State Wave function:]{} We use the $^3He$ ground state wave functions calculated by the Bochum group[@Bochum] solving Faddeev equation for different sets of realistic $NN$ potentials. Authors in Ref.[@Bochum] consider also the different models of 3N forces with the main motivation to describe the binding energy of the $A=3$ system. - [Pair Distortion:]{} To estimate the reinteraction of two outgoing slow nucleons in Eq.(12) of Ref.[@SASF1] we use the SAID group’s parameterization of the NN scattering amplitudes based on the partial wave analysis of the $NN$ scattering data[@SAID]. This parameterization successfully describes the $NN$ scattering data at a low to intermediate momentum range ($p_{LAB}\le 1.3/3$ GeV/c for $pn/pp$ scattering). - [FSI:]{} To estimate the single and double rescattering contributions in the total scattering amplitude, we use the following conventional parameterization for high energy small angle $NN$ scattering: $$f^{NN} = \sigma_{tot}^{NN}(s)(i+\alpha^{NN}(s))e^{{B^{NN}(s)\over 2}t}\delta^{h_1,h^\prime_1} \delta^{h_2,h^\prime_2}, \label{fnnp}$$ where we parameterized[@tabra] $\sigma_{tot}^{NN}$, $\alpha^{NN}(s)$ and $B^{NN}(s)$ for both $pn$ and $pp$ scattering. We use practically all published data on small angle nucleon-nucleon scattering. In performing the numerical integrations in Eqs.(21) and (27) of Ref.[@SASF1] we represent the knocked-out nucleon propagator in the integral as a sum of the pole and principal value (P.V.) terms $${1\over \Delta-k_{z}+i\epsilon}=-i\pi \delta(\Delta-k_{z}) + P.V. {1\over \Delta - k_z}, \label{polandPV}$$ where $k_z$ and $\Delta$ characterize the transfered longitudinal momenta due to rescattering. Note that strictly speaking one can use the parameterization of the free (on-shell) NN scattering amplitude in Eq.(\[fnnp\]) for the pole term of Eq.(\[polandPV\]) only. For the P.V. term, the NN rescattering amplitude is off-shell. However, we use the parameterization in Eq.(\[fnnp\]) for P.V. term too since, in the kinematics of interest, the P.V. term is only a small correction as compared to the pole contribution of Eq.(\[polandPV\]). In the following numerical analyzes our main goal is to identify the strategies which are best suited for investigation of two- and three- nucleon short range correlations. For [two nucleon]{} correlations we are particularly interested in isolation and studies of the following configurations: - [Type 2N-I]{} correlations from which a nucleon is knocked-out by virtual photon, while the third nucleon moves in the mean field of 2N SRC Fig.\[2Nsrc\](a). For the idealized case of third nucleon at rest one obtains a correlation relation: $$E_{m}^{(2N-I)} = \sqrt{m^2+p_m^2} - m - T_{A-1}, %\approx \sqrt{m^2+p_m^2} - m- {p_m^2\over 4m} \label{empm_2N-I}$$ where $T_{A-1}=\sqrt{4m^2+p_m^2}-2m$. Note that although we apply the approximation, $p^2/m^2\ll 1$ in calculation of the scattering amplitudes, we still use the relativistic form of the kinetic energy to preserve the relativistic kinematics. - [Type 2N-II]{} correlations in which virtual photon strikes a third isolated nucleon while 2N SRC breaks up at the final state of the reaction, Fig.\[2Nsrc\](b). In this case the expected correlation between $E_m$ and $\bf {p_{r2}}\approx - {\bf p_{r3}}$ for the case of $p_m\approx 0$ is: $$E_{m}^{(2N-II)} = \sqrt{m^2+p_{r2}^2} + \sqrt{m^2+p_{r3}^2} - 2m \label{empm_2N-II}$$ For [three nucleon]{} correlations we focus on two particular situations: - [Type 3N-I]{} correlations, in which case the high missing momentum of the knocked-out nucleon is balanced by two nucleons which share almost equally the missing momentum $p_m$, Fig.\[3Nsrc\](a). This corresponds to the minimal missing mass of recoil 2N system with missing energy: $$E_m^{(2N-I)} \approx \|\epsilon_A\|. \label{empm_3N-I}$$ - [Type 3N-II]{} correlations, in which case the nucleon is knocked-out from the symmetric configurations where all three nucleons have the same momentum $p_m$, Fig.\[3Nsrc\](b). This corresponds to a significantly higher value of $E_m$ as compared to the the above discussed 2N and 3N SRC cases: $$E_{m}^{(3N-II)} = 2\sqrt{m^2+p_{m}^2} - 2m - T_{A-1}. %\approx 2\sqrt{m^2+p_m^2} - 2m- {p_m^2\over 4m}. \label{empm_3N-II}$$ Spectral Function ----------------- We start with the calculation of the spectral function as it is defined in Eq.(\[srule\]). This quantity is relevant primarily for semi-exclusive $(e,e'N)$ reactions, in which only the scattered electron and knocked-out nucleon are registered. Figure \[spectral\] shows the spectral functions for the cases of knocked-out proton (a) and knocked-out neutron (b) calculated with the $^3He$ ground state wave function based on the potential[@Bochum; @Urbana]. The hatched surface represents the PWIA prediction while dotted contours show the effect of pair distortion. The correlation feature of Eq.(\[empm\_2N-I\]) is reflected in the emergence of the broad peak in $E_m$ distribution at $p_m\ge 300$ MeV/c, while the signature of 2N-II correlations is seen in the minimum of the $E_m$ distribution at $p_m\approx 0$ for the case of knocked-out neutron with two recoil protons. This minimum reflects the fact that the relative momentum distribution in the $pp$ pair has a node in the $S$-state at $\approx 420$ MeV/c. These results are in agreement with previous analyses of spectral function[@Ciofi]. The emergence of these correlations at $p_m\ge 300$ GeV/c are clearly seen in Fig.(\[spectralsrc\]), in which solid and doted lines represent calculations within PWIA and IA approximations respectively. The positions of the peaks can be related to the dominance of type 2N-I SRCs in the high missing momentum part of the nuclear wave function [@FS81; @CSFS91] according to Eq.(\[empm\_2N-I\]) (i.e. $E_{m}^{Peak}\approx E_{m}^{2N-I}$). The solid arrows in the figure correspond to the prediction of Eq.(\[empm\_2N-I\]) corresponding to the scattering off a quasi-free and stationary two-nucleon correlation. This situation is reflected in the fact that, at large nucleon momenta $p_m$, the spectral function has similar functional dependence on $p_m$ as the stationary two nucleon correlation [@FS81]. It is worth noting that there were several experimental indications[@Saclay; @NIKHEF] of $E_m-p_m$ correlations according to Eq.(\[empm\_2N-I\]). However, the lower values of $Q^2$ in these experiments did not allow direct discrimination of the short-range two-body forces from the long-range two-body currents corresponding to the meson-exchange contributions. The type 2N-II SRC can be identified clearly only for the spectral function involving two protons in the recoil kinematics, Fig.\[spectralsrc\](b). As Figs.\[spectral\] and \[spectralsrc\] demonstrate, no clear signatures are seen for 3N correlations. One expects type 3N-I correlations to dominate at the left corner of the $E_m$ distribution starting at $p_m\ge 400-500$ MeV/c, while type 3N-II correlations dominate at the right (higher value) corner of the $E_m$ distribution starting at $p_m\gtrsim 400$ MeV/c. Fig.\[spectralsrc\] also demonstrates a strong dependence of pair-distortion effects on considered values of $E_m$ and $p_m$. ### Pair Distortion Effects The pair distortion effects can be assessed quantitatively in Fig.\[spectralpd\] which shows the missing momentum, $p_m$, distribution of the ratios of the spectral functions calculated within the PWIA and IA approximations for different values of $E_m$. These calculations demonstrate that the pair distortion strongly suppresses (by a factor of five) the spectral function at the kinematic region of small $p_m$ and $E_m$. This reflects the fact that in this region two spectator nucleons have vanishing relative momentum at which the interaction cross section is very large. For the same reason, one observes significant pair distortion effects in the kinematics (see Eq.(\[empm\_3N-I\])) favorable for studies of type 3N-I correlations (high $p_m$ part of the solid curve in Fig.(\[spectralpd\])). Note that in this case pair distortion is larger for the $pn$ recoil case than for the $pp$ case. Pair distortion is also large in the kinematics of very large $E_m > E_m^{2N-I,II}$ (dotted and dash doted lines in Fig.\[spectralpd\]) where one expects the dominance of type 3N-II correlations. The large pair distortion effects in the kinematics of 3N correlations can be understood qualitatively. The pair distortion effectively represents the three-nucleon correlation, in which the initial short range $NN$ correlation between knocked-out nucleon and one of the spectator nucleons is combined with the final state $NN$ reinteraction between two recoil nucleons. For type 2N-I correlations, Fig.\[spectralpd\] (dashed lines) reveals modest pair distortion effects starting at $p_m\ge 400$MeV/c. At the same time, Fig.\[spectralpd\] demonstrates more pair distortion for type 2N-II correlations. This result can also be understood qualitatively. For type 2N-I correlations, one of the recoil nucleons is initially in the SRC, while the second one is separated from the SRC with relatively small momentum. As a result, once one of the nucleons is instantaneously removed from the SRC, the two recoil nucleons in the final state will be spatially separated, thus reducing the probability for their interaction. The situation is different for type 2N-II correlations in which two recoil nucleons are in the SRC and are spatially close to each other. This will result in much larger pair distortion effects than for type 2N-I correlations. The observed feature is consistent with the qualitative arguments of Ref.[@FS81; @FS88] that to break the SRC and release a spectator from it with minimal distortion it is preferable to knock-out a nucleon directly from the SRC. The next question we address is whether the spectral function can provide an effective framework for studies of the implication of three nucleon forces (3NFs) in the ground state wave function of $^3He$. Qualitatively, one expects that the onset of 3NFs to arise in the kinematics dominated by three-nucleon correlations. This expectation is confirmed in Fig.\[spectral3m\] where the main 3NF effects are seen at $E_m=\|\epsilon_A\|$ and $E_m> E_{m}^{2N-I,II}$. Here we consider the difference in the predictions of the spectral functions calculated based on the $^3He$ wave functions which are calculated with and without explicit inclusion of 3NFs (see e.g. [@Bochum]). Predictions are made within PWIA are denoted by dotted and dashed lines in Fig.\[spectral3m\]. However, as it was discussed above, in the same region of $E_m$ and $p_m$ we predict sizable effects due to pair distortion, which effectively imitates a three-nucleon correlation. As Fig.\[spectral3m\] demonstrates pair distortion (solid lines) will considerably diminish signatures related to 3NF effects in the spectral function. ### Final state interaction (FSI) effects {#Sec.Spec.FSI} The inclusion of the FSI of the knocked-out nucleon with spectator nucleons removes the isotropy of spectral function with respect to the direction of the virtual photon, ${\bf q}$. To asses the FSI effects quantitatively, we analyze the ratio, $T_S$, defined as follows: $$T_S = {S^N_{DWIA}(Q^2,q,E_m,p_m)\over S^N_{IA}(E_m,p_m)}, \label{trans}$$ where subscripts $DWIA$ and $IA$ denote the spectral functions calculated with and without FSI effects. Here $T_S\approx 1$ corresponds to the small effects of the FSI, while $T>1$ or $T<1$ will indicate the dominance of the absorption or the rescattering effects due to the FSI. Based on the analysis of the analytic properties of rescattering amplitudes in GEA [@ggraphs], it is possible to identify the kinematic regions in which one can isolate the FSI or the SRC as the dominant term in the scattering amplitude. Similar to the reaction of exclusive electro-disintegration of the deuteron[@deuteron], one expects the FSI to dominate in nearly transverse kinematics in which $\alpha_m=1$ and $p_{m}\ge 100$ MeV/c. Here, $\alpha_m$ in the IA describes the momentum fraction of the nucleus carried out by a nucleon “1” in the infinite momentum frame of the nucleus. It is defined as follows: $$\alpha_m = {p_{m-}\over p_{A-}} \approx {p_{f-} - q_{-}\over m}\mid_{Lab \ Frame} , \label{lcom}$$ where the “minus" components of momenta are defined by $k_- = k_0-k_z$. Fig.\[FSI\_th\] represents the $\theta_m~(\alpha_m)$ dependence of $T_S$ at different values of $p_m$ calculated at $Q^2=4$ GeV$^2$. The missing energy in these calculations is chosen for the type 2N-I correlation (Eq.(\[empm\_2N-I\])). The figure demonstrates the presence of significant FSI effects in the near-transverse kinematics, $\alpha_m\approx 1$, with the effects diminishing at parallel ($\theta_m=0^0$) and anti-parallel ($\theta_m=180^0$) kinematics [^3]. The missing energy $E_m$, gives us an additional degrees of freedom to “manipulate” the relative strength of the FSI and SRC effects in the different kinematics of electro-disintegration. This is especially important for isolating SRC structures in the spectral function as it is measured in $A(e,e'N)X$ reaction. In the beginning of Sec.\[IV\], we identified several kinematic regions in which the strength of the spectral function is largely determined by SRCs. In all these cases, the initial momenta of at least two nucleons in the nucleus exceeds $300-400$ MeV/c. The final state reinteraction of the knocked-out nucleon with the spectator nucleons can largely destroy this picture. For example, as can be seen from Eqs.(21) and (27) in Ref.[@SASF1], due to the integration in the rescattering loops, it is impossible to ensure the appearance of large values of internal momenta in the ground state wave function of the nucleus in a straightforward way. Thus, the condition $p_m\ge 300$ MeV/c or $E_m\gtrsim 100$ MeV may not ensure the dominance of the high-momentum component in the ground state wave function of the nucleus. This situation may significantly affect the identification of type 2N and 3N correlations. The problem of suppression of the FSI in probing the SRC in the $A(e,e'N)X$ reaction was addressed within the GEA in Ref.[@ggraphs]. It was observed that a trivial condition, $p_m\ge k_F$, is not sufficient to probe the SRC component of the ground state nuclear wave function. One needs to impose the following additional condition on the effective longitudinal momentum, $p_Z$, entering the single rescattering amplitude, $$p_Z \equiv p_{mz} + {q_0\over q}(E_m + {p_m^2\over 4m}) > k_F. \label{pZ}$$ With this, one will ensure that rescatterings happen within the SRC. Note that “$p_Z$”s in Eq.(\[pZ\]) correspond to the pole values of fast nucleon propagators in the rescattering amplitude (see e.g. Eq.(21) in Ref.[@SASF1]). In Fig.\[fig\_pZ\], we consider the $E_m$ ($p_Z$) dependence of $T_S$ for parallel (a) and anti-parallel (b) kinematics for values of $p_m$ equal to $400$, $500$ and $600$ MeV/c. In Fig.\[fig\_pZ\](a), the center of mass momentum of the two recoil nucleons is in the direction backward to ${\bf q}$, while in Fig.\[fig\_pZ\](b), the recoil system is produced in the forward direction. One can see from this figure that the FSI indeed decreases with increase of $\|p_Z\|$. This indicates that the FSI is increasingly confined within the SRC. An interesting feature of these results is that while the FSI contribution keeps decreasing with increase of $\|p_Z\|$ in the kinematic region relevant for type 2N-I and 3N-II correlations, it grows sharply in the region associated with the type 3N-I correlations. Both trends can be understood qualitatively. When $\|p_Z\|\le k_F$, one has dominant contributions from the FSI involving uncorrelated nucleons which have a larger probability amplitude in the ground state wave function of $^3He$. Once $\|p_Z\|> k_F$, the FSI takes place predominantly within the 2N correlation. In the latter case, for type 2N-I SRC, the third nucleon has a small momentum and is spatially separated from the NN SRC. Thus, it does not contribute substantially to the FSI. For type 3N-I correlations both spectator nucleons contribute to the FSI since they are both correlated with the knocked-out nucleon. This results in larger FSI effects as compared to the IA contribution. These calculations indicate that (except for the case of type 3N-II SRC in the parallel kinematics) the FSI generally dominates in the kinematics where one expects an enhanced contribution from 3N SRCs in the wave function of $^3He$. For type 2N-II correlations, the FSI takes place between the knocked-out nucleon, which is initially almost at rest and spatially separated from the 2N SRC (see Fig.\[2Nsrc\](b)). As a result, one expects a diminished FSI contribution in whole range of $E_m$. Such behavior can be seen in Fig.\[FSIem\_pn\_2N-II\] for the calculated $E_m$ dependence of the spectral function at $p_m=0$. This calculation shows that the type 2N-II SRC for the knocked-out neutron reaction attains the characteristic minimum in $E_m$ distribution although smeared out strongly due to rescatterings. In Fig.\[FSIem\] we summarize the predictions for the $E_m$ dependence for the spectral function at different values of missing momenta $p_m$ calculated for parallel (a) and anti-parallel (b) kinematics. These calculations demonstrate that the observed $E_m-p_m$ correlation within the IA for type 2N-I SRC generally survives the FSI for both parallel and anti-parallel kinematics. For 3N correlations, type 3N-II SRC survives FSI for parallel kinematics while type 3N-I SRC is strongly affected by FSI. Summarizing our consideration of the spectral function we can conclude that the $A(e,e'N)X$ reaction is best suited for studies of 2N correlations only. Type 2N-I SRC survives both pair distortion and the FSI while type 2N-II attains its specific feature for neutron-knock out reactions at $p_m\le 100$ MeV/c [^4]. Our calculations show that, in general, 3N correlations are strongly affected by pair distortion and FSI effects. Pair distortion, having a qualitative features of 3N correlation, strongly affects both types of 3N correlations, while the FSI contributes strongly in type 3N-I SRC and has a diminished impact on type 3N-II SRCs only at parallel kinematics. Finally, for near transverse kinematics starting at $p_m\ge 400$ MeV/c, the $A(e,e'N)X$ reactions represent an ideal tool for studying the structure of FSIs. This feature becomes especially valuable for large $Q^2>4$ GeV$^2$ for studies of color coherence phenomena for which we expect a decrease of $T_S$ with an increase of $Q^2$ as opposed to the $Q^2$ independence of $T_S$ in the GEA. Decay Function -------------- The decay function is practically an unexplored quantity and the experiments that will allow us to extract it from the exclusive cross section data are only emerging[@LW]. Our main motivation in these numerical analyses is to highlight those significant features of the decay function that are related to the short-range structure of the ground state nuclear wave function as well was to the structure of reinteractions between the nucleons in the final state of the reaction. We will consider the partially integrated decay function, which will allow us to remove the $\delta$-function in Eqs.(\[Dia\]) and (\[Ddwia\]). Namely, we consider, $$D_{int}^N({\bf p_m},E_m,p_{r2},t_{r2}) = \int D^N({\bf p_m},E_m,{\bf p_{r2}},t_{r2})p^2_{r2} d\Omega_{r2}, \label{Dfun-I}$$ where $D^N({\bf p_m},E_m,{\bf p_{r2}})$ is defined in Eqs.(\[Dia\]) or (\[Ddwia\]). This integration takes into account the fact that one of the components of ${\bf p_m}$ or ${\bf p_{r2}}$ is not independent and is fixed from the energy conservation condition for the quasi-elastic disintegration of $^3He$. Therefore, $D_{int}$ represents a quantity which could be extracted directly from a $^3He(e,e'NN)N$ experiment. Furthermore, we will refer to $D_{int}$ as a decay function. We start by analyzing the decay function in the PWIA, focusing on those features that are related to SRC signatures of the ground state wave function which are described in Figs.\[2Nsrc\] and \[3Nsrc\]. As we observed in the previous section, type 2N-I SRCs exhibited measurable (though broad) correlation between $E_m$ and $p_m$ (starting at $p_m\ge 300$ MeV/c), with the peak of the $E_m$ distribution defined by Eq.(\[empm\_2N-I\]). These correlations could be understood qualitatively as a result of strong short range interactions between the struck nucleon (with momentum $p_m$) and one of the recoil nucleons in the nucleus with the third nucleon having a relatively small momentum in the mean field of the 2N SRC pair. To check this picture, we calculate the decay function distribution as a function of $p_{r2}$ and $p_{r3}$ imposing different cuts on the missing momentum, $p_m$. Figs.\[decay\_pwia\_n\] and \[decay\_pwia\_p\] present the results of the calculation of the decay function of the reactions corresponding to the knock-out of the neutron and proton respectively. The (a), (b), (c) and (d) parts correspond to the $p_m$ cuts at $p_m>150$, $300$, $400$ and $700$ MeV/c respectively. The calculations clearly show emerging type 2N-I correlations between the knocked-out nucleon and one of the spectator nucleons with an increase of $p_m$. These correlations are dominating the landscape of the ($p_{r2}$, $p_{r3}$) distribution once $p_m\ge 300 MeV/c$. Fig.\[decay\_pwia\_p\] shows also that the $pn$ SRCs are significantly larger than the $pp$ correlations. Thus, measuring the decay function will allow a direct check of the relative strength of $pp$ and $pn$ correlations. We conclude, from the discussions of Figs.\[decay\_pwia\_n\] and \[decay\_pwia\_p\], that the decay function is strikingly more sensitive to the SRC than the $E_m-p_m$ correlation observed in the spectral function. Next we analyze the features of the decay function related to the type 2N-II correlations. From the consideration of the spectral function in the previous section, we learned that reactions corresponding to the knock-out of the neutron are best suited for studies of type 2N-II correlations since, at small $p_m<100$ MeV/c and large $E_m$, the spectral function exhibits a minimum associated with the node in the relative momentum distribution of the recoil $pp$ pair in the $S$-state. In Fig.\[decay\_2N-II\] we show the ($E_m,p_{r2}$) distribution of the decay function which demonstrates strong correlations between the minimum in the $E_m$ distribution and the value of recoil nucleon momenta at small $p_m$ ($<50$ MeV/c in the calculation). Turning now to the three nucleon correlations, we observe that for both types of 3N SRCs (Fig.\[3Nsrc\]) the two recoil nucleons have comparable momenta. This situation corresponds kinematically to the area around the saddle in Figs.\[decay\_pwia\_n\] and \[decay\_pwia\_p\]. One can check whether the expectation that type 3N-I and 3N-II correlations will dominate in the dynamics of three nucleon correlations is justified. For this we observe that the configurations of Fig.\[3Nsrc\] are characterized by a distinctive angular relations between the two recoil nucleons: for type 3N-I SRC two recoil nucleons will be produced at small relative angle (almost parallel) while for type 3N-II SRC the relative angle of the recoil nucleon momenta is $\approx 120^o$. In Fig.\[decay\_src\] we analyze the dependence of the decay function on the relative angle between the recoil nucleons, $\theta_{23}$, and the missing energy, $E_m$, for different values of missing momentum, $p_m$. Fig.\[decay\_src\](c) and (d) clearly show an emergence of peaks at small $E_m$ and $\theta_{23}$ ( type 3N-I SRCs) and at large $E_m$ and $\theta_{23}\approx 120^0$ ( type 3N-II SRCs). The appearance of the peaks is more clearly seen in Fig.\[decay\_src\_ext\] where Fig.\[decay\_src\] (d) is demonstrated from different viewpoints. Fig.\[decay\_src\_ext\] shows also the peak at $\theta_{23}\approx 180^0$ and at moderate values of $E_m$ which corresponds to the type 2N-I correlation in which only one of the recoil nucleons is correlated with the knocked-out nucleon and is produced in the direction opposite to the direction of ${\bf p_m}$. Moving to a more quantitative discussion of correlation properties of the decay function, in the following subsections we study the following question: To what degree do the detected properties of the recoil nucleon in coincidence with the knocked-out nucleon reflect the properties of preexisting short range configurations in the ground state nuclear wave function? ### Pair Distortion Effects In Fig.\[decay\_srcs\_pd\] we discuss the effects of the reinteraction between recoil nucleons (pair distortion) in the $p_{r2}$-momentum distribution of the decay function for all four types of correlations[^5]. In Fig.\[decay\_srcs\_pd\](a) one observes a significant difference in the yields of the recoil proton or neutron for a reaction in which one of the protons is knocked out from the type 2N-I correlations. Fig.\[decay\_srcs\_pd\](b) shows the momentum distribution of the recoil proton when proton (labeled “pp” in the figure) or neutron (labeled “np”) is knocked-out in the kinematics of type 2N-II correlations. This figure demonstrates significant pair distortion effects as compared to the type 2N-I correlations. Figs.\[decay\_srcs\_pd\](c) and (d) demonstrate the momentum distributions from type 3N-I and II SRC’s respectively. As in the case of the spectral function, one observes that, in general, the pair distortion interferes significantly with the three nucleon correlations. It is worth noting, however, that, due to the depleted interaction strength in the $pp$ channel at relative momenta $\sim 400$ MeV/c, pair distortion effects are suppressed in type 3N-II correlations for the neutron knock-out reaction in a recoil momentum range of $300-550$ MeV/c (see Fig.\[decay\_srcs\_pd\](d)). We will discuss in Sec.\[3NForce\] how this observation could be used to explore the type 3N-II kinematics for investigation of three nucleon forces in $^3He$. Also, Fig.\[decay\_srcs\_pd\] reveals an additional feature that allows us to discriminate between 2N and 3N SRC signals. This feature is the relative abundance of pp and pn pairs in different correlations. For 2N correlations one observes significantly smaller yield associated with $pp$ correlations as compared to $pn$ correlations while for 3N correlations these yields are comparable. ### Final State Interaction The inclusion of the final state interaction of the knocked-out nucleon with residual nucleons removes the isotropy of the decay function with respect to the momentum vector of the virtual photon, ${\bf q}$. Staying in the framework of consideration of type 2N and 3N correlations, we investigate FSI effects in angular and momentum dependences of the decay function for each type of correlation. As in Sec.\[Sec.Spec.FSI\], we consider the kinematics with fixed $Q^2=4$ GeV$^2$. [A. Type 2N-I Correlations:]{} For the type 2N-I correlations, we consider first the dependence of the decay function on the production angle of recoil nucleon $N_{r2}$ with respect to the direction of the virtual momentum ${\bf q}$. Fig.\[decay\_2n-i\_fsi\_th\] shows these dependences for different values of recoil nucleon (neutron (a) and proton (b)) momentum $p_{r2}$ for the reaction with knocked-out proton. Calculations clearly show the transition of FSI from the screening regime at momenta $p_{r2}\le 300$ MeV/c to the double scattering regime at $p_{r_2}\ge 400$ MeV/c. This picture is clearly reminiscent the double scattering signatures of the deuteron electro-disintegration[@deuteron; @ggraphs; @treview]. Calculations also predict very different angular dependences for neutron (a) and proton (b) production in recoil kinematics. It is easy to check that the maximal FSI effects are predicted at $\alpha_{r2}={E_{r2}-p^z_{r2}\over m}=1$ which is analogous to the maximums of the FSI for the spectral functions at $\alpha_m=1$ observed in Sec.\[Sec.Spec.FSI\]. Fig.\[decay\_2n-i\_fsi\_th\] shows also that our best chances to extract genuine information about type 2N-I SRC, is to concentrate on anti-parallel ($\theta_{r2}=0^0$) and parallel ($\theta_{r2}=180^0$) kinematics. It is worth noting that the range of $\theta_{r2}$ where FSI, are small is broader for the $\theta_{r2}=180^0$ case. Fig.\[decay\_2n-i\_fsi\_pr2\] represents the $p_{r2}$ momentum dependences of the decay function for $\theta_{r2}=180^0$ and $\theta_{r2}=0^0$ (marked curves) kinematics. Calculations predict qualitatively different momentum distributions for correlated $pn$ (a) and $pp$ (b) pairs. While for $pn$ at both $\theta_{r2}=180^0$ and $\theta_{r2}=0^0$ one observes qualitatively similar momentum distributions, for $pp$ pairs they are significantly different. This difference can be understood from Eq.(\[pZ\]), which defines the effective longitudinal component of missing momentum entering in the rescattering amplitude. For type 2N-I kinematics it can be written as $p_Z = -p_{r2}cos(\theta_{r2}) + {q_0\over q}(E_m+{p_{r2}^2\over 4m})$. At $\theta_{r2}=180^0$ ($\theta_{r2}=0^0$), $p_Z > (<) p^z_{r2}$ and the FSI term is defined by the effective momentum which is larger (smaller) than measured momentum, $p_{r2}$. As a result, the FSI is suppressed at $\theta_{r2}=180^0$ (backward) kinematics as compared to the $\theta_{r2}=0^0$ (forward) kinematics. For the $pp$ pair this difference is very dramatic due to the node in the relative momentum distribution of the $pp$ pair. It is worth noting that no significant triple scattering (double rescattering) contribution is observed in Figs.\[decay\_2n-i\_fsi\_th\] and \[decay\_2n-i\_fsi\_pr2\]. This serves as an additional indication that in type 2N-I kinematics the reaction is defined predominantly by two-body interactions, thus representing a nearly ideal framework for studies of the 2N processes both in terms of the short range correlations and single rescattering processes. [B. Type 2N-II Correlations:]{} To isolate type 2N-II correlations we identify kinematics similar to one presented in Fig.\[decay\_srcs\_pd\](b), in which missing momentum is fixed to $p_m=100$ MeV/c in ${\bf q}$ direction. In Fig.\[decay\_2n-ii\_fsi\_th\] we consider the $\theta_{r2}$ dependence of the decay function for both $pp$ (a) and $pn$ (b) recoil pairs. Since, in these kinematics, the knocked-out nucleon is on average at a larger distance from the recoil 2N pair, one expects the lesser FSI effects due to rescattering of the knocked-out nucleon off the recoil nucleons. Calculations presented in Fig.\[decay\_2n-ii\_fsi\_th\] confirm this expectation. The same pattern can be seen in the momentum distribution plot in Fig.\[decay\_2n-ii\_fsi\_pr2\] which shows that the FSI is diminished practically for the whole region of recoil nucleon momenta, $p_{r2}$, of interest. Comparison of type 2N-I and 2N-II correlation cases demonstrate that type 2N-II makes the best case for probing the node of the relative momentum distribution in the $pp$ pair in the $^3He$ wave function (see Figs.\[decay\_2n-ii\_fsi\_th\](a) and \[decay\_2n-ii\_fsi\_pr2\](b)). However, a definitive answer on whether the node can be observed in the experiment requires a careful accounting of non-pole effects in the pair-distortion contribution. This consideration is out of scope of present paper and a dedicated study of the node effects will be presented elsewhere[@node]. As in the case of type 2N-I correlations, the effects of double rescattering are marginal which indicates again the feasibility of isolating two-body effects without complication, due to double rescattering of knocked-out nucleon off both recoil nucleons. [C. Type 3N-I Correlations:]{} A consideration of the angular dependence of the decay function for type 3N-I correlations in Fig.\[decay\_3n-i\_fsi\_th\] reveals a significant effect of FSIs for almost all angles of pair production (or $\theta_m$). The FSI dominates especially at transverse kinematics at $\alpha_m=1$ which reveals a significant contribution from the double rescattering processes starting at $p_m\ge 400$ MeV/c. Production of $pp$ pairs (Fig.\[decay\_3n-i\_fsi\_th\](a)) in parallel kinematics provides the best condition for probing type 3N-I SRCs (albeit not without considerable pair distortion effects) starting at $p_m\ge 600$ MeV/c. This can be understood qualitatively. Since each rescattering transfers $k_z\approx \Delta, k_t\le 500$ MeV/c momentum (predominantly in transverse direction; $k_t>\Delta$), and the rescattering amplitude falls exponentially with $k^2_t$, it is kinematically infeasible to rescatter two nucleons above 600 MeV/c in the backward direction. The effects discussed in the previous paragraph can be seen in more detail in the momentum distribution in Fig.\[decay\_3n-i\_fsi\_pr2\] which confirms again that the only reasonable chance to extract the original momentum distribution in type 3N-I correlation is to measure the coherent recoil $pp$ pair production in the parallel kinematics $\theta_{m}=0^0$. Note that the significant contribution from double rescattering processes at transverse kinematics can be also understood qualitatively. To produce two coherent nucleons at large angle it is more preferable for the knocked-out nucleon to scatter consecutively off spectator nucleons. It contributes maximally in the transverse kinematics due to the eikonal nature of NN rescattering which dominates at $\alpha_m=1$. This situation is reminiscent of the dynamics relevant to the form-factors of few-body systems measured in high momentum transfer reactions, in which case one needs significant rescatterings between constituents of the system to produce coherent combination of the subsystem. [D. Type 3N-II Correlations:]{} In considering the angular dependence of the decay function for type 3N-II correlations in Fig.\[decay\_3n-ii\_fsi\_th\], we observe overall large FSI effects except in the kinematics in which both recoil nucleons are produced in the backward direction compared to that of $\bf q$. In this figure this corresponds to the situation when the recoil nucleon with momentum $p_{r2}$ is produced at $\theta_{r2}=120^0$ and the fast nucleon is knocked out in parallel kinematics ($\theta_m=0^0$). This automatically puts the production angle of the second recoil nucleon at $\theta_{r3}=120^0$ in the other half of the scattering plane. One can understand the suppression of FSI in this kinematics qualitatively. It is very improbable with one single rescattering to produce two nucleons in the backward hemisphere of knocked-out energetic nucleon. One may expect that double rescattering can contribute to production of such configurations. However, as it will be discussed in the next section, it is dominant only at $\alpha_{r2}\approx \alpha_{r3}\approx 1$ which is significantly away from the considered kinematics. Note that the different angular dependence for recoil $pp$ (a) and $pn$ (b) pairs at momentum range $300-500$ MeV/c is related to the qualitative difference in the relative momentum distribution of these pairs (namely to an existing node in $pp$ distribution). Note that recoil nucleon angles $110 < \theta_{r2} < 130$ at $\theta_m=0^0$ and $p_{r2}=600$ MeV/c are kinematically forbidden since in this case $\alpha_m+\alpha_{r2}+\alpha_{r3}\ge 3$. However, the $600$ MeV/c curve shows that the FSI is small at broader angular ranges starting at $80^0$ to $150^0$. This may be very important for probing larger internal momenta in type 3N-II SRCs without substantial FSI effects. The momentum distribution of the decay function in Fig.\[decay\_3n-ii\_fsi\_pr2\] confirms the observed, in Fig.\[decay\_3n-ii\_fsi\_th\], smallness of FSIs at parallel kinematics with two recoil nucleons produced in the backward hemisphere at 120$^0$. This situation provides a unique window for accessing type 3N-II correlations. They will be discussed in more details in Sec.\[3NForce\]. In Fig.\[decay\_3n-ii\_fsi\_pr2\] we also compare the momentum distribution for anti-parallel ($\theta_m=180^0$) kinematics when recoil nucleons are produced in forward $60^0$ angles. Here we observe significant FSIs which enhances the cross section by almost an order of magnitude. ### Double Rescattering Effects {#Sec.DFSI} Previous considerations demonstrated that in practically all cases the maximal FSI is generated in kinematics in which $\alpha_m=1$. To enhance the effects of double rescattering, relative to the single rescattering contribution, the strategy is to keep $\alpha_m=1$ and choose kinematics in which two recoil nucleons are produced in symmetric and transverse kinematics. Such configurations boost the double rescattering effects, since in these cases the most economic way to produce two recoil nucleons with large transverse momentum is to have two consecutive rescatterings of the knocked-out nucleon off the spectator nucleons in $^3He$. One such kinematics corresponds to the type 3N-I correlations, in which two recoil nucleons are produced with almost vanishing relative momentum. Fig.\[decay\_3n-i\_fsi\_th\] demonstrates a large contribution to double rescattering when two coherent recoil nucleons are produced at almost transverse angles with respect to the direction of $\bf q$. In Fig.\[decay\_n-nn\_double\] we consider further the kinematics of the type 3N-I correlation by calculating the $p_m$ dependence of the decay function at $\alpha_m=1$. The result is the significant enhancement of the double rescattering effects starting at $p_m\ge 300$ MeV/c. In the range of $300 \le p_m \le 700$ MeV/c, double rescattering screens the single rescattering contribution through its destructive interference with the single rescattering amplitude (see Eqs.(21) and (27) in Ref.[@SASF1]). However, starting with $p_m \ge 700$ MeV/c, our calculations show that the decay function is determined predominantly by the square of the double rescattering amplitude of Eq.(27) in Ref.[@SASF1]. Since the internal nucleon momenta in the $^3He$ wave function are small in the double rescattering amplitude, the relativistic effects are expected to be small in spite of $p_m>700$ MeV/c. Another kinematical region which provides the symmetric configuration for recoil nucleons is the one close to the type 3N-II correlations in which the light cone momentum fractions of all three nucleons are chosen $\alpha_m=\alpha_{r2}= \alpha_{r3}=1$, and two recoil nucleons are produced in the opposite halves of the scattering plane. In this case, one again expects the enhanced contribution from double rescattering. The analysis of single (Eq.(21) in Ref.[@SASF1]) and double rescattering (Eq.(27) in Ref.[@SASF1]) amplitudes shows that the double rescattering contribution is maximized in the kinematics where the interference of single and IA (Eq.(11) in Ref.[@SASF1]) amplitudes cancel the square of the single rescattering amplitudes, which happens at $p_m\approx 300$ MeV/c. This can be seen in Fig.\[decay\_n-n-n\_double\] which displays missing momentum dependence of the decay function in the above kinematical region discussed above. One observes that at $p_m\approx$ 300 MeV/c the double rescattering diagram is dominant. It is interesting to note that in this case the dominant part comes from the interference between IA (Eq.(11) in Ref.[@SASF1]) and double (Eq.(27) in Ref.[@SASF1]) rescattering amplitudes.\ [Digression: Color Transparency:]{} The ability to isolate the double rescattering contribution in $^3He$ electro-disintegration may play a significant role in the ongoing studies of color transparency (CT) phenomena. The main premise of CT is that at sufficiently large $Q^2$ the knocked-out nucleon is produced in a point like configuration (PLC) which, due to the color neutrality of the object, will have a reduced hadronic interaction strength. Thus, CT phenomena will be manifest in the decreasing of the $f_{PLC-N}$ amplitude of rescattering with an increase of $Q^2$. This is in contrast to the $Q^2$ independence of $f_{NN}$ in eikonal approximation. Presently, two complementary experimental approaches are used to find the signatures for CT phenomenon. One is the attenuation experiments[@Ent] in which nuclear transparency is measured in $(e,e'N)$ reaction off nuclei with $A\ge 2$ and the other is the deuteron electro-disintegration reactions[@KE] which are aimed at measuring the single rescattering terms in $d(e,e'N)N$ reactions. While attenuation experiments are sensitive to $\sim f_{PLC-N}$, the single-rescattering experiments can provide sensitivity up to $\sim f_{PLC-N}^2$. The possibility of isolating the double rescattering term in $^3He$ electro-disintegration will allow us to gain unprecedented sensitivity - up to $\sim f_{PLC,N}^4$ (see Eq.(27) in Ref.[@SASF1]). Furthermore, isolating the double rescattering amplitude will allow us to address such an intricate questions as whether the first rescattering will destroy the PLC coherence formed by the initial high $Q^2$ $\gamma^N$ scattering. ### Probing Three Nucleon Forces {#3NForce} Three nucleon forces (3NFs) are one of the most elusive features in nuclear physics. The existence of them for the triton was assumed for the first time by Wigner[@Wigner] even before the triton was discovered experimentally. There is little theoretical guidance for systematic building of 3NFs and the main experimental evidence used to constraint the different 3NF models is the binding energy of $A=3$ nuclei (for review of the present status of 3NFs see [@Friar]). The importance of 3NFs was emphasized in the studies of the binding energies of $A=3$ nuclei. Furthermore, it allowed for improvements in the calculation of binding energies of nuclei $A = 4 -9$[@PPWC]. The 3NFs can significantly modify the present models of equations of state of nuclear matter[@HeisPand], therefore, understanding 3NFs can have a significant impact on our understanding of the physics relevant to the superdense nuclear matter such as neutron stars. However, as it is mentioned in Ref.[@Friar], an accuracy of $1\%$ in calculations are needed in order to systematically disentangle 3NFs forces from the overwhelming 2N interactions in few nucleon systems. In our consideration of 3NFs, we proceed from the assumption that, just by the nature of 3N forces, they will dominate in type 3N-II correlations. Conversely, in the case of the type 2N-I correlations, it is possible to suppress 3NFs significantly by restricting the momentum of the third spectator nucleon to be close to the zero. Thus, our expectation is that one should observe significantly different contributions from 3NFs in 3N-II and 2N-I correlations. Furthermore, we recall our discussions of type 2N-I and 3N-II correlations in Secs.V.B(2A) and V.B(2D) where we found that one can significantly suppress FSI effects in type 2N-I correlations in parallel kinematics $(\theta_m=0^0)$ (see Fig.\[decay\_2n-i\_fsi\_pr2\]) and in type 3N-II correlations in kinematics where we choose two recoil nucleons to be produced in the backward hemisphere with respect to ${\bf q}$ (see Fig.\[decay\_3n-ii\_fsi\_pr2\]). Thus we expect that these kinematic windows are optimal for probing 3NFs. To quantify our statement, in Fig.\[decay\_3n-ii\_3nf\] we compare the momentum distributions of the decay function calculated in the type 3N-II kinematics when two recoil protons are produced at $120^0$ relative to the $\bf q$. Calculations are done for two cases: in one case we have only $NN$ interactions while in the other case 3NFs are included according to the Tucson-Melbourne model [@TM]. Our calculations show a factor of two difference between calculations including only 2N forces and calculation including 2N+3NF forces. In contrast, the calculations in the kinematics of the type 2N-I correlations presented in Fig.\[decay\_2n-i\_3nf\], using the same two models of NN interactions, predict little difference between momentum dependences of the decay function. These considerations show that we can identify the domains in the kinematics of type 3N-II correlations where FSI effects are relatively small and one has a good chance to extract the genuine information about 3NFs. This assumes one is doing simultaneous measurement also in 2N-I kinematics where the same 3NFs will give a negligible contribution. Summary {#V} ======= Based on generalized eikonal approximation, developed in the part-I of this work[@SASF1], for high $Q^2$ electro-disintegration of $A=3$ system we perform numerical studies of exclusive $^3He(e,e'N)NN$ reactions. As an input, in our numerical studies we use: (i) The Bochum group’s [@Bochum] calculation of ground state $^3He$ wave function for different sets of realistic NN interaction potentials as well as calculations which explicitly include three nucleon forces in the ground state. (ii) The SAID group’s[@SAID] parameterization of low-to-intermediate energy NN scattering amplitudes to calculate the two-nucleon continuum state wave function which is needed in order to evaluate the interaction between recoil nucleons in the final state of the reaction (pair distortion effects). (iii) We use our updated parameterization of high energy NN scattering amplitudes in the calculation of the small angle rescattering of struck nucleon off spectator nucleons[@tabra]. To describe the exclusive $^3He(e,e'N)NN$ reaction we use the formalism of the decay function which is related to the conventional spectral function through the integration of the phase space of the recoil nucleons. In the numerical analyses of both spectral and decay functions, we concentrate on studies of two main types of 2N and 3N correlations. For 2N correlations, we consider the ones in which the struck nucleon is initially correlated with one of the recoil nucleons while the third nucleon is spatially isolated (type 2N-I SRC). Also for the case of 2N correlations we consider the case in which the two recoil nucleons are in 2N correlation with the struck nucleon in the mean field of the SRC pair (type 2N-II SRC). For 3N correlations we consider the correlations in which struck nucleon with large missing momenta is correlated with the pair of coherent nucleons (type 3N-I SRC). For the 3N correlations we also consider the case in which all three nucleons have relative momenta exceeding the characteristic mean field momentum in the nucleus and have a relative angle $\approx 120^0$ (type 3N-II SRC). In discussing the spectral function, we demonstrate that it exhibits several unique features related to the structure of 2N correlations. These are the correlation between missing momenta and missing energy and the minimum in the spectral function associated with the node in the relative momentum distribution of the $pp$ pair. These results are in agreement with previous analyzes of spectral function (see e.g. [@Ciofi]). There are, however, no clear cut signatures in the spectral function related to the 3N correlations. Within PWIA, 3N SRCs reveal only through the strength of spectral function at very high values of missing momenta and/or energy. Our new result in considering the spectral function is that the strength related to the 3N SRCs is practically washed out by the pair distortion and the FSI effects. Considering the decay function, which is practically an unexplored quantity, we observe that within PWIA it clearly exhibits the main features of 2N and 3N correlations. Subsequent analysis of pair distortion and FSI effects revealed that the additional degrees of freedom associated with the full detection of the decay products of the reaction allows us to pinpoint unambiguously the kinematics in which the FSI or SRC plays a dominant role. Our conclusion is that the comprehensive study of the decay function in exclusive reactions will allow an unprecedented access to the 2N and 3N correlations in the nuclei. We highlight two particular cases. One is the possibility to isolate double rescattering effects which can provide us a new and powerful tool in studying color transparency phenomena. The other is identifying a kinematic window that will allow us to probe directly the effects associated with three-nucleon forces in the ground state wave function of $^3He$. Our overall conclusion is that the investigation of the decay function opens up a completely new venue in studies of short range nuclear properties and allows us to discriminate between different orders of final state reinteractions. The latter will provide a powerful tool in studies of color transparency phenomena. We are grateful to Andreas Nogga for providing us with Bochum group’s wave functions of $^3He$. Special thanks to Richard Arndt and Igor Strakovsky for their help in using the SAID program. We thank Ted Rogers for careful reading of the manuscript and for many valuable comments. This work is supported by DOE grants under contract DE-FG02-01ER-41172 and DE-FG02-93ER40771 as well as by the Israel-USA Binational Science Foundation Grant. M.M.S. gratefully acknowledges also a contract from Jefferson Lab under which this work was done. The Thomas Jefferson National Accelerator Facility (Jefferson Lab) is operated by the Southeastern Universities Research Association (SURA) under DOE contract DE-AC05-84ER40150. M.M. Sargsian, T.V. Abrahamyan, M.I. Strikman, L.L. Frankfurt, [Exclusive Electro-Disintegration of $^3$He at high $\boldmath Q^2$: I. Generalized Eikonal Approximation]{}, Part-I, arXiv:nucl-th/0406020, Phys. Rev. [C]{} ([in press]{}). L.L. Frankfurt, M.M. Sargsian and M.I. Strikman, Phys. Rev. [C56]{} (1997) 1124. M. M. Sargsian, Int. J. Mod. Phys. E [10]{}, 405 (2001). L.L. Frankfurt, J.A. Miller and M.I. Strikman, [Ann. Rev. Nucl. Part. Sci.]{} [44]{}, 501 (1994). C. Ciofi degli Atti and L. P. Kaptari, Phys. Rev. C [66]{}, 044004 (2002). L.L. Frankfurt and M.I. Strikman, Phys. Rep. [160]{}, 235 (1988). T. De Forest, Nucl. Phys. A [392]{}, 232 (1983). G. Bizard [et al.]{}, Nucl. Phys. B [85]{}, 14 (1975); M. N. Kreisler [et al.]{},Nucl. Phys. B [84]{}, 3 (1975); E. L. Miller [et al.]{}, Phys. Rev. Lett. [26]{}, 984 (1971). Y. I. Azimov, Sov.Phys. JETP, v.16, 1640 (1963); Phys.Lett., v.3, 195 (1963). W. R. Gibbs and B. Loiseau, Phys. Rev. C [50]{}, 2742 (1994). A. Nogga A. Kievsky, H. Kamada, W. Gloeckle, L. E. Marcucci, S. Rosati and M. Viviani, Phys. Rev. C [67]{}, 034004 (2003). R. A. Arndt, I. I. Strakovsky and R. L. Workman, Phys. Rev. C [62]{} (2000) 034005. T.V. Abrahamyan, [Master’s Thesis]{}, unpublished, FIU, 2003. R. B. Wiringa, V. G. J. Stoks and R. Schiavilla, Phys. Rev. C [51]{}, 38 (1995) C. Ciofi degli Atti, L. P. Kaptari and D. Treleani, Phys. Rev. C [63]{}, 044601 (2001) L.L. Frankfurt and M.I. Strikman, Phys. Rept. [76]{}, 215 (1981). C. Ciofi degli Atti, S. Simula, L. L. Frankfurt and M. I. Strikman, Phys. Rev. C [44]{}, 7 (1991). C. Marchand [et al.]{}, Phys. Rev. Lett. [60]{}, 1703 (1988). D. L. Groep [et al.]{}, Phys. Rev. C [63]{}, 014005 (2001). L.L. Frankfurt, W.R. Greenberg, G.A. Miller, M.M. Sargsian and M.I. Strikman, Z.Phys. [A352]{}, 97 (1995). R. A. Niyazov and L. B. Weinstein \[CLAS Collaboration\], Phys. Rev. Lett. [92]{}, 052303 (2004) T. Abrahamyan and M. Sargsian [in preparation]{}. K. Garrow [et al.]{}, Phys. Rev. C [66]{}, 044613 (2002). [Jefferson Lab Proposal E94-102]{}, 1994. K. Sh. Egiyan, K. A. Griffioen and M.I. Strikman(spokespersons) 1994 Measuring Nuclear Transparency in Double Rescattering Processes, [Jefferson Lab Proposal E94-019]{}, 1994. E. Wigner, Phys. Rev. [43]{}, 252 (1933). J. L. Friar, Nucl. Phys. A [684]{}, 200 (2001). S. C. Pieper, V. R. Pandharipande, R. B. Wiringa and J. Carlson, Phys. Rev. C [64]{}, 014001 (2001). H. Heiselberg and V. Pandharipande, Ann. Rev. Nucl. Part. Sci. [50]{}, 481 (2000). S. A. Coon and H. K. Han, Few Body Syst. [30*]{}, 131 (2001). [^1]: Note, that in some cases $\sigma_{eN}$ is defined without the flux factor (see e.g. [@deFor]). In these cases factor ${j_N\over j_A}$ should not be included in Eq.(\[crsIA\]). [^2]: The cases in which only a single rescattering is considered we refer to as IA+FSI1. [^3]: Everywhere in the text parallel/anti-parallel kinematics correspond to $\theta_m=0^0/180^0$ [^4]: As a note of caution, we point out that the minimum in the $E_m$ distribution is very narrow. Hence its experimental observation will require resolution in $E_m$ on the order of a few MeV. [^5]: Herewith, we will label calculations by $(N_f,N_{r2})$, in which $N_f$ and $N_{r2}$ describe the type of the knocked-out and recoil nucleons with momenta $\bf p_{f}=\bf p_m+\bf q$ and $\bf p_{r2}$ respectively.	{ "pile_set_name": "ArXiv" }
--- author: - 'R. Martín-Doménech , E. Dartois , and G. M. Muñoz Caro' date: 'Received - , 2016; Accepted - , 2016' subtitle: 'III. Diffusion of photo-produced H$_{2}$ as a function of temperature' title: 'Vacuum ultraviolet photolysis of hydrogenated amorphous carbons.' --- [Hydrogenated amorphous carbon (a-C:H) has been proposed as one of the carbonaceous solids detected in the interstellar medium. Energetic processing of the a-C:H particles leads to the dissociation of the C-H bonds and the formation of hydrogen molecules and small hydrocarbons. Photo-produced H$_{2}$ molecules in the bulk of the dust particles can diffuse out to the gas phase and contribute to the total H$_{2}$ abundance.]{} [We have simulated this process in the laboratory with plasma-produced a-C:H and a-C:D analogs under astrophysically relevant conditions to investigate the dependence of the diffusion as a function of temperature.]{} [Experimental simulations were performed in a high-vacuum chamber, with complementary experiments carried out in an ultra-high-vacuum chamber. Plasma-produced a-C:H analogs were UV-irradiated using a microwave-discharged hydrogen flow lamp. Molecules diffusing to the gas-phase were detected by a quadrupole mass spectrometer, providing a measurement of the outgoing H$_{2}$ or D$_{2}$ flux. By comparing the experimental measurements with the expected flux from a one-dimensional diffusion model, a diffusion coefficient $D$ could be derived for experiments carried out at different temperatures.]{} [Dependance on the diffusion coefficient $D$ with the temperature followed an Arrhenius-type equation. The activation energy for the diffusion process was estimated ($E_{D}$(H$_{2}$) = 1660 $\pm$ 110 K, $E_{D}$(D$_{2}$) = 2090 $\pm$ 90 K), as well as the pre-exponential factor ($D_{0}$(H$_{2}$) = 0.0007$^{+0.0013}_{-0.0004}$ cm$^{2}$ s$^{-1}$, $D_{0}$(D$_{2}$) = 0.0045$^{+0.005}_{-0.0023}$ cm$^{2}$ s$^{-1}$)]{} [The strong decrease of the diffusion coefficient at low dust particle temperatures exponentially increases the diffusion times in astrophysical environments. Therefore, transient dust heating by cosmic rays needs to be invoked for the release of the photo-produced H$_{2}$ molecules in cold PDR regions, where destruction of the aliphatic component in hydrogenated amorphous carbons most probably takes place.]{} Introduction {#intro} ============ Dust particles in the interstellar medium (ISM) include minerals (silicates and oxides) and carbonaceous matter of various types. In the diffuse ISM, carbonaceous solids are observed through both emission and absorption bands in the mid-infrared (mid-IR) region of the spectrum. The so-called aromatic infrared bands (AIBs) are a group of emission bands ubiquitously observed notably at 3.3, 6.2, 7.7, 8.6, and 11.3 $\mu$m, generally associated with the infrared fluorescence of polycyclic aromatic hydrocarbons [PAHs, @leger84; @allamadola85] upon absorption of ultraviolet (UV) photons. The observed AIBs spectral variabilities have been classified in three main types (named A, B, and C), reflecting the evolution of the carriers in the environment [@peeters02; @vanDied04], which would account for 4-5% of the total cosmic carbon abundance [@draine07]. An absorption band observed at 3.4 $\mu$m toward several lines of sight [@soifer76; @allen80; @mcfadzean89; @adamson90; @sandford91; @pendleton94; @bridger94; @imanishi00a; @imanishi00b; @spoon04; @dartois07] has been assigned to hydrogenated amorphous carbons (a-C:Hs or HACs), that harbor 5-30% of the cosmic carbon, depending on the assumed carrier [see, e.g., @sandford91]. The 3.4 $\mu$m ($\sim$2900 cm$^{-1}$) feature arises from the contribution of the symmetric and asymmetric C-H stretching modes of the methyl (-CH$_{3}$) and methylene (-CH$_{2}$-) groups. The corresponding bending modes are also observed at 7.25 $\mu$m ($\sim$1380 cm$^{-1}$), and 6.85 $\mu$m ($\sim$1460 cm$^{-1}$), respectively. These bands are accompanied by a broad absorption between 6.0 and 6.4 $\mu$m attributed to aromatic and olefinic C=C stretching modes, although the carrier is predominantly aliphatic [@dartois07]. Several laboratory analogs have been proposed to fit these observed IR features [@schnaiter98; @lee93; @furton99; @mennella99; @mennella03; @dartois05; @godard10; @godard11]. In this paper, we focus on plasma-produced a-C:H analogs [see @godard10; @godard11 and Section \[lab\]]. The 3.4 $\mu$m absorption band is not detected, though, in the dense ISM. Laboratory simulations on the energetic processing of HAC analogs under astrophysically relevant conditions have resulted in a decrease of their hydrogen content and, therefore, of the aliphatic C-H spectral features [see, e.g., @mennella03; @godard11; @alata14; @alata15 and references therein]. This energetic processing is driven by the interaction of the hydrogenated amorphous carbon particles with ultraviolet (UV) photons in the diffuse ISM [dehydrogenation by cosmic rays is negligible in these regions, see @mennella03] while destruction by cosmic rays (directly, or indirectly through the generated secondary UV field) dominates in the dense ISM. The presence of atomic hydrogen allows re-hydrogenation in the diffuse ISM. However, this process is inhibited in the dense ISM, possibly leading to a more aromatic carbonaceous solid (amorphous carbons or a-Cs) if not fully destroyed at earlier times. The destruction/transformation of the aliphatic C-H component most probably takes place in intermediate regions such as translucent clouds or photo-dominated regions, where dehydrogenation by both UV photons and cosmic rays is still active while hydrogen is in molecular form, thus not allowing re-hydrogenation [@godard11]. Hydrogen atoms resulting from the rupture of C-H bonds recombine to form H$_{2}$ molecules. Molecular hydrogen subsequently diffuses to the surface and is lost from the a-C:H [@wild87; @moller87; @adel89; @maree96; @godard11; @alata14]. This process thus constitutes an alternative pathway for H$_{2}$ formation within the bulk of the hydrogenated amorphous carbon particles in the ISM, in addition to the previously studied formation on the surface of interstellar solids from physisorbed and/or chemisorbed H atoms [see, e.g., @pirronello97; @katz99; @habart05; @cazaux11; @gavilan12]. Although molecular hydrogen is the main product resulting from the photolysis of a-C:H analogs, small hydrocarbons with up to four carbon atoms are also detected [@alata14; @alata15]. Production of these small hydrocarbons in the bulk of carbonaceous solids is proposed as an additional source that may account for the abundance of these species in some regions where pure gas-phase models face difficulties in predicting them [@pety05; @alata15]. In this work, we investigate the diffusion of molecular hydrogen through the plasma-produced hydrogenated amorphous carbon analogs under astrophysically relevant conditions. Deuterated analogs have been preferently used to avoid confusion with background H$_{2}$ during the first experiments. We have focused on the variation of the diffusion coefficient as a function of the temperature and, in particular, we have estimated $D_{0}$ and $E_{D}$ for the diffusion of H$_{2}$ (D$_{2}$) molecules through the a-C:H (a-C:D) analogs. The paper is organized as follows: Section \[lab\] describes the experimental setup and the theoretical models used to evaluate the diffusion coefficient from the experiments. Section \[results\] presents the experimental results, while their astrophysical implications are discussed in Sect. \[astro\]. Finally, conclusions are summarized in Sect. \[conclusiones\]. Methods {#lab} ======= SICAL-X {#sicalx} ------- The majority of the experiments were carried out using the SICAL-X setup described in previous papers [see, e.g., @alata14] at the Institut d’Astrophysique Spatiale. The SICAL-X setup consists in a high-vacuum (HV) chamber with a base pressure of about 2 $\times$ 10$^{-8}$ mbar. At this pressure, residual H$_{2}$ inside the chamber is not negligible. Therefore, deuterated amorphous carbon analogs were preferentially used to study diffusion of in situ photo-produced D$_{2}$ within this material. The a-C:D analogs produced in a different setup (SICAL-P, see Sect. \[sicalp\]) on a MgF$_{2}$ substrate were introduced in the chamber and cooled down to the working temperature thanks to the combination of a closed-cycle helium cryostat and a resistive-type heater. The sample temperature was monitored with a thermocouple, reaching a sensitivity of 0.1 K. Solid samples were monitored with a Bruker Vertex 80v Fourier transform infrared (FTIR) spectrometer. Spectra were collected at a resolution of 1 cm$^{-1}$, covering the spectral region between 7500 and $\sim$1500 cm$^{-1}$ (due to the low transmittance of the MgF$_{2}$ substrate below $\sim$1500 cm$^{-1}$). The analogs thicknesses $l$ were estimated from the interference pattern (or fringes) in the IR spectra, using the formula $$l = 1/(2 \cdot n \cdot \Delta\sigma \cdot cos(\alpha_{IR})), \label{fringes}$$ where $n$ is the refractive index of the analogs, estimated to be 1.7 $\pm$ 0.2, $\Delta\sigma$ the fringe spacing, and $\alpha_{IR}$ the angle of incidence of the IR beam with the sample normal (45$^{\circ}$ in SICAL-X). To study the diffusion of D$_{2}$ molecules through the a-C:D analogs, samples were irradiated with vacuum-ultraviolet (VUV) photons, leading to the dissociation of C-D bonds and the formation of D$_{2}$ in the analogs. Photochemical properties of a-C:H and a-C:D analogs are expected to be similar [see @alata14]. UV photons reached the surface of the analogs directly in contact with the MgF$_{2}$ substrate, and the photo-produced D$_{2}$ molecules diffused through the samples to the opposite surface, subsequently passing into the gas phase (see Fig. \[setup\]). The mean penetration depth for the VUV photons in the a-C:D analogs was approximately 80 nm as measured with VUV-dedicated experiments [@gavilan15]. Since the plasma-produced analogs had a thickness of around 1 $\mu$m, the region affected by the VUV photons was negligible compared to the total thickness of the sample, which remained almost unaltered. The surface of the a-C:D analogs was around 1 cm$^{2}$, much larger than the thickness of the samples. Therefore, only diffusion in the direction orthogonal to the substrate was considered. ![image](setup.eps){width="12cm"} UV irradiation was performed using a microwave-discharged hydrogen flow lamp (MDHL), with a VUV-photon flux at the sample position of 2.7 $\times$ 10$^{14}$ photons cm$^{-2}$ s$^{-1}$, measured by actinometry [@alata14]. Hydrogen pressure was set to 0.7 mbar in the lamp. The VUV spectrum of the MDHL has been studied in previous papers [see, e.g., @gustavo14; @alata14 and references therein]. An MgF$_{2}$ window was used as the interface between the lamp and the chamber interior. This window, along with the substrate, led to a cutoff in the lamp emission at about 115 nm. The mean photon energy was 8.6 eV. A 20 mm diameter metallic shutter was placed 5 cm away in front of the substrate to prevent the VUV photons from reaching the sample when necessary, without turning off the lamp and changing the conditions inside the chamber during the experiments (blank conditions). The initial D$_{2}$ concentration throughout the a-C:D analogs was zero. Photo-production of D$_{2}$ at the irradiated a-C:D surface in contact with the MgF$_{2}$ substrate established a constant D$_{2}$ flux entering the samples, since the IR spectra of the analogs did not change after UV irradiation. Deuterium molecules that were subsequently diffused through the samples eventually reached the opposite surface, where they inmediately passed into the gas-phase. Deuterium concentration at the sample surface in contact with the gas phase was thus negligible during the experiments. D$_{2}$ molecules in the gas phase were detected by a Quadera 200 quadrupole mass spectrometer (QMS) located at $\sim$10 cm from the samples. A heated filament in the QMS produced a stable current of energetic electrons ($\sim$70 eV), which ionized the D$_{2}$ molecules by electron bombardment. Ions were subsequently detected by a secondary-emission multiplier (SEM) detector. The ion current corresponding to the m/z = 4 mass fragment provided a measure of the D$_{2}$ flux through the surface of the a-C:D analogs in contact with the gas phase, considering the background level as zero flux. Therefore, upon onset of the VUV irradiation, D$_{2}$ concentration throughout the a-C:D analogs, as well as the measured outgoing flux of D$_{2}$ molecules, increased to a steady-state value [see @alata14]. Once the steady state was reached, the shutter was placed in front of the sample, blocking the VUV photons and stopping the D$_{2}$ production, thus eliminating the entering D$_{2}$ flux at the sample surface in contact with the substrate. As a consequence, the measured m/z = 4 ion current (i.e., the outgoing D$_{2}$ flux) decreased back to the background value. The evolution of the measured D$_{2}$ flux with time during irradiation of the sample (flux increasing to a steady-state value), and after irradiation (when D$_{2}$ formation is stopped and the m/z = 4 ion current decreases from the steady-state back to the background value) depends on the diffusion coefficient $D$ of the D$_{2}$ molecules through the a-C:D analogs. The diffusion coefficient can thus be estimated from the one-dimensional diffusion models that provide outgoing flux values that best fit the measured ion currents during the experiment (see Sect. \[modelo\]). Since the deuterium molecules were produced in situ in the a-C:D analogs during irradiation of the samples, we could not a priori disentangle the different steps taking place in the process: rupture of the C-D bonds, diffusion of the D atoms and recombination of two D atoms or direct neighbor D-abstraction from a C-D bond to form D$_{2}$ molecules, and the diffusion process itself. Therefore, the derived diffusion coefficients should be seen, in principle, as so-called apparent coefficients, describing the convolution of all these steps. However, D$_{2}$ molecules passing into the gas phase were detected very early on, once UV irradiation was established, subsequently increasing the observed flux with time (see Figures \[irradiationhac\], \[irradiationhac2\], and left panel of Fig. \[ajuste\], in Sect. \[results\] where the experimental results are presented). This means that all processes prior to the diffusion of molecules were probably taking place in a much shorter timescale than the diffusion itself, which could be considered the limiting step. In particular, Fig. \[irradiationhac2\] in Sect. \[results\] shows the symmetry between the measured m/z = 4 ion current during irradiation (increasing curve) and the decreasing signal observed once the UV beam was blocked after reaching the steady-state (when the diffusion equilibrium is achieved and the film is full of D$_{2}$). This shows that the measurement is dominated by the diffusion step with respect to the molecule formation timescale. Otherwise, if D$_{2}$ formation was the limiting step, we would expect to observe a delay and asymmetry in the m/z = 4 signal for the increasing curve owing to the D$_{2}$ formation limiting steps at the beginning, compared to the decreasing curve, when the beam is off and the bulk of the film is full of of previously formed D$_{2}$ molecules. Other measured behaviors supported the fact that the D$_{2}$ formation steps occurred at much shorter timescales than the diffusion step for the film thicknesses used and at the temperatures we performed the experiments, as explained below. The diffusion coefficient $D$ of a diffusing species through a given material is temperature dependent. Therefore, consecutive experiments following the above explained protocol were carried out with the same sample at different temperatures, allowing the evaluation of the diffusion coefficient as a function of the temperature. After every experiment, the a-C:D analogs were warmed up to 250 K with a heating rate of 5 K/min, to evacuate the eventual remaining deuterium from the sample, and set the D$_{2}$ concentration back to zero before performing the experiment at a different temperature. According to Equation \[diffT\] (see Sect. \[modelo\]), the diffusion coefficient increases with increasing temperature. Therefore, the diffusion time decreases with increasing temperature for a given sample thickness. At the same time, diffusion of deuterium through thicker samples takes longer, since molecules have to go through a longer distance to reach the analog surface in contact with the gas phase. A set of a-C:D analogs with different thicknesses were used to study diffusion in a wide range of temperatures while keeping the duration of the experiments within reasonable limits. In this way, diffusion at low temperatures (95 K - 140 K) was preferentially probed with thinner samples (0.2 $\mu$m - 2.6 $\mu$m), since diffusion times were shorter despite the lower diffusion coefficient; while diffusion at high temperatures (110 K - 170 K) was studied with thicker analogs (3.4 $\mu$m - 5.2 $\mu$m). When thinner films were irradiated at the same temperature (i.e., the diffusion length was short and bulk diffusion timescale was rapid), the molecular D$_{2}$ release was observed immediately after turning the UV lamp on, and stopped immediately after switching it off, with no delay that would otherwise indicate a longtime scale limiting the D$_{2}$ formation step (meaning with timescales of the order of seconds, which was the QMS scanning time). These test measurements did not allow us to measure the bulk diffusion and are thus not shown in Sect. \[results\]. In addition, the diffusion timescales changed with the thickness according to what was expected for different films at the same temperature (see above), whereas the production rate of D$_{2}$ on one side was confined to the same small thickness ($\sim$80 nm, as explained above) for all films, supporting the fact that diffusion was the limiting step. ISAC ---- Complementary experiments were carried out with a-C:H analogs (also produced in the SICAL-P setup on MgF$_{2}$ substrates, see Sect. \[sicalp\]) using the ISAC setup [@munozcaro10], to evaluate the diffusion coefficient $D$ of H$_{2}$ molecules through hydrogenated amorphous carbon analogs as a function of the temperature, Since H$_{2}$ molecules are smaller than D$_{2}$ molecules, diffusion coefficient of the former is expected to be higher than that of the latter at a given temperature. However, dependence of the diffusion coefficient of H$_{2}$ through the a-C:H analogs as a function of the temperature is expected to be similar to that of the diffusion coefficient of D$_{2}$ through a-C:D analogs, leading to similar $E_{D}$ values. The ISAC setup consists in an ultra-high-vacuum (UHV) chamber with a base pressure of about 4 $\times$ 10$^{-11}$ mbar, three orders of magnitude lower than that of the SICAL-X setup (see Sect. \[sicalx\]), thus enabling us to work with H$_{2}$ instead of D$_{2}$. The a-C:H analogs were introduced in the chamber and cooled down to the working temperature, also using a closed-cycle helium cryostat and a resistive-type heater. The temperature was controlled thanks to a silicon-diode sensor and a LakeShore Model 331 controller, reaching a sensitivity of 0.1 K. As in the SICAL-X setup, solid samples were monitored with a Bruker Vertex 70 FTIR spectrometer. Spectra were collected with a spectral resolution of 2 cm$^{-1}$, covering the range between 6000 and $\sim$1500 cm$^{-1}$. The angle of incidence of the IR beam with the sample normal was 0$^{\circ}$ in this setup. Diffusion of H$_{2}$ molecules through the a-C:H analogs were studied using the experimental protocol described in Sect. \[sicalx\]. The VUV photon flux of the MDHL at the sample position was about 2 $\times$ 10$^{14}$ photons cm$^{-2}$ s$^{-1}$, measured by CO$_{2}$ $\to$ CO actinometry [@munozcaro10]. A Pfeiffer Prisma QMS with a Channeltron detector located at $\sim$17 cm apart from the sample was used to detect the H$_{2}$ molecules in the gas phase, which were also ionized by electron bombardment with energetic ($\sim$70 eV) electrons. The ion current of the m/z = 2 mass fragment provided a measure of the outgoing H$_{2}$ flux. No shutter was present in the ISAC setup, and we had to turn off the lamp after the steady state was reached to stop the production of H$_{2}$ molecules. Subsequent changes in the background m/z = 2 ion current were taken into account. SICAL-P {#sicalp} ------- The a-C:H and a-C:D analogs were prepared by a plasma-enhanced vapor chemical deposition (PECVD) method [@godard10; @godard11], using CH$_{4}$ and CD$_{4}$ as gas precursors, respectively. Radicals and ions of the low pressure plasma produced by radio frequency (RF) at 2.45 GHz were deposited on a MgF$_{2}$ substrate. The precursor pressure was kept at $\sim$1 mbar in the SICAL-P vacuum chamber, and the power applied to the plasma was set to 100 W for all samples. Consequently, structure of all plasma-produced analogs was expected to be similar. Deposition time ranged from a few seconds to half an hour, leading to a wide range of sample thicknesses. Hydrogenated and deuterated samples were subsequently transferred to the ISAC or the SICAL-X setup, respectively. Theoretical models {#modelo} ------------------ Produced H$_{2}$ in the bulk of the a-C:H particles diffuses to the surface and is subsequently released from the solid. Diffusion of H$_{2}$ in a particular direction inside the HAC material can be described by the Fick’s first law: $$F(x,t) = - D \cdot \frac{\partial C(x,t)}{\partial x}, \label{fick1}$$ where $F(x,t)$ is the rate of transfer of H$_{2}$ molecules through unit area of HAC section (i.e., the H$_{2}$ flux), in cm$^{-2}$ s$^{-1}$, $D$ is the diffusion coefficient of molecular hydrogen in the hydrogenated amorphous carbon material, in cm$^{2}$ s$^{-1}$, and $\tfrac{\partial C(x,t)}{\partial x}$ is the concentration gradient of H$_{2}$ molecules in the particular direction $x$, in cm$^{-4}$. The negative sign indicates that diffusion takes place in the direction of decreasing concentration. The diffusion coefficient $D$ usually depends on the diffusing species, the material through which it is diffusing, and the temperature. The dependence of $D$ with the temperature generally follows an Arrhenius-type equation: $$D = D_{0} \cdot e^{\tfrac{-E_{D}}{T}}, \label{diffT}$$ where the pre-exponential factor $D_{0}$ is the diffusion at infinite temperature, in cm$^{2}$ s$^{-1}$, $E_{D}$ is the activation energy for the diffusion process in K, and $T$ the temperature in K. As mentioned in Sect. \[sicalx\], the diffusion coefficient $D$ can be estimated from one dimensional diffusion models that provides outgoing flux values that best fit the measured H$_{2}$ or D$_{2}$ ion currents during the experiments. Diffusion models are particular solutions of the fundamental differential equation of diffusion, which is also known as the Fick’s second law: $$\frac{\partial C(x,t)}{\partial t}=D \cdot \frac{\partial^{2} C(x,t)}{\partial x^{2}}. \label{fick2}$$ This equation results from combining Equation \[fick1\] with a mass balance equation applied to a differential element of volume of the material through which diffusion is being studied. Particular solutions of Equation \[fick2\] depend on the initial and surface conditions, leading to a series of diffusion models. Concentration of H$_{2}$ or D$_{2}$ in our samples during UV-irradiation (i.e., when a constant flux $F_{i}$ is established at the irradiated surface ($x = 0$) of a sample of thickness $l$, the initial H$_{2}$ or D$_{2}$ concentration is zero throughout the sample, and concentration at $x=l$ is $C(l,t)=0$ at all times) is modelled by the following equation [@early78]: $$\begin{split} C(x,t) = \frac{F_{i}}{D} (l - x) - \frac{8 F_{i} l}{\pi^{2} D} \times \sum_{n = 0}^{\infty} & \frac{(-1)^{n}}{(2n + 1)^{2}} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 l^{2}}]\\ \cdot & sin\frac{(2n+1) \pi (l - x)}{2l}. \label{csubida1b} \end{split}$$ Using Eq. \[fick1\] at $x=l$, the outgoing flux measured by the QMS, according to this model, can be calculated from Equation \[csubida1b\]: $$F(l,t) = 1 - \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 l^{2}}], \label{subida1b}$$ normalized to the steady-state value of the outgoing flux ($F(l,\infty) = 1$). On the other hand, concentration of H$_{2}$ or D$_{2}$ in the amorphous carbon analogs after UV-irradiation (i.e., when no flux is introduced at the surface $x = 0$ of a sample of thickness $l$, the initial H$_{2}$ or D$_{2}$ concentration is $C_{0}$ at $x=0$, and concentration at $x=l$ is $C(l,t)=0$ at all times) is modelled by the following equation [@carslaw]: $$\begin{split} C(x,t) = \frac{8 C_{0}}{\pi^{2}} \times \sum_{n = 0}^{\infty} & \frac{1}{(2n + 1)^{2}} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 l^{2}}]\\ \cdot & cos\frac{(2n+1) \pi x}{2l}. \label{cbajada1D} \end{split}$$ According to this model, the outgoing flux measured by the QMS can be derived from Equation \[cbajada1D\] using Equation \[fick1\] at $x=l$: $$F(l,t) = \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 l^{2}}], \label{bajada1D}$$ normalized to the initial (steady-state) value ($F(l,0) = 1$). We note that Eq. \[subida1b\] = 1 - Eq. \[bajada1D\]. As shown in Fig. \[irradiationhac2\], evolution of the H$_{2}$ or D$_{2}$ ion current after irradiation is symmetric with respect to the rise of the signal during irradiation. In our case, evolution of the m/z = 2 and m/z = 4 ion currents during the experiments was found to be better described by adding a second term in Equations \[subida1b\] and \[bajada1D\], introducing an additional diffusion coefficient $D'$. This led to $$\begin{split} F(l,t) =1 - &p \cdot \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 l^{2}}]\\ + &(1 - p) \cdot \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D' \cdot t}{4 l^{2}}], \label{subida2D} \end{split}$$ and $$\begin{split} F(l,t) = & p \cdot \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 l^{2}}]\\ + & (1-p) \cdot \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D' t}{4 l^{2}}], \label{bajada2D} \end{split}$$ respectively, where $p$ is a free parameter varying between 0.5 and 1, indicative of how close our modified model is to the models described with Equations \[subida1b\] and \[bajada1D\]. In this work, we refer to the diffusion coefficient $D$ when the opposite is not specified. The physical meaning of the additional diffusion coefficient $D'$ was studied independently in Sects. \[diffd2\] and \[diffh2\]. Experimental results {#results} ==================== Two different a-C:H analogs with thicknesses 0.9 $\mu$m and 2.2 $\mu$m were produced under the same conditions in the SICAL-P setup and subsequently studied in the ISAC setup. Diffusion experiments were carried out at 85 K, 95 K, 105 K, and 110 K for the former sample, and at 95 K, 105 K, 110 K, and 120 K for the latter (thinner samples allow the study of diffusion at lower temperatures, since diffusion of molecules through these samples takes less time even though the diffusion coefficient is lower, see Sect. \[sicalx\]). In the case of the a-C:D analogs, up to five samples with thicknesses 0.2 $\mu$m, 1.3 $\mu$m, 2.6 $\mu$m, 3.4 $\mu$m, and 5.2 $\mu$m were studied in the SICAL-X setup. Experiments were performed at a wide range of temperatures (from 95 K to 170 K). As explained in Sect. \[isac\], diffusion coefficients for the D$_{2}$ molecules in the a-C:D analogs are lower than those of the H$_{2}$ molecules through the a-C:H analogs at the same temperature. Therefore, to keep the duration of the experiments within reasonable limits, experiments with deuterated analogs were carried out at slightly higher temperatures than those performed with hydrogenated analogs of similar thickness. IR spectra of the a-C:H and a-C:D analogs {#ir} ----------------------------------------- ![IR transmittance spectrum of an a-C:H analog (left panel), and an a-C:D analog (right panel) deposited on an MgF$_{2}$ substrate.Spectra were collected at 110 K.[]{data-label="IR"}](IR.ps){width="9.25cm"} Figure \[IR\] shows the mid-IR transmittance spectra of an a-C:H analog (left panel, $l$ = 2.2 $\mu$m), and an a-C:D analog (right panel, $l$ = 2.6 $\mu$m), collected at 110 K, between 4000 cm$^{-1}$ and 1300 cm$^{-1}$. Transmittance of the MgF$_{2}$ substrate starts decreasing strongly below 1500 cm$^{-1}$. As explained in Sect. \[intro\], the asymmetric C-H stretching modes corresponding to methyl (-CH$_{3}$) and methylene (-CH$_{2}$-) groups led to two absorption peaks at 2955 cm$^{-1}$ and 2925 cm$^{-1}$, respectively, in the left panel of Fig. \[IR\]. The symmetric stretching modes at 2873 cm$^{-1}$ and 2857 cm$^{-1}$, respectively, are blended, which leads to one absorption peak in the red side of the 3.4 $\mu$m absorption band. The corresponding bending modes are observed at 1460 cm$^{-1}$ and 1380 cm$^{-1}$ for the methylene and methyl groups, respectively. Absorption at $\sim$1600 cm$^{-1}$ is assigned to the C=C stretching mode, which corresponds to the olefinic fraction of the analog. The same modes are shifted to lower frequencies in the a-C:D analogs. The assymetric stretching modes, located at 2220 cm$^{-1}$ and 2200 cm$^{-1}$, are blended in the right panel of Fig. \[IR\], as well as the symmetric modes at 2073 cm$^{-1}$ and 2100 cm$^{-1}$. The corresponding bending modes are also shifted to lower frequencies, where the absorption of the MgF$_{2}$ substrate prevents their detection. Since only a small fraction of the analogs were photoprocessed during the experiments (see Sect. \[sicalx\]), IR spectra did not change after UV irradiation, which is a condition to the hypothesis of a constant flux at $x = 0$. QMS measurements of the outgoing H$_{2}$ or D$_{2}$ flux -------------------------------------------------------- As previously explained, the measured ion current of the mass fragments m/z = 2 and m/z = 4 above the background level provided a measure of the outgoing H$_{2}$ and D$_{2}$ fluxes through the $x=l$ surface of the hydrogenated and deuterated amorphous carbon analogs, respectively, during the experiments. ![Measured m/z = 4 ion current corresponding to the outgoing D$_{2}$ flux during three experiments performed at $\sim$140 K (red solid line), $\sim$130 K (yellow solid line), and $\sim$120 K (blue solid line) on a 3.4 $\mu$m thick a-C:D analog. Irradiation starts at $t=0$.[]{data-label="irradiationhac"}](irradiation_HAC_aa.ps){width="9.25cm"} Figure \[irradiationhac\] shows the m/z = 4 ion current measured during three experiments performed at three different temperatures with a 3.4 $\mu$m thick a-C:D analog, as an example. After onset of the UV irradiation at $t=0$, the outgoing flux of the photo-produced D$_{2}$ molecules increased rapidly at first, and then more slowly, until the steady state was reached. Diffusion times depended on the diffusion coefficient $D$. At higher temperatures, a higher $D$ value led to a faster diffusion, and therefore, to a faster increase of the m/z = 4 ion current, while diffusion at lower temperatures was slower. The ion current value at the steady state depends not only on the diffusion coefficient (which in turn depends on the temperature), but also on the VUV photon flux reaching the sample, which may change slightly from one experiment to another. Parallel experiments with different VUV photon fluxes were carried out to check that the derived value of the diffusion coefficient at a given temperature did not vary with the value of the UV flux. Once the steady state was reached, the shutter was placed in front of the sample, preventing the VUV photons from reaching the analog, and thus stopping the D$_{2}$ production. The m/z = 4 ion current then decreased back to the background value, again rapidly at first and more slowly later. As for the increase of the outgoing D$_{2}$ flux, decay time also depended on the diffusion coefficient. As expected from Equations \[subida1b\] and \[bajada1D\], once the background level was subtracted, the normalized ion current (which is equivalent to the normalized D$_{2}$ flux) during irradiation was equal to unity minus the normalized ion current during decay of the D$_{2}$ flux (see Fig. \[irradiationhac2\] as an example). ![Normalized m/z = 4 ion current after background substraction (equivalent to the normalized D$_{2}$ flux) during three experiments performed at $\sim$140 K (left panel), $\sim$130 K (middle panel), and $\sim$120 K (right panel) on a 3.4 $\mu$m thick a-C:D analog. Black solid lines correspond to the normalized D$_{2}$ flux during irradiation (irradiation starts at $t=0$), while colored solid lines correspond to 1 - normalized D$_{2}$ flux after irradiation (irradiation stops at $t=0$).[]{data-label="irradiationhac2"}](irradiation_HAC_aa_2.ps){width="9.25cm"} Evolution of the normalized ion current (i.e., normalized H$_{2}$ or D$_{2}$ flux through the analog surface $x=l$) during and after irradiation is modeled by Equations \[subida1b\] and \[bajada1D\], respectively. The value of $D,$ which best fits the measured ion currents during the experiments, would be the diffusion coefficient of H$_{2}$ (D$_{2}$) molecules through the a-C:H (a-C:D) analogs at a given temperature. Evolution of the diffusion coefficient with the temperature is described by Eq. \[diffT\]. Most of the experiments were carried out with a-C:D analogs in the SICAL-X setup. Derived $D$ values for the D$_{2}$ molecules at different temperatures enabled us to calculate the pre-exponential factor $D_{0}$ and the activation energy $E_{D}$ for the diffusion of D$_{2}$ through the deuterated analogs. Results are presented in Sect. \[diffd2\]. Complementary experiments were performed with two a-C:H analogs in the ISAC setup. Calculated $D_{0}$ and $E_{D}$ values are presented in Sect. \[diffh2\]. $E_{D}$ was expected to be similar in both cases. ### Modeling of the D$_{2}$ diffusion through a-C:D analogs {#diffd2} ![image](ajuste_130K_aa.ps){width="18.5cm"} As explained above, we estimated for every experiment a diffusion coefficient $D$ for the diffusion of D$_{2}$ molecules through the a-C:D analogs at a given temperature by fitting the models described with Equations \[subida1b\] and \[bajada1D\] to the experimental evolution of the normalized m/z = 4 ion current during and after the irradiation of the sample, respectively, with the linfit procedure programmed with the IDL programming language. This procedure finds the free parameter $D$ which minimizes the $\chi^{2}$ parameter. Left panel of Fig. \[ajuste\] shows the evolution of the normalized m/z = 4 ion current, i.e., the normalized outgoing D$_{2}$ flux, during irradiation of a 2.6 $\mu$m a-C:D analog at 130 K, along with the model described by Eq. \[subida1b\] that best fits the experimental results. The diffusion coefficient derived for that experiment is also shown in the figure. This value is the same within 10% to the diffusion coefficient derived from the decay of the normalized outgoing D$_{2}$ flux in the same experiment (middle panel of Fig. \[ajuste\]). In this case, instead of subtracting the background level of the m/z = 4 ion current, we included it as a free parameter $b$ in Eq. \[bajada1D\]: $$F(l,t) = (1 - b) \cdot \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 l^{2}}] + b. \label{bajada1Dback}$$ We note that $\chi^{2}$ values in the left and middle panels of Fig. \[ajuste\] cannot be directly compared since they are not normalized. However, since there is a larger dispersion in the experimental data collected at the steady state than in the background level, we consider the diffusion coefficient derived from the decay curves more accurate, and they have thus been used preferently. As explained in Sect. \[modelo\], experimental data were better described by a modified model including a second term with an additional diffusion coefficient $D'$. Adding a second term to Eq. \[bajada1Dback\] results in $$\begin{split} F(l,t) = & p \cdot \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 l^{2}}]\\ + & (1-p-b) \cdot \frac{4}{\pi} \times \sum_{n = 0}^{\infty}\frac{(-1)^{n}}{(2n + 1)} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D' t}{4 l^{2}}]\\ + & b, \label{bajada2Dback} \end{split}$$ which reduced the $\chi^{2}$ value of the fit (see right panel of Fig. \[ajuste\]). Estimated diffusion coefficients $D$ with the modified model were of the same order as those found with the model described by Eq. \[bajada1Dback\]. Values of the free parameter $p$ usually varied between $\sim$0.7 and $\sim$0.9. The physical meaning of the $D'$ diffusion coefficient was studied independently (see below). The estimated 1$\sigma$ errors presented in Fig. \[ajuste\] for the diffusion coefficient $D$ are model-dominated. We note that diffusion coefficients derived with different experiments at the same temperature can vary up to a factor $\sim$6 (see Fig. \[ajusteT\]). This was taken into account for the $D_{0}$ and $E_{D}$ estimation. A total of 19 experiments at temperatures between 95 K and 170 K were carried out with five a-C:D analogs. The diffusion coefficient at intermediate (120 K - 140 K) temperatures was probed with most of the analogs, while $D$ value at low (high) temperatures was probed only with thin (thick) analogs. As explained in Sect. \[intro\], dependence of the diffusion coefficient $D$ with the temperature follows an Arrhenius-type equation. Equation \[diffT\] changes into a line by applying natural logarithm to both sides of the equation $$ln[D] = ln[D_{0}] - E_{D} \cdot \frac{1}{T}, \label{diffTlin}$$ where the slope $E_{D}$ corresponds to the activation energy of the diffusion process. Fitting Eq. \[diffTlin\] to the derived diffusion coefficients $D$ from the experiments with the linfit procedure programmed with the IDL programming language led to an activation energy of $E_{D}$ = 2090 $\pm$ 90 K for the diffusion of D$_{2}$ molecules through the a-C:D analogs, and $D_{0}$ = 0.0045$^{+ 0.0050}_{- 0.0023}$ cm$^{2}$ s$^{-1}$ for the pre-exponential factor. This activation energy falls in the same range as that found in @falconneche01 for the diffusion of rare gases and small molecules through carbonaceous polymer materials. Figure \[ajusteT\] shows the evolution of the experimental and modeled diffusion coefficient with the temperature, along with the associated 3$\sigma$ limits. ![Evolution of the experimental diffusion coefficient $D$ for the diffusion of D$_{2}$ molecules through a-C:D analogs with the temperature (black circles), along with the model described by Eq. \[diffT\] that best fits the experimental data (blue solid line), and the associated 3$\sigma$ limits (blue dashed lines).[]{data-label="ajusteT"}](diffusionb1_aa_log.ps){width="9.25cm"} On the other hand, the additional diffusion coefficient $D'$ introduced to improve the model did not depend on the temperature, but on the thickness of the analogs, as seen in Fig. \[diff2\]. ![Additional diffusion coefficient $D'$ derived for the experiments with a-C:D analogs of thickness 5.2 $\mu$m (red circles), 3.4 $\mu$m (yellow circles), 2.6 $\mu$m (green circles), 1.3 $\mu$m (blue circles), and 0.2 $\mu$m (purple circles), using the model described by Eq. \[bajada2Dback\].[]{data-label="diff2"}](diffusionb2.ps){width="9.25cm"} The second term in Equations \[subida2D\] and \[bajada2D\] could be an artifact that accounts for the differences between the model described by Equations \[subida1b\] and \[bajada1D\], and the real process taking place during the experiments. For example, the entering D$_{2}$ flux established during irradiation is assumed to enter the sample at the surface $x = 0$ with infinitesimal thickness, while in the real process a finite (yet negligible) thickness of the sample is processed by the VUV photons. In which case, the diffusion coefficient $D'$ would have no physical meaning. Alternatively, this second term could account for a parallel diffusion process taking place along with the “main” studied diffusion represented by the coefficient $D$. In that case, the diffusion coefficient $D'$ would describe this parallel diffusion, which could be taking place, for example, through the pores or cracks of the samples. Since $D'$ increases with the analogs’ thickness (see Fig. \[diff2\]), pores or cracks should be larger in thicker plasma-produced analogs. ### Modelling of the H$_{2}$ diffusion through a-C:H analogs {#diffh2} A total of seven complementary experiments were carried out with 2 a-C:H analogs, at temperatures between 85 K and 120 K. The modified model described by Eq. \[bajada2Dback\] was also used to fit the normalized m/z = 2 ion current measured during the experiments. Estimated diffusion coefficients $D$ for the diffusion of H$_{2}$ molecules through the a-C:H analogs were found to be approximately one order of magnitude higher than those of D$_{2}$ molecules through a-C:D analogs measured at the same temperatures. However, dependence of the diffusion coefficient with temperature was similar in both cases. The estimated activation energy and diffusion at infinite temperature were $E_{D}$ = 1660 $\pm$ 110 K, and $D_{0}$ = 0.0007$^{+ 0.0013}_{- 0.0004}$ cm$^{2}$ s$^{-1}$, respectively (see Fig. \[ajusteTH\]). The errors could be slightly larger due to the assumptions made for the m/z = 2 background level changes when switching the VUV lamp on and off, since the metallic shutter used to block the VUV photons in the SICAL-X setup after irradiation was not available in the ISAC setup. The additional diffusion coefficient in Eq. \[bajada2Dback\] followed the same behavior as in Sect. \[diffd2\], as shown in Fig. \[diff2H\], with averaged values for similar thicknesses higher by a factor of $\sim$2. ![Evolution of the experimental diffusion coefficient $D$ for the diffusion of H$_{2}$ molecules through a-C:H analogs with the temperature (black circles), along with the model described by Eq. \[diffT\] that best fits the experimental data (blue solid line), and the associated 3$\sigma$ limits (blue dashed lines).[]{data-label="ajusteTH"}](diffusionb1H_aa_log.ps){width="9.25cm"} ![Additional diffusion coefficient $D'$ derived for the experiments with a-C:H analogs of thickness 2.2 $\mu$m (green circles), and 0.9 $\mu$m (blue circles), using the model described by Eq. \[bajada2Dback\].[]{data-label="diff2H"}](diffusionb2H.ps){width="9.25cm"} Astrophysical implications {#astro} ========================== The mobility and desorption of H$_{2}$ and D$_{2}$ on water ice, mineral, or graphite surfaces are high (see, e.g., Amiaud et al. 2015, Acharyya 2014, Vidali & Li 2010, Fillion et al. 2009, Haas et al. 2009, and references therein), with typical activation energies of only a few hundreds of K. This ensures the rapid molecular hydrogen desorption in high dust grain temperature regions, such as some PDRs, but at the same time it constitutes an issue when trying to explain the formation of H$_2$ molecules by the surface recombination of H atoms, since they may not stay physisorbed for long enough on the surface to recombine. Alternatively to these formation paths, a-C:Hs are energetically processed in the ISM, leading to the destruction of the aliphatic C-H component (which is not detected in dense regions), and the formation of hydrogen molecules. Recent models help in explaining the H$_{2}$ formation at such apparent high temperatures, because of the time fraction spent by small grains at low temperature between transient heating events [@bron16]. The subsequently formed hydrogen molecules diffuse out of the carbonaceous particles and contribute to the total H$_{2}$ abundance in the ISM [@alata14]. In the diffuse ISM, re-hydrogenation of the amorphous carbon particles by atomic H equilibrates the destruction of the aliphatic C-H component by UV photons (mainly) and cosmic rays. In the dense ISM, re-hydrogenation is much less efficient. However, the interstellar UV field cannot penetrate dense clouds, and destruction only by cosmic rays (directly or indirectly through the generated secondary UV field) cannot account for the disappearance of the 3.4 $\mu$m absorption band. @godard11 state that this dehydrogenation should therefore take place in intermediate regions such as translucent clouds or PDRs. In this scenario, when H$_{2}$ molecules are produced in the bulk by photolytic reactions, they diffuse out slowly outward of the grains contrary to the production and immediate release of H$_2$ that are formed on the surface of interstellar solids. Surface Eley-Rideal or Langmuir–Hinshelwood mechanism [see, e.g., @roser03; @hornekaer03; @islam07; @vidali07; @latimer2008; @cazaux2008; @mennella08; @vidali09; @lemaire10; @sizun2010; @vidali13; @hama13; @bron2014; @amiaud2015] led to the H$_{2}$ molecule desorption with debated amounts of internal excitation energy. By contrast, the diffusion process in the bulk allows the molecules to thermalize during their escape path toward the surface. One thus expects no high vibrational level excitation for these molecules contrary to the possible hydrogen surface recombination processes. Formation of hydrogen molecules and subsequent diffusion in the bulk of the a-C:H particles also enables us to thermalize the ortho-to-para (OPR) ratio to the dust temperature, which is lower than the gas temperature in PDRs. For dust temperatures of 55-70 K typical of a warm PDR such as the Orion Bar nebula [see, e.g., @guzman11 and references therein], an OPR of $\sim$1 is expected. Observed OPR values in PDRs are indeed around 1, lower than the value of $\sim$3 expected from the excitation temperature of the H$_{2}$ rotational lines [@fuente99; @habart03; @habart11]. Alternatively, @bron16 propose the ortho-to-para conversion of physisorbed H$_{2}$ molecules on dust grain surfaces to explain these low OPR values. This process is only efficient on cold dust grains and, therefore, dust temperature fluctuations need to be invoked in that case. Diffusion of H$_{2}$ through the a-C:H particles is characterized by the diffusion coefficient $D$, which depends on the dust temperature according to Eq. \[diffT\]. With the $D_{0}$ and $E_{D}$ values estimated in Sect. \[diffd2\] we can extrapolate $D$ values to the temperature of a PDR region, and obtain a typical decay-time constant for the release of the hydrogen molecules from the a-C:Hs in space. Interstellar a-C:Hs can be approximated to a slab of thickness $2r$, $r$ being the typical radius of a carbonaceous dust particle. If the a-C:H is initially filled with a concentration $C_{0}$ of H$_{2}$ molecules, and the molecules are released to the gas phase through the surfaces $x = -r$ and $x = r$ maintained at zero concentration, then the average hydrogen concentration in the slab at time $t$ is given by $$C_{av}(t) = \frac{4 C_{0}}{\pi^{2}} \times \sum_{n = 0}^{\infty}\frac{1}{(2n + 1)^{2}} \cdot exp[-\frac{(2n+1)^{2} \pi^{2} D t}{4 r^{2}}], \label{slab}$$ according to @carslaw. We define the decay-time constant $\tau$ as the time interval needed for the average hydrogen concentration in the slab to be 10% of the initial value (i.e., $\frac{C_{av}(\tau)}{C_{0}}=0.1$). Adopting a typical $\sim$0.1 $\mu$m grain radius, we calculated $\tau$ for two different temperatures: 30 K, representative of a cold PDR such as the Horsehead nebula; and 55 K, the lower limit of a warmer PDR such as the Orion Bar nebula [see, e.g., @guzman11 and references therein]. While at 55 K calculated $\tau_{55}$ is about 19 years, at 30 K $\tau_{30}$ would be about 1.1 $\times$ 10$^{15}$ years, since the diffusion coefficient depends strongly on the temperature. If we instead adopt half the typical radius for a dust grain ($r$ $\sim$ 0.05 $\mu$m), decay times are reduced to a quarter (see Eq. \[slab\]), leading to $\tau_{55}$ $\sim$ 5 years and $\tau_{30}$ $\sim$ 3 $\times$ 10$^{14}$ years. The latter decay-time is way longer than the typical dynamical time of a PDR [see, e.g., @goldsmith07; @glover07]. To decrease the desorption time of the photo-produced hydrogen molecules in the coldest interstellar regions, several solutions can be invoked. On one hand, transient heating episodes of the a-C:H particles by, for example, cosmic rays could increase the dust temperature (and therefore the diffusion of the H$_{2}$ molecules), thus reducing significantly the release time. On the other hand, the presence of open channels in the grains, or a higher surface-to-volume ratio than that of the compact films used in the laboratory simulations could also assist the hydrogen release from the a-C:H particles. Conclusions {#conclusiones} =========== We have explored the diffusion of photo-produced H$_{2}$ (D$_{2}$) molecules through a-C:H (a-C:D) analogs. Hydrogenated amorphous carbon particles (which harbor between 5% and 30% of the total C cosmic abundance) are energetically processed in the ISM, leading to the loss of the aliphatic C-H component and the formation of hydrogen molecules that diffuse out of the particles, contributing to the total H$_{2}$ abundance. This constitutes an alternative additional formation pathway to the hydrogen surface recombination process that allows the hydrogen molecules to thermalize to the dust temperature before passing into the gas phase, leading to no high vibrational level excitation, and to OPR values similar to those observed in PDRs. We have simulated this process in the laboratory using plasma-produced a-C:H and a-C:D analogs. The surface of the analogs in contact with the substrate was irradiated by VUV photons under astrophysically relevant conditions. Photo-produced H$_{2}$ and D$_{2}$ molecules subsequently diffused through the analogs, eventually reaching the opposite surface, and passing into the gas phase. Molecules released from the analogs were detected by a QMS. The measured m/z = 2 and m/z = 4 ion current corresponded to the outgoing H$_{2}$ or D$_{2}$ flux, respectively, which was compared to the expected flux from the diffusion model that best fitted the experimental measurements, enabling us to derive a diffusion coefficient that described the diffusion process. A modified diffusion model with two different diffusion coefficients, $D$ and $D'$ was used. The diffusion coefficient $D$ described the diffusion of the molecules through the HAC material, which depended on the temperature of the analogs, following an Arrhenius-type equation. Experiments at several temperatures were carried out, estimating a diffusion coefficient $D$ for every experiment. This allowed us to derive an activation energy $E_{D}$ of the diffusion process. Estimated $E_{D}$ was 1660 $\pm$ 110 K for the diffusion of H$_{2}$ through the a-C:H analogs, and 2090 $\pm$ 90 K for the diffusion of D$_{2}$ through the a-C:D analogs. The pre-exponential factors were also derived ($D_{0}$(H$_{2}$) = 0.0007$^{+0.0013}_{-0.0004}$ cm$^{2}$ s$^{-1}$, and $D_{0}$(D$_{2}$) = 0.0045$^{+0.005}_{-0.0023}$ cm$^{2}$ s$^{-1}$). The additional diffusion coefficient $D'$ did not depend on the temperature, but on the thickness of the analogs. This coefficient could trace the differences between the model and the real process taking place in the laboratory, or, alternatively, it could be accounting for a parallel diffusion process taking place, for example, through the pores or cracks of the analogs. Using these experimental values, we extrapolated the diffusion coefficient $D$ to two different temperatures representative of PDR regions, where the destruction of the C-H bonds and formation of H$_{2}$ molecules is expected to take place. A typical decay-time constant $\tau$ was calculated characterizing the release of the H$_{2}$ molecules from the a-C:H particles. Transient heating episodes of the dust particles or other alternative solutions need to be invoked for the release of the hydrogen molecules in cold regions where the typical diffusion times exceed the dynamical time of these regions. This research has been financed by the ANR and French INSU-CNRS program Physique et Chimie du Milieu Interstellaire (PCMI), and by the Spanish MINECO under projects AYA-2011-29375 and AYA2014-60585-P. R.M.D. benefited from a FPI grant from Spanish MINECO. The authors acknowledge funding support from the PICS (Projet International de Coopération Scientifique) between the CNRS and CSIC which consolidated this French and Spanish teams’ cooperation. We also thank the anonymous reviewer for constructive remarks Acharyya, K. 2014, , 443, 1301 Adamson, A. J., Whittet, D. C. B., & Duley, W. W. 1990, MNRAS, 243, 400 Adel, M. E., Amir, O., Kalish, R., & Feldman, L. C. 1989, J. Appl. Phys., 66, 3248 Alata, I., Cruz-Díaz, G. A., Muñoz Caro, G. M., & Dartois, E. 2014, A&A, 569, A119 Alata, I., Jallat, A., Gavilan, L., et al. 2015, A&A, 584, A123 Allamandola, L. J., Tielens, A. G. G. M., & Bakrer, J. R. 1985, ApJ, 290, L25 Amiaud, L., Fillion, J.-H., Dulieu, F., Momeni, A., & Lemaire, J.-L. 2015, Physical Chemistry Chemical Physics (Incorporating Faraday Transactions), 17, 30148 Bridger, A., Wright, G. S., & Geballe, T. R., 1994, in Infrared Astronomy with Arrays: The Next Generation, ed. I. S. McLean, Astrophysics and Space Science Library, 190, 537 Bron, E., Le Bourlot, J., & Le Petit, F. 2014, , 569, A100 Bron, E., Le Petit, F., & Le Bourlot, J. 2016,A&A, accepted Carslaw, H. S., & Jaeger, J. C. 1959, in Conduction of Heat in Solids, Oxford Univ. Press Cazaux, S., Caselli, P., Cobut, V., & Le Bourlot, J. 2008, , 483, 495 Cazaux, S., Morisset, S., Spaans, M., & Allouche, A. 2011, A&A, 535, A27 Cruz-Díaz, G. A., Muñoz Caro, G. M., Chen, Y.-J., & Yih, T.-S. 2014, A&A, 562, A119 Dartois, E., Muñoz Caro, G. M., Deboffle, D., Montagnac, G., & D’Hendecourt, L. 2005, A&A, 432, 895 Dartois, E., & Muñoz Caro, G. M. 2007, A&A, 476, 1235 Draine, B. T., & Li, A. 2007, ApJ, 657, 810 Early, J. G. 1978, Acta Metall., 26, 1215 Falconnèche, B., Martin, J., & Klopffer, M.H. 2001, Oil & Gas Science and Technology, 56, 271 Fillion, J.-H., Amiaud, L., Congiu, E., et al. 2009, Physical Chemistry Chemical Physics (Incorporating Faraday Transactions), 11, 4396 Fuente, A., Martín-Pintado, J., Rodríguez-Fernández, N. J., et al. 1999, ApJ, 518, L45 Furton, D. G., Laiho, J. W., & witt, A. N. 1999, ApJ, 526, 752 Gavilan, L., Alata, I., Le, K. C., et al. 2016, A&A, 586, A106 Gavilan, L., Lemaire, J. L., & Vidali, G. 2012, MNRAS, 424, 2961 Glover, S. C. O., & Mac Low, M. 2007, A&A, 463, 635 Godard, M., & Dartois, E. 2010, A&A, 519, A39 Godard, M., Féraud, G., Chabot, M., et al. 2010, A&A, 529, A146 Goldsmith, P. F., Li, D., & Krčo, M. 2007, ApJ, 654, 273 Guzman, V. V., Pety, J., Goicoechea, J. R., Gerin, M., & Roueff, E. 2011, A&A, 534, A49 Haas, O.-E., Simon, J. M., & Kjelstrup, S. 2009, J. Phys. Chem. C, 113, 20281 Habart, E., Abergel, A., Boulanger, F., et al. 2011, A&A, 527, A122 Habart, E., Boulanger, F., Verstraete, L., et al. 2003, A&A, 397, 623 Habart, E., Walsmley, M., Verstraete, L., et al. 2005, Space, Sci. Rev., 119, 71 Hama, T., & Watanabe, N. 2013, Chem. Rev., 113, 8783 Hornekaer, L., Baurichter, A., Petrunin, V. V., Field, D., & Luntz, A. C. 2003, Science, 302, 1943 Imanishi, M. 2000a, MNRAS, 313, 165 Imanishi, M. 2000b, MNRAS, 319, 331 Islam, F., Latimer, E. R., & Price, S. D. 2007, J. Chem. Phys., 127, 064701 Katz, N., Furman, I., Biham, O., Pirronello, V., & Vidali, G. 1999, ApJ, 522, 305 Latimer, E. R., Islam, F., & Price, S. D. 2008, Chemical Physics Letters, 455, 174 Lee, W., & Wdowiak, T. J. 1993, ApJ, 417, L49 Leger, A., & Puget, J. L. 1984, A&A, 137, L5 Lemaire, J. L., Vidali, G., Baouche, S., et al. 2010, ApJ, 725, L516 Marée, C., Vredenberg, A., & Habraken, F. 1996, Mater. Chem. Phys., 46, 198 McFadzean, A. D., Whittet, D. C. B., Bode, M. F., Adamson, A. J., & Longmore, A. J. 1989, MNRAS, 241, 873 Mennella, V., ApJ, 684, L25 Mennella, V., Baratta, G. A., Esposito, A., Ferini, G., & Pendleton, Y. J. 2003, ApJ, 587, 727 Mennella, V., Brucato, J. R., Colangeli, L., & Palumbo, P. 1999, ApJ, 524, L171 Möller, W., & Scherzer, B. M. U. 1987, Appl. Phys. Lett. 50, 1870 Muñoz Caro, G. M., Jiménez-Escobar, A., Martín-Gago, J.Á. et al. 2010, A&A, 522, A108 Peeters, E., Hony, S., van Kerckhoven, C., et al. 2002, A&A, 390, 1089 Pendleton, Y. J., Sandford, S. A., Allamandola, L. J., Tielens, A. G. G. M., & Sellgren, K. 1994, ApJ, 437, 683 Pety, J., Teyssier, D., Fossé, D., et al. 2005, A&A, 435, 885 Pirronello, V., Liu, C., Shen, L., & Vidali, G. 1997, ApJ, 475, L69 Roser, J. E., Swords, S., Vidali, G., Manico, G., & Pirronello, V. 2003, ApJ, 596, L55 Sandford, S. A., Allamandola, L. J., Tielens, A. G. G. M., et al. 1991, ApJ, 371, 607 Schnaiter, M., Mutschke, H., Dorschner, J., Henning, T., & Salama, F. 1998, ApJ, 498, 486 Sizun, M., Bachellerie, D., Aguillon, F., & Sidis, V. 2010, Chemical Physics Letters, 498, 32 Soifer, B. T., Russel, R. W., & Merrill, K. M., 1976, ApJ, 207, L83 Spoon, H. W. W., Armus, L., Cami, J., et al. 2004, ApJS, 154, 184 van Diedenhoven, B., Peeters, E., van Kerckhoven, C., et al. 2004, ApJ, 611, 928 Vidali, G. 2013, Chem. Rev., 113, 8752 Vidali, G., & Li, L. 2010, Journal of Physics Condensed Matter, 22, 304012 Vidali, G., & Li, L. Roser, J. E., & Badman, R. 2009, Adv. Space Res., 43, 1291 Vidali, G., Pirronello, V., Li, L., et al. 2010, J. Chem. Phys. A, 111, 12611 Wickramasinghe, D. T., & Allen, D.A. 1980, Nature, 287, 518 Wild, C. & Koidl, P. 1987, Appl. Phys. Let., 51, 1506	{ "pile_set_name": "ArXiv" }
--- abstract: 'We reconsider the Hung and Shuryak arguments in favour of searching for the deconfinement phase transition in heavy ion collisions [downward]{} from the nominal SPS energy, at $E_{lab} \approx 30 \ GeV/A$ where the fireball lifetime is the longest one. Using the recent lattice QCD data and the mixed phase model, we show that the deconfinement transition might occur at the bombarding energies as low as $E_{lab}=3 - 5 \ GeV/A$. Attention is drawn to the study of the mixed phase of nuclear matter in the collision energy range $E_{lab}= 2-10 \ GeV/A$.' author: - \| M. Płoszajczak$^{ \dagger}$, A.A. Shanenko $^{\ddagger}$ and V.D. Toneev$^{ \dagger , \ddagger} $\ $^{\dagger} $ Grand Accélérateur National d’Ions Lourds (GANIL), BP 5027,\ F-14021 Caen Cedex, France\ $^{ \ddagger}$ Bogoliubov Laboratory of Theoretical Physics, Joint Institute for Nuclear Research,\ 141980 Dubna, Moscow Region, Russia\ ** title: \| Towards a New Strategy\ of Searching for QCD Phase Transition\ in Heavy Ion Collisions --- -40pt [PACS codes]{} : 24.85.+p, 12.38.Aw, 12.38.Mh, 21.65.+f, 64.60.-i, 25.75.+r [Keywords*]{} : Deconfinement, Nuclear matter, Mixed phase model, Equation of State, Heavy ion collisions Over the last ten years, a fundamental QCD prediction of the phase transition from hadrons into a state of free quarks and gluons (quark-gluon plasma) has been studied actively. Extensive lattice QCD calculations not only allowed to specify the deconfinement temperature, the order of phase transition and its flavor dependence both for pure gluon matter and for the plasma with dynamical quarks, but also gave insight into physical phenomena above and below the deconfinement temperature. By now, there is a rather long list of various signatures which can signal the quark-gluon plasma formation. Many of these signatures were tested experimentally at CERN-SPS with the $^{16}O$ and $^{32}S$ beams at $ 200 \ GeV/A$ where crucial conditions for the deconfinement transition are expected to be reached. Indeed, some predicted effects such as the strangeness enhancement, $J/\Psi$ suppression or $\phi /(\rho +\omega )$ enhancement have been observed but their hadronic interpretation cannot be excluded. General belief is that for more definite conclusions [heavier ions]{} and [higher beam energies]{} should be used. Recently, a new strategy of experimental search for the QCD phase transition in heavy ion collisions has been advocated [@HS94]. The equation of state (EOS) is very ’soft’ in a narrow range of temperatures ($\Delta T \simeq 10 \ MeV$) around the transition temperature, leading to a significant reduction in [transverse]{} expansion of the fireball formed in heavy ion collisions [@SZ79]. This ’softness’ of the EOS, estimated by Hung and Shuryak at the energy density $\varepsilon_{sp} \simeq 1.5 \ GeV/fm^3$ and the beam energy $E_{lab} \approx 30 \ GeV/A$ [@HS94], affects not only the transverse but also the [longitudinal* ]{} expansion and results in a longest lifetime of the excited system. Radical changes in the hydrodynamic space-time evolution around the ’softest point’ of the EOS should lead for certain observables to a sharp and specific dependence on the [heavy-ion beam energy ]{}. Therefore, some aspect of the deconfinement transition might be better studied at lower collision energy, [downward ]{} from the nominal SPS energy. Hung and Shuryak are confident that even in more sophisticated models than used in ref.[@HS94], the total lifetime of a fireball should have a maximum near the indicated collision energy region. From our point of view, this conclusion is not evident because of the two crucial assumptions made. Firstly, the crossover behaviour of the deconfinement transition was simulated in ref.[@HS94] by an [arbitrary smoothing]{} of the results of a simple two-phase model exhibiting the first order phase transition. It is not clear how well this description approximates the lattice QCD results. Secondly, the EOS of the [baryon-free matter]{}, $n_B=0$, has been used [@HS94]. Due to a considerable stopping power, there is little hope to create a baryonless matter in nucleus-nucleus collisions up to $E_{lab} \simeq 200 \ GeV/A$ (especially for heavy systems). In this paper, we shall consider how the longest-lived fireball is sensitive to the details of the EOS and how it survives in baryon-rich matter inherent in systems formed at energies $E_{lab} \le 200 \ GeV/A$. Our analysis is based on the statistical model taking properly into account the mixed phase in which unbound quarks and gluons coexist with hadrons [@SYY93a; @SYY95]. The EOS is the key quantity and thermodynamic properties of excited matter near QCD phase transition should be calculated from the first principles in a non-perturbative manner. At present, such calculations are possible only in the lattice QCD. This approach shows that gluonic matter in pure gauge $SU(2)$ and $SU(3)$ theories exhibits a phase transition of the second and first order, respectively (for a recent review see ref.[@U95]). For the more realistic case of the $SU(3)$ theory with dynamical quarks, recent lattice studies [@U95] show that there is a smooth crossover rather than a distinct phase transition if the quark masses are close to the physical ones. For the most important case of the lattice QCD at finite baryon number density, nothing is really known and statistical models should be invoked to describe thermodynamic properties of the excited nuclear matter. There are many versions of the statistical model (see ref.[@CGS86] and references quoted therein) but all of them predict the deconfinement phase transition of the first order and, therefore, do not reproduce the lattice QCD results with dynamical quarks mentioned above. The only exception is the statistical mixed phase model of the deconfinement [@SYY93a; @SYY95], which we will use in our considerations. The application of the mixed phase models to the deconfinement transition is supported strongly by the discussion of deconfinement as a color screening effect and the observation that the hadron-like excitations survive above the critical temperature (see for example ref.[@SS95]). The specific feature of the approach developed by Shanenko, Yukalova and Yukalov (SYY) [@SYY93a; @SYY95] is to consider the coexistence of spatially non-separated the hadron phase and the quark-gluon plasma phase.The latter one consists of unbound ’generic’ particles (quarks and gluons in our case) while hadron phase is composed of quark-gluon clusters. In the mixed phase model [@SYY93a], one assumes beforehand the separation of cluster degrees of freedom, i.e. an exact Hamiltonian $H(\psi)$ is replaced by an effective cluster Hamiltonian $H_{c} (\psi_c )$. Here, $\psi$ denotes field operators of the generic particles and $ \psi_c \equiv \{ \psi_n;n=1,2,\ldots \} $ stands for quasiparticle operators ($n>1$ corresponds to clusters and $n=1$ to unbound generic particles). A large variety of quasiparticles leads to an enormous number of possible states. In the equilibrium, the physical state of a system corresponds to the extremum of the thermodynamic potential $F(H_c)$. The following two important points should be emphasized. Firstly, the effective Hamiltonian $H_c(\psi_c )$ may acquire extra dependence on thermodynamic parameters like temperature $T$ and cluster densities $\rho_c$: $ H_c \equiv H_c (\psi_c,T,\rho_c ) $ where $\rho_c \equiv \{\rho_n;n=1,2,\ldots \}$. The appearance of a density-dependent interaction is a distinctive feature of the SYY approach as compared to other statistical models [@CGS86] [^1]. Secondly, one must ensure that $H(\psi )$ and $H_c(\psi_c )$ are thermodynamically equivalent, i.e. their thermodynamic characteristics are the same in the thermodynamic limit of $V \to \infty $ keeping constant $\rho_c$ [@Y91]. These demands of the thermodynamic equivalence and the thermodynamic self-consistency impose additional conditions: $$\begin{aligned} \lim_{V \to \infty \atop \rho_c =const } \ \frac{1}{V} \left[ \ F(H) - F(H_c) \ \right] = 0 \ ,\end{aligned}$$ $$\begin{aligned} \lim_{V \to \infty \atop \rho_c =const } \ \frac{1}{V} \left[ \ dF(H) -dF(H_c) \ \right] = 0 \label{eq2a}\end{aligned}$$ which lead to $$\begin{aligned} \left< \frac{\partial H_c}{\partial T} \right> &=& 0 \ , \label{eq3a} \\ \nonumber \\ \left< \frac{\partial H_c}{\partial \rho_n} \right> &=& 0, \ \ \ \ \ \ n=1,2,\ldots \label{eq3}\end{aligned}$$ and essentially define the form of the cluster Hamiltonian. In the mean-field approximation, the cluster Hamiltonian becomes [@SYY93a]: $$\begin{aligned} H_c = \sum_n \sum_s \int d{\bf r} \ \psi^+_n({\bf r},s) \ \left( \ K_n+U_n(T,\rho_c) \ \right) \ \psi_n({\bf r},s) - C \cdot V \ , \label{eq4}\end{aligned}$$ where $n$ enumerates the clusters and $s$ stands for their internal degrees of freedom. $K_n=\sqrt{-\nabla^2 + M^2_n}$ is the kinetic energy and $U_n$ is a mean field acting on the $n$-particle cluster. In the same approximation, the condition of thermodynamic equivalence (\[eq3a\]) becomes: $$\begin{aligned} \frac{\partial U_n(T,\rho_c)}{\partial T} = 0 \ \ \ \ \ n=1,2,\ldots , \label{eq5}\end{aligned}$$ i.e. the mean fields $U_n=U_n(\rho_c )$ depend on the temperature only through the temperature dependence of the densities $\rho_n$. The $c$-number term $C\cdot V$ in eq.(\[eq4\]) is necessary to satisfy the conditions (\[eq3a\]) and (\[eq3\]) in which case $C$ is a function of $U_n$’s [@Y91]. Unbound particles are treated as trivial clusters with $n=1$. If masses $M_n$ of the isolated clusters and their quantum numbers are known either experimentally or from other calculations, then, to apply the mixed phase model, one has to define only $U_n$. The mean field $U_1$ acting on unbound gluons or quarks can be approximated as follows [@SYY93a; @SYY95]: $$\begin{aligned} U_1 = \frac{A}{\rho^{\gamma}} \ \ \ \ \ \ \ \ \ \ \ 0 < \gamma < 1 \label{eq6}\end{aligned}$$ where $\rho = \sum_n n \ \rho_n$ is the total quark-gluon density. The presence of $\rho$ in (\[eq6\]) corresponds to the inclusion of the interaction between all components of the mixed phase. Note that when there are no hadrons, i.e. $\rho_n=0$ for $n>1$, the expression (\[eq6\]) is just the same as used earlier to describe the thermodynamic properties of the quark-gluon plasma with a density-dependent quark mass [@B85]. For the case of $n>1$, alongside with the mean-field term (\[eq6\]) depending on the constants $A$ and $\gamma$, there is also a cluster-cluster interaction potential which is proportional to the hadron density and the function $\Phi_{nm}(r)$ characterizing the interaction strength between clusters of $n$ and $m$ generic particles [@SYY93a; @SYY95]. For long-ranged cluster interactions it is possible to get a recurrent relation [@Y91]: $$\begin{aligned} \Phi_{nm} (r) \sim nm \ \Phi_0(r) \ , \label{eq7}\end{aligned}$$ and to reduce all unknown interactions to the single interaction potential $\Phi_0(r)$ between simplest non-trivial clusters, e.g. two-gluon glueballs in the ground state, lightest mesons or baryons. In the Hartree approximation to $H_c$, one needs to know only one constant: $ {\bar{\Phi}}_0 \equiv \int d{\bf r} \ \Phi_0(r)$ to describe the cluster-cluster interactions. Thus, the Hamiltonian (\[eq4\]) is completely defined and thereby any thermodynamic characteristics of the mixed phase system can be found if three parameters $A$, $\gamma$ and ${\bar{\Phi}}_0$ have been fixed [@SYY93a; @SYY95]. This 3-parameter set was found [@SYY93a] by fitting a temperature dependence of the energy density $\varepsilon$ and the pressure $p$ calculated within the lattice QCD for the pure gluonic matter in the gauge $SU(2)$ [@EFR89] and $SU(3)$[@K89] theories. It is worth emphasizing that $A$ depends only on the colour group and $\gamma$ is constant for all gauge systems. So, fitting the $SU(3)$ pure gluonic QCD lattice data allows to fix $A$ and $\gamma$ parameters which can then be used for the $SU(3)$ system with quarks. As to ${\bar{\Phi}}_0$ in this case, it can be found from a nucleon-nucleon potential by referring to the relation (\[eq7\]). In the following, we use $\gamma =0.62$, $A^{1/(3\gamma +1)} = 225 \ MeV $ and ${\bar{\Phi}}_0 = 4.1\cdot 10^{-5} \ MeV^{-2}$. The mixed phase model predictions for the $SU(3)$ system with two light flavours and $n_B=0$ are shown in Fig. 1. When confronted with the lattice QCD data, the mixed phase model [@SYY93a] gives a very similar temperature dependence of the energy density $\varepsilon$ and pressure $p$. As follows from these results, the mixed phase model estimates the deconfinement temperature to be $T_{dec}=150 \ MeV$ and predicts the crossover-type phase transition [@SYY93a; @SYY95] in full agreement with the QCD lattice data. Some overshooting of the lattice data at $T>1.5 \ T_{dec}$ is related to neglecting the negative Coulomb-like term of the quark-gluon interactions. Two different sets of the best available lattice QCD results plotted in Fig. 1, correspond to two different scheme of including quarks. A peak-like structure of $\varepsilon$ near the deconfinement temperature $T_{dec}$ for the Kogut-Susskind scheme [@B94] seems to be unrealistic, vanishing for temperatures below $\sim 0.9 \ T_{dec}$. The quark mass in the Kogut-Susskind calculations amounts to $m_q \approx 0.1 \ T_{dec}$ which is close to the physical mass used in the mixed phase model. In the Wilson scheme we do not really know the quark mass used. In this case, the value $m_q \sim T_{dec}$ given in [@CES86] seems to be enormously large because for $T<T_{dec}$ there is a good agreement between the Wilson scheme results and the ideal meson gas calculations [@CES86]. The EOS in the form advocated by Hung and Shuryak [@HS94] is represented in Fig.2. Here, all data of Fig.1 are replotted alongside with the curve used in [@HS94] to simulate the crossover transition. The $p/\varepsilon$-functions for [lattice data themselves]{} show [a minimum]{} which is just associated with the softest point of the EOS where a fireball of the excited nuclear matter lives longest. In the $p/\varepsilon$-representation, the lattice QCD data for two schemes of accounting for dynamical quarks are strikingly different but, nevertheless, they yield the same position of the minimum at $\varepsilon_{sp} \approx 0.5-0.6 \ GeV/fm^3$ which can be reached with the $Au$ beam energies of about $3-5 \ GeV/A$. It is noteworthy that this value od $\varepsilon_{sp}$ corresponds roughly to the value of the bag constant, so the fireball near the softest point may be considered as a ’big hadron’. The mixed phase model predicts the position of a minimum near $\varepsilon_{sp} \approx 0.3 \ GeV/fm^3$ which is rather close to the lattice results. On the contrary, Hung and Shuryak obtain much higher value, $\varepsilon_{sp} \approx 1.5 \ GeV/fm^3$ [@HS94] and the minimum is seen more distinctly than in the lattice data. Since $\varepsilon \sim T^4$, the difference in deconfinement temperatures is small. The Kogut-Susskind and Wilson schemes predict the deconfinement temperatures $T_{dec}$ as high as $157 \ MeV$ [@B94] and $150 \ MeV$ [@U95] respectively, and the mixed phase model gives $T_{dec}=150 \ MeV$. The Hung-Shuryak approximation does not come from a fit to the lattice QCD data but is an arbitrary smoothing of the two-phase model results where $T_{dec}=160 \ MeV$ is defined by the choosen bag constant [@HS94]. The mixed phase model can be naturally generalized to the case of the non-zero baryon density $n_B$ [@SYY95]. The $n_B$-dependent pressure in the SYY approach coincides with that of the Walecka-like model up to $n_B \approx 3 \ n_0$ [@HBB95]. At higher baryon densities, the mixed phase model gives lower values of pressure due to the quark admixture. As to the predicted energy density, the difference between these two models is even smaller. As is seen in Fig.3, the $p/\varepsilon $-function changes drastically with increasing the baryon density of a system: the position of $\varepsilon_{sp}$ remains unchanged at $n_B < n_0$ though the minimum is gradually vanishing with increasing $n_B$ and disappears at $n_B \approx n_0$ . So, we arrive at somewhat controversial demands: to reach the condition of the longest-lived fireball in heavy ion collisions one should go [downward]{} in bombarding energies as far as $E_{lab} \approx 3-5 \ GeV/A$ but, on the other hand, high baryon density of a fireball formed at these energies will suppress much the effect of the softest point. Due to smallness of $\varepsilon_{sp}$ it may be more favourable to look for the softest point effect in non-central high-energy collisions in the target fragmentation region. To see how the lifetime of a fireball will change with the beam energy, one needs detailed dynamical calculations with EOS of the mixed phase model [^2]. It is of interest to note that this EOS is quite different from the EOS for the pure hadronic phase as illustrated in Fig.3 in the case of an ideal pion gas. Thus, the lattice QCD data do really predict a minimum in the $p/\varepsilon$-representation of the EOS whose position, according to Hung and Shuryak [@HS94], defines the beam energy at which the fireball formed has the longest lifetime. This representation is significantly more sensitive, both to details of the lattice calculations and approximations involved therein than the conventional thermodynamic quantities $\varepsilon (T)$ and $p(T)$. The simulation of the EOS with a crossover [@HS94] results in the softest point at $\varepsilon_{sp} \approx 1.5 \ GeV/fm^3$, which is noticeably higher than the lattice value $\varepsilon_{sp} \approx 0.5 \ GeV/fm^3$. Describing correctly the order and temperature of the deconfinement QCD transition for $n_B=0$, the mixed phase model [@SYY93a; @SYY95] predicts $\varepsilon_{sp} \approx 0.3 \ GeV/fm^3$ and the agreement with the lattice data may be further improved by a more accurate treatment of cluster-cluster interactions. All this implies that the proposed beam-energy tuning for identification of the deconfinement transition should be done at bombarding energies of $E_{lab} \approx 3-5 \ GeV/A$, much below the value $E_{lab} \simeq 30 \ GeV/A$ advocated by Hung and Shuryak [@HS94]. In the recent paper [@MO95], heavy ion collisions were considered within relativistic dissipative hydrodynamics with EOS taken from the lattice QCD calculations for baryonless matter. A strong enhancement of the width of the shock front has been found at $E_{lab} \simeq 6 \ GeV/A$ as a manifestation of the softest point. This value of the beam energy is in a fine agreement with our estimate, if one remembers that the old QCD data used in ref.[@MO95] correspond to higher deconfinement temperature $T_{dec} = 200 \ MeV$. Note that these hydrodynamic calculations are complementary to the Hung and Shuryak estimate and show that the softest point effect influences not only the disassembly stage but also the compression stage. The mixed phase model predicts also a strong dependence of the EOS on the baryon density of the system: a minimum of the $p/\varepsilon $ function is washed out for $n_B \approx n_0$. Since the state with $\varepsilon = \varepsilon_{sp}$ is a transitional one, we expect that the change of the fireball lifetime with $E_{lab}$ will not be as large as predicted in [@HS94]. Nevertheless, we would like to draw attention to heavy ion collisions at moderate ($E_{lab}=2-10 \ GeV/A$) energies for studying the mixed phase of quarks and hadrons. As has been shown above, a pure hadronic EOS is quite different from the EOS predicted by the mixed phase model near the softest point. A possibility of forming the mixed quark-hadron state at energies $E_{lab}\approx 2-10 \ GeV/A$ has been noted previously [@G90]. However, being based on the two-phase models, this consideration predicts a sharp decrease of temperature just above the deconfinement threshold resulting in the formation of a ’cold’ plasma. Generally speaking, the known signals of quark-gluon plasma are not applicable to the cold plasma case. In contrast, the mixed phase model has no threshold and no temperature fall-off, thus such signals should persist but their strength will be proportional to an unbound quark abundance. It is worth mentioning that some enhancement of the $\Lambda$-hyperon production as a specific plasma formation signature has been observed at the energy as low as $3.5 \ GeV/A$ [@O95]. Interferometry measurements deserve a special attention since they are sensitive to ’granularity’ of an emitting source [@P94]. In this respect, the correlation length or screening length of unbound quarks provides a new length scale additional to the source radius. We are indebted to W. Bauer, J. Cleymans, M. Gorenstein, E. Okonov, S. Pratt, F. Webber and G. Zinoviev for stimulating discussions and comments. V.D.T. acknowledges the warm hospitality of the theory group of GANIL, Caen where a part of this work has been done. The work of V.D.T. was supported by Grant $N^0_-$ 3405 from INTAS (International Association for promotion of cooperation with scientists from the independent states of the former Soviet Union). [99]{} C.M. Hung and E.V. Shuryak, Preprint SUNY-NTG-94-59 (1994), to be published in Phys. Rev. Lett. E. Shuryak and O.V. Zhirov, Phys. Lett. B 89 (1979) 253;\ L. Van Hove, Z. Phys. C 21 (1983) 93;\ K. Kajantie, L. Mc Lerran and P.V. Ruuskanen, Phys. Rev. D 34 (1986) 2746;\ S. Chakrabarty at al., Phys. Rev. D 46 (1992) 3802. A.A. Shanenko, E.P. Yukalova and V.I. Yukalov, Nuovo Cim. A 106 (1993) 1269; Physics of Atomic Nuclei 56 (1993) 372. A.A. Shanenko, E.P. Yukalova and V.I. Yukalov, JINR Rapid Communications 1 \[69\] (1995) 19; Physics of Atomic Nuclei 58 (1995) 335. A. Ukawa, Report UTHEP-302 (1995). J. Cleymans, R.V. Gavai and E. Suhonen, Phys. Rep. 130 (1986) 217. T. Schäfer and E.V. Shuryak, Phys. Lett. B 356 (1995) 147. A.A. Shanenko, E.P. Yukalova and V.I. Yukalov, Physica A 197 (1993) 629. J. Engels et al., Z. Phys. C 42 (1989) 341. F. Karsch, Preprint CERN-TH-5498/89 (1989). K.A. Olive, Nucl. Phys. B 190 (1981) 483; [ibid]{} B 198 (1982) 461;\ D.H. Boal, J. Schachter and R.M. Woloshin, Phys. Rev. D 26 (1982) 3245;\ M. Plumer, S. Raha and R.M. Weiner, Nucl. Phys. A 418 (1984) 549c;\ D. Blaschke et al., Phys. Lett. B 151 (1985) 439;\ I.V. Moskalenko and D.E. Kharzeev, Soviet Journal of Atomic Physics 48 (1988) 713. T. Celik, J. Engels and H. Satz, Nucl. Phys. B 256 (1986) 670. T. Blum et al., Preprint AZPH-TH/94-22 (1994). G. Hejc, W. Bentz and H. Beier, Nucl. Phys. B 582 (1995) 401. T. Biro and J. Zymanyi, Nucl. Phys. A 395 (1983) 25;\ H. Stöker, Nucl. Phys. A 417 (1984) 587;\ N. Glendenning, Nucl. Phys. A 512 (1990) 737. L. Mornas and U. Ornik, Nucl. Phys. B 587 (1995) 828. E.O. Okonov, Nucl. Phys. B 583 (1995) 711;\ M. Gaździcki et al., Z. Physik C 33 (1986) 895. S. Pratt, Phys. Rev. C 49 (1994) 2722. \ Fig.1. Temperature dependence of energy density $\varepsilon$ and pressure $p$ (relative to corresponding values in the Stefan-Boltzmann limit) for $SU(3)$ gauge theory of baryon-free matter with massive dynamical quarks. Curves are calculated within the mixed phase model [@SYY93a]. Triangles and squares are lattice data for Wilson [@CES86] and Kogut-Susskind [@B94] schemes of accounting for dynamical quarks, respectively. Fig.2. The ratio of pressure and energy density $p/\varepsilon$ versus $\varepsilon$. Notation is the same as in Fig.1. The dashed curve corresponds to the approximation used in ref.[@HS94]. Data are given for $\varepsilon > 0.005 \ GeV fm^{-3}$. Fig.3. $p/\varepsilon$-representation of the EOS for the baryonic matter predicted by the mixed phase model [@SYY95]. Numbers near the curves show the baryon density in units of the normal nuclear density. The case of ideal pion gas is given by the dashed line. [^1]: Similar situation is met in the case when interaction is taken into account by the excluded-volume method. [^2]: These calculations within a hydrodynamic model are now in progress.	{ "pile_set_name": "ArXiv" }
--- abstract: 'The aim of this work is to rigorously formulate the non-commutative calculus within the framework of quantum field theory. In so doing, we will consider the application of integrable and differential structures to local algebras. In the application of integrable structures to local algebras, we make use of a new approach based on quantum Orlicz spaces. We specifically propose regularity conditions which ensure good behaviour of field operators as observables in the context of local algebras. This complements earlier work by Buchholz, Driessler, Summers & Wichman, etc, on generalized $H$-bounds. The pair of Orlicz spaces we explicitly use are respectively built on the exponential function (for the description of regular field operators) and on an entropic type function (for the description of the corresponding states). These spaces form a dual pair. This formalism has been shown to be well suited to a description of quantum statistical mechanics, and in the present work we show that it is also a very useful and elegant tool for Quantum Field Theory. Secondly as far as differential structures are concerned, we argue that the formalism based on local algebras is also well suited to the development of a non-commutative differential geometric structure along the lines of the du Bois-Violette approach to such theory. In that way we obtain a complete depiction: integrable structures based on local algebras provide a static setting for an analysis of Quantum Field Theory and an effective tool for describing regular behaviour of field operators, whereas differentiable structures posit indispensable tools for a description of equations of motion. Finally we indicate that the formalism presented here is also relevant for local algebras based on Lorentzian manifolds.' address: - \| DST-NRF CoE in Math. and Stat. Sci,\ Unit for BMI,\ Internal Box 209, School of Comp., Stat., $\&$ Math. Sci.\ NWU, PVT. BAG X6001, 2520 Potchefstroom\ South Africa - \| Institute of Theoretical Physics and Astrophysics, Gda[ń]{}sk University, Wita Stwosza 57, 80-952 Gda[ń]{}sk, Polandand DST-NRF CoE in Math. and Stat. Sci,\ Unit for BMI,\ Internal Box 209, School of Comp., Stat., $\&$ Math. Sci.\ NWU, PVT. BAG X6001, 2520 Potchefstroom\ South Africa author: - 'L. E. Labuschagne' - 'W. A. Majewski' title: Integral and differential structures for quantum field theory --- [^1] Preliminaries; some basic ideas derived from QFT. ================================================= Generally, it would seem that in Quantum Mechanics there are two schemes for a description of physical systems, cf. [@Bor1]. The first method uses bounded operators. The idea of introducing the norm topology on the set of observables was strongly advocated by I. Segal [@Segal]. To argue in favor of this idea one can say that in a laboratory a physicist deals with bounded functions of observables only! However, as it was already remarked by Borchers [@Bor1], in this method “some detailed information about a physical system is usually lost”. Furthermore, this scheme admits “non-physical states” having badly defined entropy, see [@Maj1] and the references given there. The second method uses unbounded operators. The motivation for this method can be taken from representations of canonical commutation relations, Wightman’s formulation of quantum field theory and the theory of Lie algebras. Although mathematical aspects of algebras of unbounded operators have been analyzed in much detail, see [@AIT], [@Schmud], [@Bag], it is well known that formal calculations can be misleading; see Section VIII.5 in [@RSI]. Here, we will argue that non-commutative integration theory offers a third alternative lying between the above discussed approaches, see also [@studyguide]. Besides other technical conditions it relies of selecting “more” regular unbounded operators, where “more” regular means $\tau$-measurability (see the following pages for definitions and details). In this method one not only has the advantage of having access to a very well-behaved \-algebra of unbounded operators (the algebra of $\tau$-measurable operators), but this unbounded framework also appears as a natural limiting case of the bounded framework. (Any semifinite von Neumann algebra is dense in the associated algebra of $\tau$-measurable operators.) Our objective in this paper, is therefore not to analyse some specific model, but rather to on the one hand introduce this framework, and on the other to show how various aspects of Quantum Field Theory may be harmonised within this framework. The algebraic approach to relativistic quantum field theory was formulated in the sixties by R. Haag, D. Kastler, H. Araki, H. J. Borchers and others, see [@haag], [@araki], [@BH]. The basic object of this approach is a net of von Neumann algebras, ${{\mathcal O}}\mapsto {{\mathfrak M}}({{\mathcal O}})$, on a Hilbert space ${{\mathcal H}}$, labeled by subsets ${{\mathcal O}}$ of (Minkowski) space-time ${{\rm I\!R}}^4$. It satisfies, see [@araki]: 1. Isotony: ${{\mathcal O}}_1 \subset {{\mathcal O}}_2$ implies ${{\mathfrak M}}({{\mathcal O}}_1) \subset {{\mathfrak M}}({{\mathcal O}}_2).$ 2. Covariance: for $g= (a,\Lambda) \in P^{\uparrow}_+$, there is a representation $\alpha_g$ in $Aut({{\mathfrak M}})$ such that $$\alpha_g({{\mathfrak M}}({{\mathcal O}})) = {{\mathfrak M}}(g{{\mathcal O}}), \quad g{{\mathcal O}}= \{ \Lambda x +a; x \in {{\mathcal O}}\},$$ $P^{\uparrow}_+$ stands for the Poincaré group, where the Lorentz group is restricted, homogeneous, see Section 3.3 in [@araki]. This action is realised by a strong operator continuous unitary group $U(a, \Lambda)$ acting on the underlying Hilbert space ${{\mathcal H}}$. 3. Locality: if ${{\mathcal O}}_1$ and ${{\mathcal O}}_2$ are spacelike separated then ${{\mathfrak M}}({{\mathcal O}}_1)$ and ${{\mathfrak M}}({{\mathcal O}}_2)$ commute. 4. Weak Additivity: $${{\mathfrak M}}= \left( \bigcup_{x \in {{\rm I\!R}}^4} {{\mathfrak M}}({{\mathcal O}}+ x) \right)^{''}$$ for all open ${{\mathcal O}}$. 5. Vacuum vector: there is a normalized vector $\Omega \in {{\mathcal H}}$, unique up to a phase factor that satisfies $$U(a, \Lambda) \Omega = \Omega,$$ for the Poincaré transformations $(a, \Lambda)$. 6. Positivity: The generator of translation has spectrum lying in the forward light cone. In an analysis of Poincaré-Lorentz transformations it is convenient to distinguish those related with one frame, $F$, and those which involve another frame, $F^{\prime}$, which moves with velocity $v$ relative to $F$. In other words there are two basic types of Poincaré-Lorentz transformations: 1. transformations defined in terms of inertial frames with no relative motion, i.e. the frames are simply tilted. In particular, there are rotations (but without continuous rotation) and translations. 2. transformations describing relative motion with constant (uniform) velocity and without rotations of the space coordinates. Such transformations are called boosts.* It is worth pointing out that that the spectral conditions, mentioned in the point (6) above, are relevant for the first type of Poincaré-Lorentz transformations. In particular, the spectral conditions are not applicable to generators of boosts. The vital consequences of the observation just presented, will be described in the discussion on regularity conditions of fields operators in the next section. On the other hand, in the fifties, Wightman and Gärding, see [@GW1], [@GW2], [@W], formulated postulates for general quantum field theory in terms of (unbounded) operators on a Hilbert space. Depending on the context, there are various subtle variations of these postulates. However we are not interested in a detailed application of the resultant theory to a specific context, but rather in the overarching mathematical framework and how this framework may be harmonised. The basic mathematical ingredients of these postulates that are relevant to our study, may be expressed as below (see [@araki]). The reader interested in finer detail, may refer to [@GW1], [@GW2] and [@W]. 1. \[F1\] Quantum fields: The operators $\phi_1(f), ...,\phi_n(f)$ are given for each $C^{\infty}$-function with compact support in the Minkowski space ${{\rm I\!R}}^4$. Each $\phi_i(f)$ and its hermitian conjugate operator $\phi_j^(f)$ are defined on at least a common dense linear subset ${{\mathcal D}}$ of the Hilbert space ${{\mathcal H}}$ and ${{\mathcal D}}$ satisfies $$\phi_j(f){{\mathcal D}}\subset {{\mathcal D}}, \quad \phi_j^(f) {{\mathcal D}}\subset {{\mathcal D}},$$ for any $f$, $j=1,...,n$. For any $v,w \in {{\mathcal D}}$ $$f \mapsto (v, \phi_j(f)w)$$ is a complex valued distribution. 2. Relativistic symmetry: There is a well defined strongly continuous unitary representation $U(a, A)$ of $P^{\uparrow}_+$ ($a \in {{\rm I\!R}}^4, \ A \in SL(2, {{\setbox0=\hbox{ $\displaystyle\rm C$}\hbox{\hbox to0pt{\kern0.6\wd0\vrule height0.9\ht0\hss}\box0}}})$) such that $$U(a, \Lambda) {{\mathcal D}}= {{\mathcal D}}$$ and $$U(a, A)\phi_j(f) U(a, A)^* = \sum S(A^{-1})_{jk} \phi_k(f_{(a,A)}),$$ where the matrix $(S(A)_{j,k})$ is $n$-dimensional representation of $A \in SL(2,{{\setbox0=\hbox{ $\displaystyle\rm C$}\hbox{\hbox to0pt{\kern0.6\wd0\vrule height0.9\ht0\hss}\box0}}})$, and $f_{(a,A)}(z) = f(\Lambda(A)^{-1}(z - a)).$ 3. Local commutativity: if the supports of $f$ and $g$ are space-like separated, then for any vector $v \in {{\mathcal D}}$ $$[\phi_j(f)^{\diamond}, \phi_k(g)^{\diamond}]_{\mp} (v) = 0,$$ where $\diamond$ denotes the following possibilities: no $$, one $$, and both operators $\phi$ have a $$. 4. Vacuum; there exists a well defined vacuum state, i.e. a vector $\Omega \in {{\mathcal H}}$, invariant with respect to the Poincaré group such that the following spectrum condition is satisfied : the spectrum of the translation group $U(a, {{\mathbb{I}}})$ on $\Omega^{\bot}$ is contained in $V_m = \{ p ; (p,p) \geq m^2, p^0 > 0 \}$, $m>0$. Properties of field operators versus noncommutative integration. {#locint} ================================================================ As a first step we note the following: For any field operator $\phi(f)$ satisfying Wightman’s postulates, one has $$\label{fmoment} (v,\phi^n(f)w) \in {{\setbox0=\hbox{ $\displaystyle\rm C$}\hbox{\hbox to0pt{\kern0.6\wd0\vrule height0.9\ht0\hss}\box0}}},$$ for any $v,w \in {{\mathcal D}}$, and any $n \in {{\rm I\!N}}$. In other words, the number $(v,\phi^n(f)w)$ is finite for any $n \in {{\rm I\!N}}$. For states $\omega_x(\cdot) \equiv (x, \cdot x)$ with $x \in {{\mathcal D}}$, field operators in QFT enjoy the property of having all moments finite. We remind that this feature is a starting point for an analysis of applications of the quantum Orlicz space formalism to Statistical Physics. However, to argue that the scheme based on noncommutative Orlicz spaces is applicable to QFT, we must additionally show that there is good reason to postulate that field operators may in a natural way be embedded into the noncommutative Orlicz space consisting of the regular observables of the von Neumann algebra appearing in the algebraic approach to QFT. The reasonableness of this postulate, is demonstrated in a step-wise process. We specifically show that: - field operators are affiliated to the von Neumann algebras appearing in the algebraic approach to QFT, - there is good reason to assign criteria to these operators which ensure that they embed into a larger algebra of unbounded operators (the so-called algebra of $\tau$-measurable operators (definition loc. cit.), - these criteria ensure that the field operators may be represented as elements of the noncommutative Orlicz space generated by the regular observables of the system, - and that these criteria lead to physically reasonable consequences. The affiliation of field operators to local algebras ---------------------------------------------------- To indicate that field operators satisfy the first requirement, we follow Araki [@araki] by beginning with a brief description of arguments showing how a field operator can be associated to a net of von Neumann algebras, see [@DSW], [@buch1], [@araki]. Let ${{\mathcal P}}$ be a family of operators with a common dense domain of definition ${{\mathcal D}}$ in a Hilbert space ${{\mathcal H}}$ (cf. Wightman’s rule presented above) such that if $\phi \in {{\mathcal P}}$ then also $\phi^\|_{{{\mathcal D}}} \equiv \phi^{\dagger} \in {{\mathcal P}}$. The weak commutant ${{\mathcal P}}^w$, of ${{\mathcal P}}$ is defined as the set of all bounded operators $C$ on ${{\mathcal H}}$ such that $(v,C\phi w) = (\phi^{\dagger}v, C w)$, for all $v,w \in {{\mathcal D}}$. For simplicity of our arguments we will restrict ourselves to one type of real scalar field $\phi$; i.e. $\phi(f)^$ coincides with $\phi(\overline{f})$ on ${{\mathcal D}}$. Furthermore, apart from the Wightman postulates we assume: 1. \[I\] ${{\mathcal P}}({{\mathcal O}}^p_q)^w$ is an algebra for any double cone ${{\mathcal O}}^p_q \equiv \{x; p-x \in V_+, x - q \in V_+ \}$, where $V_+ = \{ \rm{ positive \ timelike \ vectors} \ \}$. 2. \[II\] The vacuum vector $\Omega$ is cyclic for the union of ${{\mathcal P}}(D^{\prime})^w$ over all double cones $D$, where $D^{\prime}$ is the causal complement of $D$. The following theorem is taken from [@araki] , but stems from results given in [@DSW], [@buch1]. \[2.3\] Assume that both conditions (A1) and (A2) hold. For each double cone $D$, define $${{\mathfrak M}}(D) = \left( {{\mathcal P}}(D)^w \right)^{\prime}.$$ Then ${{\mathfrak M}}(D)$ is a von Neumann algebra and the net $D \mapsto {{\mathfrak M}}(D)$ satisfies conditions (L1)-(L3) for local algebras (cf the first section). On defining ${{\mathfrak M}}$ to be the von Neumann algebra generated by $\cup_D {{\mathfrak M}}(D)$, the state on ${{\mathfrak M}}$ determined by $\Omega$ is then a pure vacuum state for which - $\Omega$ is cyclic for each ${{\mathfrak M}}(D)$, - and each operator $\phi \in {{\mathcal P}}(D)$ has a closed extension $\phi_e \subset \phi^{\dagger, }$ which is affiliated with ${{\mathfrak M}}(D)$. (Here, $\phi_e \subset A$ means that the domain of $\phi_e$ is contained in the domain of $A$ and that $\phi_e = A$ on the domain of $\phi_e$.) Theorem \[2.3\] yields Field operators lead to operators affiliated to the von Neumann algebra ${{\mathfrak M}}(D)$. We remind that this property is the starting point for the definition of measurable operators. Moreover, one has There are sufficient conditions, motivated by physical requirements, for conditions (A1) and (A2) (given prior to Theorem \[2.3\]) to hold, see [@DSW], [@buch1], [@BY]. As an example of such condition we quote from [@BY] \[regular\] Let $\phi$ be a Wightman field and let $H$ denote its Hamiltonian. The field satisfies a generalized H-bound if there exists a nonnegative number $\alpha <1$, such that $\phi(f)^{*} e^{- H^{\alpha}}$ is a bounded operator for all $f$. It worth reiterating the fact pointed out in [@DSW], that the physical significance of such conditions, is that they select models with slightly more regular high energy behaviour. Local algebras and the crossed product construction --------------------------------------------------- To proceed with an analysis of the $\tau$-measurability criteria of $\phi_e$ operators, a containing von Neumann algebra with a faithful normal semifinite trace is necessary. However (see the comprehensive review [@Y]) the local algebras ${{\mathfrak M}}(D)$ are, under physically plausible assumptions, the same for all relativistic quantum field theories, namely they are isomorphic to the unique hyperfinite type $III_1$ factor. But type $III$ factors are known not to have any nontrivial traces (see vol I and II of Takesaki [@Tak], [@Tak]). Therefore, we should employ the superalgebra, the crossed product, $\mathfrak{M} \rtimes_{\sigma} {{\rm I\!R}}$, which will be defined below. We can do this, as due to the Reeh-Schlieder property of vacuum states, see [@ReSch] (cf also Theorem 4.14 in [@araki]), we can consider such vacuum states as those whose GNS-vector $\Omega$ is cyclic and separating for the von Neumann algebra ${{\mathfrak M}}(D)$, where $D$ is a bounded region in the Minkowski space. Consequently, the Tomita-Takesaki theory can be employed. In particular, the modular action for the triple $({{\mathfrak M}}(D), \Omega, {{\mathcal H}})$ exists and it will be denoted by $\sigma_t, \ t \in {{\rm I\!R}}.$ One can generalize the line of reasoning given below by replacing the vacuum state $\omega(\cdot) = (\Omega, \cdot \Omega)$ by a faithful normal semifinite weight. We remind that such a weight always exists for a von Neumann algebra. Here the vacuum state provides the connection between the approach suggested above, and the rules of Quantum Field Theory. The great significance of the above mentioned enlarged algebra, follows from the fact that the crossed product ${{\mathcal M}}\equiv \mathfrak{M} \rtimes_{\sigma} {{\rm I\!R}}$ with respect to the modular automorphism group of the canonical weight $\omega$ (see overleaf for the construction), is a semifinite von Neumann algebra when ${{\mathfrak M}}$ is type III, cf [@Tak]. In particular, $\mathfrak{M} \rtimes_{\sigma} {{\rm I\!R}}$ can be equipped with a faithful normal semifinite trace $\tau$. This object, $\tau$, is a necessary tool for a definition of $\tau$-measurable operator, see [@Tak], [@terp], or [@nelson]. \[measure\] Let ${{\mathcal M}}$ be a semifinite von Neumann algebra acting on a Hilbert space ${{\mathcal H}}$, and equipped with a faithful normal semifinite trace $\tau$. A closed and densely defined linear operator $a$ on ${{\mathcal H}}$ is called $\tau -measurable$ if $a$ is affiliated with ${{\mathcal M}}$ (that is, $au=ua$ for all unitary operators in the commutant ${{\mathcal M}}^{\prime}$ of ${{\mathcal M}}$) and there exists a $0< \lambda_0 \in {{\rm I\!R}}$ such that $\tau(E^{\|a\|}(\lambda_0, \infty)) < \infty$ (where $E^{\|a\|}$ denotes the spectral measure of the selfadjoint operator $\|a\|$). The space of all $\tau$-measurable operators is denoted by $\widetilde{{{\mathcal M}}}$ and can be equipped with a topology of convergence in measure, with respect to which it proves to be a complete metrisable \-algebra. Details may be found in [@Tak §IX.2]. This \-algebra is the noncommutative analogue of the completion of $L^\infty(X,\Sigma,\nu)$ with respect to the topology of convergence in measure. To proceed with an analysis of the $\tau$-measurability criteria of field operators, we have to describe ${{\mathcal M}}\equiv \mathfrak{M} \rtimes_{\sigma} {{\rm I\!R}}$ in some detail. To this end we will follow Haagerup’s modification of Takesaki’s construction as presented in Lemma 5.2 of [@uffe3]. This was developed in series of three papers [@uffe1; @uffe2; @uffe3], culminating in the result just mentioned. The bulk of the discussion below is taken from these papers, with occasional references from other sources where appropriate. For the sake of clarity we will extract only the very basic points of the exposition, without going to the point of sacrificing its essential content. Some modifications, which we made, are necessary in order to be able to follow the scheme of Quantum Field Theory as closely as possible. Recall that the basic ingredient of the operator algebraic approach to Quantum Field Theory is a net of local algebras having the properties described in the first section. A large part of our task is then to indicate how some of the subtleties of such algebras fit into the crossed product construction, rather than just presenting an abstract mathematical formalism. Let ${{\rm I\!R}}^4 \ni {{\mathcal O}}\mapsto {{\mathfrak M}}({{\mathcal O}})$ be a net of local observables. In general, ${{\mathfrak M}}({{\mathcal O}})$ is type III. We fix a region ${{\mathcal O}}_0$, so we can restrict ourselves to one von Neumann algebra ${{\mathfrak M}}({{\mathcal O}}_0) \equiv {{\mathfrak M}}$. Further there is well defined vacuum state $\omega$ on ${{\mathfrak M}}$, having the properties described in the first section and at the beginning of the second section. Moreover on passing to GNS representation if necessary, there is no loss of generality in assuming that ${{\mathfrak M}}$ is a von Neumann algebra acting on a Hilbert space ${{\mathcal H}}$ with cyclic and separating vector $\Omega$. As $({{\mathfrak M}}, \Omega, {{\mathcal H}})$ is then in the so-called standard form, there is a modular operator $\Delta$ and a modular automorphism $\sigma_t(\cdot) = \Delta^{it} \cdot \Delta^{-it}$, $t \in {{\rm I\!R}}$, of ${{\mathfrak M}}$. Denote the Hilbert space of all square integrable ${{\mathcal H}}$-valued functions on ${{\rm I\!R}}$ by $L^2({{\rm I\!R}}, {{\mathcal H}})$. Define representations $\pi_{\sigma}$ of ${{\mathfrak M}}$, and $\lambda$ of ${{\rm I\!R}}$, as follows: $$\label{cr1} (\pi_{\sigma}(a)\xi)(t) = \sigma_{-t}(a)\xi(t), \quad a \in {{\mathfrak M}}, \ t \in {{\rm I\!R}}, \xi \in L^2({{\rm I\!R}}, {{\mathcal H}})$$ and $$\label{cr2} (\lambda(t)\xi)(s) = \xi(s - t), \quad s,t \in {{\rm I\!R}}, \ \xi \in L^2({{\rm I\!R}}, {{\mathcal H}}).$$ The von Neumann algebra generated by $\pi_{\sigma}({{\mathfrak M}})$ and $\lambda({{\rm I\!R}})$ on $L^2({{\rm I\!R}}, {{\mathcal H}})$, is called the crossed product of ${{\mathfrak M}}$ by $\sigma$, and is denoted by $\mathfrak{M} \rtimes_{\sigma} {{\rm I\!R}}$. In the sequel we will write ${{\mathcal M}}\equiv \mathfrak{M} \rtimes_{\sigma} {{\rm I\!R}}$ and also identify ${{\mathfrak M}}$ with $\pi_\sigma({{\mathfrak M}})$ to simplify the notation. We know that any given field operator $\phi_e$ is affiliated with ${{\mathfrak M}}$ (that is taking the polar decomposition of $\phi_e$, $\phi_e = v \|\phi_e\|$, one has that $v$ and the spectral projections of $\|\phi_e\|$ are in ${{\mathfrak M}}$). But then $\phi_e$ will trivially also be a densely defined closed operator on $L^2({{\rm I\!R}}, {{\mathcal H}})$ for which $v$ and the spectral projections of $\|\phi_e\|$ are in ${{\mathcal M}}$. Hence it must be affiliated with ${{\mathcal M}}$. Now, our task is to find a trace $\tau$ on ${{\mathcal M}}$ which can be useful for an analysis of measurability. This is done in several steps. As a first step, one defines a dual action of $\mathbb{R}$ on ${{\mathcal M}}$ in the form of a one-parameter group of automorphisms $(\theta_s)$ by means of the prescription $$\label{dual} \theta_s(a)=a, \quad \theta_s(\lambda(t))=e^{-ist}\lambda(t) \mbox{ for all } a\in {{\mathfrak M}}\mbox{ and } s,t\in\mathbb{R}.$$ A fact not immediately clear from the definition of this action is that ${{\mathfrak M}}$ is not just contained in the fixed points of this dual action, but in fact ${{\mathfrak M}}=\{a\in {{\mathcal M}}:\theta_s(a)=a\mbox{ for all }s\in \mathbb{R}\}$. The next step is to define a so called faithful normal semifinite operator valued weight from the extended positive part of ${{\mathcal M}}$ to the extended positive part of ${{\mathfrak M}}$. We pause to clarify these concepts. Observe that for each $a\in {{\mathfrak M}}^+$ and each $\rho\in({{\mathfrak M}}_)^+$, the prescription $\rho\to\rho(a)$ yields an affine lower-semicontinuous map from $({{\mathfrak M}}_)^+$ onto $[0,\infty)$. With this as a starting point, the extended positive part $\hat{{{\mathfrak M}}}^+$ of ${{\mathfrak M}}$, is then defined to be the collection of maps $m: ({{\mathfrak M}}_)^+\to [0,\infty]$ for which - $m(\lambda\rho)=\lambda m(\rho)$ for all $\lambda\geq 0$, $\rho \in ({{\mathfrak M}}_)^+$ (where by convention $0.\infty=0$); - $m(\rho_1+\rho_2)=m(\rho_1)+m(\rho_2)$ for all $\rho_1, \rho_2 \in ({{\mathfrak M}}_)^+$; - $m$ is lower-semicontinuous. Haagerup goes on to develop a theory of these objects (see [@uffe2]), showing amongst other facts that any normal semifinite weight on ${{\mathfrak M}}$, extends to a “weight” on $\hat{{{\mathfrak M}}}^+$ [@uffe2 Proposition 1.10]. In addition any element $m$ of $\hat{{{\mathfrak M}}}^+$ is shown to admit of a spectral resolution of the form $$m(\rho)=\int_0^\infty\lambda \,\mathrm{d}\rho(E^\lambda)+\infty.p,$$where $\{E^\lambda: \lambda\in [0,\infty)\}$ is an increasing family of projections in ${{\mathfrak M}}$, and $p={{\mathds {1}}}-\lim_{\lambda\to\infty}E^\lambda$ [@uffe2 Theorem 1.5]. In the context of the relation of ${{\mathfrak M}}$ to ${{\mathcal M}}$, one may use this description to show that the group $(\theta_s)$ extends to automorphisms on $\hat{{{\mathcal M}}}^+$, the extended positive part of ${{\mathcal M}}$. In addition one can also show that $\hat{{{\mathfrak M}}}^+$ can be identified with the subset $\{m\in \hat{{{\mathcal M}}}^+: \theta_s(m)=m\mbox{ for all }s\in \mathbb{R}\}$ of $\hat{{{\mathcal M}}}^+$. An operator valued weight $T$ from say ${{\mathfrak M}}$ to a von Neumann subalgebra $\mathfrak{N}$, is then a map $T$ from ${{\mathfrak M}}^+$ into the extended positive part $\hat{\mathfrak{N}}^+$ of $\mathfrak{N}$, which satisfies - $T(\lambda x)=\lambda T(x)$ for all $\lambda\geq 0$, $x \in {{\mathfrak M}}^+$; - $T(x+y)=T(x)+T(y)$ for all $x, y\in {{\mathfrak M}}^+$; - $T(a^xa)=a^T(x)a$ for all $x\in {{\mathfrak M}}^+$, $a\in \mathfrak{N}$. Such an operator valued weight, is then some sort of generalised conditional expectation. This operator valued weight of interest to us, is defined on ${{\mathcal M}}^+$ by means of the formula $$T(x)=\int_{\mathbb{R}}\theta_s(x)\,ds\qquad x\in {{\mathcal M}}^+.$$Observe that by construction, we will have that $\theta_s(T(x))=T(x)$ for any $x\in {{\mathcal M}}^+$ – a fact which ensures that (as claimed) we do then indeed have that $T(x)\in \hat{{{\mathfrak M}}}^+$. Starting with the faithful normal semifinite weight $\omega$, the prescription $\hat{\omega}=\widetilde{\omega}\circ T$ yields a corresponding faithful normal semifinite weight on ${{\mathcal M}}$ (the so-called the dual weight of $\omega$), where $\widetilde{\omega}$ denotes the extension of $\omega$ to a normal weight on $\hat{{{\mathfrak M}}}^+$ - the extended positive part of ${{\mathfrak M}}$ [@uffe3]. It is now possible to show that the action on ${{\mathcal M}}$ of the modular automorphism group $\tilde{\sigma}_{t}(\cdot)$ produced by this dual weight, is implemented by the maps $\lambda(t)$ ($t\in\mathbb{R})$) in the sense that $\tilde{\sigma}_{t}(a)=\lambda(t)a\lambda(t)^$. (See the proof of [@uffe3 Lemma 5.2].) However with $\sigma_t$ denoting the modular automorphism group on ${{\mathfrak M}}$ produced by $\omega$, it is not difficult to see that on embedding ${{\mathfrak M}}$ into ${{\mathcal M}}$ we have by construction that $\sigma_s(a) = \lambda(s)a\lambda(s)^$ for all $s\in\mathbb{R}$. So one of the benefits of passing to the crossed product and equipping it with the dual weight, is that we now have a modular group with an inner action. By the Stone-von Neumann theorem there exists a densely defined positive operator $h$ on $L^2({{\rm I\!R}}, {{\mathcal H}})$ such that $h^{it} =\lambda(t)$. But then by [@Tak Theorem X.3.14], ${{\mathcal M}}$ is not only semifinite, but $h$ must also be affiliated ${{\mathcal M}}$. Since $\hat{\omega}$ is in fact faithful, $h$ must be non-singular (dense-range and injective). The affiliation of $h$ to ${{\mathcal M}}$ also ensures that $(\theta_s)$ has an action on $h$. It is now an exercise to conclude from the fact that $h^{it} =\lambda(t)$, that the action of $(\theta_s)$ as defined on the $\lambda(t)$’s, ensures that $\theta_s(h)=e^{-s}h$. We in fact have that $h^{it} = {\tilde{\Delta}_{\omega}}^{it}$ where ${\tilde{\Delta}_{\omega}}^{it}$ is the modular operator associated with the triple $({{\mathcal M}}, \hat{\omega},L^2({{\rm I\!R}}, {{\mathcal H}}))$. (In the context of the left Hilbert algebra approach to crossed products, ${\tilde{\Delta}_{\omega}}$ can be defined in a standard way as described see Lemma 1.15, Chapter X in [@Tak Lemma X.1.15].) The fact that $h^{it} = {\tilde{\Delta}_{\omega}}^{it}$ can be verified by arguing as in the uniqueness part of the proof [@Tak Theorem VIII.1.2] (see page 93). Following Takesaki one may now propose the weight $$\tau(\cdot) = \hat{\omega}( h^{-1/2} \ \cdot \ h^{-1/2}).$$ as a trace $\tau$ on ${{\mathcal M}}$. The verification that this is indeed a well-defined trace on ${{\mathcal M}}$, may be found in the [@PT Theorem 7.4]. (Take note that when those authors write $\omega( h^{-1} \cdot)$ in the proof of this theorem, what they have in mind is $\omega( h^{-1/2} \cdot h^{-1/2})$. See the notational convention introduced at the start of section 4 of this paper.) This is precisely the trace which we were looking for! Further computation then shows that the fact that $\theta_s(h)=e^{-s}h$, translates to the fact that for this trace, we have that $\tau\circ\theta_s=e^{-s}\tau$ for all $s\in\mathbb{R}$. The relation just stated also ensures that the automorphisms $\theta_s$ are continuous with respect to the topology of convergence in measure induced by $\tau$ on ${{\mathcal M}}$, and hence that they allow of unique extensions to $\widetilde{{{\mathcal M}}}$. In the final picture, the operator $h$ then turns out to be the density $\frac{d\hat{\omega}}{d\tau}$ of the dual weight $\hat{\omega}$ with respect to the trace just constructed. The fact that $\omega$ is a state, ensures that in this case $h$ is in fact $\tau$-measurable. [@terp Corollary II.6]. Having identified a suitable trace, we now have a framework for identifying the natural $\tau$-measurability criteria of field operators. However before doing so, we pause to comment on the specific nature of the operator $h$ as it relates to local algebras under discussion. One could write $$h = {\tilde{\Delta}_{\omega}}= e^{-H_{equilibrium}} \equiv e^{-K},$$ where, using “KMS” ideology, $K$ can be interpreted as the Hamiltonian describing the equilibrium dynamics for some fixed (non-moving) frame. For this (modular) dynamics, $\omega$ is moreover an equilibrium (KMS)-state. It is important to remember that the spectrum $\sigma(K)$ of $K$, in general, satisfies $$\sigma(K) = {{\rm I\!R}}.$$ Note that this property is a characteristic feature of modular groups of type $III_1$ factors! To fully appreciate this observation we remind the reader that $\widetilde{\sigma}_t(A)=\sigma_t(a)$ for all $a\in{{\mathfrak M}}$, where $\widetilde{\sigma}_t$ (respectively $\sigma_t$) is the modular group produced by the dual weight $\widehat{\omega}$ (respectively the weight $\omega$). (Compare the earlier discussion following the introduction of the operator-valued weight.) On the other hand, it is worth pointing out that, in general, $K$ is not equal to $H$. Thus, the spectral conditions assumed for $H$, are not applicable to $K$! To clarify this point and emphasise its connection to physical criteria, we will follow some arguments given in Chapter V.4 in [@haag]. Let $W$ denote the wedge in Minkowski space $\mathbb{M}$. $$W = \{r \in \mathbb{M}: r^1 > \|r^0\|, \ r^2,r^3 \quad \rm{arbitrary} \}$$ and $U(\Lambda(s))$ ($U(r)$) the unitary operators implementing the boosts $\Lambda(s)$ (the spacetime translations respectively). The corresponding generators will be denoted by $K$ and $P_{\mu}$, i.e. $$U(\Lambda(s)) = e^{iKs}, \quad U(r) = e^{iP_{\mu}}r^{\mu}.$$ Finally, let $\Theta$ stand for CPT-operator (cf Chapter II in [@haag]) and $R_1(\pi)$ denote a special rotation through the angle of $\pi$ around the $1$-axis. Bisognano and Wichmann proved, see [@BW1], [@BW2] \[BW\] Let $J_W$, $\Delta_W$ denote the modular conjugation and the modular operator for the pair $(\mathfrak{M}(W), \Omega)$. Then $$J_W = \Theta U(R_1(\pi)), \quad \Delta_W = e^{-2\pi K}.$$ Moreover, the modular automorphisms $\sigma_t$, $t \in {{\rm I\!R}}$, act geometrically as the boosts! We pause to note that although the Bisognano-Wichmann result is a theorem about ${{\mathfrak M}}(W)$, it is also a theorem about the geometry of the canonical modular group. Hence with $(\sigma_t)$ denoting the modular group, the fact that inside the crossed product $\sigma_t$ may be realised by $\sigma_t(a)=\lambda(t)a\lambda(t)^$, provides an elegant way of connecting their result with the crossed product approach. Field operators and $\tau$-measurability ---------------------------------------- Within the algebra of $\tau$-measurable operators affiliated to ${{\mathcal M}}$, one may identify spaces which represent the quantum $L^p$ and Orlicz spaces for $\mathfrak{M}$. We will revisit this point shortly. First we explore natural ways in which to assign $\tau$-measurability criteria to the field operators. In this ensuing discussion we will identify $\mathfrak{M}$ with $\pi_\sigma(\mathfrak{M})$ in order to simplify notation. Let us now apply the above ideas to the action of the field operators on ${{\mathfrak M}}(D)$ where $D$ is some double cone. At this point it is very seductive to suggest that one could simply require the field operators themselves to be measurable. But there is a problem with that. We noted earlier that the action of each $\theta_s$ extends to affiliated operators, hence also to the field operators. Each positive operator affiliated to $\mathfrak{M}(D)$ can of course be written as a pointwise increasing limit of positive elements of $\mathfrak{M}(D)$, and may therefore be regarded as an element of the extended positive part of ${{\mathfrak M}}(D)$. But we know from our earlier analysis, that the elements of $\hat{{{\mathfrak M}}}^+(D)$ are fixed points of this dual action. In other words we must have $\theta_s(\|\phi(f)\|)=\|\phi(f)\|$ for all $s$. So if $\phi(f)$ was in fact $\tau$-measurable, it would follow from Proposition II.10 of [@terp] that in the context of ${{\mathfrak M}}(D)$ the action of these operators would be bounded! So when imposing a regularity assumption related to $\tau$-measurability, we need to take care that we impose a regularity restriction which allows for unboundedness of the field operators. In search of such a criterion we turn to [@ML] for clues. In that paper a strong case was made that regular observables find their home in the Orlicz space $L^{\cosh-1}(\mathfrak{M}(D))$. The significance of this space is that it “contains” $\mathfrak{M}$ properly, and hence allows for unboundedness of observables. (What we at this point mean by containment should be explained. When in the tracial case we speak of containment, we mean standard set theoretic containment. However in the crossed product paradigm this translates to the statement that the canonical adjoint preserving embedding of embedding of $\mathfrak{M}$ into $L^{\cosh-1}(\mathfrak{M}(D))$, maps $\mathfrak{M}$ onto a proper subspace of $L^{\cosh-1}(\mathfrak{M}(D))$.) So a natural way in which we can assign $\tau$-measurability criteria to the field operators, is to ask that they canonically embed into the space of regular observables, namely $L^{\cosh-1}(\mathfrak{M}(D))$. To understand what this means, some background is necessary. Coming back to the general theory of quantum $L^p$-spaces, we have already noted that ${{\mathfrak M}}=\{a\in\widetilde{{{\mathcal M}}}:\theta_s(a)=a\mbox{ for all } s\in\mathbb{R}\}$. On a similar note Haagerup also showed that ${{\mathfrak M}}_\equiv\{a\in\widetilde{{{\mathcal M}}}:\theta_s(a)=e^{-s}a\mbox{ for all }s\in\mathbb{R}\}$. So if ${{\mathfrak M}}$ and ${{\mathfrak M}}_$ respectively represent the quantum spaces $L^\infty({{\mathfrak M}})$ and $L^1({{\mathfrak M}})$, it makes sense to suggest that the $L^p({{\mathfrak M}})$ spaces $1\leq p\leq\infty$ may well be represented by the scale of spaces $L^p({{\mathfrak M}})=\{a\in\widetilde{{{\mathcal M}}}:\theta_s(a)=e^{-s/p}a\mbox{ for all }s\in\mathbb{R}\}$. Haagerup’s triumph was in showing that these are all Banach spaces with respect to the subspace topology inherited from the topology of convergence in measure on $\widetilde{{{\mathcal M}}}$, and that there exists a tracial functional $tr$ on $L^1({{\mathfrak M}})$ (different from $\tau$) in terms of which one may define a dual action of $L^{p'}$ on $L^p$, in terms of which one may then develop a theory which reproduces much of the classical theory of $L^p$-spaces with a remarkable degree of faithfulness. By [@terp Lemma II.5, Definitions II.13 & 11.14], the natural norm on each of the Haagerup $L^p$-spaces may be realised by the formula $$\\|a\\|_p=\inf\{\varepsilon>0: \tau(E^{\|a\|}(\varepsilon, \infty))\leq 1\}.$$ Moreover in the tracial case this construction produces spaces which are canonically isometric to their more classically defined tracial counterparts. Details may be found in Chapter II of [@terp]. We are however interested in spaces more general than $L^p$-spaces, namely Orlicz spaces. In order to introduce Orlicz spaces, we first need to acquaint readers with the concept of a Young’s function. A convex function $\Psi : [0, \infty) \to [0, \infty]$ is called a Young’s function if - $\Psi(0) = 0$ and $\lim_{u \to \infty} \Psi(u) = \infty$; - it is neither identically zero nor infinite valued on all of $(0, \infty)$, - and is left continuous at $b_\Psi = \sup\{u > 0 : \Psi(u) < \infty\}$. A noteworthy aspect of Young’s functions, is that they come in complementary pairs, with the complementary function defined by the formula $\Psi^(u) = \sup_{v > 0}(uv - \Psi(v))$. Together the pair $(\Psi,\Psi^)$ satisfy the Hausdorff-Young inequality $$st\leq \Psi(s)+\Psi^(t)\mbox{ for all }s,t\geq 0.$$ To each Young’s function $\Psi$, we may associate a corresponding Orlicz space. With $L^0$ denoting the space of all measurable functions on some $\sigma$-finite measure space $(\Omega, \Sigma, m)$, the Orlicz space associated with a given Young’s function $\Psi$, may be defined by the prescription: $f\in L^0$ belongs to $L^{\Psi} \Leftrightarrow \int\Psi(\lambda \|f\|)\,d\nu<\infty$ for some $\lambda = \lambda(f) > 0.$ These spaces admit of several natural norms, but ultimately these norms turn out to be equivalent. (See [@BS §4.8].) For the sake of completeness we mention the two most important norms. - Luxemburg-Nakano norm: $\\|f\\|_\Psi = \inf\{\lambda > 0 : \\|\Psi(\|f\|/\lambda)\\|_1 \leq 1\}.$ - Orlicz norm: $\\|f\\|^O_\Psi = \sup\{\|\textstyle{\int_\Omega} fg\, dm\| : g\in L^{\Psi^}, \\|g\\|_{\Psi^}\leq 1\}.$ A notational convention used by many authors to indicate the norm in view, is to write $L^\Psi$ when the Luxemburg norm is used, and $L_\Psi$ when the Orlicz norm is used. There is an interesting connection between the spaces produced by a complementary pair of Young’s functions, namely that the spaces produced by such a pair, are Köthe duals of each other! By the claim that $L_{\Psi^}(X, \Sigma, \nu)$ is the Köthe dual of say $L^{\Psi}(X, \Sigma, \nu)$, we specifically mean that a measurable function $f$ belongs to $L_{\Psi^}(X, \Sigma, \nu)$ if and only if $fg\in L^1$ for every $g\in L^\Psi$, and that the norm on $L_{\Psi^}$ is realised by the formula $\sup\{\|\textstyle{\int_\Omega} fg\, dm\| : g\in L^{\Psi}, \\|g\\|_{\Psi}\leq 1\}$. (See [@BS §4.8]) To be able to define Orlicz spaces for type III algebras, we need the concept of a fundamental function, specifically fundamental functions of Orlicz spaces $\{L^\Psi(0,\infty), L_\Psi(0,\infty)\}$. For each such space, say $X$, the associated fundamental function* is defined by the prescription $$\varphi_X(t)=\\|\chi_E\\|_X \mbox{ where } \nu(E)=t.$$ In the case $X=L^\Psi(0,\infty)$ we will write $\varphi_\Psi$ for the fundamental function, and in the case $X=L_\Psi(0,\infty)$ write $\widetilde{\varphi}_\Psi$. These functions turn out to be so-called “quasi-concave functions” on $[0,\infty)$ (see [@BS §2.5] for a detailed discussion of such functions). For these fundamental functions the following facts are known to hold (see [@BS Theorem 2.5.2, Corollary 4.8.15 & Lemma 4.8.17]): $$\label{fundeqn} \varphi_\Psi(t)=\frac{1}{\Psi^{-1}(1/t)} \mbox{ and }\varphi_\Psi(t)\widetilde{\varphi}_{\Psi^}(t)=t\mbox{ for all }t.$$ Given $\Psi$, the Orlicz space $L^\Psi({{\mathfrak M}})$ associated with ${{\mathfrak M}}$ may then be defined by means of the following prescription (see [@LM2 Lemma 5.11]): Let $d_s=\widetilde{\varphi}_{\Psi^}(e^{-s}h)^{-1}\widetilde{\varphi}_{\Psi^}(h)$. Then $$L^\Psi({{\mathfrak M}}) =\{a\in \widetilde{{{\mathcal M}}}: \theta_s(a)=e^{-s}d_s^{1/2}ad_s^{1/2} \mbox{ for all } s\leq 0\}.$$To define $L_\Psi({{\mathfrak M}})$ one simply uses $\varphi_{\Psi^}$ in the definition instead of $\widetilde{\varphi}_{\Psi^}$. We point out that in the case where $\Psi(t)=t^p$, this prescription yields exactly Haagerup’s definition of $L^p$-spaces. As with the Haagerup $L^p$-spaces, these turn out to be complete quasi-normed spaces under the topology of convergence in measure inherited from $\widetilde{{{\mathcal M}}}$ with the natural quasinorm on each of these spaces once again being given by the prescription $$\\|a\\|_\Psi=\inf\{\varepsilon>0: \tau(E^{\|a\|}(\varepsilon, \infty))\leq 1\}.$$ (See [@L Proposition 3.11].) In the case where $\mathfrak{M}$ is semifinite and $\omega$ a trace, we will up to Fourier transform have that ${{\mathcal M}}=\mathfrak{M}\otimes L^\infty(\mathbb{R})$ (see [@terp pp 62-63]), where here the von Neumann algebra tensor product is in view. The canonical trace on ${{\mathcal M}}$ will in this case be $\omega(\cdot) \otimes \int_{\mathbb{R}}\cdot e^{-t}dt$ with the elements of for example the Orlicz space $L^\Psi(\mathfrak{M})$ being simple tensors of the form $g\otimes \varphi_{\Psi}(e^t)$, where $g$ is an element of the “tracial” Orlicz space $L^\Psi(\mathfrak{M},\omega)=\{a\in\widetilde{\mathfrak{M}}: \Psi(\lambda\|a\|)\in \widetilde{\mathfrak{M}}\mbox{ and }\omega(\Psi(\lambda\|a\|))<\infty\}$. Moreover the map sending $g$ to $g\otimes \varphi_{\Psi}(e^t)$ turns out to be a linear isometry from $L^\Psi(\mathfrak{M},\omega)$ onto $L^\Psi(\mathfrak{M})$ when each is equipped with its natural (quasi)-norm. (The natural norm on $L^\Psi(\mathfrak{M},\omega)$ is given by the “classical” prescription $\\|g\\|_\Psi= \inf\{\lambda > 0 : \tau(\Psi(\|g\|/\lambda)) \leq 1\}$. (See for example [@L Corollary 2.3].) So in the tracial case, this prescription produces exactly the “classical” Orlicz spaces. Coming back to field operators and their action in the context of ${{\mathfrak M}}(D)$, we remind the reader that the self-adjoint elements of the space $L^{\cosh-1}(\mathfrak{M}(D))$ represents the space of regular observables. If then the field operators are to be regular* in this sense, a natural restriction to place on them would be to require them to embed into the space $L^{\cosh-1}(\mathfrak{M}(D))$ in a natural way. This leads us to the following definition A field operator $\phi(f)$ affiliated to $\mathfrak{M}(D)$ is said to satisfy an $L^{\cosh-1}$ regularity restriction if the strong product $\varphi_{\cosh-1}(h)^{1/2}\phi(f)\varphi_{\cosh-1}(h)^{1/2}$ is a closable operator for which the closure is $\tau$-measurable, i.e. the closure is an element of the space $\widetilde{{{\mathcal M}}}$. Given that each of the $\phi(f)$’s are fixed points of the action of $(\theta_s)$, it is a simple matter to verify that an $L^{\cosh-1}$ regularity restriction on some $\phi(f)$, ensures that the closure of $\varphi_{\cosh-1}(h)^{1/2}\phi(f)\varphi_{\cosh-1}(h)^{1/2}$ satisfies the membership criteria for $L^{\cosh-1}(\mathfrak{M}(D))$ described above. To see this observe that $\theta_s(\varphi_{\cosh-1}(h)^{1/2}\phi(f)\varphi_{\cosh-1}(h)^{1/2}) = \varphi_{\cosh-1}(e^{-s}h)^{1/2}\phi(f)\varphi_{\cosh-1}(e^{-s}sh)^{1/2}$. Now observe that with $\Psi^$ denoting the conjugate Young’s function to ${\cosh-1}$ we may use the second formula in equation \[fundeqn\] to show that $$\varphi_{\cosh-1}(e^{-s}h)=e^{-s}[\varphi_{\Psi^}(e^{-s}h)^{-1}\varphi_{\Psi^}(h)]\varphi_{\cosh-1}(h).$$ We re-emphasise that the $L^{\cosh-1}$ regularity restriction allows room for unboundedness of the field operators! This restriction clearly has the same flavour as the generalised $H$-boundedness restriction on the field operators $\pi_{\sigma}(\phi_e)$ mentioned earlier, which implied their affiliation to the local algebras $\mathfrak{M}(D)$. The difference here is that we have a “symmetrised” product, and that we ask for this product to be $\tau$-measurable rather than bounded. We note that the $L^{\cosh-1}$ regularity condition is inherited by certain “translates” of open bounded regions ${{\mathcal O}}$ for which this condition is known to hold. Specifically given $g= (a,\Lambda) \in P^{\uparrow}_+$ and ${{\mathcal O}}\subset \mathbb{M}$, it follows from (L2) that there is an automorphism $\alpha_g$ in $Aut({{\mathfrak M}})$ such that $\alpha_g({{\mathfrak M}}({{\mathcal O}})) = {{\mathfrak M}}(g{{\mathcal O}})$. If $g$ is chosen so that $\alpha_g$ and $\sigma_t$ commute, then we may once again use [@HJX Theorem 4.1] and [@LM2 Corollary 4.5] to show that the action of the automorphism $\alpha_g$ extends to a map $\widehat{\alpha}_g$ which maps $\mathfrak{M}({{\mathcal O}}) \rtimes_{\sigma} {{\rm I\!R}}$ onto $\mathfrak{M}(g{{\mathcal O}}) \rtimes_{\sigma} {{\rm I\!R}}$, and $h_{{{\mathcal O}}}= \frac{d\hat{\omega}_{{{\mathcal O}}}}{d\tau_{{{\mathcal O}}}}$ onto $h_{g{{\mathcal O}}}=\frac{d\hat{\omega}_{g{{\mathcal O}}}}{d\tau_{g{{\mathcal O}}}}$. So $\phi(f)$ is a field operator affiliated to $\mathfrak{M}({{\mathcal O}})$, an $L^{\cosh-1}$ regularity restriction on $\phi(f)$ in terms of ${{\mathfrak M}}({{\mathcal O}})$, translates to an $L^{\cosh-1}$ regularity restriction on $\widehat{\alpha}_g(\phi(f))$ in terms of ${{\mathfrak M}}(g{{\mathcal O}})$. The concept of $L^{\cosh-1}$-regularity may of course be considered in terms of any of the local algebras ${{\mathfrak M}}({{\mathcal O}})$. However in some sense local algebras ${{\mathfrak M}}(W)$ corresponding to a wedge in Minkowski space occupy a special place in this regard. To support this claim we remind the reader that it is precisely the result of Bisognano and Wichman regarding such algebras (Theorem \[BW\]) which served as the starting point for motivating the utility of the crossed product construction for local algebras. In addition, as we will see below, such algebras serve as the context for proving that the concept of $L^{\cosh-1}$ regularity, behaves well with regard to “subtheories”. Before we are able to prove such good behaviour of $L^{\cosh-1}$-regularity with respect to subtheories, some background is necessary. In particular we need Borchers’ concept of “modular covariant subalgebras”, and a description of how that concept relates to the construction of subtheories. (See sections VI.1 and VI.3 of [@Bor2].) In the language of Borchers, a modular covariant von Neumann subalgebra ${{\mathfrak N}}$ of a von Neumann algebra ${{\mathfrak M}}$, is an algebra which satisfies the requirement that $\Delta_{{{\mathfrak M}}}^{it}{{\mathfrak N}}\Delta^{-it}_{{{\mathfrak M}}}={{\mathfrak N}}$ for all $t\in \mathbb{R}$ [@Bor2 Def VI.1.1]. But as Borchers notes [@Bor2 Theorem VI.1.3], it follows from a result of Takesaki [@Tak Theorem IX.4.2], that this requirement is equivalent to the existence of a faithful normal conditional expectation $\mathcal{E}$ from ${{\mathfrak M}}$ onto ${{\mathfrak N}}$, which leaves the canonical faithful normal weight $\omega$ on ${{\mathfrak M}}$, invariant in the sense that $\omega =\omega \circ \mathcal{E}$. To apply these concepts to local algebras, one needs the notion of “coherent” subalgebras. Specifically, following Borchers [@Bor2 Def VI.3.1], we make the following definition: Let ${{\mathfrak M}}(W)$ be a von Neumann algebra associated with a wedge $W$, and admitting a cyclic and separating vector. We say that a collection of modular covariant subalgebras ${{\mathfrak N}}(W)$ of ${{\mathfrak M}}(W)$ (where $W$ ranges over all wedges in Minkowski space), is coherent if all of the projections $E_W$ from $\mathcal{H}$ onto $[{{\mathfrak N}}(W)\Omega]$ coincide. In [@Bor2] Borchers defines the concept of coherence for subalgebras corresponding to double cones as well. However for the purpose of recovering his main theorem regarding subtheories (stated below), we only need information regarding the coherence of subalgebras corresponding to wedges. Following Borchers, we assume that the considered quantum field fulfills the Bisognano-Wichmann property, i.e. that for every wedge the modular groups acts locally like the associated group of Lorentz boosts on the underlying space* [@Bor2 Definition III.1.4]. Furthermore again following Borchers, we set ${{\mathfrak M}}(D)=\cap\{{{\mathfrak M}}(gW): D\subset gW, g\in P^{\uparrow}_+\}$. \[B-subtheory\] Let $\{{{\mathfrak M}}(D), U(\Lambda,x), \mathcal{H}, \Omega\}$ be a family of local observables fulfilling the conditions (L1) – (L6) described in the introduction, and assume that this family is in the vacuum sector. (In other words that each ${{\mathfrak M}}(D)$ is a von Neumann algebra and that $\Omega$ is cyclic and separating.) Given any coherent family ${{\mathfrak N}}(W)$ of modular covariant subalgebras of ${{\mathfrak M}}(W)$ (where $W$ ranges over all wedges in Minkowski space), there exists a family of local observables $\{{{\mathfrak N}}(D), U(\Lambda,x), E\mathcal{H}, \Omega\}$ fulfilling those same conditions. In other words $\{{{\mathfrak N}}(D), U(\Lambda,x), E\mathcal{H}, \Omega\}$ determines a subtheory. In particular for every wedge we have that ${{\mathfrak N}}(W)=\bigvee\{{{\mathfrak N}}(D): D\subset W\}$. (Here ${{\mathfrak N}}(D)$ is defined to be $\cap\{{{\mathfrak N}}(W): D\subset W\}$, and $E$ is the projector onto $[{{\mathfrak N}}(W)\Omega]$.) We may now further refine the above theorem to show that such subtheories also behave well with regard to the $L^{\cosh-1}$-regularity condition. Assume that the conditions stated in Theorem \[2.3\] hold and that $\{{{\mathfrak M}}(D), U(\Lambda,x), \mathcal{H}, \Omega\}$ is the net of local algebras implied by that theorem. Let $\{{{\mathfrak N}}(D), U(\Lambda,x), E\mathcal{H}, \Omega\}$ be a subtheory constructed as described in the preceding theorem from some coherent family ${{\mathfrak N}}(W)$ of modular covariant subalgebras of ${{\mathfrak M}}(W)$ (where $W$ ranges over all wedges in Minkowski space) Let $\phi(f)$ denote the extension of the field operators described in Theorem \[2.3\] which are affiliated to ${{\mathfrak M}}(W)$. Then for any wedge $W$ or double cone $D$, a field operator $\phi(f)$ which is affiliated to ${{\mathfrak N}}(W)$ rather than ${{\mathfrak M}}(W)$ (respectively ${{\mathfrak N}}(D)$) will fulfill an $L^{\cosh-1}$-regularity condition with respect to ${{\mathfrak N}}(W)$ (respectively ${{\mathfrak N}}(D)$) if and only if it fulfills such a condition with respect to ${{\mathfrak M}}(W)$ (respectively ${{\mathfrak M}}(D)$). The proofs being entirely analogous, we prove the claim for some double cone $D$. To start off with we note that the hypothesis of the theorem ensures that there exists a faithful normal conditional expectation $\mathcal{E}_D$ from ${{\mathfrak M}}(D)$ onto ${{\mathfrak N}}(D)$, which leaves the faithful normal state $\omega$ on ${{\mathfrak M}}(D)$ invariant. (See [@Bor2 Lemma 6.3.3].) It now follows from ([@HJX Theorem 4.1], [@LM2 Corollary 4.5] [@Gol Lemma 4.8]), that we may regard ${{\mathcal N}}_{D}={{\mathfrak N}}(D) \rtimes {{\rm I\!R}}$ as a subspace of ${{\mathcal M}}_D={{\mathfrak M}}(D) \rtimes {{\rm I\!R}}$ in such a way that the respective canonical traces satisfy $\tau_{{{\mathcal N}}}=\tau_{{{\mathcal M}}}\|_{{{\mathcal N}}_{D}}$, with in addition $h_{{{\mathcal N}}_D} = \frac{d\hat{\omega}_{{{\mathfrak N}}}}{d\tau_{{{\mathcal N}}}} = \frac{d\hat{\omega}_{{{\mathfrak M}}}}{d\tau_{{{\mathcal M}}}} = h_{{{\mathcal M}}_D}$. Once these identifications have been made, the conclusion of the theorem is obvious. Consequences of $L^{\cosh-1}$ regularity of field operators ----------------------------------------------------------- There is further evidence that this restriction is a physically reasonable one to make for field operators, namely that membership of $L^{\cosh-1}(\mathfrak{M}(D))$, will ensure that the “generalised moments” of the field operators are all finite. This restriction may therefore be seen as a requirement which complements the requirements noted in for example Equation \[fmoment\] and Axiom (F1). What we mean by these statement is contained in the following Proposition. (For the sake of simplicity of notation, we will simply write $\mathfrak{M}$ for $\mathfrak{M}(D)$ in the remainder of this subsection. Let $\mathfrak{M}$ be $\sigma$-finite and let ${{\mathcal M}}$ and $h$ be as before. If for each $f$ we have that $\varphi_{\cosh-1}(h)^{1/2}\phi(f)\varphi_{\cosh-1}(h)^{1/2}\in L^{\cosh-1}(\mathfrak{M})$, then for any $1\leq p<\infty$ we will have that $h^{1/(2p)}\phi(f)h^{1/(2p)}\in L^p(\mathfrak{M})$. Hence in their action on $\mathfrak{M}$, the generalised moments of the field operators will all be finite, in the sense that for any $a\in \mathfrak{M}$, $tr(a[h^{1/(2p)}\phi(f)h^{1/(2p)}]^p)$ will always be a finite complex number. (Here $tr$ is the tracial functional on $L^1$, not $\tau$.) Given any $2\leq p <\infty$, select $m\in \mathbb{N}$ so that $2m\leq p <2(m+1)$. Then of course $t^p\leq t^{2m}+t^{2(m+1)}$ for all $t\geq 0$. On considering the Maclaurin expansion of $\cosh(t)-1$, it now trivially follows that $\frac{1}{(2(m+1))!}t^p \leq\frac{1}{(2(m+1))!}(t^{2m}+t^{2(m+1)})\leq \cosh(t)-1$ for all $t\geq 0$. It is now a straightforward exercise to conclude from this fact that $\left(\frac{t}{(2(m+1))!}\right)^{1/p}\leq [\mathrm{arccosh}(t^{-1}+1)]^{-1}=\varphi_{\cosh-1}(t)$. In other words the function $\gamma_p(t)= \frac{t^{1/p}}{\varphi_{\cosh-1}(t)}$ is well-defined bounded continuous function on $(0,\infty)$. Hence $\gamma_p(h)$ is a bounded operator. So if $\varphi_{\cosh-1}(h)^{1/2}\phi(f)\varphi_{\cosh-1}(h)^{1/2}$ is $\tau$-measurable, then so is $\gamma_p(h)^{1/2}[\varphi_{\cosh-1}(h)^{1/2}\phi(f)\varphi_{\cosh-1}(h)^{1/2}]\gamma_p(h)^{1/2}= h^{1/(2p)}\phi(f)h^{1/(2p)}$. It is now a simple matter to verify that this $\tau$-measurable operator satisfies all the membership criteria for $L^p(\mathfrak{M})$. It remains to consider the case $1\leq p <2$. For $\sigma$-finite algebras it is however known that $L^2(\mathfrak{M})$ contractively embeds into any $L^p(\mathfrak{M})$ (where $1\leq p <2$) in a way which will send $h^{1/4}\phi(f)h^{1/4}$ to $h^{1/(2p)}\phi(f)h^{1/(2p)}$ (see [@kos]). This proves the proposition. Although thus far our focus has been on analysing the link between field operators and the space $L^{\cosh-1}(\mathfrak{M})$, in concluding this section we make some brief comments about the space $L\log(L+1)(\mathfrak{M})$. This is an Orlicz space produced by the Young’s function $t\log(t+1)$. To clarify our interest in this Orlicz space, we remind the reader that entropy is a crucial tool in the thermodynamical description of quantum systems. However the standard approach to statistical mechanics leads to serious problems with the definition of entropy (see Wehrl [@weh]). The Orlicz space $L\log(L+1)$, being dual to $L^{\cosh-1}$, can be considered to be the natural home for the states acting on regular observables. Importantly in [@ML] a strong case was made that the space $L^1\cap L\log(L+1)$ is home for the states with good entropy (see [@ML Proposition 6.8]). So in addition to the duality noted above, the space $L\log(L+1)$ can also be considered as the space generated by states with “good” entropy. A definition of entropy for type III algebras --------------------------------------------- As a first step in achieving this outcome, we note that classically $L^1\cap L\log(L+1)$ is an Orlicz space produced by the Young’s function $$\Psi_{ent}(t)=\max(t,t\log(t+1))=\left\{ \begin{array}{ll} t & 0\leq t\leq e-1\\ t\log(t+1) & e-1\leq t\end{array} \right.$$So in order to extend the ideas of [@ML] to the type III case, we need to faithfully follow the algorithm for constructing Orlicz spaces for type III algebras, in order to produce a type III analogue of the space $L^1\cap L\log(L+1)$. Getting back to the classical setting, we will for simplicity of computation assume that each of $L\log(L+1)(0,\infty)$ and $L^1\cap L\log(L+1)(0,\infty)$ are equipped with the Luxemburg norm. It is then an exercise to see that the fundamental function of $L^1\cap L\log(L+1)(0,\infty)$ is of the form $\varphi_{ent}(t)=\max(t,\varphi_{\log}(t))$. It is this fundamental function that we use to construct our type III analogue of $L^1\cap L\log(L+1)$ according to the prescription. Let us denote this space by $L^{ent}(\mathfrak{M})$. From the above computations, it is clear that the functions $\zeta_1(t)= \frac{t}{\varphi_{ent}(t)}$, and $\zeta_{\log}(t)= \frac{\varphi_{\log}(t)}{\varphi_{ent}(t)}$ are both continuous and bounded above (by 1) on $(0,\infty)$. Hence the operators $\zeta_1(h)$ and $\zeta_{\log}(h)$ are both contractive elements of ${{\mathcal M}}$. It is now an exercise to see that the prescriptions $x\to\zeta_1(h)^{1/2}x\zeta_1(h)^{1/2}$ and $x\to\zeta_{\log}(h)^{1/2}x\zeta_{\log}(h)^{1/2}$ respectively yield continuous embeddings of $L^{ent}(\mathfrak{M})$ into $L^1(\mathfrak{M})$ and $L\log(L+1)(\mathfrak{M})$. Let $x\in L^{ent}(\mathfrak{M})^+$ be given. For such an element, the quantity $$\inf_{\varepsilon >0}[\varepsilon\tau(E^{\|\zeta_{\log}(h)^{1/2}x\zeta_{\log}(h)^{1/2}\|}(\varepsilon, \infty)) + \log(\varepsilon)\\|\zeta_1(h)^{1/2}x\zeta_1(h)^{1/2}\\|_1]$$will then be well-defined (albeit possibly infinite-valued). It is this quantity that we propose as the type III analogue of von Neumann entropy. To see that this does indeed make sense, recall that in the case where $\mathfrak{M}$ is semifinite and $\omega$ a trace, ${{\mathcal M}}$ may be identified with ${{\mathfrak M}}\otimes L^\infty(\mathbb{R})$, and that with respect to this identification, any given Orlicz space $L^\Psi(\mathfrak{M})$ will consist of the simple tensors in ${{\mathfrak M}}\otimes L^\infty(\mathbb{R})$ of the form $f\otimes \varphi_{\Psi}(e^t)$, where $f$ is an element of the “tracial” Orlicz space $L^\Psi(\mathfrak{M},\omega)=\{g\in\widetilde{\mathfrak{M}}: \Psi(\lambda\|g\|)\in \widetilde{\mathfrak{M}}\mbox{ and }\omega(\Psi(\lambda\|g\|))<\infty\}$. It therefore easily follows from [@L Corollary 2.3], that for $x=f\otimes \varphi_{ent}(e^t)$, the quantity $$\inf_{\varepsilon >0}[\varepsilon\tau(E^{\|\zeta_{\log}(h)^{1/2}x\zeta_{\log}(h)^{1/2}\|}(\varepsilon, \infty)) + \log(\varepsilon)\\|\zeta_1(h)^{1/2}x\zeta_1(h)^{1/2}\\|_1]$$ may be rewritten as $$\inf_{\varepsilon >0}[\varepsilon\omega((f/\varepsilon)\log((f/\epsilon)+{{\mathds {1}}})) +\log(\varepsilon)\omega(f)]=\inf_{\varepsilon >0}\omega(f\log(f+\varepsilon{{\mathds {1}}})).$$By [@ML Proposition 6.8], this yields exactly $\omega(f\log(f))$. Hence that above prescription for entropy, is a faithful extension of the existing descriptions on semifinite von Neumann algebras Concluding remarks ------------------ The analysis in the preceding section leads to the following conclusions: \[1\] The $\tau$-measurability regularity conditions on the field operators are strongly related to the principle of relativity, i.e. on the one hand the affiliation of the field operators to the local algebra $\mathfrak{M}(D)$ results from some form of regularity with respect to the generator $H$ of time translations, whilst the embedding of the field operators into the space $L^{\cosh-1}(\mathfrak{M}(D))$, is a consequence of regularity with respect to the generator $K$ of boosts. Consequently, both of the basic transformations employed by the principle of relativity are used! and \[2\] The axioms of QFT imply that field operators should have all moments finite. This requirement is intimately related to the requirement that these operators embed into the space $L^{\cosh-1}(\mathfrak{M})$. The states which come from the space $L^{ent}(\mathfrak{M})$ (the type III analogue of $L^1\cap L\log(L+1)$ described above) moreover all admit a good definition of entropy. Consequently, the strategy based on the quantum Pistone-Sempi theorem, see [@LM], [@ML] leads to the conjecture that the proper formalism for QFT is that based on the pair of quantum Orlicz spaces $\langle L^{\cosh-1}(\mathfrak{M}), L\log(L+1)(\mathfrak{M})\rangle$. Local algebras and graded algebras of differential forms ======================================================== Having considered the application of integrable structures to local algebras, we now turn our attention to differential structures. We note that such structures are indispensable tools for the description of time evolution of quantum systems. In other words, integrable structures provide a rather static setting for an analysis of systems, whilst differential structures are employed for an examination of time evolution of these systems. Throughout $M$ will be a (smooth) $d$-dimensional, connected time-oriented globally hyperbolic Lorentzian manifold, $\mathbf{g}$ a Lorentzian metric on $M$, and $\mathbb{M}$ d-dimensional Minkowski space-time. The axiomatisation of local algebras for manifolds rather than Minkowski space, was pioneered by Dimock, et al, as early as 1980 [@D1; @D2]. However it was not until 2003 that Brunetti, Fredenhagen and Verch added the all important principle of locality to the covariance axioms for these algebras. For the sake of background we review some material from [@BFV], [@BF]. All these results are formulated in strongly categorical language. However the crucial result for us, is the description of such local algebras given in [@BFV Theorem 2.3]. For a local algebra fulfilling their criteria, Brunetti, Fredenhagen and Verch obtain the result below. This encapsulates the structural information which will form the starting point of our subsequent modelling. Since in their theory the local algebras can be either $C^$-algebras or von Neumann algebras, we here follow their convention of denoting the local algebras by ${{\mathcal A}}({{\mathcal O}})$. Here ${{\mathcal K}}(M)$ denotes the set of all open subsets in $M$ which are relatively compact and which for each pair of points $x$ and $y$, also contain all $\mathbf{g}$-causal curves in $M$ connecting $x$ and $y$. Moreover, as before ${{\mathcal A}}({{\mathcal O}})$ denotes the $C^$-algebra generated by field operators $\phi(f)$ with test function supported in ${{\mathcal O}}$, i.e. $\mathrm{supp} f \subseteq {{\mathcal O}}$. We emphasize that $M$ is now a much more general manifold than $\mathbb{M}$, which was the focus in the previous section. [@BFV] \[localnet\] Let ${\mathscr A}$ be a covariant functor with the properties stated in [@BFV Def. 2.1], and define a map ${{\mathcal K}}(M,\mathbf{g}) \owns O \mapsto {{\mathcal A}}(O) \subset {\mathscr A}(M,\mathbf{g})$ by setting $${{\mathcal A}}(O) := \alpha_{M,O}({\mathscr A}(O,\mathbf{g}_O))\,,$$ having abbreviated $\alpha_{M,O} \equiv \alpha_{\iota_{M,O}}$. The following statements hold: - Isotony, i.e. $${{\mathcal O}}_1 \subset {{\mathcal O}}_2 \Rightarrow {{\mathcal A}}({{\mathcal O}}_1) \subset {{\mathcal A}}({{\mathcal O}}_2) \qquad \mbox{for all}\qquad {{\mathcal O}}_1,{{\mathcal O}}_2 \in {{\mathcal K}}(M)\,.$$ - If there exists a group $G$ of isometric diffeomorphisms $\kappa: M \to M$ (so that $\kappa_{\mathbf{g}} = {\mathbf{g}}$) preserving orientation and time-orientation, then there is a representation by C$^$-algebra automorphisms ${\alpha}_{\kappa} : {{\mathcal A}}(M) \to {{\mathcal A}}(M)$ such that $${\alpha}_{\kappa}({{\mathcal A}}({{\mathcal O}})) = {{\mathcal A}}(\kappa({{\mathcal O}}))\,, \quad {{\mathcal O}}\in {{\mathcal K}}(M)\,.$$ - If the algebra belongs to the “causal” category of their theory, then it also holds that $$[{{\mathcal A}}({{\mathcal O}}_1),{{\mathcal A}}({{\mathcal O}}_2)] = \{0\}$$ for all ${{\mathcal O}}_1,{{\mathcal O}}_2 \in {{\mathcal K}}(M)$ with ${{\mathcal O}}_1$ causally separated from ${{\mathcal O}}_2$. - If the theory of which this algebra is part fulfills the time-slice axiom, and $\Sigma$ is a Cauchy-surface in $M$ with $S \subset \Sigma$ open and connected, then for each ${{\mathcal O}}\in {{\mathcal K}}(M)$ with ${{\mathcal O}}\supset S$ it holds that $${{\mathcal A}}({{\mathcal O}}) \supset {{\mathcal A}}(S^{\perp}{}^{\perp})$$ where $S^{\perp}{}^\perp$ is the double causal complement of $S$, and ${{\mathcal A}}(S^{\perp}{}^{\perp})$ is defined as the smallest $C^$-algebra formed by all ${{\mathcal A}}({{\mathcal O}}_1)$, ${{\mathcal O}}_1 \subset S^{\perp}{}^{\perp}$, ${{\mathcal O}}_1 \in {{\mathcal K}}(M)$. To formulate the next result some preliminaries are necessary. We wish to consider local algebras generated by field operators which are solutions of the Klein-Gordon equation. As was stated, $M$ stands for a manifold satisfying the conditions given at the beginning of this section. $C^{\infty}_0(M)$ will denote the space of smooth, real valued functions on $M$ which have compact support. Following Dimock [@D1] (see also [@BFV]), we will describe the CCR algebra of bosonic fields on the manifold $M$ given by solutions of the Klein-Gordon equation. The crucial point in his construction, is that global hyperbolicity of $M$ entails existence of global fundamental solutions $E$ for the Klein-Gordon equation $(\Box + m^2 + \xi R)\phi = 0$, where $m\geq 0$ , $\xi \geq0$ are constants, and $R$ is the scalar curvature of the metric on $M$. In particular, $E = E^{adv} - E^{ret}$, where $E^{adv/ret}$ (advanced/retarded, respectively) are well defined maps such that $E^{adv/ret}: C^{\infty}_0(M) \to C^{\infty}(M)$. It was shown by Dimock [@D1], that the property of bosonic field operators being solutions of the Klein-Gordon equation, is characterized by the following relations: $$e^{i\phi(f)} e^{i \phi(f^{\prime})} = e^{i\phi(f^{\prime})} e^{i \phi(f)} e^{-i<f,Ef^{\prime}>}.$$ This leads to the following form of Weyl relations for $W(f) \equiv e^{i\phi(f)}$: $$W(f) W(f^{\prime}) = e^{- \frac{i}{2} <f, E f^{\prime}>} W(f+f^{\prime}).$$ Denote by ${{\mathcal R}}$ the real vector space $E\left(( C^{\infty}_0(M)\right)$. We note that $\sigma(f,g) = \int_M f Eg dV_M$, where $dV_M$ is the volume form on $M$, gives a symplectic form on ${{\mathcal R}}$, i.e. $({{\mathcal R}}, \sigma)$ is the symplectic space. The $C^$-algebra of canonical commutation relations over a symplectic form $({{\mathcal R}}, \sigma)$, writen as $\mathfrak{M}({{\mathcal R}}, \sigma)$, is by definition the $C^$-algebra generated by elements $$W(-f) = W(f)^, \quad W(f) W(g) = e^{-\frac{i}{2} \sigma(f,g)} W(f+g).$$ Let $(M_1, g_1)$ and $(M_2, g_2)$ be two given manifolds satisfying the prescribed conditions. Assume that $\psi: M_1 \to M_2$ is a diffeomorphism satisfying $\psi_g_1 = g_2\|_{\psi(M_1)}$ and which also preserves causality and orientation; see [@BFV] for details. Denote by $E^{\psi}$ the global fundamental solution of the Klein-Gordon equation on $\psi(M)$, and write ${{\mathcal R}}^{\psi} \equiv E^{\psi}\left(C^{\infty}_0(\psi(M))\right)$. Dimock [@D1], has shown that $E^{\psi} = \psi_ \circ E \circ \psi^{-1}_$ and ${{\mathcal R}}^{\psi} = \psi_ {{\mathcal R}}$, where $\psi_* f = f \circ \psi^{-1}$. Moreover $$\sigma(Ef,Eg) = \sigma^{\psi}(\psi_Ef,\psi_Eg),$$ where $\sigma^{\psi}$ is a symplectic form on ${{\mathcal R}}^{\psi}$ which is defined in an analogous fashion to $\sigma$. Then the prescription $$\label{3.5} \tilde{\alpha}_{\psi}(W(f)) = W^{\psi}(\psi_f), \quad f \in {{\mathcal R}},$$ for the generators $\{W^{\psi} \}$ of the CCR algebra over $({{\mathcal R}}^{\psi}, \sigma^{\psi})$, leads to $^$-isomorphisms between the corresponding algebras. To summarize, the CCR algebras described above, yield a version of Theorem \[localnet\] adapted to solutions of the Klein-Gordon equation. The theorem as stated below is due to Brunetti, Fredenhagen and Verch (see [@BFV Theorem 2.2]). To be faithful to their comprehension and formulation of this result, we first define the category $\textbf{\textsf{Loc}}$ $\textbf{\textsf{Loc}}$ : The class of objects $\mathrm{obj}(\textbf{\textsf{Loc}})$ is formed by all (smooth) $d$-dimensional ($d\ge 2$ is held fixed), globally hyperbolic Lorentzian spacetimes $M$ which are oriented and time-oriented. Given any two such objects $M_1$ and $M_2$, the morphisms $\psi\in\mbox{\rm hom}_{\textbf{\textsf{Loc}}}(M_1,M_2)$ are taken to be the isometric embeddings $\psi: M_1 \to M_2$ of $M_1$ into $M_2$ but with the following constraints; - if $\gamma : [a,b]\to M_2$ is any causal curve and $\gamma(a),\gamma(b)\in\psi(M_1)$ then the whole curve must be in the image $\psi(M_1)$, i.e., $\gamma(t)\in\psi(M_1)$ for all $t\in ]a,b[$; - any morphism preserves orientation and time-orientation of the embedded spacetime. Composition is composition of maps, the unit element in $\mbox{\rm hom}_{\textbf{\textsf{Loc}}}(M,M)$ is given by the identical embedding ${\rm id}_M : M\mapsto M$ for any $M\in\mathrm{obj}(\textbf{\textsf{Loc}})$. \[ccr\] For each $M \in \mathrm{obj}(\textbf{\textsf{Loc}})$ define the $C^$-algebra ${{\mathcal A}}(M)$ as the CCR-algebra $\mathfrak{W}({\mathcal R}(M),\sigma_{M})$ of the Klein-Gordon equation $$\label{KGeqn} (\nabla^a\nabla_a + m^2 + \xi R)\varphi = 0$$ (with $m,\xi$ fixed for all $M$ and $\mathcal{R}$ the scalar curvature), and for each $\psi \in \mbox{\rm hom}_{\textbf{\textsf{Loc}}}(M,M')$ define the $C^$-algebraic endomorphism by $\alpha_{\psi} = \tilde{\alpha}_{\iota_{\psi}} \circ \tilde{\alpha}_{\psi}: {{\mathcal A}}(M) \to {{\mathcal A}}(M')$ where $\tilde{\alpha}_{\psi}$ and $\tilde{\alpha}_{\iota_{\psi}}$ are respectively given by $$\label{Bog1} \tilde{\alpha}_{\psi}(W(\varphi)) = W^{\psi}(\psi_(\varphi))\,, \quad \varphi \in {\mathcal R}\,$$ and $$\label{Bog2} \tilde{\alpha}_{\iota_{\psi}}(W^{\psi}(\phi)) = W'(T^{\psi}\phi)\,, \quad \phi \in {\mathcal R}^{\psi}\, .$$ (Here $W^{\psi}(\,.\,)$ are as before, $W(.)$ the generators of the Weyl-generators of $\mathfrak{W}({\mathcal R},\sigma)$, and $W'(\,.\,)$ the Weyl-generators of $\mathfrak{W}({\mathcal R}',\sigma')$, while $T^\psi$ is the corresponding symplectic map from $({\mathcal R}^\psi,\sigma^\psi)$ into $({\mathcal R}',\sigma')$.) Then the corresponding local algebra fulfills all the criteria of the preceding theorem including causality and the time-slice axiom.* Realising ${{\mathcal A}}(\mathbb{M})$ as a “tangent algebra” of ${{\mathcal A}}(M)$ ------------------------------------------------------------------------------------ Our first task is to show that the picture described above can be further refined to allow for tangential phenomena. In particular we introduce the concept of tangentially conditioned local algebras. \[tancon\] We say that a local algebra ${{\mathcal A}}(M)$ is tangentially conditioned if ${{\mathcal A}}(M)$ behaves well with respect to the atlas on $M$ in the following sense: Given a quadruple $(p, \iota_p, {{\mathcal O}}_p, \tilde{{{\mathcal O}}}_0)$ from this atlas, where ${{\mathcal O}}_p$ is a neighbourhood of some $p\in M$, $\tilde{{{\mathcal O}}}_0$ a corresponding diffeomorphic neighbourhood of $0\in \mathbb{M}$, and $\iota_p$ the diffeomorphism on $\tilde{{{\mathcal O}}}_0$ identifying these neighbourhoods, it must be true that the algebras ${{\mathcal A}}({{\mathcal O}}_p)$ and ${{\mathcal A}}(\tilde{{{\mathcal O}}}_0)$ are $$-isomorphic by means of a $$-isomorphism $\beta_p$, which implements the diffeomorphism in the sense that for any open subset ${{\mathcal O}}_1$ of ${{\mathcal O}}_p$, the restriction of $\beta_p$ to ${{\mathcal A}}({{\mathcal O}}_1)$ yields a $$-isomorphism from ${{\mathcal A}}({{\mathcal O}}_1)$ onto ${{\mathcal A}}(\iota_p({{\mathcal O}}_1))$. It is clear from the definition that tangentially conditioned algebras ${{\mathcal A}}(M)$, are those which at a local level “look like” the algebra ${{\mathcal A}}(\mathbb{M})$. The important fact to note, is that the CCR-algebras considered earlier in Theorem \[localnet\] provide concrete examples of tangentially conditioned algebras! Theorem \[ccr\] also provides concrete examples of tangentially conditioned algebras, on condition that the Klein-Gordon equation is locally solvable, i.e. there is an atlas such that for each chart $({{\mathcal O}}_p, \iota_p)$ there is a (local) fundamental solutions $E_{{{\mathcal O}}_p}$. In other words, ${{\mathcal O}}_p$ considered as a manifold should enjoy all manifold properties assumed in Theorem \[ccr\]. \[1.4\] The CCR algebras 1. ${{\mathcal A}}(M)=\mathfrak{W}((M),\sigma_{M})$ generated on the symplectic space $\left( \mathrm{span}_{{{\mathbb{C}}}}\left(C^{\infty}_0(M)\right), \sigma(f,g) = Im \int_M \overline{f} g dV_M \right)$; $\mathrm{span}_{{{\mathbb{C}}}}$ stands for the complex span of $C^{\infty}_0(M)$, 2. ${{\mathcal A}}_0(M)=\mathfrak{W}({\mathcal R}(M),\sigma^E_{M})$ generated on the symplectic space $\left( C^{\infty}_0(M), \sigma(f,g) = \int_M f Eg dV_M\right)$ for the locally solvable Klein-Gordon equation, are tangentially conditioned. To prove the first claim, let us consider a quadruple $(p, \iota_p, {{\mathcal O}}_p, \tilde{{{\mathcal O}}}_0)$ where, as in Definition \[tancon\], ${{\mathcal O}}_p$ is a neighborhood of some $p \in M$, $\tilde{{{\mathcal O}}}_0$ a corresponding diffeomorphic neighborhood of $0 \in \mathbb{M}$, and $\iota_p :{{\mathcal O}}_p \to \tilde{{{\mathcal O}}}_0$ the diffeomorphism on $\tilde{{{\mathcal O}}}_0$ identifying these neighborhoods. The subspace formed by $\{f\in C^{\infty}_0(M); \mathrm{supp} f \subseteq {{\mathcal O}}_p \}$ ($\{f\in C^{\infty}_0(\mathbb{M}); \mathrm{supp} f \subseteq \tilde{{{\mathcal O}}}_0 \}$) will be denoted by ${{\mathcal H}}_p$ (by ${{\mathcal H}}_0$ respectively). Define the map $T:{{\mathcal H}}_p \to {{\mathcal H}}_0$ by $$(Tf)(x) = f(\iota_p^{-1} x), \quad x \in \tilde{{{\mathcal O}}}_0.$$ It is easy to see that $T$ is a linear bijection from ${{\mathcal H}}_p$ onto ${{\mathcal H}}_0$. We will show that $T$ is an isometry. To this end we note that $$\begin{aligned} \langle Tf,Tg\rangle_{{{\mathcal H}}_0} &=& \int_{\tilde{{{\mathcal O}}}_0} \overline{(Tf)}(x)(Tg)(x)dV_{\mathbb{M}}(x)\\ &=& \int_{\tilde{{{\mathcal O}}}_0}\overline{f}\circ \iota_p^{-1}(x) g\circ\iota_p^{-1}(x) dV_{\mathbb{M}}(x)\\ &=& \int_{\tilde{{{\mathcal O}}}_0}\overline{f}\circ \iota_p^{-1}(x) g\circ\iota_p^{-1}(x) dV_{\mathbb{M}}\left(\iota_p \circ \iota_p^{-1} (x)\right)\\ &=& \int_{\tilde{{{\mathcal O}}}_0}\overline{f}\circ \iota_p^{-1}(x) g\circ\iota_p^{-1}(x) (\iota_p)_dV_{\mathbb{M}}(\iota_p^{-1} (x))\\ &=& \int_{{{\mathcal O}}_p} \overline{f} g dV_{{M}} = \langle f,g\rangle_{{{\mathcal H}}_p},\end{aligned}$$ where $(\iota_p)_$ is the pull-back of $\iota_p$. But, then $T$ preserves the symplectic form $\sigma$. Hence, by [@BR Theorem 5.2.8], the prescription $$\alpha_T(\mathfrak{W}(f)) = \mathfrak{W}(Tf), \quad f \in C^{\infty}_0(M)$$ yields a \-isomorphism between from ${{\mathcal A}}({{\mathcal O}}_p)$ onto ${{\mathcal A}}(\tilde{{{\mathcal O}}}_0)$. The second claim follows by arguments given prior to Theorem \[ccr\]. Tangentially conditioned algebras and local flows along contours ---------------------------------------------------------------- We now turn to the question of dynamics on algebras ${{\mathcal A}}(M)$. When studying dynamics, it is important to identify the appropriate mode of continuity with which to describe such a dynamical flow. We are particularly interested in the appropriate mode of continuity that may be assigned to the translation automorphisms in the representation of the Poincaré group. Property 1 of the Gärding-Wightman postulates for field operators (see [@RS2 §IX.8]), as well as the behavior of local algebras which fulfil the spectrum condition (see [@Sak p. 33]), both suggest that it is a physically reasonable assumption to make, that these translation automorphisms are implemented by a strongly continuous unitary group acting on the underlying Hilbert space. That translates to SOT continuity of the translation automorphisms. This mode of continuity is however more suited to a von Neumann algebraic rather than a $C^$-algebraic framework. Hence in the remainder of the paper we will restrict ourselves to von Neumann local algebras, for which the group of translation automorphisms on ${{\mathfrak M}}(\mathbb{M})$ is SOT continuous. The only comment we will make about $C^$-framework, is that each of the subsequent results will under appropriate restrictions admit of a $C^$-version. With the framework in which we will work now clear, that leads us to the question of how one may realise a dynamical flow on the manifold $M$, at the algebra level. Part (b) of Theorem \[localnet\] assures us that in the case where one is fortunate enough to have a globally defined group $G$ of isometric diffeomorphisms on $M$ preserving orientation and time-orientation, the dynamics described by that group canonically lifts to the algebra setting. But what is equally clear from this theorem, is that not all algebras ${{\mathfrak M}}(M)$ have this property. If however one is content to settle for fairly strong locality as far as dynamics is concerned, the situation improves. Specifically for any tangentially conditioned local algebra ${{\mathfrak M}}(M)$, the “local” dynamics on $M$ does indeed lift to the algebra level. We point out that this behaviour is in line with the classical setting. See the discussion on page 35 of [@Thi]. In order to be able to deliver on our promise, some background is necessary. For general $d$-dimensional $C^\infty$-manifolds $M$ the $C^\infty$ derivations of $C^\infty(M)$ correspond to local flows on $M$. (To see this combine the comment preceding Theorem 2.2.24 of [@Thi] with [@Thi Remark 2.3.11(1)].) It is moreover known that on any chart $U$ of $M$, these derivations are up to a diffeomorphism of the form $\sum_{i=1}^df_i\frac{\partial}{\partial x_i}$, where for each $i$ we have that $f_i\in C^\infty(U)$ [@Thi 2.2.27(8) & 2..4.3(1)]. If therefore in the context of ${{\mathfrak M}}(\mathbb{M})$ we are able to identify the appropriate analogues of $\frac{\partial}{\partial x_i}$ and $C^\infty(U)$, we will at a formal level be able to give a “chart-wise” description of the quantum smooth local flows associated with a tangentially conditioned algebra ${{\mathfrak M}}(M)$. However the manifolds in view here are Lorentzian. We therefore briefly pause to describe how the above idea may be refined to this context. By [@BaF Chapter 2, Theorem 1] any (smooth) $d$-dimensional, connected time-oriented globally hyperbolic Lorentzian manifold $M$, may be written in the form $M=\mathbb{R}\times S$ where each ${t}\times S$ is a Cauchy hypersurface. (Here $\mathbb{R}$ models the time variable.) A careful consideration of part (3) of this result, shows that $S$ is in its own right a $(d-1)$–dimensional Riemannian Manifold. So at a local level $S$ “looks like” $\mathbb{R}^{d-1}$. Using this fact, we may select our charts for $M$ in such a way that the local diffeomorphisms which compare $\mathbb{M}$ to $M$, maps points of the form $(t, x_1, x_2,\dots, x_{d-1})$ in say $\tilde{{{\mathcal O}}}_0\subset \mathbb{M}$, onto points $(r(t),s)$ in ${{{\mathcal O}}}_p\subset M$, where $(r(t),s)\in \mathbb{R}\times S$, with $t$ going to $r(t)$ and $(x_1, x_2,\dots, x_{d-1})$ to $s$. Suppose that this is the case and let ${{\mathcal O}}_p$ be a neighbourhood of some $p\in M$ which is in the above sense diffeomorphic to a neighbourhood $\tilde{{{\mathcal O}}}$ of $0\in \mathbb{M}$ by means of some diffeomorphism $\iota_p$ of the above type. Based on the above discussion and the assumptions made therein, on the chart ${{\mathcal O}}_p$ a continuous local dynamical flow along some contour on $M$ passing through $p=(t_p,s_p)$, may in principle be regarded as the image under $\iota_p$ of a continuous local dynamical flow along a contour flowing through $0\in \mathbb{M}$, where the dynamical flow on $\mathbb{M}$ corresponds to a set of points $(t, x_1(t), x_2(t),\dots, x_{d-1}(t))\in \tilde{{{\mathcal O}}}_0$ which varies continuously as $t$ varies over the interval $(-\varepsilon, +\varepsilon)$, with $(0, x_1(0), x_2(0),\dots, x_{d-1}(0))=0$, and with $\iota_p(0)=p$. For the sake of simplicity let us write $g_t$ for $(t, x_1(t), x_2(t),\dots, x_{d-1}(t))$ and $\beta_p$ for the \-isomorphism from ${{\mathfrak M}}({{\mathcal O}}_p)$ to ${{\mathfrak M}}(\tilde{{{\mathcal O}}}_0)$. In the context of ${{\mathfrak M}}(\mathbb{M})$, this local dynamics may then formally be lifted to the algebra level by using the “translation automorphisms” in the representation of the Poincaré group on ${{\mathfrak M}}(\mathbb{M})$. Specifically if on $\mathbb{M}$ the contour is described by the set of points $g_t$ (indexed by the time variable), we may pass to the set $\alpha_{g_t}$ where each $\alpha_{g_t}$ satisfies $\alpha_{g_t}({{\mathfrak M}}(\tilde{{{\mathcal O}}}))= {{\mathfrak M}}(g_t+\tilde{{{\mathcal O}}})$ ($\tilde{{{\mathcal O}}}$ an open subset of $\mathbb{M}$). The natural domain of the “restriction” of the action of $\alpha_{g_t}$ to the ${{\mathfrak M}}(\tilde{{{\mathcal O}}}_0)$ context, is then $[{{\mathfrak M}}(\tilde{{{\mathcal O}}}_0)\cap\alpha_{g_t}^{-1}({{\mathfrak M}}(\tilde{{{\mathcal O}}}_0))]$. For any $g_t$ this natural domain will include all subalgebras ${{\mathfrak M}}(\tilde{{{\mathcal O}}}_t)$ of ${{\mathfrak M}}(\tilde{{{\mathcal O}}}_0)$, for which $\tilde{{{\mathcal O}}}_t$ is a subset of $\tilde{{{\mathcal O}}}_0$ small enough to ensure that $g_t+\tilde{{{\mathcal O}}}_t\subset \tilde{{{\mathcal O}}}_0$. As $g_t$ gets “closer” to 0, we expect the size of the sets $\tilde{{{\mathcal O}}}_t$ we are able to select, to increase. We pause to explain how one can make these ideas exact. By passing to a subset if necessary, we may assume that the set of $\tilde{{{\mathcal O}}}_0$ is an open double cone $K=\tilde{{{\mathcal O}}}_0$ centred at 0. It is then an exercise to see that $K=\cup_{n=1}^\infty\frac{n}{n+1}K$. We may further find a decreasing sequence $\varepsilon_n>0$ such that $\frac{n}{n+1}K +g \subset K$ whenever $g=(t,x_1, \dots, x_{d-1})$ is an element of $\mathbb{M}$ for which $\\|g\\|_2^2=\|t\|^2 +\sum_{k=1}^{d-1}\|x_k\|^2<\varepsilon_n$. As far as our local dynamical flow along the given contour is concerned, for each $\varepsilon_n>0$ we may by assumption find some $\delta_n>0$ such that $\\|g(t)\\|_2^2=\|t\|^2 +\sum_{k=1}^{d-1}\|x_k(t)\|^2<\varepsilon_n$ whenever $\|t\|< \delta_n$. At the operator level, each $\alpha_{g_t}$ for which $\|t\|< \delta_n$, will then yield a well defined operator from ${{\mathfrak M}}(\frac{n}{n+1}K)$ into ${{\mathfrak M}}(\frac{n}{n+1}K+g(t))\subset {{\mathfrak M}}(K)$. So the subalgebra $\cup_{n=1}^\infty{{\mathfrak M}}(\frac{n}{n+1}K)$ of ${{\mathfrak M}}(K)$ then represents a space of observables inside ${{\mathfrak M}}(K)$, for which the restriction of the operators $\alpha_{g(t)}$ to the context of ${{\mathfrak M}}(K)$ yield a well defined dynamics for short times along this contour (where the “shortness” of the time depends on the specific observable). If now the algebra ${{\mathfrak M}}$ was additive in the sense of [@araki Definition 4.13], we would have that $[\cup_{n=1}^\infty{{\mathfrak M}}(\frac{n}{n+1}K)]''={{\mathfrak M}}(K)$. That is for additive systems, the subalgebra of ${{\mathfrak M}}(K)$ for which we obtain dynamics for short times, is weak\* dense in ${{\mathfrak M}}(K)$. We summarise the conclusions of the above discussion in the following theorem: Let ${{\mathfrak M}}(M)$ be a local algebra tangentially conditioned to ${{\mathfrak M}}(\mathbb{M})$. If ${{\mathfrak M}}(\mathbb{M})$ is additive in the sense of [@araki Definition 4.13], then for any smooth contour $C$ through $p\in M$, the dynamics along this contour will for a small enough neighbourhood ${{\mathcal O}}_p\subset M$ of $p$, lifts to dynamics for short times on a weak\* dense subalgebra of ${{\mathfrak M}}({{\mathcal O}}_p)$. The dynamics at the algebra level is determined by the local action of the translation automorphisms on ${{\mathfrak M}}(\mathbb{M})$. (For these same ideas to work in the $C^$-algebra context, we need our local algebra to satisfy what we might call “strong” additivity in the sense that for any open double cone $K$ and any collection of open subsets $\{{{\mathcal O}}_\lambda\}$ of $K$ covering $K$, we will need $\cup_{n=1}^\infty{{\mathcal A}}({{\mathcal O}}_\lambda)$ to be norm-dense in ${{\mathcal A}}(K)$.) The space of generators of local flows -------------------------------------- Each locus of points of either the form $(t, 0, 0, \dots)$ or the form $(0, 0, \dots, 0, x_k, 0, \dots, 0)$ is a copy of $\mathbb{R}$. We may denote these loci by $\mathbb{R}_t$ and $\mathbb{R}_k$ ($2\leq k\leq d$) respectively. The groups $\alpha_x$ corresponding to translation by $x$, where $x$ belongs to either $\mathbb{R}_t$ or $\mathbb{R}_k$, are then one-parameter groups on ${{\mathfrak M}}(\mathbb{M})$. Hence the derivatives at 0, namely $\delta_t$ and $\delta_k$, are densely defined closed \-derivations. For the sake of simplicity, we will in the discussion hereafter write $\delta_0$ for $\delta_t$. Our first result in this subsection, shows that these derivations are the appropriate noncommutative analogues of the partial differential operators $\frac{\partial}{\partial x_i}$. This result also shows that in a very real sense, the space $\mathrm{span}\{\delta_k: 0\leq k\leq (d-1)\}$ acts as a space of infinitesimal generators of the action of the translation automorphisms on ${{\mathfrak M}}(\mathbb{M})$. \[dspace\] Let $C$ be a smooth contour through $0\in \mathbb{M}$ parametrised by say $x(t)$ where $-1\leq t\leq 1$, and $x(0)=0$. Let ${{\mathcal O}}$ be a neighbourhood of $0\in \mathbb{M}$. Then for any $f\in{{\mathfrak M}}({{\mathcal O}})\cap[\cap_{k=0}^{(d-1)}\mathrm{dom}(\delta_k)]$, the derivative at 0 of the set $\alpha_{x(t)}(f)$ ($t\in [-1,1]$) exists in the weak\* topology on ${{\mathfrak M}}(\mathbb{M})$, and corresponds to $\sum_{k=0}a_k\delta_k(f)$ where $x'(0)=(a_0, a_1, \dots, a_{(d-1)})$. Let $f$ be as in the hypothesis. Since each $\delta_k$ is adjoint-preserving, we may clearly assume that $f=f^$. As in the previous proof we will write $t\langle k \rangle$ for the vector with $t\in \mathbb{R}$ in the $k$-th coordinate and 0’s elsewhere. The fact that the translation automorphisms are implemented by a strongly continuous unitary group acting on the underlying Hilbert space, ensures that for each $k$, we have that $\frac{1}{t}[\alpha_{t\langle k\rangle}(f)-f]$ converges strongly to $\delta_k(f)$ as $t\to 0$. Let $x(t)$ be of the form $x(t)=(x_0(t), x_1(t), \dots, x_{(d-1)}(t))$. We denote the vector $(0,\dots, 0, x_k(t), 0, \dots, 0)$ by $\hat{x}_k(t)$. We first prove that for any $\xi$ in the underlying Hilbert space and any $0\leq k\leq (d-1)$, we have that $\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)\xi$ converges in norm to $x'_k(0)\delta_k(f)\xi=a_k\delta_k(f)\xi$. If $x'_k(0)=a_k\neq 0$, there must exist some $\varepsilon>0$ such that $\frac{x_k(t)}{t}\neq 0$ for every $0<\|t\|<\varepsilon$. Since $x_k(t)\to 0$ as $t\to 0$, the claim will in this case follow by rewriting $\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)\xi$ as $\frac{x_k(t)}{t}[\frac{1}{x_k(t)}(\alpha_{\hat{x}(t)}(f)-f)\xi]$ for all $0<\|t\|<\varepsilon$, and then letting $t\to 0$. If $a_k=0$, then for the claim to be true, we must have that $\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)\xi \to 0$ as $t \to 0$. Suppose this is not the case. In that case there must exist some $\varepsilon > 0$ and a sequence $\{t_n\}$ tending 0 such that $\\|\frac{1}{t_n}(\alpha_{\hat{x}_k(t_n)}(f)-f)\xi\\|\geq \varepsilon$ for all $n$. Since $(\alpha_{0}(f)-f)\xi=0$, we must then also have that $x_k(t_n)\neq 0$ for all $n$. But then we may write $\frac{1}{t_n}(\alpha_{\hat{x}_k(t_n)}(f)-f)\xi$ as $\frac{x_k(t_n)}{t_n}[\frac{1}{x_k(t_n)}(\alpha_{\hat{x}(t_n)}(f)-f)\xi]$. But as $n\to \infty$, this expression must converge in norm to $a_k\delta_k(f)\xi=0$, which is a clear contradiction. Hence our assumption that $\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)\xi$ does not converge to 0 as $t\to 0$, must be false. We claim that for each $k$, the terms $\Pi_{i=k+1}^{(d-1)}\alpha_{\hat{x}_i(t)}([\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)]-a_k\delta_k(f))$ converge to 0 in the weak\ topology as $t\to 0$. Suppose that for some $k$ this is not the case. Then there must exist a normal state $\omega$ of ${{\mathfrak M}}(\mathbb{M})$ such that $\omega(\Pi_{i=k+1}^{(d-1)}\alpha_{\hat{x}_i(t)}([\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)]-a_k\delta_k(f)))\not\to 0$, or equivalently there exists a normal state $\omega$ and a sequence $\{t_n\}$ tending to 0 such that for some $\widetilde{\varepsilon}>0$ we have that $$\|\omega(\Pi_{i=k+1}^{(d-1)}\alpha_{\hat{x}_i(t_n)}([\frac{1}{t_n}(\alpha_{\hat{x}_k(t_n)}(f)-f)]-a_k\delta_k(f)))\|\geq \widetilde{\varepsilon}$$for each $n\in \mathbb{N}$. Note that with $w(k,t)$ denoting the vector $(0,\dots, 0, x_{k+1}(t),\dots, x_{(d-1)}(t)$, we have that $\Pi_{i=k+1}^{(d-1)}\alpha_{\hat{x}_i(t)}=\alpha_{w(k,t)}$ for each $k$. For the sake of simplicity we will in the ensuing argument make these substitutions. Since for each $\xi$ in the underlying Hilbert space we have that $[\frac{1}{t_n}(\alpha_{\hat{x}_k(t_n)}(f)-f)-a_k\delta_k(f)]\xi\to 0$ as $n\to \infty$, we know from the Banach-Steinhaus theorem that $\sup_n\\|\frac{1}{t_n}(\alpha_{\hat{x}_k(t_n)}(f)-f)-a_k\delta_k(f)\\|<\infty$. But if that is the case then by [@BR Proposition 2.4.1], $[\frac{1}{t_n}(\alpha_{\hat{x}_k(t_n)}(f)-f)-a_k\delta_k(f)]$ must in fact converge to 0 in the $\sigma$-strong topology, and not just strongly. Since by assumption $f=f^$, the convergence is actually in the $\sigma$-strong\ topology. Now recall that $\alpha_{w(k,t_n)}$ converges strongly to the identity map on ${{\mathfrak M}}({{\mathcal O}})$ as $n\to \infty$. By [@Bla III.3.2.2], this convergence also takes place in the weak\* ($\sigma$-weak) topology. So for any normal state $\nu$ on ${{\mathfrak M}}(\mathbb{M})$, each $\nu\circ \alpha_{w(k,t_n)}$ is again a normal state with $\lim_{n\to\infty}\nu(\alpha_{w(k,t_n)}(a))=\nu(a)$ for every $a\in {{\mathfrak M}}(\mathbb{M})$. If now we apply [@Tak Proposition III.5.5] we will have that $$\lim_{n\to\infty}\omega(\alpha_{w(k,t_m)}([\frac{1}{t_n}(\alpha_{\hat{x}_k(t_n)}(f)-f)]-a_k\delta_k(f)))=0$$uniformly in $m\in\mathbb{N}$. But this contradicts our earlier assumption that $$\|\omega(\alpha_{w(k,t_n)}([\frac{1}{t_n}(\alpha_{\hat{x}_k(t_n)}(f)-f)]-a_k\delta_k(f)))\|\geq\widetilde{\varepsilon}$$for each $n\in\mathbb{N}$. Hence the assumption that for some $k$ the net $\alpha_{w(k,t)}([\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)]-a_k\delta_k(f))$ is not weak\* convergent to 0, must be false. Therefore each $\alpha_{w(k,t)}([\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)]-a_k\delta_k(f))$ is weak\* convergent to 0 as $t\to 0$. Moreover by [@Bla III.3.2.2], we also have that each $a_k\alpha_{w(k,t)}(\delta_k(f))$ is weak\* convergent to $a_k\delta_k(f)$ as $t\to 0$. We are now ready to prove the primary claim of the theorem. To do this we simply note that $$\frac{1}{t}(\alpha_{x(t)}(f)-f)-\sum_{k=0}^{(d-1)}a_k\delta_k(f)$$may be rewritten as $$\sum_{k=0}^{(d-1)} \Pi_{i=k+1}^{(d-1)}\alpha_{\hat{x}_i(t)}[\frac{1}{t}(\alpha_{\hat{x}_k(t)}(f)-f)-a_k\delta_k(f)] +\sum_{k=0}^{(d-1)}a_k[\Pi_{i=k+1}^{(d-1)}\alpha_{\hat{x}_i(t)}(\delta_k(f))-\delta_k(f)]$$and then apply the foregoing conclusions to see that $\frac{1}{t}(\alpha_{x(t)}(f)-f)$ converges to $\sum_{k=0}a_k\delta_k(f)$ in the weak\* topology as $t\to 0$. Having identified the objects that serve as the quantum analogues of the partial differential operators $\frac{\partial}{\partial x_i}$, it is natural to then use these object to identify a subalgebra of ${{\mathfrak M}}(\mathbb{M})$, which is the quantum analogue of $C^\infty(M)$. This is done in the next theorem, which also shows that this subalgebra of “smooth” elements of ${{\mathfrak M}}(\mathbb{M})$, is in fact a weak\* dense subalgebra. For any local algebra ${{\mathfrak M}}(\mathbb{M})$ satisfying the SOT-continuity assumption regarding the translation automorphisms, the algebra ${{\mathfrak M}}^\infty(\mathbb{M})=\{a\in {{\mathfrak M}}(\mathbb{M}): a\in \mathrm{dom}(\delta_{\pi(1)}\dots\delta_{\pi(k)}),\ k\in \mathbb{N},\ 0\leq \pi(i)\leq (d-1)\}$ is weak\* dense in ${{\mathfrak M}}(\mathbb{M})$. We pause to point out that ${{\mathfrak M}}^\infty(\mathbb{M})$ can in a very natural way be described as a Fréchet space (see the discussion preceding [@Br Theorem 2.2.3]). It is also an exercise to see that each element of the space $\mathrm{span}\{\delta_k: 0\leq k\leq (d-1)\}$ will map ${{\mathfrak M}}^\infty(\mathbb{M})$ back into itself. The proof is based on a modification of [@pazy Theorem 2.7]. Hence at some points we will not give full details, but instead refer the reader to the corresponding argument in [@pazy]. Let $\mathfrak{C}$ be the space of all complex-valued $C^\infty$ functions on the open cell $(0,\infty)^d$ which are compactly supported. Given any $x\in \mathbb{M}$, we will throughout write $\alpha_x$ for the automorphism corresponding to translation by $x$ in the sense that $\alpha_x({{\mathfrak M}}({{\mathcal O}}))={{\mathfrak M}}(x+{{\mathcal O}})$. We will further write $t\langle k \rangle$ for the vector with $t\in \mathbb{R}$ in the $k$-th coordinate and 0’s elsewhere, and $\varphi_k$ for the partial derivative of $\varphi$ with respect to the $k$-th coordinate. Given any $a\in {{\mathfrak M}}(\mathbb{M})$ and $\varphi\in \mathfrak{C}$, then with $\mathbf{s}$ denoting $(s_0,\dots,s_{(d-1)})$, we set $$a(\varphi)=\int_0^\infty\dots\int_0^\infty\varphi(\mathbf{s})\alpha_{\mathbf{s}}(a)\,ds_0\,ds_1\dots ds_{(d-1)}.$$In view of the assumption regarding the SOT continuity of the representation $x\to \alpha_x$, this integral converges in the SOT topology, which in turn ensures that $a(\varphi)\in {{\mathfrak M}}(\mathbb{M})$. Assuming that $h>0$ it is then an exercise to see that for example $$\begin{aligned} &&\frac{1}{h}(\alpha_{h\langle 0\rangle}(a(\varphi))-a(\varphi))\\ &=& \int_0^\infty\dots\int_0^\infty\tfrac{1}{h}[\varphi(s_0-h,s_1\dots,s_{(d-1)})-\varphi(s_0,s_1\dots,s_{(d-1)})]\alpha_{\mathbf{s}}(a)\,ds_0\,ds_1\dots ds_{(d-1)}\\ &\to& (-1)\int_0^\infty\dots\int_0^\infty\varphi_{s_0}(\mathbf{s})\alpha_{\mathbf{s}}(a)\,ds_0\,ds_1\dots ds_{(d-1)}.\end{aligned}$$ (Again convergence is in the SOT topology.) Hence $a(\varphi)\in \mathrm{dom}(\delta_0)$ with $\delta_0(a(\varphi))= (-1)a(\varphi_{s_0})$. Here we chose the coordinate $k=0$ for no other reason than simplicity of notation. Hence for any $k$ we have that $a(\varphi)\in \mathrm{dom}(\delta_k)$ with $\delta_k(a(\varphi))=(-1)a(\varphi_k)$. This in particular ensures that the space $Y=\mathrm{span}\{a(\varphi): a\in {{\mathfrak M}}(\mathbb{M}), \varphi\in \mathfrak{C}\}$ is contained in ${{\mathfrak M}}^\infty(\mathbb{M})$. To conclude the proof we therefore need to show that $Y$ is weak\* dense in ${{\mathfrak M}}(\mathbb{M})$. Suppose that this is not the case. Then by the Hahn-Banach theorem there must exists some non-zero element $\rho$ of $({{\mathfrak M}}(\mathbb{M}))_$, which vanishes on $Y$. But each such $\rho$ is SOT continuous [@Bla Theorem III.2.1.4]. Hence for each $a\in {{\mathfrak M}}(\mathbb{M})$ and each $\varphi\in\mathfrak{C}$ we have that $$0=\rho(a(\varphi))= \int_0^\infty\dots\int_0^\infty\varphi({\mathbf{s}})\rho(\alpha_{\mathbf{s}}(a))\,ds_0\,ds_1\dots ds_{(d-1)}.$$This in turn ensures that for each $a\in {{\mathfrak M}}(\mathbb{M})$, the SOT-continuous map $\mathbf{s}\to \rho(\alpha_{\mathbf{s}}(a))$ is identically 0 on the open cell $(0,\infty)^d$. Letting $\mathbf{s}\to (0, \dots,0)$, it follows that $\rho(a)=0$ for all $a\in {{\mathfrak M}}(\mathbb{M})$. But this contradicts the assumption that $\rho\neq 0$. We must therefore have that $Y$ is weak\-dense in ${{\mathfrak M}}(\mathbb{M})$. This concludes the proof. Having thus identified the appropriate noncommutative analogues of the partial differential operators $\frac{\partial}{\partial x_i}$ and of $C^\infty(\mathbb{M})$, we are now in a position to identify the objects that may be regarded as quantum “local flows” on $\mathbb{M}$. We have already noted that for a general manifold $M$ $C^\infty$ local flows are in a 1-1 correspondence with the derivations on $C^\infty(M)$. (Combine [@Thi Remark 2.3.11(1)] with the comment preceding [@Thi Theorem 2.2.24].) Taking the insightful work of Bratteli [@Br] as a point of reference for further development, one may on this basis propose the space of all weak\-closable weak\-densely defined derivations on ${{\mathfrak M}}(\mathbb{M})$ that map ${{\mathfrak M}}(\mathbb{M})$ back into itself, as the quantum analogue of local flows on $\mathbb{M}$. (See for example the introduction to [@Br Chapter 2].) We shall denote this space of derivations by $\Delta_{\mathbb{M}}$. Since the space ${{\mathfrak M}}^\infty(\mathbb{M})$ was constructed using a finite-dimensional space of derivations, results like [@Br Theorem 2.3.6] (due to Batty) suggest that the elements of $\Delta_{\mathbb{M}}$ are not far from being generators of groups of transformations on ${{\mathfrak M}}(\mathbb{M})$, which ties in well with the classical theory. The local correspondence of tangentially conditioned algebras to ${{\mathfrak M}}(\mathbb{M})$, allows one to then at least on chart-wise basis attempt to lift these ideas from ${{\mathfrak M}}(\mathbb{M})$ to ${{\mathfrak M}}(M)$. Quantum graded algebras for ${{\mathfrak M}}(\mathbb{M})$ --------------------------------------------------------- The discussion at the end of the previous subsection, shows that the space of derivations $\Delta_{\mathbb{M}}$ in some sense provides the technology for giving a chart-wise description of the “quantum local flows” of a given tangentially conditioned algebra ${{\mathfrak M}}(M)$. We have in addition seen that under mild restrictions, the space ${{\mathfrak M}}^\infty(\mathbb{M})$ is weak\* dense in ${{\mathfrak M}}(\mathbb{M})$. Hence a quantum graded algebra of differential forms for $\mathbb{M}$ constructed using these objects, should in principle be “chart-wise” relevant for tangentially conditioned algebras ${{\mathfrak M}}(M)$. Using the work of Michel du Bois-Violette as a template (see the excellent review in [@dje]), one may now construct a graded algebra of differential forms from the pair $({{\mathfrak M}}^\infty(\mathbb{M}), \Delta_{\mathbb{M}})$. The actual construction of such a quantum graded algebra, can be done exactly as in section 2.5 of [@dje], with the only difference being that we replace the pair $({{\mathfrak M}}, \mathrm{Der}_{{{\mathfrak M}}})$ used by Djemai, et al, by the pair $({{\mathfrak M}}^\infty(\mathbb{M}), \Delta_{\mathbb{M}})$. All other ingredients remain exactly the same. We pause to justify this replacement before going on to explain why this framework is sufficient for the construction to go through. Recall that by assumption the translation automorphisms are induced by a strongly continuous unitary group acting on the underlying Hilbert space. Using this fact, one is able to conclude that each of the derivations $\delta_k$ is of the form $\delta_k(a)= i[H_k,a]$. For the time variable $H_0$ is just the Hamiltonian, with $H_k$ being a momentum operator for $k= 1,\dots, (d-1)$. These operators are all necessarily unbounded, and hence so are each of the $\delta_k$’s. These are therefore clearly not defined on all of ${{\mathfrak M}}(\mathbb{M})$, and hence if we want a model incorporating the information encoded in the $\delta_k$’s, we cannot a priori insist on using ${{\mathfrak M}}(\mathbb{M})$ in the construction, as this will exclude these operators. However each $\delta_k$ is defined everywhere on a smooth part of the algebra, namely ${{\mathfrak M}}^\infty(\mathbb{M})$. The replacement of ${{\mathfrak M}}(\mathbb{M})$ with ${{\mathfrak M}}^\infty(\mathbb{M})$, therefore allows one to incorporate the $\delta_k$’s into the picture. In fact Djemai himself reveals an implicit concern for “smoothness”, when at the start of [@dje Subsection 2.5.2] he points out that the constructs described at that point may be applied to $C^\infty(M)$. For the readers who are concerned about the validity of the claim that the construction in [@dje §2.5] carries over to the pair $({{\mathfrak M}}^\infty(\mathbb{M}), \Delta_{\mathbb{M}})$, we hasten to point out that the construction in [@dje §2.5] is entirely algebraic, and that all we in principle need is a $$-algebra, and a space of derivations on that algebra which admits a left-module action of that algebra, and that we do have. We pause to further justify the claim that a $$-algebra rather than a $C^$-algebra will suffices. Note for example that although the author invokes the “topological” tensor product at the start of section 2.2 of [@dje], the algebraic tensor product will do just as well for this part of the construction. Note further that in the construction described in sections 2.1-2.5 of [@dje], there are three crucial ingredients. These are Property 1 on page 808, Property 2 on page 808, and Proposition 3 on page 809. Although full details are not given in the actual text of [@dje], it can be seen from [@bour Chapter III, §X] that Property 1 is a purely algebraic property. For the other two aspects Djemai cites [@Con] as a reference. In that paper Connes announces a 7 step programme (on p 264), with the bulk of [@Con] devoted to step II. This is the part required by the construction in [@dje]. But as can be seen from the declaration at the top of page 262 of [@Con], the content of [@Con Part II] is purely algebraic! Readers that have some concern that at some point Part II of [@Con] has a hidden reliance on the more topological Part I, will be reassured by the discussion on page 310, where Alain Connes describes the relationship between Parts I and II. The above discussion leads us to the following conclusion: The algebra ${{\mathfrak M}}(\mathbb{M})$ admits the construction of a smooth quantum graded algebra of differential forms which on a chart-wise basis models the action of smooth quantum local flows on a tangentially conditioned algebra ${{\mathfrak M}}(M)$. In closing we wish to point out that there is a very comprehensive theory of the derivational approach to noncommutative differential geometry. Our goal in this paper was to demonstrate how one aspect of this theory may be incorporated into the theory of local algebras. There is surely more that can be done in this regard, but that is not the concern of the present paper. Conclusions. ============ In the present paper we have continued our study on the new approach based on Orlicz spaces to the analysis of large systems, i.e. systems having an infinite number of degrees of freedom, see [@LM], [@LM2], [@ML]. Having shifted our focus to quantum field theory, we here show that this new strategy initially developed for quantum statistical mechanics, can in a very natural and elegant way be applied to quantum field theory, obviously with suitable modifications. It is important to note that the modifications necessary for the application of this strategy to quantum field theory, are drawn from the very basic ingredients of the principle of relativity; see Corollary \[1\]. Having thus well established the static setting of quantum fields, one should take into account that to examine physical laws and the mathematical equations describing these laws, one needs differential structures. To give an elementary example, of the usefulness of these structures, we note that even Maxwell equations fit naturally into differential geometry; specifically the calculus on manifolds. We have shown that the presented formalism is well suited to a du Bois-Violette approach to non-commutative differential geometry. But, as was hinted at the end of the previous section, the theory of quantum local flows on local algebras is incomplete. In other words, we conclude the paper with the acknowledgment that there remain many questions, and that the community of researchers struggling with this problem are still far from achieving a “theory of everything”. We do however hope that the present contribution will in some small way help to illuminate the next step in this grand quest. [99]{} J-P Antoine, A. Inoue, C. Trapani Partial $^$-Algebras and Their Operator Realizations Mathematics and Its Applications, vol. 553, Kluwer, Dordrecht, NL, 2002 H. Araki, Mathematical Theory of Quantum Fields, Oxford University Press F. Bagarello, Algebras of unbounded operators and physical applications: A survey. Rev. Math. Phys. 19 231 (2007) C Bär & K Fredenhagen (Editors), Quantum Field Theory for curved spacetimes :Concepts and mathematical foundations, Lecture Notes in Physics Vol 178, Springer-Verlag, Berlin-Heidelberg, 2009. G Bennet and R Sharpley, Interpolation of Operators, Academic Press, London, 1988. J. J. Bisognano, E. H. Wichmann, On the duality condition for Hermitean scalar fields, J. Math. Phys. 16, 985 (1975) J. J. Bisognano, E. H. Wichmann, On the duality condition for quantum fields, J. Math. Phys. 17, 303 (1976) B Blackadar, Operator Algebras, Springer, 2006. H. J. Borchers, Algebraic aspects of Wightman quantum field theory in International Symposium on Mathematical Methods in Theoretical Physics, ed. H. Araki, Lecture Notes in Mathematics, vol. 39, Springer, (1975) pp 283-292. H. J. Borchers, On revolutionizing quantum field theory with Tomita’s modular theory, J. Math. Phys. 41, no. 6, 3604–3673 (2000). H. J. Borchers, J. Yngvason, Positivity of Wightman functionals and the existence of local nets, Commun. Math. Phys. 127 607-615 (1990) N Bourbaki, Algèbre I, Hermann, Paris, 1970 O Bratteli, Derivations, Dissipations and Group Actions on $C^\ast$-algebras, Springer-Verlag, Berlin-Heidelberg-New York, 1986. O. Bratteli, D. Robinson, Operator algebras and Quantum Statistical Mechanics , Text and Monographs in Physics, Springer Verlag; vol. I, 1979; vol. II, 1981 R Brunetti, K Fredenhagen, Algebraic Approach to Quantum Field Theories. in: Encyclopedia of Mathematical Physics, Amsterdam: Elsevier, 2007. - ISBN: 9780125126601 R Brunetti, K Fredenhagen, R Verch, The generally covariant locality principle—a new paradigm for local quantum field theory, Comm. Math. Phys. 237 (2003) no. 1-2, 31-–68. D. Buchholz, On quantum fields that generate local algebras, J. Math. Phys. 1839-1846 (1990) D. Buchholz, R. Haag, The quest for understanding in relativistic quantum physics, J. Math. Phys. 41, 3674-3697 (2000). A Connes, Noncommutative differential geometry, Inst. Hautes Etudes Sci. Publ. Math., No. 62 (1985), 257–360. J. Dimock, Algebras of local observables on a manifold, Comm. Math. Phys. 77 (1980), 219–228. J. Dimock, Dirac quantum fields on a manifold, Trans. Amer. Math. Soc. 269 (1982) no. 1, 133–147. A E F Djemai, Introduction to Dubois-Violette’s noncommutative differential geometry, International Journal of Theoretical Physics 34 (1995) No. 6, 801–887. W. Driessler, S. J. Summers, E. H. Wichmann, On the connection between quantum fields and von Neumann algebras of local operators, Commun. Math. Phys. 105, 49-84 (1986) L. Gärding, A. S. Wightman, Representation of the anticommutation relations, Proc. Nat. Acad. Sci. 40, 617 (1954) L. Gärding, A. S. Wightman, Representation of the commutation relations, Proc. Nat. Acad. Sci. 40, 622 (1954) S Goldstein, Conditional expectation and stochastic integrals in non-commutative $L^p$ spaces’, Math Proc Camb Phil Soc 110(1991), 365–383 R. Haag, Local Quantum Physics. Fields, Particles, Algebras, Springer, 2nd edition, 1996 U. Haagerup, Normal weights on W\-Algebras, J. Funct. Anal.* 19 302-317 (1975) U. Haagerup, Operator valued weights in von Neumann algebras: I, J. Funct. Anal. 32 175-206 (1979) U. Haagerup, Operator valued weights in von Neumann algebras: II, J. Funct. Anal. 33 339-361 (1979) U. Haagerup, M. Junge, Q. Xu, A reduction method for noncommutative $L_p$-spaces and applications, TAMS, 362 2125-2165 (2010) RV Kadison and JR Ringrose, Fundamentals of the Theory of Operator Algebras: Vol 1, Academic Press, New York, 1983. H. Kosaki, Applications of complex interpolation method to a von Neumann algebra (Non-commutative $L^p$-spaces) J. Funct. Anal. 56 29-78 (1984) LE Labuschagne, A crossed product approach to Orlicz spaces, Proc LMS 107 (3) (2013), 965-1003. LE Labuschagne, WA Majewski, Maps on non-commutative Orlicz spaces, Illinois J. Math. [55]{}, 1053-1081, (2011) LE Labuschagne and WA Majewski, Quantum dynamics on Orlicz spaces, arXiv:1605.01210 \[math-ph\]. WA Majewski, LE Labuschagne, On applications of Orlicz spaces to Statistical Physics, Ann. H. Poincare., [15]{}, 1197-1221, (2014) W. A. Majewski, L. E. Labuschagne, Why are Orlicz spaces useful for Statistical Physics? in Noncommutative Analysis, Operator Theory and Applications; Eds. D. Alpay et al. Birkhauser-Basel, Series: Linear Operators and Linear Systems, vol 252, 271-283 (2016) W. A. Majewski, On quantum statistical mechanics; A study guide, Adv. Math. Phys - to appear. arXiv 1608.06766v2 \[math-ph\] E. Nelson, Notes on non-commutative integration, J. Funct. Anal. 15 (1974), 103 A Pazy, Semigroups of linear operators and applications to partial differential equations, Springer-Verlag, New York, 1983 G. Pedersen and M. Takesaki, The Radon-Nikodym theorem for von Neumann algebras, Acta. Math.130, 53-87, (1973) M. Reed, B. Simon, Methods of Modern Mathematical Physics. I. Functional Analysis Academic Press, New York and London, 1972 M Reed & B Simon, Fourier Analysis, Self-Adjointness : Methods of Modern Mathematical Physics, Vol. 2 (1st Ed), Academic Press, 1975. H. Reeh, S. Schlieder, Bemerkungen zur Unitäräquivalenz von Lorentzinvarianten Feldern, Nuovo Cim. 19 787-793 (1961) S. Sakai, Operator Algebras in Dynamical Systems, Cambridge University Press, Cambridge, 1991. K. Schm[ü]{}dgen, Unbounded Operator Algebras and Representation Theory Akademie-Verlag, Berlin, 1990 I. E. Segal, Postulates for General Quantum Mechanics, Ann. Math. 48 930-948 (1947) M Takesaki, Theory of Operator Algebras, Vol I, II, III, Springer, New York, 2003. M. Terp, [$L^p$ spaces associated with von Neumann algebras]{}. K[ø]{}benhavs Universitet, Mathematisk Institut, Rapport No 3a (1981). W. Thirring, [A Course in Mathematical Physics I: Classical Dynamical Systems (1st edition)]{}, Springer-Verlag, 1978. A. Wehrl, General properties of entropy, [Rev. Mod. Phys.]{}, [50]{}, 221-260, 1978 A. S. Wightman, Quelque problèmes mathématique de la théorie quantique relativiste. In: Lecture Notes, Faculté de Sciences. Univ. de Paris, 1957. J. Yngvason, The role of type III factors in Quantum Field Theory, Rep. Math. Phys. 55 135-147 (2005) [^1]: The contribution of L. E. Labuschagne is based on research partially supported by the National Research Foundation (IPRR Grant 96128). Any opinion, findings and conclusions or recommendations expressed in this material, are those of the author, and therefore the NRF do not accept any liability in regard thereto.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We study a particular graded ring structure on the set of all loopfree matroids on a fixed labeled ground set, which occurs naturally in tropical geometry. The product is given by matroid intersection and the additive structure is defined by assigning to each matroid the indicator vector of its chains of flats. We show that this ring is generated in corank one, more precisely that any matroid can be written as a linear combination of products of corank one matroids. Moreover, we prove that a basis for the graded part of rank $r$ matroids is given by the set of nested matroids and that the total number of these is a Eulerian number. Derksen’s $\mathcal{G}$-invariant then defines a ${\mathbb{Z}}$-linear map on this ring, which implies for example that the Tutte polynomial is linear on it as well. Finally we show that the ring is the cohomology ring of the toric variety of the permutohedron and thus fulfills Poincaré duality.' address: \| Technische Universität Berlin\ Institut für Mathematik, Sekretariat MA 6-2\ Straße des 17. Juni 136, 10623 Berlin author: - Simon Hampe bibliography: - 'bibliography.bib' title: The intersection ring of matroids --- Introduction {#section_introduction} ============ In matroid theory, one can associate various algebraic invariants to individual matroids, such as, for example, the Tutte polynomial. However, one can also try to define algebraic structures on sets of matroids. For example, Crapo and Schmitt study coalgebra and Hopf algebra structures on matroids [@sincidencehopf; @csprimitivematroidminor; @csuniquefactorization]. In [@abgwhomology], the authors define homology groups of matroids. Recently, Giansiracusa and Giansiracusa [@gggrassmannalgebra] defined the Grassmann algebra of valuated matroids as an idempotent analogue of the classical Grassmann algebra. Our approach is inspired by intersection theory in tropical geometry. Tropical geometry can be seen as a combinatorial or polyhedral version of algebraic geometry, though it has ramifications into (among others) optimization, number theory, biological statistics and economics. We recommend the book by Maclagan and Sturmfels [@MaclaganSturmfelsBook] as an introduction to the subject. Matroids have played an important role in tropical geometry ever since Sturmfels discovered that the tropicalization of a linear space only depends on the underlying matroid [@ssolving]. Speyer [@stropicallinear] generalized the concept of a tropical linear space to valuated matroids [@DressWenzel]. He also showed that there exists a product structure on valuated matroids by proving that the intersection product of two tropical linear spaces is a tropical linear space. It is this product structure that we wish to exploit, though we will only consider ordinary matroids, i.e. trivial valuations. Many geometric operations on tropical linear spaces have matroid-theoretic counterparts, which makes them very well-behaved and well-understood. Thus matroid theory has provided a very useful tool for tropical geometers (see for example [@smatroidintersection; @frdiagonalintersection]). Here we intend to go the opposite way: Use the geometric intuition behind tropical geometry to learn something about matroids. Indeed, the operations we will define to obtain the intersection ring of matroids might seem odd and unnatural to the matroid theorist. However, from the point of view of tropical geometry it is immediately clear that this ring is the right object to consider. It is the intersection ring of tropical linear spaces (with trivial valuation), where the operations are the obvious ones. As we will show, this ring exhibits a very rich structure that is tightly connected to the underlying matroids. In particular, it has the striking property that the $\mathcal{G}$-invariant induces a ${\mathbb{Z}}$-linear map on it. This invariant was introduced by Derksen in [@dsymmetricpolymatroids] as a generalization of various matroid invariants. In particular, all invariants which can be derived linearly from the ${\mathcal{G}}$-invariant automatically induce linear maps on the matroid intersection ring. This includes: The Tutte polynomial, all Tutte-Grothendieck invariants, the number of flats of fixed rank and the number of cyclic flats of fixed rank. Derksen and Fink showed in [@dfvaluative] that ${\mathcal{G}}$ is the universal invariant for valuations on matroid polytopes. We will see in Section \[subsection\_poincare\_duality\] that as a ${\mathbb{Z}}$-module the intersection ring of matroids is a quotient of the matroid polytope module they construct. The ring is naturally graded by corank and each graded piece is a free ${\mathbb{Z}}$-module whose dimension is a well-known combinatorial quantity: The Eulerian number $A_{r,n}$, which by definition counts permutations on $n$ elements with $r$ ascents. We show that it is also the number of nested matroids of rank $r+1$ on $n$ labeled elements. To our knowledge, this is a new result – only the number of isomorphism classes of nested matroids had been determined so far. The basic idea for defining the intersection ring of matroids is this (a formal definition will be made in Section \[section\_preliminaries\]): We identify each loopfree matroid of rank $r$ on $n$ labeled elements with its indicator vector of maximal chains of flats. We denote the ${\mathbb{Z}}$-module thus obtained by ${\mathbb{M}}_{r,n}$. We then define a product on ${\mathbb{M}}_n = \bigoplus_{r \geq 1} {\mathbb{M}}_{r,n}$ via matroid intersection (which is the dual operation of matroid union): For any two matroids $M,N$, we set $M \cdot N := M \wedge N$ if the latter is loopfree, and $0$ otherwise. The product is extended to arbitrary linear combinations by distributivity. Phrased like this it is not even clear that this is well-defined. However, tropically this is clear: A linear combination of matroids corresponds to a linear combination of tropical linear spaces, which is an actual geometric object, to which an intersection product can be applied. Well-definedness then follows from the distributivity of the intersection product. We now summarize our main results: - With the operations defined above, $({\mathbb{M}}_n,+,\cdot)$ is a commutative ${\mathbb{Z}}$-algebra with $1 = U_{n,n}$. It is graded by corank, i.e. $${\mathbb{M}}_{r,n} \cdot {\mathbb{M}}_{s,n} \subseteq {\mathbb{M}}_{r+s-n,n}$$ (where ${\mathbb{M}}_{k,n} = 0$ if $k \leq 0$). Furthermore, it is generated in corank one: every matroid can be written as a linear combination of products of corank one matroids. - A basis of the free ${\mathbb{Z}}$-module ${\mathbb{M}}_{r,n}$ is given by the set of all loopfree nested matroids of rank $r$. The number of these matroids is the Eulerian number $A_{r-1,n}$. - The $\mathcal{G}$-invariant induces a ${\mathbb{Z}}$-module homomorphism ${\mathbb{M}}_n \to {\mathbb{Z}}$ via $$M \mapsto {\mathcal{G}}(M) \textnormal{ for all matroids } M\enspace.$$ - ${\mathbb{M}}_n \cong A^(X(\textnormal{Perm}_n))$, the cohomology ring of the toric variety corresponding to the normal fan of the permutohedron of order $n$. In particular, it fulfills Poincaré duality. That is, for every $1 \leq r \leq n$, the intersection product induces a perfect pairing $${\mathbb{M}}_{r,n} \times {\mathbb{M}}_{n-r+1,n} \to {\mathbb{M}}_{1,n} \cong {\mathbb{Z}}\enspace.$$ - ${\mathbb{M}}_n$ is isomorphic to the subalgebra of McMullen’s polytope algebra generated by all matroid polytopes. As the theorem suggests, a central role in our arguments is played by nested matroids. These matroids have occurred under a variety of names, such as generalized Catalan matroids* [@bmlatticepathmatroids] or shifted matroids [@acatalanmatroid] and seem to have been first defined by Crapo [@csingleelement]. They are transversal matroids whose transversal presentation is a chain of sets. They are minor- and dual-closed, well-quasi-ordered and have an infinite list of excluded minors [@oprmatroidsdomains; @bmlatticeofcyclic]. Their relevance to our approach stems from the fact that they can equivalently be characterized by the fact that their lattice of cyclic flats is a chain. Cyclic flats are flats which are unions of circuits. They encode the full lattice of flats and nested matroids can thus be seen as the basic building blocks for constructing more complicated matroids. The last part of the theorem above emphasizes the relevance of the multiplicative structure. The Eulerian numbers are symmetric, thus implying that ${\mathbb{M}}_{r,n}$ is isomorphic to ${\mathbb{M}}_{n-r+1,n}$ as a free ${\mathbb{Z}}$-module. This is surprising at first, as the isomorphism clearly cannot be induced by any obvious matroid operation such as taking duals (in fact, dualizing is not even a well-defined map on ${\mathbb{M}}_n$, as it may produce loops). However, the corresponding tropical linear spaces do have complementary dimension and so immediately suggest Poincaré duality. The layout of the paper is as thus: In Section \[section\_preliminaries\], we mainly introduce the central notions from matroid theory we will need – in particular the notion of cyclic flats. A rigorous definition of the matroid intersection ring will be made. We also give a very brief summary of the most basic definitions from tropical geometry. In Section \[section\_linear\_combinations\] we will only be concerned with the additive structure on ${\mathbb{M}}_{r,n}$, the product will not yet play a role. We show that the nested matroids are linearly independent and that they form, in fact, a basis. In Theorem \[theorem\_basis\_presentation\] we will give an explicit formula for the representation of an arbitrary matroid in terms of this basis, which is based on the lattice of chains of cyclic flats of the matroid. To this end we introduce the notion of cyclic reductions, which are special weak maps. In \[subsection\_tutte\] we show that the $\mathcal{G}$-invariant and various other matroid invariants define ${\mathbb{Z}}$-linear maps on ${\mathbb{M}}_n$. Our proofs build on results of Bonin and Kung [@bkcatenary]. Section \[section\_counting\_nested\] is quite short: We prove that the number of loopfree nested matroids of rank $r$ on $n$ labeled elements is the Eulerian number $A_{r-1,n}$. Section \[section\_intersection\_product\] is dedicated to studying the multiplication on ${\mathbb{M}}_n$. In \[subsection\_chain\_matroids\] we introduce the notion of chain products, which are special products of corank one matroids and we show that these are the same as nested matroids. In \[subsection\_vanishing\] we study in which cases the product of a matroid and a nested matroid vanishes. In the last part \[subsection\_poincare\_duality\] we show that ${\mathbb{M}}_n$ is isomorphic to two familiar algebraic objects: The cohomology ring of a smooth, complete toric variety and the subring of the polytope algebra generated by matroid polytopes. The first result then immediately implies Poincaré duality (in fact, this is also a special case of a more general result by Adiprasito, Huh and Katz [@ahkhodgetheory]. Note also that Poincaré duality was known to hold for the full polytope algebra [@bstructurepolytopealgebra]). The second result shows that ${\mathbb{M}}_n$ is a quotient of the matroid polytope module considered by Derksen and Fink. The last part, Section \[section\_outlook\], contains some suggestions for further research. The author was supported by DFG grant JO366/3-2, which is part of the DFG priority project SPP 1489 ([www.computeralgebra.de](www.computeralgebra.de)). I would like to thank Michael Joswig and Benjamin Schröter for many helpful discussions and the anonymous referees for their very constructive comments. Many of the results in this paper were first discovered by computational means, using the author’s software `a-tint` [@hatint], which is an extension for `polymake` [@gjpolymake]. The latter is a software package for polyhedral and combinatorial computations. Functionality for computing in the intersection ring of matroids will be included in its next release. Preliminaries {#section_preliminaries} ============= In this section, we mostly collect the definitions and results from matroid theory and combinatorics that are relevant to this paper. Part \[subsection\_tropical\] contains definitions from tropical geometry and illuminates the origin of the ring structure that we study here. We will assume familiarity with basic notions of matroid theory, for which we recommend [@omatroidtheory; @wtheoryofmatroids] as references. All matroids are matroids on a labeled ground set, i.e. we do not consider isomorphism classes of matroids. If not explicitly stated otherwise, all matroids are assumed to be loopfree. The complement of a set $A$ is written as $A^c$. We write $U_{r,E}$ for the uniform matroid of rank $r$ on the ground set $E$ and $U_{r,n}$ if $E = \{1,\dots,n\}$. Let $M$ be a matroid on the ground set $E$. A flat of $M$ is a set $F$ such that for any $x \notin F$, ${\textnormal{rank}}_M(F \cup x) = {\textnormal{rank}}_M(F) +1$. The set of flats of $M$ is denoted by ${\mathcal{F}}(M)$. The set of flats of rank $s$ is ${\mathcal{F}}_s(M)$. The closure of a set $A$ in $M$, denoted by ${\textnormal{cl}}_M(A)$, is the smallest flat containing $A$. A spanning set of $M$ is a set $A$ such that ${\textnormal{cl}}_M(A) = E$. The set of all spanning sets is ${\mathcal{S}}(M)$. The set of bases of $M$ is written ${\mathcal{B}}(M)$. The corank of a matroid on $n$ elements is ${\textnormal{corank}}(M) = n - {\textnormal{rank}}(M)$. Similarly, the corank of a set $A$ in $M$ is ${\textnormal{corank}}_M(A) = {\textnormal{rank}}(M) - {\textnormal{rank}}_M(A)$. The nullity of a set $A$ is ${\textnormal{null}}_M(A) = {\left\lvert A \right\rvert} - {\textnormal{rank}}_M(A)$. The dual of $M$ is denoted by $M^$. A matroid $N$ on $E$ is a quotient* of a matroid $M$ on $E$ if every flat of $N$ is also a flat of $M$. Equivalently, $N$ is a quotient of $M$ if and only if every circuit of $M$ is a union of circuits of $N$. Posets, lattices and the Möbius function {#subsection_lattices} ---------------------------------------- Let $({\mathcal{P}}, \leq)$ be a poset. - Let $x,y,z \in {\mathcal{P}}$. We say that $z$ is the join of $x$ and $y$ in ${\mathcal{P}}$, written $z = x \vee_{{\mathcal{P}}} y$, if it is the unique minimal element that is larger than or equal to both $x$ and $y$. Similarly, $z$ is the meet $x \wedge_{{\mathcal{P}}} y$ if it is the unique maximal element which is smaller than or equal to both $x$ and $y$. - We call ${\mathcal{P}}$ a lattice if for any two elements $x,y \in {\mathcal{P}}$ both join and meet exist. - We say that ${\mathcal{P}}$ is join-contractible if there exists an $a \in {\mathcal{P}}$, such that $x \vee_{{\mathcal{P}}} a$ exists for all $x \in {\mathcal{P}}$. - For a poset ${\mathcal{P}}$ we denote by ${\textnormal{Ch}}({\mathcal{P}})$ the poset of nonempty chains of elements of ${\mathcal{P}}$, partially ordered by inclusion. Here a chain means a set of pairwise comparable elements. - The order complex $\Delta({\mathcal{P}})$ of ${\mathcal{P}}$ is the simplicial complex on the vertex set ${\mathcal{P}}$ whose faces are all chains in ${\mathcal{P}}$. - We denote by $\hat{{\mathcal{P}}} := {\mathcal{P}} \cup \{\hat{0},\hat{1}\}$ the poset obtained from ${\mathcal{P}}$ by adjoining artificial minimal and maximal elements, so that $\hat{0} < x < \hat{1}$ for all $x \in {\mathcal{P}}$. - We say that an element $x$ of ${\mathcal{P}}$ covers another element $y$ if $x > y$ and there is no $z$ such that $x > z > y$. The basic example of a lattice in the context of matroids is the set of flats ${\mathcal{F}}(M)$ of a matroid $M$. A partial ordering is induced by set inclusion, the meet of two flats is their intersection and the join is the closure of the union. Let ${\mathcal{P}}$ be a poset. We define the Möbius function $\mu_{{\mathcal{P}}} : {\mathcal{P}} \times {\mathcal{P}} \to {\mathbb{Z}}$ in the following, recursive manner: - $\mu_{{\mathcal{P}}}(x,y) = 0$ if $x \nleq y$. - $\mu_{{\mathcal{P}}}(x,x) = 1$. - Let $x \leq y$ and assume $\mu_{{\mathcal{P}}}(x,z)$ has been defined for all $x \leq z < y$. Then $$\mu_{{\mathcal{P}}}(x,y) = - \sum_{x \leq z < y} \mu_{{\mathcal{P}}}(x,z)\enspace.$$ We define the Möbius number of ${\mathcal{P}}$ to be $\mu({\mathcal{P}}) := \mu_{\hat{{\mathcal{P}}}}(\hat{0},\hat{1}).$ There is an abundance of literature on the Möbius function and its various properties, see for example [@rfoundationscombinatorial; @acombinatorialtheory] and the unpublished notes by Godsil [@gintroductionmoebius]. Our notation follows the latter. We will need the following well-known results (see also [@btopologicalmethods] for a more topological formulation): \[lemma\_moebius\] Let ${\mathcal{P}}$ be a poset and $x,z \in {\mathcal{P}}$. Then $$\sum_{x \leq y \leq z} \mu_{{\mathcal{P}}}(y,z) = 0\enspace.$$ \[lemma\_join\_contractible\] Let ${\mathcal{P}}$ be a join-contractible poset. Then $$\mu({\mathcal{P}}) =0\enspace.$$ The next result is a consequence of the fact that the Möbius number of ${\mathcal{P}}$ is the reduced Euler characteristic of $\Delta({\mathcal{P}})$ (e.g. [@rfoundationscombinatorial Prop. 3.6]) and that $\Delta({\mathcal{P}})$ is homeomorphic to $\Delta(\textnormal{Ch}({\mathcal{P}}))$ (the latter is the barycentric subdivision of the first). \[prop\_moebius\_chains\] Let ${\mathcal{P}}$ be a poset. Then $$\mu({\mathcal{P}}) = \mu({\textnormal{Ch}}({\mathcal{P}}))\enspace.$$ Cyclic flats, transversal and nested matroids {#subsection_cyclic_flats} --------------------------------------------- \[def\_cyclic\_flats\] For each flat $F$ of a matroid $M$, we define the cyclic part of $F$ to be $${\textnormal{cyc}}_M(F) := \bigcup_{\substack{C \textnormal{ circuit of }M\\ C \subseteq F}} C\enspace.$$ This is again a flat of $M$. We write ${\textnormal{free}}_M(F) := F {\backslash}{\textnormal{cyc}}_M(F)$ and ${\textnormal{frk}}_M(F) = {\left\lvert {\textnormal{free}}(F) \right\rvert}$. We note some obvious properties: - $M_{\mid F} = M_{\mid {\textnormal{cyc}}(F)} \oplus U_{{\textnormal{frk}}(F),{\textnormal{free}}(F)}$ and ${\textnormal{rank}}_M(F) = {\textnormal{rank}}_M({\textnormal{cyc}}_M(F)) + {\textnormal{frk}}(F)$. - Every set ${\textnormal{cyc}}_M(F) \subseteq A \subseteq F$ is again a flat of $M$. We call a flat $F$ cyclic if $F = {\textnormal{cyc}}_M(F)$. The set of all cyclic flats of $M$ is denoted by ${\mathcal{Z}}(M)$. We say that $M$ is nested if ${\mathcal{Z}}(M)$ is a chain of sets. \[remark\_cyclic\_construction\] Thomas Brylawski pointed out in [@baffinerepresentation] that knowing the ground set, all cyclic flats and their ranks is sufficient to determine the whole matroid. The actual construction will be relevant to some of our arguments, so we recall it here. We partition the set of all flats ${\mathcal{F}}(M)$ into the sets $${\mathcal{F}}(M)_{s,m} := \{F \in {\mathcal{F}}(M); {\textnormal{rank}}(F) = s \textnormal{ and } {\textnormal{frk}}(F) = m\}\enspace.$$ We then recursively build up ${\mathcal{F}}(M)$ in the following manner: 1. ${\mathcal{F}}(M)_{s,0}$ is the set of all cyclic flats of rank $s$. 2. For $m > 0$, ${\mathcal{F}}(M)_{s,m}$ is the set of all $F \cup \{p\}$, such that: - $F \in {\mathcal{F}}(M)_{s-1,m-1}$ and $p \notin F$. - $F \cup \{p\}$ is not contained in any $G$, where $G \in {\mathcal{F}}(M)_{s,m'}$ and $m' < m$. Bonin and de Mier later proved that this information in fact provides a cryptomorphic characterization of matroids. \[thm\_cyclic\_axioms\] Let ${\mathcal{Z}}$ be a collection of subsets of $E$ and $r$ an integer-valued function on ${\mathcal{Z}}$. Then ${\mathcal{Z}}$ is the collection of cyclic flats of a matroid $M$ and $r$ the restriction of the rank function on $M$ if and only if the following hold: - ${\mathcal{Z}}$ is a lattice under inclusion. - $r(0_{{\mathcal{Z}}}) = 0$, where $0_{{\mathcal{Z}}}$ is the minimal element of ${\mathcal{Z}}$. - $0 < r(Y) - r(X) < {\left\lvert Y - X \right\rvert}$ for all sets $X,Y$ in ${\mathcal{Z}}$ with $X \subsetneq Y$. - For all sets $X,Y$ in ${\mathcal{Z}}$, $$r(X) + r(Y) \geq r(X \vee_{{\mathcal{Z}}} Y) + r(X \wedge_{{\mathcal{Z}}} Y) + {\left\lvert (X \cap Y) - (X \wedge_{{\mathcal{Z}}} Y) \right\rvert}\enspace.$$ Note that in a matroid, $X \vee_{{\mathcal{Z}}(M)} Y = {\textnormal{cl}}_M(X \cup Y)$ is the usual join of flats, but $X \wedge_{{\mathcal{Z}}(M)} Y$ is the union of all circuits contained in $X \cap Y$ and can in general be strictly smaller than the flat $X \cap Y$. Also, $0_{{\mathcal{Z}}(M)}$ is the set of loops and $1_{{\mathcal{Z}}(M)}$ is the union of all circuits and thus the complement of the coloops of $M$. \[ex\_cyclic\_flats\] 1. We define a rank two matroid $M$ on $E = \{1,\dots,4\}$ via its lattice of flats: $${\mathcal{F}}(M) := \{ \emptyset, \{1,4\},\{2,3\}, E\}\enspace.$$ Its circuits are $\{1,4\}$ and $\{2,3\}$, so ${\mathcal{Z}}(M) = {\mathcal{F}}(M).$ In particular, $M$ is not nested. 2. The circuits of the uniform matroid $U_{r,n}$ are all the sets of size $r+1$. Its flats are ${\mathcal{F}}(U_{r,n}) = \{ F \subseteq E,\; {\left\lvert F \right\rvert} < r\} \cup \{E\}.$ Hence ${\mathcal{Z}}(U_{r,n}) = \{\emptyset, E\}$, so any uniform matroid is nested. Let ${\mathcal{A}} := (A_1,\dots,A_m)$ be subsets of $E$ (which need not be distinct). A partial transversal of $E$ is a subset $S \subseteq E$ such that there is a bijection $\psi: J \to S$ from a set $J \subseteq [m]$ fulfilling $\psi(j) \in A_j$ for all $j \in J$. The set of all partial transversals forms the set of independent sets of a matroid (see for example [@omatroidtheory Theorem 1.6.2]), which we denote by $M[A_1,\dots,A_m]$. We call a matroid of this form a transversal matroid. \[remark\_transversal\_presentations\] The transversal presentation, i.e. the choice of set system for a transversal matroid is not unique. However, the following holds [@bwsomeresults]: - Every rank $s$ transversal matroid has a presentation with $s$ sets $A_1,\dots,A_s$. More precisely, if $M = M[A_1,\dots,A_k]$ has a basis $B$ which is a transversal of $A_{i_1},\dots,A_{i_s}$, then $M = M[A_{i_1},\dots,A_{i_s}]$. - Every rank $s$ transversal matroid $M$ has a unique maximal presentation, i.e. one cannot increase any of the sets $A_i$ without changing the matroid. This maximal presentation is constructed as follows: 1. Assume $M = M[A_1',\dots,A_s']$ has a presentation with $s$ sets. 2. For each $i = 1,\dots,s$, replace $A_i'$ by $$A_i := A_i' \cup R_i\enspace,$$ where $R_i$ is the set of coloops of $M_{\mid A_i'^c}$. In particular, a presentation $M[A_1,\dots,A_s]$ is maximal if and only if $M_{\mid A_i^c}$ has no coloops. - The restriction of a transversal matroid is again transversal. To be precise, if $T \subseteq E$, we have $$M[A_1,\dots,A_s]_{\mid T} = M[A_1 \cap T,\dots,A_s \cap T]\enspace.$$ The survey [@bintroductiontransversal] is a good introduction to the subject of transversal matroids. They are relevant to our discussion mainly because of the following result, a proof of which can for example be found in [@oprmatroidsdomains]: \[theorem\_nested\_matroids\] A matroid $M$ is nested if and only if $M = M[A_1,\dots,A_k]$ for a chain of sets $A_1 \subseteq \dots \subseteq A_k$. \[ex\_transversal\] The matroid $M$ from Example \[ex\_cyclic\_flats\],(1) has the maximal presentation $M = M[\{1,4\}, \{2,3\}]$. The uniform matroid $U_{r,n}$ is the transversal matroid $M[E,\dots,E]$, where $E = [n]$ occurs $r$ times. The intersection ring of matroids {#subsection_intersection_ring} --------------------------------- The notion of matroid intersection seems to have gotten very little attention from matroid theorists. It is the dual operation of a much more actively studied object, the matroid union (see for example [@omatroidtheory Chapter 11.3] or [@wtheoryofmatroids Chapter 7.6]): Let $M,N$ be matroids on a ground set $E$. The union $M \vee N$ is the matroid on $E$, whose independent sets are $${\mathcal{I}}_{M \vee N} := \{ I \cup J; I \in {\mathcal{I}}_M, J \in {\mathcal{I}}_N\}\enspace.$$ The intersection of $M$ and $N$ is then defined as $$M \wedge N = (M^* \vee N^)^\enspace.$$ We note a few properties of matroid intersection: 1. It is known that both $M$ and $N$ are quotients of $M \vee N$. By duality it follows that $M \wedge N$ is a quotient of both $M$ and $N$ (see [@omatroidtheory Chapter 7.3]). 2. The spanning sets of $M \wedge N$ are given by: $${\mathcal{S}}(M \wedge N) = \{ S \cap T; S \in {\mathcal{S}}(M), T \in {\mathcal{S}}(N)\}\enspace.$$ 3. As bases are the minimal spanning sets of a matroid and since $M \wedge N$ is a quotient of both $M$ and $N$, this implies $${\mathcal{B}}(M \wedge N) = \{B \cap B'; B \in {\mathcal{B}}(M), B' \in {\mathcal{B}}(N), {\left\lvert B \cap B' \right\rvert} = {\textnormal{rank}}(M \wedge N)\}\enspace.$$ 4. Matroid intersection commutes with contraction, i.e. $$(M \wedge N)/A = (M/A) \wedge (N/A)\enspace.$$ \[ex\_intersection\] Let $M$ be any matroid on $[n]$ of rank $r > 1$. It is an easy combinatorial exercise to see that the bases of $M \wedge U_{n-1,n}$ are all the sets of the form $\{B' \subseteq B; B \textnormal{ a basis of }M\textnormal{ and }{\left\lvert B' \right\rvert} = r - 1\}$. This is also called the truncation $T(M)$ of $M$. Inductively, we see that $U_{n-k,n}$ is the $k$-fold intersection of $U_{n-1,n}$ with itself for any $k$, so $M \wedge U_{n-k,n}$ is just the $k$-fold truncation of $M$. In particular, $M \wedge U_{n-r+1,n} = U_{1,1}$ for all loopfree matroids $M$. \[defn\_matroid\_ring\] For $1 \leq r \leq n = {\left\lvert E \right\rvert}$, let $\mathfrak{C}_{r,n}$ be the set of all chains of sets of length $r$, i.e. which are of the form $$\emptyset \subsetneq F_1 \subsetneq \dots \subsetneq F_r = E\enspace.$$ We denote by $V_{r,n} = {\mathbb{Z}}^{\mathfrak{C}_{r,n}}$ the free ${\mathbb{Z}}$-module whose coordinates are indexed by the elements of $\mathfrak{C}_{r,n}$. Let ${\mathbb{M}}_{r,n}^{{\textnormal{free}}}$ be the free ${\mathbb{Z}}$-module with generators the set of all loopfree matroids of rank $r$ on the ground set $\{1,\dots,n\}$. We define a homomorphism $$\Phi_{r,n}: {\mathbb{M}}_{r,n}^{{\textnormal{free}}} \to V_{r,n}; M \mapsto v_M\enspace,$$ where for each chain ${\mathcal{C}}$ we define $$(v_M)_{{\mathcal{C}}} := \begin{cases} 1, &\textnormal{if } {\mathcal{C}} \textnormal{ is a chain of flats in }M\\ 0, &\textnormal{otherwise.} \end{cases}$$ The intersection ring of matroids on $[n]$ is the ${\mathbb{Z}}$-module $${\mathbb{M}}_n = \bigoplus_{r=1}^n {\mathbb{M}}_{r,n}\enspace,$$ with ${\mathbb{M}}_{r,n} = {\mathbb{M}}_{r,n}^{{\textnormal{free}}} / \ker \Phi_{r,n}.$ It becomes a ring with the product defined by $$M \cdot N := \begin{cases} M \wedge N, &\textnormal{if } M \wedge N \textnormal{ is loopfree}\\ 0, &\textnormal{otherwise,} \end{cases}$$ extended to linear combinations of matroids via distributivity. It is not at all obvious that this is well-defined, i.e. that the definition of the product is compatible with the additive structure on ${\mathbb{M}}_n$. However, it follows implicitly from the fact that it is a tropical intersection product (see Remark \[remark\_speyer\_isomorphism\]). \[ex\_modules\] The modules ${\mathbb{M}}_{r,n}$ are easy to write down for $r \in \{1,n\}$. In both cases there is only one loopfree matroid of rank $r$ on $n$ elements: $U_{1,n}$ and $U_{n,n}$, respectively. Hence ${\mathbb{M}}_{1,n} \cong {\mathbb{M}}_{n,n} \cong {\mathbb{Z}}$. We will postpone a nontrivial calculation until Example \[ex\_relation\], where we can make use of the geometric intuition of tropical cycles. Tropical geometry {#subsection_tropical} ----------------- In this section we will only sketch the most important definitions from tropical geometry. For a more in-depth account we recommend the book by Maclagan and Sturmfels [@MaclaganSturmfelsBook] and the book in progress by Mikhalkin and Rau [@mrtropicalgeometry]. Note that we use the $\min$-convention in our definition of matroid fans and that all coordinates are tropical projective coordinates in ${\mathbb{R}^{n}/\textbf{1}} := {\mathbb{R}}^n/{\left\langle (1,\dots,1) \right\rangle}$. A tropical cycle $(X,\omega_X)$ is a pure-dimensional, rational polyhedral complex $X$ in ${\mathbb{R}^{n}/\textbf{1}}$ together with a function $\omega_X: X^{\max} \to {\mathbb{Z}}$ defined on its maximal cells that fulfills a certain balancing condition: For a cone $\sigma$, we write $V_\sigma = {\left\langle a-b; a,b \in \sigma \right\rangle}$ and $\Lambda_\sigma = V_\sigma \cap {\mathbb{Z}}^n/\textbf{1}$. Then at every codimension one face $\tau$ of $X$ we must have $$\sum_{\sigma > \tau} \omega_X(\sigma) u_{\sigma/\tau} = 0 \;(\textnormal{mod } V_\tau)\enspace,$$ where $u_{\sigma/\tau}$ is the primitive generator of the group $\Lambda_\sigma / \Lambda_\tau \cong {\mathbb{Z}}$ pointing towards $\sigma$. The support of $X$ is the set ${\left\lvert X \right\rvert} := \bigcup_{\sigma \in X^{\max}: \omega_X(\sigma) \neq 0} \sigma$. We will consider two tropical cycles to be the same if their supports have a common refinement respecting both weight functions (in particular, cells of weight zero are considered irrelevant). For any point $p \in {\left\lvert X \right\rvert}$, we define the Star of $X$ at $p$ to be the fan $${\textnormal{Star}}_X(p) := \{ {\mathbb{R}}_{\geq 0} \cdot (\sigma - p); p \in \sigma\}\enspace,$$ with weight function $\omega_{\textnormal{Star}}({\mathbb{R}}_{\geq 0} (\sigma - p)) = \omega_X(\sigma)$. It is easy to see that this is a tropical cycle. The sum of two $k$-dimensional cycles $X,Y$ in ${\mathbb{R}^{n}/\textbf{1}}$ is defined as follows: Choose refinements of $X$ and $Y$ such that they are defined on the same polyhedral structure (possibly defining some weights to be 0). Then $$X + Y := X \cup Y \textnormal{ with weight function }\omega_{X + Y} := \omega_X + \omega_Y\enspace.$$ This operation makes the set of all $k$-dimensional cycles in ${\mathbb{R}^{n}/\textbf{1}}$ into a group, which we denote by $Z_k({\mathbb{R}^{n}/\textbf{1}})$. Let $M$ be a loopfree matroid on $[n]$. We define its matroid fan $B(M)$ to be the fan in ${\mathbb{R}^{n}/\textbf{1}}$ consisting of all cones $${\textnormal{cone}}({\mathcal{C}}) = \left\{ \sum_{i=1}^k \alpha_i v_{F_i}; \alpha_i \geq 0\right\}\enspace,$$ where ${\mathcal{C}} = (F_1 \subsetneq \dots \subsetneq F_k)$ is a chain of flats in $M$ and $v_F = \sum_{i \in F} e_i \in {\mathbb{R}^{n}/\textbf{1}}$. This is a polyhedral fan of pure dimension ${\textnormal{rank}}(M)-1$. If one equips all maximal cells with weight 1, it becomes a tropical cycle. The notation $B(M)$ is in honor of George Bergman [@blogarithmiclimit], who studied objects like these as logarithmic limit sets of algebraic varieties. The polyhedral structure given above was discovered by Ardila and Klivans [@akbergman]. The interested reader can find more information on matroid fans in the context of tropical geometry in [@MaclaganSturmfelsBook Chapter 4]. Note that each matroid fan is a subfan of the fan whose set of maximal cones is $\{{\textnormal{cone}}({\mathcal{C}})\}$, where ${\mathcal{C}}$ runs over all chains $\emptyset \subsetneq F_1 \subsetneq \cdots \subsetneq F_r = E$. Thus we can also identify a matroid fan with its indicator vector of chains $v_M \in V_{r,n}$ and the sum of two matroid fan cycles is just the sum of the indicator vectors. Hence we have: ${\mathbb{M}}_{r,n}$ is isomorphic to the subgroup of $Z_{r-1}({\mathbb{R}^{n}/\textbf{1}})$ generated by matroid fans. Note that under this identification, a linear combination of matroids now has actual geometric meaning: It is the tropical cycle sum of the corresponding matroid fans. This makes dealing with the product much easier in this context, since we will define it on arbitrary tropical cycles. \[ex\_relation\] We encounter the first nontrivial linear relations for $n = 4$ and $r = 2$. We define four matroids on $E = \{1,\dots,4\}$ in terms of their flats: $$\begin{aligned} {\mathcal{F}}(M_1) &:= \{\emptyset, \{1\},\{2\},\{3\},\{4\},E\}\enspace,\\ {\mathcal{F}}(M_2) &:= \{\emptyset, \{1,4\},\{2,3\},E\}\enspace,\\ {\mathcal{F}}(M_3) &:= \{\emptyset, \{1,4\},\{2\},\{3\},E\}\enspace,\\ {\mathcal{F}}(M_4) &:= \{\emptyset, \{1\},\{2,3\},\{4\},E\}\enspace. \end{aligned}$$ Then one sees easily that $M_1 + M_2 = M_3 + M_4$ in ${\mathbb{M}}_{2,4}$. The corresponding tropical cycles are depicted in Figure \[figure\_tropical\_sum\]. ; There is a notion of intersection product $X \cdot Y$ of two tropical cycles $X,Y$, which makes $Z_{n-1} = \bigoplus_{k \in {\mathbb{Z}}} Z_k({\mathbb{R}^{n}/\textbf{1}})$ into a ring (we set $Z_k = 0$ for $k \notin \{0,\dots,n-1\}$). There are various equivalent definitions of this product ([@mtropicalapplications; @arfirststeps; @jystableintersection]). For the sake of legibility, we adopt the following description from [@jystableintersection] of the set $X \cdot Y$ and omit the definition of the weights. Since we will only consider products of matroid fans, which – as we shall shortly see – are again matroid fans, all occurring weights are one anyway. Let $X,Y$ be two tropical cycles in ${\mathbb{R}^{n}/\textbf{1}}$. Then $${\left\lvert X\cdot Y \right\rvert} = \{ p \in {\left\lvert X \right\rvert} \cap {\left\lvert Y \right\rvert}; \dim( {\textnormal{Star}}_X(p) \boxplus {\textnormal{Star}}_Y(p)) = n-1\}\enspace,$$ where $A \boxplus B = \{a + b; a \in A, b \in B\}$ is the Minkowski sum of sets. Let $X = Y = B(U_{2,3})$. This is the one-dimensional fan in ${\mathbb{R}^{3}/\textbf{1}}$, whose three rays are spanned by the vectors $e_i, i=1,\dots,3$. We wish to compute the support of $X \cdot Y$. At every point $p \neq 0$, ${\textnormal{Star}}_X(p) = {\textnormal{Star}}_Y(p) = {\textnormal{Star}}_X(p) \boxplus {\textnormal{Star}}_Y(p)$ is an actual line. For $p = 0$, we have ${\textnormal{Star}}_X(p) \boxplus {\textnormal{Star}}_Y(p) = X \boxplus Y = {\mathbb{R}^{3}/\textbf{1}}$, so ${\left\lvert X \cdot Y \right\rvert} = \{0\}$, which is in fact the support of $B(U_{1,3} = U_{2,3} \wedge U_{2,3})$ (see also Figure \[figure\_intersection\]). (0,0) – (2,0) node\[right\] [$v_{\{1\}}$]{}; (0,0) – (0,2) node\[right\] [$v_{\{2\}}$]{}; (0,0) – (-1.5,-1.5) node\[left\] [$v_{\{3}\}$]{}; (0,1) circle (2pt) node\[right=5pt\] ; ; (0,0) circle (2pt); (0,-.5) node\[right=1pt\] ; ; \[remark\_intersection\_product\] $(Z_{n-1},+,\cdot)$, with cycle sum as addition and intersection product as multiplication is a ${\mathbb{Z}}$-algebra, graded by codimension (see [@arfirststeps] for details) and with multiplicative neutral element $1_{Z_{n-1}} = {\mathbb{R}^{n}/\textbf{1}} = B(U_{n,n})$. In particular, we have $$\begin{aligned} Z_k \cdot Z_l &\subseteq Z_{k+l-n}\enspace,\\ X \cdot (Y + Z) &= X \cdot Y + X \cdot Z\enspace. \end{aligned}$$ \[remark\_speyer\_isomorphism\] David Speyer proved in [@stropicallinear Theorem 4.11] that $$B(M)\cdot B(N) = B(M \cdot N)$$ (where $B(0) = 0$, obviously). He proved this in the more general context of valuated matroids and tropical linear spaces. Together with remark \[remark\_intersection\_product\], this already implies the well-definedness of the product on ${\mathbb{M}}_n$ and that the resulting ring is a familiar object to tropical geometers: ${\mathbb{M}}_n$ is isomorphic to the subring of the intersection ring of tropical cycles in ${\mathbb{R}^{n}/\textbf{1}}$ which is generated by matroid fans. Hence it is a commutative ring with multiplicative neutral element $U_{n,n}$ and it is graded by corank. Linear combinations of nested matroids {#section_linear_combinations} ====================================== This Section is dedicated to proving that the nested matroids of rank $r$ form a basis of ${\mathbb{M}}_{r,n}$. We will first prove linear independence in a more or less elementary manner. To show that each matroid can be written as a linear combination of nested matroids – for which we will give an explicit formula – requires more work. Linear independence {#subsection_linear_independence} ------------------- \[remark\_cyclic\_cardinality\] By Remark \[remark\_cyclic\_construction\] a nested matroid $M$ is uniquely determined by the list of tuples $$(Z_0 = \emptyset,r_0 =0),\dots,(Z_k,r_k),$$ where $Z_0 \subsetneq \dots \subsetneq Z_k$ is the chain of cyclic flats of $M$ and $r_i = {\textnormal{rank}}_M(Z_i)$. We note some obvious properties of this data: $$\begin{aligned} {\textnormal{null}}_M(Z_k) &= {\textnormal{corank}}(M)\textnormal{, as $Z_k$ is the complement of the coloops of $M$}.\\ {\textnormal{null}}_M(Z_i) &< {\textnormal{null}}_M(Z_j)\textnormal{ for all $i < j$, due to (Z2) of Theorem }\ref{thm_cyclic_axioms}. \end{aligned}$$ The set of loopfree nested matroids of rank $r$ is linearly independent in ${\mathbb{M}}_{r,n}$. For a nested matroid $M$ given by a tuple $$(Z_0 = \emptyset,r_0 := 0) , (Z_1,r_1), \dots, (Z_k, r_k)$$ as above, we define its gap measure to be the tuple $$\gamma(M) := (d_i)_{i=1,\dots,r} \in {\mathbb{N}}^r, \textnormal{ where } d_i := \begin{cases} r_i - r_{i-1}, &\textnormal{if } i \leq k\\ 0, &\textnormal{otherwise}. \end{cases}$$ Let ${\mathcal{N}}$ be the set of all loopfree nested matroids of rank $r$ on $[n]$. Assume there is a linear relation $$\sum_{M \in {\mathcal{N}}} a_M M = 0$$ in ${\mathbb{M}}_{r,n}$. We will show by lexicographic induction on $\gamma(M)$ that $a_M =0$ for all $M$. If $\gamma(M) = (1,\dots,1)$, we have $r_i = i$ for all $i$. We complete the chain of cyclic flats to a chain of flats of $M$ of length $r$ in an arbitrary manner. We claim that $M$ is the only nested matroid containing this chain. In particular $a_M$ must be 0. Assume $N$ is a nested matroid of rank $r$ whose lattice of flats contains this chain. The empty set must be cyclic in $N$, as the matroid is loopfree. Inductively, we assume $Z_0,\dots,Z_{j-1}$ are cyclic in $N$. If $Z_j$ is noncyclic, it contains the cyclic flat $G := {\textnormal{cyc}}_N(Z_j)$ with ${\left\lvert Z_j {\backslash}G \right\rvert} = {\textnormal{rank}}_N(Z_j) - {\textnormal{rank}}_N(G)$. However, as $r_i = i$ for all $i$, we know that all the cyclic flats of smaller rank are $Z_0,\dots,Z_{j-1}$ and since $Z_j$ is cyclic in $M$ we know that ${\left\lvert Z_j {\backslash}Z_i \right\rvert} > {\textnormal{rank}}(Z_j) - {\textnormal{rank}}(Z_i)$ for any $i < j$. Hence $Z_j$ must be cyclic for all $j$. As ${\textnormal{null}}_N(Z_k) = {\textnormal{null}}_M(Z_k) = {\textnormal{corank}}(M) = {\textnormal{corank}}(N)$, $N$ can contain no cyclic flat larger than $Z_k$ and we conclude that $N = M$. If $\gamma(M) =: (d_1,\dots,d_r) > (0,\dots,0)$, we again complete the chain of cyclic flats to a chain of flats of $M$ of length $r$ in an arbitrary manner. We want to show that any nested matroid $N \neq M$ containing that chain must fulfill $\gamma(N) <_{\textnormal{lex}}\gamma(M)$, so by induction $a_N = 0$ and thus finally also $a_M = 0$. So let $N$ be such a matroid and write $\gamma(N) =: (c_1,\dots,c_r)$. First assume all the $Z_i$ are cyclic in $N$. As $N \neq M$, there must be a minimal $j \geq 1$ and a minimal cyclic flat $G$ of $N$ such that $Z_{j-1} \subsetneq G \subsetneq Z_j$. Hence $c_i = d_i$ for all $i < j$ and $c_j < d_j$, so $\gamma(N) <_{\textnormal{lex}}\gamma(M)$. Now let $1 \leq j$ be minimal such that $Z_j$ is not a cyclic flat of $N$. We can assume that $Z_0,\dots,Z_{j-1}$ are all the cyclic flats of $N$ of rank at most $r_{j-1}$, since otherwise we again have $\gamma(N) <_{\textnormal{lex}}\gamma(M)$. In particular, $c_i = d_i$ for all $i < j$. Since $Z_j$ is not cyclic, we can use a similar argument as in the case $\gamma(M)=(1,\dots,1)$ to see that there must be a cyclic flat $G'$ of $N$ of rank $r_j' < r_j$ such that $Z_{j-1} \subsetneq G' \subsetneq Z_j$. Hence $c_j \leq r_j' - r_{j-1} < r_j - r_{j-1} = d_j$, so $\gamma(N) <_{\textnormal{lex}}\gamma(M)$. Cyclic reductions {#subsection_cyclic_reductions} ----------------- This section is dedicated to the notion of cyclic reductions $N$ of a matroid $M$, which are special cases of rank-preserving weak maps $N \stackrel{\textnormal{id}}{\to} M$. A weak map between matroids $M,M'$ on ground sets $E,E'$ is a map $\varphi: E \to E'$, such that for all $X \subseteq E$, we have ${\textnormal{rank}}_M(X) \geq {\textnormal{rank}}_{M'}(\varphi(X))$. It is rank-preserving if ${\textnormal{rank}}(M) = {\textnormal{rank}}(M')$ (see for example [@omatroidtheory Chapter 7.3] for more on weak maps). We will mainly be concerned with the question when a flat of $N$ is a flat of the same rank in $M$ and vice versa. Let $M,N$ be matroids on $E$. We say that $N$ is a cyclic reduction of $M$ if $\{\emptyset, {\textnormal{cyc}}_M(E)\} \subseteq {\mathcal{Z}}(N) \subseteq {\mathcal{Z}}(M)$ and the rank function on ${\mathcal{Z}}(N)$ is the one given by $M$, i.e. ${\textnormal{rank}}_N(Z) = {\textnormal{rank}}_M(Z)$ for all $Z \in {\mathcal{Z}}(N)$. The easiest way to create a cyclic reduction of a loopfree matroid $M$ is to pick a chain $\emptyset = Z_0 \subsetneq \dots \subsetneq Z_k = {\textnormal{cyc}}_M(E)$ and define $N$ to be the matroid with cyclic flats ${\mathcal{Z}}(N) = \{Z_i; i=0,\dots,k\}$ with ${\textnormal{rank}}_N(Z_i) = {\textnormal{rank}}_M(Z_i)$. In particular, $U_{r,n}$ is a cyclic reduction of any loop- and coloopfree matroid of rank $r$ on $E = [n]$. \[lemma\_flatcontainer\] Let $N$ be a cyclic reduction of $M$. For every flat $F$ of $N$, there exists a flat $G$ of $M$ such that $F \subseteq G$, ${\textnormal{rank}}_N(F) = {\textnormal{rank}}_M(G)$ and ${\textnormal{frk}}_M(G) \leq {\textnormal{frk}}_N(F)$. In particular, ${\textnormal{rank}}_M(A) \leq {\textnormal{rank}}_N(A)$ for all $A \subseteq E$, so $N \stackrel{\textnormal{id}}{\to} M$ is a rank-preserving weak map. The statement is clearly true if $s := {\textnormal{rank}}(F) = 0$, i.e. $F = \emptyset$. Now assume $s > 0$. If $F$ is cyclic, then we can choose $G = F$. Otherwise, let $m := {\textnormal{frk}}(F)$. Then by Remark \[remark\_cyclic\_construction\] $F$ is of the form $F = F' \cup \{x\}$, where $F' \in {\mathcal{F}}_{s-1,m-1}(N)$ and $x \notin F'$. By induction there exists a flat $G' \in {\mathcal{F}}(M)_{s-1,j}$ with $j \leq m-1$ and $F' \subseteq G'$. If $x \in G'$, then we can pick any $y \notin G'$. Then there exists a flat $G' \cup \{y\} \subseteq G \in {\mathcal{F}}(M)_{s,k}$, with $k \leq j+1 \leq m$. If $x \notin G'$, we can pick $y = x$ and apply the same argument. \[prop\_passdown\_flats\] Let $N$ be a cyclic reduction of $M$. Let $F$ be a flat of $M$. Then $F$ is a flat of $N$ of the same rank if and only if ${\textnormal{cyc}}_M(F)$ is a cyclic flat of $N$. For the “if” direction, assume ${\textnormal{cyc}}_M(F)$ is a cyclic flat of $N$. We prove that $F$ is a flat of $N$ by induction on ${\textnormal{frk}}(F)$. If ${\textnormal{frk}}(F) = 0$, then $F = {\textnormal{cyc}}_M(F)$ and we are done. So assume $m := {\textnormal{frk}}(F) > 0$ and write $s := {\textnormal{rank}}_M(F)$. Pick any $p \in F {\backslash}{\textnormal{cyc}}_M(F)$. The set $F' := F {\backslash}\{p\}$ is again a flat of $M$ and ${\textnormal{cyc}}_M(F') = {\textnormal{cyc}}_M(F)$. In particular, $F' \in {\mathcal{F}}_{s-1,m-1}(M)$, so by induction it is also in ${\mathcal{F}}_{s-1,m-1}(N)$. If $F \notin {\mathcal{F}}_{s,m}(N)$, then by Remark \[remark\_cyclic\_construction\] there must be a flat $G \in {\mathcal{F}}_{s,l}(N)$, with $l < m$ and $F \subsetneq G$. By Lemma \[lemma\_flatcontainer\] there exists a flat $H \in {\mathcal{F}}_s(M)$ with $F \subsetneq G \subseteq H$, which is clearly impossible. Hence $F \in {\mathcal{F}}_{s,m}(N)$. For the “only if” direction, assume $Z := {\textnormal{cyc}}_M(F)$ is not a cyclic flat of $N$, but that $F$ is a flat of both $M$ and $N$, such that ${\textnormal{rank}}_M(F) = {\textnormal{rank}}_N(F)$. As $N$ is a cyclic reduction of $M$, we must have $Z' := {\textnormal{cyc}}_N(F) \subsetneq Z$. We then have $$\begin{aligned} {\textnormal{rank}}_M(Z) + {\left\lvert F {\backslash}Z \right\rvert} &= {\textnormal{rank}}_M(F) = {\textnormal{rank}}_N(F) \\ &= {\textnormal{rank}}_N(Z') + {\left\lvert F {\backslash}Z' \right\rvert} = {\textnormal{rank}}_M(Z') + {\left\lvert F {\backslash}Z' \right\rvert}\enspace. \end{aligned}$$ On the other hand, property (Z2) of Theorem \[thm\_cyclic\_axioms\] tell us that $$\begin{aligned} {\left\lvert F {\backslash}Z' \right\rvert} - {\left\lvert F {\backslash}Z \right\rvert} = {\left\lvert Z {\backslash}Z' \right\rvert} > {\textnormal{rank}}_M(Z) - {\textnormal{rank}}_M(Z')\enspace,\end{aligned}$$ which is a contradiction. Let $N$ be a cyclic reduction of $M$ and $F \in {\mathcal{F}}(N)$ be a flat of $N$. We call $Z \in {\mathcal{Z}}(M)$ an abundant flat for $F$ in $M$ if $${\left\lvert Z \cap F \right\rvert} \geq {\textnormal{null}}_N(F) + {\textnormal{rank}}_M(Z)\enspace.$$ Equivalently, ${\left\lvert F {\backslash}Z \right\rvert} \leq {\textnormal{rank}}_N(F) - {\textnormal{rank}}_M(Z)$. Note that this implies in particular that ${\textnormal{rank}}_M(Z) \leq {\textnormal{rank}}_N(F)$. Moreover, if $Z \notin {\mathcal{Z}}(N)$, we call $Z$ a witness flat for $F$ in $M$. We denote the set of all abundant flats for $F$ by $A_M(F)$ and the set of witness flats by $W_M(F)$. \[remark\_cyclic\_abundant\] It follows from the remarks in Definition \[def\_cyclic\_flats\] that ${\textnormal{cyc}}_N(F) \in A_M(F)$. Also, any flat which is strictly contained in ${\textnormal{cyc}}_N(F)$ can never be an abundant flat for $F$ by axiom (Z2) of Theorem \[thm\_cyclic\_axioms\]. \[prop\_liftup\_flats\] Let $N$ be a cyclic reduction of $M$. Let $F \in {\mathcal{F}}_k(N)$. Then $$A_M(F) {\backslash}W_M(F) = \{ {\textnormal{cyc}}_N(F) \}\enspace.$$ Furthermore, $F \in {\mathcal{F}}_k(M)$ if and only if $W_M(F) = \emptyset$. We know from Remark \[remark\_cyclic\_abundant\] that ${\textnormal{cyc}}_N(F) \in A_M(F)$. So let $Z \in {\mathcal{Z}}(N)$ and assume ${\left\lvert Z \cap F \right\rvert} \geq {\textnormal{null}}_N(F) + {\textnormal{rank}}_M(Z) = {\left\lvert F \right\rvert} - (k - {\textnormal{rank}}_M(Z))$. Using the construction from Remark \[remark\_cyclic\_construction\], we can inductively build flats $G_i$ of $N$ of rank ${\textnormal{rank}}_M(Z) + i$ such that ${\left\lvert G_i \cap F \right\rvert} \geq {\left\lvert Z \cap F \right\rvert} + i$. In particular, there is a flat $Z \subseteq G$ of rank $k$ such that $F \subseteq G$. Hence $G = F$. By construction of $G$ we have $Z \subseteq F$, so $Z \subseteq {\textnormal{cyc}}_N(F)$. As we assumed $Z \in A_M(F)$, this implies ${\textnormal{cyc}}_N(F) = Z$ by Remark \[remark\_cyclic\_abundant\]. That proves the first statement. For the “only if” part of the second statement, assume that $F \in {\mathcal{F}}_k(M)$ as well. Let $Z \in {\mathcal{Z}}(M)$ and assume $Z \in W_M(F)$. We construct a flat $G \in {\mathcal{F}}_k(M)$ in the same manner as before and conclude again that $F = G$ and $Z = {\textnormal{cyc}}_M(F)$. By Proposition \[prop\_passdown\_flats\], $Z \in {\mathcal{Z}}(N)$, which is a contradiction. For the “if” direction, assume $F \in {\mathcal{F}}_k(N) {\backslash}{\mathcal{F}}_k(M)$. By Lemma \[lemma\_flatcontainer\], there exists a flat $F \subsetneq G$ in $M$ of rank $k$. Let $Z := {\textnormal{cyc}}_M(G)$. Then by Proposition \[prop\_passdown\_flats\], $Z \notin {\mathcal{Z}}(N)$, since otherwise $G$ would be a flat of $N$. Let $Z' := {\textnormal{cyc}}_N(F)$. Then $Z' \subseteq Z$, so ${\left\lvert Z \cap F \right\rvert} \geq {\left\lvert Z' \right\rvert} = {\left\lvert F \right\rvert} - (k - {\textnormal{rank}}_M(Z')) \geq {\left\lvert F \right\rvert} - (k - {\textnormal{rank}}_M(Z)).$ Hence $Z \in W_M(F)$. \[lemma\_witness\_join\_or\_meet\] Let $N$ be a cyclic reduction of $M$. Let $F \subseteq F'$ be flats of $N$ and assume $Z \in A_M(F)$ and $Z' \in A_M(F')$. Then at least one of the following is true: - $Z \wedge_{{\mathcal{Z}}(M)} Z' \in W_M(F)\enspace.$ - $Z \vee_{{\mathcal{Z}}(M)} Z' \in A_M(F')\enspace.$ If $Z' \in W_M(F')$, then either $Z \wedge_{{\mathcal{Z}}(M)} Z' \in W_M(F)$ or $Z \vee_{{\mathcal{Z}}(M)} Z' \in W_M(F')$. As a shorthand, write $X := Z \wedge_{{\mathcal{Z}}(M)} Z'$ and $Y := Z \vee_{{\mathcal{Z}}(M)} Z'$. Assume that $X \notin W_M(F)$. Then by Proposition \[prop\_liftup\_flats\] ${\left\lvert X \cap F \right\rvert} \leq {\textnormal{null}}_N(F) + {\textnormal{rank}}_M(X)$, with equality if and only if $X = {\textnormal{cyc}}_N(F)$. We have to show that $Y$ contains sufficiently many elements of $F'$. So we compute $$\begin{aligned} {\left\lvert Y \cap F' \right\rvert} &\geq {\left\lvert (Z \cup Z') \cap F' \right\rvert} = {\left\lvert Z \cap F' \right\rvert} + {\left\lvert Z' \cap F' \right\rvert} - {\left\lvert Z \cap Z' \cap F' \right\rvert} \\ \textnormal{\tiny as $F \subseteq F'$}\quad&= {\left\lvert Z \cap F \right\rvert} + {\left\lvert Z' \cap F' \right\rvert} - {\left\lvert Z \cap Z' \cap F' \right\rvert} + {\left\lvert Z \cap (F' {\backslash}F) \right\rvert}\\ \textnormal{\tiny as $Z \in A_M(F), Z' \in A_M(F)$}\quad&\geq {\textnormal{null}}_N(F) + {\textnormal{null}}_N(F') + {\textnormal{rank}}_M(Z) + {\textnormal{rank}}_M(Z') \\ &\qquad - {\left\lvert Z \cap Z' \cap F' \right\rvert} + {\left\lvert Z \cap (F' {\backslash}F) \right\rvert}\\ \textnormal{\tiny by (Z3)}\quad&\geq ({\textnormal{null}}_N(F) + {\textnormal{rank}}_M(X)) + ({\textnormal{null}}_N(F') + {\textnormal{rank}}_M(Y))\\ &\qquad + {\left\lvert (Z \cap Z') {\backslash}X \right\rvert} - {\left\lvert Z \cap Z' \cap F' \right\rvert} + {\left\lvert Z \cap (F' {\backslash}F) \right\rvert}\\ \textnormal{\tiny as $X \notin W_M(F)$}\quad&\geq {\textnormal{null}}_N(F') + {\textnormal{rank}}_M(Y)\\ &\qquad + \underbrace{{\left\lvert X \cap F \right\rvert}+ {\left\lvert (Z \cap Z') {\backslash}X \right\rvert} - {\left\lvert Z \cap Z' \cap F' \right\rvert} + {\left\lvert Z \cap (F' {\backslash}F) \right\rvert}}_{=: \delta}\enspace.\\ \end{aligned}$$ So we only have to show that $\delta \geq 0$. For this note that it is trivially true that ${\left\lvert Z \cap (F' {\backslash}F) \right\rvert} \geq {\left\lvert X \cap (F' {\backslash}F) \right\rvert}$. Also, we have $$\begin{aligned} {\left\lvert Z \cap Z' \cap F' \right\rvert} - {\left\lvert X \cap F \right\rvert} - {\left\lvert X \cap (F' {\backslash}F) \right\rvert} &= {\left\lvert Z \cap Z' \cap F' \right\rvert} - {\left\lvert X \cap F' \right\rvert} \\ &= {\left\lvert ((Z \cap Z'){\backslash}X) \cap F' \right\rvert}\enspace. \end{aligned}$$ Hence we have $$\delta \geq {\left\lvert (Z \cap Z') {\backslash}X \right\rvert} - {\left\lvert ( (Z \cap Z') {\backslash}X) \cap F' \right\rvert} \geq 0\enspace.$$ Finally, assume that $Z' \in W_M(F')$. As we already proved that $Y \in A_M(F)$, then by Proposition \[prop\_liftup\_flats\] we only need to prove that $Y \neq {\textnormal{cyc}}_N(F')$. But if that was the case, we would have $Z' \subseteq Y = {\textnormal{cyc}}_N(F')$, so by Remark \[remark\_cyclic\_abundant\] $Z'$ cannot be a witness flat for $F'$. Representations of arbitrary matroids {#subsection_representations} ------------------------------------- We will begin by stating how an arbitrary matroid can be written as a linear combination of nested matroids. Let $M$ be a matroid. For any maximal chain ${\mathcal{C}} = (F_0 = \emptyset \subsetneq F_1 \subsetneq \dots \subsetneq F_r = E$ of flats we define its cyclic set to be ${\textnormal{cyc}}({\mathcal{C}}) = \{ {\textnormal{cyc}}_M(F_i); i = 0,\dots,r\} \subseteq {\mathcal{Z}}(M)$. The cyclic chain lattice of $M$ is the set $${{{\mathcal{C}}_{{\mathcal{Z}}}}}(M) := \{ T \subseteq {\mathcal{Z}}(M) \textnormal{ a chain with } \emptyset, 1_{{\mathcal{Z}}} \in T\} \cup \{ \hat{1}\}\enspace,$$ with partial order induced by set inclusion and $\hat{1}$ as an artificial maximal element, i.e.: - $T < \hat{1}$ for all chains $T$. - If $T,T' \neq \hat{1}$, then $T \leq T'$ if and only if $T \subseteq T'$. This is a lattice with $T \wedge T' = T \cap T'$ and $$T \vee T' = \begin{cases} T \cup T', &\textnormal{if } T \cup T' \textnormal{ is a chain,}\\ \hat{1}, &\textnormal{otherwise.} \end{cases}$$ Note that each element $\hat{1} \neq T \in {{{\mathcal{C}}_{{\mathcal{Z}}}}}(M)$ defines a chain of cyclic flats of $M$ (with ranks given by ${\textnormal{rank}}_M$) and thus a nested matroid, which we denote by $M{_{\left(T\right)}}$. Furthermore, we will use the following shorthand: For any $T \in {{{\mathcal{C}}_{{\mathcal{Z}}}}}(M)$, let $$\mu_1(T) := \mu_{{{{\mathcal{C}}_{{\mathcal{Z}}}}}(M)}(T,\hat{1})\enspace.$$ \[theorem\_basis\_presentation\] Let $M$ be a loopfree matroid of rank $r$ on $[n]$. Then the following equality holds in ${\mathbb{M}}_{r,n}$: $$M = \sum_{\substack{T \in {{{\mathcal{C}}_{{\mathcal{Z}}}}}(M)\\T \neq \hat{1}}} \left( - \mu_1(T) \right) M{_{\left(T\right)}}\enspace.$$ Let ${\mathcal{C}} = (\emptyset = F_0 \subsetneq F_1 \subsetneq \dots \subsetneq F_r = E)$ be a chain of length $r = {\textnormal{rank}}(M)$. First assume ${\mathcal{C}}$ is a chain of flats of $M$ and denote by $T$ its cyclic set. Then by Proposition \[prop\_passdown\_flats\], ${\mathcal{C}}$ is a chain of flats in $M{_{\left(T'\right)}}$ for an element $\hat{1} \neq T' \in {{{\mathcal{C}}_{{\mathcal{Z}}}}}(M)$ if and only if $T \leq T'$. So by Lemma \[lemma\_moebius\] the coefficient of ${\mathcal{C}}$ on the right hand side in the above equation is $$\sum_{\hat{1} \neq T' \geq T} - \mu_1(T') = \mu_1(\hat{1}) = 1\enspace.$$ Now assume ${\mathcal{C}}$ is a chain of some $M{_{\left(T\right)}}$, but not of $M$. Let $T_1,\dots,T_k$ be the minimal elements of ${{{\mathcal{C}}_{{\mathcal{Z}}}}}(M)$ such that ${\mathcal{C}}$ is a chain in $M{_{\left(T_i\right)}}, i = 1,\dots,k$. In particular, by Proposition \[prop\_passdown\_flats\] every $Z \in T_i$ is the cyclic part of some $F_j$. Now if $T$ is such that ${\mathcal{C}}$ is a chain in $M{_{\left(T\right)}}$, there is a unique $i = 1,\dots,k$ such that $T \geq T_i$: If $T \geq T_i, T_j$, then by Proposition \[prop\_passdown\_flats\], both $T_i$ and $T_j$ contain ${\textnormal{cyc}}_{M{_{\left(T\right)}}}({\mathcal{C}})$. Hence so does $T_i \cap T_j$, which is a contradiction to the minimality assumption. For $i = 1,\dots,k$ we define the following subposet of ${{{\mathcal{C}}_{{\mathcal{Z}}}}}(M)$: $$R_i := \{\hat{1} \neq T > T_i;\; {\mathcal{C}} \textnormal{ is not a chain in } M{_{\left(T\right)}}\}\enspace.$$ Proposition \[prop\_passdown\_flats\] implies that if $T \in R_i$ and $T' > T$, then $T' \in R_i$. It follows from Lemma \[lemma\_moebius\] that $$\label{eq_ri_sum} \mu(R_i) = \mu_{\hat{R_i}}(\hat{0},\hat{1}) = - \sum_{T \in R_i \cup \{\hat{1}\}} \mu_{\hat{R_i}}(T,\hat{1}) = - \sum_{T \in R_i \cup \{\hat{1}\}} \mu_1(T)\enspace.$$ We write $W_i := \{Z \in {\mathcal{Z}}(M); T_i \cup \{Z\} \in R_i\}$. Equivalently, this is the set of cyclic flats $Z \notin T_i$ such that $T_i \cup \{Z\}$ is a chain and $Z \in W_M(F)$ for some $F$ in ${\mathcal{C}}$. Note that if $W_i$ is not empty, it is a join-contractible subposet of ${\mathcal{Z}}(M)$: Pick any minimal element $Z$ of $W_i$ and let $Z' \in W_i$ be arbitrary. If $Z \subseteq Z'$, then $Z \vee_{{\mathcal{Z}}(M)} Z' = Z' \in W_i$. Otherwise, the meet of $Z$ and $Z'$ cannot lie in $W_i$ due to the minimality of $Z$. By Lemma \[lemma\_witness\_join\_or\_meet\] $Z \vee_{{\mathcal{Z}}(M)} Z'$ then lies in $W_i$. By the above considerations and Proposition \[prop\_liftup\_flats\], we have a poset isomorphism $${\textnormal{Ch}}(W_i) \to R_i,\quad \{Z_1,\dots,Z_k\} \mapsto T_i \cup \{Z_1,\dots,Z_k\}\enspace.$$ Applying Lemma \[lemma\_join\_contractible\] and Proposition \[prop\_moebius\_chains\] we see that if $R_i \neq \emptyset$, we have $$\label{eq_wi_mu} \mu(R_i) = \mu({\textnormal{Ch}}(W_i)) = \mu(W_i) = 0\enspace.$$ We want to show that $R_i \neq \emptyset$ or, equivalently, that $W_i \neq \emptyset$. We need to construct a witness flat that forms a chain with $T_i$. As we assumed that ${\mathcal{C}}$ is not a chain in $M$, Proposition \[prop\_liftup\_flats\] tells us that there must be a noncyclic flat $F$ in ${\mathcal{C}}$ and a witness flat $Z \in W_M(F)$. We choose $F$ maximal, such that a witness flat exists for it and let $Z$ also be a maximal element of $W_M(F)$. Denote by $Z_1 := {\textnormal{cyc}}_{M{_{\left(T_i\right)}}}(F)$. Then $Z_1$ is not equal to $1_{{\mathcal{Z}}(M)}$, otherwise there could be no witness flats for $F$. So let $Z_2$ be the smallest element of $T_i$ that strictly contains $Z_1$. As $Z_1 \wedge_{{\mathcal{Z}}(M)} Z \subseteq Z_1$, it cannot be a witness flat for $F$ by Remark \[remark\_cyclic\_abundant\]. Hence, by Lemma \[lemma\_witness\_join\_or\_meet\] we know that $Z_1 \vee_{{\mathcal{Z}}(M)} Z \in W_M(F)$. Due to the maximality of $Z$, we must have $Z_1 \vee_{{\mathcal{Z}}(M)} Z = Z$, so $Z_1 \subseteq Z$. Due to the minimality of $T_i$, there must be a flat $F \subsetneq F'$ such that $Z_2 = {\textnormal{cyc}}_{M{_{\left(T_i\right)}}}(F')$. In particular, $Z_2 \in A_M(F')$. Let $X := Z_2 \wedge_{{\mathcal{Z}}(M)} Z$ and $Y := Z_2 \vee_{{\mathcal{Z}}(M)} Z$. By Lemma \[lemma\_witness\_join\_or\_meet\] there are two possibilities: If $X \in W_M(F)$, we have $Z_1 \subseteq X \subseteq Z_2$, so $X \in W_i$. If $Y \in A_M(F')$, then by our choice of $F$ there are no witness flats for $F'$, so $Y = Z_2$ by Proposition \[prop\_liftup\_flats\]. But then $Z_1 \subseteq Z \subseteq Y = Z_2$, so $Z \in W_i$. We can now finally use Equations \[eq\_ri\_sum\] and \[eq\_wi\_mu\], as well as Lemma \[lemma\_moebius\] to see that the coefficient of ${\mathcal{C}}$ on the right hand side above is $$- \sum_{i=1}^k\left( \sum_{T \geq T_i} \mu_1(T) - \sum_{T' \in R_i \cup \{\hat{1}\}} \mu_1(T') \right) = - \sum_{i=1}^k (0 - 0) =0\enspace.$$ \[corollary\_basis\] The set of loopfree nested matroids of rank $r$ on $[n]$ is a basis for ${\mathbb{M}}_{r,n}$, i.e. every matroid can be written as a unique linear combination of nested matroids. It should be noted that Derksen and Fink show in [@dfvaluative] that the polytopes of nested matroids (which they call Schubert matroids) also form a basis for their module of polytopes. In their case the representation of an arbitrary matroid is given as a sum over all possible chains of sets (for details see their Theorem 4.2). It would be interesting to study the precise relation of these two presentations and what it implies for the combinatorics of the matroids involved (see also Remark \[remark\_derksen\_fink\]). \[ex\_representation\] 1. We recall the matroid $M$ on $E = \{1,\dots,4\}$ from example \[ex\_cyclic\_flats\], whose flats were given by ${\mathcal{F}}(M) = {\mathcal{Z}}(M) = \{\emptyset, \{1,4\}, \{2,3\},E\}$. The theorem tells us that $$M = M{_{\left(\emptyset, \{1,4\}, E\right)}} + M{_{\left(\emptyset, \{2,3\},E\right)}} - M{_{\left(\emptyset,E\right)}}\enspace,$$ which is the same relation we already encountered in Example \[ex\_relation\]. 2. We consider the matroid $M$ of rank 4 on $E := \{1,\dots,8\}$ given by the lattice of cyclic flats depicted in Figure \[figure\_torsion\_poset\]. The cyclic chain poset is drawn below. According to Theorem \[theorem\_basis\_presentation\], we have $$\begin{aligned} M = &\;M{_{\left(S_1,R,U_1\right)}} + M{_{\left(S_2,R,U_1\right)}} + M{_{\left(S_1,R,U_2\right)}} + M{_{\left(S_2,R,U_2\right)}}\\ &\; - M{_{\left(R,U_1\right)}} - M{_{\left(S_1,R\right)}} - M{_{\left(S_2,R\right)}} - M{_{\left(R,U_2\right)}} \\&\;+ M{_{\left(R\right)}} \end{aligned}$$ (Note that we omit $\emptyset,E$ in the description of each chain set). ; The G-invariant {#subsection_tutte} --------------- As mentioned in the introduction, the $\mathcal{G}$-invariant is an important matroid invariant. We will use the characterization by Bonin and Kung [@bkcatenary] in terms of catenary data as a definition. Assume $n \geq 1$ and $1 \leq r \leq n$. Let ${\mathcal{C}} = (F_0 \subsetneq F_1 \subsetneq \dots \subsetneq F_r = E)$ be a chain of sets. Its composition is the tuple $(a_0,\dots,a_r)$ with $a_0 := {\left\lvert F_0 \right\rvert}$ and $a_i := {\left\lvert F_i {\backslash}F_{i-1} \right\rvert}$ for $i > 0$. For a matroid $M$ of rank $r$ and a fixed composition $a := (a_0,\dots,a_r)$ we define $\nu(M;a)$ to be the number of chains of flats of $M$ of length $r$ with composition $a$. Now let ${\mathcal{G}}(n,r)$ be the free abelian group on all possible compositions $a$. We will denote the generator corresponding to $a$ by $\gamma(a)$ in accordance with the notation in [@bkcatenary]. The ${\mathcal{G}}$-invariant of $M$ is $${\mathcal{G}}(M) := \sum_a \nu(M;a) \gamma(a) \in {\mathcal{G}}(n,r)\enspace,$$ where the sum runs over all possible compositions $a$. \[theorem\_g\_invariant\] For each $n$ and $r$ the ${\mathcal{G}}$-invariant induces a ${\mathbb{Z}}$-module homomorphism via $${\mathbb{M}}_{r,n} \to {\mathcal{G}}(r,n),\; M \mapsto {\mathcal{G}}(M)\enspace.$$ We only need to show that for each composition $a$, there is an induced ${\mathbb{Z}}$-module homomorphism ${\mathbb{M}}_{r,n} \to {\mathbb{Z}},\, M \mapsto \nu(M;a)$. But this is obvious from the definition of ${\mathbb{M}}_{r,n}$ as a submodule of $V_{r,n}$: Project to the coordinates of chains with composition $a$, then take the sum of the coordinates. The significance of the $\mathcal{G}$-invariant can also be recognized from the fact that many other matroid invariants can be derived from it. In particular, Derksen [@dsymmetricpolymatroids] showed that the Tutte polynomial $$t_M(x,y) := \sum_{S \subseteq E} (x-1)^{{\textnormal{corank}}_M(S)} (y-1)^{{\textnormal{null}}_M(S)} \in {\mathbb{Z}}[x,y]$$ can be computed from the $\mathcal{G}$-invariant by applying a certain linear map to $\mathcal{G}(n,r)$. This immediately implies the following: The Tutte polynomial induces a ${\mathbb{Z}}$-module homomorphism via $${\mathbb{M}}_{r,n} \to {\mathbb{Z}}[x,y],\; M \mapsto t_M(x,y)\enspace.$$ This implies of course that the same statement holds for all generalized Tutte-Grothendieck invariants and Tutte-Grothendieck group invariants, as defined in [@botutteapplications], such as for example the characteristic polynomial, the beta invariant, the Whitney numbers of the first kind or the number of bases. However, there are also invariants which can not be derived from the Tutte polynomial, but which are encoded in the ${\mathcal{G}}$-invariant. We consider a particular example from [@bkcatenary], again adopting their notation: For a matroid $M$, we denote by $F_{h,k}(M;s_h,\dots,s_k)$ the number of chains of flats $F_h \subsetneq \dots \subsetneq F_k$ of $M$ such that ${\textnormal{rank}}(F_i) = i$ and ${\left\lvert F_i \right\rvert} = s_i$ for all $i = h,\dots,k$. For fixed $h,k$ and $s := (s_h,\dots,s_k)$, there is a ${\mathbb{Z}}$-module homomorphism ${\mathbb{M}}_{r,n} \to {\mathbb{Z}}$ induced by $M \mapsto F_{h,k}(M;s_h,\dots,s_k)$. In [@bkcatenary Proposition 5.2], Bonin and Kung show how the numbers $F_{h,k}(M;s)$ are derived from the $\mathcal{G}$-invariant. We follow their argument to see that we indeed obtain a linear map. For $k = r$, they show there exists a linear specialization map $\textnormal{spec}: \mathcal{G}(n,r) \to {\mathbb{Z}}$ such that the following identity holds: $$F_{h,r}(M;s) = \textnormal{spec}\left((1/s_h!)\sum_a \nu(M; a') \gamma(a)\right)\enspace,$$ where $a$ runs over a certain set of compositions depending on $h$ and $s$ and $a'$ is a composition that depends only on $a$ and $s$. We saw in Theorem \[theorem\_g\_invariant\] that $\nu(M;a')$ induces a linear map on ${\mathbb{M}}_{n,r}$. Hence $F_{h,r}(M;s)$ induces a linear map. For $k = r-1$ the statement follows from the observation that $$F_{h,r-1}(M;s) = F_{h,r}(M;s,n)\enspace.$$ For $k < r-1$, note that $F_{h,k}(M;s) = F_{h,k}(T^{r-k-1}M;s)$, where $T^iM$ denotes again the $i$-fold truncation of $M$. Recall from example \[ex\_intersection\] that $T^iM = M \cdot U_{n-i,n}$, so $M \mapsto T^iM$ induces a linear map. Thus the claim follows by induction. Let $f_k(M;s,c)$ denote the number of flats $F$ of $M$ of size $s$ and rank $k$ such that $M_{\mid F}$ has $c$ coloops. This induces a ${\mathbb{Z}}$-module homomorphism ${\mathbb{M}}_{r,n} \to {\mathbb{Z}}$. In particular, the number of flats of rank $k$ and the number of cyclic flats of rank $k$ induce ${\mathbb{Z}}$-module homomorphisms as well. The statement is trivial for $k = r$. It is shown in [@bkcatenary Proposition 5.5] that for $k=r-1$ we have $$\sum_{j=c}^n f_{r-1}(M;s,j) \frac{j!}{(j-c)!} = F_{r-1-c,r-1}(M;s-c,s-c+1,\dots,s)\enspace.$$ An easy inductive argument then shows that for $k=r-1$ we obtain a linear map. The general case then follows by using truncations as in the proof above. We again look at the matroid $M$ on $E = \{1,\dots,4\}$ from Examples \[ex\_cyclic\_flats\] and \[ex\_representation\],(1). From the latter we recall the linear relation $$M = M{_{\left(\emptyset,\{1,4\},E\right)}} + M{_{\left(\emptyset,\{2,3\},E\right)}} - M{_{\left(\emptyset,E\right)}}\enspace.$$ Computing the ${\mathcal{G}}$-invariants on the right hand side, we see that $$\begin{aligned} {\mathcal{G}}(M) &= ( 2\gamma(0,1,3) + \gamma(0,2,2)) + (2\gamma(0,1,3) + \gamma(0,2,2)) - 4 \gamma(0,1,3)\\ &= 2\gamma(0,2,2)\enspace. \end{aligned}$$ Indeed, $M$ has two maximal chains of flats, $(\emptyset, \{1,4\},E)$ and $(\emptyset,\{2,3\},E)$. Counting nested matroids {#section_counting_nested} ======================== We have shown that nested matroids are a basis for ${\mathbb{M}}_{r,n}$, so naturally we want to determine their exact number. It was already pointed out in [@oprmatroidsdomains] that the number of isomorphism classes of nested matroids of rank $r$ on $n$ elements is $\binom{n}{r}$. We will show that without taking isomorphisms into account, we still get a familiar number. Let $n \geq 1$ and $0 \leq r < n$. The Eulerian number $A_{r,n}$ is the number of permutations on $\{1,\dots,n\}$ with $r$ ascents. An ascent of a permutation $\sigma$ is a number $i \in \{1,\dots,n-1\}$, such that $\sigma(i) < \sigma(i+1)$. Another interpretation of these numbers is the following: For $0 < k < n$, define the hypersimplex $$\Delta_{k,n} := \left\{(x_1,\dots,x_n), x_i \in [0,1], \sum x_i = k\right\}\enspace.$$ Then $A_{r,n}$ is the lattice volume of $\Delta_{r+1,n+1}$ (see [@lpalcoved] for a discussion of a concrete unimodular triangulation of the hypersimplex). We note a few properties of these numbers here, which can be found in most standard textbooks on combinatorics such as [@gkpconcretemath]: - By definition, we have $\sum_{r=0}^{n-1} A_{r,n} = n!$. - The Eulerian numbers are symmetric: $A_{r,n} = A_{n-r-1,n}$. - The generating function of the Eulerian numbers is $$\sum_{n=0}^\infty \sum_{r=0}^\infty A_{r,n} \frac{x^rt^n}{n!} = \frac{x-1}{x - e^{(x-1)t}}\enspace.$$ We denote by $N_{r,n}$ the number of loopfree nested matroids of rank $r$ on $n$ labeled elements. \[lemma\_nested\_recursive\] $N_{r,n}$ is determined by the following recursive relation: $$N_{r,n} = 1 + \sum_{k=1}^{r-1} \left(\sum_{s=k+1}^{k+n-r} \binom{n}{s} N_{r-k,n-s}\right) \textnormal{ if }r > 1\enspace,$$ and $N_{1,m} = 1$ for any $m \geq 1$. There is only one loopfree matroid of rank 1 on $m$ elements, the uniform matroid, which is also nested. Hence $N_{1,m} = 1$. We build a nested matroid of rank $r > 1$ by recursively constructing its chain of cyclic flats. The first one is always the empty set of rank zero, as the matroid is loopfree. We consider the choices we have for the first nonempty cyclic flat $F$: We can choose its rank $k$ and its size $s$. If the rank is $r$, then $F = E$ and there is only one nested matroid of this form. Otherwise pick any rank $1 \leq k < r$. By (Z2) of Theorem \[thm\_cyclic\_axioms\], the size $s$ of $F$ has to be at least $k+1$ and can be at most $k+n-r$. We clearly have $\binom{n}{s}$ possibilities to choose such an $F$. For each nested matroid of rank $r$ on $[n]$ whose chain of cyclic flats is of the form$(\emptyset,0) \subsetneq (F_1 = F, r_1 = k) \subsetneq \dots \subsetneq (F_l,r_l)$, the deletion of $F$ gives a nested matroid of rank $r-k$ on $n-s$ elements. Conversely, we can take any such matroid and lift it to a nested matroid of rank $r$ on $n$ elements, whose chain of cyclic flats starts with $(\emptyset,0) \subsetneq (F,k)$. This proves the claim. \[theorem\_counting\_nested\] For any $n \geq 1$ and $1 \leq r \leq n$, we have $$N_{r,n} = A_{r-1,n}\enspace.$$ We prove this by showing equality of generating functions. As stated above, we have $$\sum_{n=0}^\infty \sum_{r=0}^\infty A_{r,n} \frac{x^rt^n}{n!} = \frac{x-1}{x - e^{(x-1)t}}\enspace.$$ Here we set $A_{0,0} = 1$ and $A_{r,n} = 0$ if $0 < n \leq r$ or $\min\{n,r\} < 0$. In accordance with this, we define $$N_{1,0} = 1\textnormal{ and }N_{a,b} = 0\textnormal{ for all }(a,b) \notin \{(r,n); 1 \leq r \leq n, 1 \leq n\} \cup \{(1,0)\}\enspace.$$ We can then rewrite the formula of Lemma \[lemma\_nested\_recursive\] as $$N_{r,n} = \sum_{k=1}^\infty \sum_{s=k+1}^\infty \binom{n}{s} N_{r-k,n-s}, \textnormal{ if } r > 1\enspace.$$ We thus compute: $$\begin{aligned} f(x,t) &= \sum_{n=0}^\infty \sum_{r=0}^\infty N_{r+1,n} \frac{x^r t^n}{n!}\\ &= \sum_{n=0}^\infty N_{1,n} \frac{t^n}{n!} + \sum_{n=0}^\infty \sum_{r=1}^\infty \left( \sum_{k=1}^\infty \sum_{s=k+1}^\infty \binom{n}{s} N_{r-k+1,n-s}\right) \frac{x^rt^n}{n!}\\ &= e^t + \sum_{k=1}^\infty \sum_{s=k+1}^\infty \frac{x^kt^s}{s!}\underbrace{\left( \sum_{n=0}^\infty \sum_{r=1}^\infty N_{r-k+1,n-s} \frac{x^{r-k} t^{n-s}}{(n-s)!} \right)}_{= f(x,t)}\\ &= e^t + f(x,t)\left( \sum_{k=1}^\infty \left( e^t - \sum_{i=0}^k \frac{t^i}{i!} \right)x^k\right) \\ &= e^t + f(x,t)\left(\frac{xe^t - e^{tx}}{1-x} + 1\right)\enspace.\\\end{aligned}$$ Solving for $f$, we get $$f(x,t) = \frac{-e^t (1-x)}{xe^t - e^{tx}} = \frac{x-1}{x - e^{(x-1)t}}\enspace.$$ The intersection product {#section_intersection_product} ======================== In this section, we will study the properties of the intersection product on ${\mathbb{M}}_n$. We will first prove that each nested matroid is a certain product of corank one matroids. We will then study when certain products of matroids vanish in ${\mathbb{M}}_n$. Finally we will show that ${\mathbb{M}}_n$ fulfills a Poincaré-type duality. Chain products {#subsection_chain_matroids} -------------- Fix a ground set $E$ of size $n$. One sees easily that a loopfree corank one matroid is uniquely determined by fixing its set of coloops $G$, where $0 \leq {\left\lvert G \right\rvert} \leq n-2$. It must then be of the form $$H_G := U_{{\left\lvert G \right\rvert},G} \oplus U_{{\left\lvert G^c \right\rvert}-1,G^c}\enspace.$$ Given a chain of sets ${\mathcal{G}} = (G_1 \subsetneq \dots \subsetneq G_k)$ with ${\left\lvert G_k \right\rvert} \leq n-2$, we define its chain product to be $$M_{{\mathcal{G}}} := H_{G_1} \wedge \dots \wedge H_{G_k}\enspace.$$ We also fix the following notation for set systems: For a set $S$ and a positive integer $l$, we write $S^{\oplus l}$ for the $l$-fold concatenation of $S$, i.e. $$S^{\oplus l} = \underbrace{(S,\cdots,S)}_{l \textnormal{ times}}\enspace.$$ We make a few observations about chain products and corank one matroids that are easily verified: - $M_{{\mathcal{G}}}$ is a loopfree matroid of rank $n-k$. In fact, its bases are given by $${\mathcal{B}}(M_{{\mathcal{G}}}) = \left\{E{\backslash}\{j_1,\dots,j_k\}; j_i \notin G_i \textnormal{ for all }i \textnormal{ and }{\left\lvert \{j_1,\dots,j_k\} \right\rvert} = k\right\}.$$ Hence we have $M_{{\mathcal{G}}} = H_{G_1} \cdot \,\dots\, \cdot H_{G_k}$. - For any loopfree matroid $M$ with ${\textnormal{rank}}(M) > 1$ and any corank one matroid $H_G$, their matroid intersection fulfills ${\textnormal{rank}}(M \wedge H_G) = {\textnormal{rank}}(M) -1$ (it may have loops, though). \[lemma\_chain\_bases\] The bases of $M_{{\mathcal{G}}}$ are $$\left\{B \in \binom{[n]}{n-k}; {\left\lvert G_i {\backslash}B \right\rvert} \leq i - 1 \textnormal{ for all } i = 1,\dots,k\right\}\enspace.$$ By our remark above the bases of $M_{{\mathcal{G}}}$ are of the form $E {\backslash}\{j_1,\dots,j_k\}$, with $j_i \in G_i^c$. In particular, $G_i {\backslash}B$ can contain at most $j_1,\dots,j_{i-1}$. Thus any basis of $M_{{\mathcal{G}}}$ is of the given form. Conversely let $B \in \binom{[n]}{n-k}$ with ${\left\lvert B \cap G_i \right\rvert} \geq {\left\lvert G_i \right\rvert} - (i - 1)$ for all $i$. In particular $B = E {\backslash}\{j_1,\dots,j_k\}$ for some $j_i \in [n]$. We define an ordering on $[n]$ in the following manner. For $j \in [n]$, let $m_{{\mathcal{G}}}(j) := \min\{i; j \in G_i\}$. Then we say that $j <_{{\mathcal{G}}} j'$ if and only if $m_{{\mathcal{G}}}(j) < m_{{\mathcal{G}}}(j')$ or equality holds and $j < j'$. We can assume without loss of generality that $j_1 <_{{\mathcal{G}}} \dots <_{{\mathcal{G}}} j_k$. By assumption, $j_1$ must lie in $G_1^c$, so $m_{{\mathcal{G}}}(j_1) \geq 2$. Using the fact that ${\left\lvert B^c \cap G_2 \right\rvert} \leq 1$, one sees that $j_2$ lies in $G_2^c$. One can continue inductively to see that in fact $j_i \in G_i^c$ for all $i$, which concludes the proof. \[prop\_matroid\_transversal\] Let ${\mathcal{G}} = (G_1 \subsetneq \cdots \subsetneq G_k)$ with ${\left\lvert G_k \right\rvert} \leq n-2$ and $k \geq 1$. Then $M_{{\mathcal{G}}}$ is a transversal matroid. More precisely: $$M_{{\mathcal{G}}} = M[{\mathcal{A}}_{{\mathcal{G}}}] := M\left[ G_1^{\oplus {\left\lvert G_1 \right\rvert}}, G_2^{\oplus {\left\lvert G_2 {\backslash}G_1 \right\rvert}-1},\cdots, G_k^{\oplus {\left\lvert G_k {\backslash}G_{k-1} \right\rvert}-1}, E^{\oplus {\left\lvert G_k^c \right\rvert}-1}\right]\enspace.$$ For this one only needs to verify that the bases of $M[{\mathcal{A}}_{{\mathcal{G}}}]$ are of the form given in Lemma \[lemma\_chain\_bases\], which is obvious. \[remark\_cardinality\_of\_coloops\] Pick any chain ${\mathcal{G}} = (G_1,\dots,G_k)$ with ${\left\lvert G_i \right\rvert} = i-1.$ Then the proposition above tells us that $$M_{{\mathcal{G}}} = M\left[E^{\oplus n-k} \right] = U_{n-k,n}\enspace.$$ So we see that one matroid can have multiple representations as a chain product. However, the transversal presentation given above identifies it uniquely, as we will shortly prove. This tells us that for any two representations $$N = H_{G_1} \cdot \dots \cdot H_{G_k} = H_{G_1'} \cdot \dots \cdot H_{G_k'}\enspace,$$ we must have ${\left\lvert G_i \right\rvert} = {\left\lvert G_i' \right\rvert}$ for all $i = 1,\dots,k$. Also if ${\left\lvert G_i {\backslash}G_{i-1} \right\rvert} > 1$, we actually must have $G_i = G_i'$. \[lemma\_maximal\_presentation\] The presentation ${\mathcal{A}}_{{\mathcal{G}}}$ is maximal. By Remark \[remark\_transversal\_presentations\], maximality is fulfilled if we can show that for all $i = 1,\dots,k$, the matroid $M_i := M[{\mathcal{A}}_{{\mathcal{G}}}]_{\mid G_i^c}$ is coloop-free. We can rewrite Lemma \[lemma\_chain\_bases\] to see that the bases of $M_{{\mathcal{G}}}$ are $$\left\{B \in \binom{[n]}{n-k}; {\left\lvert B \cap G_i^c \right\rvert} \leq {\left\lvert G_i^c \right\rvert} -k + (i - 1) \textnormal{ for all } i = 1,\dots,k\right\}\enspace.$$ In addition, one can easily construct a basis where the above inequality is an equality (i.e., equivalently, ${\left\lvert G_i {\backslash}B \right\rvert} = i-1$ for all $i$). Hence the bases of $M_i$ are all subsets of $G_i^c$ of size ${\left\lvert G_i^c \right\rvert} -k + (i-1)$, so $M_i$ is a uniform matroid of corank $k-i+1 \geq 1$, which is coloop-free. \[prop\_chain\_matroids\_are\_nested\] The set of chain products is equal to the set of loopfree nested matroids. From Proposition \[prop\_matroid\_transversal\] and Theorem \[theorem\_nested\_matroids\] it is obvious that any chain product is a nested matroid. Now let $M := M[A_1,\dots,A_k]$ be a nested matroid and assume without loss of generality that $k = {\textnormal{rank}}(M)$. Since $M$ is loopfree, we must have $A_k = E$. Assume there is a $j$ such that $A_{j+1} {\backslash}A_j = \{x\}$ for some $x \in E$. Then $M_{\mid A_j^c} = M[ \{x\}, A_{j+2}{\backslash}A_j,\dots,A_k {\backslash}A_j]$ has $x$ as a coloop so we can augment the presentation and replace $A_j$ by $A_j \cup \{x\}$ without changing the matroid. Using similar arguments, one can finally assume that $M = M[A_1^{\oplus t_1},\dots,A_k^{\oplus t_k}]$, such that - $\emptyset \subsetneq A_1 \subsetneq \cdots \subsetneq A_k = E$. - ${\left\lvert A_i {\backslash}A_{i-1} \right\rvert} \geq 2$ for all $i > 1$. - $0 < t_1 \leq {\left\lvert A_1 \right\rvert}$. - $0 < t_i < {\left\lvert A_i {\backslash}A_{i-1} \right\rvert}$ for all $i > 1$. - $\sum_{i=1}^k t_i = {\textnormal{rank}}(M)$. One can then easily construct a chain that produces a chain product with this presentation. \[corollary\_corank\_one\] The ring ${\mathbb{M}}_n$ is generated in corank one, more precisely: every matroid can be written as a linear combination of products of corank one matroids. By Theorem \[theorem\_basis\_presentation\], every matroid is a linear combination of nested matroids. By Proposition \[prop\_chain\_matroids\_are\_nested\], every nested matroid is a product of corank one matroids. Vanishing conditions {#subsection_vanishing} -------------------- In this section we study when certain products in ${\mathbb{M}}_n$ vanish. More precisely, we will give necessary and sufficient criteria for a product of a matroid and a nested matroid (written as a chain product) to be zero. \[lemma\_vanishing\] Let $M$ be a matroid on $E$ of rank at least 2. Let $H_G$ be a corank one matroid. Then the following are equivalent: 1. $M \cdot H_G = 0$. 2. $M$ has a rank one flat $F$ such that $F \cup G = E$. Assume $M$ has a rank one flat $F$ such that $F \cup G = E$. Pick any element $f \in F$. As the elements from $F$ are parallel, any basis $B$ of $M$ containing $f$ must fulfill $B \cap F = \{f\}$. In other words, $B {\backslash}\{f\} \subseteq G$. Since any basis $B'$ of $H_G$ containing $f$ also contains all of $G$, we have $B \cap B' = B$, which is too large to be a basis of $M \wedge H_G$. Hence $f$ is a loop of $M \wedge H_G$ and $M \cdot H_G = 0$. Conversely, assume $M$ has no such flat and pick $e \in E$. Then there exists an element $e' \notin G$, such that $\{e,e'\}$ is independent in $M$. Thus there is a basis $B$ of $M$ containing $\{e,e'\}$. $B' := E {\backslash}\{e'\}$ is a basis of $H_G$, so $B \cap B' = B {\backslash}{e'}$ is a basis of $M \wedge H_G$ containing $e$. Hence, $M \wedge H_G$ is loopfree. \[lemma\_flats\_of\_product\] Let $M$ be a matroid of rank at least 2 and $H_G$ a corank one matroid such that $M \cdot H_G \neq 0$. Then the flats of the intersection product are given by: $${\mathcal{F}}(M \cdot H_G) = \{F \in {\mathcal{F}}(M) \cap {\mathcal{F}}(H_G); M/F \cdot H_G/F \neq 0\}.$$ An equivalent formulation is the following: Let $F$ be a flat of both $M$ and $H_G$. Then $F$ is also a flat of $M \cdot H_G$ if and only if one of the following two conditions is met: 1. $F \cup G = E$. 2. There is no flat $F'$ of $M$ that covers $F$ and such that $F' \cup G = E$. As $M' := M \cdot H_G = M \wedge H_G$ is loopfree, a set $F$ is a flat of $M'$ if and only if $M'/F$ is loopfree. But $M'/F = (M \wedge H_G)/F = (M/F) \wedge (H_G/F)$. As $F$ is a flat of both $M$ and $H_G$, both factors are loopfree. This proves the first statement. The second statement follows from the first, Lemma \[lemma\_vanishing\] and the fact that $$H_G / F = \begin{cases} U_{{\left\lvert G {\backslash}F \right\rvert}, G {\backslash}F},&\textnormal{if } F \cup G = E,\\ H_{G {\backslash}F}, &\textnormal{otherwise.} \end{cases}$$ If $F \cup G = E$, then $M/F \cdot H_G/F = M/F \neq 0$, so $F$ is a flat of $M \cdot H_G$. Otherwise, $F$ is a flat if and only if $M/F \cdot H_{G {\backslash}F} \neq 0$. As the rank one flats of $M/F$ are all flats $F' {\backslash}F$, where $F'$ covers $F$ in ${\mathcal{F}}(M)$, this is equivalent to the second condition above. \[lemma\_rank\_down\] Let $M$ be a matroid on $[n]$, $H_G$ a corank one matroid and assume $M \cdot H_G \neq 0$. Let $\emptyset \neq F$ be a flat of both $M$ and $M \cdot H_G$. Then $${\textnormal{rank}}_{M \cdot H_G}(F) = \begin{cases} {\textnormal{rank}}_M(F) -1, &\textnormal{if } F \cup G = E\\ {\textnormal{rank}}_M(F), &\textnormal{otherwise.} \end{cases}$$ We prove this by induction on $s := {\textnormal{rank}}_M(F)$. $s = 1$ is obvious, so assume $s > 1$. As ${\textnormal{rank}}(M \cdot H_G) = {\textnormal{rank}}(M)-1$, it is clear that ${\textnormal{rank}}_{M \cdot H_G}(F) \in \{{\textnormal{rank}}_M(F), {\textnormal{rank}}_M(F)-1\}.$ Hence we want to prove that $${\textnormal{rank}}_{M \cdot H_G}(F) = {\textnormal{rank}}_M(F) -1 \textnormal{ if and only if }F \cup G = E.$$ Assume $F \cup G = E$. Then by Lemma \[lemma\_flats\_of\_product\], any flat $H$ covered by $F$ cannot be a flat of $M \cdot M_G$. Hence ${\textnormal{rank}}_{M \cdot H_G}(F) = {\textnormal{rank}}_M(F) -1$. Conversely, assume ${\textnormal{rank}}_{M \cdot H_G}(F) = {\textnormal{rank}}_M(F) -1$, but $F \cup G \subsetneq E$. Pick any flat $H$ of $M$ that is covered by $F$. If $H$ is a flat of $M \cdot H_G$, then we can use induction to see that ${\textnormal{rank}}_{M \cdot H_G}(F) = {\textnormal{rank}}_{M \cdot H_G}(H) + 1= {\textnormal{rank}}_M(H) + 1 = {\textnormal{rank}}_M(F)$, which is a contradiction. Hence $H$ is not a flat of $M \cdot H_G$. By Lemma \[lemma\_flats\_of\_product\], there must be a flat $H'$ covering $H$ such that $H' \cup G = E$. In particular ${\textnormal{rank}}_M(H') = {\textnormal{rank}}_M(F)$ and flat axioms tell us that $F \cap H' = H$. Denote by $K = \textnormal{cl}_M(F \cup H')$. By semimodularity, we have ${\textnormal{rank}}_M(K) \leq {\textnormal{rank}}_M(F) + {\textnormal{rank}}_M(H) - {\textnormal{rank}}_M(F \cap H') = {\textnormal{rank}}_M(F)+1$. Thus, $K$ is a flat of $M$ covering $F$ such that $K \cup G = E$. But this implies that $F$ is not a flat of $M \cdot H_G$, which is a contradiction. \[prop\_nested\_intersection\_zero\] Let $M$ be a matroid on $[n]$ of rank at least 2 and $M_{{\mathcal{G}}} = H_{G_1} \cdot \dots \cdot H_{G_c}$ a nested matroid with $c < n - {\textnormal{corank}}(M)$. Then $M \cdot M_{{\mathcal{G}}} = 0$ if and only if the following holds: There exists an $i = 1,\dots,c$ and a flat $F$ of $M$ such that 1. ${\textnormal{rank}}_M(F) = c - i + 1$. 2. $F \cup G_i = E$. We prove this by induction on $c$. The case $c = 1$ is a reformulation of Lemma \[lemma\_vanishing\]. Now let $c > 1$. We write $M' := M \cdot H_{G_1}$ and $N' := H_{G_2} \cdot \dots \cdot H_{G_c}$. Hence $N'$ is a nested matroid of corank $c-1$. First assume $M \cdot M_{{\mathcal{G}}} = 0$. Then either $M' = 0$ or $M' \cdot N' = 0$. If $M' = 0$, Lemma \[lemma\_vanishing\] tells us that there exists a rank one flat $F$ of $M$ such that $F \cup G_1 = E$ and we are done. If $M' \neq 0$, but $M' \cdot N' = 0$, we can use induction to see that there exists an $i' = 1,\dots,c-1$ and a flat $F$ of $M'$ such that 1. ${\textnormal{rank}}_{M'}(F) = c-i'$. 2. $F \cup G_{i'+1} = E$. As $M'$ is a quotient of $M$, $F$ is also a flat of $M$. We now distinguish two cases. If ${\textnormal{rank}}_M(F) = {\textnormal{rank}}_{M'}(F) + 1$, then by Lemma \[lemma\_rank\_down\], we must have $F \cup G_1 = E$ and since ${\textnormal{rank}}_M(F) = c-i'+1 \leq c$, we can replace $F$ by any flat of rank $c$ that contains it. If ${\textnormal{rank}}_M(F) = {\textnormal{rank}}_{M'}(F)$, we can pick $i = i'+1$ and the two conditions are fulfilled. Conversely, assume we have a flat $F$ of $M$ and an $i \in \{1,\dots,c\}$ with the required properties. If $i = 1$, then $F$ is a flat of rank one in $M \cdot H_{G_1} \cdot \dots H_{G_{c-1}}$ by Lemmas \[lemma\_flats\_of\_product\] and \[lemma\_rank\_down\]. So $M \cdot M_{{\mathcal{G}}} = 0$ by Lemma \[lemma\_vanishing\]. Hence we assume $i > 1$. First assume that $F$ is not a flat of $M'$. If $F$ is not even a flat of $H_{G_1}$, then we must have $F \cup G_1 = E {\backslash}\{x\}$ for some $x \in E$ by definition of $H_{G_1}$. The properties of the flats of a matroid dictate that there has to be a flat $F'$ of $M$ covering $F$ such that $x \in F'$, so $F' \cup G_1 = E$. If $F$ is a flat of $H_{G_1}$, then by Lemma \[lemma\_flats\_of\_product\], there also exists such a flat $F'$. As ${\textnormal{rank}}_M(F') = {\textnormal{rank}}_M(F) + 1 = c-i+2 \leq c$, we can simply replace $F$ by a larger flat fulfilling the two conditions (1) and (2). Thus we can assume that $F$ is indeed a flat of $M'$. Now we distinguish two cases: If ${\textnormal{rank}}_M(F) = {\textnormal{rank}}_{M'}(F) +1$, then $F \cup G_1 = E$ by Lemma \[lemma\_rank\_down\] and since $G_1 \subseteq G_2$, we can apply induction to see that $M' \cdot N' = 0$. If ${\textnormal{rank}}_M(F) = {\textnormal{rank}}_{M'}(F)$, then we have ${\textnormal{rank}}_{M'}(F) = c -i +1 = (c-1) - (i-1) + 1$ and again we have $M' \cdot N' = 0$ by induction. Poincaré duality and the matroid polytope algebra {#subsection_poincare_duality} ------------------------------------------------- By Theorem \[theorem\_counting\_nested\] and the symmetry of the Eulerian numbers, we already know that the free ${\mathbb{Z}}$-modules ${\mathbb{M}}_{r,n}$ and ${\mathbb{M}}_{n-r+1,n}$ are isomorphic. From a purely matroid-theoretic perspective this might seem somewhat odd, as one would maybe rather have expected ${\mathbb{M}}_{r,n}$ to be isomorphic to ${\mathbb{M}}_{n-r,n}$. However, from a geometric perspective, the statement makes immediate sense, as the corresponding matroid fans have complementary dimensions. We will now see that there is another geometric interpretation of the ring ${\mathbb{M}}_n$, which makes it immediately clear what the isomorphism must be. \[prop\_cohomology\] The ring ${\mathbb{M}}_n$ is isomorphic to the cohomology ring $A^(X(\textnormal{Perm}_n))$ of the toric variety corresponding to the normal fan of the permutohedron of order $n$. By [@fstoricintersection], $A^(X(\textnormal{Perm}_n))$ is the ring of all tropical cycles which are supported on some skeleton of the normal fan of the permutohedron. This normal fan is $B(U_{n,n})$, the fan of all chains, so we see that ${\mathbb{M}}_n$ is in fact a subring of $A^(X(\textnormal{Perm}_n))$. It is a classical fact that the permutohedral variety has Eulerian numbers as Betti numbers, so the claim follows from Theorem \[theorem\_counting\_nested\]. \[theorem\_poincare\] Let $2 \leq n, 1 \leq r \leq n$. The intersection product on ${\mathbb{M}}_n$ induces a perfect pairing $${\mathbb{M}}_{r,n} \times {\mathbb{M}}_{n-r+1,n} \to {\mathbb{M}}_{1,n} \cong {\mathbb{Z}}\enspace,$$ i.e. it induces an isomorphism ${\mathbb{M}}_{r,n} \to \textnormal{Hom}({\mathbb{M}}_{n-r+1,n},{\mathbb{Z}}) \cong {\mathbb{M}}_{n-r+1,n}.$ This is immediate from Proposition \[prop\_cohomology\], as $X(\textnormal{Perm}_n)$ is a smooth and complete toric variety. A more general version of this can be found in [@ahkhodgetheory Theorem 6.19], where Poincaré duality is shown for the ring of cycles supported on skeleta of an arbitrary Bergman fan. It has been shown [@fstoricintersection; @jystableintersection] that the intersection ring $Z_{n-1}^{\textnormal{fan}}$ of all tropical fan cycles in ${\mathbb{R}^{n}/\textbf{1}}$ is isomorphic to McMullen’s polytope algebra $\Pi_{n-1}$ [@mpolytopealgebra] (considered over ${\mathbb{Q}}$). This is the algebra generated by symbols $[P]$ for each polytope in ${\mathbb{R}}^{n-1}$, modulo translations and the identity $[P \cup Q] = [P] + [Q] - [P \cap Q]$ whenever $P \cup Q$ is a polytope. The isomorphism is defined by mapping the class $[P]$ of a polytope to $$\exp(P) = \sum_{i=0}^{n-1} \frac{1}{i!} H_P^i\enspace,$$ where $H_P$ denotes the tropical hypersurface dual to $P$. As a set, this is just the codimension one skeleton of the normal fan of $P$. The weight of a maximal cell of $H_P$ is the lattice length of the dual edge of $P$. $H_P^i$ is the $i$-fold intersection product of the hypersurface. It follows that ${\mathbb{M}}_n^{\mathbb{Q}}:= {\mathbb{M}}_n \otimes_{\mathbb{Z}}{\mathbb{Q}}$ is isomorphic to a subring of $\Pi_{n-1}$. We can identify this subring precisely. For a matroid $M$ (possibly with loops), its matroid polytope* is $$P_M := {\textnormal{conv}}\left\{\sum_{i \in B} e_i;\; B \textnormal{ a basis of }M\right\} \subseteq {\mathbb{R}}^n\enspace.$$ Forgetting the last coordinate is a linear equivalence on $P_M$, so we can consider $[P_M]$ as an element of $\Pi_{n-1}$. Also, the normal fan of $P_M$ has a lineality space containing $\textbf{1}$, so its hypersurface can be seen as a cycle in ${\mathbb{R}^{n}/\textbf{1}}$. \[corollary\_polytope\_algebra\] Under the isomorphism $\exp$, we have $${\mathbb{M}}_n^{\mathbb{Q}}\cong {\mathbb{Q}}[ [P_M]; M \textnormal{ a matroid on }\{1,\dots,n\}]\enspace.$$ Note that $M$ is allowed to have loops. It is easy to see that for any matroid polytope $P_M$, its hypersurface is contained in the codimension one skeleton of the normal fan of the permutohedron. So by Proposition \[prop\_cohomology\] $\exp$ maps $[P_M]$ into ${\mathbb{M}}_n$. It remains to see that $\exp$ is surjective onto ${\mathbb{M}}_n^{\mathbb{Q}}$. By Corollary \[corollary\_corank\_one\], it suffices to show that every corank one matroid $H_G$ is in the image. But the matroid fan of $H_G$ is the hypersurface dual to the polytope ${\textnormal{conv}}\{e_i, i \notin G\}$, which is the matroid polytope of $(H_G)^$. \[remark\_derksen\_fink\] We already mentioned the matroid polytope modules studied by Derksen and Fink in [@dfvaluative]. The additive structure they consider is basically the same as on McMullen’s polytope algebra, except that polytopes differing by a translation are not considered equal. In particular, it follows from Corollary \[corollary\_polytope\_algebra\] that ${\mathbb{M}}_{r,n}$ is a quotient module of their module $P_M(n,r)$. One of their main results is the fact that ${\mathcal{G}}(M)$ is the universal valuative matroid invariant, in particular it is a linear map on $P_M(n,r)$. Hence Theorem \[theorem\_g\_invariant\] would also follow from their result and the fact that ${\mathcal{G}}$ is compatible with translations of polytopes. Outlook {#section_outlook} ======= In this section we outline some interesting questions and connections for further research. Matroid homology ---------------- For each $r \geq 1$ and each $i \in [n]$ there are two natural ${\mathbb{Z}}$-module homomorphisms $d_i: {\mathbb{M}}_{r,n} \to {\mathbb{M}}_{r,n-1}$ and $c_i: {\mathbb{M}}_{r,n} \to {\mathbb{M}}_{r-1,n-1}$ given on matroids by $$\begin{aligned} d_i(M) &:= \begin{cases} M{\backslash}i, &\textnormal{if } i\textnormal{ is not a coloop of }M,\\ 0,&\textnormal{otherwise}. \end{cases}\\ c_i(M) &:= \begin{cases} M/i, &\textnormal{if } {\textnormal{cl}}_M(\{i\}) = \{i\},\\ 0,&\textnormal{otherwise.} \end{cases}\end{aligned}$$ It is not obvious that these give well-defined maps. Tropical geometry comes to the rescue also in this case: Geometrically, $d_i$ corresponds to the push-forward* of tropical cycles along a coordinate projection and $c_i$ is the “intersection product with the hyperplane at $x_i = \infty$”. In both cases it is known that these operations commute with taking sums of cycles (see [@smatroidintersection] for proofs). These operations are defined to be zero whenever the dimension of the result does not match the expected dimension. Let ${\mathbb{M}}= \oplus_{n \geq 0} {\mathbb{M}}_n$. In [@abgwhomology], the authors show that one can define boundary maps on the free abelian group over all matroids using alternating sums of deletions or contractions. It is not hard to see that the same works for the maps $d_i$ and $c_i$, i.e. if we set $$\begin{aligned} \partial_d: &{\mathbb{M}}\to {\mathbb{M}}, M \mapsto \sum (-1)^i d_i(M)\\ \partial_c: &{\mathbb{M}}\to {\mathbb{M}}, M \mapsto \sum (-1)^i c_i(M)\enspace,\end{aligned}$$ then $\partial_d^2 = \partial_c^2 = 0$. This allows us to define homology groups of matroids or minor-closed classed of matroids. Computational experiments suggest that when taking all matroids these homology groups always vanish, which seems not at all obvious. The matroid of matroids and the polytope of matroids ---------------------------------------------------- Identifying each matroid with its indicator vector of chains $v_M$ makes the set of all loopfree matroids of rank $r$ into a matroid. So far, we have proven rather little about this matroid. We know its rank, which is $A_{r-1,n}$. Furthermore it is, by definition, realizable over any field of characteristic zero. Of particular interest are the circuits. More precisely, it would be interesting to understand the kernel of the map $\Phi_{r,n}: {\mathbb{M}}_{r,n}^{\textnormal{free}}\to V_{r,n}$, especially since the total number of loopfree matroids of rank $r$ on $n$ labeled elements is obviously $A_{r-1,n} + \dim \ker \Phi_{r,n}$. If one considers the vector space ${\mathbb{M}}_n^{\mathbb{Q}}$ in the coordinates given by the basis of nested matroids, one can take the convex hull of the points corresponding to matroids. One can show that this is an empty lattice polytope (i.e. it has no interior lattice points), whose vertices are exactly the matroids. Since various matroid invariants define linear maps on ${\mathbb{M}}_n$, as shown in \[subsection\_tutte\], this provides a new approach to open extremality questions, such as the one posed by Bonin and de Mier [@bmlatticeofcyclic] about the maximal possible number of cyclic flats of a matroid. Hence it seems essential to understand the combinatorial structure of this polytope of matroids. Regular subdivisions of matroid polytopes ----------------------------------------- It was suggested in [@ahrrationalequivalence] that there should be an analogue of the polytope algebra which is isomorphic to the intersection ring of all tropical cycles (not just fans). The correct object would likely be an algebra of regular subdivisions of polytopes. One could again consider the subalgebra generated by matroid polytopes and their regular subdivisions and study its tropical counterpart. This might provide an interesting approach to understanding regular subdivisions of matroid polytopes better.	{ "pile_set_name": "ArXiv" }
--- author: - Denis Bashkirov title: 'Quantum Field Theory, Causal structures and Weyl transformations.' --- Physical motivation =================== Let us start with the instrumentalist point of view. Suppose one is given a theory, that is a collection of local fields and the form of the action, for example, but is not told on what Lorentzian background one is placed. To what extent can one reconstruct the metric? In order to more easily illustrate several important points let us be concrete and consider $\phi^4$ theory, although every statement is valid in general, without reference to any Lagrangians. The first step is obvious – to determine the causal structure, one measures commutators of local observables at separate spacetime points. To find the full answer one can measure correlation functions with separate-points insertions of local operators. However, not all information contained in them is ’physical’: for any $c-$function $f(x)$ which is nowhere zero, operators ${\cal O}'(x)=f(x){\cal O}(x)$ are neither worse nor better than operators ${\cal O}(x)$ – these are also local operators. This introduces ambiguity in correlation functions of local operators as devices to measure the Lorentzian metric since it is not clear a priori which of the two local operators corresponds to the local field $\phi(x)$ in the action for the $\phi^4$ theory. Indeed, if one uses a set of local operators $\{{\cal O}_i(x)\}$ to conclude/calculate that the metric is $g_{\mu\nu}$, then an alternative set of operators $\{{\cal O}'_a(x)\}$, local with respect to the first, meaning that $$\begin{aligned} \widehat{\cal O}'_a(x)=\sum_{i}f^i_a(x)\widehat{\cal O}_i(x)+f_i(x)\widehat 1\end{aligned}$$ may find a different Lorentzian metric $g_{\mu\nu}'$. This is because it is not clear beforehand which of the two operators one should identify with the field $\phi(x)$ in the action if, f.e., one considers a $\phi^4$ theory. So the full metric may be ambiguous since it is not directly measurable. As we mentioned in the beginning, there is a part of the metric on which all observers, wether they prefer $\{{\cal O}(x)\}$ or $\{{\cal O}'(x)\}$, would agree: the causal structure: $g_{\mu\nu}$ and $g'_{\mu\nu}$ should produce the same causal structure on the Lorentzian manifold. Indeed, this is a question of wether commutators of local operators at separate points of spacetime are zero or nonzero – this does not change with local fields redefinitions. All these commutators sit in correlations functions and form the invariant under change of descriptions, physical part, of them. For example, the mean value of the commutator of two local operators in a given state is the imaginary part of the time-ordered two-point correlator: $$\begin{aligned} {\langle\Psi\,\|}[{\cal O}(x),\tilde{\cal O}(y)]{\|\Psi\,\rangle}=i\hbox{sign}(x^0-y^0)\hbox{Im}{\langle\Psi\,\|}T{\cal O}(x)\tilde{\cal O}(y){\|\Psi\,\rangle}\end{aligned}$$ So, the metrics $g_{\mu\nu}$ and $g'_{\mu\nu}$ reconstructed from correlators of local operators ${\cal O}_i(x)$ and ${\cal O}_a(x)$ related by local transformations have necessarily the same causal structure. But that means that they are conformally related: $$\begin{aligned} g'(x)=\Omega^2(x)g(x).\end{aligned}$$ Now, on one hand, this is obviously true in Conformal Field Theories. On the other hand, the physical arguments above do not seem to single them out from the class of all Quantum Field Theories. Indeed, in this operational setting it should be true that one should only be able to unambiguously reconstruct the metric only up to a conformal factor. We have seen that there is ambiguity in matching of local operators with local fields which allow the conformal equivalence to exist, but in general it may not be enough. To allow the conformal equivalence exist more generally, one needs to have more ambiguous information specializing a QFT to have at ones disposal – these are parameters or coupling constants which specify a theory. We already know form the conventional RG that those are ambiguous – they ’run’. So, it must be them that change from one ’conformal frame’ frame to another to allow for equivalent descriptions. In other words, to allow the metric to be unambiguous only up to a conformal factor. To put still differently, to guarantee that a QFT is defined not for the set of all Lorentzian metrics but for the set of all causal structures. Of course, this point of view requires that all coupling constants be promoted to functions on spacetime. Not all choices of these coupling functions are allowed, though. For example, in flat Minkowski space, one should not allow the mass function $m^2(x)$ to be negative in some regions of space-time, as this will lead to super-luminal propagation in those region, that is, tho the loss of causality. Thus we are lead to the suggestion that a QFT lives not on Lorentzian manifolds $\{{\cal M}, g\}$, but on the equivalence classes of Lorentzian manifolds $\{{\cal M}, {\cal C}\}$ with the same causal structure ${\cal C}$, or, equivalently on the equivalence classes $\{{\cal M}, g/Weyl\}$ whose representatives are related by Weyl transformations. In practice, one picks a representative (a ’conformal’ frame) with metric $g$ and coupling constants/parameters (now functions on space-time) $\lambda(x)$ and studies correlators of local operators ${\cal O}(x)$. One can equally choose another representative $g'=\Omega^2g$ with a different set of parameters $\lambda'(x)$ and study correlators with this metric and this set of coupling functions. The equivalence of descriptions is in the fact that there will be a choice of local operators ${\cal O}^{\Omega}_a(x)$ in the second frame related in a local way to the operators in the first frame ${\cal O}^{\Omega}_a(x)=\sum_{i}f^i_a(x){\cal O}_i(x)+f_a(x)1$ and $\lambda'(x)=f^{\Omega}(\lambda(x))$ such that triples $(g(x),{\cal O}(x),\lambda(x))$ and $(g'=\Omega^2g,{\cal O}^{\Omega}(x),f^{\Omega}(\lambda(x)))$ will yield the same correlators. The case when there is a mixing with the identity operator in the relation $$\begin{aligned} {\cal O}^{\Omega}_a(x)=\sum_{i}f^i_a(x){\cal O}_i(x)+f_a(x)1\end{aligned}$$ deserves a special mention – this is the case of a conformal anomaly which is discussed in the next section. Let us motivate our suggestion from a complimentary perspective by considering the standard Callan-Symanzik equation.[^1] What is the meaning of the standard Callan-Symanzik equation for correlation functions of ${\cal O}$ in position space in a general QFT? Suppose for simplicity that we consider a local operator which does not have an anomalous dimension. Then the statement of the Callan-Symanzik equation is the following: $$\begin{aligned} <\phi(x)\phi(y)>_{g,c}=\Omega^{-\Delta}\Omega^{-\Delta}<\phi(x)\phi(y)>_{\Omega^2g,c'(c,\Omega)}=F(x,y)\end{aligned}$$ where $c'(c, \Omega)$ is a new value of the coupling constant $c$ obtained by running from the previous value by the RG under the rescaling by $\Omega$. Now try to relate this to an experiment. The correspondence between measuring devices and local fields (which is a computational tool), that is the correspondence between experiment and theory is not given a priori. One needs to match the two, so it is in general, for any QFT, is matching of pairs $(\hbox{parameters},{\cal O})$ (in a fixed state) to pairs $(\hbox{measuring apparatus},\hbox{outcomes})$. In our particular case it is the triple $(g,c,{\cal O})$ that needs to be matched to the measuring apparatus and measurements outcomes. So suppose one measures correlation functions with a measuring device to be $F(x,y)$. Then one theorist says: this matches to measuring the field $\phi$, and so the metric and the coupling constant are $(g,c)$, but another theorist says: no, it corresponds to the field $\phi'=\phi/\Omega^\Delta$, and the metric is $g'=\Omega^2g$, and the coupling constant is $c'$ – those are two interpretations on equal footing, and both interpretations are valid. It is better to recast the (solution of) CS equation into its usual form: $$\begin{aligned} Z(\lambda)^{i_1}_{j_1}...Z(\lambda)^{i_k}_{j_k}<\phi^{j_1}(x_1)...\phi^{j_k}(x_k)>_{\lambda^2g,c_a(\lambda)}=const(\lambda)= <\phi^{i_1}_0(x_1)...\phi^{i_k}_0(x_k)>\end{aligned}$$ Here $\{c_a\}$ is the set of coupling constants, and the RHS is interpreted as mysterious ’bare’ operators. They are actual expectation values of local operators ${\cal O}(x)$ which live in the Hilbert space ${\cal H}_{\cal C}$ and correspond to the concrete measuring device, while $\phi^i(x)$ are local fields which come with each conformal frame, so they part of a frame dependent description. We measure the ’bare’ correlators – expectation values of physical local operators ${\cal O}(x)$ corresponding to local measuring apparatus in some state ${\|\Psi\,\rangle}_{\cal C}$. They are obviously frame-independent. So $$\begin{aligned} <\phi^{i_1}_0(x_1)...\phi^{i_k}_0(x_k)>_{\Psi}={\langle\Psi\,\|}T\widehat{\cal O}^{i_1}(x_1)...\widehat{\cal O}^{i_k}_0(x_k){\|\Psi\,\rangle}\end{aligned}$$ Even if in some conformal frame all $Z_i^j$ become the identity matrix, it doesn’t mean the metric corresponding to this frame is the ’real’ metric – it’s again a matter of convention. One physicist might like the identity matrix, another – some other matrix. The point is the very fact that the expectation value CAN be interpreted differently, and the matter of easiness/convenience is irrelevant in principle (but not for practical calculation, of course). The conclusion is that the standard RG invariance implies that metric itself is not physical – the physical part is only the scaling equivalence class. In a CFT the redundancy is enlarged to the full Weyl group, which means that physical (unambiguous) is only the causal structure. The redundancy cannot be larger than the Weyl group because all matchings agree on wether commutators are zero or not. But is the scaling redundancy the full redundancy in non-conformal QFTs? - First of all, unlike the CFT case, there is no clear physical expalnation/motivation for this redundancy. - In the spirit of locality, it seems natural that the redundancy should work locally. Let us return to the first example of two-point correlator with a single coupling $\lambda$ for simplicity. If it works locally it means that the triples $(g(x),c(x),\phi(x))$ and $(\Omega^2(x)g(x),\lambda^\Omega(x),\phi^\Omega(x))$ should give the same correlator $F(x,y)={\langle\Psi\,\|}T{\cal O}(x){\cal O}(y){\|\Psi\,\rangle}$. Here $c^\Omega(x)$ is a local function $c^\Omega(x)=f(c(x),\Omega(x))$ (and their derivatives) and $\phi^\Omega(x)$ is a local rescaling of $\phi(x)$. But then it is nothing else then the statement of the existence of a local RG flow! Furthermore, now this local redundancy has a physical explanation/motivation: the only unambiguous information that any QFT contains about Lorentzian metric is its causal structure. The rest is ambiguous – this is how the local RG flow appears – it’s just comparison of equivalent descriptions in different ’conformal’ frames. It is the correlators of local operators that are frame independent (for any tuple of spacetime points $(x_1,...,x_k)$ it is just a number when $\mu$ is fixed) and what all frames $(g(x),c(x),\phi(x))$ give. Coupling functions/sources {#coupling-functionssources .unnumbered} ========================== Coupling functions play a double role in QFT. On one hand, a set of coupling functions ${\cal J}=\{J_I(x)\}$ in a fixed conformal frame defines a particular theory for a fixed causal structure. On the other hand they serve as sources for correlation functions of local operators through the relation $$\begin{aligned} {\langle\Psi\,\|}T{\cal O}_1(x_1){\cal O}_2(x_2)...{\cal O}_k(x_k){\cal O}{\|\Psi\,\rangle}_{\{g,J\}}=\frac{\delta^k}{\delta J_1(x_1)\delta J_2(x_2)...\delta J_k(x_k)}{\langle\Psi\,\|}{\cal O}{\|\Psi\,\rangle}_{\{g,J\}}\end{aligned}$$ Here ${\cal O}$ stands for some insertions of some other operators. In particular, an insertion of the stress-tensor corresponds to a variation with respect to the metric. This is why a QFT should be defined on general (or at least generic) Lorentzian manifolds even if at the end of the day one only studies it in Minkowski space. We will have more to say about this in Section 4. A predictive QFT is characterized by a finite set of coupling functions (modulo Weyl equivalence), while every QFT contains infinitely many local operators to each of which there corresponds a source. The Weyl group acts on this infinite set of sources, so a predictive QFT is characterized by the property that there is an ’almost fixed point’ in this infinite-dimensional space – a ’fixed subspace’ of a finite dimension. Furthermore, if there is a true fixed point in this space, then it is natural to make it the origin for the sources, that is, to introduce the convention that all the coupling functions are zero at this point which corresponds to a Conformal Field Theory. Such Conformal Field Theories exist in odd number of dimensions, but strictly speaking there are no such fixed points in even dimension due to the conformal anomaly. More precisely, in odd dimensions there is a conformal anomaly due to spacetime-dependent coupling functions but it can be set to zero for a certain choice of coupling functions (zero coupling functions by the above convention). In even dimensions this cannot be done. So one can have at best an Anomalous Conformal Field Theory. A conformal anomaly is the situation when there is an additive (spacetime-dependent) shift in the coupling function for the identity operator (it can be called ’vacuum energy density’). Due to this it is not possible to fix this coupling function for the entire conformal class, so there is no fixed point, but there is a pretty harmless fixed ’line’ which does not affect correlators at separate points. Due to the double role of coupling functions this is exactly related to the presence of mixing with identity operator discussed in the previous section. It should be stressed that above we considered arbitrary Quantum Field Theories, and not just Conformal Field Theories. Now, states live on causal structures/conformal classes, one just probes/describes them differently (but equivalently) in different conformal frames. So if one considers states in the Schrodinger picture as functionals $$\begin{aligned} \Psi_{\{g,J\}}[\phi(x)],\end{aligned}$$ one has to make sure this functional is not changed under conformal transformation of sources $\{g,J\}$ and fields $\phi(x)$. Units and dimensional analysis in Quantum Field Theory ====================================================== In Quantum Field Theory there is only one unit – the lengths unit, but more often one talks about the mass unit which is just the inverse of length. All operators have definite length (or, equivalently, mass) dimensions: every operator is measured in a power of length which is called its length dimension equal to minus mass dimension. There are two ways to implement dimensional analysis. The first is what one finds in classical physics and in quantum mechanics (of finitely many degrees of freedom), when for a dimensionless quantity $F$ which is a function of dimensionful variables $\{R_i\}$ there is a single formula $$\begin{aligned} F=F(R_i).\end{aligned}$$ Consider for concreteness the case of two variables $R_1$ and $R_2$ with length dimension $1$ and $2$ respectively. Then by dimensional analysis the dimensionless quantity $F$ is a function of the dimensionless ratio $R_1^2/R_2$: $$\begin{aligned} F=F(R_1,R_2)=f(R_1^2/R_2).\end{aligned}$$ This is not the only option to make the dimensional analysis work. The second possibility is that for a dimensionless quantity $F$ which is a function of dimensionful variables $\{R_i\}$ there are infinitely many expressions – one for each choice of units. Let us denote these functions by $F_\xi(R_i)$ - there is one parameter family of functions with $\xi$ running over the real line ${\mathbb R}$. Those are expressions for numerical values of all quantities: one for when one measures in meters, one for centimeters, one for millimeters etc. Each formula describes the relation between numerical values of all quantities. Because in each ’unit frame’ (choice of length unit) all variables are just dimensionless variables, each function $F_\xi(R_i)$ may be more complicated than $F(R_i)$ from the previous option, but since they all express a dimensionless quantity which is independent of choice of units, there must be a relation between $F_\xi(R_i)$ for different $\xi$ which ensures this. Return to the previous example of two dimensionful variables $R_1$ and $R_2$ with length dimension $1$ and $2$ correspondingly. Suppose we measure everything in centimeters and find the following relation between $F,R_1,R_2$ in these units: $$\begin{aligned} F=10R_1R_2.\end{aligned}$$ Next we switch to using millimeters. The numerical values for $R_1$ get multiplied by 10 and for $R_2$ – by 100 while numbers for $F$ stay the same. This means that in millimeters unit the formula has the form $$\begin{aligned} F=10\times10\times 10^2R_1R_2.\end{aligned}$$ The point is that for a chosen unit the formula may take any form, but then for all other choices of units (in all other ’unit frames’)it is fixed. So, in this example $$\begin{aligned} F=\xi R_1R_2\end{aligned}$$ where under change in units $R_1\to\lambda R_1$, $R_2\to\lambda^2 R_2$, $F\to F$, so that $\xi\to\lambda^{-3}\xi$. The parameter $\xi$ transforms as the third power of a mass parameter, so it is convenient to denote it $\mu^3$. Then the formula takes form $$\begin{aligned} F=F(R_1,R_2;\mu)=\mu^3R_1R_2\end{aligned}$$ Now if one treats the parameter $\mu$ as a dummy mass, one is back to the first option of implementing dimensional analysis at the expense of introducing a dummy mass parameter $\mu$. This second option is more general than the first one, obviously, as it allows for more functional dependence on physical variables $\{R_i\}$. Furthermore, it contains the first option since the original formula is reproduced if one requires the dependence on $\mu$ to be trivial. More importantly, it is the second option that is realized in QFT: the introduction of the dummy mass parameter $\mu$ is forced by the renormalization procedure. A finite $\mu$ remains at the end of the day after the theory is made finite – this is the essence of the renormalization procedure. Inconsistency of minimal coupling to geometrical background =========================================================== As was noted above a QFT must be defined on a general Lorentzian manifold in order for variations with respect to the metric produce a stress-tensor. Naively, it seems that given a QFT in flat space there are many ways to put it on curved backgrounds. To be concrete, consider the massive $\phi^4$ scalar theory from [@CCJ] with flat space action $$\begin{aligned} S=\int d^4x(\eta^{\mu\nu}\partial_\mu\phi(x)\partial_\nu\phi(x)+m^2\phi^2(x)+\lambda\phi^4(x)).\end{aligned}$$ It seems that there is a preferred way to put this theory on a curved spacetime – the minimal coupling. One just promotes the Minkowski metric to a general metric and gets $$\begin{aligned} S=\int d^4x\sqrt{g}(g^{\mu\nu}\partial_\mu\phi(x)\partial_\nu\phi(x)+m^2(x)\phi^2(x)+\lambda(x)\phi^4(x)).\end{aligned}$$ According to the suggestion that a QFT ’lives’ on causal structures rather than metrics, this does not correspond to putting a single $\phi^4$ theory on arbitrary curved background: the action does not preserve its form under conformal transformations because of the kinetic term and therefore is not defined on causal structures only. Thus it is putting a family of $\phi^4$ theories (parameterized by the metric $g_{\mu\nu}$ up to a constant scale factor) each in a single curved metric (up to a constant scale factor), and not a single theory in arbitrary background. Thus when one performs a variation with respect to the metric to get an insertion of the stress-tensor, one does not vary the metric for a fixed theory, but performs a variation along a section in the space of $\phi^4$ theories and metrics. Correspondingly, one expects that this stress-tensor is not well-defined even in Minkowski space. Furthermore, one expects that problems come from the trace part of the stress-tensor, as it is this part that corresponds to varying the conformal factor of the metric, and it is the relation between conformal frames which does not work for this minimal coupling. This is exactly the result of [@CCJ] who found that such tensor does not exist because it has infinite matrix elements due to the trace part. The correct way to put this theory on curved backgrounds is to include the conformal coupling to the curvature $$\begin{aligned} S=\int d^4x\sqrt{g}(g^{\mu\nu}\partial_\mu\phi(x)\partial_\nu\phi(x)+\frac{1}{6}R(x)\phi^2(x)+m^2(x)\phi^2(x)+\lambda(x)\phi^4(x))\end{aligned}$$ In this case, one indeed varies metrics for a fixed $\phi^4$ theory. The same authors [@CCJ] showed that the stress-tensor obtained by varying this action (called new improved stress-tensor) has finite matrix elements. Because this direct coupling to curvature does not disappear in local inertial frames, the equivalence principle is violated in our interpretation of QFT in the presence of scalars (like in the Standard Model). A quick remark about scale but not (anomalous) conformal theories is appropriate at this point. Consider the example of the free ${\mathbb R}$ (not $U(1)$) Maxwell theory in three dimensions. It is scale but not conformally invariant unless one introduces monopole operators [@ERN]. On a topologically trivial Lorentzian manifold one can dualize this theory to the theory of free massless scalar. Not including monopole operators allows only minimal coupling of this scalar to background metric which is not consistent. Presumably, on the flat spacetime this will show in some correlation functions with stress-tensor insertions. It is possible that the same problem happens for other cases of scale but not conformal theories. Callan-Symanzik equation ======================== In this section we show that for conformal frames related by spacetime-independent conformal factors the equivalence of descriptions is equivalent to the requirement that separate-points correlators satisfy Callan-Symanzik equation. In fact, this was done 25 years ago by Hugh Osborn [@Osb][^2]. Namely, instead of correlators he considered the generating functional $W[g,J;\mu]$ for the connected correlators. In the presence of massless degrees of freedom this functional is not defined for any background metric and sources – one needs to choose them so that ground state degeneracy is lifted, then take variational derivatives to obtain connected correlators (which always exist) and only after that take the limit of sources/metric to land in the desired vacuum. In this description the double role of coupling functions is obvious. For completeness, here we give a quick recap of how it goes. The requirement of equivalence of descriptions in different conformal frames is the requirement of invariance of $W[g,J;\mu]$ under conformal transformations up to a conformal anomaly: $$\begin{aligned} W[\Omega^2(x)g(x),J^\Omega(x);\mu]=W[g(x),J(x);\mu]+{\cal A}[g(x),J(x);\Omega(x)],\end{aligned}$$ where the conformal anomaly ${\cal A}[g(x),J(x);\Omega(x);\mu]$ is a local functional of its arguments. The dummy mass parameter is not a function on spacetime! If one introduces the source $\Lambda(x)$ (vacuum energy density) for the identity operator and defines $W[g(x),J(x),\Lambda(x);\mu]\equiv W[g(x),J(x);\mu]+\int\Lambda(x)$, the equation can be recast into a manifestly form-preserving form. $$\begin{aligned} W[\Omega^2(x)g(x),J^\Omega(x),\Lambda^\Omega(x);\mu]=W[g(x),J(x),\Lambda(x);\mu],\end{aligned}$$ with $\Lambda^\Omega(x)=\Lambda(x)-a(x)$ where $a(x)$ is the density of the local anomaly functional ${\cal A}[g(x),J(x);\Omega(x)]$. We will use the first option with the explicit anomaly. The second equation expresses the requirement of dimensional analysis: since $W$ is dimensionless, it should not depend on the choice of units. Because of diffeomorphism covariance the change of units affects the arguments of $W$ in the following way: $$\begin{aligned} \mu\to\lambda\mu,\qquad g(x)\to\lambda^{-2}g(x),\qquad J(x)\to\lambda^{m[J]}J(x).\end{aligned}$$ Here $m[J]$ is the mass (inverse scale) dimension of $J$. If at small scales the flow ends in a CFT for which the coupling functions are by convention take to be zero, then mass dimensions of all local operators are just their UV conformal dimensions. Thus $m[J]=\Delta[J]$. So, the second equation which ensures that dimensional analysis works is $$\begin{aligned} W[\lambda^{-2}g(x),\lambda^{\Delta[J]}J(x),\lambda\mu]=W[g(x),J(x),\mu].\end{aligned}$$ Next, for small values of $J(x)$ (in the units of $\mu$)(close to the UV CFT) the action of the Weyl group on the coupling functions are $$\begin{aligned} J^{\Omega(x)}(x)=\Omega^{\Delta[J]}(x)J(x)+\gamma[J(x),\Omega(x)]J,\end{aligned}$$ where the anomalous scaling $\gamma[J,g]$ is of order $J$ (in the case when $\Delta[J]=0$ the first term is absent and the second does not have $J$ multiplying $\gamma$). This corresponds to the anomalous dimension of the corresponding local operator. There is in general a mixing of operators which we omit in order not to clutter the notation. One should stress that the scaling of the metric in the first equation is physical, while in the second equation it is just a change of length unit. Next, one considers a ’diagonal’ section of this physical and dummy scalings so that the explicit expression for the metric stays the same, combines the two equation and arrives at the Callan-Symanzik equation for scale transformations: $$\begin{aligned} (\mu\frac{\partial}{\partial\mu}+\int d^4x\gamma[J(x),\lambda]J\frac{\delta}{\delta J(x)})W[g,J;\mu]={\cal A}[g,J;\lambda]\end{aligned}$$ Taking variational derivatives at separate spacetime points one gets the Callan-Symanzik equation for connected correlation functions. The anomaly contributes only to the 1-point insertion in a CFT, but for any QFT it is zero (for constant rescalings) in flat spacetime. We should stress that although the metric does not appear to be varied in this equation, it actually is just an appearance coming from combining physical variation with a change of length unit. Summary ======= In this paper we suggest the following definition of a Quantum Field Theory: - A QFT is a quantum mechanical system for which for every causal structure ${\cal C}$ of spacetime there correspond a Hilbert space ${\cal H}_{\cal C}$ together with local fields ${\{\cal O}_i\}$ (operators when smeared over a finite open set) acting on it. Elements of the Hilbert space ${\cal H}_{\cal C}$ are states ${\|\Psi\,\rangle}_{\cal C}$. - The set of causal structures on spacetime has a description as the quotient of the set of all lorentzian metrics by the action of the Weyl group. Thus to describe a causal structure one can pick a representative in the equivalence class – a Lorentzian metric and describe a QFT in this ’conformal’ frame. All such descriptions are equivalent (though, of course, some may be easier). - In practice a QFT is described by the set of local (and also higher-dimension) fields together with a parameter – in each conformal frame this parameter is a set of coupling functions $J_A(x)$ on spacetime. All conformal frames are equivalent and describe the same physics. The change of a conformal frame induces the local action of the Weyl group on the space of coupling functions and local fields: the two description $(g(x), J(x), {\cal O}(x))$ and $(\Omega^2(x)g(x),J^{\Omega(x)}(x),{\cal O}^{\Omega(x)}(x))$ of the QFT corresponding to two conformal frames are equivalent – they give the same correlation functions. This has the interpretation of a local RG flow. - The usual RG flow corresponds to comparing descriptions in conformal frames related by a constant conformal factor $\Omega$. - This must be supplemented by additional physical requirements like unitary time evolution in conformal frames which allow global time-like Killing vectors (in those frames it makes sense to take coupling functions to be time-independent), and the choice of coupling functions which do not violate causality (like locally negative mass function $m^2(x)$ in flat conformal frame.) Acknowledgements ================ This research is supported by the Perimeter Institute for Theoretical Physics. Research at the Perimeter Institute is supported by the Government of Canada through Industry Canada and by Province of Ontario through the Ministry of Economic Development and Innovation. We would like to thank the hospitality of the Simons Center for Geometry and Physics where work that has lead to this publication started during the 2015 Simons Summer workshop in mathematics and physics. [99]{} J. Callan, Curtis G., S. R. Coleman, and R. Jackiw, Annals Phys. 59, 42 (1970). S. El-Showk, Y. Nakayama and S. Rychkov, “What Maxwell Theory in $D\lessgtr4$ teaches us about scale and conformal invariance,” Nucl. Phys. B 848, 578 (2011) \[arXiv:1101.5385 \[hep-th\]\]. H. Osborn, Weyl consistency conditions and a local renormalization group equation for general renormalizable field theories, Nucl. Phys. B363 (1991) 486. F. Baume, B. Keren-Zur, R. Rattazzi, and L. Vitale, The local Callan-Symanzik equation: structure and applications, JHEP 08 (2014) 152, \[arXiv:1401.5983\]. [^1]: The form of the Callan-Symanzik equation we consider here is not quite standard: usually one keeps the expression for the metric the same and varies the dummy mass parameter $\mu$, while we keep $\mu$ fixed and vary the metric. These two points of view are equivalent – see section 4. [^2]: For a recent summary and some extension see [@Retal]	{ "pile_set_name": "ArXiv" }
--- author: - 'Laurent Bienvenu, Alexander Shen' title: Algorithmic information theory and martingales --- Introduction ============ What is probability? What is (or should be) the subject of probability theory? How this mathematical theory is (or should be) applied to the “real world”? These questions were debated for centuries, and these discussions go far beyond the scope of our paper. However, there is a clear dividing line between two kinds of different approaches; some of them attempt to define mathematically the notion of an “individual random object” while the others move this notion completely to the grey zone between “pure” probability theory (understood as a part of mathematics) and its practical applications. In practice, almost all mathematicians (and most non-mathematicians), looking at the winning numbers of a lottery for the last year and suddenly noticing that they are all even, will conclude that something wrong happens. The same feeling would arise if (as in the “Rosenkrantz and Guildenstern are dead”, the play by Tom Stoppard) the long sequence of heads appears while tossing a (presumably fair) coin. However, classical probability theory assigns to this sequence (say, $100$ heads) the same probability $2^{-100}$ as to any other sequence and does not try to explain why this sequence looks “non-random” and raises the suspicion. This paradox (sequences with various regularities or symmetries in them appear less random to us, even when each of them is just as probable as any other outcome), occupied probabilists already in the nineteenth century, including Laplace.[^1] However, the attempts to define mathematical notions that somehow capture the intuition of an individual random object (in some idealized way) are not that old. Richard von Mises suggestion (at the beginning of XXth century) was to base probability theory on the notion of the so-called “Kollektiv” (an individual random sequence). These ideas were developed, critically analyzed and made rigorous in 1930s by Wald, Ville and Church (the latter gave a first precise definition of a “random sequence”). In 1960s and 1970s these notions were related to the notion of complexity (amount of information, defined in algorithmic terms), and now different definitions of randomness are well studied in the framework of recursion theory and algorithmic information theory. In this paper we try to describe the main stages of this development and its main achievements from the mathematical viewpoint focusing on the role played by martingales. This paper is based on published sources, discussion at the Dagstuhl meeting (Seminar 06051, 29 January – 3 February 2006; C. Calude, C.P. Schnorr, P. Vitanyi gave talks that were recorded and made available at `http://www.hutter1.net/dagstuhl` by Marcus Hutter) and contributions of Leonid Bassalygo, Cristian Calude, Peter Gács, Leonid Levin, Vladimir A. Uspensky, Vladimir Vovk, Vladimir Vyugin and others. It was initiated by Glenn Shafer whose historical comments about Kolmogorov and Ville became a starting point. (Of course, the people mentioned are not responsible in any way for the authors’ flaws.) Collectives =========== The first well known attempt to define mathematically the notion of an individual random object was done by Richard von Mises in his 1919 paper [@mises-1919]. Then he elaborated his ideas in the book published in 1928 [@mises-wsw]. He also made some clarifying comments is his address delivered on September 11, 1940 at the meeting of the Institute of Mathematical Statistics in Hanover, N.H. (USA) and published in 1941 [@mises-1941; @mises-1941a]. Mises explains that probability theory studies a special class of natural phenomena, like tossing a coin, rolling a dice, or other repetitive experiments. Geometry tries to capture and axiomatize the real-world notion of space; in a similar way probability theory captures and axiomatizes the properties of random phenomena, called “collectives” (German: Kollektiv) in Mises’ paper. Informally speaking, collectives are (according to Mises) plausible sequences of outcomes we can get by performing infinitely many independent trials of some experiment. He formulated two axioms for the notion of collectives. For simplicity, we state them for a collective with two values, e.g., the sequence of heads and tails obtained by coin tossing (where the coin is potentially unbalanced, i.e., the outcome “tails” may appear more (or less) often than “heads”): I. There exists a limit frequency: if $s_N$ is the number of heads among the first $N$ coin tosses, the ratio $s_N/N$ converges to some real $p$ as $N\to\infty$. II. This limit frequency is stable: if we select a subsequence according to some “selection rule”, then the resulting subsequence (if infinite) has the same limit frequency. Axiom I is quite natural: if we want to explain informally what probability is, we say something like “repeat the experiment many times until the frequency of some event (say, head on a coin) becomes almost stable; this stable value is called a probability of the event”. What is the second axiom needed for? Remember that collectives should represent plausible sequences of outcomes of independent trials. Suppose somebody tells you that flipping a coin produced the sequence $$0101010101010101010101010101\ldots$$ where $0$ (heads) and $1$ (tails) alternate. Would you believe this? Probably not. Globally, the limit frequency of $0$ and $1$ in this sequence exists and is equal to $1/2$. But this sequence does not look plausible as a sequence of outcomes, as it presents some highly suspicious regularity. This is where axiom II comes into place: if one selects from this sequence the bits in even positions, one gets a new sequence $$1111111111111111111111111111\ldots$$ in which the frequency of ones is different ($1$ instead of $1/2$). Probability theory, according to Mises, needs to define its subject, and this subject is the properties of collectives and operations that transform collectives into other collectives. Mises uses the following example: take a collective (a sequence of zeros of ones) and cut it into 3-bit groups. Then replace each group by an individual bit according to the majority rule. Probability theory has to find the limit frequency of the resulting sequence if the limit frequency of the original one is known. In his early papers Mises explained in quite informal way which selection rules are allowed: the selection rule should decide whether a term is selected or not, using only the values of the preceding terms but not the value of the term in question. For example, selection rule may select terms whose numbers are prime, or terms that immediately follow heads in the sequence, but not the terms that are heads themselves. The existence of collectives, according to von Mises, is an observation confirmed by our experience, e.g., by thousands of people who invented different systems to beat the casino but all failed in the long run (principle of “ausgeschlossenen Spielsystem”, as Mises said). Clarifications. Wald’s theorem ============================== Of course, Mises’ approach was quite vulnerable from the mathematical viewpoint. What is a selection rule? Do collectives exist at all? Answering these objections, Mises adopted a more formal definition of a selection rule suggested by A. Wald (see, e.g., [@wald] and [@mises-1941]). Assume for simplicity that a sequence is formed by zeros and ones. The selection rule is a total function $s\colon\{0,1\}^\to\{0,1\}$. Here $\{0,1\}^$ is a set of all finite binary strings. Applying selection rule $s$ to an infinite binary sequence $\omega_1\omega_2\ldots$ means that we select all terms $\omega_i$ such that $s(\omega_1\omega_2\ldots\omega_{i-1})=1$; the selected terms are listed in the same order as in the initial sequence. The condition II for a selection rule $s$ says that for a collective the selected subsequence either should be finite or should have the same limit frequency as the entire sequence. Therefore we get a formal definition of a collective as soon we fix some class of selection rules. The evident problem here is that if we consider all selection rules of the described type, collectives (non-trivial ones, with limit frequency not equal to $0$ or $1$) do not exist. Indeed. for every set $S$ of natural numbers there exists a selection rule that selects the terms $\omega_i$ for $i\in S$ (the function $s$ depends only on the length of its argument). Using for a given sequence $\omega_1\omega_2\ldots$ the set $S$ of all $i$ such that $\omega_i=0$ (or $\omega_i=1$), we get a contradiction. Wald [@wald][^2] provided a kind of solution for this problem. He proved that for any countable family of selection rules and for any $p\in(0,1)$ there is a continuum of sequences that satisfy the axioms I (with limit frequency $p$) and II for this class of selection rules. Today this statement looks almost trivial: indeed, if a given selection rule $s$ is applied to a $B_p$-randomly chosen sequence, where $B_p$ is Bernoulli distribution with parameter $p$, the selected subsequence has the same distribution $B_p$, so the Strong Law of Large Numbers guarantees that the set of sequences that do not satisfy II for a given $s$ has $B_p$-measure zero; the countable union of null sets is a null set and its complement has continuum cardinality. However, Wald wanted to give a constructive proof of this result; Theorem V ([@wald], p. 49) says that if a “konstruktiv definiertes abzählebare System von Auswahlvorschriften” is given, “so kann man Kollektiv $\langle\ldots\rangle$ konstruktiv definieren” (if a countable system of selection rules is defined constructively, there exists a constructively defined collective). Note that there is no formal definition of “constructive” objects in Wald’s paper; he just provides a construction of a collective that refers to selection rules (uses them as an oracle, in modern terminology). The collective sequence is constructed inductively. Let us explain the idea of the construction in a simple case when only finitely many selection rules $s_1,\ldots,s_n$ are considered and sequence of zeros and ones has limit frequency $1/2$. At the $i$th step of the construction we should decide whether $\omega_i$ is $0$ or $1$. At that time we already know which of the rules $s_1,\ldots,s_n$ would include $\omega_i$ in the selected subsequence. In other terms, we know a Boolean vector of length $n$. The entire sequence (that we have to construct) would be therefore split into $2^n$ subsequences that correspond to $2^n$ values of this Boolean vector. Now the main idea: each of these $2^n$ sequences should be $0101010\ldots$ (zeros and ones alternate starting with zero). This determines the sequence $\omega$ uniquely. Since $\omega$ is a mixture of $2^n$ sequences that have limit frequency $1/2$, the entire sequence $\omega$ has the same limit frequency. And if we apply selection rule $s_i$ to $\omega$, we get a mixture of $2^{n-1}$ of these subsequences (corresponding to $2^{n-1}$ Boolean vectors where $s_i$ is playing). Each sequence has limit frequency $1/2$, and their mixture has therefore the same limit frequency. In fact the construction for countably many selection rules is quite similar: we just have to add new rules one by one when the sequence is so long that the boundary effects cannot destroy the limit frequency. In fact Wald proves more: he considers not only the two-element set $\{0,1\}$, but any finite set (Theorem I, p. 45). Then he considers the case of infinite set $M$ (Theorem II–IV, pp. 45–47; we do not go into details here, but to get a reasonable definition of a collective for infinite $M$ one should either consider countable $M$ or a restricted class of events). Theorems V–VI (p. 49) observe that the resulting collectives are “constructive”. Based on Wald’s results, Mises [@mises-1941] concludes that the notion of colletive can be studied without contractictions: we can consider all the selection rules we want to use and their combinations; though we do not know them in advance, one may reasonably assume that they form a finite or countable set and therefore collectives (with respect to this set) do exist. Wald’s results show, in a sense, that the requirements I and II are not too strong. But other objections to the notion of collective, raised by Ville in his book [@ville-etude], say that these requirements are too weak: not only collectives exist, but one can construct some collective in the sense of Mises’ definition that does not look random. Ville’s objections. Martingales {#martingale} =============================== Let us explain Ville’s objections. The requirement II can be reformulated in terms of games as follows. (For simplicity we consider the case when limit frequency is $1/2$.) A player comes into a casino where a coin is tossed infinitely many times, and can (for each tossing) decide to make a bet or to skip it depending on the results of a previous tossings (according to the selection rule she has in mind). Her initial credit is \$0, and she is allowed to incur arbitrarily large debts. All bets are for the same amount of money, say \$1, which the player loses or doubles, depending on whether her guess was correct or not. Let $c_N$ be the player’s capital after $N$ games. The player wins (after infinitely many games) if she makes infinitely many bets and the ratio $c_N/N$ does not converge to zero. (This game deviates from the original idea of a selection rule: instead of just choosing of a subsequence we are allowed also to reverse some of the terms chosen. However, this gives an equivalent definition since we may consider separately the “positively” and “negatively” chosen terms; if both subsequences have limit frequencies $1/2$, the ratio $c_N/N$ does converge to $0$. Note also that this definition assumes that the coin is fair.) We have reformulated Mises’ definition in terms of a game, but this game looks rather unnatural. Yes, for a “really random coin” we would expect that $c_N/N$ converges to $0$ (at least after we learned the strong law of large numbers). But is it the only thing we would expect? Imagine, for example, that $c_N$ is always positive and goes slowly but steadily to infinity, so $c_N/N\to 0$ but $c_N\to+\infty$. This would mean that the player manages to make arbitrarily large amounts of money without incurring debts. In that case, would we agree with the assumption that she is playing with a fair coin? Ville suggested a different kind of gambling games, which are much more natural. In his games we come to the casino with some fixed amount of money (say, \$1) and can use it (in whole or in part) for betting, but cannot go negative. In other terms, if we have $s$ before the next game, we can bet any amount $s'\le s$ on zero or one. If our guess is incorrect, the money is lost, and our capital becomes $s-s'$, otherwise the money is doubled, and our capital is then $s+s'$. Mathematically such a strategy is represented by a function $m$ whose arguments are finite binary strings and values are non-negative reals. The value $m(\omega_1\ldots\omega_n)$ is our capital after we have played $n$ times getting outcomes $\omega_1,\ldots,\omega_n$; the value $m(\Lambda)$ (where $\Lambda$ denotes the string of length zero) is the initial capital, which we assume to be positive. The rules of the game dictate that $$m(x)=\frac{m(x0)+m(x1)}{2} \eqno()$$ Here $x$ is some binary string (representing some moment in the game), $x0$ and $x1$ are obtained by adding $0$ or $1$ to $x$ and correspond to two possible outcomes in the next round. The requirement says that $m(x)$ is the average between two possibilities, i.e., our possible gain and loss are balanced. Ville used the name martingale* for functions that have property $()$. (One may also allow the martingales to have negative values, but we use only non-negative martingales in the sequel.) A martingale $m$ (i.e., the player that uses corresponding strategy) wins* against a sequence $\omega_1\omega_2\ldots$ if the values $m(\omega_1\omega_2\ldots\omega_n)$ are unbounded. Now we can switch from Mises’ selection rules to martingales and say that a sequence $\omega=\omega_1\omega_2\ldots$ is a collective (in a new sense) if all martingales from some (countable) family do not win against $\omega$. To support this change in the class of games, Ville notes that: - Martingales provide a generalization of Mises’ games (with limit frequency $1/2$): for any selection rule one can construct a martingale that wins against every sequence that does not satisfy axiom II when this selection rule is applied. - The notion of martingale matches well the notion of a null set (set of measure $0$) used in classical probability theory: for every martingale $m$, the set of all sequences against which $m$ wins is a null set (has measure $0$) according to the uniform Bernoulli distribution. - The reverse statement is also true: for every null subset $X\subset\{0,1\}^\infty$ there exists a martingale $m$ that wins against every element of $X$. (Together with the strong law of large number this implies the first statement in the list). (The proofs are quite natural: first we prove the finite versions of these results saying that (1) the probability to transform initial capital $1$ into some $C$ during $N$ games does not exceed $1/C$; (2) for every $N$ and for every set of $N$-bit sequences that contains $\varepsilon$-fraction of all sequences of length $N$, there is a martingale that wins $1/\varepsilon$ on every sequence from this set.) Martingales have some other nice properties. One may ask why our winning condition says that martingale is unbounded: isn’t it more natural to require that its values tend to $+\infty$ (a strong winning condition)? The answer is that it does not matter much, as the following simple observation shows: for every martingale $m$ there exist another martingale $m'$ that strongly wins against a sequence $\omega$ if $m$ wins against $\omega$. (The martingale $m'$ should save part of the capital when the capital reaches some bound and use only the remaining part for playing, waiting until it has enough to save again, etc.) Another nice property is the possibility of combining martingales: if $m_i$ are arbitrary martingales, the weighted sum $\sum_i \alpha_i m_i$ (where $\alpha_i$ are some positive reals with sum $1$) is a martingale that wins against a sequence $\omega$ if and only if at least one of $m_i$ wins against $\omega$. (Recall that we consider only non-negative martingales.) Ville’s example =============== The arguments above may convince you that martingales have more nice properties than just selection rules.[^3] But is this difference essential? If we switch from selection rules to martingales, do we get stronger requirements for random sequences (collectives)? Ville showed that it is indeed the case, proving the following result. > For any countable family $\mathcal{S}$ of selection rules there exists a sequence $\omega$ that satisfies requirement II (with limit $1/2$) when rules from $\mathcal{S}$ are used but every prefix of $\omega$ has at least as many zeros as ones ([@ville-etude], p. 63, Remarque). (In fact, Ville proved more; Theorem 4, p. 55, provides also some bounds for the speed of convergency.) This proof raises a historical question. In fact, Ville’s argument is very close to Wald’s argument used in [@wald]: the sequence is splitted into subsequences and inductive construction is performed; Wald does not discuss the one-sided convergence explicitly, but it is obtained in a straightforward way as a byproduct of Wald’s conctruction. Indeed, let us say that a sequence is “biased” if every prefix has at least as many zeros as ones (frequency of ones does not exceed $1/2$). If we merge biased sequences, the result is also a biased sequence; note also that the sequence $01010101\ldots$ is biased. However, Ville does not mention this similarity (though Wald’s paper is mentioned many times in Ville’s book and the existence result is quoted with reference to Wald). It is especially strange since the explanations given in Wald’s paper are quite clear — probably more clear than Ville’s argument, which is written in a rather technical way. May be this heavy technical style of Ville’s paper was the reason why other authors prefer to give their own reconstruction of the proof instead of following the details of Ville’s paper (see, e.g., [@lieb] and references within). More about Ville’s example ========================== Establishing the difference between selection-based and martingale-based definitions of randomess, Ville also showed that there is a martingale that wins against every “biased” sequence (a sequence whose prefixes have more zeros than ones). This is a consequence of the law of iterated logarithm; it implies that the set of all biased sequences has measure zero, so we can use the results mentioned in Section \[martingale\]. However, let us provide a simple direct construction of such a martingale just for illustration. Let $\omega$ be a binary sequence; let $d_n$ be the difference between the numbers of zeros and ones in $n$-bit prefix of $\omega$. We assume that the difference $d_n$ is always non-negative. The limit $d=\liminf d_n$ is then also non-negative; it can be finite or $+\infty$. It is easy to construct a martingale that wins against any biased sequence with ${d=+\infty}$. Imagine that you come into a casino knowing in advance that (1) the number of ones never exceeds the number of zeros and (2) the difference between them tends to infinity. How can you become infinitely rich? Just bet a fixed amount (not exceeding the initial capital) at every step. The condition (1) guarantees that you will never go negative and always have enough money to bet; the condition (2) guarantees that your capital tends to infinity. Now assume that the casino sequence is biased and $d$ is finite. How can you win then? In this case the difference goes below $d$ only finitely many times, and starting from some time $T$ it is at least $d$ being equal to $d$ infinitely many times. A conclusion: if you see (after the initial period of length $T$) that the difference is $d$, you know that the next coin tossing provides a head, so you bet on it with no risk. This allows you to become infinitely rich if you know $d$ and $T$ in advance. So we have one martingale $m$ that wins against any biased sequence with $d=+\infty$ and a countable family $m_{d,T}$ of martingales who win against sequences with given $d$ and $T$. As we have noted, this countable family of martingales can be combined into one martingale. There is a large variety of possible interpretation of Ville’s example. One can treat this example as a failure of Mises’ approach: it shows that requirements I and II that guarantee frequency stability (and therefore establish the very notion of probability) are not strong enough to provide a satisfactory definition of a random sequence (collective): a martingale cannot win against a “real coin” but still can win against a collective formally defined in terms of selection rules. One may say also that axioms I and II do not pretend to capture all properties of “really random” sequence but only some of them needed to define the notion of probability, and therefore the Mises’ notion of collective can be considered as an upper bound for the class of “really random” sequences. Finally, one can say also that replacing selection rules by a stronger martingale requirement, we harmonize the idea of a random sequence with the measure-theoretic understanding of laws of probability theory, therefore giving new life to Mises’ approach and getting a better notion of randomness. It would be interesting to reconstruct the real attitude of Mises, Ville, Frechet and others; however, this again goes far beyond the scope of the article. Let us note nevertheless that the only place where Ville is mentioned in [@mises-1964] has nothing to do with martingales (it is a paper on game theory). Things become even more complicated when we try to interpret Mises’ remark in [@mises-1919] when he says: “Solange man etwa nur die Zahlen $1$–$10000$ betrachtet, bietet die Anordnung der Ziffern an der 5. Stelle \[in the table of logarithms\] tatsächlich das ungefähre Bild eines empirisches Kollektivs und man kann auch die Sätze der Wahrscheinlichkeitsrechnung näherungsweise darauf anwenden.” This quote shows that for him (at least at that moment) the behavior of the 5th decimal digit in the table of logarithms of integers $1$–$10000$ looks like “empirical collective” and this sequence satisfies the laws of probability theory to a certain extent (while for bigger numbers the regularities show up). Note that logarithms are computable, so there exists a computable selection rule that selects only zeros from this sequence. One may speculate that Mises had in mind some notion of “pseudorandom” sequence that satisfies the axiom II only for simple enough selection rules, but this remark remains isolated in his paper and it is hard to say what he really meant. Church definition of randomness =============================== Approximately at the same time, in 1930s, a theory of computable functions was developed by Kleene, Church, Turing and others. It provided a very natural class of selection rules: computable rules, where the function $s\colon\{0,1\}^\to\{0,1\}$ is a total computable function. This class contains almost all rules we can think of; it also has nice closure properties needed to prove theorem about collectives. For example, it is closed under composition, and this can be used to prove that a sequence obtained from a collective by a selection rule is again a collective. This step (combining recursion theory with Mises’ approach) was done in 1940 by Church [@church-random]: he called a sequence random if it has limiting frequency and, moreover, any computable selection rule produces either finite sequence or a sequence with the same limit frequency. In fact, Church could do the same with Ville’s definition and define random sequences using computable martingales. But probably he did not realize the importance of martingales. More details about the evolution of the randomness notion from Mises to Church can be found in a historical survey of Martin-Löf [@martin-lof-1969]. An intermission =============== In the 1940s and 1950s the notion of an individual random sequence did not attract much attention. At that time the measure-theoretic approach to probability theory became gradually more and more popular (and, in particular, the notion of martingale was embedded into the framework of measure theory). Another important change during these 20 years was the development of the theory of computation. In 1930s theory of computation appeared as a kind of exotic thing developed by logicians that is using strange tools like recursive functions (with quite unnatural definition), $\lambda$-calculus (even more peculiar definition) or fictional devices called “Turing machines”. But after twenty years the notion of a computer program became quite familiar; many mathematicians played with computers (i.e., programmed them — computer games for dummies were almost unknown at that time) as a part of their job or just for fun. This prepared a next step in the development of randomness notion when the connections with the complexity (incompressibility) was understood. Complexity and randomness in 1960s ================================== Recall the question we started with: why does the long sequence of zeros (heads) look suspicious while the other sequence of the same length (having the same probability $2^{-n}$ according to the classical theory) looks OK? What is the difference between these two sequences? Now, when the notion of computer program became familiar, the difference between them is evident: the first sequence (zeros) can be generated by a short program while the other one (non-suspicious) cannot. So there is no surprise that the ideas of complexity of a finite object (defined as the length of a shortest program that generates this object) were developed independently in different places and by different people. This kind of complexity is often called description* complexity, as opposed to computation complexity, since we ignore the time needed to generate an object and look only at the length of the generating program. There were other (not related to randomness) reasons to consider description complexity. One of these reasons was the quantitative analysis of undecidability. “Undecidable algorithmic problems were discovered in many fields, including algorithms theory, mathematical logic, algebra, analysis, topology and mathematical linguistics. Their essential property is their generality: we look for an algorithm that can be applied to every object from some infinite class and always gives a correct answer. This general formulation makes the question not very practical. A practical requirement is that algorithm works for every object from some finite, though probably very large, class. On the other hand, the algorithm itself should be practical. $\langle\ldots\rangle$ Algorithm is some instruction, and it is natural to require that this instruction is not too long, since we need to invent this algorithm… So an algorithmic problem could be unsolvable in some practical sense even if we restrict inputs to some finite set” (A.A. Markov [@markov-1967], p. 161; this paper provides proofs for the results announced in [@markov-1964]) Note also that the idea of measuring the complexity of a message as the length of its shortest “encoding” was quite familiar due to Shannon information theory (though the encodings considered there are very restricted). Earlier (in [@solomonoff-1964a; @solomonoff-1964b]; these papers are based on technical reports that go back to 1960 and 1962) R. Solomonoff considered similar notions in the context of inductive inference (somebody gives us a long sequence; we want to know what is the reasonable way to predict the next term of this sequence knowing the preceding terms). G. Chaitin [@chaitin-home] tells that entering a Bronx High School of Science (in 1962) he wrote an essay where the idea of randomness as an absence of short description was mentioned; later, in 1965, after his first year in City College, he wrote a paper that was submitted to the Journal of the ACM and finally published in two parts [@chaitin-1966; @chaitin-1969]. In [@chaitin-1966] he defines a complexity measure of a binary string in terms of the size of a Turing machine; in [@chaitin-1969] the complexity is defined in more general terms (in the same way as in Kolmogorov paper [@kolmogorov-1965], see below).[^4] L.A. Levin [@levin-interview; @levin-memoir] tells that being a student of a high school for gifted children in Kiev (USSR, now Ukraine) in 1963/4, he was thinking about the length of the shortest arithmetic predicate that is provable for a single value of its parameter but did not know how make this definition invariant (how to make the complexity independent of the specific formalization of arithmetics). Next year (1964/1965) he moved to Moscow where a special boarding school for gifted children was founded by A. Kolmogorov, and told about this idea to A. Sossinsky who was at that time a teacher in this school. Sossinsky asked Kolmogorov and Kolmogorov replied that in one of his forthcoming papers this question was answered.[^5] This was the paper [@kolmogorov-1965] that soon became the main reference for the definition of complexity; now the complexity defined as the length of the shortest program is often called “Kolmogorov complexity”. The paper was called “Three approaches to the quantitative definition of information”, and one of the approaches (the algorithmic one) defined the complexity of a binary string as the length of the shortest program producing it, assuming the programming language is optimal, and proves the existence of such an optimal language (for the technical details see the paper or any of the tutorials on Kolmogorov complexity, e.g., [@shen-notes]). This Kolmogorov paper had several historical reasons to become most popular (among many expositions of the same ideas, including the above mentioned). It was the first publication where the rigorous definition of complexity was given and universality theorem was proved. (This was done also in the second part of Chaitin’s article submitted in November 1965, after Kolmogorov’s publication, and published only in 1969. Solomonoff’s papers did not contain an explicit definition of complexity.) Second, Kolmogorov was famous as one of the greatest mathematicians of his time, and therefore his papers attracted a lot of attention. And being one of the founders of probability theory, he has a clear vision of the role that complexity can play in the foundations of probability theory (in the definition of individual random object and in information theory). So his paper was concise and well written. [^6] Therefore it is no wonder that among many people who came to very close ideas, Kolmogorov got the most attention.[^7] The introduction of the complexity notion allowed to identify randomness (for finite bit strings and fair coin) with incompressibility. One should have in mind, however, that one cannot hope to draw a sharp dividing line between random and non-random strings of a given finite length, and the complexity function $K(x)$ is defined up to a $O(1)$ term, so, strictly speaking, only asymptotic statements are possible. Martin-Löf definition of randomness =================================== To obtain such a sharp borderline one needs to consider infinite sequences. A natural idea: to define randomness of an infinite sequence in terms of complexity of its prefixes. The first attempt was to say that a sequence $\omega_1\omega_2\ldots$ is random if $K(\omega_1\ldots\omega_n)$ is maximal up to a constant, i.e., $$K(\omega_1\ldots\omega_n)= n+O(1).$$ But Martin-Löf[^8] found that it is not possible (sequence with this property do not exist). Taking this difficulty into account, Martin-Löf tried a different approach and gave a definition of a random sequence based on effectively null sets, making it more measure-theoretic. The idea of this approach can be explained as follows. Let us define a random bit sequence (for simplicity we consider only the case of a fair coin) as a sequence that satisfies all probability laws. And probability law is a property of sequences that is true for almost all sequences, i.e., for all sequences outside some null set. Finally, a subset $X$ of the Cantor space $\{0,1\}^\infty$ (of all infinite binary sequences) is a null set if its uniform measure is $0$ (equivalent formulation: if for every $\varepsilon>0$ there exists an infinite sequence of intervals that covers $X$ whose total measure is at most $\varepsilon$). The problem with this definition is that random bit sequences defined in this way do not exist at all. Indeed, for every sequence $\alpha$ the singleton $\{\alpha\}$ is a null set, so its complement $\{0,1\}^\infty\setminus\{\alpha\}$ can be considered as a probability law, and $\alpha$ does not satisfy this law. Martin-Löf pointed out that if we restrict ourselves to effectively null sets, this plan becomes quite reasonable. A set $X$ is an effectively null set if there exists an algorithm that (given positive rational $\varepsilon$) generates a sequence of intervals that cover $X$ and have total measure at most $\varepsilon$. (Replacing algorithms with arbitrary functions, we get a classical definition of null sets.) It is easy to see that the union of all effectively null sets is a null set, since there are only countably many algorithms. Therefore random sequences (defined as sequences that do not belong to any effectively null set) exist and the set of random sequences has measure $1$. Moreover, Martin-Löf have proved that the union of all effectively null sets is an effectively null set (in other terms, there exists the largest effectively null set). This maximal set consists of all non-random sequences. A set $X$ is effectively null if and only if $X$ is a subset of this maximal effectively null set, i.e., $X$ does not contain any random sequence. We can formulate this in the following way. Let $P$ be some property of binary sequences. Then the statements $P(\alpha)$ is true for every random sequence $\alpha$ and the set of sequences $\alpha$ that do not satisfy $P$ is an effectively null set are equivalent in the word “random” in understood in Martin-Löf sense. This is nice because people often say, for example, that “for a random sequence $\alpha$ the limit frequency is equal to $1/2$” (the strong law of large numbers) having in mind that the set of sequences that do not have this property is a null set. Now this sentence can be understood literally (if a null set is an effective null set, which is true in most cases). Martin-Löf published this definition in 1966 in [@martin-lof-definition]). His results were also covered by a detailed survey paper [@zvonkin-levin]. written by two Kolmogorov’s young colleagues, Leonid Levin and Alexander Zvonkin (by Kolmogorov’s initiative; Kolmogorov carefully reviewed this paper once it was finished and suggested many corrections). This survey included Martin-Löf results as well as other results about complexity and randomness obtained by the Kolmogorov school in Moscow. In particular, a proof of the symmetry of information (an important result obtained independently by Levin and Kolmogorov) was included there. [^9] Martin-Löf definition of randomness at first seems to be purely measure-theoretic, it has nothing to do with selection rules, martingales, and complexity. However, it turned out to be closely related to these notions, and it was soon found by different authors. Randomness and martingales: Schnorr =================================== During the next decade (1965–1975; recall that Kolmogorov published his definition of complexity in 1965 and Martin-Löf published his definition of randomness in 1966) a lot of work was done by different authors who provided missing links between complexity, randomness and games (martingales). One of these authors was C.P. Schnorr. As he tells [@schnorr-talk], after finishing his Ph.D. he was looking for new topics. Martin-Löf gave a course in Erlangen, and the lecture notes of this course were distributed. So this field become known in Germany, Schnorr heard a talk about complexity and randomness and became interested. He wrote several papers and then a book in Lecture Notes in Mathematics series [@schnorr-ln] based on his 1970 lectures (the book is in German; it contains references to his other papers, including [@schnorr-unified] where many of the results from the book are presented in English). His habilitation was based on the results obtained in these papers. In this book for the first time the notion of martingale was used in connection with algorithmic randomness.[^10] Schnorr defined a class of computable (berechenbare) and lower semicomputable (subberechenbare) martingales. A function $f$ (arguments are strings, values are reals) is called computable if there is an algorithm that computes the values of $f$ with any given precision: given $x$ and positive rational $\varepsilon$, the algorithm computes some rational $\varepsilon$-approximation to $f(x)$. A function is lower semicomputable if there is an algorithm that, given $x$, generates all rational numbers that are less than $x$. It is easy to see that $f$ is computable if and only if both $f$ and $-f$ are lower semicomputable. Schnorr then proved that a sequence is Martin-Löf random if and only if no semicomputable martingale wins against it, thus providing a criterion of Martin-Löf randomness in terms of martingales. (A technical remark: note that the initial capital can be non-computable in our setting.) Schnorr, however, was not satisfied with this notion (lower semicomputability). He found it rather counter-intuitive: there is no evident reason why we should generate approximations from below (but not above) to martingale values. So he thought that this class of martingales is too broad and, therefore, the corresponding class of sequences is too narrow. So he called Martin-Löf random sequences “hyperzufällig” (“hyperrandom”; this name is not in use now). He proved that there exists a sequence that wins against all computable martingales but is not Martin-Löf random. Schnorr also defined the notion of lower semicomputable supermartingale. A function $m$ is a supermartingale if it satisfies the supermartingale inequality, $$m(x)\ge \frac{m(x0)+m(x1)}{2}.$$ In game terms this means that player is allowed to throw away her money during the game. Schnorr proved that lower semicomputable supermartingales can be used for Martin-Löf randomness criterion in place of martingales. Trying to find a better definition of randomness, Schnorr considered a smaller class of effectively null sets (now called sometimes “Schnorr null sets”). As we have said, for an effectively null set $X$ there exists an algorithm that given $\varepsilon>0$ generates a sequence of intervals that cover $X$ and have total measure at most $\varepsilon$. Schnorr introduced a stronger requirement: this total measure should be equal to $\varepsilon$. (This sounds a bit artificial; more natural equivalent definition asks for a computably converging series of the length of covering intervals.) The sequences that are outside all Schnorr null sets are called “zufällig” (now they are sometimes called “Schnorr random” sequences). Schnorr proved that this is indeed a broader class of sequences than “hyperzufällig” (Martin-Löf random). He also proved a criterion in terms of computable martingales: a sequence is zufällig if and only if no computable martingale “computably wins” on it (“computably wins” means that there exists a non-decreasing unbounded computable function $h(n)$ such that the player’s capital after $n$ steps is greater than $h(n)$ for infinitely many $n$). Schnorr’s papers and book contain a lot of other interesting things which were developed much later. For example, he considers how fast player’s capital increases during the game and proves that if a sequence does not satisfy the strong law of large numbers, then there exists a computable martingale that wins exponentially fast against it (much later, in 2000s, the growth of martingales was explored farther in connection to the notions of effective dimension). As Schnorr explains, one of his goals was to approach the notion of “pseudorandomness”. Sometimes even a sequence generated by an algorithm looks similar to a random one; such sequences may be used when the source of physical randomness is unavailable and sometimes are called “pseudorandom”, though this term may have different more or less precise meanings. One of the possible approaches to this phenomenon is that a “pseudorandom” object may have a short description, but the time needed for the decompressing algorithm to process this description is unreasonably large.[^11] So Schnorr considers also complexity with bounded resources in his book. Supermartingales and semimeasures ================================= Schnorr’s lower semicomputable supermartingales are closely related to other notion that appeared in Zvonkin and Levin’s 1970 paper [@zvonkin-levin], the notion of a semicomputable semimeasure. It is easy to see that martingale (as defined above) is just a ratio of two measures on the Cantor space: an arbitrary one and the uniform one. More formally, let $Q$ be any measure on Cantor space and let $P$ be the uniform Bernoulli measure. Then the ratio $Q(I_x)/P(I_x)$, where $I_x$ is the interval rooted at binary string $x$ (the set of all extensions of $x$), is a martingale. Moreover, every martingale can be represented in this way. The supermartingales correspond in the same way to objects that Levin called “semimeasures”. A semimeasure is a measure on the set $\Sigma$ of all finite and infinite binary sequences. Let $\Sigma_x$ be the set of all extensions (finite and infinite) of a binary string $x$. Then $\Sigma_x=\Sigma_{x0}\cup \Sigma_{x1}\cup\{x\}$. If $Q$ is a measure on $\Sigma$, the inequality $$Q(\Sigma_x)\ge Q(\Sigma_{x0})+Q(\Sigma_{x1})$$ holds; moreover, any non-negative real-valued function $q$ on finite strings that satisfies the inequality $q(x)\ge q(x0)+q(x1)$, determines a measure on $\Sigma$. The difference between both sides of this inequality is the probability of the finite string $x$. A semimeasure is lower semicomputable if the function $x\mapsto q(x)=Q(\Sigma_x)$ is lower semicomputable. Lower semicomputable semimeasures are considered in [@zvonkin-levin]; Levin proved that they can be equivalently defined as output distributions of probabilistic machines that have no input, use internal fair coin and generate their output sequentially (bit by bit). Levin proved also that there exists a maximal lower semicomputable semimeasure (universal semimeasure, sometimes called a priori probability on the binary tree). This notion can be also considered as a formalization of Solomonoff’s ideas. The connection between semimeasures and supermartingales: supermartingales can be defined as fractions where the numerator is a semimeasure and denominator is the uniform Bernoulli measure (similar to the description of martingales as fractions of two measures). Lower semicomputable semimeasures correspond to lower semicomputable supermartingales. This representation of (semi)martingales as ratios can be easily generalized to other probability distributions, e.g., to the case of a non-symmetric coin. If $P$ is the distribution declared by the game organizers (now not necessarily uniform), then in the “fair” game the player’s capital is a $P$-martingale, i.e., the ratio $Q/P$ where $Q$ is some measure. (The notion of martingale with respect to a non-uniform measure was also considered by Schnorr in [@schnorr-ln].) In a similar way $P$-supermartingales (that allow the player to discard some money at each step) can be defined as ratios $Q/P$ where $Q$ is a semimeasure. This implies, for example, that for any computable measure $P$ there exists a maximal lower semicomputable $P$-supermartingale: it is the ratio $A/P$ where $A$ is the a priori probability (the largest lower semicomputable semimeasure). The last observation provides a connection between maximal $P$-supermartingales for different $P$; as Levin points in one of the letters to Kolmogorov (see the Appendix) the advantage of the a priori probability notion is that the same notion can be compared to different measures. When switching from (semi)measures to (super)martingales one object (the a priori probability) is transformed into a family of seemingly different objects (maximal lower semicomputable supermartingales with respect to different computable measures). However, a natural goal: “to obtain a criterion of randomness (for infinite sequences) in terms of complexity of their prefixes” (the idea to relate complexity and randomness was present already in the 1965 Kolmogorov publication [@kolmogorov-1965]) was not achieved either in Zvonkin and Levin paper or in Schnorr’s book. This was done few years later when new versions of complexity (monotone and prefix complexities) appeared. Prefix complexity ================= Prefix complexity was introduced by Levin and Chaitin. Since the introduction of prefix complexity sometimes becomes a source of unnecessary controversy, some historical clarifications would be useful here. To put the story short, the first publications where (1) the prefix complexity was defined in terms of self-delimiting codes and as the logarithm of the maximal lower semicomputable converging series, and (2) the claim that these definitions coincide was made (without proofs), are [@levin-1974; @gacs-1974]. These publications appeared in 1974 in Russian; English translations of these two papers were published in 1976 and 1975 respectively (see [@gacs-review]); the logarithm of the maximal lower semicomputable converging series (but not the self-delimiting descriptions) was considered also in unpublished thesis of Levin in 1971.[^12] In 1970 paper [@zvonkin-levin] an a priori probability (on a binary tree, as defined in this paper) of a sequence $0^n1$ is considered (last paragraph on p. 107) and some properties of this quantity are proved, though no name is given for it; this quantity coincides with a maximal lower semicomputable converging series (up to $O(1)$ factor, as usual). At the same time Chaitin independently came to the same two definitions (self-delimited complexity and logarithm of probability) in [@chaitin-1975]; this paper, submitted in 1974, contained, among other results, the first published proof of their equivalence. (See more about the history of this paper below.) The prefix complexity, as we have said, can be defined in different ways. The first approach defines prefix complexity of $x$ as the length of the shortest program that produces $x$, but the programming language must satisfy an additional requirement. In Levin’s paper [@levin-1974] this requirement is formulated as follows: if a bit string $p$ considered as a program produces some output $x$, then its extensions either produce the same $x$ or do not produce anything. The 1974 paper refers for details to Gacs’ paper of the same year [@gacs-1974][^13] and to other Levin’s paper (then unpublished; it was published only in 1976 [@levin-1976]). In Chaitin’s paper mentioned above [@chaitin-1975][^14] a slightly different requirement is used: if a bit string $p$ considered as a program outputs $x$, then none of $p$’s extension could produce any output. Both restrictions reflect the intuitive idea of a self-delimiting program (that does not contain an end-marker; the machine should be able to find out when the program ends) though in technically different ways. Another way to define prefix complexity uses probabilities; as we have mentioned, it appeared in Levin’s thesis (1971) that remained unpublished. Consider the lower semicomputable series of non-negative reals with sum at most $1$ ($\sum p_n\le 1$ where $p_n\ge 0$ and the function $n\mapsto p_n$ is lower semicomputable). These series correspond to machines that use internal fair coin to produce some integer (or, may be, do not produce anything) if we let $p_n$ be the probability of output $n$. Among those series there exists a maximal one (up to $O(1)$ factor). It is called a priori probability on integers (and is closely related to the a priori probability on bit strings considered in Zvonkin–Levin paper [@zvonkin-levin]: a priori probability of a bit string $0^n1$ coincides with the a priori probability of integer $n$ up to $O(1)$ factor). A very important property of these notions: minus binary logarithm of an a priori probability equals prefix complexity (up to $O(1)$ additive term). This property is mentioned without proof both in [@gacs-1974] and [@levin-1974]; the proof was published for the first time in [@chaitin-1975]. This proof implies also that two version of prefix-free requirements mentioned above lead to the same complexity function (up to $O(1)$ additive term). Another advantage of prefix complexity, also discovered independently by Levin (the proof, attributed to Levin, is published in [@gacs-1974]) and Chaitin (the proof is published in [@chaitin-1975]) is a more precise (up to $O(1)$-term) formula for the complexity of a pair in terms of conditional complexities. This formula is an improvement of the symmetry of information theorem that was earlier proved for plain complexity with bigger (logarithmic) error terms by Kolmogorov and Levin. Randomness criterion: Schnorr and Levin ======================================= It was soon understood by Schnorr and Levin that the original goal of describing randomness in terms of complexity can be achieved if one changes a bit the definition of complexity making it monotonic in some sense. Schnorr suggested such a modification in his talk at 4th STOC in 1972 [@schnorr-1972]. The idea of the modification was to take into account that prefixes of a sequence are not separate binary strings but prefixes of one infinite sequence. As Schnorr puts it ([@schnorr-1972], pp. 168–169), “it has already been observed that there must be some difference in the concept of regularity of finite objects which do not involve a direction (for instance a natural number) and the concept of regularity of infinite sequences (as well as finite subsequences \[prefixes\] of an infinite sequence) where a natural direction is involved. For example, he who wants to understand a book will not read it backwards, since the comments or facts which are given in his first part will help him to understand subsequent chapters (this means they help him to find regularities in the rest of the book). Hence anyone who tries to detect regularities in a process (for example an infinite sequences or an extremely long finite sequence) proceeds in the direction of the process. Regularities that have ever been found in an initial segment of the process are regularities for ever. Our main argument is that the interpretation of a process (for example to measure his complexity) is a process itself that proceeds in the same direction.”[^15] Then he gives a formal definition of monotone complexity, called “process complexity” in his paper, and notes that “basic properties of processes have been developed independently in \[5\] and \[8\]” (i.e., [@schnorr-ln] and [@zvonkin-levin] in our list; note that none of these two publications includes a definition of monotone/process complexity). Using his definition, Schnorr proves that a sequence in Martin-Löf random if and only if its $n$-bit prefix has monotone complexity $n+O(1)$. Levin [@levin-monotone-1973] proves essentially the same result using a slightly different version of the monotone complexity (used also in subsequent paper of Schnorr [@schnorr-1977]). Levin also notes that the same proof works for the so-called “a priori complexity”, the minus logarithm of the a priori probability on the binary tree. This statement is equivalent to Schnorr’s characterization of randomness in terms of semicomputable supermartingales (though Levin does not say anything about martingales). Chaitin in [@chaitin-1975] suggested prefix complexity as a tool to define randomness. He calls an infinite sequence $\omega_1\omega_2\ldots$ random if there exists $c$ such that $$H(\omega_1\ldots\omega_n)\ge n-c$$ for all $n$ (he used letter $H$ to denote prefix complexity; Levin used $KP$; now the letter $K$ is most often used), and writes: “C.P. Schnorr (private communication) has shown that this complexity-based definition of a random infinite string and P. Martin-Löf statistical definition of this concept are equivalent”. As Schnorr remembers in his talk [@schnorr-talk], “I knew the first paper of Chaitin that has been published one year later after the Kolmogorov’s 1965 paper but it was the next paper which really made Chaitin also one of the basic investigators of complexity. This was a paper on self-delimiting or prefix-free descriptions and this was published in 1975 in the Journal of the ACM. In fact I was a referee of this paper and I think Chaitin knew this because I’ve sent my personal comments and suggestions to him and he used them”. Lower semicomputable random reals ================================= One more result about randomness in [@chaitin-1975] is an example of a lower semicomputable random real number, now well known as “Chaitin’s $\Omega$ number”. It is related to a philosophical question: can we specify somehow an individual random sequence? One would expect at first the negative answer: if a sequence has some description that defines it uniquely, how can we treat it as random? This negative answer is supported by the (evident) result: a computable sequence is not Martin-Löf random (for the case of a fair coin, i.e., the uniform Bernoulli distribution). However, if we do not insist that description is an algorithm that computes our sequence and let it be less direct, the answer becomes positive. Indeed, in [@zvonkin-levin] the following result attributed to Martin-Löf is stated (Theorem 4.5): there exists a $\Sigma^0_2$-sequence that is Martin-Löf random. This means that there exist a decidable property $R(n,p,q)$ of three natural numbers such that the sequence $\omega$ defined by equivalence $$\omega_n = 1 \ \Leftrightarrow\ \exists p\,\forall q\, R(n,p,q)$$ is Martin-Löf random. This provides an example of an individual explicitly described (though in a non-constructive way) random sequence. The example of a random $\Sigma_2^0$-sequence appears also in Theorem 4.3 in Chaitin’s 1975 paper [@chaitin-1975], but Chaitin went farther in this direction. He noticed that a Martin-Löf random sequence can be a binary representation of a lower semicomputable real number. Speaking about random reals, we identify real numbers in the interval $(0,1)$ with their binary representations. (The collisions like $0.0011111\ldots=0.0100000\ldots$ do not matter since this can happen only for non-random sequences.) Recall that a real number $x$ is lower semicomputable if there is an algorithm that enumerates all rational numbers less than $x$. (Equivalent definition: if $x$ is a limit of an increasing computable sequence of rational numbers.) It is easy to see that all lower semicomputable reals $x\in(0,1)$ have binary representations in $\Sigma^0_2$ but the reverse statement is not true. This alone wouldn’t make Chaitin’s example of lower semicomputable random real so popular. In fact, Section 4.4 of [@zvonkin-levin] (proof ot Theorem 4.5 mentioned above) already constructs a specific example of a random real, i.e., the smallest real outside an effective open set of small measure that covers all non-random reals. Zvonkin and Levin used the language of binary sequences, not reals (which makes the description a bit more tedious) and did not mention explicitly the lower semicomputability (which follows immediately from the construction). But the main reason why Chaitin’s example became so famous is in the form of the description. Chaitin’s lower semicomputable real $\Omega$ has simple and intuitive meaning: it is the probability that the universal machine used in the definition of prefix complexity terminates on a randomly chosen program. This could create an impression that we really have a random real “in our hands”: this is the probability of the event “the universal machine terminates on random input”.[^16] Subsequent achievements ======================= The study of randomness as a mathematical object had clearly a philosophical motivation related to the foundations of probability theory. However, the mathematical theory has its own logic of development: answering some philosophically motivated questions, it introduces new notions and new questions related to these notions. So the mathematical theory of randomness (and related algorithmic information theory) became a rich mathematical subject. In the last decade it attracted a lot of attention from the recursion theorists who used advanced techniques developed in recursion theory to understand the randomness definitions better. For example, they looked at one of the first definitions of randomness (from Kolmogorov’s papers) and proved that it coincides with Martin-Löf randomness relativized to $\mathbf{0}'$-oracle [@nies; @miller]. The other thread that has some philosophical and historical interest is related to non-monotonic selection rules and martingales. In Mises definition the terms of the sequence are revealed in some fixed order (time order, if we look at casino’s example). He never explicitly mentioned other possibilities (though he sometimes writes about data whose ordering is not clear, like statistical data about deaths used by an insurance company). When he was forced to provide a formal definition of a selection rule, this monotonicity is explicitly present in the definition. However, one can consider other examples that motivate non-monotonic selection. Imagine that casino prepares random bits and write them on cards which are then placed on a table (so that bits are invisible). The player is then allowed to look at the cards in any order and also make bets (before the card is turned). Imagine that she manages to win systematically; does it implies that the sequence is not random? As D. Loveland [@loveland-1966] explains this: “Consider the following “practical” situation. A manufacturer produces very cheaply and quickly some item which has a large fluctuation in life expectancy from item to item, with the fluctuation passing through a threshold of acceptance. The producer would naturally wish to cull out the unaccepted items but (it is presumed) cannot test the item to be used for life expectancy without destroying it. He must then look for “systematic fluctuations” in the process so that he can select the items to be used based on the knowledge of the process including knowledge of tested items then ineligible for use. If the process were random in the aforementioned sense, then no system of testing previously manufactured items would indicate whether the next item manufactured should be chosen for use or whether one should choose, rather, some future item after more testing. However, suppose the manufacturer numbers each item consecutively as it is produced and allows it to fall it into a bin from which items are drawn to be tested or selected for use. Then he may test higher numbered items before digging down in the bin to select a specific item for use.” Earlier the same extension was suggested by Kolmogorov in a footnote in his paper [@kolmogorov-1963]. It led to many interesting questions. For example, how complex should be prefixes of a sequence that is random in the sense of Mises–Church definition and in this extended Mises–Kolmogorov–Loveland definition? Kolmogorov claimed [@kolmogorov-1968-69] that in both cases complexity could be logarithmic, but later An. Muchnik has shown that it is not the case (see [@uspensky-semenov-shen]) for Mises–Kolmogorov–Loveland randomness (while for Mises–Church randomness Kolmogorov was right). Many other interesting results are obtained but their description goes far beyond the scope of this paper. Concluding remarks ================== Remember that Mises’ initial reason to consider collectives was the desire to explain what probability is and why and how the mathematical probability theory can be applied to the real world. The question “why” is rather philosophical one, but one can try to answer to second part, “how”, and describe the current best practice. Here is an attempt to provide such a description taken from [@uspensky-semenov-shen; @shen]. “The application of probability theory has two stages. At the first stage we try to estimate the concordance between some statistical hypothesis and experimental results. The rule “the actual occurrence of an event to which a certain statistical hypothesis attributes a small probability is an argument against this hypothesis” (Polya [@polya], Vol. II, Ch. XIV, part 7, p. 76), it seems, could be made more correct if we are allowed to consider only “simply described” events. It is clear that the event “$1000$ tails appeared” can be described more simply that the event “a sequence $A$ appeared” where $A$ is a “random” sequence of $1000$ heads and tails (these two events have the same probability). This difference may explain why our reactions to these events (we have in mind the hypothesis of a fair coin) are so different. To clarify the notion of a “simply described event” the notion of complexity of the constructive object (introduced by Kolmogorov) may be useful. Let us assume that we have already chosen a statistical hypothesis concordant (as we think) with the result of observations. Then we come to the second stage and derive some conclusions from the hypothesis chosen. Here we have to admit that probability theory makes no predictions but can only recommend something: if the probability (computed on the basis of the statistical hypothesis) or an event $A$ is greater than the probability of an event $B$, then the possibility of the event $A$ must be taken into consideration to a greater extent than the possibility of the event $B$. One can conclude that events with very small probabilities may be ignored. Borel [@borel-1913] writes “…Fewer than a million people live in Paris. Newspapers daily inform us about the strange events or accidents that happen to some of them. Our life would be impossible if we were afraid of all adventures we read about. So one can say that from a practical viewpoint we can ignore events with probability less that one millionth… Often trying to avoid something bad we are confronted with even worse… To avoid this we must know well the probabilities of different events” (Russian ed., pp. 159–160). Sometimes the criterion for selection of a statistical hypothesis and the rule for its application are united in the statement “events with small probabilities do not happen”. For example, Borel writes “One must not be afraid to use the word “certainty” to designate a probability that is sufficiently close to $1$.” ([@borel-1950], Russian ed., p. 7). But we prefer to distinguish between these two stages, because at the first stage the existence of a simple description of an event with small probability is important, and at the second stated it seems unimportant. (We can expect, however, that events interesting to us have simple descriptions because of their interest.)” This description (which, we believe, still describes adequately the current best practice of probability theory application) uses the notions of algorithmic information theory only once (when describing when we reject a statistical hypothesis), but this use seems to be important. Let us note also that this description shows that quantum mechanics does not make a real difference compared to probability theory and statistical mechanics: we just replace “small probability” by “small amplitude” in the scheme described. (However, to provide a foundation for the measurement procedure, one should prove a quantum counterpart for the law of large numbers: the amplitude of the event “measured frequency of some outcome diverges significantly from the square of the assumed amplitude of this outcome” is small.) More detailed discussion can be found in [@shen-reachability]. Appendix A: Abstracts of Kolmogorov’s talks {#appendix-a-abstracts-of-kolmogorovs-talks .unnumbered} =========================================== Some talks at the meetings of Moscow Mathematical Society have short abstracts published in the journal “Успехи математических наук” (Uspekhi matematicheckikh nauk, partially translated as “Russian mathemathical surveys”; these abstracts were not translated). Here we reproduce abstracts of three talks given by A.N. Kolmogorov devoted to algorithmic information theory (translated by Leonid Levin). I. \[vol. 23, no. 2, March-April 1968\]. 1\. A.N. Kolmogorov, “Several theorems about algorithmic entropy and algorithmic amount of information”. Algorithmic approach to the foundations of information theory and probability theory was not developed far in several years from its appearance since some questions raised at the very start remained unanswered. Now the situation has changed somewhat. In particular, it is ascertained that the decomposition of entropy $H(x,y)\sim H(x)+H(y\|x)$ and the formula $J(x\|y)\sim J(y\|x)$ hold in algorithmic concept only with accuracy $O([\log H(x,y)])$ (Levin, Kolmogorov). Stated earlier cardinal distinction of algorithmic definition of a Bernoulli sequence (a simplest collective) from the definition of Mises-Church is concretized in the form of a theorem: there exist Bernoulli (in the sense of Mises-Church) sequences $x=(x_1,x_2,...)$ with density of ones $p=\frac 1 2$, with initial segments of entropy (“complexity”) $H(x^n)=H(x_1,x_2,...,x_n)=O(\log n)$ (Kolmogorov). For understanding of the talk an intuitive, not formal, familiarity with the concept of a computable function suffices. (Moscow Mathematical Society meeting, October 31, 1967) II. \[vol. 27, no. 2, 1972\] [1]{}. A.N. Kolmogorov. “Complexity of specifying and complexity of constructing mathematical objects”. 1. Organizing machine computations requires dealing with evaluation of (a) complexity of programs, (b) the size of memory used, (c) duration of computation. The talk describes a group of works that consider similar concepts in a more abstract manner. 2. It was noticed in 1964-1965 that the minimal length $K(x)$ of binary representation of a program specifying construction of an object $x$ can be defined invariantly up to an additive constant (Solomonoff, A.N. Kolmogorov). This permitted using the concept of [definition complexity]{} $K(x)$ of constructive mathematical objects as a base for a new approach to foundations of information theory (A.N. Kolmogorov, Levin) and probability theory (A.N. Kolmogorov, Martin-Löf, Schnorr, Levin). 3. Such characteristics as “required memory volume,” or “required duration of work” are harder to free of technical peculiarities of special machine types. But some results may already be extracted from axiomatic “machine-independent” theory of broad class of similar characteristics (Blum, 1967). Let $\Pi(p)$ be a characteristic of “construction complexity” of the object $x=A(p)$ by a program $p$, and $\Lambda(p)$ denotes the length of program $p$. The formula $K^n\Pi(x)= \inf(\Lambda(p): x=A(p), \Pi(p)=n)$ defines “$n$-complexity of definition” of object $x$ (for unsatisfiable condition the $\inf$ is considered infinite). 4. Barzdin’s Theorem on the complexity $K(M_\alpha)$ of prefixes $M_\alpha$ of an enumerable set of natural numbers (1968) and results of Barzdin, Kanovich, and Petri on corresponding complexities $K^n\Pi(M_\alpha)$, are of general mathematical interest, as they shed some new light on the role of extending previously used formalizations in the development of mathematics. The survey of the state of this circle of problems was given in the form free from cumbersome technical apparatus. (Moscow Mathematical Society meeting, November 23, 1971) III. \[Vol. 29,. no. 4 (155), 1974\] 1\. A.N. Kolmogorov. “Complexity of algorithms and objective definition of randomness”. To each constructive object corresponds a function $\Phi_x(k)$ of a natural number $k$ – the log of minimal cardinality of $x$-containing sets that allow definitions of complexity at most $k$. If the element $x$ itself allows a simple definition, then the function $\Phi$ drops to $1$ even for small $k$. Lacking such a definition, the element is “random” in a negative sense. But it is positively “probabilistically random” only when function $\Phi$, having taken the value $\Phi_0$ at a relatively small $k=k_0$, then changes approximately as $\Phi(k)=\Phi_0-(k-k_0)$. (Moscow Mathematical Society meeting, April 16, 1974) Appendix B. Levin’s letters to Kolmogorov {#appendix-b.-levins-letters-to-kolmogorov .unnumbered} ========================================= These letters do not have dates but were written after submission of [@zvonkin-levin] in August 1970 and before Kolmogorov went (in January 1971) to the oceanographic expedition (“Dmitry Mendeleev” ship). Copies provided by L. Levin (and translated by A. Shen). I. {#i. .unnumbered} -- Dear Andrei Nikolaevich! Few days ago I’ve obtained a result that I like a lot. May be it could be useful to you if you work on these topics while traveling on the ship. This result gives a formulation for the foundations of probability theory different from Martin-Löf. I think it is closer to your initial idea about the relation between complexity and randomness and is much clearer from the philosophical point of view (as, e.g., \[Yu. T.\] Medvedev says). Martin-Löf considered (for an arbitrary computable measure $P$) an algorithm that studies a given sequence and finds more and more deviation from $P$-randomness hypothesis. Such an algorithm should be $P$-consistent, i.e., find deviations of size $m$ only for sequences in a set that has measure at most $2^{-m}$. It is evident that a number $m$ produced by such an algorithm on input string $x$ should be between $0$ and $-\log_2 P(x)$. Let us consider the complementary value $(-\log_2 P(x))-m$ and call it the “complementary test” (the consistency requirement can be easily reformulated for complementary tests). Theorem. The logarithm of a priori probability $-\log_2 R(x)$ is a $P$-consistent complementary test for every measure $P$ and has the usual algorithmic properties. Let me remind you that by a priori probability I mean the universal semicomputable measure introduced in our article with Zvonkin. \[See [@zvonkin-levin].\] It is shown there that it \[its minus logarithm\] is numerically close to complexity. Let us consider a specific computable measure $P$. Compared to the universal Martin-Löf test $f$ (specific to a given measure $P$) our test is not optimal up to an additive constant, but is asymptotically optimal. Namely, if the universal Martin-Löf test finds a deviation $m$, our test finds a deviation at least $m-2\log_2 m - c$. Therefore, the class of random infinite banry sequences remains the same. Now look how nice it fits the philosophy. We say that a hypothesis “$x$ appeared randomly according to measure $P$” can be rejected with certainty $m$ if the measure $P$ is much less consistent with the appearence of $x$ than a priori probability (this means simply that $P(x) < R(x)/2^m$. This gives a law of probability theory that is violated with probability at most $2^{-m}$. Its violation can be established effectively since $R$ is \[lower\] semicomputable \[=enumerable from below\]. But if this law holds, all other laws of probability theory \[i.e., all Martin-Löf tests\] hold, too. The drawback is that it gives a bit smaller value of randomness deficiency (only $m-2\log_2 m -c$ instead of $m$), but this is a price for the universality (arbitrary probability distribution). The connection with complexity is provided because $-\log_2 R(x)$ almost coincides with complexity of $x$. Now this connection does not depend on measure. It is worth noting that the universal semicomputable measure has many interesting applications besides the above mentioned. You know its application to the analysis of randomized algorithms. Also it is ofter useful in proofs (e.g., in the proof of J.T.Schwartz’ hypothesis regarding the complexity of almost all trajectories of dynamic systems). Once I used this measure to construct a definition of intuitionistic validity. All this show that it is a rather natural quantity. L. II. {#ii. .unnumbered} --- Dear Andrei Nikolaevich! I would like to show that plain complexity does not work if we want to provide an exact definition of randomness, even for a finite case. For the uniform distribution on strings of fixed length $n$ the randomness deficiency is defined as $n$ minus complexity. For a non-uniform distribution length is replaced by minus the logarithm of probability. It turns out that even for a distribution on a finite set the randomness deficiency could be high on a set of large measure. Example. Let $$P(x)=\begin{cases} 2^{-(l(x)+100)}, \text{ if } l(x)\le 2^{100};\\ 0, \text{ if } l(x)>2^{100}. \end{cases}$$ Then $\|\log_2 P(x)\| - K(x)$ exceeds $100$ for all strings $x$. A similar example can be constructed for strings of some fixed length (by adding zero prefixes). The violation could be of logarithmic order. Let me show you how to sharpen the definition of complexity to get an exact result (both for finite and infinite sequences). Definitions. Let $A$ be a monotone algorithm, i.e., for every $x$ and every $y$ that is a prefix of $x$, if $A(x)$ is defined, then $A(y)$ is defined too and $A(y)$ is a prefix of $A(x)$. Let us define $$KM_A(x) = \begin{cases} \min\ l(p) \colon x \text{ is a prefix of } A(p);\\ \infty, \text{ if there is no such } p \end{cases}$$ The complexity with respect to an optimal algorithm is denoted by $KM(x)$. Let $P(x)$ be a computable distribution on the Cantor space $\Omega$, i.e., $P(x)$ is the measure of the set $\Gamma_x$ of all infinite extensions of $x$. Theorem 1. $$KM(x) \le \|\log_2 P(x)\| + O(1);$$ Theorem 2. $$KM((\omega)_n) = \|\log_2 P((\omega)_n)\| + O(1)$$ for $P$-almost all $\omega$; here $(\omega)_n$ stands for $n$-bit prefix of $\omega$. Moreover, the probability that the randomness deficiency exceeds $m$ for some prefix is bounded by $2^{-m}$. Theorem 3. The sequences $\omega$ such that $$KM((\omega)_n) = \|\log_2 P((\omega)_n)\| + O(1);$$ satisfy all laws of probability theory (all Martin-Löf tests). Let me use this occasion to tell you the results from my talk in the laboratory \[of statistical methods in Moscow State University\]: why one can omit non-computable tests (i.e., tests not definable without a strong language). For this we need do improve the definition of complexity once more. The plain complexity $K(x)$ has the following property: Remark. Let $A_i$ be an effectively given sequence of algorithms such that $$K_{A_{i+1}}(x) \le K_{A_i(x)}$$ for all $i$ and $x$. Then there exists an algorithm $A_0$ such that $$K_{A_0}(x) = 1 + \min_i K_{A_i}(x).$$ Unfortunately, it seems that $KM(x)$ does not have this property. This can be corrected easily. Let $A_i$ be an effective sequence of monotone algorithms with finite domain (provided as tables) such that $$KM_{A_{i+1}}(x) \le KM_{A_i(x)}$$ for all $i$ and $x$. Let us define then $$\overline{KM}_{A_i}(x) = \min_i KM_{A_i}(x).$$ Among all sequences $A_i$ there exists an optimal one, and the compexity with respect to this optimal sequence is denoted by $\overline{KM}(x)$. This complexity coincides with the logarithm of an universal semicomputable semimeasure \[=a priori probability on the binary tree\]. Theorem 4. $\overline{KM}(x)$ is a minimal semicomputable function that makes Theorem $2$ true. Therefore no further improvements of $\overline{KM}$ are possible. Now consider the language \[=set\] of all functions computable with a fixed noncomputable sequence \[oracle\] $\alpha$. Assume that $\alpha$ is complicated enough, so this set contains the characteristic function of a universal enumerable set \[$\mathbf{0}'$\]. We can define then a relativized \[“языковую” in the Russian original\] complexity $\overline{KM}_\alpha(x)$ replacing algorithms by algorithms with oracle $\alpha$, i.e., functions from this language. Definition. A sequence $\omega$ is called normal if $$\overline{KM}((\omega)_n) = \overline{KM}_\alpha((\omega)_n)+O(1).$$ For a finite sequence $\omega_n$ we define the “normality deficiency” as $$\overline{KM}(\omega_n) - \overline{KM}_\alpha(\omega_n).$$ Theorem 5. A sequence obtained by an algorithm from a normal sequence is normal itself. Theorem 6. Let $P$ be a probability distribution that is defined (in a natural encoding) by a normal sequence. Then $P$-almost every sequence is normal. This theorem exhibits a law of probability theory that says that a random process cannot produce a non-normal sequence unless the probability distribution itself is not normal. This is a much more general law than standard laws of probability theory since it does not depend on the distribution. Moreover, Theorem 5 shows that this law is not restricted to probability theory and can be considered as a univeral law of nature: Thesis. Every sequence that appears in reality (finite or infinite) has normality deficiency that does not exceed the complexity of the description (in a natural language) of how it is physically produced, or its location etc. It turns out that this normality law (that can be regarded as not confined in probability theory) and the law corresponding to the universal computable test together imply any law of probability theory (not necessary computable) that can be described in the language. Namely,the following result holds: Theorem 7. Let $P$ be a computable probability distribution. If a sequence $\omega$ is normal and passes the universal computable $P$-test, then $\omega$ passes any test defined in our language (i.e., every test computable with oracle $\alpha$). Note that for every set of measure $0$ there exists a test (not necessary computable) that rejects all its elements. Let us give one more iunteresting result that shows that all normal sequences have similar structure. Theorem 8. Every normal sequence can be obtained by an algorithm from a sequence that is random with respect to the uniform distribution. III. {#iii. .unnumbered} ---- (This letter has no salutation. Levin recalls that he often gave notes like this to Kolmogorov, who rarely had much time to hear lengthy explanations and preferred something written in any case.) We use a sequence $\alpha$ that provides a “dense” coding of a universal \[recursively\] enumerable set. For example, let $\alpha$ be the binary representation of \[here the text “the sum of the a priori probabilities of all natural numbers” is crossed out and replaced by the following:\] the real number $$\sum_{p\in A} \frac{1}{p\cdot \log^2 p}$$ where $A$ is the domain of the optimal algorithm. A binary string $p$ is a “good” code for $x$ if the optimal algorithm converts the pair $(p, K(x))$ into a list of strings that contains $x$ and the logarithm of the cardinality of this list does not exceed $K(x)+3\log K(x)-l(p)$. (The existence of such a code means that $x$ is “random” when $n\ge l(p)$.) We say that a binary string $p$ is a canonical code for $x$ if every prefix of $p$ either is a “good” code for $x$ or is a prefix of $\alpha$, and $l(p)=K(x)+2\log K(x)$. Theorem 1. Every $x$ (with finitely many exceptions) has a canonical code $p$, and $p$ and $x$ can be effectively transformed into each other if $K(x)$ is given. Therefore, the “non-randomness” in $x$ can appear only due to some very special information (a prefix of $\alpha$) contained in $x$. I cannot imagine how such an $x$ can be observed in (extracted from) the real world since $\alpha$ is not computrable. And the task “to study the prefixes of a specific sequence $\alpha$” seems to be very special. [99]{} Borel, Émile, Le hazard, Alcan, Paris, 1913. (Russian translation: Случай, Госиздат, Москва – Петроград, 1923.) Borel, Émile, Probabilité et certitude, Presses Univ. de France, Paris, 1950. (Russian translation: Вероятность и достоверность, Физматгиз, Москва, 1961.) Calude, Cristian S., Information and Randomness. An Algorithmic Perspective. (Texts in Theoretical Computer Science.) Springer-Verlag, 1994. Second edition, 2002. Calude, Cristian S., A talk given at Dagstuhl seminar, 29 January – 3 February 2006, downloaded from `http://www.hutter1.net/dagstuhl/calude.mp3` Chaitin, Gregory J., On the length of programs for computing finite binary sequences, Journal of the ACM, v. 13 (1966), pp. 547–569. Available also at Chaitin’s home page [@chaitin-home]. Chaitin Gregory J., On the length of programs for computing finite binary sequences: statistical considerations, Journal of the ACM, v. 16 (1969), pp. 145–159. Available also at Chaitin’s home page [@chaitin-home]. Chaitin, Gregory J., Computational complexity and Gödel’s incompleteness theorem, ACM SIGACT News, No. 9 (April 1971), pp. 11–12. Chaitin, Gregory J., A theory of program size formally identical to information theory, Journal of the ACM, vol. 22 (1975), pp. 329–340. Received April 1974; revised December 1974. в разделе 2 даётся определение префиксной машины (которая должна остановиться ровно в месте конца программы на входной ленте) и устанавливается эквивалентность этого определения требованию отсутствия префиксов в области определения доказывается существование оптимальной условная сложность $H(s\|t)$ определяется с условием $t^$, являющимся первой программой, порождающей $t$. Это эквивалентно добавлению условия $H(t)$, так как количество минимальных программ есть $O(1)$. Вероятность $P(s)$ определяется как вероятность получить $s$ при случайной программе, то есть как сумма $2^{-\|p\|}$ по всем программам, дающим на выходе $s$. интерпретация вероятности для машин со случайным датчиком (заранее написанным на ленте) Информация $I(x:y)$ определяется как $H(y)-H(y\|x)$, где условие понимается как кратчайшая программа для $x$ (то есть добавляется $H(x)$) доказывается лемма о шторах (теорема 3.2); доказательство ссылается на N.Pippenger’а (и содержит стандартный инвариант про объединение отрезков разных длин). теорема 3.4 доказывает совпадение сложности и логарифма вероятности теорема 3.9 содержит формулу для сложности пары теорема 4.1 говорит, что для сходящегося вычислимого ряда $\sum 2^{-f(n)}$ выполнено неравенство $H(n)\le f(n)+O(1)$ (но перечислимые снизу ряды не рассматриваются) теорема 4.2: максимум сложности слов длины $n$ равен $n+H(n)+O(1)$; доля слов длины $n$, у которых сложность отличается от максимума на $k$, не больше $2^{-k}$; для бесконечной последовательности с вероятностью $1$ сложность начальных отрезков больше $H(n)$ при всех $n$, кроме конечного числа. определение случайности: $H(\omega_1\ldots\omega_n)\ge n-O(1)$ “C.P.Schnorr (private communication) has shown that this complexity-based definition of a random infinite string and P.Martin-Löf’s statistical definition of this concept are equivalent” Определение числа $\omega$. Теорема 4.3: оно перечислимо снизу, случайно и принадлежит классу $\Sigma_2$. добавление: $H(s,H(s))=H(s)$, $H(H(s)\|s)$ не есть $O(1)$. Chaitin, Gregory J., Algorithmic information theory. Some recollections. 25 May 2007. Downloaded from the homepage of Gregory Chaitin, `http://www.cs.auckland.ac.nz/ chaitin/60.html` Church, Alonzo, On the concept of a random sequence. Bull. Amer. Math. Soc, 1940, v. 46, no. 2, pp. 130–135. Гач, Петер, О симметрии алгоритмической информации. Доклады Академии наук СССР, vol. 218 (1974), No. 6, pp. 1265–1267. Submitted: April 9, 1974. Translated as: Gacs, Peter, On the symmetry of algorithmic information, Soviet Math. Dokl., vol. 15 (1974), No. 5. Gács, Peter, Komplexität und Zufälligkeit, Inaugural–Dissertation zur Erlangung des Doktorgrades der Naturwissenschaften vorgelegt beim Fachbereich Mathematik der Johann Wolfgang Goethe Universität zu Frankfurt am Main, Tag der mündlichen Prüfung 12.12.1978. 41 pp. Gács, Peter, Review of Algorithmic Information Theory* by Gregory J. Chaitin, The Journal of Symbolic Logic, Vol. 54, No. 2 (June 1989), pp. 624–637. Gács, Peter, private communication, 2008. Goldreich, Oded, An Introduction to Cryptography. Vol. 1. Cambridge University Press, 2001. Kolmogorov, Andrei N., On tables of random numbers, Sankhyã: The Indian Journal of Statistics, Series A, Vol. 25, Part 4, 1963, pp. 369–376. Колмогоров, Андрей Николаевич, Три подхода к определению понятия “количество информации”, Проблемы передачи информации, v. 1, No. 1, 1965, pp. 3–11. (Three approaches to the quantitative definition of information; reprinted in the collection [@kolmogorov-collection-3]) Колмогоров, Андрей Николаевич, К логическим основам теории информации и теории вероятностей, Проблемы передачи информации, 1969, v. 5, No. 3, pp. 3–7. The translation is published as: Kolmogorov A.N., Logical basis for information theory and probability theory, IEEE Trans. Inform. Theory, 14, 662–664 (1968). Footnote: “Manuscript received December 13, 1967. The work is based on an invited lecture given at the International Symposium on Information Theory, San Remo, Italy, September, 1967. Translation courtesy of AFOSR, USAF. Edited by A.V.Balakrishnan.” Колмогоров, Андрей Николаевич, Избранные труды. Том 3. Теория информации и теория алгоритмов. М.:Наука, 1987. English translation: Selected Works of A.N. Kolmogorov. Vol.III: Information Theory and the Theory of Algorithms. Kluwer, 1993. Laplace, M. le Marquis de, in: Œuvres compétes de Laplace, publiées sous le auspices de l’Académie des science, par MM. les secrétaires perpétuels. Tome septième. Paris, Gauthier-Villars, imprimeur-libraire de l’École Polytechnique, du bureau des longitudes, 1886. Levin, Leonid A., Some Syntactic Theorems on the Finite Problems Calculus of Ju.T. Medvedev, Soviet Math. Doklady, vol. 10 (1969), pp. 288-290. (Translation of an article in Доклады Академии наук СССР, v. 185 (1969), pp. 32–33.) Левин, Леонид Александрович, О понятии случайной последовательности, Доклады Академии наук СССР, 1973, v. 212, no. 3, pp. 548–550. Submitted July 1, 1972. Translated as: Levin, Leonid A., On the notion of a random sequence, Soviet Math. Dokl., vol. 14, pp. 1413–1416, 1973. Левин, Леонид Александрович, Законы сохранения (невозрастания) информации и вопросы обоснования теории вероятности. Проблемы передачи информации, vol. 10, 1974, No. 3, pp. 30–35. Submitted: Jan. 9, 1974, the journal issue was sent to the printer in August 1974. Translated as: Levin L.A., Laws of information conservation (nongrowth) and aspects of the foundation of probability theory, Problems of Information Transmission, v. 10 (1974), pp. 206–210. Левин, Леонид Александрович, О различных мерах сложности конечных объектов (аксиоматическое описание). Доклады Академии наук СССР, vol. 227 (1976), no. 4, pp. 804–807. Submitted: June 7, 1975. Translated as: Levin, Leonid A., The various measures of the complexity of finite objects (an axiomatic description), Soviet Math. Dokl., v. 17 (1976), pp. 522–526. Levin, Leonid A., An interview (June 2008, unpublished) Levin, Leonid A., Колмогоров глазами школьника и студента (Kolmogorov as seen by a high school and university student). In Колмогоров в воспоминаниях учеников (Kolmogorov remembered by his disciples). Moscow, 2006, Наука, МЦНМО. ISBN 5-94057-198-0. Loveland, Donald, A new interpretation of the von Mises’ concept of random sequence, Zeitschrift für Mathematische Logik und Grundlagen der Mathematik, v. 12, no. 3, pp. 279–294, 1966. Lieb, Elliot H.; Osherson Daniel; Weinstein, Scott, Elementary Proof of a Theorem of Jean Ville, see `arxiv:cs/0607054v1` at `arxiv.org`. Марков, Андрей Андреевич, О нормальных алгорифмах, вычисляющих булевы функции. Доклады Академии наук СССР, vol. 157, no. 2, pp. 262–264, 1964 Марков, Андрей Андреевич, О нормальных алгорифмах, связанных с вычислением булевых функций. Известия Академии наук СССР, серия математическая, vol. 31, pp. 161–208, 1967. Martin-Löf, Per (Мартин-Лёф П.) О понятии случайной последовательности. (On the notion of a random sequence.) Теория вероятностей и её применения. V. 11, No. 1, pp. 198–200, 1966. Martin-Löf, Per, The Definition of Random Sequences, Information and Control, Vol. 9, No. 6, pp. 602–619, 1966. Martin-Löf, Per, The Literature on von Mises’ Kollektivs Revisited, Theoria, vol. 35(1), 1969, pp. 12–37. Martin-Löf, Per, Complexity Oscillations in Infinite Binary Sequences, Zeitschrift für Wahrscheinlichkeitstheorie und verwandte Gebiete, vol. 19, pp. 225–230, 1971. Martin-Löf, Per, private communication, 2008. Miller, Joseph, Every 2-random real is Kolmogorov random, Journal of Symbolic Logic, v. 69, no. 2, pp. 555–584, 2004. von Mises, Richard, Grundlagen der Wahrscheinlichkeitsrechnung, Mathematische Zeitschrift, Bd. 5, 191, S. 52–99. (Reprinted in: Selected Papers of Richard von Mises. Volume Two. Probability and Statistics, General. American Mathematical Society, 1964. pp. 57–106.) даётся как бы определение коллектива, причём для последовательности точек в многомерном пространстве (стабильность частот попадания в подмножества, вроде бы не уточняется класс подмножеств) но, скажем, говорится, что Solange man etwa nur die Zahlen 1–10000 betrachtet, bietet die Anordnung der Ziffern an der 5. Stelle \[in the table of logarithms\] tatsächlich das ungefähre Bild eines empirisches Kollektivs und man kann auch die Sätze der Wahrscheinlichkeitsrechnung näherungsweise darauf anwenden. в основном обсуждаются преобразования распределений при преобразовании коллективов есть ещё статья 1912 года Über die Grundbegriffe der Kollektivmasslehre”, но там собственно определения коллектива (или чего-то похожего на определение) я не нашёл, но подробно не разбирал von Mises, Richard, Wahrscheinlichkeit, Statistik und Wahrheit, Wien: Springer-Verlag, 1928. 189 pp. von Mises, Richard, On the foundations of probability and statistics, Annals of Mathematical Statistics, vol. 12 (1941), pp. 191–205. (Reprinted in [@mises-1964], pp. 340–355.) выступление на собрании Американского статистического общества 1\. теория вероятностей — не метафизика и о вероятностях одиночных событий, типа существования Наполеона, ничего не говорит 2\. Примеры коллективов — бросание монеты, выживание до 41 года среди большой группы здоровых 40-летних людей (NB! ничего не говорится о порядке) на практике всё конечно, но это обычная для математики идеализация, вроде как бесконечные прямые пример задачи теории вероятностей: в коллективе из нулей мы делим на группы по три и в каждой группе берём большинство; надо вычислить новое распределение по старому We know four and only four simple, i.e., irreducible, transformations or four fundamental operations. They are called selection, mixing, partitioning and combination. By combining these basic processes we can settle all problems in probability theory 3\. Правило выбора: даётся точное определение. “A place selection is defined by an infinite set of functions $s_n(x_1,x_2,\ldots,x_{n-1})$ where $x_1,x_2,x_3,\ldots$ are the digits of an admissible number of a kollektiv and $s_n$ has one of the two values, zero or one. Here $s_n=1$ means that the $n$th digit of the sequence is retained, $s_n=0$ means that it is discarded.” “For this case, it was proven by Wald and by Copeland that, if an arbitrary distribution \[неотрицательные числа, в сумме равные единице\] is given and an arbitrary countable set $\Sigma$ of place selections, there exists a continuum of sequences each of which has the given distribution, which is not affected by any place selection belonging to $\Sigma$.” Про коллективы с вещественными значениями: “I cannot enter here into a discussion of the more complicated case where the range within which the elements of the kollektiv vary, is infinite, either a countable set or a continuum. At any rate, all problems of principle connected with the notion of the kollektiv can be settled satisfactorily by considering these general types of sequences as limiting cases of kollektivs with a finite set of attributes.” 4\. Разбирается пример разбиения последовательности игр на группы по три, а также более сложный вариант, в котором третья игра не проводится, если первые две имеют одинаковый исход и третья не влияет на большинство. Мизес говорит, что теория вероятностей должна обосновать частоту выигрышей в группах в этом случае, и обычная теория вероятностей этого сделать не может. в конце: “I have to make some specific remarks about the so-called measure theory of probability”. Footnote: “What I call measure theory is essentially that proposed by Kolmogoroff in his book of 1933. As to the theory discussed by Doob in his following paper (where the space of all logically possible sequences is used in establishing the measures) see my comments” in [@mises-1941a]. 5\. Probability as measure. “In particular A. Kolmogoroff sets up a complete system of such restriction. We cannot ask for the existence of the limiting frequency in any arbitrary subset $A$.” “All axioms of Kolmogoroff can be accepted within the framework of our theory as a part of it but in no way as a substitute for our definition of probability.” далее довольно смутные возражения про множества меры нуль и вывод: “In recapitulation this paragraph I must say: First, the axioms of Kolmogoroff are concerned with the distribution function within one kollektiv and are supplementary to my theory, not a substitute for it. Second, using the notion of measure zero in an absolute way without reference to the arbitrarily assumed measure system, leads to essential inconsistencies.” 6\. Statistical estimations. von Mises, Richard, Doob, J. L., Discussion of papers on probability theory. Annals of Mathematical Statistics, vol. 12 (1941), pp. 215–217. (Reprinted in [@mises-1964], pp. 356–359.) Заключительные замечания к статье Дуба (не перепечатанной в собрании трудов Мизеса, естественно). Объясняется, что Дуб предлагает ввести меру на множестве $\Omega$ всех последовательностей. Мизес говорит: It seems to me that my conception is simpler in the application and closer to reality, while the other model may be considered more satisfactory from a logical standpoint since it avoids the difficulties connected with the concept of “all place selections”. Дуб отвечает, что у Мизеса всё получается математически сложно, а с мерами всё проще (в том числе и пример с группировкой игр). как рассказывает Shafer, это сильно приглаженная версия гораздо более ожесточённой полемики Selected Papers of Richard von Mises. Volume Two. Probability and Statistics, General. American Mathematical Society, 1964. Nies, Andre; Stephan, Frank; Terwijn, Sebastiaan A., Randomness, relativization and Turing degrees, Journal of Symbolic Logic, v. 70, no. 2, 515–535, 2005. Pólya, George (György), Mathematics and plausible reasoning. I. Induction and analogy in mathematics. II. Patterns of plausible inference. Princeton University Press, Princeton, NJ, 1954. Schnorr, Claus-Peter, A Unified Approach to the Definition of Random Sequences, Mathematical Systems Theory, v. 5, no. 3, pp. 246–258 (1971). Schnorr, Claus-Peter, Zufälligkeit und Wahrscheinlichkeit. Eine algorithmische Begründung der Wahrscheinlichkeitstheorie, Lecture notes in mathematics, v. 218. IV+212 S. Springer, 1971. Schnorr, Clauss-Peter, The process complexity and effective random tests, Proceedings of the fourth annual ACM symposium of Theory of Computing, held in Denver, Colorado, United States, May 01–03, 1972. Journal version: Process complexity and Effective Random Tests, Journal of computer and system sciences, vol. 7, 376–388, 1973. Schnorr, Claus-Peter, A survey of the theory of random sequences, In: Butts R.E., Hintikka J., eds., Basic problems in methodology and linguistics, Dordrecht, D. Reidel, pp. 193–211. Schnorr, Claus-Peter, A talk during Dagstuhl seminar 06051, 29 January – 3 February 2006, downloaded at `http://www.hutter1.net/dagstuhl/schnorr.mp3`. The paper where it is explained that even Feller reviewing Ville’s 1936 paper did not understood the intuitive meaning of martingales saying that Ville replaces selection rules by martingales by no reason. Шень, Александр, К логическим основам применения теории вероятностей, Семиотические аспекты формализации интеллектуальной деятельности. Школа-семинар. Г. Телави, 29 октября – 6 ноября 1983 г. Тезисы докладов и сообщений. Москва, ВИНИТИ, 1983. 253 pp., pp. 144-146. Shen, Alexander, Algorithmic Information Theory and Kolmogorov Complexity, Uppsala Universitet, Technical Report 2000-034. Available at: `http://www.it.uu.se/research/publications/reports/2000-034`. Shen, Alexander, Algorithmic information theory and foundations of probability. Submitted to 2009 Reachability Problems conference, September 2009. Solomonoff, Ray J., A formal theory of inductive inference, Part I, Information and Control, v. 7, no. 1, pp. 1–22 (March 1964) Solomonoff Ray J., A formal theory of inductive inference, Part II, Information and Control, v. 7, no. 2, pp. 224-254 (June 1964) Solomonoff Ray J., Complexity-Based Induction Systems: Comparisons and Convergence theorems. IEEE Transactions on Information Theory, Vol. IT-24, No. 4, pp. 422–432, July 1978. Solovay, Robert, Draft of paper (or series of papers) on Chaitin’s work, unpublished notes, May 1975, 215 pp. Успенский, Владимир Андреевич, Лекции о вычислимых функциях, М., Физматгиз, 1960 Успенский, Владимир Андреевич; Семёнов, Алексей Львович; Шень, Александр, Может ли (индивидуальная) последовательность нулей и единиц быть случайной? Успехи математических наук, v. 45, no. 1 (271), pp. 105–162, 1990. English translation: Uspenskii, Vladimir A.; Semenov, Alexei L.; Shen’, Alexander, Can an individual sequence of zeros and ones be random? Russian Math. Surveys, vol. 45, no. 1, pp. 121–189, 1990. Ville, Jean, Etude critique de la notion de collectif, Monographies des Probabilitiés, Paris, 1939. Wald Abraham, Sur la notion de collectif dans le calcul des probabilités. (On the notion of collective in probability theory), présentée par M. Émile Borel. Comptes rendus, 202, pp. 180–183 (séance du 20 janvier 1936). Wald Abraham, Die Wiederspruchsfreiheit des Kollektivbegriffes der Wahrscheinlichkeitsrechnung, Ergebnisse eines matematischen Kolloquiums, vol. 8, pp. 38–72, 1937. Reprinted in: Menger, Karl, Ergebnisse eines Mathematischen Kolloquiums, Edited by E. Dirker, K. Siegmund, With contributions by J.W. Dawson jr., R. Engelking, W. Hildenbrand, Foreword by G. Debreu, Afterword by F. Alt, Springer, Wien, New York, 1998. Звонкин, Александр Калманович; Левин, Леонид Анатольевич, Сложность конечных объектов и обоснование понятий информации и случайности с помощью теории алгоритмов. Успехи математических наук, v. 25, no. 6 (156), pp. 85–127 (1970). English translation: Zvonkin, A.K., Levin, L.A., The complexity of finite objects and the development of the concepts of information and randomness by means of the theory of algorithms. Russian Math. Surveys, 25:6 (1970), pp. 83–124. [^1]: “C’est ici le lieu de définir le mot extraordinaire. Nous rangeons, par la pensée, tous les événements possibles en diverses classes, et nous regardons comme extraordinaires ceux des classes qui en comprenement un très petit nombre. Ainsi, a jou de croix ou pile, l’arrivée de croix cent fois de suite nous parait extraordinaire, parce ques le nombre presque infini des combinaisons quit peuvent arriver en cent coups, étant partagé en séries régulières ou dans lesqulles nous voyone régner un ordre facile à saisir, et en séries irrégulières, celles-ci sont incomparablement plus nombreuses. La sortie d’une boule blanche d’une urne qui, sur un million de boules, n’en contient qu’une seule de cette couleur, les autres étant noires, nous parait encore extraordinaire, parce que nous ne formons que deux classes d’événement ordinaire, relatives aux deux couleurs. Mais la sortir du n$^\circ~475813$, par exemple, d’une urne qui renferme un million de numéros nous semble un événement ordinaire, parce que, comparant individuallement les numéros les uns aux autres, sans les partager en classes, nous n’avons aucune raisone de croire que l’un d’eux sortira plutòt que les autres.” (“Essai philosophique sur les Probabilités” [@laplace], VI Principe). Peter Gács, who used this passage as an opening quote for his Dissertation [@gacs-thesis], comments: “Laplace makes two informal suggestions (withouth strictly distinguishing them). First, he considers various classes of events, and views as extraordinary the small ones. (To make this precise, one would need to restrict attention to “simple” classes.) Second, he makes the assertion (without proof or even exact statement) that all outcomes of a given length having some regularity in them, grouped together, would still form a small class. (To make this precise, regularity must be defined appropriately.)” [@gacs-private] [^2]: A short note without proofs was published earlier [@wald-1936]. [^3]: In fact, at Ville’s time these arguments did not sound very convincing even to some experts: W. Feller wrote in his Zentralblatt review of one of the first Ville’s papers: “Aus unerfindlichen Gründen will nun Verf. den Auswahlbegriff so abändern (“martingale” statts Auswahl) daß jede Nullmenge als Ausnamemenge bei passendem $S$ autreten kann”, both reproducing the main argument of Ville (the possibility to exclude any null set) and finding it unconvincing (“unerfindlichen Gründen”), see [@shafer]. [^4]: The most famous discovery of Chaitin is probably the proof of Gödel incompleteness theorem based on the Berry paradox [@chaitin-1971]; we don’t discuss it here. [^5]: Here is the Russian quotation from [@levin-memoir]: “Тема, которой Андрей Николаевич тогда увлекался — общие понятия сложности, случайности, информации — волновала меня чрезвычайно. Как многие молодые люди, я искал самых фундаментальных концепций. Но такие “первичные” теории, как логика или теория алгоритмов, смущали меня своей “качественной” природой — там нечего было “посчитать”. На самом деле, я ещё в Киеве пытался дать определение сложности (я называл её “неестественность”), но не мог доказать её инвариантности. В Москве я рассказал о своих неудачах Сосинскому, он спросил Колмогорова и принёс мне поразительный ответ: Колмогоров как раз доказал то, что я не смог и уже вот-вот выйдет его подробная статья! Тогда я решил во что бы то ни стало поступить в МГУ и стать учеником Андрея Николаевича.” [^6]: Chaitin’s papers start with a lot of technical details related to the counting of Turing machines states. Solomonoff’s paper [@solomonoff-1964a] contains passages like “The author feels that Eq. (1) is likely to be correct or almost correct, but that the methods of working the problems of Sections 4.1 to 4.3 are more likely to be correct than Eq. (1). If Eq. (1) is found to be meaningless, inconsistent or somehow gives results that are intuitively unreasonable, then Eq. (1) should be modified in ways that do not destroy the validity of the methods used in Sections 4.1 to 4.3” — not very encouraging for the readers, to say the least. Levin remembers that when he was instructed by Kolmogorov to read and cite the work of Solomonoff, he was frustrated by this kind of attitude and soon gave up. Section 3.2.1 of [@solomonoff-1964a] contains the following sentence: “Although a proof \[of some statement, related to a definition called Eq. (1); this definition contained an error, as Solomonoff found later\] is not available, an outline of the heuristic reasoning behind this statement will give clues as to the meanings of the terms used and the degree of validity to be expected of the statement itself”. But later in the same paragraph a very clear proof of universality theorem is provided for the readers who are not confused by previous remarks and are able to extract its statement out of the proof. This paper also contained a lot of other ideas that were developed much later; e.g., in Section 3.2 Solomonoff gives a nice simple formula for predictions in terms of the conditional a priori probability, using monotonic machines much before Levin and Schnorr. (In 1978 Solomonoff formally proved that this formula works for all computable probability distributions, see [@solomonoff-1978].) In fact, Solomonoff’s main interest was inductive inference. He tried to formalize the “Occam’s Razor” principle in the following way: base your prediction on the simplest “law” that fits the data, say the simplest program that could generate it. This requires a definition of “simplecity”, and it was in this context that Solomonoff defined complexity in terms of description length and proved its invariance. (His actual prediction formula uses conditional a priori probability, based on all possible programs that fit the data, with longer programs entering with smaller weights.) [^7]: When Kolmogorov has came to the definition of complexity? In his 1963 paper [@kolmogorov-1963] Kolmogorov makes some remarks that partially explain how he came to the complexity notion: “I have already expressed the view $\langle\ldots\rangle$ that the basis for the applicability of the results of the mathematical theory of probability to real ‘random phenomena’ must depend on some form of the frequency concept of probability, the unavoidable nature of which has been established by von Mises in a spirited manner. However, for a long time I had the following views: (1) The frequency concept based on the notion of limiting frequency as the number of trials increases to infinity, does not contribute anything to substantiate the applicability of the results of probability theory to real practical problems where we have always to deal with a finite number of trials. (2) The frequency concept applied to a large but finite number of trials does not admit a rigorous formal exposition within the framework of pure mathematics. Accordingly I have sometimes put forward the frequency concept which involves the conscious use of certain not rigorously formal ideas about ‘practical reliability’, ‘approximate stability of the frequency in a long series of trials’, without the precise definition of the series which are ‘sufficiently large’… I still maintain the first of the two theses mentioned above. As regards the second, however, I have come to realize that the concept of random distribution of a property in a large finite population can have a strict formal mathematical exposition. In fact, we can show that in sufficiently large populations the distribution of the property may be such that the frequency of its occurrence will be almost the same for all sufficiently large sub-populations, when the law of choosing these is sufficiently simple. Such a conception in its full development requires the introduction of a measure of the complexity of the algorithm. I propose to discuss this question in another article. In the present article, however, I shall use the fact that there cannot be a very large number of simple algorithms.” In this quote Kolmogorov suggested a finitary Mises-style approach that uses selection rules of bounded complexity, but does not explain what complexity is; also he does not speak here about definition of randomness in terms of complexity (directly, without using selection rules). Asked when Kolmogorov came to his definition of complexity, Martin-Löf writes [@martin-lof-interview]: “Kolmogorov must have arrived at his complexity definition before autumn 1964, since Lyonya Bassalygo \[Леонид Бассалыго\] told me about it then. \[Bassalygo confirms this; he remembers a walk during late autumn or early spring when Kolmogorov tried to explain him the complexity definition that was quite difficult to grasp at first.\] On the other hand, it should be later than the randomness definition proposed in the Sankhya paper [@kolmogorov-1963] which was received April 1963 by the journal. Those considerations pin down the time of discovery to 1963–64, more exactly. (Kolmogorov never told me anything about the history of his discovery.) \[On the other hand,\] in his obituary note in the Journal of Applied Probability, Vol. 25, No. 1, pp. 445–450, March 1988, K.R. Parthasarathy writes: “Immediately after his arrival in Calcutta, Andrei Nikolaevich lost no time in plunging into discussions with the young students at the Institute about his recent research work on tables of random numbers, and the measurement of randomness of a sequence of numbers using ideas borrowed from mathematical logic. This piece of research was carried out by him during his travel by ship from the USSR to India; the ship was actually proceeding on an oceanographic expedition.” This seems to fix the time of the discovery of the complexity definition of randomness to 1962 \[at least in some preliminary form\] and to locate it to the ship that brought him to India for the reception of the degree of Doctor Honoris Causa at the University of Calcutta.” Kolmogorov gave several talks at the Moscow Mathematical Society but for most of them only the titles are known, and we may only guess what was there: Редукция данных с сохранением информации (Data reduction that conserves information, March 22, 1961), Что такое “информация”? (What is information?, April 4, 1961), О таблицах случайных чисел (On the tables of random numbers, October 24, 1962, probably corresponding to Sankhya paper [@kolmogorov-1963]), Мера сложности конечных двоичных последовательностей (A complexity measure for finite binary strings, April 24, 1963), Вычислимые функции и основания теории информации и теории вероятностей (Computable functions and the foundations of information theory and probability theory, November 19, 1963), Асимптотика сложности конечных отрезков бесконечной последовательности (Asymptotic behavior of the complexities of finite prefixes of an infinite sequence, December 15, 1964; the title suggest that the last talk was about Martin-Löf results, though Martin-Löf remembers discussing these results with Kolmogorov only next spring, see below). Three later talks about algorithmic information theory (1968–1974) have short published abstracts (see Appendix A.) [^8]: Per Martin-Löf, a mathematician from Sweden, studied Russian during his military service and then decided to make use of his knowledge by coming to Moscow and working with Kolmogorov. Martin-Löf tells in [@martin-lof-interview]: …I had not worked on randomness before coming to Moscow in 1964–65. Kolmogorov first gave me a statistical problem in discriminant analysis, which I solved, although I did not find it challenging enough. It was a problem that I might just as well have worked on at home in Stockholm. But I got to know Leonid (Lyonya) Bassalygo \[Леонид Бассалыго\], and he told me about Kolmogorov’s new ideas about complexity and randomness, which I found very exciting. This was in late autumn 1964. So I started to learn the necessary recursive function theory from Uspenskij’s book [@uspensky-lectures]… It was only when I told Kolmogorov about my first results on complexity oscillations in infinite binary sequences in early 1965 that complexity and randomness became the subject of our discussions. (So I did not learn about Kolmogorov complexity directly from Kolmogorov but only indirectly from Bassalygo). \[As to the motivation,\] I studied the previous literature on random sequences only after I had made my own first contributions. This resulted in the paper The Literature on von Mises’ Kollektivs Revisited published in the Swedish philosophical journal Theoria [@martin-lof-1969]. \[As to the predecessors,\] I have been most interested in Borel, particularly because he was the most important of the early French constructivists, which Brouwer called the pre-intuitionists. My affection for him may also have to do with the fact that I inherited a copy of Borel’s Lecons sur la Théorie des Fonctions, with its many interesting Notes at the end, when my grandfather died in 1958 and I was aged 16. When trying to require the complexities of the finite initial segments to be as big as possible, I discovered the unavoidable complexity oscillations about which I wrote my first paper on the subject (in Russian and typed by Nataliya Dmitrievna Svetlova \[Наталья Дмитриевна Светлова (Солженицына)\], who became Solzjenitsyn’s wife in her second marriage). This led me to try the new approach of suitably interpreting the definition of null set in the sense of recursion theory. I should add that my primary reason for being interested in infinite rather than finite random sequences was to get rid of the additive constants that cropped up everywhere, and whose arbitrariness I found annoying. \[This paper,\] the first one of my two Russian papers was never published, but a typed copy of it should still exist somewhere in my unsorted archive. However, the results contained in it were subsequently published in English in my paper [@martin-lof-1971]. The paper [@martin-lof-1966r] is the second of the two papers that I have written in Russian. It summarizes a talk that I apparently gave in Moscow on 2 June 1965 and shows very clearly that I had not yet reached the definition of my Information and Control paper [@martin-lof-definition] though I was on my way. Kolmogorov was immediately very interested in my two theorems on the unavoidable complexity oscillations in infinite binary sequences, which I told him about in the train on our way to Caucasus, more precisely, Bakuriani \[Armenia\] in early March 1965. In fact, he was so positive that he asked me to present my results as a sequel to a guest lecture that he gave in Tbilisi on our way back in late March. I do not think that he had thought himself about the problem of defining infinite random sequences by means of his complexity measure before then. So I think it is correct to say… that he was more interested in finite random sequences. In a way, even if I have myself been interested in getting a good definition of randomness for infinite sequences, it is more striking that one can give a sensible definition of randomness already for finite sequences. Concerning finite random sequences, my own only contribution was the observation that the random elements of a finite population should be the ones whose conditional complexity given the population is maximal, that is, approximately equal to the logarithm to the base $2$ of the number of elements of the population, whereas Kolmogorov’ original suggestion was to use the unconditional complexity. So, in the case of a completely random sequence of length $n$, we should use $K(x_1\ldots x_n\|n)$ rather than $K(x_1\ldots x_n)$, and, in the case of Bernoulli sequences, $K(x_1\ldots x_n\|n,s_n)$, where $s_n = x_1 +\ldots+ x_n$. I never had the opportunity of discussing my own definition of randomness for infinite sequences with Kolmogorov, simply because I did not find it until after I left Moscow in July 1965. It must have been sometimes during the academic year 1965–66. (End of quote.) [^9]: Levin recalls that being an undergraduate student he wanted to convince Kolmogorov to be his advisor and hoped that this result would impress Kolmogorov. But Kolmogorov was rather busy, and the appointment was postponed several times from February to August 1967. Finally, when Levin called him again, Kolmogorov said something like: “O yes, come to see me, I have very interesting results, the information is symmetric”. — “But, Andrei Nikolaevich, this is exactly what I wanted to tell you.” — “But do you know that the symmetry is only up to logarithmic terms?” — “Yes.” — “And you can give a specific example?” — “Yes.” Then Levin came to see Kolmogorov, they discussed these results (later announced in [@kolmogorov-1968-69] without proof; the first proof appeared in [@zvonkin-levin]). Levin indeed worked with Kolmogorov during his undergraduate years and even earlier (the first Levin’s result was obtained under Kolmogorov’s supervision when Levin was in high school and published later as [@levin-1969]) but V.A. Uspensky was officially listed as his undergraduate advisor for some formal reasons (see below). [^10]: As Schnorr said in his talk [@schnorr-talk], he had not read Ville’s book, but learned the notion of martingale indirectly through other sources. [^11]: Later a more practical theory of pseudorandom sequences was developed by Yao, Blum, Micali and others. Now it is a very important part of computational cryptography, see, e.g., the textbook [@goldreich]. Schnorr later also worked in the field of computational cryptography. [^12]: Let us add some historical remarks about situation in the Mathematics Department of Moscow State University and in Russia at that time. The typical track of a future mathematician at that time was 5 years of undergraduate studies (высшее образование) plus 3 years of graduate school (аспирантура). After the graduate school student is assumed to defend a thesis and get a title “kandidat fiziko-matematicheskih nauk” (кандидат физико-математических наук) which is a rough equivalent of Ph.D. Unlike the US universities, the student of Moscow State University (and other Soviet universities) had to decide what is his major before entering the university: e.g., the mathematics and physics programs are administered by different departments, have no common courses, different entrance procedures etc. After two years of undergraduate studies at mathematics department, a student had to choose a division (кафедра) which he wants to join for three remaining years, and a scientific advisor in the chosen division. (It could be, say, Algebra Division, or Geometry and Topology Division, etc.) At the end of the 5th year student writes a thesis (дипломная работа). Sometimes this thesis is considered as something close to the Master thesis in the US. To enter the graduate school after finishing 5 years of undergraduate studies, one needed a good academic record and (a very important condition!) a recommendation from the local communist party and komsomol (комсомол) organization. Komsomol (an abbreviation for коммунистический союз молодёжи, communist union of the young people), like Hitlerjugend in Germany, was almost obligatory, and included people of age 14–28, so most university students were komsomol members (комсомольцы), though there were some exceptions and this requirement was never formalized as a law. Levin was a student of a special boarding school founded by Kolmogorov (unofficially called Kolmogorov’s boarding school, колмогоровский интернат); during 1963/4 academic year he was a student of a similar school in Kiev (now Ukraine) and then managed to move to Moscow for 1964/5 academic year. Then (in January 1966) he entered the Moscow university becoming a first-year undergraduate in the middle of the academic year (there was some exceptional procedure for the students of Kolmogorov’s school in this year related to the change in the education system in the USSR that moved from 11-years to 10-years education program). Being not only Jewish (already a handicap at that time) but also a kind of non-conformist, Levin as an undergraduate student created a lot of troubles for local university authorities. As a member of komsomol, he became elected local komsomol leader but he defied the policies established by the Communist Party supervisors (and this was mentioned in his graduation letter of recommendation, a very important document at the time). No wonder he was effectively barred from applying to any graduate school when he finished undergraduate studies at the Mathematical Logic Division (кафедра математической логики) in 1970. (His official undergraduate advisor was Vladimir A. Uspensky who was Kolmogorov’s student in 1950s. Kolmogorov officially did not belong to Mathematical Logic division and asked his former student Uspensky to replace him in this capacity.) However, Kolmogorov managed to secure a research scientist position for Levin (with the help of the University rector, a prominent mathematician and a very decent person, I.G. Petrovsky) in the University statistical laboratory (Kolmogorov was a head of this laboratory). Being there, in 1971 Levin wrote a “kandidat” thesis (that contained mostly Levin’s results included in [@zvonkin-levin], but also some others, including the probabilistic definition of prefix complexity) and tried to find a place for its defense. (According to the rules, the thesis defence was not technically connected to a graduate school (if any) of defendant’s affiliation, only a recommendation from the institution where dissertation was prepared was required; in this case the person was called “соискатель”. Though most graduate students in the USSR were defending their thesis in the same institution (sometimes a few years later after their studies in the graduate school), the thesis defense was not a university affair, but regulated by a special government institution, called “Высшая Аттестационная Комиссия”.) In Moscow it was clearly impossible, and finally the defense took place in Novosibirsk (in Siberia). The thesis received strong approvals from official reviewers (J. Barzdin, B. Trachtenbrot and his lab), the reviewing institutions (Leningrad Division of Steklov Mathematical Institute) and the advisor (Kolmogorov and his lab). Nevertheless, the defence was unsuccessful (quite untypical event). According to Levin, the most active negative role during the council meeting was played by Yu.L. Ershov (recursion theorist, now a member of the Russian Academy of Sciences) but Levin believes that Ershov did not have other choice unless he was ready to get into career troubles himself; however, Ershov did also something “above and beyond the call of duty” (as Levin puts it) as a Soviet scientific functionaire — he insisted that the “unclear political position” of Levin should be mentioned in the council decision. This effectively prevented Levin’s defense in any other place in the Soviet Union (even with a new thesis) and therefore barred a scientific career in Soviet Union for him. Fortunately, Levin got a permission to leave Soviet Union and emigrated to US where he got many well known results in different areas of theoretical computer science (about one-way functions, holographic proofs et al.). As Levin recalls, KGB made it known that they think going away would be the best option for him; they even asked Kolmogorov to deliver this advice (which Kolmogorov did, though he did not indicated whether he himself agrees…) Now we can make jokes about these events (Levin once noted that a posteriori Ershov’s behaviour was a favor for him: it was a motivation to leave Soviet Union) but at that time things were much more dramatic. But while being still in the USSR after this unsuccessful defense, Levin followed an advice of some friend, who told that Levin should publish his results while he is still allowed to publish papers in Soviet journals (this was not a joke, the danger was quite real) and published a bunch of papers in 1973–1977. These papers were rather short and cryptic, a lot of things was stated there without proofs, so many ideas from them were really understood only much later. We will trace only two main contributions made in these papers: the prefix complexity, and the randomness criterion in terms of monotone complexity. [^13]: Peter Gacs came to Moscow State University for 1972/3 academic year from Hungary where he became interested in this topic after reading Kolmogorov paper [@kolmogorov-1965], Martin-Löf lecture notes from Erlangen and Zvonkin and Levin’s paper [@zvonkin-levin] and started correspondence with Levin by sending him some paper about randomness characterization in terms of complexity. When Gacs came to Moscow in 1972, Levin explained his criterion of randomness in terms of monotone complexity which looked much better to Gacs so his paper was never published. Then Levin explained the notion of prefix complexity to Gacs and asked whether it is symmetric (with $O(1)$ precision). The negative answer obtained by Gacs became part of his paper [@gacs-1974] that included also some Levin’s results, including the $O(1)$-formula for the prefix complexity of a pair (attributed to Levin). The prefix complexity is very briefly introduced in the beginning of this paper with the remark “considered in detail by Levin”. [^14]: This paper was written [@chaitin-home] in 1974 during the visit to the IBM Watson Lab in Yorktown Heights for a few months. Chaitin’s work there has another important implication: an unpublished manuscript by R. Solovay [@solovay-manuscript]. In his talk [@calude-talk] Cristian Calude tells the story: “When I started reading and trying to understand the subject to write my book “Information and Randomness” [@calude-book], I discussed this with Greg Chaitin and he told me: look, if you really want to write a good book, you have to read Solovay’s manuscript… So I started asking around, and eventually wrote to Solovay: Greg Chaitin told me that I should read your manuscript; could I have a copy? Solovay answered: I had one, but I don’t have it any more. This was in 1991, I think. I tried again to get it and eventually I contacted Charles Bennett, and he had one copy; he was very kind to send me a copy of this copy. That is also an interesting story which Greg Chaitin told me about how this book \[manuscript\] was written. Solovay was for one year at IBM on a sabbatical leave and he was asked to write a report about Chaitin’s work. Probably most of us would write a report of two or three pages and forget forever about it. But Solovay took it very seriously, so he rewrote many parts of the theory in his completely different new style, and he solved also a substantial number of open problems at that stage. This was a kind of shock: look, this guy is so bright, he has nothing to do with this field, he comes, he reads this bunch of papers, he produces this beautiful manuscript solving so many problems and at the end of the day he does not want even to publish anything! Solovay never published this manuscript. I sent Solovay a copy of his ‘lost’ manuscript and he said: well, if you have a student or whoever would like to read and edit and publish the book, fine with me, but I am not interested in working on it. It had to wait until Rod Downey and Denis Hirschfeldt had the force to get through and recuperate most of the results in this manuscript.” [^15]: This argument sounds convincing; however, one may expect that randomness of a binary sequence is invariant under computable permutation of its terms while Schnorr’s criterion of randomness in terms of monotone complexity is not. Recently A. Rumyantsev pointed out the following simple invariant criterion: $\textit{KP}(A,\omega(A))\ge \|A\|-O(1)$. Here $\textit{KP}$ stands for the prefix complexity of a pair; $A$ is a finite set of indices of size $\|A\|$ and $\omega(A)$ is a restriction of $\omega$ onto $A$ (a bit string of length $\|A\|$). [^16]: A similar thing was done once to test early Unix utilities: they were fed with random bits and crashed quite often! In fact, standard programming languages and executable file formats satisfy Chaitin’s requirements for universal machine if we ignore that machine word has finite size, usually between $8$ and $64$ bits.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We present the analysis and results of recent high-energy gamma-ray observations of the BL Lac object 3C 66A conducted with the Solar Tower Atmospheric Cherenkov Effect Experiment (STACEE). During the 2003-2004 observing season, STACEE extensively observed 3C 66A as part of a multiwavelength campaign on the source. A total of 33.7 hours of data was taken on the source, plus an equivalent-duration background observation. After cleaning the data set a total of 16.3 hours of live time remained, and a net on-source excess of 1134 events was seen against a background of 231742 events. At a significance of $2.2$ standard deviations this excess is insufficient to claim a detection of 3C 66A, but is used to establish flux upper limits for the source.' author: - 'D.A. Bramel, J. Carson, C.E. Covault, P. Fortin, D.M. Gingrich, D.S. Hanna, A. Jarvis, J. Kildea, T. Lindner, R. Mukherjee, C. Mueller, R.A. Ong, K. Ragan, R.A. Scalzo, D. A. Williams, J. Zweerink' title: Observations of the BL Lac Object 3C 66A with STACEE --- Introduction ============ To date, all confirmed extragalactic sources of TeV ($10^{12}$ eV) photons have been low-redshift, high-frequency-peaked BL Lac objects (HBLs) [@2004NewAR..48..527H]. Of all BL Lac objects, it is reasonable that nearby HBLs, with their very energetic synchrotron emission, would be the first to be detected at TeV energies. As the energy thresholds of ground-based gamma-ray telescopes decrease, and their sensitivities increase, there is significant potential for growth in the number of very high-energy (VHE) gamma-ray sources to include higher redshift objects and low-frequency BL Lacs (LBLs). The LBL object 3C 66A is a likely candidate for detection with future VHE telescopes because it was detected in the 30 MeV-20 GeV energy band by the EGRET gamma-ray satellite instrument on the Compton Gamma Ray Observatory, has a higher-energy synchrotron peak than most LBL objects, and already has an unconfirmed TeV detection. 3C 66A also has a higher redshift than any confirmed TeV source. In the context of the possible absorption of gamma rays by intergalactic radiation fields, the greater redshift of the source may indicate that detectors with improved sensitivity to energies at or below 100 GeV, such as the Solar Tower Atmospheric Effect Experiment (STACEE), will be able to confirm its detection. 3C 66A was first optically identified by [@1974ApJ...190L..97W]. It is highly variable in the optical and X-ray bands and shows significant optical polarization . It has been extensively observed in the radio and optical, but the host galaxy surrounding the blazar jet has not been resolved. Observations with the VLBA by [@2001ApJS..134..181J] show a highly superluminal jet, confirming that beamed emission likely plays a major role in the observed flux. The redshift for the source is widely quoted at 0.444, but the scant data supporting this value are highly uncertain (see §\[sec:Redshift\]). At MeV-GeV energies, 3C 66A is associated with the EGRET source 3EG J0222+4253 [@1999ApJS..123...79H]. This association is not unique, however, as the error box for 3EG J0222+4253 also covers a nearby pulsar, PSR 0218+42. The EGRET source position is consistent with both 3C 66A and the pulsar, but the significance of the association varies with energy . Position contours derived from low-energy (100-300 MeV) photons favor the pulsar location, while high energy ($>1$ GeV) contours exclude the pulsar and correlate well with 3C 66A. From this, it is concluded that the pulsar is the primary source of the softer photons detected by EGRET, and 3C 66A is the source of the harder component of the spectrum. Thus it can be expected that the 3EG J0222+4253 spectral index ($-2.01\pm0.14$) is a lower limit for the 3C 66A spectral index, and that 3C 66A should produce more high-energy photons than otherwise would be predicted. Above the EGRET energy range, 3C 66A observations have been largely unsuccessful. Repeated detections above 900 GeV by the Crimean Astrophysical Observatory’s GT-48 imaging atmospheric Cherenkov telescope [@1998AstL...24..134N; @2002ARep...46..634S] at an average integral flux level of $2.4\times10^{-11}$ cm$^{-2}$s$^{-1}$ have yet to be confirmed at other observatories. Observations by the Smithsonian Astrophysical Observatory’s Whipple 10-m gamma-ray instrument produced a 99.9% integral flux upper limit above 350 GeV of $<1.9\times10^{-11}$ cm$^{-2}$ s$^{-1}$ in 1993 [@1995ApJ...452..588K] and again of $<0.35\times10^{-11}$ cm$^{-2}$ s$^{-1}$ in 1995 [@2004ApJ...603...51H], while observations by the HEGRA telescope array in 1997 produced a 99% upper limit above 630 GeV of $<1.4\times10^{-11}$ cm$^{-2}$ s$^{-1}$. This uncertainty grants 3C 66A a ‘C-’ rating in the TeV source summary of [@2004NewAR..48..527H], placing it among the most tenuous of TeV sources. As a potential gamma-ray source, 3C 66A is interesting for what it could reveal about the density of extragalactic background light (EBL). At TeV energies, gamma rays can interact with infrared EBL photons, causing a decrease in the observed TeV flux that is related to the column density of the EBL photons [@1999APh....11...93P; @2001ApJ...555..641M]. A redshift of 0.444 would place 3C 66A further away than any confirmed TeV source to date and would make any observed TeV flux highly sensitive to the effects of EBL absorption. Extrapolation of the 3EG spectrum with no EBL absorption predicts a 100 GeV source flux around 0.2 Crab, but actual 3C 66A emission could in fact be higher given the pulsar confusion in the 3EG source. Modeling of the 3C 66A gamma-ray flux by predicts a moderate flux of $7.0-9.6\times10^{-11}$ cm$^{-2}$ s$^{-1}$ above 40 GeV, but negligible flux ($<0.14\times10^{-11}$ cm$^{-2}$ s$^{-1}$) above 300 GeV. Without EBL absorption, 3C 66A has the potential to be an easily detected TeV source and for this reason detections and upper limits of the source are very useful for EBL studies. A low predicted EBL cutoff energy of 100-200 GeV [@1999APh....11...93P] puts 3C 66A out of reach for many older Cherenkov detectors, but newer instruments may be able to push below the cutoff energy. The Solar Tower Atmospheric Cherenkov Effect Experiment (STACEE) is a wavefront-sampling Cherenkov detector sensitive to photons above 100 GeV. No previous observations of 3C 66A have been reported by any instrument with a gamma-ray energy threshold in the 100-300 GeV range. The expected high-energy cutoff for 3C 66A is thus a unique and interesting challenge for lower-energy ground-based Cherenkov detectors such as STACEE. The observations described in this paper were taken as part of a 3C 66A multiwavelength campaign [@2004HEAD....8.0410B]. This campaign took place during the 2003-2004 observing season and included optical monitoring by the Whole Earth Blazar Telescope (WEBT) collaboration, X-ray monitoring by the Rossi X-Ray Timing Explorer (RXTE), VHE gamma-ray observations by STACEE and VERITAS [@2002APh....17..221W], and long-term radio monitoring. In addition, 9 high-spatial-resolution observations using the VLA were carried out during the campaign and throughout 2004 to follow possible structural changes of the source. In this paper we describe only the high-energy gamma-ray observations with STACEE (see also [@MyThesis]). The 3C 66A Redshift {#sec:Redshift} =================== The high redshift associated with 3C 66A is one of the main features that sets it apart from most potential TeV sources. The commonly quoted value of 0.444, we believe, may have caused many in the TeV field to dismiss 3C 66A as an undetectable source due to EBL considerations. As part of our work on this object, we have conducted an extensive literature investigation of the widely quoted redshift value and find that the data that back it up are remarkably uncertain. Although not central to the results published in this paper, it is important that the nature of this redshift value be made known to the blazar community, particularly as 3C 66A may play a significant role in future TeV measurements of EBL absorption. To date, redshift measurements of 3C 66A have been reported in only two papers. The first, [@1978bllo.conf..176M], is most widely credited with the z=0.444 measurement. Unfortunately, the redshift value given in this paper for 3C 66A has been so widely quoted in catalogs and summaries over the intervening years that the qualifications of the original work have been generally disregarded. In the paper, @1978bllo.conf..176M call 3C 66A “one of two [\[]{}objects[\]]{} in our study for which we feel we still have not measured a definitive redshift”, and note of the single emission feature detected on the source that “we are not certain of the reality of the feature.” They summarize with the admonition that “the redshift 0.444 cannot be considered reliable, and the object deserve[\[]{}s[\]]{} more attention.” The second paper on the topic is that of [@1991ApJS...75..645K], which reports on a reprocessing of International Ultraviolet Explorer (IUE) data taken in the early 1980s. In this reanalysis it is noted that a “weak feature near 1750 $\mbox{{\AA}}$ could be Ly$\alpha$ emission at the object redshift of 0.444”, but @1991ApJS...75..645K also indicate that [@1990PASP..102..463C] found a detector artifact at the same location. For a point-source spectrum, @1990PASP..102..463C find that the region around 1750 $\mbox{{\AA}}$ shows an excess of $2-3\times10^{-15}$ erg cm$^{-2}$ s$^{-1}$ $\mbox{{\AA}}^{-1}$ that is created by the IUE instrument itself. The weak feature mentioned by @1991ApJS...75..645K is not more than $3\times10^{-15}$ erg cm$^{-2}$ s$^{-1}$ $\mbox{{\AA}}^{-1}$ above the continuous source background. A close examination of the 3C 66A IUE spectrum in this region reveals a strong resemblance, both in structure and in amplitude, to the IUE point-source artifact spectrum. Without reanalysis that explicitly accounts for detector artifacts, the ability of the IUE data to confirm the redshift is questionable. There are no other detections of emission lines for 3C 66A in the literature, though there have been several attempts (e.g. @1974ApJ...190L..97W [@1976ApJS...31..143W]). The data for both the initial detection and later confirmation of the 0.444 redshift are not, in our opinion, sufficient to ensure a reliable result. We find it reasonable to hold the value of 3C 66A’s redshift in question, and we encourage further spectroscopic observations of this source. The STACEE Detector =================== The Solar Tower Atmospheric Cherenkov Effect Experiment (STACEE) [@NIMpaper; @ChantellEtAl] is operational at the National Solar Thermal Test Facility near Albuquerque, NM. It uses 64 large (37$m^{2}$) heliostat mirrors to focus Cherenkov light from gamma-ray-initiated extensive air showers onto an array of 64 photomultiplier tubes (PMTs), with each heliostat mapped onto a unique PMT. High-speed electronics measure the charges and relative arrival times of the PMT pulses. A multi-level coincidence trigger [@MartinRagan] is used to select Cherenkov events. In this fashion the detector samples the Cherenkov wavefront at 64 separate locations on the ground, making STACEE a wavefront-sampling detector. Other gamma-ray detectors utilizing a similar technique include CELESTE [@2002ApJ...566..343D] and Solar-Two [@2002AAS...200.2503T]. The large mirror area obtained with the heliostats allows for the detection of faint Cherenkov light pulses and grants a lower energy threshold than all but the newest imaging-Cherenkov telescopes. STACEE utilizes a two-level trigger to discriminate Cherenkov shower events from randomly coincident night-sky background (NSB) photons. The 64-channel array is broken up into eight clusters, each containing eight PMT channels. The signals from each PMT are sampled by discriminators set at a fixed threshold of approximately five photoelectrons. The first-level (L1) trigger requires at least five out of eight channels to have a discriminator hit within a 16 ns window. The second-level (L2) trigger then requires at least five of the eight clusters to trigger within the same 16 ns coincidence window before recording a Cherenkov event. For each Cherenkov trigger, 1 GS/s Flash ADCs (FADCs) are used to digitize the pulse on each channel, preserving all the information contained in the wavefront sample. STACEE has been fully operational since 2001, and has detected gamma rays from the Crab [@2001ApJ...547..949O] and Mrk 421 [@2002ApJ...579L...5B]. It has also collected data on a number of active galactic nuclei [@JKHeidelberg]. A full description of the STACEE detector can be found in @GINGRICHNSS. Data and Analysis ================= Observations with STACEE are performed in on-off pairs, whereby a source is observed for 28 minutes followed by a 28-minute observation of an equivalent off-source area of dark sky. Each off-source observation covers the same azimuth and elevation range as the corresponding on-source observation. By using this technique, each instant of on-source data has exactly one equivalent instant of off-source data, and comparisons between halves of a pair are inherently corrected for observation-angle-dependent systematics. The difference between the on-source and off-source shower-detection rates is attributed to photons coming from the direction of the source. Since the off-source observations are required for background estimation, an on-off pair is considered to be the base unit of a STACEE observation. Background for STACEE consists almost entirely of hadron-initiated air showers. During the 2003-2004 observation season, 85 on-off pairs were taken on 3C 66A. These pairs totaled 33.7 hours of on-source live time prior to the data-cleaning process, plus an equal amount of off-source live time. A series of data-cleaning criteria, or cuts, were applied to remove known hardware malfunctions, including removal of sections in which one or more heliostats were not operational, the high voltage system had tripped off, or parts of the data acquisition system were inoperative. After the application of hardware cuts, a total of 29.3 hours on-source remained. Data-quality Cuts ----------------- A set of cuts was applied to the data that were focused on removing apparent or potential problems caused by unfavorable weather conditions. A cut was made to remove all data that could potentially be contaminated by frost build-up on the heliostat mirrors, as even a small amount of condensation on the optics causes significant degradation in the detector performance. The frost potential was determined by looking at the records of the atmospheric frost index (defined to be the difference between the atmospheric temperature and the dew point) during data taking. Any data taken while the frost index was below $5.5\degr$ C were discarded. A second data-quality cut was made to remove data biased by clouds or changing atmospheric opacity. The L1 cluster rates are driven primarily by accidental coincidences stemming from night-sky background (NSB) photons and thus are sensitive to changes in background light levels. At the STACEE site, light pollution is high enough that any increase in sky opacity due to clouds, haze, or other atmospheric phenomena causes a distinct increase in the NSB rate. Accordingly, the L1 trigger rate provides a measure of the stability of the observing conditions. Stable nights produce steady L1 rates, while unstable nights show significant L1 rate fluctuations. Since the NSB noise is elevation-dependent, no attempt was made to constrain L1 rates to an absolute range. Instead, we used the cuts to constrain the correlation between on-source and off-source L1 rates. The data were divided into 30-second intervals, and the average L1 rates on- and off-source were calculated for each interval. A correlation statistic $\theta_{L1}$ was defined for each interval based on the L1 rates $$\theta_{L1}=\ln\left(\frac{L1_{on}}{L1_{off}}\right)\label{eq: L1 Angle}$$ giving a characteristic measure of the correlation between the on-source and off-source L1 rates. Histograms of this statistic for each cluster were then compiled over the entire data set (see Figure \[fig: L1 Angle Graphs\]). For data that are unaffected by sky opacity changes, this distribution of $\theta_{L1}$ for each cluster should be Gaussian, indicating only random variations of the correlation statistic. In practice, a Gaussian peak with extended tails is obtained, indicating non-random variation caused by changing cloud cover. By performing a Gaussian fit to the peaks of the distributions and discarding data in which the correlation statistic of any cluster lies more than two standard deviations from the mean value for that cluster, sections of data that have been contaminated by clouds and haze were removed. A second, similar type of quality cut is made with the occupancy of each channel. The occupancy of a channel is defined as the average fraction of triggers in which the channel has a discriminator hit. With stable observing and hardware conditions, the on-source and off-source occupancies should be highly correlated. We define an occupancy correlation statistic for a single channel similar to that of equation \[eq: L1 Angle\] $$\theta_{occ}=\ln\left(\frac{occ_{on}}{occ_{off}}\right)\label{eq: Occ angle}$$ and cut on the distribution of this quantity in the same way as with the L1 statistic. Because of the larger number of channels that are being constrained, there is a higher likelihood of a large fluctuation in good-quality data than with the L1 cuts. To avoid eliminating these data, we remove only data with three or more channels having occupancy statistic values more than three standard deviations from the mean value for that channel or with any one channel having an occupancy statistic value more than five standard deviations from the mean value for that channel. After application of all data-quality selection cuts, a total of 16.8 hours of on-source data remained. Padding ------- The measured background rate can be systematically biased by variations in the brightness of the sky in the field of view of the detector, often referred to as the field brightness. When an observing field is bright, the increased NSB photon count results in increased levels of event promotions (dim Cherenkov showers that trigger simply because they are boosted above threshold by random NSB photons). Thus we expect bright fields to yield higher trigger rates than dark fields. The field brightness can change between on-source and off-source observations due to weather instability, but even very stable nights show a field brightness difference due to the different stars in each field. The brightness difference due to field stars is often rather small: the number and magnitude of stars falling inside the $\sim1\degr$ STACEE field of view tend to be roughly constant, and the promotion contributions from each field cancel each other out. However, for certain sources an individual bright star in one field upsets the balance (see, e.g. @2002ApJ...579L...5B); the result is a highly distorted background measurement that, if followed blindly, results in a spurious source excess or deficit. To counter the effects of field brightness differences, a technique called software padding is employed. Software padding effectively increases the light level of the darker half of the pair by adding a sample trace containing only NSB background to each event’s FADC trace. Once the light level is increased, a software trigger criterion is applied to both halves of each pair. This padding technique, described in detail in @2004ApJ...607..778S, has been shown to effectively remove the effects of field brightness differences in STACEE data. Data used in the 3C 66A analysis presented here were padded using the waveform library technique of @2004ApJ...607..778S. Cuts to remove sections of the data where the padding algorithm was unable to execute were also applied, leaving a total net live time of 16.3 hours on-source. Results ======= After all time cuts and padding, we are left with 1134 excess on-source events over a net observation time of 16.3 hours. The excess events are against an off-source background of 231742 events, yielding an on-source excess significance of $2.2$ standard deviations using the method of [@1983ApJ...272..317L]. This excess significance is not sufficient to claim a detection of the source. There were no significant transient events in the data, as shown by the histogram of significances for each of the 85 pairs in Figure \[fig: Pairwise Sigma\]. Detector Simulations ==================== In order to best understand the results of the observations of 3C 66A, simulations of the STACEE detector were carried out to closely mimic the 3C 66A data set. Using the CORSIKA air-shower simulation package [@CORSIKA], sets of showers were simulated with gamma-ray, proton, and helium primaries. These were generated over a range of energies and source hour angle ($\mathcal{H}$) positions. Cherenkov photons from each shower were passed through a custom-made optics simulation package and converted into photoelectrons by simulated PMTs. These resultant photoelectrons were run through a custom-made electronics simulation to determine which showers would have triggered the array. Parameters for the custom simulation packages were specified to match operational detector parameters as closely as possible. All fixed detector parameters, such as coincidence conditions and electronics configurations, were set in simulation to be identical to the detector as it was during data taking. Variable parameters that affect detector performance, such as PMT currents, were set to the average values taken from data with hour angle similar to that of the simulated showers. In this way, each simulated shower was processed with simulated parameters as close as possible to the data it was intended to simulate. The simulation is able to reproduce several diagnostic quantities seen in the data, including L1 rates and cosmic-ray trigger rate as a function of source hour angle. Once the simulated showers were processed through the simulation pipeline, analysis was carried out in the same manner as for real data. Effective Area -------------- To determine the sensitivity of the STACEE detector we have to determine its effective area, the sensitive area presented to incoming gamma rays. The effective area of the array was determined for each shower type by scattering simulated showers across the detector, then multiplying the triggered fraction by the total scattering area. Scattering areas were circular in the plane normal to the arrival direction of the primary particle, with radii sufficiently large that triggers near the edge of the area were negligible. Effective areas were calculated as a function of energy $E$ at several hour angles $\mathcal{H}$ for each shower primary type. We interpolated between the discrete effective area points and then weighted the effective area by the source observation time to obtain a net effective area as a function of energy$$A_{Eff}\left(E\right)=\frac{\int A_{Eff}\left(\mathcal{H},E\right)\times X\left(\mathcal{H}\right)d\mathcal{H}}{\int X\left(\mathcal{H}\right)d\mathcal{H}}\label{eq: Net Eff. Area}$$ where the exposure $X\left(\mathcal{H}\right)$ was taken from the post-cut live time in the real data set. This net effective area, computed separately for each primary particle type, is the effective area of STACEE relevant for the 3C 66A data set. Energy Threshold and Flux Upper Limit ------------------------------------- Using the simulated effective area we calculated detector energy thresholds and source flux upper limits. Following convention, the energy threshold $E_{th}$ is defined as the peak of the response curve generated when the effective area of the detector is convolved with a source spectrum, as shown in Figure \[fig: gamma response\]. The lack of any known spectral properties for 3C 66A in the STACEE energy range adds a degree of uncertainty to the estimation of the energy threshold and flux upper limits. These estimates depend heavily on the assumed shape of the source spectrum. At EGRET energies, the spectrum of 3C 66A is known to be quite hard, fitting a power-law photon differential index no softer than $-2.01\pm0.14$. At higher energies above 20 GeV the spectrum no doubt falls off more steeply due to intrinsic softening and EBL absorption, but neither of these effects is well constrained. The amount of EBL absorption is particularly uncertain because the redshift value of the source is, as argued above, very unreliable. We present in Table \[tab: Ethresh + Flux CI\] the energy thresholds and flux upper limits derived from the observations, for a variety of power-law and EBL-absorbed power-law spectra. [ccccc]{} $-2.0$&$200$&$<1.0$&$150$&$<1.9$\ $-2.5$&$184$&$<1.2$&$150$&$<1.9$\ $-3.0$&$150$&$<1.7$&$142$&$<2.1$\ $-3.5$&$147$&$<1.8$&$137$&$<2.3$\ \[tab: Ethresh + Flux CI\] Summary and Discussion ====================== The STACEE experiment observed the BL Lac object 3C 66A for a total of 33.7 hours in the fall of 2003. After all cuts and padding, 16.3 hours of data yielded an on-source excess with a significance of 2.2 standard deviations, consistent with no detected flux. Flux upper limits derived from simulated effective areas are given in Table \[tab: Ethresh + Flux CI\]. Figure \[fig: SED\] shows the full spectral energy distribution for 3C 66A, made using non-contemporaneous data. Radio - X-ray data have been compiled from the literature and are plotted with 1$\sigma$ error bars. Various EGRET detections are shown as open symbols. These include the data of , who estimated and subtracted the contribution from the nearby pulsar PSR 0218+42; note that the pulsar-subtracted measurement suggests that the high energy peak continues to rise steeply into the STACEE energy band. The STACEE limits (plus signs) assume an unabsorbed spectrum with the four spectral indices listed in Table 1. They are at a lower energy threshold and higher flux than the TeV limits from the Whipple telescope [@2004ApJ...603...51H] and HEGRA telescope array . A homogeneous, one-zone, synchrotron self-Compton model from is also shown. All of the gamma-ray observations to date, including the STACEE flux limits, are consistent with this model. In conjunction with simultaneous data at other wavelengths, the STACEE limits presented in Table \[tab: Ethresh + Flux CI\] have the potential to constrain models of the source emission mechanism. However, very little physical modeling of 3C 66A has been published in the literature. Modeling of high-energy gamma-ray emission from BL Lac objects is very complex and depends extensively on simultaneous constraints from lower-energy emission. The 2003-2004 multiwavelength campaign, of which the STACEE observations are a part, will produce the first simultaneous set of broadband spectral data on 3C 66A from optical to gamma-ray energies. Broadband modeling has been successful when applied to simultaneous multiwavelength observations of sources similar to 3C 66A, such as W Comae and BL Lacertae. All three of these BL Lacs share the similarity that their synchrotron emission is dominant at X-ray energies. According to the ROSAT all-sky survey data, the X-ray spectrum of 3C 66A is soft (spectral index of 1.6) [@1998MNRAS.299..433F], indicating that the synchrotron component of 3C 66A extends to the X-ray regime. The ROSAT measurement of a soft spectrum for 3C 66A is supported by more recent observations by XMM [@2003MNRAS.346.1041C] and by BEPPO/SAX . Higher-energy synchrotron emission by these LBLs indicates that they lie closer to the HBL end of the BL Lac spectrum than most LBLs and thus are more likely to have significant TeV emission. In the past, W Comae and BL Lacertae have been the subject of extensive multiwavelength campaigns. These data have been interpreted using the fully time-dependent leptonic jet simulation code of @2002ApJ...581..127B as well as hadronic Synchrotron-Proton Blazar (SPB) models [@2002astro.ph..6164M]. In the case of W Comae, time-dependent modeling of the X-ray variability of the source was found to yield quite different model predictions of the GeV-TeV flux for hadronic or leptonic models [@2002ApJ...581..143B]. W Comae has thus proved to be a promising target for VHE gamma-ray and coordinated broadband observations, as it may serve to distinguish between leptonic and hadronic jet models for blazars. STACEE flux limits on the $>$100 GeV emission from W Comae begin to constrain hadronic emission models [@2004ApJ...607..778S]. Similarly, BL Lacertae was also the subject of an extensive multiwavelength monitoring campaign in 2000, and the data were modeled using leptonic and hadronic jet models to fit the observed broadband spectra and spectral variability patterns [@2004ApJ...609..576B]. Such multiwavelength modeling studies with blazars tell us about cooling timescales, magnetic fields, and Doppler factors associated with blazar jets. Hadronic models for blazar jets predict significantly greater TeV flux than leptonic models, and these differences can be tested by future TeV observations of BL Lac by VERITAS or MAGIC [@2004MmSAI..75..232M]. Although such modeling is outside the scope of this current work, studies are in progress to understand data from the recent 3C 66A multiwavelength campaign using the fully time-dependent leptonic jet simulation code of @2002ApJ...581..127B and to use the data from radio through X-ray energies to make predictions about high-energy emission [@3C_MW_model in prep.]. The implications of the STACEE upper limits on 3C 66A could then be evaluated in the context of the properties of the relativistic jet in 3C 66A, testing leptonic and hadronic models for this source. Model-dependent flux predictions at VERITAS energies can also be made, given a specific model and the STACEE flux upper limits. argue that the multiwavelength spectrum and modeling studies of 3C 66A could be applied toward resolving the controversy regarding the identification of 3C 66A/PSR 0218+4232 with the EGRET source. With the pulsar accounting for a majority of the low-energy EGRET photons, the spectral energy distribution (SED) of 3C 66A in the EGRET energy range must be steeply rising. point out that it is difficult to fit the 3C 66A SED from to a log-parabolic approximation of the synchrotron self-Compton (SSC) peak. A simple, smooth model of the SSC peak favors a 3C 66A spectrum that included the photons attributed to the pulsar by . Simultaneous modeling of the broadband 3C 66A spectrum with data at X-ray and TeV energies would be able to help resolve the issue by showing whether or not an additional component is able to fit the spectrum. Detection of $>$100 GeV photons from 3C 66A would be a very interesting result for gamma-ray astronomy. Currently all blazars detected by TeV experiments have been high-frequency-peaked BL Lac (HBL) objects; 3C 66A is an LBL object, with a synchrotron peak at lower energies than HBL objects. Confirmation of $>$100 GeV photons from this source would open the door on a new class of objects for TeV astronomy. In addition, if the redshift of 0.444 is correct, such a confirmation would make 3C 66A the most distant known TeV source. This would make it a very interesting source for constraining EBL spectral models. It is our hope that the results presented here will motivate further study of this source with a more sensitive detector. The possibility that there may be detectable flux at energies near 100 GeV makes 3C 66A an interesting target for the new generation of imaging Cherenkov telescopes such as VERITAS and MAGIC. Together with GLAST [@2004HEAD....8.3006T] and AGILE [@2004HEAD....8.1605P] at lower energies, observations of this source will provide a comprehensive data set at sub-TeV gamma-ray energies that will be important for modeling studies. [44]{} natexlab\#1[\#1]{} , F. A., [Akhperjanian]{}, A. G., [Barrio]{}, J. A., [Bernl[" o]{}hr]{}, K., [Bojahr]{}, H., [Calle]{}, I., [Contreras]{}, J. L., [Cortina]{}, J., [Daum]{}, A., [Deckers]{}, T., [Denninghoff]{}, S., [Fonseca]{}, V., [Gonzalez]{}, J. C., [Heinzelmann]{}, G., [Hemberger]{}, M., [Hermann]{}, G., [He[ß]{}]{}, M., [Heusler]{}, A., [Hofmann]{}, W., [Hohl]{}, H., [Horns]{}, D., [Ibarra]{}, A., [Kankanyan]{}, R., [Kestel]{}, M., [Kettler]{}, J., [K[" o]{}hler]{}, C., [Konopelko]{}, A., [Kornmeyer]{}, H., [Kranich]{}, D., [Krawczynski]{}, H., [Lampeitl]{}, H., [Lindner]{}, A., [Lorenz]{}, E., [Magnussen]{}, N., [Mang]{}, O., [Meyer]{}, H., [Mirzoyan]{}, R., [Moralejo]{}, A., [Padilla]{}, L., [Panter]{}, M., [Petry]{}, D., [Plaga]{}, R., [Plyasheshnikov]{}, A., [Prahl]{}, J., [P[" u]{}hlhofer]{}, G., [Rauterberg]{}, G., [Renault]{}, C., [Rhode]{}, W., [R[" o]{}hring]{}, A., [Sahakian]{}, V., [Samorski]{}, M., [Schmele]{}, D., [Schilling]{}, M., [Schr[" o]{}der]{}, F., [Stamm]{}, W., [V[" o]{}lk]{}, H. J., [Wiebel-Sooth]{}, B., [Wiedner]{}, C., [Willmer]{}, M., & [Wittek]{}, W. 2000, , 353, 847 , M. 2005, in prep. , M. & [Chiang]{}, J. 2002, , 581, 127 , M., [Mukherjee]{}, R., & [Reimer]{}, A. 2002, , 581, 143 , M. & [Reimer]{}, A. 2004, , 609, 576 , M., [Joshi]{}, M., [Fossati]{}, G., [Smith]{}, I. A., [Mukherjee]{}, R., [Bramel]{}, D., [Cui]{}, W., & [WEBT Collaboration]{}. 2004, AAS/High Energy Astrophysics Division, 8. , L. M., [Hinton]{}, J. A., [Bramel]{}, D., [Chae]{}, E., [Covault]{}, C. E., [Fortin]{}, P., [Gingrich]{}, D. M., [Hanna]{}, D. S., [Mukherjee]{}, R., [Mueller]{}, C., [Ong]{}, R. A., [Ragan]{}, K., [Scalzo]{}, R. A., [Schuette]{}, D. R., [Th[' e]{}oret]{}, C. G., & [Williams]{}, D. A. 2002, , 579, L5. Bramel, D. 2005, Ph.D. Thesis, Columbia University. , M. C., [Bhattacharya]{}, D., [Covault]{}, C. E., [Dragovan]{}, M., [Fernholz]{}, D. T., [Gregorich]{}, D. T., [Hanna]{}, D. S., [Marion]{}, G. H., [Ong]{}, R. A., [Oser]{}, S. M., [Tumer]{}, T. O., & [Williams]{}, D. A. 1998, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 408, 468 , L. & [Ghisellini]{}, G. 2002, , 384, 56 , D. M., [Bruegman]{}, O. W., & [Norman]{}, D. J. 1990, , 102, 463 , J. H., [Hardcastle]{}, M. J., [Birkinshaw]{}, M., & [Worrall]{}, D. M. 2003, , 346, 1041 , M., [Holder]{}, J., [Bazer-Bachi]{}, R., [Bergeret]{}, H., [Bruel]{}, P., [Cordier]{}, A., [Debiais]{}, G., [Dezalay]{}, J.-P., [Dumora]{}, D., [Durand]{}, E., [Eschstruth]{}, P., [Espigat]{}, P., [Fabre]{}, B., [Fleury]{}, P., [H[' e]{}rault]{}, N., [Hrabovsky]{}, M., [Incerti]{}, S., [Le Gallou]{}, R., [M[" u]{}nz]{}, F., [Musqu[\` e]{}re]{}, A., [Olive]{}, J.-F., [Par[' e]{}]{}, E., [Qu[' e]{}bert]{}, J., [Rannot]{}, R. C., [Reposeur]{}, T., [Rob]{}, L., [Roy]{}, P., [Sako]{}, T., [Schovanek]{}, P., [Smith]{}, D. A., [Snabre]{}, P., & [Volte]{}, A. 2002, , 566, 343 Dingus, B.L., Bertsch, D.L., Digel, S.W., Eposito, J.A., Fichtel, J.M., Fierro, J.M., Hartman, R.C., Hunter, S.D., Kanbach, G., Kniffen, D.A., Lin, Y.C., Mattox, J.R., Mayer-Hasselwander, H.A., Michelson, P.F., von Montigny, C., Mukherjee, R., Nolan, P.L., Schneid, E., Sreekumar, P., Thompson, D.J., Willis, T.D. 1996, , 467, 589. , G., [Maraschi]{}, L., [Celotti]{}, A., [Comastri]{}, A., & [Ghisellini]{}, G. 1998, , 299, 433 , D. M., [Boone]{}, L. M., [Bramel]{}, D. A., [Carson]{}, J., [Covault]{}, C. E., [Fortin]{}, P. F., [Hanna]{}, D. S., [Hinton]{}, J. A., [Jarvis]{}, A., [Kildea]{}, J., [Lindner]{}, T., [Mueller]{}, C., [Mukherjee]{}, R., [Ong]{}, R. A., [Ragan]{}, K., [Scalzo]{}, R. A., [Theoret]{}, C. G., [Williams]{}, D. A., & [Zweerink]{}, J. A. 2004, in Proceedings of the 2004 IEEE Nuclear Science Symposium and Medical Imaging Conference , D. S., [Bhattacharya]{}, D., [Boone]{}, L. M., [Chantell]{}, M. C., [Conner]{}, Z., [Covault]{}, C. E., [Dragovan]{}, M., [Fortin]{}, P., [Gregorich]{}, D. T., [Hinton]{}, J. A., [Mukherjee]{}, R. M., [Ong]{}, R. A., [Oser]{}, S. M., [Ragan]{}, K., [Scalzo]{}, R. A., [Schuette]{}, D. R., [Theoret]{}, C. G., [Tumer]{}, T. O., [Williams]{}, D. A., & [Zweerink]{}, J. A. 2002, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 491, 126 , R. C., [Bertsch]{}, D. L., [Bloom]{}, S. D., [Chen]{}, A. W., [Deines-Jones]{}, P., [Esposito]{}, J. A., [Fichtel]{}, C. E., [Friedlander]{}, D. P., [Hunter]{}, S. D., [McDonald]{}, L. M., [Sreekumar]{}, P., [Thompson]{}, D. J., [Jones]{}, B. B., [Lin]{}, Y. C., [Michelson]{}, P. F., [Nolan]{}, P. L., [Tompkins]{}, W. F., [Kanbach]{}, G., [Mayer-Hasselwander]{}, H. A., [M[" u]{}cke]{}, A., [Pohl]{}, M., [Reimer]{}, O., [Kniffen]{}, D. A., [Schneid]{}, E. J., [von Montigny]{}, C., [Mukherjee]{}, R., & [Dingus]{}, B. L. 1999, , 123, 79 , D., [Knapp]{}, J., [Capdevielle]{}, J., [Schatz]{}, G., & [Thouw]{}, T. 1998, T. Rep. FZKA 6019 , D., [Badran]{}, H. M., [Bond]{}, I. H., [Boyle]{}, P. J., [Bradbury]{}, S. M., [Buckley]{}, J. H., [Carter-Lewis]{}, D. A., [Catanese]{}, M., [Celik]{}, O., [Cui]{}, W., [Daniel]{}, M., [D’Vali]{}, M., [de la Calle Perez]{}, I., [Duke]{}, C., [Falcone]{}, A., [Fegan]{}, D. J., [Fegan]{}, S. J., [Finley]{}, J. P., [Fortson]{}, L. F., [Gaidos]{}, J. A., [Gammell]{}, S., [Gibbs]{}, K., [Gillanders]{}, G. H., [Grube]{}, J., [Hall]{}, J., [Hall]{}, T. A., [Hanna]{}, D., [Hillas]{}, A. M., [Holder]{}, J., [Jarvis]{}, A., [Jordan]{}, M., [Kenny]{}, G. E., [Kertzman]{}, M., [Kieda]{}, D., [Kildea]{}, J., [Knapp]{}, J., [Kosack]{}, K., [Krawczynski]{}, H., [Krennrich]{}, F., [Lang]{}, M. J., [Le Bohec]{}, S., [Linton]{}, E., [Lloyd-Evans]{}, J., [Milovanovic]{}, A., [Moriarty]{}, P., [Muller]{}, D., [Nagai]{}, T., [Nolan]{}, S., [Ong]{}, R. A., [Pallassini]{}, R., [Petry]{}, D., [Power-Mooney]{}, B., [Quinn]{}, J., [Quinn]{}, M., [Ragan]{}, K., [Rebillot]{}, P., [Reynolds]{}, P. T., [Rose]{}, H. J., [Schroedter]{}, M., [Sembroski]{}, G. H., [Swordy]{}, S. P., [Syson]{}, A., [Vassiliev]{}, V. V., [Wakely]{}, S. P., [Walker]{}, G., [Weekes]{}, T. C., & [Zweerink]{}, J. 2004, , 603, 51 , D. & [Weekes]{}, T. C. 2004, New Astronomy Review, 48, 527 , S. G., [Marscher]{}, A. P., [Mattox]{}, J. R., [Wehrle]{}, A. E., [Bloom]{}, S. D., & [Yurchenko]{}, A. V. 2001, , 134, 181 , A. D., [Akerlof]{}, C. W., [Biller]{}, S., [Buckley]{}, J., [Carter-Lewis]{}, D. A., [Cawley]{}, M. F., [Chantell]{}, M., [Connaughton]{}, V., [Fegan]{}, D. J., [Fennell]{}, S., [Gaidos]{}, J., [Hillas]{}, A. M., [Kwok]{}, P. W., [Lamb]{}, R. C., [Lappin]{}, T., [Lessard]{}, R., [McEnery]{}, J., [Meyer]{}, D. I., [Mohanty]{}, G., [Quinn]{}, J., [Rose]{}, H. J., [Rovero]{}, A. C., [Sembroski]{}, G., [Schubnell]{}, M. S., [Punch]{}, M., [Weekes]{}, T. C., [West]{}, M., [Wilson]{}, C., & [Zweerink]{}, J. 1995, , 452, 588 , J., [Alabiso]{}, A., [Bramel]{}, D., [Carson]{}, J., [Covault]{}, C. E., [Fortin]{}, P., [Gingrich]{}, D. M., [Hanna]{}, D., [Jarvis]{}, A., [Lindner]{}, T., [Mueller]{}, C., [Mukherjee]{}, R., [Ong]{}, R. A., [Ragan]{}, K. J., [Scalzo]{}, R. A., [Williams]{}, D. A., & [Zweerink]{}, J. 2004, in Gamma 2004 International Symposium on High Energy Gamma-Ray Astronomy , A. L., [Bohlin]{}, R. C., [Blades]{}, J. C., & [York]{}, D. G. 1991, , 75, 645 , L., [Hermsen]{}, W., [Verbunt]{}, F., [Thompson]{}, D. J., [Stairs]{}, I. H., [Lyne]{}, A. G., [Strickman]{}, M. S., & [Cusumano]{}, G. 2000, , 359, 615 Lamb, R.C. & Macomb, D.J. 1997, , 488, 872. , T.-P. & [Ma]{}, Y.-Q. 1983, , 272, 317 , D., [Garilli]{}, B., [Schild]{}, R., & [Tarenghi]{}, M. 1987, , 178, 21 , M. A. & [Stecker]{}, F. W. 2001, , 555, 641 , J. P. & [Ragan]{}, K. 2000, in Proc. IEEE Nuclear Science Symposium (Lyon) , J. S., [French]{}, H. B., & [Hawley]{}, S. A. 1978, in Pittsburgh Conference on BL Lac Objects, Pittsburgh, Pa., April 24-26, 1978, Proceedings. (A79-30026 11-90) Pittsburgh, Pa., University of Pittsburgh, 1978, p. 176-187; Discussion, p. 187-191., 176–187 , A. & [the MAGIC collaboration]{}. 2004, Memorie della Societa Astronomica Italiana, 75, 232 , A., [Protheroe]{}, R. J., [Engel]{}, R., [Rachen]{}, J. P., & [Stanev]{}, T. 2002, ArXiv Astrophysics e-prints , Y. I., [Stepanyan]{}, A. A., [Kalekin]{}, O. P., [Fomin]{}, V. P., [Chalenko]{}, N. N., & [Shitov]{}, V. G. 1998, Astronomy Letters, 24, 134 , S., [Bhattacharya]{}, D., [Boone]{}, L. M., [Chantell]{}, M. C., [Conner]{}, Z., [Covault]{}, C. E., [Dragovan]{}, M., [Fortin]{}, P., [Gregorich]{}, D. T., [Hanna]{}, D. S., [Mukherjee]{}, R., [Ong]{}, R. A., [Ragan]{}, K., [Scalzo]{}, R. A., [Schuette]{}, D. R., [Th[' e]{}oret]{}, C. G., [T[" u]{}mer]{}, T. O., [Williams]{}, D. A., & [Zweerink]{}, J. A. 2001, , 547, 949 , M., [Massaro]{}, E., [Giommi]{}, P., [Capalbi]{}, M., [Nesci]{}, R., [Tagliaferri]{}, G., [Ghisellini]{}, G., [Ravasio]{}, M., & [Miller]{}, H. R. 2003, , 407, 453 , C., [Tavani]{}, M., & [AGILE Team]{}. 2004, AAS/High Energy Astrophysics Division, 8, , J. R., [Bullock]{}, J. S., [Somerville]{}, R. S., & [MacMinn]{}, D. 1999, Astroparticle Physics, 11, 93 , R. A., [Boone]{}, L. M., [Bramel]{}, D., [Carson]{}, J., [Covault]{}, C. E., [Fortin]{}, P., [Gauthier]{}, G., [Gingrich]{}, D. M., [Hanna]{}, D., [Jarvis]{}, A., [Kildea]{}, J., [Lindner]{}, T., [Mueller]{}, C., [Mukherjee]{}, R., [Ong]{}, R. A., [Ragan]{}, K. J., [Williams]{}, D. A., & [Zweerink]{}, J. 2004, , 607, 778 , A. A., [Neshpor]{}, Y. I., [Andreeva]{}, N. A., [Kalekin]{}, O. R., [Zhogolev]{}, N. A., [Fomin]{}, V. P., & [Shitov]{}, V. G. 2002, Astronomy Reports, 46, 634 , L. O., [Sillanpaeae]{}, A., [Pursimo]{}, T., [Lehto]{}, H. J., [Nilsson]{}, K., [Teerikorpi]{}, P., [Heinaemaeki]{}, P., [Lainela]{}, M., [Kidger]{}, M., [de Diego]{}, J. A., [Gonzalez-Perez]{}, J. N., [Rodriguez-Espinosa]{}, J.-M., [Mahoney]{}, T., [Boltwood]{}, P., [Dultzin-Hacyan]{}, D., [Benitez]{}, E., [Turner]{}, G. W., [Robertson]{}, J. W., [Honeycut]{}, R. K., [Efimov]{}, Y. S., [Shakhovskoy]{}, N., [Charles]{}, P. A., [Schramm]{}, K. J., [Borgeest]{}, U., [Linde]{}, J. V., [Weneit]{}, W., [Kuehl]{}, D., [Schramm]{}, T., [Sadun]{}, A., [Grashuis]{}, R., [Heidt]{}, J., [Wagner]{}, S., [Bock]{}, H., [Kuemmel]{}, M., [Pfeiffer]{}, M., [Heines]{}, A., [Fiorucci]{}, M., [Tosti]{}, G., [Ghisellini]{}, G., [Raiteri]{}, C. M., [Villata]{}, M., [de Francesco]{}, G., [Bosio]{}, S., [Latini]{}, G., [Poyner]{}, G., [Aller]{}, M. F., [Aller]{}, H. D., [Hughes]{}, P., [Valtaoja]{}, E., [Teraesranta]{}, H., & [Tornikoski]{}, M. 1996, , 120, 313 , D. J. & [GLAST Large Area Telescope Collaboration]{}. 2004, AAS/High Energy Astrophysics Division, 8, , S. M., [Bhattacharya]{}, D., [Lizarazo]{}, J., [Marleau]{}, P., [Mohanty]{}, G., [Mohideen]{}, U., [Murray]{}, P., [Tom]{}, H., [Tumer]{}, T., [Xing]{}, G., & [Zweerink]{}, J. 2002, Bulletin of the American Astronomical Society, 34, 676 , T. C., [Badran]{}, H., [Biller]{}, S. D., [Bond]{}, I., [Bradbury]{}, S., [Buckley]{}, J., [Carter-Lewis]{}, D., [Catanese]{}, M., [Criswell]{}, S., [Cui]{}, W., [Dowkontt]{}, P., [Duke]{}, C., [Fegan]{}, D. J., [Finley]{}, J., [Fortson]{}, L., [Gaidos]{}, J., [Gillanders]{}, G. H., [Grindlay]{}, J., [Hall]{}, T. A., [Harris]{}, K., [Hillas]{}, A. M., [Kaaret]{}, P., [Kertzman]{}, M., [Kieda]{}, D., [Krennrich]{}, F., [Lang]{}, M. J., [LeBohec]{}, S., [Lessard]{}, R., [Lloyd-Evans]{}, J., [Knapp]{}, J., [McKernan]{}, B., [McEnery]{}, J., [Moriarty]{}, P., [Muller]{}, D., [Ogden]{}, P., [Ong]{}, R., [Petry]{}, D., [Quinn]{}, J., [Reay]{}, N. W., [Reynolds]{}, P. T., [Rose]{}, J., [Salamon]{}, M., [Sembroski]{}, G., [Sidwell]{}, R., [Slane]{}, P., [Stanton]{}, N., [Swordy]{}, S. P., [Vassiliev]{}, V. V., & [Wakely]{}, S. P. 2002, Astroparticle Physics, 17, 221 , B. J. & [Wills]{}, D. 1974, , 190, L97+ , D. & [Wills]{}, B. J. 1976, , 31, 143	{ "pile_set_name": "ArXiv" }
--- author: - 'Neeraj Gupta, Raghunathan Srianand, Patrick Petitjean & Cédric Ledoux' date: 'Received date/ Accepted date' title: ' Outflowing material in the = 4.92 BAL QSO SDSS J160501.21$-$011220.0[^1] ' --- Introduction ============ Broad absorption line systems (BALs) seen in the spectra of QSOs are characterized by absorption features with large velocity widths ( $\Delta v \sim$ a few $1000$ ) and, usually, high ionization states (Turnshek et al. 1988; Weymann et al. 1991). The gas giving rise to BALs is believed to be material ejected by the quasar but still located very close to the central regions. Thus studying metallicities in the absorbing gas is a direct probe of the chemical enrichment in the very central regions of AGNs (see Hamann & Ferland 1999 for the review). The associated absorption line systems by and large show large metallicities with a tendency of nitrogen being over-abundant with respect to oxygen and other $\alpha-$ process elements (e.g. Petitjean et al. 1994; Korista et al. 1996; Hamann 1997; Petitjean & Srianand 1999). Detailed studies of emission line properties of $z\ge 4$ QSOs have revealed Super-Solar metallicities in the line emitting gas in broad line regions (BLR), suggesting very rapid star formation process (e.g. Hamann & Ferland 1992; Dietrich et al. 2003). However, no such metallicity estimates at such high redshifts based on absorption lines using echelle spectra are available. Unlike in the case of most normal galaxies, it is believed that the high metal enrichment usually observed in QSOs has taken place in less than a few 10$^8$ years through rapid star formation similar to what is expected to happen in the center of ellipticals (Hamann & Ferland 1993; Matteucci & Padovani 1993). Thus, studying BAL QSOs at $ \ge 5$, where the age of the Universe is t$\le 1.07\times 10^9$ yr (for H$_{\rm o}$ = 75 km s$^{-1}$ Mpc$^{-1}$, $\Omega_\Lambda$ = 0.70, $\Omega_{\rm m}$ = 0.30), is important for understanding the star-formation history at a time close to the epoch of reionization and possibly also for constraining cosmological parameters. In the initial list of SDSS QSOs (Fan et al. 2000) there are few QSOs at $z\ge 4$ with associated absorption lines. We obtained a UVES spectrum of the QSO SDSS J160501.21-011220.0, the highest redshift BAL QSO known at that time, with the aim to study the outflowing gas in detail. In order to derive a realistic estimate of the absorbing gas metallicity one needs to have a good handle on the ionization corrections. Associated systems being close to the QSO are most probably ionized by the QSO light rather than the diffuse intergalactic background. However the ionizing spectrum from the BAL QSOs is poorly known. It is known that BAL QSOs are under-luminous in X-rays (Bregman 1984; Singh et al. 1987; Green & Mathur 1996). Recent, deep Chandra observations have shown that the optical to X-ray spectral index, $\alpha_{OX}$, measured for BAL QSOs is systematically lower than that of non-BAL QSOs (Green et al. 2001). It is not clear whether this X-ray weakness is due to the QSO being intrinsically X-ray quiet (i.e., this is a property of the central engine) or due to line-of-sight absorption (i.e., radiative transfer effects). Using a complete sample of optically selected QSOs, Brandt et al. (2000) have shown a significant anti-correlation between $\alpha_{OX}$ and the total equivalent width of the absorption lines. In addition it is noticed that the emission line properties of the BAL and non-BAL QSOs are [very similar]{} even though the observed X-ray properties differ significantly (Weymann et al. 1991; Korista et al. 1993). Thus, it is most likely that the relative X-ray weakness of BAL QSOs is due to intrinsic absorption. It must be noted that all existing models of BAL systems assume standard QSO spectrum. However estimated metallicities will be different if the intrinsic $\alpha_{OX}$ is different from that of a typical QSO spectrum. Indeed, Srianand & Petitjean (2000) have shown that the inferred metallicities decrease with decreasing value of $\alpha_{OX}$. Also, if the absorption is responsible for the suppression of X-rays then it is important to investigate whether it is caused by the gas responsible for the BAL troughs (common absorbers) or by a distinct gas component. It is likely that the common absorber picture is correct for the BAL towards PHL 5200 (Mathur et al. 1995) and that a distinct Compton thick screen is required for the BAL towards PG 0946+301 (Gallagher et al. 1999; Mathur et al. 2000; see also Kraemer et al. 2002). In this paper we present a detailed analysis of two well detached BAL outflow components that are seen in the spectrum of QSO J160501.21-011220.0 ($z_{em}$ = 4.92). Details of the observations are discussed in Sect. 2 which is followed by the description of the BALs and column density estimates in Sect. 3. In Sect. 4 we investigate the physical conditions of the gas.The results are summarised in Sect. 5. Observations and data reduction =============================== The Ultraviolet and Visible Echelle Spectrograph (UVES; Dekker et al. 2000), installed at the ESO VLT 8.2 m telescope, unit Kueyen, on Mount Paranal in Chile, was used on June 15-17, 2001, to obtain high resolution spectra of QSO J160501.21$-$011220.0. A non-standard setting with cross-disperser \#4 and central wavelength 8420 Å was used in the Red arm of UVES. Full wavelength coverage was obtained this way from 6514 to 8300 Å and from 8494 to 9300 Åaccounting for the gap between the two Red-arm CCDs. The CCD pixels were binned $2\times 3$ (namely, twice in the spatial direction and three times in the dispersion direction) and the slit width was either fixed to $1\farcs 2$ or $1\farcs 5$, yielding an overall spectral resolution $R\sim 30000$. The total integration time 3h30min was split into 3 exposures. The data were reduced in the dedicated context of MIDAS [^2], the ESO data reduction system, using the UVES pipeline (Ballester et al. 2000) in an interactive mode. The main characteristics of the pipeline is to perform a precise inter-order background subtraction for science frames and master flat-fields, and an optimal extraction of the object signal rejecting a number of cosmic ray impacts and subtracting the sky spectrum simultaneously. The pipeline products are checked step by step. The wavelength scale of the spectra reduced by the pipeline was then converted to vacuum-heliocentric values and individual 1-D exposures scaled, weighted and combined altogether using the NOAO [onedspec]{} package of the IRAF[^3] software. During this process, the spectra were rebinned to 0.08 Å pix$^{-1}$. In order to increase the signal-to-noise ratio in this near-IR part of the optical range (and as the lines of interest are broad), we applied a Gaussian filter smoothing with two pixel FWHM and get an effective spectral resolution of $\sim 12$ km s$^{-1}$. The resulting signal-to-noise ratio per pixel is of the order of 20 or more over most of the wavelength range considered here. The final spectrum is shown in Fig. \[fig1\]. The continuum fitting in the red side of the emission has been performed using a smooth low order polynomial considering only the absorption free regions. In the Lyman-$\alpha$ forest, we approximated the QSO spectrum by a powerlaw, $f_\nu\propto\nu^{-0.6}$, that we fitted on the low-dispersion SDSS spectrum made available to us by Dr. Fan. The normalization of the power-law was done using the flux in the region between the and emission lines. In our normalised spectrum the mean transmission flux in the redshift range $z$=4.39 to $z$=4.60 (that is not contaminated by any BAL absorption lines) is 0.15. This is consistent with the values derived in the same redshift range using non-BAL QSOs (Becker et al. 2001). Broad absorption line systems ============================= Based on the unabsorbed emission line we estimate the emission redshift of the QSO to be =4.92. Two well detached absorption line systems are identified based on the and absorptions over the redshift ranges, $z_{\rm abs}$=4.83$-$4.88 and $z_{\rm abs}$ = 4.62$-$4.75, respectively. For mean redshifts = 4.855 and 4.685, this corresponds to ejection velocities of $\approx$3,330 and $\approx$ 12,160 . The spectrum in the Lyman-$\alpha$ forest is heavily affected by the high opacity of the intergalactic medium. Even though excess absorption is seen at the expected positions of BAL absorption lines (see Fig. \[fig1\]) it is difficult to estimate column densities in this region. Therefore, in the following sections we use only lines observed on the red side of the emission line to investigate the physical conditions in the outflowing gas. The red component at $z_{\rm abs}$ = 4.855 ------------------------------------------ The absorption trough in this component is $\sim2000$ wide. The absorption profiles of different transitions centered at $z_{abs}=4.855$ are shown in Fig. \[fig2\]. The absorption profile of the doublet that is comparatively less saturated than that of the doublet suggests the presence of 2 relatively narrow components denoted by l1 and l2 and 3 broad subcomponents denoted by a, b and c. While and are detected in all these components, absorption is [detected]{} only in the broad subcomponents (see Fig. \[fig2\]). All the obviously saturated absorption lines have zero residual (C [iv]{} and Lyman-$\alpha$, see Fig. \[fig2\]). The covering factor of the gas is therefore unity. In case of complete coverage the observed residual intensities R$_1$ and R$_2$ in the first and second member of the doublet are related as R$_1$=R$_2^2$ (see Srianand and Shankarnaraynan 1999). Indeed, even though part of the absorption line is contaminated by atmospheric absorption (see Fig. \[fig3\]), the velocity ranges in the profiles that are free of blending are consistent with complete coverage (see top panel in Fig. \[fig3\]). As emission in this wavelength range is contributed by both continuum and broad emission lines the corresponding absorbing gas has to be located outside the broad line region (BLR). As the lines are weak we can derive the N [v]{} column densities. However, due to saturation effects, we can determine only lower limits for and column densities. It can be seen on Fig. \[fig2\] that Si [iv]{}$\lambda$1402 is only partly saturated which means that, although very uncertain, we can estimate the corresponding column density. We use also the Si [iv]{}$\lambda$1402 profile as a template to determine upper limits on the column densities of non detected species (see below). The column density in each velocity pixel was obtained using the relation $$N(v)=3.768\times 10^{+14}~\tau/f\lambda~{\rm cm^{-2}~km^{-1}s}$$ where, $\tau$, $f$ and $\lambda$ are the optical depth, oscillator strength and rest wavelength respectively. The total column density is obtained by integrating $N(v)$ over the velocity range $-$1120 to 1160 . The results are presented in Table \[tab1\]. Rest wavelengths and oscillator strengths used here are taken from Verner et al. (1994). The two lines of the C [iv]{} doublet are blended and we use an effective oscillator strength, $f=f_1+f_2$, and mean wavelength, ($\lambda_1+\lambda_2$)/2, to compute the lower limit on the column density. In the case of N [v]{}, we use the line for component a and line for components b and c that are free of blending. Contamination by atmospheric features (see Fig. \[fig3\]) is corrected by carefully masking the atmospheric contamination using the normalised spectrum of Q 1122$-$1628. Singly ionized species such as Si [ii]{}, C [ii]{} and Al [ii]{} are absent and the spectrum at the expected position of Al [iii]{} lines is unfortunately very noisy. As said above, we evaluate upper limits for column densities of the latter species using the profile (which is not completely saturated and least affected by atmospheric contamination) extending from $-$1120 to 1160 as a template. The scaling factor $k$ = \[$Nf\lambda$\]$_{X^+}$/\[$Nf\lambda$\]$_{template}$ between the two optical depths $\tau$$_{template}$ and $\tau$$_{X^+}$ for species $X^+$ is then obtained by minimizing $$\alpha = \Sigma(\tau_{\rm X^+} - k\tau_{template})^{2}$$ For Si [ii]{} and C [ii]{}, $k$ comes out to be 0.031 and 0.035 respectively. The corresponding upper limits on the column densities are given in Table \[tab1\]. Note that the error in the column density of is mainly due to continuum placement uncertainties. We note that most of the atmospheric absorption seen in the expected wavelength range have consistent equivalent width. This suggests that our continuum fitting does not underpredict the absorption in the region. No flux is detected at the expected position of Si [iii]{}$\lambda$1206 (see Fig. \[fig1\]). However, we could not use this to derive a lower limit on Si [iii]{} as the same region is contaminated by a probable absorption line from the blue component (see Fig. \[fig4\]). Finally we notice that the velocity separation between the two narrow components l1 and l2 of the red component is very close to the Si [iv]{} doublet splitting. This is shown in Fig. \[fig5\]. This is consistent with growing evidence for line-locked flows and radiative acceleration in BAL systems (see Srianand et al. 2002 and references therein). species ------- ----------- --------------- 4.855 $\ge$15.67 C [ii]{} $\le$14.21 C [iv]{} $\ge$15.64 Si [ii]{} $\le$14.13 Si [iv]{} 15.33: N [v]{} 14.63$-$14.71 4.685 $\le 16.00$ $\ge$15.94 Si [ii]{} $\le$14.73 Si [iv]{} $\ge$15.40 : Column densities estimated using apparent optical depth method with complete coverage. \[tab1\] ------- --------------- -------------- -------------- --------------- Solar Metal 1.0 Z$_\odot$ 1.0 Z$_{SB}$ 0.5 Z$_{SB}$ 0.20 Z$_{SB}$ C -3.44 -3.90 -4.22 -4.63 N -4.03 -4.60 -5.34 -6.51 O -3.13 -3.00 -3.30 -3.70 Si -4.44 -4.43 -4.74 -5.14 ------- --------------- -------------- -------------- --------------- : Chemical compositions considered in the models: \[abu\] \ This form of Starburst abundances correspond to chemical evolution model M5a of Hamann & Ferland [(1993)]{}. The blue component at $z_{\rm abs}$ = 4.685 ------------------------------------------- The absorption profiles of different transitions centered at $z_{abs}=4.685$ are shown in Fig. \[fig4\]. Doublet partners for and are partially blended together. The non-zero residual flux for the blended corresponds to a covering factor, $f_{\rm c}$ $\ge$0.90. It is interesting to note that the absorption is well detached from the emission line profile. Thus the partial coverage reflects either $\simeq 10\%$ contribution from the scattered light or partial coverage of the continuum source. Lower limits to column densities for the and were obtained assuming full coverage (see Table \[tab1\]). Doing otherwise would only increase these limits. Note that absorption line extends over a larger velocity range as compared to absorption profile (see Fig. \[fig4\]). Like in the red component, singly ionized species and are not detected which indicates that the neutral hydrogen column density cannot be large (remember that metallicity in this kind of gas is usually large). N [v]{} may be present but its redshifted position in the forest overlaps with that of H [i]{} and Si [iii]{} from the red component (see Fig. \[fig1\]). The mean transmitted flux at the position is consistent with saturated absorption line. In the Fig. \[fig4\] the intervening Lyman-$\alpha$ forest is clearly observed over the whole range corresponding to the BAL H [i]{} Lyman-$\alpha$ trough. This means that the H [i]{} optical depth cannot be large except if the covering factor of H [i]{} is much smaller than that of C [iv]{}. This is not impossible (see e.g. Srianand et al. 2002) but would be surprising given the non detection of both C [ii]{} and Si [ii]{}. By comparing the mean transmission in this region with that devoid of BAL absorption lines we estimate that $\tau(v)$ for is less than 2. If we use the velocity range covered by the profile ($\simeq 4000$ ), we see that the column density can not much larger than 10$^{16}$ cm$^{-2}$. [llclccc]{} & & [log(N(H(total)))\]{}\ Spectrum & enrichment& Metallicity(Z) & log N(H [i]{})\& ratios & observed N&\ \ MF & Solar &0.5Z$_{\odot}$,\[Si/N\]=3.0\[Si/N\]$_{\odot}$ & 17.00 & (-2.25,-2.15) & $\approx$-2.2,-2.1& (19.99,20.11)\ &&&&\ BAL & “ & 1.0Z$_{\odot}$ & 16.00 & (-1.5,-1.1) & $\approx$-1.1 & 20.00\ ” & “ & 1.5Z$_{\odot}$ & ” & (-1.4,-0.95) & $\approx$-1.0 & 19.90\ &&&&&&\ “ & ” & 0.05$_{\odot}$ & 17.00 & (-1.7,-0.75) & $\approx$-0.7 & 21.42\ “ & ” & 0.1Z$_{\odot}$ & “ & (-1.7,-0.7) & (-0.9,-0.8) & (21.53,21.62)\ ” & “ & 0.2Z$_{\odot}$ & ” & (-1.65,-0.65) & $\approx$-1.0 & 21.36\ “ & ” & 1.0Z$_{\odot}$ & “ & (-1.4,-0.4) & – & –\ &&&&&&\ ” & Starburst & 1.0Z$_{SB}$ & 16.00 & (-1.4,-0.65) & $\approx$-0.7 & 20.30\ “ & ” & 5.0Z$_{SB}$ & “ & (-1.4,-0.7) & $\approx$(-1.0,-0.9) & (19.32,19.41)\ ” & “ & 9.0Z$_{SB}$ & ” & (-1.15,-0.8) & $\approx$-1.05 & 19.11\ &&&&&&\ “ & ” & 0.2Z$_{SB}$ & 17.00 & (-1.7,FP) & $\approx$0.1 & 22.42\ “ & ” & 0.5Z$_{SB}$ & “ & (-1.5,0.1) & $\approx$(-0.6,-0.5)& (21.60,21.70)\ ” & “ & 1.0Z$_{SB}$ & ” & (-1.3,0.1) & $\approx$-0.8 & 21.12\ &&&&&&\ \[tab3\] \* in cm$^{-2}$ Discussion ========== Physical conditions in the outflowing gas ----------------------------------------- From the discussion in the previous Section, it is clear that constraints are not strong enough to probe the physical conditions in the blue component ( = 4.685). We therefore mainly concentrate on the red component ( = 4.855). Recent Chandra observations (Vignali et al. 2001) fail to detect X-rays from the QSO SDSS J160501.21-011220.0. As discussed before, the X-ray non-detection could be either due to large X-ray absorbing column density or a consequence of the QSO being intrinsically X-ray quiet. Indeed, there are already some reports that the X-ray continuum shapes of QSOs may evolve at $z\ge$2.5 (e.g. Vignali et al. 1999; Blair et al. 2000; Vignali et al. 2003). Assuming that the lack of X-rays in this source is due to absorption, Vignali et al. (2001) infer that the X-ray non-detection for this object implies a total hydrogen column density of N(H(total)) $\ge 5.0 \times 10^{23}$ cm$^{-2}$, characteristic of low-ionization BAL QSOs at low redshift. Note that the limit on the X-ray flux further implies a small optical to X-ray spectral index, $\alpha$$_{\rm ox}$ $<$ $-$1.82, for this object. The absence of singly ionized species and suggests that H [i]{} is optically thin at the Lyman limit. Thus, to probe the physical conditions in the red component, we run grids of photoionization models using Cloudy (Ferland 1996) in the range log $N$ (H [i]{}) (cm$^{-2}$)$\sim$ 10$^{16}$ to 10$^{17}$ cm$^{-2}$, considering the gas being ionized by either an unattenuated QSO spectrum given by Mathews & Ferland (1987) (hereafter MF spectrum; see Fig. \[fig6\]) or a modified spectrum with little X-rays (hereafter BAL spectrum; see Fig. \[fig6\]) mimicking attenuation of a typical QSO spectrum by a large column density of ionized gas (see below). The models are run for different chemical composition, either the Solar one or the so-called Starburst one (see Table \[abu\]). In all the models, the calculations are stopped when the [total]{} neutral hydrogen column density reaches the limit we set (log $N$(H [i]{}) (cm$^{-2}$) = 16 or 17). The input parameters of the models, that reproduce the observed ratios as well as the column densities of individual species, are given in Table \[tab3\]. The corresponding total hydrogen column density in the allowed range of ionization parameters is given in the last Column of the Table \[tab3\]. Results for some selected models with different ionizing spectra and chemical composition for log $N$(H [i]{}) (cm$^{-2})$) = 17 are given in Fig. \[fig7\]. In this figure we have plotted the column density of different species as a function of ionization parameter. The horizontal dotted lines mark limits on the observed column densities of , , and . The minimum and maximum of the allowed range for log $U$ are marked with verticle dashed lines. The former is obtained using either of the ratios $N$(Si [ii]{})/$N$(Si [iv]{}) or $N$(C [ii]{})/N(C [iv]{}) and the latter is obtained from the observed $N$(N [v]{})/N(Si [iv]{}) ratio for metallicities assumed in the model. The range of ionization parameters for which the observed N [v]{} column density is reproduced is indicated as a verticle shaded region. We note that, if the column density is close to the observed value (i.e. few 10$^{16}$ cm$^{-2}$) one needs overall metallicities larger than Solar to explain the observed column densities. We therefore mainly concentrate on models with $N$(H [i]{})$ = 10^{17}~{\rm cm}^{-2}$ that will give a conservative lower limit on the metal enrichment in the system. If the gas is ionized by a MF type spectrum and has a Solar composition, it can be seen from panel (a) in Fig. \[fig7\] that the observed constraints on the ion ratios and the and column densities can be consistently reproduced for log U $\simeq-1.9$ (see the shaded region). However the predicted column density of is lower than the observed value by more than an order of magnitude. It is clear that the observed value can be reproduced only if \[Si/H\]$\ge$\[Si/H\]$_\odot$. Increasing the overall abundance actually reduces the required overabundance of Si with respect to N. For example if one assumes Z = 0.5 Z$_\odot$ the observed column densities can be reproduced consistently with lower ionization parameter and a \[Si/N\] ratio higher than Solar by a factor of 3 only (see Table. \[tab3\]). Larger metallicities are not possible however as C [ii]{} would then be detectable. The enhancement of $N$(Si [iv]{}) with respect to $N$(N [v]{}) can be achieved by either enhancing the silicon metallicity with respect to the nitrogen one as discussed above or by suppressing the N [iv]{} ionizing photons. The first possibility can be achieved in a chemical enrichment model where the secondary production of N has not yet begun to dominate. The latter possibility naturally arise if the cloud sees an attenuated X-ray spectrum, similar to what we observe directly for SDSS J160501.21$-$011220.0. We explore this possibility in the following. We model the absorbed spectrum (called BAL spectrum in the following) as the composite of a black body spectrum with temperature $T$ = 150,000 K, plus a power-law spectrum with $\alpha_{uv}$ = $-$0.5 and $\alpha_{x}$ = $-$1.0 and a relative scaling $\alpha$$_{ox}$ = $-$2.0. From Fig. \[fig6\], it is clear that for a given number of hydrogen ionizing photons this spectrum has less photons to ionize Si [iii]{} and N [iv]{} compared to the MF spectrum. Thus a given ionization state will be produced by higher values of log U and hence higher total hydrogen column densities. However the major difference is the presence of a soft X-ray excess in the MF spectrum that implies more N [iv]{} ionizing photons for a given number of Si [iii]{} ionizing photons compared to the BAL spectrum. For a column density of log $N$ (H [i]{}) (cm$^{-2}$)$\approx$17.00, observations can be reproduced by models with metallicities 0.05Z$_{\odot}$$\le$Z$\le$0.2Z$_{\odot}$ (Table \[tab3\]). This is also demonstrated in panel (b) of Fig. \[fig7\]. As expected, in these models the results are consistent with observations at an ionization parameter an order of magnitude higher than that required for the MF ionizing spectrum. Thus, we can reproduce the observed column densities with Sub-Solar metallicity and Solar abundance ratios. It is usual in the case of the BLR clouds to explain an enhanced nitrogen abundance by rapid star formation nucleosynthesis. We consider such a chemical enrichment in our models using option “starburst abundances” in Cloudy (see Table \[abu\]). The details of the models are given in Hamann & Ferland (1993). These models have problems in reproducing the observations when a MF ionizing spectrum is considered (see panel (c) in Fig. \[fig7\]). However, the observed ratios are reproduced if the BAL spectrum is used instead (see Table \[tab3\] and panel (d) in Fig. \[fig7\]) for overall metallicity $\ge 0.2$ and log [N(H [i]{})]{} (cm$^{-2}$) in the range 16 to 17. It is interesting to note that for all the models considered here the total hydrogen column density is at least an order of magnitude less than the amount required (i.e. $N$(H(total)) $\ge5\times 10^{23}$ cm$^{-2}$) to explain the lack of X-rays in this QSO by absorption of a standard QSO spectrum (Table \[tab3\]). We also run models by fixing the total hydrogen column density to be $N$(H(total)) $\ge5\times 10^{23}$ cm$^{-2}$. For MF ionizing spectrum we do not find any model that can reproduce the constraints noted in Table [\[tab1\]]{}. Thus, we can conclude that either SDSS J160501.21$-$011220.0 is intrinsically X-ray weak or, if there is any Compton thick screen between the quasar and us, it has nothing to do with the gas that is associated with the red component. The similar characteristics of the blue component together with the low $N$(H [i]{}) column density of the system strongly suggests that the blue component as well is not Compton thick. Formation redshift of the blackhole ----------------------------------- We estimate the mass of the black hole from the Eddington accretion. For this we estimate the rest frame luminosity at 2500 Å using the low dispersion spectra obtained by the SDSS group (Fan et al. 2000) to be L$_\nu$$=$1.37$\times$10$^{31}$ erg s$^{-1}$ Hz$^{-1}$. Applying bolometric correction as suggested by Elvis et al. (1994), we obtain a bolometric luminosity of L$_{bol}$=9.21$\times$10$^{46}$ erg s$^{-1}$. Assuming Eddington accretion we derive the black hole mass, $M_{\rm BH}$ = 7.1$\times$10$^{8}$$\eta^{-1}$ M$_{\odot}$ (see Haiman & Loeb 2001). Here, $\eta$ is the efficiency parameter (i.e. accretion luminosity given in the units of Eddington luminosity). If $\epsilon$=L$_{bol}$/M$_{BH}$c$^2$ is the radiative efficiency for a mass accretion rate M$_{BH}$ then the natural e-folding time scale for the growth of a single seed can be written as $$t=M_{BH}/$\.M$_{BH}=4.0\times10^7\epsilon_{0.1}\eta^{-1} yr$$ It will therefore take about ln(7.1$\times$10$^{8}$$\eta^{-1}$M$_{\odot}$/10 M$_{\odot}$) = 18.1 e-folding times (or t $\ge 0.7\epsilon_{0.1}\eta^{-1}$ Gyr) for the black hole to grow to the above estimated mass from a stellar-mass seed of 10 M$_\odot$. For the assumed value of $\epsilon$ $\approx$ 0.1 and $\eta$ $\approx$ 1.0 this corresponds to a formation epoch of the black hole close to $z \approx11$ for the cosmological model considered here. Summary ======= We have presented the analysis of broad absorption lines (BALs) seen in the spectrum of the $\simeq$4.92 QSO SDSS J160501.21-011220.0. Our high spectral resolution UVES spectrum shows two well detached absorption line systems at = 4.685 and 4.855. The system at = 4.855 covers the background source completely suggesting that the gas is located outside the broad emission line region. On the contrary the system at = 4.685, which is redshifted on top of the quasar continuum, has a covering factor of the order of 0.9. Physical conditions are investigated in the BAL system at = 4.855 using detailed photoionization models. The observed column density and the limits on and absorptions suggest that log $N$(H [i]{}) (cm$^{-2}$) is in the range 16–17. The observed column densities of , and in the = 4.855 component require, unlike what is derived when analysing broad emission lines, that nitrogen is underabundant by more than a factor of 3 compared to silicon if the gas is photoionized by a typical QSO spectrum. Thus, if the gas is ionized by a standard MF spectrum, the chemical enrichment of the cloud is different from that required by emission line clouds. We show however that the relative supression in the column densities can be reproduced for Solar abundance ratios or abundance ratios typical of rapid Starburst nucleosynthesis if we use an ionizing spectrum that is devoid of X-rays. Thus if the composition of BAL is like that of the emission line regions it is most likely that the cloud sees an ionizing spectrum similar to what we observe from this QSO that is strongly attenuated in the X-rays. This is consistent with the fact that none of our models have high enough Compton optical depth to be able to remove X-rays from the QSO. Similar arguments lead to the conclusion that the system at = 4.685 as well is not Compton thick. Using simple Eddington arguments we show that the mass of the central black hole is $\sim 8\times 10^8$ M$_\odot$. This suggests that the accretion onto a seed black hole must have started as early as $z$ $\sim$ 11. This gives a typical formation epoch for the host galaxy of the QSO. acknowledgments {#acknowledgments .unnumbered} =============== This work was supported in part by the European Communities RTN network “The Physics of the Intergalactic Medium”. We wish to thank Dr. Fan for making the low dispersion data on SDSS J160501.21$-$011220.0 available to us. Ballester P., Modigliani A., Boitquin O. et al., 2000, The Messenger, 101, 31 Becker R. et al., 2001, AJ, 122, 2850 Blair A. J., Stewart G. C., Georgantopoulous I., Boyle B. J., Griffiths, R. E., Shanks, t & Almaini, O., 2000, MNRAS, 314,138 Brandt W. N., Laor A., Wills B. J., 2000, ApJ, 528, 637 Bregman J. N., 1984, ApJ, 276, 423 Dekker H. et al., 2000, SPIE, 4008, 534 Dietrich M. et al., 2003, A&A, 398, 891 Elvis M. et al., 1994, ApJS, 95, 1 Fan X. et al., 2000, AJ, 119, 1 Ferland G. J., 1996, HAZY- a brief introduction to Cloudy. Univ.Kentucky, Dept. Physics & Astron. internal report Gallagher S., Brandt W. N., Sambruna R., Mathur S., Yamamski N., 1999, ApJ, 519, 549. Green P.J., Aldcroft, T. L., Mathur, S., Wilkes B. J., Elvis M., 2001, ApJ, 558, 109 Green P. J., Mathur S., 1996, ApJ, 462, 637 Haiman Z., Loeb A., 2001, ApJ, 552, 459 Hamann F., 1997, ApJS, 109, 279 Hamann F., Ferland G. J., 1992, ApJ, 391L, 53 Hamann F., Ferland G. J., 1993, ApJ, 418, 11 Hamann F., Ferland G. J., 1999, ARA&A, 37, 487 Korista K. et al., 1993, ApJ, 413, 445 Korista K. et al., 1996, ApJ, 461, 641 Kraemer S. B., Crenshaw D. M., Yaqoob T. et al., 2002, astro-ph/0208478 Mathews W. G., Ferland G. J., 1987, ApJ, 323, 456 Mathur S., Elvis M., Singh K. P., 1995, ApJ, 455, 9 Mathur S. et al., 2000, ApJ, 533, 79 Matteucci F., Padovani P., 1993, ApJ, 419, 485 Petitjean P., Rauch M., Carswell R. F., 1994, A&A, 291, 29 Petitjean P., Srianand R., 1999, A&A, 345, 73 Singh K. P., Westergaard N. J., Schnopper H. W., 1987, A&A, 172, 11 Srianand R., Petitjean P., 2000, A&A, 357, 414 Srianand R., Petitjean P., Ledoux C., Hazard C., 2002, MNRAS, 336, 753 Srianand R., Shankaranarayanan S., 1999, ApJ, 518, 672 Turnshek D. A., 1988, in STScI Symposium Series 2, QSO Absorption lines: Probing the Universe, ed. J. C. Blades, D. Turnshek, & C. Norman (Cambridge:Cambridge Univ. Press) Verner D. A., Barthel P. D., Tytler D., 1994, A&AS, 108, 287 Vignali C. et al., 1999, ApJ, 516, 582 Vignali C. et al., 2001, AJ, 122, 2143 Vignali C. et al., 2003, AJ, 125, 418 Weymann R. J., Morris S. L., Foltz C. B., Hewett P. C., 1991, ApJ, 373, 23 [^1]: Based on observations carried out at the European Southern Observatory (ESO) under programmes 67.A-0078 and 69.A-0457 with the UVES spectrograph installed at the Nasmyth focus B of the VLT 8.2m telescope,unit Kueyen,on Cerro Paranal in Chile [^2]: MIDAS: Munich Image Data Analysis System, trademark of the European Southern Observatory (ESO) [^3]: IRAF: The Image Reduction and Analysis Facility is distributed by the National Optical Astronomy Observatories, which is operated by the Association of Universities for Research in astronomy, Inc. (AURA), under cooperative agreement with the National Science Foundation.	{ "pile_set_name": "ArXiv" }
--- address: \| Research Center for Nuclear Physics, Osaka University,\ Mihogaoka 10-1, Ibaraki 567-0047, Japan author: - 'H. Suganuma, K. Amemiya, H. Ichie, H. Matsufuru, Y. Nemoto and T.T. Takahashi' title: Quark Confinement Physics from Lattice QCD --- Nonabelian Feature in QCD ========================= Nowadays, quantum chromodynamics (QCD) is established as the fundamental theory of the strong interaction. In spite of the simple form of QCD lagrangian \_[QCD]{}=-[1 2]{} [tr]{} G\_G\^ +\|q (i -m\_q) q, it is still hard to understand nonperturbative QCD (NP-QCD) phenomena, such as color confinement and dynamical chiral-symmetry breaking, in the infrared strong-coupling region. In fact, QCD tells us the “rule” of the elementary interaction between quarks and gluons, but to solve QCD is another difficult problem. In this point, QCD may resemble the Rubik Cube, where the rule is quite simple, but to solve this puzzle is rather difficult. Like QCD, the difficulty of the Rubik cube comes from the non-commutable procedures of the “nonabelian” rotational process as shown in Fig.1. Here, a kind of “local procedure” on 3 $\times$ 3 $\times$ 3 subcubes leads to extremely large number of configurations, which makes this puzzle difficult and interesting. Then, the Rubik cube may be regarded as a miniature of QCD.[@NEWS99] ![ The difficulty and the interest of the Rubik cube originate from non-commutable operations based on the nonabelian feature of the rotational group. The configuration after Step 1 and Step 2 depends on the order of these operations.](Fig1.eps){height="7cm"} The main difficulties of QCD originate from the nonabelian and the strong-coupling features in the infrared region below 1 GeV. In the ultraviolet region, the QCD coupling is weak and then we can use the perturbative QCD, where the three and the four gluon vertices stemming from the nonabelian nature of QCD are treated as perturbative interactions. In the infrared region, however, the nonabelian and the strong-coupling features are significant. In the electro-magnetism, the superposition of solutions is possible, because of the linearity of the field equation $ \partial_\mu F^{\mu\nu}=j^\nu, $ so that we can consider the partial electro-magnetic field formed by each charge and the total electro-magnetic field can be obtained by summing up the individual field solution. On the other hand, the QCD field equation becomes nonlinear as \_G\^ + ig \[A\_, G\^\]=j\^, due to the nonabelian feature, and then it is difficult to solve it even at the classical level, because the superposition of solutions is impossible. So, we cannot divide the color-electromagnetic field into each part formed by each quark. Instead, the QCD system is to be analyzed as a whole system. In this way, for the analysis of QCD, the nonabelian feature provides one of serious difficulties. Abelianization of QCD in MA Gauge ================================= The nonabelian nature of QCD in the infrared region is, however, removed in the MA gauge. In fact, QCD is reduced into an abelian gauge theory with color-magnetic monopoles, keeping essence of infrared nonperturbative features. Dual Superconductor Theory and QCD ---------------------------------- In QCD, to understand the confinement mechanism is one of the most difficult problems remaining in the particle physics. As is indicated by hadron Regge trajectories and lattice QCD calculations, the confinement phenomenon is characterized by [one-dimensional squeezing]{} of the color-electric flux and the [string tension]{} $\sigma \simeq 1{\rm GeV/fm}$, which is the key quantity of confinement. On the confinement mechanism, Nambu first proposed the [dual superconductor theory]{} for quark confinement,[@N74] based on the electro-magnetic duality in 1974. In this theory, there occurs the one-dimensional squeezing of the color-electric flux between quark and anti-quark by the [dual Meissner effect]{} due to condensation of bosonic color-magnetic monopoles. However, there are [two large gaps]{} between QCD and the dual superconductor theory.$^{3-5}$ 1. The dual superconductor theory is based on the [abelian gauge theory]{} subject to the Maxwell-type equations, where electro-magnetic duality is manifest, while QCD is a nonabelian gauge theory. 2. The dual superconductor theory requires condensation of color-magnetic monopoles as the key concept, while QCD does not have color-magnetic monopoles as the elementary degrees of freedom. These gaps may be simultaneously fulfilled by taking [MA gauge fixing,]{} which reduces QCD to an abelian gauge theory including color-magnetic monopoles. MA Gauge Fixing and Relevant Mode for Confinement ------------------------------------------------- In Euclidean QCD, the maximally abelian (MA) gauge is defined so as to minimize the total amount of the off-diagonal gluons,$^{3-5}$ $$\begin{aligned} R_{\rm off} [A_\mu ( \cdot )] \equiv \int d^4x \ {\rm tr} \left\{ [\hat D_\mu ,\vec H][\hat D_\mu ,\vec H]^\dagger \right\} ={e^2 \over 2} \int d^4x \sum_\alpha \|A_\mu ^\alpha (x)\|^2, \end{aligned}$$ by the SU($N_c$) gauge transformation. Here, we have used the Cartan decomposition, $A_\mu (x)=\vec A_\mu (x) \cdot \vec H +\sum_\alpha A_\mu^\alpha (x)E^\alpha $. Since the ${\rm SU}(N_c)$ covariant derivative operator $\hat D_\mu \equiv \hat \partial_\mu+ieA_\mu $ obeys the adjoint gauge transformation, the local form of the MA gauge condition is easily derived as $[\vec H, [\hat D_\mu , [\hat D_\mu , \vec H]]]=0.$ $^{3-5}$ In the MA gauge, the gauge symmetry $G \equiv {\rm SU}(N_c)_{\rm local}$ is reduced into $H \equiv {\rm U(1)}_{\rm local}^{N_c-1} \times {\rm Weyl}^{\rm global}_{N_c}$, where the global Weyl symmetry is a subgroup of ${\rm SU}(N_c)$ relating the permutation of $N_c$ bases in the fundamental representation. In the MA gauge, off-diagonal gluons behave as charged matter fields like $W_\mu^{\pm}$ in the Standard Model, and provide the color-electric current in terms of the residual abelian gauge symmetry. In addition, according to the reduction of the gauge symmetry, color-magnetic monopoles appear as topological objects reflecting the nontrivial homotopy group$^{3-8}$ \_2([SU]{}(N\_c)/[U(1)]{}\^[N\_c-1]{})=\_1([U(1)]{}\^[N\_c-1]{}) =[Z]{}\^[N\_c-1]{}\_, in a similar manner to similarly in the GUT monopole. Here, the global Weyl symmetry and color-magnetic monopoles are relics of nonabelian nature of QCD. Thus, in the MA gauge, QCD is reduced into an abelian gauge theory including color-magnetic monopoles, which is expected to provide a theoretical basis of the dual superconductor theory for quark confinement. Furthermore, recent lattice QCD studies show remarkable features of [abelian dominance]{} and [monopole dominance]{} for NP-QCD in the MA gauge. 1. Without gauge fixing, all the gluon components equally contribute to NP-QCD, and it is difficult to extract relevant degrees of freedom for NP-QCD. 2. In the MA gauge, QCD is reduced into an abelian gauge theory including the electric current $j_\mu $ and the magnetic current $k_\mu $, which forms the [global network of the monopole world-line covering the whole system.]{} In the MA gauge, lattice QCD shows [abelian dominance]{} for NP-QCD (confinement, chiral symmetry breaking [@M95], gluon propagators [@AS99]): only the diagonal gluon, which remains an abelian gauge field, is relevant for NP-QCD, while off-diagonal gluons do not contribute to NP-QCD. 3. By the Hodge decomposition, the diagonal gluon is decomposed into the “photon part” and the “monopole part”, corresponding to the separation of $j_\mu$ and $k_\mu$. In the MA gauge, lattice QCD shows [monopole dominance]{}$^{9,11-12}$ for NP-QCD: the monopole part ($k_\mu \ne 0$, $j_\mu=0$) leads to NP-QCD, while the photon part ($j_\mu \ne 0, k_\mu=0$) seems trivial like QED and does not contribute to NP-QCD. For example, on the $Q$-$\bar Q$ potential, the purely linear confinement potential appears in the monopole part, while the Coulomb potential appears in the photon part like QED.[@P97] In fact, by taking the MA gauge, the [relevant collective mode for NP-QCD]{} can be extracted as the color-magnetic monopole.[@SAIT00; @SITA98] Strong Random Phase of Off-diagonal Gluon in MAQCD ================================================== To find out essence of the MA gauge, we study the feature of the off-diagonal gluon field $A_\mu^\pm \equiv \frac1{\sqrt{2}}(A_\mu^1 \pm i A_\mu^2)$ in the MA gauge in SU(2) lattice QCD. In SU(2) lattice QCD, the SU(2) link variable is factorized as $U_\mu(s)=M_\mu(s)u_\mu(s)$, according to the Cartan decomposition ${\rm SU(2)/U(1)}_3 \times {\rm U(1)}_3$. Here, $u_\mu(s)\equiv \exp\{i\tau^3\theta^3_\mu(s)\}\in {\rm U(1)}_3$ denotes the abelian link variable, and the abelian projection is defined by the replacement as $U_\mu(s) \rightarrow u_\mu(s)$. The off-diagonal matrix $M_\mu(s)\in {\rm SU(2)/U(1)}_3$ is parameterized as $$\begin{aligned} M_\mu(s)\equiv e^{i\{\tau^1\theta^1_\mu(s)+\tau^2\theta^2_\mu(s)\}} =\left( \begin{array}{cc} {\rm cos}{\theta_\mu}(s) & i e^{-i\chi_\mu(s)}{\rm sin}{\theta_\mu}(s) \\ i e^{i\chi_\mu(s)}{\rm sin}{\theta_\mu}(s) & {\rm cos}{\theta_\mu}(s) \end{array} \right).\end{aligned}$$ In the continuum limit, $\chi_\mu(s)$ coincides with the off-diagonal gluon phase as $A_\mu^\pm(x)=e^{\pm i \chi_\mu(x)}\|A_\mu^\pm(x)\|$, and $\theta_\mu(s)$ is proportional to the off-diagonal gluon amplitude as $\theta_\mu(s)=\frac1{\sqrt{2}}a\|eA_\mu^{\pm}(s)\|$ with the lattice spacing $a$. In the MA gauge, the diagonal element $\cos \theta_\mu(s)$ in $M_\mu(s)$ is maximized by the SU(2) gauge transformation, [e.g.]{} $\langle\cos \theta_\mu(s)\rangle_{\rm MA}\simeq 0.93$ at $\beta=2.4$. Accordingly, the off-diagonal gluon $A_\mu^\pm(x)=e^{\pm i \chi_\mu(x)}\|A_\mu^\pm(x)\|$ has two relevant features in the MA gauge.$^{3-5}$ 1. The off-diagonal gluon amplitude $\|A_\mu^{\pm}(x)\|$ (or $\|\sin \theta_\mu(s)\|$ on lattices) is strongly suppressed by SU($N_c$) gauge transformation in the MA gauge. 2. The off-diagonal gluon phase $\chi_\mu(x)$ tends to be random, because $\chi_\mu(x)$ is not constrained by MA gauge fixing at all, and only the constraint from the QCD action is weak due to a small accompanying factor $\|A_\mu^\pm\|$ or $\|\sin \theta_\mu\|$. ![ (a) The distribution $P(\Delta \chi)$ of the difference $\Delta \chi \equiv \|\chi_\mu(s)-\chi_\mu(s+\hat \nu)\| ({\rm mod} \pi)$ in the MA gauge plus U(1)$_3$ Landau gauge at $\beta$=0 ($a=\infty$, thin line), $\beta$=1.0 ($a \simeq$ 0.57fm, dotted curve), $\beta$=2.4 ($a \simeq$ 0.127fm, solid curve), $\beta$=3.0 ($a \simeq$ 0.04fm, dashed curve). (b) The off-diagonal gluon contribution to the Wilson loop $W_C^{\rm off}$ v.s. the perimeter length $L \equiv 2(R+T)$ in the MA gauge on $16^4$ lattice with $\beta$=2.4. The thick line denotes the theoretical estimation in Eq.(10) with the microscopic input $\langle \ln \{ \cos \theta_\mu(s) \} \rangle_{\rm MA} \simeq -0.082$ at $\beta=2.4$.](Fig2a.eps "fig:"){height="4cm"} ![ (a) The distribution $P(\Delta \chi)$ of the difference $\Delta \chi \equiv \|\chi_\mu(s)-\chi_\mu(s+\hat \nu)\| ({\rm mod} \pi)$ in the MA gauge plus U(1)$_3$ Landau gauge at $\beta$=0 ($a=\infty$, thin line), $\beta$=1.0 ($a \simeq$ 0.57fm, dotted curve), $\beta$=2.4 ($a \simeq$ 0.127fm, solid curve), $\beta$=3.0 ($a \simeq$ 0.04fm, dashed curve). (b) The off-diagonal gluon contribution to the Wilson loop $W_C^{\rm off}$ v.s. the perimeter length $L \equiv 2(R+T)$ in the MA gauge on $16^4$ lattice with $\beta$=2.4. The thick line denotes the theoretical estimation in Eq.(10) with the microscopic input $\langle \ln \{ \cos \theta_\mu(s) \} \rangle_{\rm MA} \simeq -0.082$ at $\beta=2.4$.](Fig2b.eps "fig:"){height="3.9cm"} Now, we consider the behavior of $\Delta \chi \equiv \|\chi_\mu(s)-\chi_\mu(s+\hat \nu)\| ({\rm mod} \pi)$ in the MA gauge. If the off-diagonal gluon phase $\chi_\mu(x)$ is a continuum variable, as lattice spacing $a$ goes to 0, $\Delta \chi \simeq a \|\partial_\nu \chi_\mu\|$ must go to zero, and hence $P(\Delta \chi)$ approaches to the $\delta$-function with the peak at $\Delta \chi=0$. However, as shown in Fig.2(a), $P(\Delta \chi)$ is almost $a$-independent and almost flat. These features indicate the [strong randomness of the off-diagonal gluon phase]{} $\chi_\mu(x)$ in the MA gauge. Then, $\chi_\mu(x)$ is approximately a random angle variable in the MA gauge.[^1] Randomness of Off-diagonal Gluon Phase and Abelian Dominance ------------------------------------------------------------ Within the random-variable approximation$^{3-5}$ for the off-diagonal gluon phase $\chi_\mu(s)$ in the MA gauge, we analytically prove abelian dominance of the string tension. Here, we use $$\begin{aligned} \langle e^{i\chi_\mu(s)}\rangle_{\rm MA} \simeq \frac1{2\pi} \int_0^{2\pi} d\chi_\mu(s)\exp\{i\chi_\mu(s)\}=0. \end{aligned}$$ In the Wilson loop $\langle W_C[U]\rangle \equiv \langle{\rm tr}\Pi_C U_\mu(s) \rangle= \langle{\rm tr}\Pi_C\{M_\mu(s)u_\mu(s)\}\rangle$, the off-diagonal matrix $M_\mu(s)$ is reduced to a $c$-number factor, $$\begin{aligned} M_\mu(s) \rightarrow \cos \theta_\mu(s) \ {\bf 1},\end{aligned}$$ and then the SU(2) link variable $U_\mu(s)$ is reduced to a [diagonal matrix]{} as $$\begin{aligned} U_\mu(s)\equiv M_\mu(s)u_\mu(s) \rightarrow \cos \theta_\mu(s) u_\mu(s), \end{aligned}$$ after the integration over $\chi_\mu(s)$. For the $R \times T$ rectangular $C$, the Wilson loop $W_C[U]$ in the MA gauge is approximated as $$\begin{aligned} \langle W_C[U]\rangle &=& \langle{\rm tr}\Pi_{i=1}^L \{M_{\mu_i}(s_i)u_{\mu_i}(s_i)\}\rangle \simeq \langle\Pi_{i=1}^L \cos \theta_{\mu_i}(s_i) \cdot {\rm tr} \Pi_{j=1}^L u_{\mu_j}(s_j)\rangle_{\rm MA} \nonumber \\ &\simeq& \langle\exp\{\Sigma_{i=1}^L \ln (\cos \theta_{\mu_i}(s_i))\}\rangle_{\rm MA} \ \langle W_C[u]\rangle_{\rm MA} \end{aligned}$$ with perimeter $L \equiv 2(R+T)$ and the abelian Wilson loop $W_C[u] \equiv {\rm tr}\Pi_{i=1}^L u_{\mu_i}(s_i)$. Replacing $\sum_{i=1}^L \ln \{\cos(\theta_{\mu_i}(s_i))\}$ by its average $L \langle \ln (\cos \theta_\mu) \rangle_{\rm MA}$ in a statistical sense, we derive the [perimeter law]{} of the [off-diagonal gluon contribution to the Wilson loop]{} as W\_C\^[off]{} W\_C\[U\]/W\_C\[u\]\_[MA]{} {L (\_) \_[MA]{}}, which is confirmed in lattice QCD, as shown in Fig.2(b). Near the continuum limit, we find also the relation between the [macroscopic]{} quantity $W_C^{\rm off}$ and the [microscopic]{} quantity of the off-diagonal gluon amplitude $\langle \|eA_\mu^\pm\|^2 \rangle_{\rm MA}$ as W\_C\^[off]{} W\_C\[U\]/W\_C\[u\]\_[MA]{} {-L \|eA\_\^\|\^2 \_[MA]{}}. In this way, [perfect abelian dominance for the string tension]{}, $\sigma_{\rm SU(2)}=\sigma_{\rm Abel}$, is analytically derived within the random-variable approximation for the off-diagonal gluon phase in the MA gauge. Strongly Random Phase and Large Mass of Off-diagonal Gluons ----------------------------------------------------------- As another remarkable fact, [strong randomness of off-diagonal gluon phases leads to rapid reduction of off-diagonal gluon correlations.]{} In fact, if $\chi_\mu(x)$ is a complete random phase, Euclidean off-diagonal gluon propagators exhibit $\delta$-functional reduction, A\_\^+(x) A\_\^-(y) \_[MA]{} &=& \|A\_\^+(x)\|\|A\_\^-(y)\|e\^[i{\_(x)-\_(y)}]{} \_[MA]{}\ &=& \|A\_\^(x)\|\^2 \_[MA]{} \_\^4(x-y), which means the infinitely large mass of off-diagonal gluons. Of course, the real off-diagonal gluon phases are not complete but approximate random phases. Then, off-diagonal gluon mass would be large but finite. In this way, [strong randomness of off-diagonal gluon phases is expected to provide a large effective mass of off-diagonal gluons.]{} Large Mass Generation of Off-diagonal Gluons in MA Gauge : Essence of Infrared Abelian Dominance ================================================================================================ Using SU(2) lattice QCD, we actually investigate the Euclidean gluon propagator $G_{\mu \nu }^{ab} (x-y) \equiv \langle A_\mu ^a(x)A_\nu ^b(y)\rangle$ ($a,b =1,2,3$) and the off-diagonal gluon mass $M_{\rm off}$ in the MA gauge. As for the residual U(1)$_3$ gauge symmetry, we take U(1)$_3$ Landau gauge, to extract most continuous gluon configuration under the MA gauge constraint and to compare with the continuum theory. The continuum gluon field $A_\mu^a(x)$ is derived from the link variable as $U_\mu(s)={\rm exp}\{iaeA_\mu^a(s) \frac{\tau^a}{2}\}$. We show in Fig.3(a) the scalar-type gluon propagators $G_{\mu \mu}^3(r)$ and $G_{\mu\mu}^{+-}(r) \equiv \langle A_\mu^{+}(x)A_\mu^{-}(y)\rangle = \frac12 \{G_{\mu\mu}^1(r)+G_{\mu\mu}^2(r)\}$, which depend only on the four-dimensional Euclidean distance $r \equiv \sqrt{(x_\mu- y_\mu)^2}$, in SU(2) lattice QCD with $2.2 \le \beta \le 2.4$ with various sizes ($12^3 \times 24$, $16^4$, $20^4$). We find [infrared abelian dominance for the gluon propagator in the MA gauge]{}: only the abelian gluon $A_\mu^3(x)$ propagates over the long distance and can influence the long-distance physics.$^{4,5,10}$ Since the four-dimensional Euclidean propagator of the massive vector boson with the mass $M$ takes a Yukawa-type asymptotic form as G\_(r) = 3[4\^2]{} K\_1(Mr)+1[M\^2]{}\^4(x-y) , we investigate the effective mass $M_{\rm off}$ of off-diagonal gluons $A_\mu^{\pm}(x)$ from the slope of the logarithmic plot of $r^{3/2} G_{\mu\mu}^{+-}(r)\sim \exp\{-M_{\rm off}r\}$ in Fig.3(b). 1. The off-diagonal gluon $A_\mu^{\pm}(x)$ behaves as a massive field with a large mass about 1 GeV for $r \ge 0.2 {\rm fm}$ in the MA gauge. 2. From the fitting analysis of the lattice QCD data with $r \ge 0.2 {\rm fm}$, the off-diagonal gluon mass is evaluated as $M_{\rm off} \simeq 1.2~{\rm GeV}$ in the MA gauge. We perform also the mass measurement of off-diagonal gluons from the temporal correlation of the zero-momentum projected operator $O_\mu^\pm(\tau)$, \_\^[+-]{}() O\_\^+()O\_\^-(0) , O\_\^() d[x]{} A\_\^([x]{},), in lattice QCD in the MA gauge plus ${\rm U(1)}_3$ Landau gauge. We find the off-diagonal gluon mass $M_{\rm off} \simeq 1.2 {\rm GeV}$ again from the slope of the logarithmic plot of $\Gamma_{\mu\mu}^{+-}(\tau)$ in SU(2) lattice QCD with $2.3 \le \beta \le 2.35$ with $16^3\times 32$ and $12^3\times 24$. Thus, [off-diagonal gluons $A_\mu^\pm$ acquire a large effective mass $M_{\rm off} \simeq 1.2 {\rm GeV}$ in the MA gauge]{}, which is [essence of infrared abelian dominance]{}.$^{4,5,10}$ In the MA gauge, due to the large effective mass $M_{\rm off}\simeq 1.2 {\rm GeV}$, off-diagonal gluons $A_\mu^\pm$ can propagate only within a short range as $r {~^{<~}_{\sim~}}M_{\rm off}^{-1} \simeq 0.2{\rm fm}$, and becomes [infrared inactive]{} like weak bosons in the Standard Model. Then, in the MA gauge, off-diagonal gluons $A_\mu^\pm$ cannot contribute to the infrared NP-QCD, which leads to infrared abelian dominance.$^{4,5,7,10,14}$ ![ (a) The scalar-type gluon propagator $G_{\mu \mu }^a(r)$ as the function of the four-dimensional distance $r$ in the MA gauge in SU(2) lattice QCD with $2.2 \le \beta \le 2.4$ with various sizes ($12^3 \times 24$, $16^4$, $20^4$). (b) The logarithmic plot of $r^{3/2} G_{\mu \mu}^a(r)$ v.s. $r$. The off-diagonal gluon propagator behaves as the Yukawa-type function, $G_{\mu \mu } \sim {\exp(-M_{\rm off}r) \over r^{3/2}}$. (c) The logarithmic plot of the temporal correlation $ \Gamma_{\mu\mu}^{+-}(\tau) \equiv \langle O_\mu^+(\tau)O_\mu^-(0) \rangle $ as the function of the temporal distance $\tau$ in SU(2) lattice QCD with $2.3 \le \beta \le 2.35$ with $16^3 \times 32$ and $12^3 \times 24$. From the slope of the dotted lines in (b) and (c), the effective mass of the off-diagonal gluon $A_\mu^\pm$ is estimated as $M_{\rm off}\simeq 1.2 {\rm GeV}$.](Fig3a.eps "fig:"){height="3.3cm"} ![ (a) The scalar-type gluon propagator $G_{\mu \mu }^a(r)$ as the function of the four-dimensional distance $r$ in the MA gauge in SU(2) lattice QCD with $2.2 \le \beta \le 2.4$ with various sizes ($12^3 \times 24$, $16^4$, $20^4$). (b) The logarithmic plot of $r^{3/2} G_{\mu \mu}^a(r)$ v.s. $r$. The off-diagonal gluon propagator behaves as the Yukawa-type function, $G_{\mu \mu } \sim {\exp(-M_{\rm off}r) \over r^{3/2}}$. (c) The logarithmic plot of the temporal correlation $ \Gamma_{\mu\mu}^{+-}(\tau) \equiv \langle O_\mu^+(\tau)O_\mu^-(0) \rangle $ as the function of the temporal distance $\tau$ in SU(2) lattice QCD with $2.3 \le \beta \le 2.35$ with $16^3 \times 32$ and $12^3 \times 24$. From the slope of the dotted lines in (b) and (c), the effective mass of the off-diagonal gluon $A_\mu^\pm$ is estimated as $M_{\rm off}\simeq 1.2 {\rm GeV}$.](Fig3b.eps "fig:"){height="3.3cm"} ![ (a) The scalar-type gluon propagator $G_{\mu \mu }^a(r)$ as the function of the four-dimensional distance $r$ in the MA gauge in SU(2) lattice QCD with $2.2 \le \beta \le 2.4$ with various sizes ($12^3 \times 24$, $16^4$, $20^4$). (b) The logarithmic plot of $r^{3/2} G_{\mu \mu}^a(r)$ v.s. $r$. The off-diagonal gluon propagator behaves as the Yukawa-type function, $G_{\mu \mu } \sim {\exp(-M_{\rm off}r) \over r^{3/2}}$. (c) The logarithmic plot of the temporal correlation $ \Gamma_{\mu\mu}^{+-}(\tau) \equiv \langle O_\mu^+(\tau)O_\mu^-(0) \rangle $ as the function of the temporal distance $\tau$ in SU(2) lattice QCD with $2.3 \le \beta \le 2.35$ with $16^3 \times 32$ and $12^3 \times 24$. From the slope of the dotted lines in (b) and (c), the effective mass of the off-diagonal gluon $A_\mu^\pm$ is estimated as $M_{\rm off}\simeq 1.2 {\rm GeV}$.](Fig3c.eps "fig:"){height="3.2cm"} (a)(b)(c) Lattice-QCD Evidence of Monopole Condensation ============================================= In the MA gauge, there appears the global network of the monopole world-line covering the whole system as shown in Fig.4(a), and this monopole-current system (the monopole part) holds essence of NP-QCD. Using SU(2) lattice QCD, we examine the dual Higgs mechanism by monopole condensation in the NP-QCD vacuum in the MA gauge.$^{4,5}$ Since QCD is described by the “electric variable” as quarks and gluons, the “electric sector” of QCD has been well studied with the Wilson loop or the inter-quark potential, however, the “magnetic sector” of QCD is hidden and still unclear. To investigate the magnetic sector directly, it is useful to introduce the “dual (magnetic) variable” as the [dual gluon field]{} $B_\mu $, which is the dual partner of the diagonal gluon and directly couples with the magnetic current $k_\mu $. ![The SU(2) lattice-QCD results in the MA gauge. (a) The monopole world-line projected into ${\bf R}^3$ on the $16^3 \times 4$ lattice with $\beta =2.2$ (the confinement phase). There appears a global network of monopole currents covering the whole system. (b) The inter-monopole potential $V_M(r)$ v.s. the 3-dimensional distance $r$ in the monopole-current system on the $20^4$ lattice. The solid curve denotes the Yukawa potential with $m_B=0.5$GeV. The dotted curve denotes the Yukawa-type potential including the monopole-size effect. (c) The scalar-type dual-gluon correlation $\ln (r_E^{3/2} \langle B_\mu(x)B_\mu(y)\rangle_{\rm MA})$ as the function of the 4-dimensional Euclidean distance $r_E$ on the $24^4$ lattice. The slope corresponds to the dual gluon mass $m_B$.](Fig4a.eps "fig:"){height="3.4cm"} ![The SU(2) lattice-QCD results in the MA gauge. (a) The monopole world-line projected into ${\bf R}^3$ on the $16^3 \times 4$ lattice with $\beta =2.2$ (the confinement phase). There appears a global network of monopole currents covering the whole system. (b) The inter-monopole potential $V_M(r)$ v.s. the 3-dimensional distance $r$ in the monopole-current system on the $20^4$ lattice. The solid curve denotes the Yukawa potential with $m_B=0.5$GeV. The dotted curve denotes the Yukawa-type potential including the monopole-size effect. (c) The scalar-type dual-gluon correlation $\ln (r_E^{3/2} \langle B_\mu(x)B_\mu(y)\rangle_{\rm MA})$ as the function of the 4-dimensional Euclidean distance $r_E$ on the $24^4$ lattice. The slope corresponds to the dual gluon mass $m_B$.](Fig4b.EPSF "fig:"){height="3.0cm"} ![The SU(2) lattice-QCD results in the MA gauge. (a) The monopole world-line projected into ${\bf R}^3$ on the $16^3 \times 4$ lattice with $\beta =2.2$ (the confinement phase). There appears a global network of monopole currents covering the whole system. (b) The inter-monopole potential $V_M(r)$ v.s. the 3-dimensional distance $r$ in the monopole-current system on the $20^4$ lattice. The solid curve denotes the Yukawa potential with $m_B=0.5$GeV. The dotted curve denotes the Yukawa-type potential including the monopole-size effect. (c) The scalar-type dual-gluon correlation $\ln (r_E^{3/2} \langle B_\mu(x)B_\mu(y)\rangle_{\rm MA})$ as the function of the 4-dimensional Euclidean distance $r_E$ on the $24^4$ lattice. The slope corresponds to the dual gluon mass $m_B$.](Fig4c.EPSF "fig:"){height="3.0cm"} (a)(b)(c) Owing to the absence of the electric current $j_\mu$ in the monopole part, the dual gluon $B_\mu $ can be introduced as the regular field satisfying $(\partial \land B)_{\mu\nu}={^\!F}_{\mu\nu}$ and the dual Bianchi identity, $ {\partial^{\mu}} {^\!(}\partial \land B)_{\mu\nu}=j_\nu=0. $ By taking the dual Landau gauge $\partial_\mu B^\mu=0$, the field equation is simplified as $\partial^2 B_\mu ={\partial_\alpha} {^\!F}_{\alpha \mu}=k_\mu$, and therefore we obtain the dual gluon field $B_\mu$ from the monopole current $k_\mu$ as B\_(x) = ( \^[-2]{} k\_)(x)= - d\^4y . Since the dual gluon $B_\mu $ is massive under monopole condensation, we investigate the dual gluon mass $m_B$ as the evidence of the dual Higgs mechanism. First, we put test magnetic charges in the monopole-current system in the MA gauge in SU(2) lattice QCD, and measure the inter-monopole potential $V_M(r)$ to get information about monopole condensation. Since the dual Higgs mechanism provides the [screening effect on the magnetic flux]{}, $V_M(r)$ is expected to be short-range Yukawa-type, if monopole condensation occurs. The potential between the monopole and the anti-monopole can be derived as V\_[M]{}(R) = -\_[T ]{} [1 T]{} W\_D(R,T) , using [the dual Wilson loop]{} $W_D$ as the loop-integral of the dual gluon,$^{4,5}$ W\_D(C) {i[e 2]{}\_C dx\_B\^}= {i[e 2]{}d\_[\^\F]{}\^}, which is the [dual version of the abelian Wilson loop]{} $$\begin{aligned} W_{\rm Abel}(C) \equiv \exp\{i{e \over 2}\oint_C dx_\mu A_3^\mu \} =\exp\{i{e \over 2}\int\!\!\!\int d\sigma_{\mu\nu}{F}^{\mu\nu}\}\end{aligned}$$ and the test monopole charge is set to be $e/2$. In Fig.4(b), we show $V_M(r)$ in the monopole part in the MA gauge.$^{4,5}$ Except for the short distance, the inter-monopole potential is well fitted by the Yukawa potential $$\begin{aligned} V_M(r) = -{{(e/2)}^2 \over 4\pi}{e^{-m_Br} \over r}, \end{aligned}$$ and thus the [magnetic screening]{} is observed. In the MA gauge, the dual gluon mass is estimated as $m_B \simeq {\rm 0.5GeV}$ from the infrared behavior of $V_M(r)$. Second, we investigate also the Euclidean scalar-type dual gluon propagator $\langle B_\mu(x)B_\mu(y)\rangle_{\rm MA}$ as shown in Fig.4(c), and estimate the dual gluon mass as $m_B = 0.4 \sim 0.5$ GeV from its long-distance behavior.$^{5}$ From these two tests, the dual gluon mass is evaluated as $m_B=0.4 \sim 0.5$ GeV, and this can be regarded as the lattice-QCD evidence for the dual Higgs mechanism by monopole condensation at the infrared scale.$^{4,5}$ To conclude, lattice QCD in the MA gauge exhibits [infrared abelian dominance]{} and [infrared monopole condensation]{}, and therefore the dual Ginzburg-Landau (DGL) theory$^{8,17-23}$ can be constructed as the infrared effective theory directly based on QCD in the MA gauge. (See Fig.5). ![ Construction of the dual Ginzburg-Landau (DGL) theory from lattice QCD in the maximally abelian gauge. ](Fig5.eps){height="14cm"} Monopole Structure in terms of the Off-diagonal Gluon ===================================================== Let us compare the QCD-monopole with the point-like Dirac monopole. There is no point-like monopole in QED, because the QED action diverges around the point-like monopole. The QCD-monopole also accompanies a large abelian action density, however, [owing to cancellation with the off-diagonal gluon contribution, the total QCD action is kept finite even around the QCD-monopole.]{}$^{3-5}$ To see this, we examine the QCD-monopole structure in the MA gauge in terms of the action density using SU(2) lattice QCD.$^{3-5}$ From the SU(2) plaquette $P^{\rm SU(2)}_{\mu \nu }(s)$ and the abelian plaquette $P^{\rm Abel}_{\mu \nu }(s) \in {\rm U(1)}_3 \subset {\rm SU(2)}$, we define the “SU(2) action density” $ S_{\mu \nu }^{\rm SU(2)}(s) \equiv 1-{1 \over 2}{\rm tr}P^{\rm SU(2)}_{\mu \nu }(s), $ the “abelian action density” $ S_{\mu \nu }^{\rm Abel}(s) \equiv 1-{1 \over 2}{\rm tr}P^{\rm Abel}_{\mu \nu }(s) $ and the “off-diagonal gluon contribution” $ S_{\mu \nu }^{\rm off}(s) \equiv S_{\mu \nu }^{\rm SU(2)}(s)-S_{\mu \nu }^{\rm Abel}(s). $ In the lattice formalism, the monopole current $k_\mu (s)$ is defined on the dual link, and there are 6 plaquettes around the monopole. Then, we consider the local average over the 6 plaquettes around the dual link, $$\begin{aligned} S(s,\mu) \equiv {1 \over 12} \sum_{\alpha\beta\gamma} \sum_{m=0}^1 \| \varepsilon_{\mu\alpha\beta\gamma} \| S_{\alpha\beta}(s + m \hat \gamma).\end{aligned}$$ We show in Fig.6(b) the probability distribution of the action densities $S_{\rm SU(2)}$, $S_{\rm Abel}$ and $S_{\rm off}$ around the QCD-monopole in the MA gauge. ![ (a) The total probability distribution $P(S)$ on the whole lattice and (b) the probability distribution $P_k(S)$ around the monopole for SU(2) action density $S_{\rm SU(2)}$ (dashed curve), abelian action density $S_{\rm Abel}$ (solid curve) and off-diagonal gluon contribution $S_{\rm off}$ (dotted curve) in the MA gauge at $\beta =2.4$ on $16^4$ lattice. Around the QCD-monopole, large cancellation between $S_{\rm Abel}$ and $S_{\rm off}$ keeps the total QCD-action small. (c) The schematic figure for the QCD-monopole structure in the MA gauge. The QCD-monopole includes a large amount of off-diagonal gluons around its center as well as the diagonal gluon.](Fig6a.EPSF "fig:"){height="2.6cm"} ![ (a) The total probability distribution $P(S)$ on the whole lattice and (b) the probability distribution $P_k(S)$ around the monopole for SU(2) action density $S_{\rm SU(2)}$ (dashed curve), abelian action density $S_{\rm Abel}$ (solid curve) and off-diagonal gluon contribution $S_{\rm off}$ (dotted curve) in the MA gauge at $\beta =2.4$ on $16^4$ lattice. Around the QCD-monopole, large cancellation between $S_{\rm Abel}$ and $S_{\rm off}$ keeps the total QCD-action small. (c) The schematic figure for the QCD-monopole structure in the MA gauge. The QCD-monopole includes a large amount of off-diagonal gluons around its center as well as the diagonal gluon.](Fig6b.EPSF "fig:"){height="2.6cm"} ![ (a) The total probability distribution $P(S)$ on the whole lattice and (b) the probability distribution $P_k(S)$ around the monopole for SU(2) action density $S_{\rm SU(2)}$ (dashed curve), abelian action density $S_{\rm Abel}$ (solid curve) and off-diagonal gluon contribution $S_{\rm off}$ (dotted curve) in the MA gauge at $\beta =2.4$ on $16^4$ lattice. Around the QCD-monopole, large cancellation between $S_{\rm Abel}$ and $S_{\rm off}$ keeps the total QCD-action small. (c) The schematic figure for the QCD-monopole structure in the MA gauge. The QCD-monopole includes a large amount of off-diagonal gluons around its center as well as the diagonal gluon.](Fig6c.eps "fig:"){height="2.9cm"} (a)(b)(c) We summarize the results on the QCD-monopole structure as follows. 1. Around the QCD-monopole, both the abelian action density $S_{\rm Abel}$ and the off-diagonal gluon contribution $S_{\rm off}$ are largely fluctuated, and their cancellation keeps the total QCD-action density $S_{\rm SU(2)}$ small. 2. The QCD-monopole has an [intrinsic structure relating to a large amount of off-diagonal gluons]{} around its center like the ’t Hooft-Polyakov monopole. At a large scale, off-diagonal gluons inside the QCD-monopole become invisible, and QCD-monopoles can be regarded as point-like Dirac monopoles. 3. From the concentration of off-diagonal gluons around QCD-monopoles in the MA gauge, we can naturally understand the [local correlation between monopoles and instantons]{}.$^{3-5,11,12,15,16}$ In fact, instantons tend to appear around the monopole world-line in the MA gauge, because instantons need full SU(2) gluon components for existence. Gluonic Higgs and Gauge Invariant Description of MA Projection =============================================================== In the MA gauge, the gauge group $G \equiv {\rm SU}(N_c)$ is partially fixed into its subgroup $H \equiv {\rm U}^{N_c-1}_{\rm local}\times {\rm Weyl}_{N_c}^{\rm global}$, and then the gauge invariance becomes unclear.[^2] In this section, we propose a [gauge invariant description of the MA projection in QCD]{}. Even without explicit use of gauge fixing, we can naturally define the MA projection by introducing a “gluonic Higgs scalar field” $\phi(x)$. For a given gluon field configuration $\{ A_\mu(x) \}$, we define a gluonic Higgs scalar $\vec \phi(x) =\Omega(x) \vec H \Omega^\dagger(x)$ with $\Omega(x) \in {\rm SU}(N_c)$ so as to minimize R\[()\] d\^4x [tr]{} {\[D\_, (x)\]\[D\_, (x)\]\^}. We summarize the features of this description as follows.$^{3}$ 1. The gluonic Higgs scalar $\vec \phi(x)$ does not have amplitude degrees of freedom but has only color-direction degrees of freedom, and $\vec \phi(x)$ corresponds to a “color-direction” of the nonabelian gauge connection $\hat D_\mu$ averaged over $\mu$ at each space-time $x$. 2. Through the projection along $\vec \phi(x)$, we can extract the abelian U(1)$^{N_c-1}$ sub-gauge-manifold which is most close to the original SU($N_c$) gauge manifold. This projection is manifestly gauge invariant, and is mathematically equivalent to the ordinary MA projection.$^3$ 3. Similar to $\hat D_\mu$, the gluonic Higgs scalar $\vec \phi(x)$ obeys the adjoint gauge transformation, and is diagonalized in the MA gauge. Then, monopoles appear at the hedgehog singularities of $\vec \phi(x)$ as shown in Fig.7.$^{3,8}$ 4. In this description, infrared abelian dominance is interpreted as infrared relevance of the gluon mode along the color-direction $\vec \phi(x)$, and QCD seems similar to a nonabelian Higgs theory. ![ The correlation between the gluonic Higgs scalar field $\phi(x)=\phi^a(x)\frac{\tau^a}{2}$ and monopoles denoted by dots in SU(2) lattice QCD with $\beta=2.4$ and $16^4$. The arrow denotes the color direction of $(\phi^1(x),\phi^2(x),\phi^3(x))$ in the SU(2) internal space. The monopole in the MA gauge appears at the hedgehog singularity of the gluonic Higgs scalar $\phi(x)$ in the Landau gauge.](Fig7.eps){height="3cm"} The 3Q Ground-State Potential in SU(3) Lattice QCD ================================================== In relation to the color-flux-tube picture for baryons,[@CI86; @BPV95] we study the three-quark (3Q) ground-state potential in SU(3)$_c$ lattice QCD at the quenched level.$^{26-28}$ In contrast with a number of studies on the Q-$\bar {\rm Q}$ potential using lattice QCD, there were only a few preliminary lattice-QCD works for the 3Q potential.$^{29-31}$ To begin with, let us consider the potential form in the Q-$\bar{\rm Q}$ and 3Q systems with respect to QCD. In the short-distance limit, perturbative QCD is applicable and the Coulomb-type potential appears as the one-gluon-exchange (OGE) result. In the long-distance limit at the quenched level, the flux-tube picture[@CI86; @BPV95] would be applicable from the argument of the strong-coupling limit of QCD[@KS75], and hence a linear confinement potential proportional to the total flux-tube length is expected to appear. Indeed, lattice QCD results for the Q-$\bar {\rm Q}$ ground-state potential are well described by $$V_{\rm Q \bar{Q}}(r)=-\frac{A_{\rm Q \bar{Q}}}{r} +\sigma_{\rm Q \bar{Q}} r+C_{\rm Q \bar{Q}} \label{QQpot}$$ at the quenched level. In fact, $V_{\rm Q \bar{Q}}$ is described by a sum of the short-distance OGE result and the long-distance flux-tube result. Also for the 3Q ground-state potential $V_{\rm 3Q}$, we try to apply this simple picture of the short-distance OGE result plus the long-distance flux-tube result. Then, the 3Q potential $V_{\rm 3Q}$ is expected to take a form of $$V_{\rm 3Q}=-A_{\rm 3Q}\sum_{i<j}\frac1{\|{\bf r}_i-{\bf r}_j\|} +\sigma_{\rm 3Q} L_{\rm min}+C_{\rm 3Q}, \label{3Qpot}$$ with $L_{\rm min}$ the minimal value of total length of flux tubes linking the three quarks. Similar to the derivation of the Q-${\bar {\rm Q}}$ potential from the Wilson loop, the 3Q static potential $V_{\rm 3Q}$ can be derived from the 3Q Wilson loop $W_{\rm 3Q}$ as $^{25-31}$ V\_[3Q]{}=-\_[T ]{} 1T W\_[3Q]{}, W\_[3Q]{} 1[3!]{}\_[abc]{}\_[a’b’c’]{} U\_1\^[aa’]{} U\_2\^[bb’]{} U\_3\^[cc’]{} with $U_k \equiv {\rm P}\exp\{ig\int_{\Gamma_k}dx^\mu A_{\mu}(x)\}$ ($k=1,2,3$) (see Fig.8). Here, the 3Q Wilson loop $W_{\rm 3Q}$ is defined in a gauge-invariant manner. The 3Q gauge-invariant state is generated at $t=0$ and is annihilated at $t=T$. For $0 < t < T$, the three quarks are spatially fixed in ${\bf R}^3$. ![The 3Q Wilson loop $W_{\rm 3Q}$. The 3Q state is generated at $t=0$ and is annihilated at $t=T$. The three quarks are spatially fixed in ${\bf R}^3$ for $0 < t < T$.](Fig8.eps){height="3.75cm"} Physically, the true ground state of the 3Q system would be expressed by the flux tubes instead of the strings, and the 3Q state which is expressed by the three strings generally includes many excited-state components such as flux-tube vibrational modes. Since the practical measurement of $\langle W_{\rm 3Q}\rangle$ is quite severe for large $T$ in lattice QCD calculations, the smearing technique for ground-state enhancement[@SMNT00; @TMNS00] is practically indispensable for the accurate measurement of the 3Q ground-state potential $V_{\rm 3Q}$. The standard smearing for link-variables[@APE87] is expressed as the iteration of the replacement of the spatial link-variable $U_i (s)$ ($i=1,2,3$) by the obscured link-variable $\bar U_i (s) \in {\rm SU(3)}_c$ which maximizes $${\rm Re} \,\, {\rm tr} \Bigl( \bar U_i^{\dagger}(s) \Big\{ \alpha U_i(s)+\sum_{\pm, \ j \ne i} U_{\pm j}(s)U_i(s \pm \hat j)U_{\pm j}^\dagger (s + \hat i) \Big\} \Bigr),$$ with a real smearing parameter $\alpha$ and $U_{-\mu}(s) \equiv U_{\mu}^{\dagger}(s-\hat \mu)$. The $n$-th smeared link-variables $U_\mu^{(n)}(s)$ $(n=1,2,..,N_{\rm smear})$ are iteratively defined starting from $U_\mu^{(0)}(s) \equiv U_\mu(s)$ as $$U_i^{(n)}(s) \equiv \bar U_i^{(n-1)}(s) \quad (i=1,2,3), \qquad U_4^{(n)}(s) \equiv U_4(s).$$ This smearing procedure keeps the gauge covariance of the “fat” link-variable $U_\mu^{(n)}(s)$ properly. In fact, the gauge invariance of $F(U_\mu^{(n)}(s))$ is ensured if $F(U_\mu (s))$ is a gauge-invariant function. Since the fat link-variable $U_\mu^{(n)}(s)$ includes a spatial extension, the “spatial line” expressed with $U_\mu^{(n)}(s)$ physically corresponds to a “flux tube” with a spatial extension. Therefore, if a suitable smearing is done, the “spatial line” of the fat link-variable is expected to be close to the ground-state flux tube. We search reasonable values of the smearing parameter $\alpha$ and the iteration number $N_{\rm smear}$ in the lattice QCD calculation. We show lattice QCD results for the 3Q ground-state potential. We generate 210 gauge configurations using SU(3)$_c$ lattice QCD Monte-Carlo simulation with the standard action with $\beta=5.7$ and $12^{3} \times 24$ at the quenched level. The lattice spacing $a \simeq 0.19 \,{\rm fm}$ is determined so as to reproduce the string tension as $\sigma$=0.89 GeV/fm in the Q-$\bar{\rm Q}$ potential $V_{\rm Q \bar{Q}}$. Here, the pseudo-heat-bath algorithm is used for updating, and the gauge configurations are taken every 500 sweeps after a thermalization of 5000 sweeps. We measure the 3Q ground-state potential $V_{\rm 3Q}$ using the smearing technique, and compare the lattice data with the theoretical form of Eq.(\[3Qpot\]). Owing to the smearing with $\alpha$=2.3 and $N_{\rm smear}$=12, the ground-state component is largely enhanced, and therefore the 3Q Wilson loop $\langle W_{\rm 3Q} \rangle$ composed with the smeared link-variable exhibits a single-exponential behavior as $\langle W_{\rm 3Q} \rangle \simeq e^{-V_{\rm 3Q}T}$ even for a small value of $T$. For each 3Q configuration, we extract $V_{\rm 3Q}^{\rm lat}$ from the least squares fit with the single-exponential form $$\langle W_{\rm 3Q}\rangle =\bar{C}e^{-V_{\rm 3Q}T} \label{EXfit}$$ in the range of $T_{\rm min}\leq T\leq T_{\rm max}$ listed in Table 2. Here, the fit range of $T$ is chosen such that the stability of the “effective mass” V(T){W\_[3Q]{}(T) / W\_[3Q]{}(T+1)} is observed in the range of $T_{\rm min}\leq T\leq T_{\rm max}-1$. For each 3Q configuration, we summarize the lattice QCD data $V_{\rm 3Q}^{\rm lat}$ as well as the prefactor $\bar{C}$ in Eq.(\[EXfit\]), fit range of $T$ and $\chi^2/N_{\rm DF}$ in Table 2. The statistical error of $V_{\rm 3Q}^{\rm lat}$ is estimated with the jackknife method. We find a large ground-state overlap as $\bar{C} > 0.8$ for all 3Q configurations. Now, we consider the potential form of $V_{\rm 3Q}$. We show in Table 1 the best fit parameters in Eq.(\[3Qpot\]) for $V_{\rm 3Q}$. We compare in Table 2 the lattice QCD data $V_{\rm 3Q}^{\rm lat}$ with the fitting function $V_{\rm 3Q}^{\rm fit}$ in Eq.(\[3Qpot\]) with the best fit parameters. [The three-quark ground-state potential $V_{\rm 3Q}$ is well described by Eq.(\[3Qpot\]) with accuracy better than a few %,]{} although $\chi^2/N_{\rm DF}=3.76$ seems relatively large, which may reflect a systematic error on the finite lattice spacing. (The fitting with $\Delta$-type flux-tube ansatz suggested in Refs.\[29,31,34\] is rather worse and shows unacceptably large $\chi^2/N_{\rm DF}=10.1$ even for the best fit.) By comparing the coefficients $(\sigma_{\rm 3Q}, A_{\rm 3Q})$ with $(\sigma_{\rm Q\bar{Q}}, A_{\rm Q\bar{Q}})$ in Table 1, we find a [universal feature of the string tension,]{} $\sigma_{\rm 3Q} \simeq \sigma_{\rm Q\bar{Q}}$, and the [OGE result for the Coulomb coefficient,]{} $A_{\rm 3Q} \simeq \frac12 A_{\rm Q\bar{Q}}$. Acknowledgments {#acknowledgments .unnumbered} =============== The authors would like to thank Professor Yoichiro Nambu for his useful suggestions. The lattice calculations were performed on NEC-SX4 at Osaka University. [99]{} H. Suganuma [et al.]{}, [Proc of “NEWS ’99”]{} (World Scientific). Y. Nambu, [Phys. Rev.]{} [D10]{}, 4262 (1974). H. Ichie and H. Suganuma, [Phys. Rev.]{} [D60]{}, 77501 (1999);\ [Nucl. Phys.]{} [B548]{}, 365-382 (1999) ; [Nucl. Phys.]{} [B574]{}, 70-106 (2000). H. Suganuma, K. Amemiya, H. Ichie, A.Tanaka, [Nucl.Phys.]{} [A670]{}, 40 (2000). H. Suganuma [et al.]{}, [Prog. Theor. Phys. Suppl.]{} [131]{}, 559-571 (1998). G. ’t Hooft, [Nucl. Phys.]{} [B190]{}, 455 (1981). Z.F. Ezawa and A. Iwazaki, [Phys. Rev.]{} [D25]{}, 2681 (1982); [D26]{}, 631 (1982). H. Suganuma, S. Sasaki and H. Toki, [Nucl. Phys.]{} [B435]{}, 207-240 (1995). O. Miyamura, [Phys. Lett.]{} [B353]{}, 91 (1995). K. Amemiya and H. Suganuma, [Phys. Rev.]{} [D60]{}, 114509 (1999). H. Suganuma [et al.]{}, [Aust. J. Phys.]{} [50]{}, 233-243 (1997). H. Suganuma [et al.]{}, [Nucl. Phys.]{} [B]{} (Proc. Suppl.) [47]{}, 302 (1996); [53]{}, 528 (1997); [65]{}, 29 (1998). M.I. Polikarpov, [Nucl. Phys.]{} [B]{} (Proc. Suppl.) [53]{}, 134 (1997). K.-I. Kondo, [Phys. Rev.]{} [D57]{}, 7467 (1998); [D58]{}, 105016 (1998). M. Fukushima [et al.]{}, [Phys. Lett.]{} [B399]{}, 141 (1997). M. Fukushima, H. Suganuma and H. Toki, [Phys. Rev.]{} [D60]{}, 94504 (1999). H. Suganuma, S. Sasaki, H. Toki and H. Ichie, [Prog. Theor. Phys. Suppl.]{} [120]{}, 57-73 (1995). S. Sasaki, H. Suganuma and H. Toki, [Prog. Theor. Phys.]{} [94]{}, 373 (1995); [Phys. Lett.]{} [B387]{}, 145 (1996). H. Ichie, H. Suganuma and H. Toki, [Phys. Rev.]{} [D52]{}, 2994 (1995);\ [Phys. Rev.]{} [D54]{}, 3382 (1996). S. Umisedo, H. Suganuma and H. Toki, [Phys. Rev.]{} [D57]{}, 1605 (1998). H. Monden, H. Ichie, H. Suganuma, H. Toki, [Phys. Rev.]{} [C57]{}, 2564 (1998). Y. Koma, H. Suganuma and H. Toki, [Phys. Rev.]{} [D60]{}, 74024 (1999). H. Toki and H. Suganuma, [Prog. Part. Nucl. Phys.]{} [45]{}, 397-472 (2000). S. Capstick and N. Isgur, [Phys. Rev.]{} [D34]{}, 2809 (1986). N. Brambilla, G.M. Prosperi and A. Vairo, [Phys. Lett.]{} [B362]{}, 113 (1995). T.T. Takahashi, H. Matsufuru, Y. Nemoto, H. Suganuma, [Proc. of “Dynamics of Gauge Fields”]{}, Tokyo, 1999, (Universal Academy Press, 2000) p.179. H. Suganuma, H. Matsufuru, Y. Nemoto and T.T. Takahashi, [Nucl. Phys.]{} [A680]{}, 159 (2000). T.T. Takahashi, H. Matsufuru, Y. Nemoto and H. Suganuma, [Phys. Rev. Lett.]{} [86]{}, 18 (2001). R. Sommer and J. Wosiek, [Phys.Lett.]{} [149B]{}, 497 (1984); [Nucl.Phys.]{} [B267]{}, 531 (1986). J. Kamesberger [et al.]{}, [Proc of “Few-Body Problems in Particle, Nuclear, Atomic and Molecular Physics”]{}, 529 (1987). H.B. Thacker, E. Eichten, J.C. Sexton, [Nucl. Phys.]{} [B]{} (Proc.) [4]{}, 234 (1988). J. Kogut and L. Susskind, [Phys. Rev.]{} [D11]{}, 395 (1975). APE Collaboration, M. Albanese [et al.]{}, [Phys. Lett.]{} [B192]{}, 163 (1987). J.M. Cornwall, [Phys. Rev.]{} [D54]{}, 6527 (1996). [^1]: However, as shown in Sect. 6, near the monopole, there remains a large amplitude of $\|\sin \theta_\mu(s)\|$ even in the MA gauge, and $\chi_\mu(s)$ is strongly constrained so as to make the QCD action small. Hence, $\chi_\mu(s)$ cannot be regarded as a random variable near monopoles. [^2]: In Refs.\[3,12\], we show a useful [gauge-invariance criterion]{} on the operator $O_{\rm MA}$ defined in the MA gauge: [If $O_{\rm MA}$ defined in the MA gauge is $H$-invariant, $O_{\rm MA}$ is also invariant under the whole gauge transformation of $G$.]{}	{ "pile_set_name": "ArXiv" }
--- abstract: 'We consider the astrophysical and cosmological implications of the existence of a minimum density and mass due to the presence of the cosmological constant. If there is a minimum length in nature, then there is an absolute minimum mass corresponding to a hypothetical particle with radius of the order of the Planck length. On the other hand, quantum mechanical considerations suggest a different minimum mass. These particles associated with the dark energy can be interpreted as the “quanta” of the cosmological constant. We study the possibility that these particles can form stable stellar-type configurations through gravitational condensation, and their Jeans and Chandrasekhar masses are estimated. From the requirement of the energetic stability of the minimum density configuration on a macroscopic scale one obtains a mass of the order of $10^{55}{\rm g}$, of the same order of magnitude as the mass of the universe. This mass can also be interpreted as the Jeans mass of the dark energy fluid. Furthermore we present a representation of the cosmological constant and of the total mass of the universe in terms of ‘classical’ fundamental constants.' author: - \| C. G. Böhmer[^1]\ Department of Mathematics, University College London,\ Gower Street, London, WC1E 6BT, UK - \| T. Harko[^2]\ Department of Physics and Center for Theoretical\ and Computational Physics, The University of Hong Kong,\ Pok Fu Lam Road, Hong Kong title: Physics of dark energy particles --- \ \ Keywords: gravitation, dark energy, minimum mass, dark energy particles\ \ PACS numbers: 03.70.+k, 11.90.+t, 11.10.Kk Introduction {#intro} ============ Several recent astrophysical observations of distant type Ia supernovae [@Ri98; @Pe99; @Ber00; @Ha00] have provided the astonishing result that around $95-96\%$ of the content of the universe is in the form of dark matter $+$ energy, with only about $4-5\%$ being represented by baryonic matter. More intriguing, around $70\%$ of the total energy-density may be in the form of what is called the dark energy, with the associated density parameter $% \Omega _{DE}$ of the order of $\Omega _{DE}\sim 0.70$. The dark energy is responsible for the recent acceleration of the Universe. The best candidate for the dark energy is the cosmological constant $\Lambda $, which is usually interpreted physically as a vacuum energy, with energy density $\rho _{\Lambda }$ and pressure $p_{\Lambda }$ satisfying the unusual equation of state $\rho _{\Lambda }=-p_{\Lambda }/c^{2}=\Lambda /8\pi G/c^2$. Its size is of the order $\Lambda \approx 3\times 10^{-56}$ cm$^{-2}$ [@PeRa03; @Pa03]. The existence of the cosmological constant modifies the allowed ranges for various physical parameters, like, for example, the maximum mass of compact stellar objects [@MaDoHa00; @Bo04; @Bo05; @BaBoNo05a], thus leading to a modifications of the “classical” Buchdahl limit [@Bu]. In conjunction with other parameters, like the Schwarzschild radius, the cosmological constant $\Lambda $ leads to a set of scales relevant not only for cosmological, but also for astrophysical applications. Hence, for example, there exists a lower and an upper cut-off on the possible velocities of test particles travelling over distances of the order of Mpc [@BaBoNo05]. Since about $70\%$ of the Universe consists of dark energy, which almost entirely determines its structure and dynamics, it is natural to consider $% \Lambda $ as a fundamental constant and to explore the possibilities which follows from enlarging the set of fundamental constants of nature, which can be considered as being the speed of light $c$, the gravitational constant $G$, Planck’s constant $\hbar $ and the cosmological constant $\Lambda $, respectively [@We04]. On the other hand, we cannot exclude a priori the possibility that the cosmological constant, which may also be interpreted as a manifestation of the vacuum energy, can also play an important role not only at galactic or cosmological scales, but also at the level of elementary particles. Therefore the presence of the cosmological constant may require a drastic modification of the basic laws of physics. In the presence of a cosmological constant, ordinary Poincaré special relativity is no longer valid, and must be replaced by a de Sitter special relativity, in which Minkowski space is replaced by a de Sitter spacetime [@Pe1]. Consequently, the ordinary notions of energy and momentum change, and will satisfy a different kinematic relation. Since the only difference between the Poincaré and the de Sitter groups is the replacement of translations by certain linear combinations of translations and proper conformal transformations, the net result of this change is ultimately the breakdown of ordinary translational invariance [@Pe2; @Pe3; @Pe4]. From the experimental point of view, therefore, a de Sitter special relativity might be probed by looking for possible violations of translational invariance. If we assume the existence of a connection between the energy scale of an experiment and the local value of the cosmological constant, there would be changes in the kinematics of massive particles which could hopefully be detected in high-energy experiments. Furthermore, due to the presence of a horizon, the usual causal structure of spacetime would be significantly modified at the Planck scale. By using dimensional analysis, Wesson [@We04] has found two different masses, which can be constructed from the set of constants $% (c,G,\hbar ,\Lambda )$. The mass $m_{P}$ relevant at the quantum scale is $$m_{P}=\left( \frac{\hbar }{c}\right) \sqrt{\frac{\Lambda }{3}}\approx 3.5\times 10^{-66}\,\mathrm{g}, \label{mp}$$ while the mass $m_{PE}$ relevant to the cosmological scale is $$m_{PE}=\left( \frac{c^{2}}{G}\right) \sqrt{\frac{3}{\Lambda }}\approx 1\times 10^{56}\,\mathrm{g}. \label{mpe}$$ The interpretation of the mass $m_{PE}$ is straightforward: it is the mass of the observable part of the universe, equivalent to $10^{80}$ baryons of mass $10^{-24}$ g each. The interpretation of the mass $m_{P}$ is more difficult. By using the dimensional reduction from higher dimensional relativity and by assuming that the Compton wavelength of a particle cannot take any value, Wesson [@We04] proposed that the mass is quantised according to the rule $m=(n\hbar /c)\sqrt{\Lambda /3}$. Hence $m_{P}$ is the minimum mass corresponding to the ground state $n=1$. With the use of the generalized Buchdahl identity [@MaDoHa00], it can be rigorously proven that the existence of a non-negative $\Lambda $ imposes a lower bound on the mass $M$ and density $\rho $ for general relativistic objects with radius $R$, which is given by [@BoHa05] $$2GM\geq \frac{\Lambda c^{2}}{6}R^{3},\qquad \rho =\frac{3M}{4\pi R^{3}}\geq \frac{\Lambda c^{2}}{16\pi G}=:\rho _{\min }. \label{minm}$$ Therefore, the existence of the cosmological constant implies the existence of an absolute minimum density in the universe. No object present in relativity can have a density that is smaller than $\rho_{\min}$. For $\Lambda >0$, this result also implies a minimum density for stable fluctuations in energy density. These results have been generalized to compact anisotropic general relativistic objects in [@Bo1], where it was shown that in the presence of the cosmological constant, a minimum mass configuration with given anisotropy does exist. For charged general relativistic objects there is also a lower bound for the mass-radius ratio [@Bo2]. By considering the total energy (including the gravitational one) and the stability of the objects with minimum mass-radius ratio, a representation of the mass and radius of the charged objects with minimum mass-radius ratio in terms of the charge and vacuum energy only has been obtained. It is the purpose of the present paper to further explore the possible physical implications of the existence of a minimum mass in the universe, given by Eq. (\[minm\]), and which is a direct consequence of the existence of a non-zero cosmological constant. In particular, we show that if a minimum length does exist in nature, then the condition (\[minm\]) does imply the existence of an absolute minimum mass. By combining the rigorous result for the minimum mass with the dimensional analysis of Wesson [@We04], we can obtain an intriguing representation of the vacuum energy as a function of the fundamental constants $c$, $G$, $\hbar $ as well as the mass $m_{e}$ and the radius $r_{e}$ of the electron. On the other hand, by considering the possibility of the gravitational condensation of the dark energy fluid we obtain the interpretation of the mass $m_{PE}$ as the Jeans mass of the gravitational dark energy condensate. By minimizing the total (matter plus gravitational) energy of a stable configuration consisting of particles with the minimum mass we provide a rigorous derivation of the cosmological mass $m_{PE}$, given by Eq. (\[mpe\]). The present paper is organized as follows. The physical implications of the existence of a minimum mass are presented in Section 2. The gravitational condensation of the dark energy particles is considered in Section 3. The total energy (including the gravitational one) for stable configurations of particles with minimum mass is obtained in Section 4. We briefly conclude and discuss our results in the last section. Minimum mass and radius of dark energy particles ================================================ At a microscopic level two basic quantities, the Planck mass $m_{Pl}$ and the Planck length $l_{Pl}$ are supposed to play a fundamental physical role. The Planck mass is derived by equating the gravitational radius $2Gm/c^{2}$ of a Schwarzschild mass with its Compton wavelength $\hbar /mc$. The corresponding mass $m_{Pl}=\left( c\hbar /2G\right) ^{1/2}$ is of the order $% m_{Pl}\approx 1.5\times 10^{-5}$ g. The Planck length is given by $% l_{Pl}=\left( \hbar G/c^{3}\right) ^{1/2}\approx 1.6\times 10^{-33}$ cm and at about this scale quantum gravity will become important for understanding physics. The Planck mass and length are the only parameters with dimension mass and length, respectively, which can be obtained from the fundamental constants $c$, $G$ and $\hbar $. The problem of the physical nature of the cosmological constant/dark energy is one of the most important issues confronting modern physics. A popular interpretation of the cosmological constant is in terms of the vacuum energy $\langle\rho_{vac}\rangle$, which is of the order $\langle\rho _{vac}\rangle\approx 2\times 10^{71}$ GeV$^4$. However, astronomical observations indicate that the cosmological constant is many orders of magnitude (around $10^{120}$) smaller than the estimate for vacuum energy. Many different approaches to the solution of this problem have been proposed, like the interpretation of the cosmological constant as an integration constant, anthropic considerations, quantum cosmology etc. [@We89]. Presently, astronomical observations suggests that dark energy could be dynamical and evolving, with the dark-energy density approaching its natural value, zero. The smallness of the dark energy is a result of the expansion of the universe and its old age [@PeRa03]. Due to the fact that in a curved space-time the vacuum is not unique, the phenomenon of particle production occurs in an expanding universe as a typical quantum effect [@BiDa82]. As the universe evolves, and the curvature changes, the vacuum state also changes, and the initial zero particle vacuum state becomes later a multiparticle state. If the universe is decelerating and is asymptotically Minkowskian at infinity, in the large time limit the particle production does not occur any more. However, observations indicate that we are living in an accelerating universe, and the mechanism of particle production can be very important. The rate of the particle production for a flat universe filled with a fluid with equation of state $p=\alpha \rho$, where $\alpha $ is arbitrary, has been obtained recently in [@BaFaHo07]. The calculations were performed for the case of a massless scalar field, for which the corresponding Klein-Gordon equation was solved. The rate of particle production is determined exactly for any value of $\alpha $, including $\alpha =-1$. When the strong energy condition is satisfied, the rate of particle production decreases as time goes on, in agreement to the fact that the four-dimensional curvature decreases with the expansion; the opposite occurs when the strong energy condition is violated. Hence the “cosmological constant” (with $\rho c^2=-p$) can be an effective source of particles, during a purely de Sitter evolutionary phase [@BaFaHo07]. An alternative approach to particle production from a dark energy vacuum fluid was suggested in [@We05], by assuming that the vacuum, with the energy density proportional to $\Lambda $, gives up energy which corresponds to a particle with rest mass $m$, so that $d\Lambda =-6\left(mc/h\right)^2$. This situation is similar to the Dirac hole theory, in which a positron is regarded as a hole created in an underlying sea of energy. Particle production from dark energy can also be interpreted in geometrical terms. The vacuum energy/cosmological constant is a “sea of energy” that curves the space-time having a curvature $L=\sqrt{3/\Lambda }$. Locally, a perturbation in the vacuum corresponds to a change in curvature, and a change in the curvature leads to a change in the quantum mechanical vacuum state, resulting in a production of a massive particle. >From Eq. (\[minm\]) one can estimate the numerical value of the minimal density for a positive $\Lambda $ as $\rho _{\min }=\Lambda c^2/16\pi G=8.0\times 10^{-30}$ g cm$^{-3}$. Since the Planck length $l_{Pl}$ is a natural minimal length scale in physics, we define the absolute minimal mass which possibly can exist in nature by $$M_{\min}=\frac{\Lambda c^{2}}{16\pi G} \frac{4\pi}{3}l_{Pl}^{3} =\frac{% \Lambda c^{2}}{12G}l_{Pl}^{3} =\frac{\Lambda }{12}\sqrt{\frac{\hbar^{3}G}{% c^{5}}} =\frac{\Lambda }{6\sqrt{2}}m_{Pl}l_{Pl}^{2} =\frac{\Lambda}{3}\frac{% \hbar}{2}\frac{t_{Pl}}{2}, \label{minmass}$$ where we denoted by $t_{Pl}$ the Planck time $t_{Pl}=l_{Pl}/c$. The numerical value of $M_{\min}$ is given by $$M_{\min} \approx 1.4 \times 10^{-127}\,\mathrm{g} \approx 7.9 \times 10^{-95}\,\mathrm{eV}.$$ If an absolute minimum length does exist in nature, then, via the first of Eqs. (\[minm\]), a positive cosmological constant implies the existence of an absolute minimum mass in nature, given by Eq. (\[minmass\]). Hypothetical particles having this value of the mass may be called cosminos. The cosminos could also be interpreted as “quanta” of the dark energy (cosmological constant), and therefore $M_{\min}$ gives the mass of the quantum of the cosmological constant. Compared with the upper bound of the electron neutrino mass $m_{\nu_{e}} < 1.8\,\mathrm{eV}$ [@tr03], we emphasize the smallness of the minimal mass $M_{\min}$. By generalizing Eq. (\[minmass\]) we propose that the mass is quantized according to the general rule $$m = n\, \frac{\Lambda }{3}\frac{\hbar }{2} \frac{t_{Pl}}{2},\qquad n\in N,$$ which is different from Wesson’s proposal [@We04]. >From a purely quantum mechanical point of view, the value of the minimum mass can be derived with the use of the uncertainty principle for energy and time, which gives $$m_{\min }c^{2}\approx \frac{\hbar }{\Delta t}.$$ By assuming that $\Delta t$ is of the same order of magnitude as the age of the Universe, $\Delta t\approx 1/H_0$, where $H_0\approx 3.24\times 10^{-18}$ s$^{-1}$ is the Hubble constant (the present value of the Hubble function), we obtain for the minimum mass the expression $$m_{\min }=\frac{\hbar H_0}{c^{2}}\approx 3. 8017\times 10^{-66}\mathrm{g}.$$ The numerical value of the minimum mass obtained from quantum mechanical considerations agrees with the value of the mass $m_{P}=(\hbar/c) \sqrt{% \Lambda /3}$ obtained by Wesson [@We04] by using purely dimensional considerations. Therefore it is natural to assume that these two masses are the same, thus obtaining $$\left(\frac{\hbar}{c}\right) \sqrt{\frac{\Lambda }{3}}=\frac{\hbar H_0}{c^{2}% },$$ which gives $$\label{a} H_0=c\sqrt{\frac{\Lambda }{3}}.$$ We propose to call particles having the mass given by $m_{\min}=m_P$ cosmons[^3]. The possibility of the existence of a very light scalar particle, also named Cosmon, a dilaton which should essentially decouple from the Standard Matter Lagrangian, but still could mediate new macroscopic forces in the submillimeter range, was proposed in [@ShSo99]. The mass of the Cosmon is given by $m_S^2=\Lambda_{QCD}^4/M^4$, where $\Lambda_{QCD}\approx 100 \mathrm{MeV}$ is the intrinsic QCD scale and $M\geq 10^{10} \mathrm{GeV}$ is some high energy scale [@ShSo99]. The mass of this particle is of the order of the neutrino mass, $m_S\approx(10^{-3}-10^{-2})\,\mathrm{eV}\approx 2\times (10^{-30}-10^{-31})\,\mathrm{g}$. Therefore it represents a very different particle as compared to the minimum mass particles considered in the present paper. By assuming however, that the minimum mass in nature is given by $% m_{P}=m_{\min}$ it follows that the radius corresponding to $m_{P}$ is given by $$R_{P}=48^{1/6}\left( \frac{\hbar G}{c^{3}}\right)^{1/3} \Lambda^{-1/6}\approx 1.9\,l_{Pl}^{2/3}\,\Lambda ^{-1/6},$$ with the numerical value $R_P=4.7\times 10^{-13} \mathrm{cm} =4.7 \mathrm{fm} $. This would also imply that the minimum length in nature could be very different from the Planck length $l_{Pl}$. In fact the radius $R_P$ is of the same order of magnitude as the classical radius of the electron $r_e=e^2/m_ec^2=2.81\times 10^{-13}$ cm. Therefore, by formally equating $R_P$ with $r_e$ and neglecting terms of the order of unity gives a representation of the cosmological constant in terms of the ‘classical’ fundamental constants as $$\label{const} \Lambda =\frac{l_{Pl}^{4}}{r_{e}^{6}}= \frac{\hbar ^{2}G^{2}m_{e}^{6}c^{6}}{% e^{12}} \approx 1.4\times 10^{-56} \mathrm{cm}^{-2}.$$ Conceptually, the identification of the radius $R_P$ to the electron radius $% r_e$ may be based on a “small number hypothesis”, representing an extension of the large number hypothesis by Dirac [@Di74], and which proposes that the numerical equality between two very small quantities with a very similar physical meaning cannot be a simple coincidence. Gravitational condensation of dark energy particles =================================================== Recently a class of hypothetical compact objects called gravastars ([gra]{}vitational [va]{}cuum stars) have been proposed as potential alternatives to explain the astrophysical phenomenology traditionally associated to black holes [@MaMo04]. According to this scenario, the quantum vacuum undergoes a phase transition at or near the location the event horizon is expected to form. Hence the gravastar consists of an interior de Sitter condensate, governed by an equation of state $\rho c^2=-p$, matched to a shell of finite thickness with an equation of state $\rho c^2=p$. The latter is then matched to an exterior Schwarzschild solution. Dark energy stars, for which the interior vacuum energy is much larger than the cosmological energy, have also been investigated [@Ch05; @Lo06]. Hence the possibility that condensation processes, like, for example, Bose-Einstein condensation, could play an essential role in astrophysical and cosmological situations cannot be excluded [a priori]{}. Generally, Bose-Einstein condensation processes take place in a Bose gas consisting of particles with mass $m$ and number density $n$ when the thermal de Broglie wave length $\lambda _{dB}=\sqrt{2\pi \hbar ^2/mkT}$, where $k$ is Boltzmann’s constant and $T$ is the temperature, exceeds the mean inter-particle distance $n^{1/3}$, and the wave packets percolate in space. The critical condensation temperature is $T\leq2\pi\hbar ^2n^{2/3}/mk$ [@BoHa07]. If we assume an adiabatic cosmological expansion of the universe, the temperature dependence of the number density of the particle is $T\propto n^{2/3}$. Hence Bose-Einstein condensation occurs if the mass of the particle satisfies the condition $m<1.87$ eV [@Bos], a condition which is obviously satisfied by both cosminos and cosmons. Hence these particles may Bose-Einstein condense to form large scale astrophysical or cosmological structures. It is tempting to assume that the cosmons with mass $M_{\min }$ or $m_{\min}=m_P$ could condense gravitationally to form stellar type stable compact objects. To study the cosmologival implications of the condensation process we assume that the cosmon fluid, with an initial density $\rho _{0}=\Lambda c^{2}/8\pi G$ and pressure $p_{0}$, satisfying the equation of state $\rho _{0}c^{2}+p_{0}=0$, condenses into a non-relativistic, dissipationless fluid, which can be characterized by a density $\rho $, a pressure $p$, a velocity $\vec{v}$ and a gravitational acceleration $\vec{g}$. The dynamics of the system is described by the continuity equation, the hydrodynamical Euler equation and the Poisson equation, which can be written as $$\frac{\partial \rho }{\partial t}+\nabla \cdot \left( \rho \vec{v}\right) =0,\frac{% \partial \vec{v}}{\partial t}+\left( \vec{v}\cdot \nabla \right) \vec{v}=-% \frac{1}{\rho }\nabla p+\vec{g}, \label{hydr1}$$ $$\nabla \times \vec{g}=0,\nabla \cdot \vec{g}=-4\pi G\rho . \label{hydr2}$$ We take as the initial (unperturbed) state of the system the state characterized by the absence of the ”real” gravitational forces, $\vec{g}=% \vec{g}_{0}=0$, of the hydrodynamical flow, $\vec{v}=\vec{v}_{0}=0$, and by constant values of the density and pressure, $\rho =\rho _{0}$ and $p=p_{0}$, respectively, with $\rho _{0}c^{2}+p_{0}=0$. The condensation process leads to the appearance of the gravitational interaction in the system, as well as to small perturbations of the hydrodynamical quantities, so that $$\rho =\rho _{0}+\rho _{1},p=p_{0}+p_{1},\vec{v}=\vec{v}_{0}+\vec{v}_{1},\vec{% g}=\vec{g}_{0}+\vec{g}_{1},$$ so that $-1<<\rho _{1}/\rho _{0}<<1$ and $-1<<p_{1}/p_{0}<<1$, respectively. In the first order approximation Eqs. (\[hydr1\]) and (\[hydr2\]) take the form $$\frac{\partial \rho _{1}}{\partial t}+\rho _0\nabla \cdot \vec{v}_{1} =0,\frac{\partial \vec{v}_{1}}{\partial t}=-\frac{v_{s}^{2}}{\rho _{0}}% \nabla \rho _{1}+\vec{g}_{1}, \label{hydr3}$$ $$\nabla \times \vec{g}_{1}=0,\nabla \cdot \vec{g}_{1}=-4\pi G\rho _{1},$$ where we have introduced the adiabatic speed of sound $v_{s}$ in the condensed cosmon fluid, defined as $v_{s}=\sqrt{p_{1}/\rho _{1}}=\sqrt{% \partial p/\partial \rho }$. By taking the partial derivative with respect to the time of the continuity equation in Eqs. (\[hydr3\]), we obtain the propagation equation of the density perturbation in the cosmon fluid in the form $$\frac{\partial ^{2}\rho _{1}}{\partial t^{2}}=v_{s}^{2}\nabla ^{2}\rho _{1}+% \frac{\Lambda c^{2}}{2}\rho _{1}.$$ By looking for a solution of the form $\rho _{1}\propto \exp \left[ i\left( \vec{k}\cdot \vec{r}-\omega t\right) \right] $, we obtain the following dispersion relation for $\omega $ $$\omega ^{2}=v_{s}^{2}\vec{k}^{2}-\frac{\Lambda c^{2}}{2}.$$ >From the dispersion equation one can see that for $k<k_{J}$, where $$\label{kj} k_{J}=\sqrt{\frac{\Lambda c^{2}}{2v_{s}^{2}}},$$ is the Jeans wave number, the angular frequency $\omega $ becomes an imaginary quantity, which corresponds to an instability of the fluid-$\rho _{1}$can increase (or decrease) exponentially, leading to a gravitational condensation (or rarefaction). Therefore, for $k<k_{J}$, $% \omega =\pm v_{s}\sqrt{k^{2}-k_{J}^{2}}=i{\rm Im}\omega $, where ${\rm Im}% \omega =\pm v_{s}\sqrt{k_{J}^{2}-k^{2}}$, and consequently $\rho _{1}\propto \exp \left[ \pm \left\| {\rm Im}\omega \right\| t\right] $. When the mass of the condensate exceeds the mass of a sphere with radius $% 2\pi /k_{J}$, a gravitational instability occurs in the cosmon fluid, and the cloud of particles would collapse . The critical mass is the Jeans mass $% M_{J}=\left( 4\pi /3\right) \left( 2\pi /k_{J}\right) ^{3}\rho _{0}$, and for the cosmon fluid it is given by $$M_{J}=\frac{8\sqrt{2}}{3}\pi ^{3}\left( \frac{v_{s}}{c}\right) ^{3}\frac{% c^{2}}{G}\Lambda ^{-1/2}\approx 1.6\times 10^{30}\times \left( \Lambda \;{\rm cm}^{-2}\right) ^{-1/2}\left( \frac{v_{s}}{c}\right) ^{3}\;{\rm g}.$$ For $\Lambda =3\times 10^{-56}$cm$^{-2}$ we obtain $M_{J}=9. 24\times 10^{57}\left( v_{s}/c\right) ^{3}$ g. By taking into account the representation of the cosmological constant in terms of the ”classical constants” given by Eq. (\[const\]), we obtain for the critical Jeans mass of the cosmon fluid the expression $$M_{J}=\frac{8\sqrt{2}}{3}\pi ^{3}\left( \frac{v_{s}}{c}\right) ^{3}\frac{% e^{6}}{\hbar G^{2}m_{e}^{3}c}.$$ The effective radius $R_{J}$ of the stable cosmon configuration is given by $$R_{J}=2^{3/2}\pi \frac{v_{s}}{c}\Lambda ^{-1/2}\approx 2^{3/2}\pi \frac{v_{s}% }{c}\frac{e^{6}}{\hbar Gm_{e}^{3}c^{3}}.$$ The theoretical value of the maximum mass $M_{Ch}$ of the stable compact astrophysical type objects, like white dwarfs and neutron stars, was found by Chandrasekhar and Landau and is given by the Chandrasekhar limit, $$\label{chandra} M_{Ch}\approx \left[ \left( \frac{\hbar c}{G}\right) m_{B}^{-4/3}\right] ^{3/2},$$ where $m_{B}$ is the mass of the particles giving the main contribution to the mass (baryons in the case of the white dwarfs and neutron stars) [@ShTe83]. Thus, with the exception of some composition-dependent numerical factors, the maximum mass of a degenerate star depends only on fundamental physical constants. The Jeans mass for cosmons can also be written in the form of a Chandrasekhar limiting mass, if we assume that the cosmons have an effective mass $m_{eff}$ given by $$m_{eff}=\left( G\hbar ^{5}c^{5}\right) ^{1/4}\frac{m_{e}^{3/2}}{e^{3}},$$ so that $M_{J}=\left[ 8\sqrt{2}\pi ^{3}\left( v_{s}/c\right) ^{3}/3\right] % \left[ \left( \hbar G/c\right) m_{eff}^{-4/3}\right] ^{3/2}$. The value of the effective mass of the cosmon is $m_{eff}\approx 8\times 10^{-20}$ g. On the other hand, one can also assume that the cosminos or the cosmons could form stellar type objects with the limiting mass given by the Chandrasekhar limit, Eq. (\[chandra\]. The mass of such an hypothetical super-massive object formed from cosminos with mass $M_{\min}$ is of the order of $M_{Ch}^{(1)}=8\times 10^{237}$ g, which exceeds by around $180$ orders of magnitude the mass of our universe. Therefore, it follows that cosminos did not condense gravitationally, and hence the particles associated with dark energy fail to represent dark matter, which is in complete agreement with the present standard model of cosmology. On the other hand, in the case of cosmons for the Chandrasekhar limiting mass $M_{Ch}^{(2)}$ we obtain $% M_{Ch}^{(2)}=2.7782\times 10^{116}$ g, which also shows that degenerate cosmon stars, having masses much larger than the mass of the universe, are very unlikely to exist. Gravitational energy of stable cosmon configurations ==================================================== The total energy (including the gravitational field contribution) inside an equipotential surface $S$ of radius $R$ can be defined, according to [@LyKa85], to be $$E=E_{M}+E_{F}=\frac{c^{4}}{8\pi G}\xi _{s}\int_{S}[K]dS,$$ where $\xi ^{i}$ is a Killing vector field of time translation, $\xi _{s}$ its value at $S$ and $[K]$ is the jump across the shell of the trace of the extrinsic curvature of $S$, considered as embedded in the 2-space $t=\mathrm{% constant}$. $E_{M}=\int_{S}T_{i}^{k}\xi^{i}\sqrt{-g}dS_{k}$ and $E_{F}$ are the energy of the matter and of the gravitational field, respectively. This definition is manifestly coordinate invariant. For a static spherically symmetric system in a Schwarzschild-de Sitter space-time the total energy is $$E=\frac{c^{4}}{G}R\left[ 1-\left( 1-\frac{2GM}{c^{2}R}-\frac{\Lambda }{3}% R^{2}\right) ^{1/2}\right] \left( 1-\frac{2GM}{c^{2}R}-\frac{\Lambda }{3}% R^{2}\right) ^{1/2}.$$ For the minimum mass particle the total energy can be expressed in terms of the radius and cosmological constant only as $$\label{en} E=\frac{c^{4}}{G}R\left[ 1-\left( 1-\frac{\Lambda}{2}R^{2}\right)^{1/2}% \right] \left( 1-\frac{\Lambda}{2}R^{2}\right) ^{1/2}.$$ For a stable configuration, the energy should have a minimum, $\partial E/\partial R=0$, a condition which determines $R$ as $$\label{rad} R_{BC}=\frac{1}{3}\sqrt{11+\sqrt{13}}\Lambda ^{-1/2} \approx 1.3\times \Lambda^{-1/2}.$$ Therefore the mass of the stable configuration can be obtained as $$M_{BC}=\frac{1.15}{6}\frac{c^{2}}{G}\Lambda ^{-1/2}\approx0.2\frac{c^{2}}{G}\frac{r_e^3}{l_{Pl}^2}\approx 0.2m_{Pl}\left(\frac{r_e}{l_{Pl}}\right)^3,$$ which gives a mass of the order $M_{BC}\approx 8.2\times 10^{54}$ g, a value which is close to the mass $m_{PE}$, which follows from dimensional considerations, and is of the same order of magnitude as the total mass of the observable universe. Therefore we may regard the observable universe as a dark energy dominated object with minimum density. For the second derivative of the energy, evaluated for $R=R_{BC}$, we obtain the expression $\left.( \partial ^{2}E/\partial R^{2})\right\|_{R=R_{BC}} =-6.89\sqrt{\Lambda }$, which shows that indeed the configuration is in a state of minimum total energy. Discussions and final remarks ============================= In the present paper we have investigated some of the possible consequences of the existence of a minimum mass and density for stable general relativistic objects, which is a direct result of the existence of the cosmological constant. The existence of a fundamental length, assumed to be the Planck scale, leads to an absolute minimum mass in nature, which could be the mass of the quanta of the dark energy (the cosminos), with radius of the order of the Planck length. However, the application of the quantum uncertainty principle for the energy shows that the mass of the elementary particles associated to the dark energy (the cosmons) is given by $m_{P}=\hbar H_0/c^2=\left( \hbar /c\right) \sqrt{\Lambda /3}$. If this is indeed the case, then the radius of such a particle is of the same order of magnitude as the classical electron radius. This leads to the intriguing possibility of the electron charge and radius, or, more generally, of the electromagnetic processes, as playing an essential role in the dark energy related phenomena. We also propose that “dark energy particles” may condensate, either Bose-Einstein or gravitationally, to form compact super-massive objects, formed of cosminos or cosmons, respectively. The mass of this condensation, which is gravitationally stable, was derived using two independent methods. Firstly, we have assumed that the dark energy fluid condenses into a dissipationless, non-relativistic fluid. The corresponding Jeans mass is proportional to $\Lambda ^{-1/2}$, and (except some numerical factors) is the same as the mass $m_{PE}$ introduced from dimensional considerations. Its numerical value is of the same order of magnitude as the total mass of the universe. The requirement that the total energy of the stable configuration formed from the particle satisfying the relation $2GM= \Lambda / 6R^3$ is a minimum leads to a second, rigorous derivation of the mass $m_{PE}$, which is of the same order of magnitude as the mass of the universe. This also shows that the only energetically stable dark energy dominated general relativistic objects must have a mass of the same order of magnitude as our universe. Therefore the general relativistic condition Eq. (\[minm\]) as combined with the thermodynamic condition of energetic stability may explain the actual value of the mass of the universe. Moreover, the total mass of the universe can also be obtained in terms of the elementary constants ${c,\hbar,e,m_e,G}$. Therefore, these two independent results may imply that our universe was born as the result of the dark energy condensation, which took place at a very high temperature and density. Hence the initial constituents of our universe may have been the cosmons. We also obtain the physical interpretation of the masses $m_{PE}$ and $M_{BC}$ as the critical Jeans mass of the Universe, that is, the mass of the gravitationally stable dark particles clouds. This result also gives a new physical interpretation of the cosmological constant. From Eq. (\[kj\]), by assuming that the speed of sound in the gravitationally condensed dark energy fluis is equal to the speed of light, $v_s=c$, it follows that $\Lambda \approx k_J^2$, that is, physically the cosmological constant represents the square of the Jeans wave number of a dark energy fluid. Alternatively, one can express the cosmological constant as $\Lambda =8\pi^2/\lambda _J^2$, where $\lambda _J=2\pi /k_J$ is the Jeans wavelength. Moreover, the mass of the universe can be expressed in terms of three fundamental quantities, the Planck mass $m_{Pl}$, the Planck length $l_{Pl}$, and the classical electron radius $r_e$, respectively. On the other hand, even that the estimation of the limiting Chandrasekhar masses [@ShTe83] for cosmons suggests the possible existence of super-massive stable degenerate dark energy objects, the existence of such stars with masses much larger than the mass of the universe is impossible in the universe we are living in. Finally, it would be very interesting to recall the cosmological constant problem again here. If it is interpreted as a measure of the vacuum energy density and from a particles physics point of view, the cosmological constant $\Lambda$ is 120 orders of magnitude too small than expected [@PeRa03]. Let us therefore assume that the cosmological constant were indeed 120 orders of magnitude larger. This would have drastic consequences for the minimal mass and we would find $M_{\min} \approx 10^{19}\,\mathrm{eV}$, in which case the minimal mass would exceed the masses of all elementary particles. From this point of view, we would like to also argue that because of the resulting problems, the interpretation of the cosmological constant as the vacuum energy density may raise some conceptual contradictions with the results derived in the present paper. Acknowledgements {#acknowledgements .unnumbered} ---------------- We would like to thanks to the two anonymous referees, whose comments helped us to significantly improve an earlier version of the manuscript. The work of TH was supported by the RGC grant No. 7027/06P of the government of the Hong Kong SAR. A. G. Riess et al., Astron. J. [116]{}, 109 (1998) S. Perlmutter et al., Astrophys. J. [517]{}, 565 (1999) P. de Bernardis et al., Nature [404]{}, 995 (2000) S. Hanany et al., Astrophys. J. [545]{}, L5 (2000) P. J. E. Peebles and B. Ratra, Rev. Mod. Phys. [75]{}, 559 (2003) T. Padmanabhan, Phys. Repts. [380]{}, 235 (2003) M. K. Mak, P. N. Dobson, Jr. and T. Harko, Mod. Phys. Lett. [A 15]{}, 2153 (2000) C. G. Böhmer, Gen. Rel. Grav. [36]{}, 1039 (2004) C. G. Böhmer, Ukr. J. Phys. [50]{}, 1219 (2005) A. Balaguera-Antolinez, C. G. Böhmer and M. Nowakowski, [Int. J. Mod. Phys.]{} D14, 1507 (2005) H. A. Buchdahl, Phys. Rev. [116]{}, 1027 (1959) A. Balaguera-Antolinez, C. G. Böhmer and M. Nowakowski, Class. Quant. Grav. [23]{}, 485 (2006) P. S. Wesson, Mod. Phys. Lett. [A 19]{}, 1995 (2004) R. Aldrovandi, J. P. Beltran Almeida and J. G. Pereira, arXiv:gr-qc/0702065 (2007) R. Aldrovandi, J. P. Beltran Almeida, C. S. O. Mayor and J. G. Pereira, arXiv:0709.3947 (2007) R. Aldrovandi, J. P. Beltran Almeida, C. S. O. Mayor and J. G. Pereira, arXiv:0710.0610 (2007) R. Aldrovandi, J. P. Beltran Almeida and J. G. Pereira, Class. Quant. Grav. [24]{}, 1385 (2007) C. G. Böhmer and T. Harko, Phys. Lett. [B 630]{}, 73 (2005) C. G. Böhmer and T. Harko, Class. Quant. Grav. [23]{}, 6479 (2006) C. G. Böhmer and T. Harko, Gen. Rel. Grav. [39]{}, 757 (2007) S. Weinberg, Rev. Mod. Phys. [61]{}, 1 (1989) N. D. Birrel and P. C. W. Davies, Quantum fields in curved space, Cambridge University Press, Cambridge (1982) A. B. Batista, J. C. Fabris and S. Houndjo, arXiv:0710.0999 (2007) P. S. Wesson, Found. Phys. Lett. [19]{}, 285 (2006) M. Trinczek et al., Phys. Rev. Lett. [90]{}, 012501 (2003) R. D. Peccei, J. Sola and C. Wetterich, Phys. Lett. [B 195]{}, 183 (1987) C. Wetterich, Nucl. Phys. [B 302]{}, 668 (1988) J. Sola, Phys. Lett. [B 228]{}, 317 (1989) J. Sola, Int. J. Mod. Phys. [A 5]{}, 4225 (1990) I. L. Shapiro and J. Sola, Phys. Lett. [B 475]{}, 236 (2000) P. A. M. Dirac, Proc. R. Soc. [A 333]{}, 439 (1974) P. O. Mazur and E. Mottola, Proc. Nat. Acad. Sci. [101]{}, 9545 (2004) G. Chapline, arXiv:astro-ph/0503200 (2005) F. S. N. Lobo, Class. Quant. Grav. [23]{}, 1525 (2006) C. G. Böhmer and T. Harko, JCAP [06]{} 025 (2007) T. Fukuyama, M. Morikawa and T. Tatekawa, arXiv:0705.3091 (2007) S. L. Shapiro and S. A. Teukolsky, Black Holes, White Dwarfs, and Neutron Stars, John Wiley & Sons, New York (1983) J. Katz, D. Lynden-Bell and W. Israel, Class. Quantum Grav. [5]{}, 971 (1988) [^1]: E-mail: [email protected] [^2]: E-mail: [email protected] [^3]: Cosmons were originally introduced by Peccei, Sola and Wetterich [@Pe] to name scalar fields that could dynamically adjust the cosmological constant to zero, see also [@So; @We; @So2].	{ "pile_set_name": "ArXiv" }
--- abstract: \| The phenomenon of quantum nucleation is studied in a ferromagnet in the presence of a magnetic field at an arbitrary angle. We consider the magnetocrystalline anisotropy with tetragonal symmetry and that with hexagonal symmetry, respectively. By applying the instanton method in the spin-coherent-state path-integral representation, we calculate the dependence of the rate of quantum nucleation and the crossover temperature on the orientation and strength of the field for a thin film and for a bulk solid. Our results show that the rate of quantum nucleation and the crossover temperature depend on the orientation of the external magnetic field distinctly, which provides a possible experimental test for quantum nucleation in nanometer-scale ferromagnets. [PACS number(s)]{}: 75.45.+j, 73.40.Gk, 75.30.Gw, 75.50.Gg address: 'Center for Advanced Study, Tsinghua University, Beijing 100084, P. R. China' author: - 'Yi Zhou, Rong Lü, Jia-Lin Zhu, and Lee Chang' title: Quantum nucleation in ferromagnets with tetragonal and hexagonal symmetries --- I. Introduction {#i.-introduction .unnumbered} =============== The tunneling of macroscopic object, known as Macroscopic Quantum Tunneling (MQT), is one of the most fascinating phenomena in condensed matter physics. In the last decade, the problem of quantum tunneling of magnetization in nanometer-scale magnets has attracted a great deal of theoretical and experimental interest.[@1] MQT in magnetic systems are interesting from a fundamental point of view as they can extend our understanding of the transition from quantum to classical behavior. On the other hand, these phenomena are important to the reliability of small magnetic units in memory devices and the designing of quantum computers in the future. And the measurement of magnetic MQT quantities such as the tunneling rates could provide independent information about microscopic parameters such as the magnetocrystalline anisotropies and the exchange constants. All this makes magnetic quantum tunneling an exciting area for theoretical research and a challenging experimental problem. The problem of quantum nucleation of a stable phase from a metastable one in ferromagnetic films is an interesting fundamental problem which allows direct comparison between theory and experiment.[@2] Consider a ferromagnetic film with its plane perpendicular to the easy axis determined by the magnetocrystalline anisotropy energy depending on the crystal symmetry. A magnetic field [H]{} is applied in a direction between perpendicular and opposite to the initial easy direction of the magnetization [M]{}, which favors the reversal of the magnetization. The reversal occurs via the nucleation of a critical bubble, which then the nucleus does not collapse, but grows unrestrictedly in volume. If the temperature is sufficiently high, the nucleation of a bubble is a thermal overbarrier process, and the rate of thermal nucleation follows the Arrhenius law $\Gamma _T\varpropto \exp \left( -U/k_BT\right) $, with $k_B$ being the Boltzmann constant and $U$ being the height of energy barrier. In the limit of $T\rightarrow 0$, the nucleation is purely quantum-mechanical and the rate goes as $\Gamma _Q\varpropto \exp \left( -{\cal S}_{cl}/\hbar \right) $, with ${\cal S}_{cl}$ being the classical action or the WKBexponent which is independent of temperature. Because of the exponential dependence of the thermal rate on $T$, the temperature $T_c$ characterizing the crossover from quantum to thermal regime can be estimated as $% k_BT_c=\hbar U/{\cal S}_{cl}$. The problem of quantum nucleation was studied by Privorotskii[@3] who estimated the exponent in the rate of quantum nucleation based on the dimensional analysis. Chudnovsky and Gunther[@4] studied the quantum nucleation of a thin ferromagnetic film in the presence of an external magnetic field along the opposite direction to the easy axis at zero temperature by applying the instanton method in the spin-coherent-state path-integral representation. Ferrera and Chudnovsky extended the quantum nucleation to a finite temperature.[@5] Kim studied the quantum nucleation in a thin ferromagnetic film placed in a magnetic field at an arbitrary angle.[@6] It is noted that the previous results[@4; @5; @6] of quantum nucleation were obtained for ferromagnetic sample with the simplest form of the magnetocrystalline anisotropy energy such as the biaxial symmetry, and the model considered in Refs. 4 and 5 was confined to the condition that the magnetic field be applied along the opposite direction to the easy axis. The purpose of this paper is to extend the previous results of quantum nucleation in ferromagnetic system with simple biaxial symmetry to that of system with a more general symmetry, such as tetragonal and hexagonal symmetry. The generic quantum nucleation problem, however, and the easiest to implement in practice, is that of ferromagnets with a general structure of magnetocrystalline anisotropy in a magnetic field applied at a some angle $\theta _H$ to the anisotropy axis. This problem does not possess any symmetry and for that reason is more difficult mathematically. It is worth pursuing, however, because of its significance for experiments.[@1] In this paper the magnetic field is applied in an arbitrary direction between perpendicular and opposite to the initial easy axis ($\widehat{z}$ axis). Our interest in studying quantum nucleation of magnetic bubbles with a more general structure of magnetocrystalline anisotropy in an arbitrarily directed magnetic field is stimulated by the fact that the corresponding experiment would be most easy to perform and to interpret. Within the instanton approach, we present the numerical results for the WKB exponent in quantum nucleation of a thin ferromagnetic film with the magnetic field applied in a range of angles $\pi /2<\theta _H<\pi $, where $\theta _H$ is the angle between the initial easy axis ($\widehat{z}$ axis) and the field. We also discuss the $\theta _H$ dependence of the crossover temperature $T_c$ from purely quantum nucleation to thermally assisted processes. Our results show that the distinct angular dependence, together with the dependence of the WKB exponent on the strength of the external magnetic field, may provide an independent experimental test for quantum nucleation in a ferromagnetic film. Quantum nucleation (the description involves space-time instantons), being a field theory problem, is more difficult than tunneling of magnetization in single-domain particles, both at the conceptual and at the technical level. Therefore, this paper provides [a nontrivial generalization of uniform rotation of magnetization vector (homogeneous spin tunneling) in single-domain magnets[@13; @15] to a nonuniform rotation of magnetization in bulk magnets with a more general structure of magnetocrystalline anisotropy in the presence of a magnetic field at an arbitrary angle]{}. Compared with the tunneling in single-domain particles, a local tunneling event in a bulk magnet can trigger instability on a much greater scale, which leads to really macroscopic consequences.[@1] In experiments, it may be easier to monitor single nucleation events in a thin film than to detect the magnetization reversal in a nanometer-scale particle. Therefore, our theoretical results for a general structure of magnetocrystalline anisotropy in an arbitrarily directed field will be more applicable for experimental tests of quantum nucleation. Besides the importance from the fundamental point of view, processes of quantum nucleation and collapse of magnetic bubbles are potentially important for quantum limitations on the density and long-term reliability of the data storage in magnetic memory devices and designing of quantum computer. This paper is structured in the following way. In Sec. II, we review briefly some basic ideas of quantum nucleation of magnetization in ferromagnets. In Secs. III and IV, we study quantum nucleation of magnetization in ferromagnets with tetragonal and hexagonal symmetry in an external magnetic field applied in the $ZX$ plane with a range of angles $\pi /2\leq \theta _H<\pi $. The conclusions and discussions are presented in Sec. V. II. The physical model {#ii.-the-physical-model .unnumbered} ====================== For a spin tunneling problem, the rate of magnetization reversal by quantum tunneling is determined by the imaginary-time transition amplitude from an initial state $\left\| i\right\rangle $ to a final state $\left\| f\right\rangle $ as $$U_{fi}=\left\langle f\right\| e^{-HT}\left\| i\right\rangle =\int {\cal D}% \left\{ {\bf M}\left( {\bf r},\tau \right) \right\} \exp \left( -{\cal S}% _E/\hbar \right) , \eqnum{1}$$ where ${\cal S}_E$ is the Euclidean action which includes the Euclidean Lagrangian density ${\cal L}_E$ as $${\cal S}_E=\int d\tau d^3{\bf r}{\cal L}_E. \eqnum{2}$$ For ferromagnets at sufficiently low temperature, all the spins are locked together by the strong exchange interaction, and therefore only the orientation of magnetization ${\bf M}\left( {\bf r},\tau \right) $ can change but not its absolute value. For that reason the field ${\bf M}\left( {\bf r},\tau \right) $ is equivalent to the fields $\theta \left( {\bf r}% ,\tau \right) $ and $\phi \left( {\bf r},\tau \right) $, which are spherical coordinates of ${\bf M}$. In this case the measure of the path integral $% {\cal D}\left\{ {\bf M}\left( {\bf r},\tau \right) \right\} $ in Eq. (1) is equivalent to $$\int {\cal D}\left\{ \theta \left( {\bf r},\tau \right) \right\} {\cal D}% \left\{ \phi \left( {\bf r},\tau \right) \right\} =\lim_{\varepsilon \rightarrow 0}\prod_{k=1}^N\left( \frac{2S+1}{4\pi }\right) \sin \theta _kd\theta _kd\phi _k, \eqnum{3}$$ where $\varepsilon =\max \left( \tau _{k+1}-\tau _k\right) $ and $% S=M_0/\hbar \gamma $ is the total spin of ferromagnet. Here $\gamma $ is the gyromagnetic ratio and $M_0$ is the magnitude of magnetization. In the spin-coherent-state representation, the magnetic Lagrangian is given by $${\cal L}_E=i\frac{M_0}\gamma \left( \frac{d\phi \left( {\bf r},\tau \right) }{d\tau }\right) \left[ 1-\cos \theta \left( {\bf r},\tau \right) \right] +E\left( \theta ,\phi \right) . \eqnum{4}$$ The first term in Eq. (4) is a total imaginary-time derivative, which has no effect on the classical equations of motion, but it is crucial for the spin-parity effects.[@1; @7; @8; @9; @10; @11] However, for the closed instanton trajectory described in this paper (as shown in the following), this time derivative gives a zero contribution to the path integral, and therefore can be omitted. The energy density in Eq. (4) is $$E\left( \theta ,\phi \right) =E_a\left( \theta ,\phi \right) +E_{ex}\left( \theta ,\phi \right) , \eqnum{5}$$ where $E_a$ includes the magnetocrystalline anisotropy energy and the energy due to the external magnetic field, and $E_{ex}$ is the exchange energy $$E_{ex}=\frac \alpha 2\left( \partial _iM_j\right) ^2=\frac \alpha 2% M_0^2\left[ \left( \nabla \theta \right) ^2+\sin ^2\theta \left( \nabla \phi \right) ^2\right] , \eqnum{6}$$ where $\alpha $ is the exchange stiffness..[@12] The magnetocrystalline anisotropy energy for tetragonal and hexagonal symmetry is shown in Sec. III and IV, respectively. In the semiclassical limit, the rate of quantum nucleation, with an exponential accuracy, is given by $$\Gamma _Q\varpropto \exp \left[ -{\cal S}_E^{\min }/\hbar \right] , \eqnum{7}$$ where ${\cal S}_E^{\min }$ is obtained along the trajectory that minimizes the Euclidean action ${\cal S}_E$. III. Ferromagnets with tetragonal symmetry {#iii.-ferromagnets-with-tetragonal-symmetry .unnumbered} ========================================== In this section, we study the quantum nucleation of magnetization in ferromagnets with tetragonal symmetry in the presence of a magnetic field at arbitrary angles in the $ZX$ plane, which has the following magnetocrystalline anisotropy energy $$E_a\left( \theta ,\phi \right) =K_1\sin ^2\theta +K_2\sin ^4\theta -K_2^{\prime }\sin ^4\theta \cos \left( 4\phi \right) -M_0H_x\sin \theta \cos \phi -M_0H_z\cos \theta , \eqnum{8}$$ where $K_1$, $K_2$ and $K_2^{\prime }$ are the magnetic anisotropy coefficients, and $K_1>0$. In the absence of the magnetic field, the easy axes of this system are $\pm \widehat{z}$ for $K_1>0$. And the field is applied in the $ZX$ plane at $\pi /2<\theta _H<\pi $. Then the total energy is given by $$\begin{aligned} E\left[ \theta \left( {\bf r},\tau \right) ,\phi \left( {\bf r},\tau \right) \right] &=&K_1\sin ^2\theta +K_2\sin ^4\theta -K_2^{\prime }\sin ^4\theta \cos \left( 4\phi \right) +\frac \alpha 2M_0^2\left[ \left( \nabla \theta \right) ^2+\sin ^2\theta \left( \nabla \phi \right) ^2\right] \nonumber \\ &&-M_0H_x\sin \theta \cos \phi -M_0H_z\cos \theta +E_0, \eqnum{9}\end{aligned}$$ where $E_0$ is a constant which makes $E\left( \theta ,\phi \right) $ zero at the initial state. By applying the similar method in Ref. 15, we can perform a Gaussian integration over the variable $\phi $ in the path integral and reduce the system to that with only one variable $\delta $ (as shown in the following). Then it is possible to perform the rest of the calculation by using the instanton method. This method simplifies the problem tremendously, compared to the problem where the action depended on $% \theta \left( \tau \right) $ and $\phi \left( \tau \right) $, though a complete mathematical equivalence to the initial problem is preserved. By introducing the dimensionless parameters as $$\overline{K}_2=K_2/2K_1,\overline{K}_2^{\prime }=K_2^{\prime }/2K_1,% \overline{H}_x=H_x/H_0,\overline{H}_z=H_z/H_0, \eqnum{10}$$ Eq. (9) can be rewritten as $$\begin{aligned} \overline{E}\left( \theta ,\phi \right) &=&\frac 12\sin ^2\theta +\overline{K% }_2\sin ^4\theta -\overline{K}_2^{\prime }\sin ^4\theta \cos \left( 4\phi \right) +\frac{\alpha M_0^2}{4K_1}\left[ \left( \nabla \theta \right) ^2+\sin ^2\theta \left( \nabla \phi \right) ^2\right] \nonumber \\ &&-\overline{H}_x\sin \theta \cos \phi -\overline{H}_z\cos \theta +\overline{% E}_0, \eqnum{11}\end{aligned}$$ where $E\left( \theta ,\phi \right) =2K_1\overline{E}\left( \theta ,\phi \right) $, and $H_0=2K_1/M_0$. At finite magnetic field, the plane given by $% \phi =0$ is the easy plane, on which $\overline{E}_a\left( \theta ,\phi \right) $ reduces to $$\overline{E}_a\left( \theta ,\phi =0\right) =\frac 12\sin ^2\theta +\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \sin ^4\theta -\overline{H}% \cos \left( \theta -\theta _H\right) +\overline{E}_0. \eqnum{12}$$ The initial angle $\theta _0$ is determined by $\left[ d\overline{E}_a\left( \theta ,0\right) /d\theta \right] _{\theta =\theta _0}=0$, and the critical angle $\theta _c$ and the dimensionless critical field $\overline{H}_c$ are determined by both $\left[ d\overline{E}_a\left( \theta ,0\right) /d\theta \right] _{\theta =\theta _c,\overline{H}=\overline{H}_c}=0$ and $\left[ d^2% \overline{E}_a\left( \theta ,0\right) /d\theta ^2\right] _{\theta =\theta _c,% \overline{H}=\overline{H}_c}=0$, which leads to $$\begin{aligned} \frac 12\sin \left( 2\theta _0\right) +\overline{H}\sin \left( \theta _0-\theta _H\right) +4\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \sin ^3\theta _0\cos \theta _0 &=&0, \eqnum{13a} \\ \frac 12\sin \left( 2\theta _c\right) +\overline{H}_c\sin \left( \theta _c-\theta _H\right) +4\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \sin ^3\theta _c\cos \theta _c &=&0, \eqnum{13b} \\ \cos \left( 2\theta _c\right) +\overline{H}_c\cos \left( \theta _c-\theta _H\right) +4\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \left( 3\sin ^2\theta _c\cos ^2\theta _c-\sin ^4\theta _c\right) &=&0. \eqnum{13c}\end{aligned}$$ Assuming that $\left\| \overline{K}_2-\overline{K}_2^{\prime }\right\| \ll 1$, we obtain the critical magnetic field and the critical angle as $$\begin{aligned} \overline{H}_c &=&\frac 1{\left[ \left( \sin \theta _H\right) ^{2/3}+\left\| \cos \theta _H\right\| ^{2/3}\right] ^{3/2}}\left[ 1+\frac{4\left( \overline{K% }_2-\overline{K}_2^{\prime }\right) }{1+\left\| \cot \theta _H\right\| ^{2/3}}% \right] , \eqnum{14a} \\ \sin \theta _c &=&\frac 1{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^{1/2}}\left[ 1+\frac 83\left( \overline{K}_2-\overline{K}% _2^{\prime }\right) \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right] . \eqnum{14b}\end{aligned}$$ In the low barrier limit, i.e., $\epsilon =1-\overline{H}/\overline{H}% _c\rightarrow 0$, by using Eqs. (13b) and (13c) we obtain the approximate equation for $\eta \left( \equiv \theta _c-\theta _0\right) $ in the order of $\epsilon ^{3/2}$, $$\begin{aligned} &&-\epsilon \overline{H}_c\sin \left( \theta _c-\theta _H\right) +\eta ^2\left[ \frac 32\overline{H}_c\sin \left( \theta _c-\theta _H\right) +3\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \sin \left( 4\theta _c\right) \right] \nonumber \\ &&+\eta \left\{ \epsilon \overline{H}_c\cos \left( \theta _c-\theta _H\right) -\eta ^2\left[ \frac 12\overline{H}_c\cos \left( \theta _c-\theta _H\right) +4\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \cos \left( 4\theta _c\right) \right] \right\} =0. \eqnum{15}\end{aligned}$$ Introducing $\delta \equiv \theta -\theta _0$ ($\left\| \delta \right\| \ll 1$ in the small $\epsilon $ limit), we derive the energy $\overline{E}\left( \theta ,\phi \right) $ as $$\begin{aligned} \overline{E}\left( \delta ,\phi \right) &=&\overline{K}_2^{\prime }\left[ 1-\cos \left( 4\phi \right) \right] \sin ^4\left( \theta _0+\delta \right) +% \overline{H}_x\left( 1-\cos \phi \right) \sin \left( \theta _0+\delta \right) \nonumber \\ &&+\frac{\alpha M_0^2}{4K_1}\left[ \left( \nabla \theta \right) ^2+\sin ^2\theta \left( \nabla \phi \right) ^2\right] +\overline{E}_1\left( \delta \right) , \eqnum{16}\end{aligned}$$ where $\overline{E}_1\left( \delta \right) $ is a function of only $\delta $ given by $$\begin{aligned} \overline{E}_1\left( \delta \right) &=&\left[ \frac 12\overline{H}_c\sin \left( \theta _c-\theta _H\right) +\left( \overline{K}_2-\overline{K}% _2^{\prime }\right) \sin \left( 4\theta _c\right) \right] \left( \delta ^3-3\delta ^2\eta \right) \nonumber \\ &&+\left[ \frac 18\overline{H}_c\cos \left( \theta _c-\theta _H\right) +\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \cos \left( 4\theta _c\right) \right] \left( \delta ^4-4\delta ^3\eta +6\delta ^2\eta ^2-4\delta ^2\epsilon \right) \nonumber \\ &&+4\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \epsilon \delta ^2\cos \left( 4\theta _c\right) . \eqnum{17}\end{aligned}$$ It can be shown that in the region of $\pi /2<\theta _H<\pi $, $0<\theta _c<\pi /2$, $\eta $ and $\delta $ are of the order of $\sqrt{\epsilon }$, the second or third term in Eq. (17) is smaller than the first term in the small $\epsilon $ limit. It is convenient to use dimensionless variables $${\bf r}^{\prime }=\epsilon ^{1/4}{\bf r}/r_0,\tau ^{\prime }=\epsilon ^{1/4}\omega _0\tau ,\overline{\delta }=\delta /\sqrt{\epsilon }, \eqnum{18}$$ where $r_0=\sqrt{\frac{\alpha M_0^2}{2K_1}}$, and $\omega _0=2\gamma K_1/M_0$. Then the Euclidean action Eq. (2) for $\pi /2<\theta _H<\pi $ becomes $$\begin{aligned} {\cal S}_E\left[ \overline{\delta }\left( {\bf r}^{\prime },\tau ^{\prime }\right) ,\phi \left( {\bf r}^{\prime },\tau ^{\prime }\right) \right] &=&% \frac{\hbar Sr_0^3}\epsilon \int d\tau ^{\prime }d^3{\bf r}^{\prime }\left\{ -i\epsilon ^{1/4}\sin \left( \theta _0+\sqrt{\epsilon }\overline{\delta }% \right) \phi \left( \frac{\partial \overline{\delta }}{\partial \tau ^{\prime }}\right) \right. \nonumber \\ &&+2\overline{K}_2^{\prime }\sin ^2\left( 2\phi \right) \sin ^4\left( \theta _0+\sqrt{\epsilon }\overline{\delta }\right) +2\overline{H}_x\sin ^2\left( \frac \phi 2\right) \sin \left( \theta _0+\sqrt{\epsilon }\overline{\delta }% \right) \nonumber \\ &&+\frac 12\epsilon ^{3/2}\left( \nabla ^{\prime }\overline{\delta }\right) ^2+\frac 12\epsilon ^{1/2}\sin ^2\left( \theta _0+\sqrt{\epsilon }\overline{% \delta }\right) \left( \nabla ^{\prime }\phi \right) ^2 \nonumber \\ &&\left. +\frac A4\epsilon ^{3/2}\left( \sqrt{6}\overline{\delta }^2-% \overline{\delta }^3\right) \right\} , \eqnum{19}\end{aligned}$$ where $$A=2\frac{\left\| \cot \theta _H\right\| ^{1/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\left[ 1+\frac 43\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \frac{7-4\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right] . \eqnum{20}$$ In Eq. (19) we have performed the integration by part for the first term and have neglected the total imaginary-time derivative. In can be showed that for $\pi /2<\theta _H<\pi $, only small values of $\phi $ contribute to the path integral, so that one can replace $\sin ^2\phi $ in Eq. (19) by $\phi ^2 $ and neglect the term including $\left( \nabla ^{\prime }\phi \right) ^2$ which is of the order $\epsilon ^2$ while the other terms are of the order $% \epsilon ^{3/2}$. Then the Gaussian integration over $\phi $ leads to $$\int {\cal D}\left\{ \delta \left( {\bf r}^{\prime },\tau ^{\prime }\right) \right\} \exp \left( -\frac 1\hbar {\cal S}_E^{eff}\right) , \eqnum{21}$$ where the effective action is $${\cal S}_E^{eff}\left[ \overline{\delta }\left( {\bf r}^{\prime },\tau ^{\prime }\right) \right] =\hbar S\epsilon ^{1/2}r_0^3\int d\tau ^{\prime }d^3{\bf r}^{\prime }\left[ \frac 12M\left( \frac{\partial \overline{\delta }% }{\partial \tau ^{\prime }}\right) ^2+\frac 12\left( \nabla ^{\prime }% \overline{\delta }\right) ^2+\frac A4\left( \sqrt{6}\overline{\delta }^2-% \overline{\delta }^3\right) \right] . \eqnum{22}$$ The effect mass in Eq. (22) is found to be $$M=\frac{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) \left[ 1+\frac 83% \left( \overline{K}_2-\overline{K}_2^{\prime }\right) \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right] }{% 1-\epsilon +16\overline{K}_2^{\prime }+4\left( \overline{K}_2-\overline{K}% _2^{\prime }\right) \frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}+128% \overline{K}_2^{\prime }\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}}. \eqnum{23}$$ Introducing the variables $\overline{\tau }=\tau ^{\prime }\sqrt{A/M}$ and $% \overline{{\bf r}}={\bf r}^{\prime }\sqrt{A}$, the effective action Eq. (22) is simplified as $${\cal S}_E^{eff}\left[ \overline{\delta }\left( \overline{{\bf r}},\overline{% \tau }\right) \right] =\hbar S\epsilon ^{1/2}r_0^3\frac{\sqrt{M}}A\int d% \overline{\tau }d^3\overline{{\bf r}}\left[ \frac 12\left( \frac{\partial \overline{\delta }}{\partial \overline{\tau }}\right) ^2+\frac 12\left( \overline{\nabla }\overline{\delta }\right) ^2+\frac 14\left( \sqrt{6}% \overline{\delta }^2-\overline{\delta }^3\right) \right] . \eqnum{24}$$ For the quantum reversal of magnetization ${\bf M}$ in a small particle of volume $V\ll r_0^3$, ${\bf M}$ is uniform within the particle and $\overline{% \delta }$ does not depend on the space $\overline{{\bf r}}$, Eq. (24) reduces to $${\cal S}_E^{eff}\left[ \overline{\delta }\left( \overline{{\bf r}},\overline{% \tau }\right) \right] =\hbar S\epsilon ^{5/4}\sqrt{MA}V\int d\overline{\tau }% \left[ \frac 12\left( \frac{d\overline{\delta }}{d\overline{\tau }}\right) ^2+\frac 14\left( \sqrt{6}\overline{\delta }^2-\overline{\delta }^3\right) \right] . \eqnum{25}$$ The corresponding classical trajectory satisfies the equation of motion $$\frac{d^2\overline{\delta }}{d\overline{\tau }^2}=\frac 12\sqrt{6}\overline{% \delta }-\frac 34\overline{\delta }^2. \eqnum{26}$$ Eq. (26) has the instanton solution $$\overline{\delta }\left( \overline{\tau }\right) =\frac{\sqrt{6}}{\cosh ^2\left( 3^{1/4}\times 2^{-5/4}\overline{\tau }\right) }, \eqnum{27}$$ corresponding to the variation of $\overline{\delta }$ from $\overline{% \delta }=0$ at $\overline{\tau }=-\infty $, to $\overline{\delta }=\sqrt{6}$ at $\overline{\tau }=0$, and then back to $\overline{\delta }=0$ at $% \overline{\tau }=\infty $. Eq. (27) agrees well with the result in Refs. 13 and 15. The associated classical action is found to be $$\begin{aligned} {\cal S}_{cl} &=&\frac{2^{17/4}\times 3^{1/4}}5\hbar S\epsilon ^{5/4} \nonumber \\ &&\times \frac{\left\| \cot \theta _H\right\| ^{1/6}\left[ 1+\frac 23\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \frac{7-2\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right] }{\sqrt{% 1-\epsilon +16\overline{K}_2^{\prime }+4\left( \overline{K}_2-\overline{K}% _2^{\prime }\right) \frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}+128% \overline{K}_2^{\prime }\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}}}. \eqnum{28}\end{aligned}$$ Now we turn to the nonuniform problem. In case of a thin film of thickness $% h $ less than the size $r_0/\epsilon ^{1/4}$ of the critical nucleus and its plane is perpendicular to the initial easy axis, we obtain the action Eq. (24) after performing the integration over the $\overline{z}$ variable, $${\cal S}_E^{eff}\left[ \overline{\delta }\left( \overline{{\bf r}},\overline{% \tau }\right) \right] =\hbar S\epsilon ^{3/4}r_0^2h\sqrt{\frac MA}\int d% \overline{\tau }d^2\overline{{\bf r}}\left[ \frac 12\left( \frac{\partial \overline{\delta }}{\partial \overline{\tau }}\right) ^2+\frac 12\left( \overline{\nabla }\overline{\delta }\right) ^2+\frac 14\left( \sqrt{6}% \overline{\delta }^2-\overline{\delta }^3\right) \right] . \eqnum{29}$$ At zero temperature the classical solution of the effective action Eq. (29) has $O\left( 3\right) $ symmetry in two spatial plus one imaginary time dimensions. Therefore, the solution $\overline{\delta }$ is a function of $u$, where $u=\left( \overline{\rho }^2+\overline{\tau }^2\right) ^{1/2}$, and $% \overline{\rho }=\left( \overline{x}^2+\overline{y}^2\right) ^{1/2}$ is the normalized distance from the ${\bf z}$ axis. Now the effective action Eq. (29) becomes $${\cal S}_E^{eff}\left[ \overline{\delta }\left( \overline{{\bf r}},\overline{% \tau }\right) \right] =4\pi \hbar S\epsilon ^{3/4}r_0^2h\sqrt{\frac MA}\int duu^2\left[ \frac 12\left( \frac{d\overline{\delta }}{du}\right) ^2+\frac 14% \left( \sqrt{6}\overline{\delta }^2-\overline{\delta }^3\right) \right] . \eqnum{30}$$ The corresponding classical trajectory satisfies the following equation of motion $$\frac{d^2\overline{\delta }}{du^2}+\frac 2u\frac{d\overline{\delta }}{du}=% \frac{\sqrt{6}}2\overline{\delta }-\frac 34\overline{\delta }^2. \eqnum{31}$$ By applying the similar method,[@4; @6] the instanton solution of Eq. (31) can be found numerically and is illustrated in Fig. 1. The maximal rotation of ${\bf M}$ is $\overline{\delta }_{\max }\approx 6.8499$ at $\overline{% \tau }=0$ and $\overline{\rho }=0$. Numerical integration in Eq. (30), using this solution, gives the rate of quantum nucleation for a thin ferromagnetic film as $$\begin{aligned} \Gamma _Q &\varpropto &\exp \left( -{\cal S}_E/\hbar \right) \nonumber \\ &=&\exp \left\{ -74.39S\epsilon ^{3/4}r_0^2h\frac{1+\left\| \cot \theta _H\right\| ^{2/3}}{\left\| \cot \theta _H\right\| ^{1/6}}\left[ 1-\frac 23% \left( \overline{K}_2-\overline{K}_2^{\prime }\right) \frac{7-6\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right] \right. \nonumber \\ &&\left. \times \frac 1{\sqrt{1-\epsilon +16\overline{K}_2^{\prime }+4\left( \overline{K}_2-\overline{K}_2^{\prime }\right) \frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}+128\overline{K}_2^{\prime }\left( \overline{K}_2-\overline{% K}_2^{\prime }\right) \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}}}\right\} . \eqnum{32}\end{aligned}$$ At high temperature, the nucleation of ${\bf M}$ is due to thermal activation, and the rate of nucleation follows $\Gamma _T\varpropto \exp \left( -W_{\min }/k_BT\right) $, where $W_{\min }$ is the minimal work necessary to produce a nucleus capable of growing. In this case the instanton solution becomes independent of the imaginary-time variable $% \overline{\tau }$. In order to obtain $W_{\min }$, we consider the effective potential of the system $$U_{eff}=\int d^3{\bf r}E=\int d^3{\bf r}\left[ \frac \alpha 2M_0^2\left( \left( \nabla \theta \right) ^2+\sin ^2\theta \left( \nabla \phi \right) ^2\right) +E_a\left( \theta ,\phi \right) \right] . \eqnum{33}$$ For a cylindrical bubble Eq. (33) reduces to $$U_{eff}=4\pi K_1h\epsilon r_0^2\int_0^\infty d\overline{\rho }\overline{\rho }\left[ \frac 12\left( \frac{d\overline{\delta }}{d\overline{\rho }}\right) ^2+\frac 14\left( \sqrt{6}\overline{\delta }^2-\overline{\delta }^3\right) \right] . \eqnum{34}$$ From the saddle point of the functional the shape of the critical nucleus satisfies $$\frac{d^2\overline{\delta }}{d\overline{\rho }^2}+\frac 1{\overline{\rho }}% \frac{d\overline{\delta }}{d\overline{\rho }}=\frac{\sqrt{6}}2\overline{% \delta }-\frac 34\overline{\delta }^2. \eqnum{35}$$ The solution can be found by numerical method similar to the one in Refs. 4 and 6. Fig. 2 shows the shape of the critical bubble in thermal nucleation, and the maximal size is $3.906$ at $\overline{\rho }=0$. Using this result, the minimal work corresponding the thermal nucleation is $$W_{\min }=41.3376K_1h\epsilon r_0^2. \eqnum{36}$$ Comparing this with Eq. (32), we obtain the approximate formula for the temperature characterizing the crossover from thermal to quantum nucleation as $$\begin{aligned} k_BT_c &\approx &0.55\frac{K_1\epsilon ^{1/4}}S\frac{\left\| \cot \theta _H\right\| ^{1/6}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\left[ 1+\frac 23% \left( \overline{K}_2-\overline{K}_2^{\prime }\right) \frac{7-6\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right] \nonumber \\ &&\times \left[ 1-\epsilon +16\overline{K}_2^{\prime }+4\left( \overline{K}% _2-\overline{K}_2^{\prime }\right) \frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}\right. \nonumber \\ &&\left. +128\overline{K}_2^{\prime }\left( \overline{K}_2-\overline{K}% _2^{\prime }\right) \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right] ^{1/2}. \eqnum{37}\end{aligned}$$ To observe the quantum nucleation one needs a large crossover temperature and not too small a nucleation rate. Eq. (37) shows that ferromagnets with large anisotropy, i.e., small ration of $K_2^{\prime }$ to $K_1$, and small saturated magnetization are preferable for experimental study. In Fig. 3, we plot the $\theta _H$ dependence of the crossover temperature $T_c$ for typical values of parameters for nanometer-scale ferromagnets: $K_1=10^7$ erg/cm$^3$, $\overline{K}_2^{\prime }=0.1$, $\overline{K}_2-\overline{K}% _2^{\prime }=0.01$, $M_0=500$ emu/cm$^3$, $\epsilon =0.01$ in a wide range of angles $\pi /2<\theta _H<\pi $. Fig. 3 shows that the maximal value of $% T_c$ is about 225 mK at $\theta _H=1.743$. The maximal value of $T_c$ as well as $\Gamma _Q$ is expected to be observed in experiment. The similar $% \theta _H$ dependence of the crossover temperature $T_c$ was first observed in Ref. 15, while the problem considered in Ref. 15 was homogeneous spin tunneling in single-domain particles with uniaxial symmetry. IV. Ferromagnets with hexagonal symmetry {#iv.-ferromagnets-with-hexagonal-symmetry .unnumbered} ======================================== In this section, we study the quantum nucleation of magnetization in nanometer-scale ferromagnets with hexagonal symmetry in an external magnetic field at an arbitrary angle in the $ZX$ plane. Now the magnetocrystalline anisotropy energy $E_a\left( \theta ,\phi \right) $ can be written as $$\begin{aligned} E_a\left( \theta ,\phi \right) &=&K_1\sin ^2\theta +K_2\sin ^4\theta +K_3\sin ^6\theta -K_3^{\prime }\sin ^6\theta \cos \left( 6\phi \right) \nonumber \\ &&-M_0H_x\sin \theta \cos \phi -M_0H_z\cos \theta , \eqnum{38}\end{aligned}$$ where $K_1$, $K_2$, $K_3$, and $K_3^{\prime }$ are the magnetic anisotropic coefficients. The easy axes are $\pm \widehat{z}$ for $K_1>0$. By choosing $% K_3^{\prime }>0$, we take $\phi =0$ to be the easy plane, at which the anisotropy energy can be written in terms of the dimensionless parameters as $$\overline{E}_a\left( \theta ,\phi =0\right) =\frac 12\sin ^2\theta +% \overline{K}_2\sin ^4\theta +\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \sin ^6\theta -\overline{H}\cos \left( \theta -\theta _H\right) +% \overline{E}_0, \eqnum{39}$$ where $\overline{K}_3=K_3/2K_1$ and $\overline{K}_3^{\prime }=K_3^{\prime }/2K_1$. Then the initial angle $\theta _0$ is determined by $\left[ d\overline{E}% _a\left( \theta ,0\right) /d\theta \right] _{\theta =\theta _0}=0$, and the critical angle $\theta _c$ and the dimensionless critical field $\overline{H}% _c$ by both $\left[ d\overline{E}_a\left( \theta ,0\right) /d\theta \right] _{\theta =\theta _c,\overline{H}=\overline{H}_c}=0$ and $\left[ d^2\overline{% E}_a\left( \theta ,0\right) /d\theta ^2\right] _{\theta =\theta _c,\overline{% H}=\overline{H}_c}=0$, which leads to $$\begin{aligned} &&\left. \frac 12\sin \left( 2\theta _0\right) +\overline{H}\sin \left( \theta _0-\theta _H\right) +4\overline{K}_2\sin ^3\theta _0\cos \theta _0+6\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \sin ^5\theta _0\cos \theta _0=0,\right. \eqnum{40a} \\ &&\left. \frac 12\sin \left( 2\theta _c\right) +\overline{H}_c\sin \left( \theta _c-\theta _H\right) +4\overline{K}_2\sin ^3\theta _c\cos \theta _c+6\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \sin ^5\theta _c\cos \theta _c=0,\right. \eqnum{40b} \\ &&\left. \cos \left( 2\theta _c\right) +\overline{H}_c\cos \left( \theta _c-\theta _H\right) +4\overline{K}_2\left( 3\sin ^2\theta _c\cos ^2\theta _c-\sin ^4\theta _c\right) \right. \nonumber \\ &&\left. +6\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \left( 5\sin ^4\theta _c\cos ^2\theta _c-\sin ^6\theta _c\right) =0,\right. \eqnum{40c}\end{aligned}$$ Under the assumption that $\left\| \overline{K}_2\right\| $, $\left\| \overline{% K}_3-\overline{K}_3^{\prime }\right\| \ll 1$, we obtain the dimensionless critical field $\overline{H}_c$ and the critical angle as $$\begin{aligned} \overline{H}_c &=&\frac 1{\left[ \left( \sin \theta _H\right) ^{2/3}+\left\| \cos \theta _H\right\| ^{2/3}\right] ^{3/2}}\left[ 1+\frac{4\overline{K}_2}{% 1+\left\| \cot \theta _H\right\| ^{2/3}}+\frac{6\left( \overline{K}_3-% \overline{K}_3^{\prime }\right) }{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] , \eqnum{41a} \\ \sin \theta _c &=&\frac 1{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^{1/2}}\left[ 1+\frac 83\overline{K}_2\frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right. \nonumber \\ &&\left. +8\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \frac{% \left\| \cot \theta _H\right\| ^{2/3}}{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] . \eqnum{41b}\end{aligned}$$ In the limit of small $\epsilon =1-\overline{H}/\overline{H}_c$, Eq. (40a) becomes $$\begin{aligned} &&-\epsilon \overline{H}_c\sin \left( \theta _c-\theta _H\right) +\eta ^2\left[ \left( 3/2\right) \overline{H}_c\sin \left( \theta _c-\theta _H\right) +3\overline{K}_2\sin \left( 4\theta _c\right) \right. \nonumber \\ &&\left. +12\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \sin ^3\theta _c\cos \theta _c\left( 5-8\sin ^2\theta _c\right) \right] +\eta \left\{ \epsilon \overline{H}_c\cos \left( \theta _c-\theta _H\right) \right. \nonumber \\ &&-\eta ^2\left. \left[ \left( 1/2\right) \overline{H}_c\cos \left( \theta _c-\theta _H\right) +4\overline{K}_2\cos \left( 4\theta _c\right) \right. \right. \nonumber \\ &&\left. \left. \left. +12\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \sin ^2\theta _c\left( 5-20\sin ^2\theta _c+16\sin ^4\theta _c\right) \right] \right\} =0,\right. \eqnum{42}\end{aligned}$$ where $\eta \equiv \theta _c-\theta _0$ which is small for $\epsilon \ll 1$. By introducing a small variable $\delta \equiv \theta -\theta _0$ $\left( \left\| \delta \right\| \ll 1\text{ in the limit of }\epsilon \ll 1\right) $, the anisotropy energy becomes $$\overline{E}_a\left( \delta ,\phi \right) =\overline{K}_3^{\prime }\left[ 1-\cos \left( 6\phi \right) \right] \sin ^6\left( \theta _0+\delta \right) +% \overline{H}_x\left( 1-\cos \phi \right) \sin \left( \theta _0+\delta \right) +\overline{E}_1\left( \delta \right) , \eqnum{43}$$ where $\overline{E}_1\left( \delta \right) $ is a function of only $\delta $ given by $$\begin{aligned} \overline{E}_1\left( \delta \right) &=&\left[ \frac 12\overline{H}_c\sin \left( \theta _c-\theta _H\right) +\overline{K}_2\sin \left( 4\theta _c\right) +4\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \left( 5\sin ^3\theta _c\cos ^3\theta _c-3\sin ^5\theta _c\cos \theta _c\right) \right] \nonumber \\ &&\times \left( \delta ^3-3\delta ^2\eta \right) +\left[ \frac 18\overline{H}% _c\cos \left( \theta _c-\theta _H\right) +\overline{K}_2\cos \left( 4\theta _c\right) +3\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \sin ^2\theta _c\left( \sin ^4\theta _c\right. \right. \nonumber \\ &&\left. \left. -10\sin ^2\theta _c\cos ^2\theta _c+5\cos ^4\theta _c\right) \right] \left( \delta ^4-4\delta ^3\eta +6\delta ^2\eta ^2-4\delta ^2\epsilon \right) +\epsilon \delta ^2\left[ 4\overline{K}_2\cos \left( 4\theta _c\right) \right. \nonumber \\ &&\left. +12\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \sin ^2\theta _c\left( \sin ^4\theta _c-10\sin ^2\theta _c\cos ^2\theta _c+5\cos ^4\theta _c\right) \right] . \eqnum{44}\end{aligned}$$ By applying the similar procedure in Sec. III, we obtain the transition amplitude Eqs. (21) and (22) by integrating out $\phi $. For this case the effective mass is $$\begin{aligned} M &=&\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) \left[ 1+\frac 83% \overline{K}_2\frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}+8\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \frac{\left\| \cot \theta _H\right\| ^{2/3}}{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] \nonumber \\ &&\times \left[ 1-\epsilon +36\overline{K}_3^{\prime }\frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}+4\overline{K}_2\left( 1+120\overline{K}_3^{\prime }% \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right) \frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}\right. \nonumber \\ &&\left. +6\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \left( 1+240% \overline{K}_3^{\prime }\frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right) \frac 1{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] ^{-1}, \eqnum{45}\end{aligned}$$ and the prefactor $A$ is $$A=2\frac{\left\| \cot \theta _H\right\| ^{1/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\left[ 1+\frac 43\overline{K}_2\frac{7-4\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}+2\left( \overline{K}_3-% \overline{K}_3^{\prime }\right) \frac{11-16\left\| \cot \theta _H\right\| ^{2/3}}{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] . \eqnum{46}$$ In case of a small ferromagnet of volume $V\ll r_0^3$, the result of quantum nucleation is $\Gamma _Q\varpropto \exp \left( -{\cal S}_{cl}/\hbar \right) $, where the classical action for hexagonal symmetry is found to be $$\begin{aligned} {\cal S}_{cl} &=&\frac{2^{17/4}\times 3^{1/4}}5\hbar S\epsilon ^{5/4}\left\| \cot \theta _H\right\| ^{1/6} \nonumber \\ &&\times \left[ 1+\frac 23\overline{K}_2\frac{7-2\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}+\left( \overline{K}% _3-\overline{K}_3^{\prime }\right) \frac{11-12\left\| \cot \theta _H\right\| ^{2/3}}{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] \nonumber \\ &&\times \left[ 1-\epsilon +36\overline{K}_3^{\prime }\frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}+4\overline{K}_2\left( 1+120\overline{K}_3^{\prime }% \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right) \frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}\right. \nonumber \\ &&\left. +6\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \left( 1+240% \overline{K}_3^{\prime }\frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right) \frac 1{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] ^{-1/2}, \eqnum{47}\end{aligned}$$ In case of a thin film of thickness $h$ less than the size $r_0/\epsilon ^{1/4}$ of the critical nucleus, we obtain the quantum nucleation as $\Gamma _Q\varpropto \exp \left( -{\cal S}_E/\hbar \right) $, with the classical action $$\begin{aligned} {\cal S}_E &=&74.39S\epsilon ^{3/4}r_0^2h\frac{1+\left\| \cot \theta _H\right\| ^{2/3}}{\left\| \cot \theta _H\right\| ^{1/6}} \nonumber \\ &&\times \left[ 1-\frac 23\overline{K}_2\frac{7-2\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}-\left( \overline{K}% _3-\overline{K}_3^{\prime }\right) \frac{11-12\left\| \cot \theta _H\right\| ^{2/3}}{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] \nonumber \\ &&\times \left[ 1-\epsilon +36\overline{K}_3^{\prime }\frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}+4\overline{K}_2\left( 1+120\overline{K}_3^{\prime }% \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right) \frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}\right. \nonumber \\ &&\left. +6\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \left( 1+240% \overline{K}_3^{\prime }\frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right) \frac 1{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] ^{-1/2}. \eqnum{48}\end{aligned}$$ And the crossover temperature for hexagonal symmetry is found to be $$\begin{aligned} k_BT_c &\approx &0.55\frac{K_1\epsilon ^{1/4}}S\frac{\left\| \cot \theta _H\right\| ^{1/6}}{1+\left\| \cot \theta _H\right\| ^{2/3}} \nonumber \\ &&\times \left[ 1+\frac 23\overline{K}_2\frac{7-2\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}+\left( \overline{K}% _3-\overline{K}_3^{\prime }\right) \frac{11-12\left\| \cot \theta _H\right\| ^{2/3}}{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] \nonumber \\ &&\times \left[ 1-\epsilon +36\overline{K}_3^{\prime }\frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}+4\overline{K}_2\left( 1+120\overline{K}_3^{\prime }% \frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right) \frac 1{1+\left\| \cot \theta _H\right\| ^{2/3}}\right. \nonumber \\ &&\left. +6\left( \overline{K}_3-\overline{K}_3^{\prime }\right) \left( 1+240% \overline{K}_3^{\prime }\frac{\left\| \cot \theta _H\right\| ^{2/3}}{1+\left\| \cot \theta _H\right\| ^{2/3}}\right) \frac 1{\left( 1+\left\| \cot \theta _H\right\| ^{2/3}\right) ^2}\right] ^{1/2}. \eqnum{49}\end{aligned}$$ V. Conclusions and discussions {#v.-conclusions-and-discussions .unnumbered} ============================== In summary we have investigated the quantum nucleation of magnetization in nanometer-scale ferromagnets in the presence of an external magnetic field at arbitrary angle. We consider the magnetocrystalline anisotropy with tetragonal symmetry and that with hexagonal symmetry, respectively. By applying the instanton method in the spin-coherent-state path-integral representation, we obtain the analytical formulas for quantum reversal of magnetization in small magnets and the numerical formulas for quantum nucleation in thin ferromagnetic film in a wide range of angles $\pi /2<\theta _H<\pi $. The temperature characterizing the crossover from the quantum to thermal nucleation is clearly shown for each case. Our results show that the rate of quantum nucleation and the crossover temperature depend on the orientation of the external magnetic field distinctly. Therefore, both the orientation and the strength of the external magnetic field are the controllable parameters for the experimental test of quantum nucleation of magnetization in nanometer-scale ferromagnets. If the experiment is to be performed, there are three control parameters for comparison with theory: the angle of the external magnetic field $\theta _H$, the strength of the field in terms of $\epsilon $, and the temperature $T$. Our results show that ferromagnetic samples with large anisotropy and small saturated magnetization are suitable for experimental study the phenomenon of quantum nucleation. Recently, Wernsdorfer and co-workers have performed the switching field measurements on individual ferrimagnetic and insulating BaFeCoTiO nanoparticles containing about $10^5$-$10^6$ spins at very low temperatures (0.1-6K).[@14] They found that above 0.4K, the magnetization reversal of these particles is unambiguously described by the Néel-Brown theory of thermal activated rotation of the particle’s moment over a well defined anisotropy energy barrier. Below 0.4K, strong deviations from this model are evidenced which are quantitatively in agreement with the predictions of the MQT theory without dissipation.[@13; @15] It is noted that the observation of quantum nucleation in ferromagnets would be extremely interesting as the next example, after single-domain nanoparticles, of macroscopic quantum tunneling. The theoretical results presented here may be useful for checking the general theory in a wide range of systems, with more general symmetries. The experimental procedures on single-domain ferromagnetic nanoparticles of Barium ferrite with uniaxial symmetry[@14] may be applied to the systems with more general symmetries. Note that the inverse of the WKB exponent $% B^{-1}$ is the magnetic viscosity $S$ at the quantum-tunneling-dominated regime $T\ll T_c$ studied by magnetic relaxation measurements.[@1] Therefore, the quantum nucleation of magnetization should be checked at any $% \theta _H$ by magnetic relaxation measurements. Over the past years a lot of experimental and theoretical works were performed on the spin tunneling in molecular Mn$_{12}$-Ac[@16] and Fe$_8$[@17] clusters having a collective spin state $S=10$ (in this paper $S=10^6$). Further experiments should focus on the level quantization of collective spin states of $S=10^2$-$10^4$. The ferromagnet is typically an insulating particle with as many as $% 10^3\sim 10^6$ magnetic moments. For the dynamical process, it is important to include the effect of the environment on quantum tunneling caused by phonons, [@18; @19] nucleation spins,[@20] and Stoner excitations and eddy currents in metallic magnets.[@21] However, many studies showed that even though these couplings might be crucial in macroscopic quantum coherence, they are not strong enough to make the quantum tunneling unobservable.[@1; @18; @19; @20; @21] The theoretical calculations performed in this paper can be extended to the AFM bubbles, where the relevant quantity is the excess spin due to the small noncompensation of two sublattices. Work along this line is still in progress. We hope that the theoretical results presented in this paper may stimulate more experiments whose aim is observing quantum nucleation in nanometer-scale ferromagnets. Acknowledgments {#acknowledgments .unnumbered} =============== Y. Zhou and R.L. would like to acknowledge Dr. Su-Peng Kou, Professor Zhan Xu, Professor Mo-Lin Ge, Professor Jiu-Qing Liang and Professor Fu-Cho Pu for stimulating discussions. R. L. would like to thank Professor W. Wernsdorfer and Professor R. Sessoli for providing their paper (Ref. 11), and Professor Kim for providing his paper (Ref. 13). For a review, see [Quantum Tunneling of Magnetization]{}, edited by L. Gunther and B. Barbara (Kluwer, Dordrecht, 1995); and E. M. Chudnovsky and J. Tejada, [Macroscopic Quantum Tunneling of the Magnetic Moment]{} (Cambridge University Press, 1997). A. P. Malozemoff and J. C. Slonczewski, [Magnetic Domain Walls in Bubble Materials]{} (Academic, New York, 1979). I. A. Privorotskii, Uspekhi Fiz. Nauk. [108]{}, 43 (1972) \[Sov. Phys. -Usp. [15]{}, 555 (1973)\]. E. M. Chudnovsky and L. Gunther, Phys. Rev. B [37]{}, 9455 (1988). A. Ferrera and E. M. Chudnovsky, Phys. Rev. B [53]{}, 354 (1996). G. -H. Kim, Phys. Rev. B [55]{}, 15053 (1998). D. Loss, D. P. DiVicenzo, and G. Grinstein, Phys. Rev. Lett. [69]{}, 3232 (1992). J. V. Delft and G. L. Henley, Phys. Rev. Lett. [69]{}, 3236 (1992). A. Garg, Europhys. Lett. [22]{}, 205 (1993). H. B. Braun and D. Loss, Europhys. Lett. [31]{}, 555 (1995). W. Wernsdorfer and R. Sessoli, Science [284]{}, 133 (1999). C. Kittle, Revs. Modern Phys. [21]{}, 552 (1949); E. P. Wohlfarth, Proc. Phys. Soc. (London) A [65]{}, 1053 (1952); C. Herring, Phys. Rev. [85]{}, 1003 (1952). G. -H. Kim and D. S. Hwang, Phys. Rev. B [55]{}, 8918 (1997); J. Appl. Phys., [84]{}, 391 (1998). W. Wernsdorfer, E. B. Orozco, K. Hasselbach, A. Benoit, D. Mailly, O. Kubo, H. Nakano, and B. Barbara, Phys. Rev. Lett. [79]{}, 4014 (1997); W. Wernsdorfer, E. B. Orozco, B. Barbara, A. Benoit, and D. Mailly, Physica B [280]{}, 264 (2000). M. -G. Miguel and E. M. Chudnovsky, Phys. Rev. B [54]{}, 388 (1996). R. Sessoli, D. Gatteschi, A. Caneschi, M. A. Novak, Nature [365]{}, 141 (1993); C. Paulsen and J. -G. Park, in [Quantum Tunneling of Magnetization]{}, edited by L. Gunther and B. Barbara (Kluwer, Dordrecht, 1995); M. A. Novak and R. Sessoli, in [Quantum Tunneling of Magnetization]{}, edited by L. Gunther and B. Barbara (Kluwer, Dordrecht, 1995); J. M. Hernández, X. X. Zhang, F. Luis, J. Bartolomé, J. Tejada, and R. Ziolo, Europhys. Lett. [35]{}, 301 (1996); L. Thomas, F. Lionti, R. Ballou, D. Gatteschi, R. Sessoli, and B. Barbara, Nature (London) [383]{}, 145 (1996); J. R. Friedman, M. P. Sarachik, J. Tejada, and R. Ziolo, Phys. Rev. Lett. [76]{}, 3820 (1996); J. M. Hernández, X. X. Zhang, F. Luis, J. Tejada, J. R. Friedman, M. P. Sarachik, and R. Ziolo, Phys. Rev. B [55]{}, 5858 (1997); F. Lionti, L. Thomas, R. Ballou, B. Barbara, A. Sulpice, R. Sessoli, and D. Gatteschi, J. Appl. Phys. [81]{}, 4608 (1997); D. A. Garanin and E. M. Chudnovsky, Phys. Rev. B [56]{}, 11102 (1997). A.-L. Barra, P. Debrunner, D. Gatteschi, C. E. Schulz, R. Sessoli, Europhys. Lett. [35]{}, 133 (1996); C. Sangregorio, T. Ohm, C. Paulsen, R. Sessoli, and D. Gatteschi, Phys. Rev. Lett. [78]{}, 4645 (1997); W. Wernsdorfer and R. Sessoli, Science [284]{}, 133 (1999). E. M. Chudnovsky and L. Gunther, Phys. Rev. Lett. [60]{}, 661 (1988). A. Garg and G. -H. Kim, Phys. Rev. Lett. [63]{}, 2512 (1989); Phys. Rev. B [43]{}, 712 (1991); H. Simanjuntak, J. Low temp. Phys. [90]{}, 405 (1992). A. Garg, Phys. Rev. Lett. [70]{}, 1541 (1993). G. Tatara and H. Fukuyama, Phys. Rev. Lett. [72]{}, 772 (1994). Figure Captions: Fig. 1 The instanton, corresponding to subbrrier bubble formation in a thin film by quantum tunneling, for $\overline{\tau }=0$, $\overline{\tau }=\pm 0.5$, $\overline{\tau }=\pm 1$, $\overline{\tau }=\pm 1.5$, and $\overline{% \tau }=\pm 2$. Fig. 2 The shape of the critical bubble in a thermal nucleation of magnetization. Fig. 3 The $\theta _H$ dependence of the crossover temperature $T_c$ for $% \pi /2<\theta _H<\pi $. Here, $K_1=10^7$ erg/cm$^3$, $\overline{K}_2^{\prime }=0.1$, $\overline{K}_2-\overline{K}_2^{\prime }=0.01$, $M_0=500$ emu/cm$^3$, and $\epsilon =0.01$.	{ "pile_set_name": "ArXiv" }
[Five-point Correlation Numbers in One-Matrix Model]{} G.Tarnopolsky[^1] L.D. Landau Institute for Theoretical Physics\ Chernogolovka, 142432, Russia\ Abstract Introduction ============ There exist two approaches to the 2d Quantum gravity. One of them is the continuous approach. In this approach the theory is determined by the functional integral over all metrics [@polyakov]. Calculation of this integral in conformal gauge leads to the Liouville field theory. Therefore this approach is called the Liouville gravity. The other way to describe sum over all surfaces is the discrete approach. It is based on the idea of approximation of two-dimensional geometry by the ensemble of planar graphs of big size. Technically the ensemble of graphs is usually defined by expansion into a series of perturbation theory of integral over matrixes. That is why this approach is called the Matrix models. References to the both approaches can be found in the review [@review]. The main objects in these two approaches are the correlation numbers $\langle O^{L}_{1}...O^{L}_{N} \rangle$ of the observables $O^{L}_{k}$ in the Liouville gravity and the correlation numbers $\langle O^{M}_{1}...O^{M}_{N} \rangle$ of the observables $O^{M}_{k}$ in the Matrix models. It was a remarkable discovery that the spectra of the gravitational dimensions [@review] in the both approaches coincide. It was reason to assume that both these theories describe the same variant of the 2d Quantum gravity and therefore the correlation numbers will coincide. However the attempt of naive identification of the correlation numbers doesn’t lead to the coincidence in general case. It was remarked in [@MSS], [@BZ2], that existence of the so-called resonances makes the identification of the correlation numbers ambiguous. In order to investigate this ambiguity it is convenient to pass from the correlation numbers to generating functions of correlators according to the formulas $$F^{L}(\lambda_{1},...,\lambda_{N}) = \langle \exp(\sum \lambda_{k}O_{k}^{L}) \rangle , \quad \mathcal{Z}(t_{1},...,t_{N}) = \langle \exp (\sum t_{k}O_{k}^{M}) \rangle.$$ In [@MSS], [@BZ2] a conjecture was proposed that there exists a “resonance” transformation $t_{k}=t_{k}(\{\lambda\})$ (from KdV frame to Liouville frame), such that the correlation functions in the both theories will coincide $$F^{L}(\lambda_{1},...,\lambda_{N}) = \mathcal{Z}(t_{1}(\{\lambda\}),...,t_{N}(\{\lambda\})).$$ The form of the transformation was conjectured in [@BZ2] for particular case, namely for the Minimal quantum gravity $\mathcal{MG}_{2/2p+1}$ and the One-matrix model with p critical points. In loc. cit. the conjectured of the coincidence was checked up to four-point correlator. In this paper we continue comparing of $\mathcal{MG}_{2/2p+1}$ and the One-matrix model. Using the “resonance” transformation [@BZ2] we find the five-point correlation numbers in One-matrix model in the Liouville frame. And we checked that the correlation numbers satisfy the fusion rules, which necessarily must be satisfied in the Minimal gravity $\mathcal{MG}_{2/2p+1}$. The article is organized in the following way. In the first part of the article there is a brief summary of the results of the paper [@BZ2]. In the second part, an expression for the five-point correlation numbers in the Liouville frame is found, and validity of the fusion rules for it is proved. One-matrix model ================ As it has been shown in the classical works on the Matrix models (see review [@review]) in the scaling limit near the $p$-critical point the partition function of the One-matrix model can be described in terms of the solution of the “string equation” $$\begin{aligned} \mathcal{P}(u)=0, \label{Str0}\end{aligned}$$ where $\mathcal{P}(u)$ is the polynomial of degree $p+1$ ($p$ is natural number) $$\begin{aligned} \mathcal{P}(u)=u^{p+1}+t_{0}u^{p-1}+\sum\limits_{k=1}^{p-1}t_{k}u^{p-k-1}, \label{Pol}\end{aligned}$$ with the parameters $t_{k}$ describing deviation from the $p$-critical point. The singular part of the partition function in the Matrix models $\mathcal{Z}(t_{0}, t_{1},...t_{p-1})$ can be described according to (\[Pol\]), as $$\begin{aligned} \mathcal{Z} = \frac{1}{2}\int_{0}^{u_{}}\mathcal{P}^{2}(u)du, \label{StS}\end{aligned}$$ where $u_{}=u_{}(t_{0},t_{1},...,t_{p-1})$ is the maximal real root [@BZ2] of the polynomial (\[Pol\]). Expression (\[StS\]) gives only the singular part of the partition function in the Matrix models, but complete the matrix integral includes also the regular part, which is analytical with respect to all parameters $t_{k}$ at the point $\{t_{1},...,t_{p-1}\}=0$. It is not worth considering it. If we calculate the correlation numbers by taking derivatives in the parameters $t_{k}$ (KdV frame), according to the formula $$\begin{aligned} \langle O^{M}_{k_{1}}O^{M}_{k_{2}}...O^{M}_{k_{N}} \rangle=\mathcal{Z}_{k_{1}...k_{N}}=\left.\frac{\partial^{N}\mathcal{Z}}{\partial t_{k_{1}}...\partial t_{k_{N}}}\right\|_{t_{1}=...=t_{p-1}=0}, \label{Corls}\end{aligned}$$ that coincidence with results of the Minimal gravity $\mathcal{MG}_{2/2p+1}$ will not take place. In the work [@BZ2] it was argued that among analytic and scale-invariant transformations $t_{k}=t_{k}(\{\lambda\})$, exists the special one, after which the correlation numbers satisfy the fusion rules, which necessarily must be satisfied in the Minimal gravity $\mathcal{MG}_{2/2p+1}$ . The fusion rules are expressed as follows $$\langle O_{k_{1}}O_{k_{2}}...O_{k_{N}} \rangle=0, \quad \textrm{if}\quad \begin{cases} k_{1}+...+k_{N-1}<k_{N}, \;\;\quad \textrm{when} \quad k_{1}+...+k_{N} \quad \textrm{is even}, \\ k_{1}+...+k_{N}<2p-1 , \quad \textrm{when} \quad k_{1}+...+k_{N} \quad \textrm{is odd}, \end{cases} \label{fusion}$$ Here it is assumed that $k_{i}$ run through the range $k_{i}=0,1,...,p-1$ and that $k_{N}$ is the maximal index, i.e. $k_{i} \leqslant k_{N}$. Later we will say that we are in even(odd) sector, if $k_{1}+...+k_{N}$ is even(odd). After this transformation the polynomial $\mathcal{P}(u)$, up to the factor $\frac{(p+1)!}{(2p-1)!!}u_{0}^{p+1}$ takes the form [@BZ2]: $$\begin{aligned} Q(x,\{s\}) =\sum\limits_{n=0}^{\infty}\sum\limits_{k_{1}...k_{n}=1}^{p-1} \frac{s_{k_{1}}...s_{k_{n}}}{n!}\frac{d^{n-1}} {dx^{n-1}}L_{p-\sum k_{i}-n}(x),\end{aligned}$$ here, we already pass from dimensional Liouville parameters $\{\lambda_{k}\}$ to dimentionless $\{s_{k}\}$, by the formula $$\begin{aligned} s_{k} = \frac{g_{k}}{g}\frac{u_{0}^{-k-2}}{2p+1}\lambda_{k}, \quad \textrm{where} \quad g_{k}= \frac{(p-k-1)!}{(2p-2k-3)!!}, \quad g=\frac{(p+1)!}{(2p+1)!!}.\end{aligned}$$ Also $x=u/u_{0}$, where $u_{0}$ is $u_{}$ at $s_{1},...,s_{p-1}=0$. The relation between $t_{0}$ and $u_{0}$ is $$\begin{aligned} t_{0}= -\frac{1}{2}\frac{p(p+1)}{2p-1}u_{0}^{2}.\end{aligned}$$ And $L_{n}(x)$ are the Legendre polynomials (see Appendix A). We assume that $\left(\frac{d}{dx}\right)^{-1}L_{p} = \int L_{p} dx =\frac{L_{p+1}-L_{p-1}}{2p+1}$. Below we use the notations $$\begin{aligned} Q_{k_{1}...k_{n}}(x)=\frac{d^{n-1}} {dx^{n-1}}L_{p-\sum k_{i}-n}(x), \quad Q_{0}(x)= \frac{L_{p+1}-L_{p-1}}{2p+1}.\end{aligned}$$ This “resonance” transformation is expressed as $$\begin{aligned} t_{k}&= \frac{g}{g_{k}}\frac{2p+1}{u_{0}^{-k-2}}\cdot\sum_{n=1}^{[\frac{2+k}{2}]} \sum_{\substack{ m_{1},...,m_{n}=0 \\m_{1}+...+m_{n}=\\=k+2-2n} }^{p-1}\frac{(2p-2k-5+2n)!!}{(2p-2k-3)!!}\cdot \frac{s_{m_{1}}...s_{m_{n}}}{n!},\end{aligned}$$ where we introduce auxiliary parameter $s_{0}=-\frac{1}{2}$. Thus the singular part of the partition function in the Matrix models expressed in terms of the new parameters can be written as $$\begin{aligned} \mathcal{Z} = \frac{1}{2}\int_{0}^{x_{}}Q^{2}(x)dx, \label{ZQ}\end{aligned}$$ where $x_{}=x_{}(s_{1},...,s_{p-1})$ is the maximal real root of the polynomial $Q(x)$, and $x_{}(0,0,...,0)=1$. The correlation numbers in the Liouville frame are defined as follows $$\begin{aligned} \langle O^{M}_{k_{1}}O^{M}_{k_{2}}...O^{M}_{k_{N}} \rangle=\mathcal{Z}_{k_{1}...k_{N}}=\left.\frac{\partial^{N}\mathcal{Z}}{\partial s_{k_{1}}...\partial s_{k_{N}}}\right\|_{s_{1}=...=s_{p-1}=0}. \label{Corls}\end{aligned}$$ These correlators will coincide with those of the Minimal gravity $\mathcal{MG}_{2/2p+1}$. In the next section we will give the answers derived from the formula (\[Corls\]) for the three- and four-point correlation numbers. Three- and four-point correlation numbers ----------------------------------------- In the next sections we will use convenient notations $$\begin{aligned} k = \sum\limits_{i=1}^{N}k_{i}, \quad \textrm{and}\quad k_{i_{1}...i_{n}}^{j_{1}...j_{m}} = (k_{i_{1}}+...+k_{i_{n}})-(k_{j_{1}}+...+k_{j_{m}}). \label{Oboz2}\end{aligned}$$ To simplify expressions we need a symbol of symmetrization, denoted by parentheses, for example $$\begin{aligned} Q_{(k_{1}k_{2}}Q_{k_{3})} =Q_{k_{1}k_{2}}Q_{k_{3}}+Q_{k_{1}k_{3}}Q_{k_{2}}+Q_{k_{2}k_{3}}Q_{k_{1}},\end{aligned}$$ notice, that the order of indexes in each $Q$-term doesn’t matter. The answer for the three-point correlation numbers can be written in the form $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}} = \frac{1}{2}\int_{-1}^{1} (Q_{(k_{1}k_{2}}Q_{{k_{3}})}+ Q_{0}Q_{k_{1}k_{2}k_{3}})dx-\\ -\frac{Q_{k_{1}}(1)Q_{k_{2}}(1)Q_{k_{3}}(1)}{Q'_{0}(1)}-\frac{Q_{0}(1) Q_{k_{1}k_{2}}(1)Q_{k_{3}}(1)}{Q'_{0}(1)}. \label{Zkkk}\end{gathered}$$ Taking into account the values of the Legendre polynomials (see Appendix A) and their derivatives in the point $x=1$, and also that $Q_{0}(x)$ is orthogonal to $Q_{k_{1}k_{2}k_{3}}(x)$, we get $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}} = -1+\frac{1}{2}\int_{-1}^{1} Q_{(k_{1}k_{2}}Q_{{k_{3}})} dx. \label{Zzkkk}\end{aligned}$$ Now we give the general final answer for $\mathcal{Z}_{k_{1}k_{2}k_{3}}$, assuming that $0 \leqslant k_{1} \leqslant k_{2} \leqslant k_{3} \leqslant p-1$ $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}} = \begin{cases} -1, \quad \textrm{if} \quad k_{3}\leqslant k_{12},\\ \;\;\, 0, \quad \textrm{if} \quad k_{3}> k_{12}, \end{cases}\end{aligned}$$ in the even sector, and $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}} = \begin{cases} -1, \quad \textrm{if} \quad k \geqslant 2p-1, \\ \textrm{reg.}, \;\,\; \textrm{if} \quad k < 2p-1, \end{cases}\end{aligned}$$ in the odd sector, where $"\textrm{reg.}"$ are the regular terms. Here we give the answer for four-point correlation numbers $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}} = \frac{1}{2}\int_{-1}^{1} \left(Q_{(k_{1}} Q_{k_{2}k_{3}k_{4})}+Q_{(k_{1}k_{2}} Q_{k_{3}k_{4})}+Q_{0}Q_{k_{1}k_{2}k_{3}k_{4}} \right)dx- \\ -\frac{Q_{(k_{1}k_{2}}Q_{k_{3}}Q_{k_{4})}}{Q'_{0}}+ \frac{Q'_{(k_{1}}Q_{k_{2}}Q_{k_{3}}Q_{k_{4})}}{(Q'_{0})^{2}}- \frac{Q''_{0}Q_{k_{1}}Q_{k_{2}}Q_{k_{3}} Q_{k_{4}}}{(Q'_{0})^{3}}, \label{Zkkkk1}\end{gathered}$$ where all $Q$-terms after the integral are taken in the point $x=1$. Now assuming, as usual, that $ 0 \leqslant k_{1} \leqslant k_{2} \leqslant k_{3} \leqslant k_{4} \leqslant p-1, $ in the even sector we have $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}} = \begin{cases} \frac{1}{2}(3p^{2}-5p-(2p-1)k +\sum k_{i}^{2} ), \quad\qquad \qquad\; p-1 < k_{12}\\ (1+k_{1})(2p-3-k)+F(k_{14})+F(k_{13}), \qquad k_{12}\leqslant p-1<k_{13},\\ (1+k_{1})(2p-3-k)+F(k_{14}),\qquad \qquad \qquad \, k_{13}\leqslant p-1<k_{14},k_{23},\\ (1+k_{1})(2p-3-k), \qquad\qquad \qquad \qquad\qquad\; k_{14}< k_{23}, \; k_{14} \leqslant p-1,\\ \frac{1}{2}(2+k_{123}-k_{4})(2p-3-k) ,\qquad\qquad\qquad\;\;\;\, k_{14}\geqslant k_{23}, \; k_{23} \leqslant p-1,\\ 0,\qquad\qquad\qquad\qquad \qquad\qquad \qquad\qquad\qquad\;\; k_{4}-k_{123}> 0,\\ \end{cases}\end{aligned}$$ where $F(k)=\frac{1}{2}(p-k-1)(p-k-2)$. And in the odd sector $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}} = \begin{cases} \frac{1}{2}(3p^{2}-5p-(2p-1)k +\sum k_{i}^{2} ), \qquad\qquad\quad\; p-1 < k_{12},\\ (1+k_{1})(2p-3-k)+F(k_{14})+F(k_{13}), \qquad k_{12}\leqslant p-1<k_{13}, \\ (1+k_{1})(2p-3-k)+F(k_{14}),\qquad\qquad\qquad\, k_{13}\leqslant p-1<k_{14},k_{23},\\ (1+k_{1})(2p-3-k), \qquad\qquad\qquad\qquad \qquad\; k_{14}< k_{23},\; k_{14} \leqslant p-1, \; k \geqslant 2p-3, \\ \frac{1}{2}(2+k_{123}-k_{4})(2p-3-k), \qquad\qquad\qquad\;\;\; k_{14}\geqslant k_{23}, \; k_{23} \leqslant p-1,\; k \geqslant 2p-3,\\ \textrm{reg. }. \qquad\qquad\qquad\qquad \qquad\qquad\qquad\qquad \quad k \leqslant 2p-5.\\ \end{cases}\end{aligned}$$ One can see, that near the “critical” regions, i.e. the regions in which relations between $k_{i}$ are already similar to those for which the fusion rules must be valid, the correlation numbers are factorized and become the simple form. In other regions the correlators has very bulky form. This interesting feature of the “simplification” of the correlator will be noted in the calculation of the five-point correlation numbers. Five-point correlation numbers =============================== We can at last begin to investigate the five-point correlation numbers in the Matrix models. The answer for it can be written as follows $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}} =& \frac{1}{2}\int_{-1}^{1} \left(Q_{(k_{1}} Q_{k_{2}k_{3}k_{4}k_{5})}+ Q_{(k_{1}k_{2}}Q_{k_{3}k_{4}k_{5})} +Q_{0}Q_{k_{1}k_{2}k_{3}k_{4}k_{5}}\right)dx- \frac{Q_{(k_{1}k_{2}k_{3}}Q_{k_{4}}Q_{k_{5})}} {Q_{0}'}-\notag\\ &-\frac{Q_{(k_{1}k_{2}}Q_{k_{3}k_{4}} Q_{k_{5})}}{Q_{0}'}+ \frac{Q'_{(k_{1}k_{2}}Q_{k_{3}}Q_{k_{4}} Q_{k_{5})}}{(Q_{0}')^{2}} +\frac{Q'_{(k_{1}}Q_{k_{2}k_{3}} Q_{k_{4}}Q_{k_{5})}}{(Q'_{0})^{2}}- \notag\\ &-\frac{Q''_{(k_{1}}Q_{k_{2}}Q_{k_{3}} Q_{k_{4}}Q_{k_{5})}} {(Q'_{0})^{2}}- \frac{2Q'_{(k_{1}}Q'_{k_{2}}Q_{k_{3}} Q_{k_{4}}Q_{k_{5})}} {(Q'_{0})^{3}}- Q''_{0}\frac{Q_{(k_{1}k_{2}}Q_{k_{3}}Q_{k_{4}} Q_{k_{5})}}{(Q'_{0})^{3}}+\notag\\ &+3Q''_{0} \frac{Q'_{(k_{1}}Q_{k_{2}}Q_{k_{3}}Q_{k_{4}} Q_{k_{5})}}{(Q'_{0})^{4}}+\left(Q'''_{0}-\frac{3(Q''_{0})^{2}}{Q'_{0}}\right) \frac{Q_{k_{1}}Q_{k_{2}} Q_{k_{3}}Q_{k_{4}}Q_{k_{5}}}{(Q'_{0})^{4}}, \label{Zkkkkk1}\end{aligned}$$ where all $Q$-terms after the integral are taken in the point $x=1$. After simplification (see Appendix C), one can get $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}} =& \frac{1}{2}\int_{-1}^{1} \left(Q_{(k_{1}}Q_{k_{2}k_{3}k_{4}k_{5})}+ Q_{(k_{1}k_{2}}Q_{k_{3}k_{4}k_{5})}\right)dx+ \left(Q'''_{0}-3(Q''_{0})^{2}\right)-\notag\\ &+\sum\limits_{i=1}^{5}(3Q''_{0}Q'_{k_{i}}-Q''_{k_{i}})+ \sum\limits_{i<j}(Q'_{k_{i}k_{j}}- 2Q'_{k_{i}}Q'_{k_{j}}-Q''_{0}Q_{k_{i}k_{j}})+ \notag\\ &+\sum\limits_{i,j,l}Q'_{k_{i}}Q_{k_{j}k_{l}}-\sum\limits_{i<j<l}Q_{k_{i}k_{j}k_{l}}- \sum\limits_{i,j,k,l}Q_{k_{i}k_{j}}Q_{k_{l}k_{m}}.\end{aligned}$$ For convenience let us divide the five-point correlator in several parts $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}= \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{I})}+ \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{J})}+ \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{0})},\end{aligned}$$ where the first two integral terms are $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{I})} =\frac{1}{2}\int_{-1}^{1} Q_{(k_{1}}Q_{k_{2}k_{3}k_{4}k_{5})}dx, \qquad \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{J})} =\frac{1}{2}\int_{-1}^{1} Q_{(k_{1}k_{2}}Q_{k_{3}k_{4}k_{5})}dx,\end{aligned}$$ and the last term is given by the formula $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{0})} = &\left(Q'''_{0}-3(Q''_{0})^{2}\right)+ \sum\limits_{i=1}^{5}(3Q''_{0}Q'_{k_{i}}-Q''_{k_{i}}) -\sum\limits_{i<j} 2Q'_{k_{i}}Q'_{k_{j}}+ \notag\\ &+\sum\limits_{i<j}(Q'_{k_{i}k_{j}}-Q''_{0}Q_{k_{i}k_{j}}) - \sum\limits_{i<j<l}Q_{k_{i}k_{j}k_{l}}+ \sum\limits_{i,j,l}Q'_{k_{i}}Q_{k_{j}k_{l}}- \sum\limits_{i,j,k,l}Q_{k_{i}k_{j}}Q_{k_{l}k_{m}}\end{aligned}$$ After the further simplification (see Appendix C), one can find $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{0})} = \\ =\frac{1}{8}(4\sum_{i=1}^{5}k_{i}^{2}-k^{2}-2k -8- \sum\limits_{m<n}k_{ijl}^{mn}(k_{ijl}^{mn}+2) \Theta(k_{mn}-p))(2p-3-k)(2p-5-k)+\\ +\sum\limits_{i<j<l}H(k_{ijl})G_{1}-\sum\limits_{i,j,l,m}F(k_{ij})F(k_{lm})G_{2}, \label{Corup}\end{gathered}$$ where $H(k)=\frac{1}{8}\prod\limits_{r=1}^{4}(p-r-k)$, and also two factors $G_{1}=\Theta(k_{ijl}-p)+ \Theta(k_{mn}-p)-1 \;$ and $G_{2}=\Theta(k_{ij}-p)\Theta(k_{lm}-p)$, and $\Theta(a-b)$ is the step-function, which is defined as $$\begin{aligned} \Theta(a-b)= \begin{cases} 1, \quad \textrm{if}\quad a>b, \\ 0, \quad \textrm{if}\quad a\leqslant b. \end{cases}\end{aligned}$$ Now we get down to consideration of the correlation numbers in the particular sectors. Let us begin with the odd sector. Odd Sector ---------- In this sector $k=k_{1}+k_{2}+k_{3}+k_{4}+k_{5}$ is odd, and $ \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{I})} = \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{J})}=0 $ due to oddness of the integrand. Thus one can get $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{odd})} = \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{0})}=\\ =\frac{1}{8}(4\sum_{i=1}^{5}k_{i}^{2}-k^{2}-2k -8- \sum\limits_{m<n}k_{ijl}^{mn} (k_{ijl}^{mn}+2)\Theta(k_{mn}-p))(2p-3-k)(2p-5-k)+\\ +\sum\limits_{i<j<l}H(k_{ijl})G_{1}-\sum\limits_{i,j,l,m}F(k_{ij})F(k_{lm})G_{2}. \label{zodd}\end{gathered}$$ If $k \leqslant 2p -7$, the terms of the partition function are regular [@BZ2]. Thus, due to the formula (\[fusion\]) we only need to check the fusion rules at the points $k=2p-5$ and $k=2p-3$. It is obvious, that if $k=2p-3$, and $k=2p-5$, the first term in the r.h.s. of the formula (\[zodd\]) equals zero. Also notice that $G_{2}=0$, because $G_{2} = 1$ at least if $k\geqslant 2p$. For $k=2p-3, 2p-5$, $G_{1}=0$ or $G_{1}=-1$. In case $G_{1}=-1$ we have the inequalities $k_{mn}< p$ and $k_{ijl}< p$, therefore $k_{ijl}$ can be equal only $p-4,\, p-3, \, p-2, \, p-1$, but for these values $H(k_{ijl}) =0$. Thus we showed, that the second term in the formula (\[zodd\]) also equals zero for $k=2p-3$ and $k=2p-5$. As the result we proved the validity of the fusion rules for $\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{odd})}$, i.e. $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{odd})}=0, \quad \textrm{when} \quad k<2p-1.\end{aligned}$$ Now let us pass to consideration of the even sector. Even Sector ----------- In the even sector $k=k_{1}+k_{2}+k_{3}+k_{4}+k_{5}$ is even. After calculation $\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{J})}$ (see formula (\[I2\])) and summation of it with $\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(0)}$ one can get $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{even})} = \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{I})}+\\ +\frac{1}{8}(4\sum_{i=1}^{5}k_{i}^{2}-k^{2}-2k -8- \sum\limits_{m<n}G_{3}k_{ijl}^{mn}(k_{ijl}^{mn}+2))(2p-3-k)(2p-5-k)+\\+ \left(\sum\limits_{i<j<l}H(k_{ijl})-\sum\limits_{i,j,l,m}F(k_{ij})F(k_{lm}) \right)G_{2},\end{gathered}$$ where $G_{3}=\Theta(k_{mn}-p)+\Theta(k_{mn}-k_{ijl}-2) \Theta(p-1-k_{ijl})\Theta(p-1-k_{mn})$ and $G_{2}=\Theta(k_{ij}-p)\Theta(k_{lm}-p)$. We as usual assume the following ordering $ 0 \leqslant k_{1} \leqslant k_{2} \leqslant k_{3}\leqslant k_{4}\leqslant k_{5} \leqslant p-1 \label{Usl5}. $ If $k_{5} > k_{1234}$ it is obvious, that $G_{2}=0$. Then by means of the simple reasoning one can prove, that $G_{3}=0$, when one of the indexes $i,j,l$ equals five. In remaining cases $G_{3}=1$. Having calculated $\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{I})}$ (see formula (\[I1\])), we derive the following formula $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{even})} = \frac{1}{8}(k_{5}^{1234}-2)(k_{5}^{1234}-4)(2p-3-k)(2p-5-k) (1-\Theta(k_{5}^{1234}-6)) \label{zeven}.\end{aligned}$$ From this formula we can see $\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{even})}=0$ if $k_{5}> k_{1234}$. Thus the fusion rules are valid. Notice that when $k_{5}=k_{1234}+2$ and $k_{5}=k_{1234}+4$ nulling of the function take place automatically, without the integral term $\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{I})}$. Conclusion ========== In this paper the five-point correlation numbers have been calculated. These numbers are necessary for the several reasons. First it is one more test of the fusion rules. Second, as we assume the correlation numbers in the two approaches are the same. Therefore the five-point correlation numbers which were found in this paper have to coincide with answers in continious approach. But the five-point correlation numbers in continious approach are still unknown. Aknowledgements =============== I am grateful to A.A. Belavin for posing the problem and attention to this work. I am also grateful to Y. Pugai, M.Lashkevich and M.Bershtein for useful discussion. This work was supported by RBRF-CNRS grant PICS-09-02-91064 and by RFBR initiative interdisciplinary project grant 09-02-12446-OFI-m and by the Russian Ministry of Science and Technology under Scientific Schools grant 3472.2008.2. The research was held within the bounds of Federal Program “Scientific and Scinetific-Pedagogical personnel of innovational Russia” on 2009-2013 , goskontrakt N P1339. Appendix {#appendix .unnumbered} ======== Legendre Polynomials ==================== The Legendre polynomials $L_{n}(x)$ are $n$-th order polynomials, which form an orthogonal system on the interval $[-1,1]$ with the weight $1$, $$\begin{aligned} \int_{-1}^{1}L_{n}(x)L_{m}(x) dx =\frac{2\delta_{m,n}}{2n+1}.\end{aligned}$$ The standard normalization is such that $$\begin{aligned} L_{n}(1)=1.\end{aligned}$$ Explicit formula for $L_{n}(x)$ is $$\begin{aligned} L_{n}(x) = \frac{2^{-n}}{n!}\frac{d^{n}}{dx^{n}}[x^{2}-1]^{n} = 2^{-n}\sum\limits_{l=0}^{[n/2]}(-1)^{l}\frac{(2n-2l)!}{l!(n-l)!(n-2l)!}x^{n-2l}.\end{aligned}$$ Expression of polynomial $L_{n}(x)$ in terms of Hypergeometric is series $$\begin{aligned} L_{n}(x)={}_{2}F_{1}\left(-n,n+1,1;\frac{1-x}{2}\right),\end{aligned}$$ from this we can obtain $$\begin{aligned} L'_{n}(1) = \frac{n(n+1)}{2}, \quad L''_{n}(1) = \frac{(n-1)n(n+1)(n+2)}{8}, \quad \textrm{etc}\end{aligned}$$ Yet another closed form is in terms of the contour integral $$\begin{aligned} L_{n}(x) = \oint_{0} \frac{(1-2xz +z^{2})^{-1/2}}{z^{n+1}}\frac{dz}{2\pi i}.\end{aligned}$$ The following relations are useful in our analysis: $$\begin{aligned} L'_{n+1}(x)-L'_{n-1}(x) = (2n+1)L_{n}(x),\end{aligned}$$ they are valid for all $n=0,1,2,3...$if one assume, that $L_{-1}(x)=0$. The following formulas are $$\begin{aligned} \frac{1}{2}\int_{-1}^{1}L'_{n}(x)L_{m}(x)dx = E_{n+m-1}\Theta_{n,m+1},\end{aligned}$$ $$\begin{aligned} \frac{1}{2}\int_{-1}^{1}L''_{n}(x)L_{m}(x)dx = E_{n+m}\Theta_{n,m+2}\frac{(n+m+1)(n-m)}{2}, \label{L''L}\end{aligned}$$ and in general $$\begin{aligned} \frac{1}{2}\int_{-1}^{1}L^{(l)}_{n}(x)L_{m}(x)dx= E_{n+m-l}\Theta_{n,m+l}\frac{2^{-l+1}}{(l-1)!}\prod\limits_{s=0}^{l-2}(n+m+l-1-2s) (n-m+l-2-2s), \label{LlL}\end{aligned}$$ where $L^{(l)}_{n}(x)$ stands for the $l$-th derivative. Here $\Theta_{n,m}=L_{n-m}(1)$ is the step function, and $$\begin{aligned} E_{n}= \begin{cases} 1,\quad \textrm{if}\quad n \textrm{\;\;is even}, \\ 0, \quad \textrm{if}\quad n \textrm{\;\;is odd}. \end{cases}\end{aligned}$$ Integrating (\[L”L\]) by parts, we have $$\begin{aligned} \frac{1}{2}\int_{-1}^{1}L'_{n}(x)L'_{m}(x)dx = E_{n+m}\left[\Theta_{m,n}\frac{n(n+1)}{2}+\Theta_{n,m}\frac{m(m+1)}{2}\right].\end{aligned}$$ The general formula which expresses $m$-th derivative of the Legendre polynomial of $n$-th order, by the sum of the Legendre polynomials is $$\begin{aligned} \frac{d^{m}}{dx^m}L_{n}(x)=\sum\limits_{k:2}^{n-m}(2k+1)B^{(m)} _{n,k}L_{k}(x), \label{Main}\end{aligned}$$ where $$\begin{aligned} B^{(m)} _{n,k}= \frac{2^{m-1}}{(m-1)!}\frac{\Gamma \left(\frac{n+k+m+1}{2}\right) \Gamma \left(\frac{n-k+m}{2}\right)}{\Gamma \left(\frac{n+k-m+3}{2}\right) \Gamma \left(\frac{n-k-m+2}{2}\right) }.\end{aligned}$$ notice, that (\[LlL\]) easily leads from this formula. Evaluation of $x_{k_{i}k_{j}}$ and $x_{k_{i}k_{j}k_{l}}$ ======================================================== From “string equation” $$\begin{aligned} Q(x,\{s\}) =0,\end{aligned}$$ where $$\begin{aligned} Q(x,\{s\})=Q_{0}(x)+\sum\limits_{k=1}^{p-1} s_{k}Q_{k}(x)+...+ \sum\limits_{k_{i}=1}^{p-1}\frac{s_{k_{1}}...s_{k_{n}}} {n!}Q_{k_{1}...k_{n}}(x)+...,\end{aligned}$$ after differentiation, we have $$\begin{aligned} \frac{\partial Q}{\partial x} dx + \sum\limits_{k=1}^{p-1}\frac{\partial Q} {\partial s_{k}}ds_{k}+...+ \sum\limits_{k_{i}=1}^{p-1}\frac{\partial Q} {\partial s_{k_{i}}}ds_{k_{i}}=0,\end{aligned}$$ and we can see that $$\begin{aligned} Q' \frac{\partial x}{\partial s _{k_{i}}} +\frac{\partial Q}{\partial s_{k_{i}}} =0, \quad \textrm{therefore} \quad x_{k_{i}} = - \frac{Q_{k_{i}}}{Q'}.\end{aligned}$$ In what follows we will use the formula $$\begin{aligned} \frac{\partial Q_{k_{i}}}{\partial s_{k_{j}}} = Q'_{k_{i}}x_{k_{j}}+Q_{k_{i}k_{j}}.\end{aligned}$$ Evaluation of $x_{k_{i}k_{j}}$ $$\begin{gathered} x_{k_{i}k_{j}} = -\frac{\partial}{\partial s_{k_{j}}} \left(\frac{Q_{k_{i}}}{Q'}\right) = -\frac{Q'_{i}x_{k_{j}}+Q_{k_{i}k_{j}}}{Q'}+ \frac{Q_{k_{i}}(Q''x_{k_{j}}+ Q'_{k_{j}})}{(Q')^{2}}=\\=-\frac{Q_{k_{i}k_{j}}}{Q'} +\frac{Q'_{k_{i}}Q_{k_{j}}+ Q'_{k_{j}}Q_{k_{i}}}{(Q')^{2}}-\frac{Q'' Q_{k_{i}} Q_{k_{j}}}{(Q')^{3}}.\end{gathered}$$ Evaluation of $x_{k_{i}k_{j}k_{l}}$ $$\begin{aligned} &x_{k_{i}k_{j}k_{l}} = \frac{\partial}{\partial s_{k_{l}}}\left(\frac{Q'_{k_{i}}Q_{k_{j}}+ Q'_{k_{j}}Q_{k_{i}}}{(Q')^{2}}-\frac{Q_{k_{i}k_{j}}} {Q'} -\frac{Q''Q_{k_{i}}Q_{k_{j}}}{(Q')^{3}}\right) =\notag\\ &= \frac{(Q''_{k_{i}}x_{k_{l}}+Q'_{k_{i}k_{l}}) Q_{k_{j}}+Q'_{k_{i}}(Q'_{k_{j}} x_{k_{l}}+Q_{k_{j}k_{l}})+ (Q'_{k_{i}}x_{k_{l}}+Q_{k_{i}k_{l}}) Q'_{k_{j}}+Q_{k_{i}}(Q''_{k_{j}}x_{k_{l}}+ P'_{k_{j}k_{l}})}{(P')^{2}} -\notag\\ &-\frac{2(Q'_{k_{i}}Q_{k_{j}}+Q_{k_{i}} Q'_{k_{j}})(Q''x_{k_{l}}+Q'_{k_{l}})} {(Q')^{3}}-\frac{Q'_{k_{i}k_{j}}x_{k_{l}}+ Q_{k_{i}k_{j}k_{l}}}{Q'}+\frac{Q_{k_{i}k_{j}}} {(Q')^{2}} (Q''x_{k_{l}}+Q'_{k_{l}})-\notag\\ &-\frac{(Q_{k_{i}}Q_{k_{j}} (Q'''x_{k_{l}}+Q''_{k_{l}})+(Q'_{k_{i}}x_{k_{l}} +Q_{k_{i}k_{l}}) Q''Q_{k_{j}}+Q_{k_{i}} Q''(Q'_{k_{j}}x_{k_{l}}+Q_{k_{j}k_{l}}))} {(Q')^{3}}+\notag\\ &+\frac{3Q''Q_{k_{i}}Q_{k_{j}}} {(Q')^{4}}(Q''x_{k_{l}}+Q'_{k_{l}}).\end{aligned}$$ Replace $x_{k_{i}}, x_{k_{j}}$ and $x_{k_{l}}$ in this formula we obtain $$\begin{aligned} &x_{k_{i}k_{j}k_{l}} = -\frac{Q_{k_{i}k_{j}k_{l}}}{Q'}+ \frac{Q'_{k_{i}k_{j}}Q_{k_{l}}+ Q'_{k_{j}k_{l}}Q_{k_{i}}+Q'_{k_{l}k_{i}}Q_{k_{j}}} {(Q')^{2}}+ \frac{Q'_{k_{i}}Q_{k_{j}k_{l}}+Q'_{k_{j}} Q_{k_{l}k_{i}}+Q'_{k_{l}}Q_{k_{i}k_{j}}} {(Q')^{2}}-\notag\\ &-\frac{Q''_{k_{i}}Q_{k_{j}}Q_{k_{l}} +Q''_{k_{j}}Q_{k_{l}}Q_{k_{i}}+ Q''_{k_{l}}Q_{k_{i}}Q_{k_{j}}}{(Q')^{3}} -\frac{2(Q'_{k_{i}}Q'_{k_{j}}Q_{k_{l}}+Q'_{k_{j}} Q'_{k_{l}}Q_{k_{i}}+ Q'_{k_{l}}Q'_{k_{i}}Q_{k_{j}})}{(Q')^{3}}-\notag\\ &-Q''\frac{Q_{k_{i}k_{j}} Q_{k_{l}}+Q_{k_{j}k}Q_{k_{i}}+ Q_{k_{l}k_{i}}Q_{k_{j}}}{(Q')^{3}}+ 3Q''\frac{Q'_{k_{i}}Q_{k_{j}}Q_{k_{l}} +Q'_{k_{j}}Q_{k_{l}}Q_{k_{i}}+ Q'_{k_{l}}Q_{k_{i}}Q_{k_{j}}}{(Q')^{4}}+\notag\\ &+Q''' \frac{Q_{k_{i}}Q_{k_{j}}Q_{k_{l}}}{(Q')^{4}}-3(Q'')^{2} \frac{Q_{k_{i}}Q_{k_{j}}Q_{k_{l}}}{(Q')^{5}}.\end{aligned}$$ Using the notation of the symmetrization we have $$\begin{gathered} x_{k_{i}k_{j}k_{l}}= -\frac{Q_{k_{i}k_{j}k_{l}}}{Q'}+\frac{Q'_{(k_{i}k_{j}} Q_{k_{l})}}{(Q')^{2}} +\frac{Q'_{(k_{i}}Q_{k_{j}k_{l})}}{(Q')^{2}}- \frac{Q''_{(k_{i}}Q_{k_{j}}Q_{k_{l})}}{(Q')^{3}}- \frac{2Q'_{(k_{i}}Q'_{k_{j}}Q_{k_{l})}}{(Q')^{3}} -Q''\frac{Q_{(k_{i}k_{j}}Q_{k_{l})}}{(Q')^{3}}+ \\+ 3Q'' \frac{Q'_{(k_{i}}Q_{k_{j}}Q_{k_{l})}} {(Q')^{4}} +Q'''\frac{Q_{k_{i}}Q_{k_{j}}Q_{k_{l}}} {(Q')^{4}}-3(Q'')^{2}\frac{Q_{k_{i}}Q_{k_{j}} Q_{k_{l}}}{(Q')^{5}}.\end{gathered}$$ Evaluation of five-point numbers ================================ From equations (\[ZQ\]) and (\[Corls\]) one can calculate that $$\begin{aligned} &\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}} = \frac{1}{2}\int_{-1}^{1} \left(Q_{(k_{1}} Q_{k_{2}k_{3}k_{4}k_{5})}+Q_{(k_{1}k_{2}} Q_{k_{3}k_{4}k_{5})}+ Q_{0}Q_{k_{1}k_{2}k_{3}k_{4}k_{5}}\right)dx+\notag\\ &+(Q_{(k_{1}k_{2}k_{3}}Q_{k_{4})}+Q_{(k_{1}k_{2}}Q_{k_{3})k_{4}} +Q_{0}Q_{k_{1}k_{2}k_{3}k_{4}})x_{k_{5}}+\notag\\ &+(Q_{(k_{1}k_{2}k_{3}} Q_{k_{5})}+ Q_{(k_{1}k_{2}}Q_{k_{3})k_{5}}+Q_{0} Q_{k_{1}k_{2}k_{3}k_{5}})x_{k_{4}}+\notag\\ &+(Q_{(k_{1}k_{2}k_{4}}Q_{k_{5})} +Q_{(k_{1}k_{2}} Q_{k_{4})k_{5}}+ Q_{0}Q_{k_{1}k_{2}k_{4}k_{5}})x_{k_{3}}+\notag\\ &+(Q'_{(k_{1}k_{2}}Q_{k_{3})}+Q_{(k_{1}k_{2}}Q'_{k_{3})} +Q'_{0}Q_{k_{1}k_{2}k_{3}} +Q_{0}Q'_{k_{1}k_{2}k_{3}})x_{k_{4}}x_{k_{5}}+\notag\\ &+(Q'_{(k_{1}k_{2}}Q_{k_{4})}+Q_{(k_{1}k_{2}}Q'_{k_{4})}+Q'_{0}Q_{k_{1}k_{2}k_{4}} +Q_{0}Q'_{k_{1}k_{2}k_{4}})x_{k_{3}}x_{k_{5}}+\notag\\ &+(Q'_{(k_{1}k_{2}}Q_{k_{5})}+Q_{(k_{1}k_{2}}Q'_{k_{5})} +Q'_{0}Q_{k_{1}k_{2}k_{5}} +Q_{0}Q'_{k_{1}k_{2}k_{5}})x_{k_{3}}x_{k_{4}}+\notag\\ &+(Q_{(k_{1}k_{2}}Q_{k_{3})}+ Q_{0}Q_{k_{1}k_{2}k_{3}})x_{k_{4}k_{5}}+\notag\\ &+(Q_{(k_{1}k_{2}}Q_{k_{4})}+ Q_{0}Q_{k_{1}k_{2}k_{4}})x_{k_{3}k_{5}}+\notag\\ &+(Q_{(k_{1}k_{2}}Q_{k_{5})}+ Q_{0}Q_{k_{1}k_{2}k_{5}})x_{k_{3}k_{4}}+\notag\\ &+(Q''_{(k_{1}}Q_{k_{2})}+2Q'_{k_{2}} Q'_{k_{1}}+ Q''_{0}Q_{k_{1}k_{2}}+2Q'_{0}Q'_{k_{1}k_{2}}+ Q_{0}Q''_{k_{1}k_{2}})x_{k_{3}}x_{k_{4}}x_{k_{5}}+\notag\\ &+(Q'_{(k_{1}}Q_{k_{2})}+ Q'_{0}Q_{k_{1}k_{2}}+ Q_{0}Q'_{k_{1}k_{2}})(x_{(k_{3}k_{5}}x_{k_{4})})+(Q_{k_{1}}Q_{k_{2}}+Q_{0} Q_{k_{1}k_{2}})x_{k_{3}k_{4}k_{5}}. \label{Zkkkkk}\end{aligned}$$ All integrated terms are taken at the point $x=1$. Substituting in this formula values for $x_{k_{i}}$, $x_{k_{i}k_{j}}$ and $x_{k_{i}k_{j}k_{l}}$, calculated in Appendix B, after simplification we find $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}} =& \frac{1}{2}\int_{-1}^{1} \left(Q_{(k_{1}} Q_{k_{2}k_{3}k_{4}k_{5})}+ Q_{(k_{1}k_{2}}Q_{k_{3}k_{4}k_{5})} +Q_{0}Q_{k_{1}k_{2}k_{3}k_{4}k_{5}}\right)dx- \frac{Q_{(k_{1}k_{2}k_{3}}Q_{k_{4}}Q_{k_{5})}} {Q_{0}'}-\notag\\ &-\frac{Q_{(k_{1}k_{2}}Q_{k_{3}k_{4}} Q_{k_{5})}}{Q_{0}'}+ \frac{Q'_{(k_{1}k_{2}}Q_{k_{3}}Q_{k_{4}} Q_{k_{5})}}{(Q_{0}')^{2}} +\frac{Q'_{(k_{1}}Q_{k_{2}k_{3}} Q_{k_{4}}Q_{k_{5})}}{(Q'_{0})^{2}}- \notag\\ &-\frac{Q''_{(k_{1}}Q_{k_{2}}Q_{k_{3}} Q_{k_{4}}Q_{k_{5})}} {(Q'_{0})^{2}}- \frac{2Q'_{(k_{1}}Q'_{k_{2}}Q_{k_{3}} Q_{k_{4}}Q_{k_{5})}} {(Q'_{0})^{3}}- Q''_{0}\frac{Q_{(k_{1}k_{2}}Q_{k_{3}}Q_{k_{4}} Q_{k_{5})}}{(Q'_{0})^{3}}+\notag\\ &+3Q''_{0} \frac{Q'_{(k_{1}}Q_{k_{2}}Q_{k_{3}}Q_{k_{4}} Q_{k_{5})}}{(Q'_{0})^{4}}+\left(Q'''_{0}-\frac{3(Q''_{0})^{2}}{Q'_{0}}\right) \frac{Q_{k_{1}}Q_{k_{2}} Q_{k_{3}}Q_{k_{4}}Q_{k_{5}}}{(Q'_{0})^{4}}.\end{aligned}$$ Because $Q_{k_{1}...k_{n}}(x) =\frac{d^{n-1}}{dx^{n-1}}L_{p-\sum k -n}(x)$, using Appendix A, we find that $$\begin{aligned} &Q_{0}(1) = 0,\\ &Q'_{0}(1) = 1,\\ &Q_{k_{i}}(1) = 1,\\ &Q''_{0}(1) = \frac{p(p+1)}{2},\\ &Q'''_{0}(1) = \frac{1}{8}(p-1)p(p+1)(p+2),\\ &Q'_{k_{i}}(1)=F_{\Theta}(k_{i}-1),\\ &Q_{k_{i}k_{j}}(1)=F_{\Theta}(k_{ij}),\\ &Q_{k_{i}k_{j}k_{l}}(1)=H_{\Theta}(k_{ijl}),\\ &Q'_{k_{i}k_{j}}(1)=H_{\Theta}(k_{ij}-1), \\ &Q''_{k_{i}}(1)=H_{\Theta}(k_{i}-2),\end{aligned}$$ here new function were introduced $$\begin{aligned} &F_{\Theta}(k) = \frac{1}{2}(p-1-k)(p-2-k)\Theta(p-1-k),\\ &H_{\Theta}(k)=\frac{1}{2}F_{\Theta}(k)F_{\Theta}(k+2)= \frac{1}{8}\Theta(p-1-k)\prod\limits_{r=1}^{4}(p-r-k).\end{aligned}$$ After partial simplification we get $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}} =& \frac{1}{2}\int_{-1}^{1} \left(Q_{(k_{1}}Q_{k_{2}k_{3}k_{4}k_{5})}+ Q_{(k_{1}k_{2}}Q_{k_{3}k_{4}k_{5})}\right)dx+ \left(Q'''_{0}-3(Q''_{0})^{2}\right)- \notag\\ &+\sum\limits_{i=1}^{5}(3Q''_{0}Q'_{k_{i}}-Q''_{k_{i}})+ \sum\limits_{i<j}(Q'_{k_{i}k_{j}}- 2Q'_{k_{i}}Q'_{k_{j}}-Q''_{0}Q_{k_{i}k_{j}})+ \notag\\ &+ \sum\limits_{i,j,l}Q'_{k_{i}}Q_{k_{j}k_{l}}-\sum\limits_{i<j<l}Q_{k_{i}k_{j}k_{l}}- \sum\limits_{i,j,k,l}Q_{k_{i}k_{j}}Q_{k_{l}k_{m}}.\end{aligned}$$ Also introduce the functions $$\begin{aligned} &F(k)=\frac{1}{2}(p-1-k)(p-2-k),\\ &H(k)=\frac{1}{2}F(k)F(k+2)= \frac{1}{8}\prod\limits_{r=1}^{4}(p-r-k),\end{aligned}$$ which already doesn’t depend from $\Theta(p-1-k)$. Divide the five-point correlation numbers on several parts $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}= \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{I})}+ \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{J})}+ \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{1})}+ \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2})},\end{aligned}$$ where $$\begin{aligned} &\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{I})} =\frac{1}{2}\int_{-1}^{1} Q_{(k_{1}}Q_{k_{2}k_{3}k_{4}k_{5})}dx, &\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{J})} =\frac{1}{2}\int_{-1}^{1} Q_{(k_{1}k_{2}}Q_{k_{3}k_{4}k_{5})}dx,\end{aligned}$$ $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{1})} =\sum\limits_{i=1}^{5}\left(\frac{3p(p+1)}{2} F_{\Theta}(k_{i}-1)-H_{\Theta}(k_{i}-2)\right)-\\ -2\sum\limits_{i<j}F_{\Theta}(k_{i}-1) F_{\Theta}(k_{j}-1) -\frac{p(p+1)(5p^{2}+5p+2)}{8},\end{gathered}$$ $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2})}= \sum\limits_{i<j}\left(H_{\Theta}(k_{ij}-1) -F_{\Theta}(k_{ij})\frac{p(p+1)}{2}+F_{\Theta}(k_{ij})\sum\limits_{l\neq i,j} F_{\Theta}(k_{l}-1)\right)- \\-\sum\limits_{i<j<l}H_{\Theta}(k_{ijl})-\sum\limits_{i,j,l,m} F_{\Theta}(k_{ij})F_{\Theta}(k_{lm}).\end{gathered}$$ Part of partition function $\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{1})}$ has such a difference, that one shouldn’t take into account influence on it of the factors $\Theta(a-b)$, due to the inequalities $k_{i}\leqslant p-1$. So we can rewrite it by means of the functions $F(k)$ and $H(k)$, as follows $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{1})} =\sum\limits_{i=1}^{5}\left(\frac{3p(p+1)}{2} F(k_{i}-1)-H(k_{i}-2)\right)-\\ -2\sum\limits_{i<j}F(k_{i}-1) F(k_{j}-1) -\frac{p(p+1)(5p^{2}+5p+2)}{8}.\end{gathered}$$ Then using that $$\begin{aligned} &H_{\Theta}(k_{ijl}) = H(k_{ijl})(1- \Theta(k_{ijl}-p)), \notag\\ &H_{\Theta}(k_{ij}-1)= H(k_{ij}-1) (1-\Theta(k_{ij}-p)),\notag\\ &F_{\Theta}(k_{ij}) = F(k_{ij})(1- \Theta(k_{ij}-p)),\notag\\ &F_{\Theta}(k_{ij})F_{\Theta}(k_{lm})= F(k_{ij})F(k_{lm})(1-(\Theta(k_{ij}-p) +\Theta(k_{lm}-p)-\Theta(k_{ij}-p)\Theta(k_{lm}-p))),\end{aligned}$$ let us write $\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2})}$, in the following way $$\begin{aligned} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2})} = \tilde{\mathcal{Z}}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2})}+ \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2}\Theta)},\end{aligned}$$ where $$\begin{gathered} \tilde{\mathcal{Z}}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2})}= \sum\limits_{i<j}\left(H(k_{ij}-1) -F(k_{ij})\frac{p(p+1)}{2}+F(k_{ij})\sum\limits_{l\neq i,j} F(k_{l}-1)\right)- \\-\sum\limits_{i<j<l}H(k_{ijl})-\sum\limits_{i,j,l,m} F(k_{ij})F(k_{lm}),\end{gathered}$$ which doesn’t already depend on the step-functions, and $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2}\Theta)} =\\ = \sum\limits_{m<n}\Theta(k_{mn}-p) \left[F(k_{mn})\left(\sum\limits_{i<j}^{}F(k_{ij}) -\sum\limits_{l\neq m,n}F(k_{l}-1)+\frac{p(p+1)}{2}\right) -H(k_{mn}-1)\right]+\\ +\sum\limits_{i<j<l}H(k_{ijl})\Theta(k_{ijl}-p)- \sum\limits_{i,j,l,m}F(k_{ij})F(k_{lm})\Theta(k_{ij}-p)\Theta(k_{lm}-p),\end{gathered}$$ in which there is all dependence on $\Theta(a-b)$. Having simplified, we have $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2}\Theta)} = - \frac{1}{8}(2p-3-k)(2p-5-k)\sum\limits_{m<n}k_{ijl}^{mn}(k_{ijl}^{mn}+2) \Theta(k_{mn}-p)+\\+\sum\limits_{i<j<l} H(k_{ijl})[\Theta(k_{ijl}-p)+\Theta(k_{mn}-p)] -\sum\limits_{i,j,l,m}F(k_{ij})F(k_{lm})\Theta(k_{ij}-p)\Theta(k_{lm}-p).\end{gathered}$$ On the other hand one can calculate $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{1})} +\tilde{\mathcal{Z}}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2})}=\\ = \frac{1}{8}(4\sum_{i=1}^{5}k_{i}^{2}-k^{2}-2k -8)(2p-3-k)(2p-5-k) - \sum\limits_{i<j<l}H(k_{ijl}).\end{gathered}$$ As the result we find that $$\begin{gathered} \mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{1})} +\mathcal{Z}_{k_{1}k_{2}k_{3}k_{4}k_{5}}^{(\textrm{2})}= \\ =\frac{1}{8}(4\sum_{i=1}^{5}k_{i}^{2}-k^{2}-2k -8- \sum\limits_{m<n}k_{ijl}^{mn}(k_{ijl}^{mn}+2) \Theta(k_{mn}-p))(2p-3-k)(2p-5-k)+\\ +\sum\limits_{i<j<l}H(k_{ijl})G_{1}-\sum\limits_{i,j,l,m}F(k_{ij})F(k_{lm})G_{2}, \label{Corup}\end{gathered}$$ where $G_{1}=\Theta(k_{ijl}-p)+ \Theta(k_{mn}-p)-1 \;$ and $G_{2}=\Theta(k_{ij}-p)\Theta(k_{lm}-p)$. In the even section we will need formulas $$\begin{gathered} \frac{1}{2}\int_{-1}^{1}Q_{k_{i}k_{j}k_{l}k_{m}}(x)Q_{k_{n}}(x)dx =\\ = \frac{1}{8}(k_{n}^{ijlm}-2)(k_{n}^{ijlm}-4)(2p-3-k)(2p-5-k)\Theta(k_{n}^{ijlm}-6), \label{I1}\end{gathered}$$ $$\begin{gathered} \frac{1}{2}\int_{-1}^{1}Q_{k_{m}k_{n}}(x)Q_{k_{i}k_{j}k_{l}}(x)dx =\\ =\left(H(k_{ijl})-\frac{k^{mn}_{ijl}(k^{mn}_{ijl}+2)(2p-3-k)(2p-5-k)} {8}\Theta(k_{mn}^{ijl}-2)\right) \Theta(p-1-k_{ijl})\Theta(p-1-k_{mn}). \label{I2}\end{gathered}$$ [99]{} A. Polyakov, “Quantum Geometry of Bosonic Strings”, Phys. Lett. [B103]{}: 207-210, (1981). P.H. Ginsparg and G.W. Moore, “Lectures on 2-D gravity and 2-D string theory”, [arXiv:9304011 \[hep-th\]]{};\ P. Di Francesco, P.H. Ginsparg, J. Zinn-Justin, “2-D Gravity and random matrices”, Phys.Rep.254:1-133,(1995); [arXiv:9306153 \[hep-th\]]{} G.W. Moore, N. Seiberg, M. Staudacher, “From loops to states in 2-D quantum gravity”, Nucl. Phys. [B362]{}, 665-709, (1991) A.A. Belavin and A.B. Zamolodchikov, “On correlation numbers in 2D minimal gravity and matrix models”, Jour. Phys. [A42]{} (2009) 304004; [arXiv:0811.0450 \[hep-th\]]{} [^1]: [email protected]	{ "pile_set_name": "ArXiv" }
--- abstract: \| The bandwidth theorem \[Mathematische Annalen, 343(1):175–205, 2009\] states that any $n$-vertex graph $G$ with minimum degree $\big(\tfrac{k-1}{k}+o(1)\big)n$ contains all $n$-vertex $k$-colourable graphs $H$ with bounded maximum degree and bandwidth $o(n)$. We provide sparse analogues of this statement in random graphs as well as pseudorandom graphs. More precisely, we show that for $p\gg \big(\tfrac{\log n}{n}\big)^{1/\Delta}$ a.a.s. each spanning subgraph $G$ of $G(n,p)$ with minimum degree $\big(\tfrac{k-1}{k}+o(1)\big)pn$ contains all $n$-vertex $k$-colourable graphs $H$ with maximum degree $\Delta$, bandwidth $o(n)$, and at least $C p^{-2}$ vertices not contained in any triangle. A similar result is shown for sufficiently bijumbled graphs, which, to the best of our knowledge, is the first resilience result in pseudorandom graphs for a rich class of subgraphs. Finally, we provide improved results for $H$ with small degeneracy, which in particular imply a resilience result in $G(n,p)$ with respect to the containment of spanning bounded degree trees for $p\gg \big(\tfrac{\log n}{n}\big)^{1/3}$. address: - '(PA) London School of Economics, Department of Mathematics, Houghton Street, London WC2A 2AE, UK' - '(JB) London School of Economics, Department of Mathematics, Houghton Street, London WC2A 2AE, UK' - '(JE) Technische Universität Hamburg, Institut für Mathematik, Am Schwarzenberg-Campus 3, 21073 Hamburg, Germany ' - '(AT) Technische Universität Hamburg, Institut für Mathematik, Am Schwarzenberg-Campus 3, 21073 Hamburg, Germany ' author: - Peter Allen - Julia Böttcher - Julia Ehrenmüller - Anusch Taraz bibliography: - 'references.bib' title: The Bandwidth Theorem in sparse graphs --- Introduction ============ A central topic in extremal graph theory is to determine minimum degree conditions which force a graph $G$ to contain a copy of some large or even spanning subgraph $H$. The prototypical example of such a theorem is Dirac’s theorem [@dirac1952], which states that if $\delta(G)\ge\tfrac12 v(G)$ then $G$ is Hamiltonian. Analogous results were established for a wide range of spanning subgraphs $H$ with bounded maximum degree such as powers of Hamilton cycles, trees, or $F$-factors for any fixed graph $F$ (see e.g. [@kuhnsurvey] for a survey). One feature that all these subgraphs $H$ have in common is that their bandwidth is small. The bandwidth of a graph $H$ is the minimum $b$ such that there is a labelling of the vertex set of $H$ by integers $1, \ldots, n$ with $\|i-j\| \leq b$ for every edge $ij$ of $H$. And indeed, it was shown in [@bottcher2009proof] that a more general result holds, which provides a minimum degree condition forcing any spanning bounded degree subgraphs of small bandwidth. This result is by now often called the bandwidth theorem. \[thm:bandwidth\] For every $\gamma >0$, $\Delta \geq 2$, and $k\geq 1$, there exist $\beta>0$ and $n_0 \geq 1$ such that for every $n\geq n_0$ the following holds. If $G$ is a graph on $n$ vertices with minimum degree $\delta(G) \geq \left(\frac{k-1}{k}+\gamma\right)n$ and if $H$ is a $k$-colourable graph on $n$ vertices with maximum degree $\Delta(H) \leq \Delta$ and bandwidth at most $\beta n$, then $G$ contains a copy of $H$. We remark that in contrast to the above mentioned earlier results for specific bounded degree spanning subgraphs the minimum degree condition in this theorem has an error term $\gamma n$, but it is known that this cannot completely be omitted in this general statement. In that sense the minimum degree condition in Theorem \[thm:bandwidth\] is best-possible. It is also known that the bandwidth condition cannot be dropped completely (see [@bottcher2009proof] for further explanations). Moreover, this condition does not limit the class of graphs under consideration unreasonably, because many interesting classes of graphs have sublinear bandwidth. Indeed, it was shown in [@bottcher2010bandwidth] that for bounded degree $n$-vertex graphs, restricting the bandwidth to $o(n)$ is equivalent to restricting the treewidth to $o(n)$ or forbidding linear sized expanding subgraphs, which implies that bounded degree planar graphs, or more generally classes of bounded degree graphs defined by forbidding some fixed minor have bandwidth $o(n)$. Generalisations of Theorem \[thm:bandwidth\] were obtained in [@BoeHeiTar; @BoeTarWue; @KnoTre; @Lee:degenerateBandwidth]. In this paper we are interested in the transference of Theorem \[thm:bandwidth\] to sparse graphs. Such transference results recently received much attention, including for example the breakthrough result on the transference of Turán’s theorem to random graphs by Conlon and Gowers [@ConGow] and Schacht [@Schacht]. The random graph model we shall consider here is the binomial random graph $G(n,p)$, which has $n$ vertices, and each pair of vertices forms an edge independently with probability $p$. The appearance of large or spanning subgraphs of $G(n,p)$ was studied since the early days of probabilistic combinatorics and by now many important results were obtained. Gems include the Johansson–Kahn–Vu theorem [@JohKahVu] which determines the threshold for $G(n,p)$ to contain an $F$-factor whenever $F$ is strictly balanced (as is the case, for example, when $F$ is a clique), and the theorem of Riordan [@Riordan] which gives a very good, and in many cases tight, upper bound on the threshold for $G(n,p)$ to contain a general spanning graph $H$. For spanning graphs $H$ with maximum degree $\Delta(H)\le\Delta$ Riordan’s theorem implies that $G(n,p)$ asymptotically almost surely (a.a.s.), that is, with probability tending to $1$ as $n$ tends to infinity, contains $H$ as a subgraph if $p\cdot n^{\frac{2}{\Delta+1}-\frac{2}{\Delta(\Delta+1)}}\to\infty$. This is not believed to be best possible. Indeed, it follows from the Johansson–Kahn–Vu theorem that the threshold for $G(n,p)$ to contain a $K_{\Delta+1}$-factor is $(\log n)^{1/\binom{\Delta+1}{2}}/n^{2/(\Delta+1)}$, and it is conjectured in [@FerLuhNgu] that above this probability we also get any other sequence of spanning graphs $H=(H_n)$ with $\Delta(H)\le\Delta$. This was proved, using the Johansson–Kahn–Vu theorem, to be true for almost spanning graphs by Ferber, Luh, and Nguyen [@FerLuhNgu]. \[thm:FerLuhNgu\] For every ${\varepsilon}>0$ and $\Delta\ge 1$, and every sequence $H=(H_n)$ of graphs with $v(H)\le(1-{\varepsilon})n$ and $\Delta(H)\le\Delta$, the random graph $G(n,p)$ a.a.s. contains $H$ if $p\cdot n^{2/(\Delta+1)}/(\log n)^{1/\binom{\Delta+1}{2}} \to\infty$. Better bounds are available if we further know that the degeneracy of $H$ is bounded by a constant much smaller than $\Delta(H)$. The degeneracy of $H$ is the smallest integer $D$ such that any subgraph of $H$ has a vertex of degree at most $D$. Surprisingly, for this class of graphs $H$ already Riordan’s theorem implies an essentially optimal bound. \[cor:Riordan\] For every $\Delta\ge 1$ and $D\ge 3$, and every sequence $H=(H_n)$ of graphs with $v(H)\le n$ and $\Delta(H)\le\Delta$ and degeneracy at most $D$, the random graph $G(n,p)$ a.a.s. contains $H$ if $p\cdot n^{1/D} \to\infty$. This is best possible because a simple first moment calculation shows that if $p\cdot n^{1/D}\to 0$ then $G(n,p)$ a.a.s. does not contain the $D$-th power of a Hamilton path, which is a $D$-degenerate graph with maximum degree $2D$. A feature that both Riordan’s theorem and the Johansson–Kahn–Vu theorem (and consequently all results which rely on them, such as Theorem \[thm:FerLuhNgu\] and Corollary \[cor:Riordan\]) have in common is that their proofs are non-constructive. Furthermore, they do not allow for so-called universality results. A graph $G$ is said to be universal for a family ${\mathcal{H}}$ of graphs if $G$ contains copies of all graphs in ${\mathcal{H}}$ simultaneously. The random graph $G(n,p)$ is known to be universal for various families of graphs, but in almost all cases we only know an upper bound on the threshold for universality, which we do not believe is the correct answer. The reason why probabilistic existence results such as Corollary \[cor:Riordan\] do not imply universality is that in $G(n,p)$ the failure probability for containing any given spanning graph $H$ without isolated vertices is at least $(1-p)^{n-1}$, the probability that a fixed vertex of $G(n,p)$ is isolated. This probability is too large to apply a union bound. Thus, to prove universality results one needs to show that any graph $G$ with some collection of properties that $G(n,p)$ a.a.s. possesses must contain any given $H\in{\mathcal{H}}$. Using this approach, and improving on a series of earlier results, Dellamonica, Kohayakawa, Rödl and Ruciński [@dellamonica2014] obtained the following universality result for the family ${\mathcal{H}}(n,\Delta)$ of $n$-vertex graphs with maximum degree $\Delta$. \[thm:universal\] For all $\Delta\ge3$ there is $C$ such that if $p\ge C\big(\frac{\log n}{n}\big)^{1/\Delta}$ then $G(n,p)$ is a.a.s. universal for ${\mathcal{H}}(n,\Delta)$. However, it is conjectured that universality and the appearance of a $K_{\Delta+1}$-factor occur together, at the threshold given in Theorem \[thm:FerLuhNgu\]. A probability bound which is better, but still far from the conjectured truth, was so far only established for almost spanning graphs by Conlon, Ferber, Nenadov and Škorić [@CFNS], who showed that for $\Delta\ge 3$, if $p\gg n^{-1/(\Delta-1)}\log^5 n$ then $G(n,p)$ is a.a.s. universal for ${\mathcal{H}}\big((1-o(1))n,\Delta)\big)$. For graphs with small degeneracy, again, the following better bound exists, but this also is far away from the threshold in Corollary \[cor:Riordan\], which is a plausible candidate for the correct answer. \[thm:Duniversal\] For all $\Delta, D\ge 1$ there is $C$ such that if $p\ge C\big(\frac{\log n}{n}\big)^{1/(2D+1)}$ then $G(n,p)$ is a.a.s. universal for all graphs in ${\mathcal{H}}(n,\Delta)$ with degeneracy at most $D$. Furthermore, one may ask how robustly $G(n,p)$ contains certain large subgraphs $H$. Questions of this type were considered by Alon, Capalbo, Kohayakawa, Rödl, Ruciński and Szemerédi [@ACKRRS], and further popularised by Sudakov and Vu [@SudVu], who introduced the term local resilience. Let ${\mathcal{P}}$ be a monotone increasing graph property and let $G$ be a graph in ${\mathcal{P}}$. The local resilience of $G$ with respect to ${\mathcal{P}}$ is defined to be the minimum $r \in \mathbb R$ such that by deleting at each vertex $v\in V(G)$ at most $r\deg(v)$ edges one can obtain a graph not in ${\mathcal{P}}$. In this language, for example, Theorem \[thm:bandwidth\] says that the the local resilience of $K_n=G(n,1)$ with respect to being universal for all $k$-colourable graphs in ${\mathcal{H}}(n,\Delta)$ with sublinear bandwidth is $\frac1k-o(1)$, and a sparse analogue asks for a similar local resilience result to hold a.a.s. for $G(n,p)$. Lee and Sudakov [@lee2012] obtained a very strong local resilience result for Hamilton cycles. Improving on [@SudVu], they showed that the local resilience of $G(n,p)$ with respect to Hamiltonicity is a.a.s. at least $\tfrac12-o(1)$ when $p=\Omega\big(\tfrac{\log n}{n}\big)$. This is optimal up to the constant factor, as below this probability $G(n,p)$ is itself a.a.s. not Hamiltonian. Triangle factors were investigated by Balogh, Lee and Samotij [@balogh2012corradi], who proved that the local resilience of $G(n,p)$ with respect to the containment of a triangle factor on $n-O\big(p^{-2}\big)$ vertices is a.a.s. $\frac13-o(1)$ if $p\gg (\frac{\log n}{n})^{1/2}$. It was observed by Huang, Lee, and Sudakov [@huang2012] that we cannot hope to cover all vertices in such a result with triangles: Already for constant $p$ it is easy to delete edges in the neighbourhood of $\Theta(p^{-2})$ vertices in $G(n,p)$ without violating the resilience condition. Very recently Noever and Steger [@NoeSte] showed that the local resilience of $G(n,p)$ with respect to containing a $(1-o(1))n$-vertex squared cycle (a cycle with all edges between vertices at distance $2$ added) is a.a.s. $\tfrac13-o(1)$ provided $p\gg n^{-1/2+o(1)}$. Even more recently, Nenadov and Škorić [@NenSko] removed the $\log$-factor from the probability bound of [@balogh2012corradi]. These results are notable in that the bounds on $p$ are close to optimal: for $p\ll n^{-1/2}$, a.a.s. most edges of $G(n,p)$ are not in triangles and one can obtain a triangle-free graph by deleting only a tiny fraction of edges at each vertex, so that the local resilience of $G(n,p)$ with respect to containing triangles is $o(1)$. Furthermore, for $p\ll n^{-1/2}$ the random graph $G(n,p)$ itself does not contain any $(1-o(1))n$-vertex squared cycle. More general subgraphs $H$ were considered by Huang, Lee and Sudakov [@huang2012], who proved an analogue of the bandwidth theorem (Theorem \[thm:bandwidth\]) in $G(n,p)$ with $0<p<1$ constant (for subgraphs $H$ with certain vertices not in triangles). A version which works for much smaller probabilities $p\gg(\frac{\log n}{n})^{1/\Delta}$ in the special case of bipartite graphs $H$ on $(1-o(1))n$ vertices (with maximum degree $\Delta$ and sublinear bandwidth) was established in [@bottcher2013almost]. An even better bound on $p$ was obtained when we restrict the problem to the class of almost spanning trees $H$: Balogh, Csaba, and Samotij [@balogh2011] proved that the local resilience of $G(n,p)$ with respect to containing copies of all trees $T$ on $(1-o(1))n$ vertices with $\Delta(T) \leq \Delta$ is asymptotically almost surely at least $1/2 - o(1)$ if $p \gg 1/n$, which is optimal. Finally, returning to $H$-factors, Conlon, Gowers, Samotij and Schacht [@CGSS] gave resilience results for almost-spanning $H$-factors which work down to the optimal probability threshold, but leave a quite large number of vertices uncovered; Nenadov and Škorić [@NenSko] substantially improved this, but (for most graphs) the number of vertices left uncovered in their result is still not the correct order of magnitude. Our results. {#our-results. .unnumbered} ------------ We prove several sparse analogues of the bandwidth theorem (Theorem \[thm:bandwidth\]). Our first result is a version for sparse random graphs, extending the resilience results of Huang, Lee and Sudakov [@huang2012] and [@bottcher2013almost]. \[thm:main\] For each $\gamma >0$, $\Delta \geq 2$, and $k \geq 1$, there exist constants $\beta^\ast >0$ and ${C^{\ast}}>0$ such that the following holds asymptotically almost surely for $\Gamma = G(n,p)$ if $p \geq {C^{\ast}}\big(\frac{\log n}{n}\big)^{1/\Delta}$. Let $G$ be a spanning subgraph of $\Gamma$ with $\delta(G) \geq\left(\frac{k-1}{k}+ \gamma\right)pn$, and let $H$ be a $k$-colourable graph on $n$ vertices with $\Delta(H) \leq \Delta$, bandwidth at most $\beta^\ast n$, and with at least ${C^{\ast}}p^{-2}$ vertices which are not contained in any triangles of $H$. Then $G$ contains a copy of $H$. Observe that the bound on $p$ achieved in this result matches the bound in the universality result in Theorem \[thm:universal\]. Hence, though we do not believe it to be optimal, improving it will most likely be hard. Moreover, as explained in conjunction with Theorem \[thm:bandwidth\], the minimum degree of $G$ cannot be decreased, nor can the bandwidth restriction be removed. As indicated above, it is also necessary that $\Theta(p^{-2})$ vertices of $H$ are not in triangles. If in addition the subgraph $H$ is also $D$-degenerate, we can prove a variant of Theorem \[thm:main\] for $p\gg (\log n/n)^{1/(2D+1)}$. Again, this probability bound matches the one in the currently best universality result for $D$-degenerate graphs given in Theorem \[thm:Duniversal\]. As before we require a certain number of vertices which are not in triangles of $H$. But, due to technicalities of our proof method, in addition these vertices are now also required not to be in four-cycles. \[thm:degenerate\] For each $\gamma >0$, $\Delta \geq 2$, and $D, k \geq 1$, there exist constants $\beta^\ast >0$ and ${C^{\ast}}>0$ such that the following holds asymptotically almost surely for $\Gamma = G(n,p)$ if $p \geq {C^{\ast}}\big(\frac{\log n}{n}\big)^{1/(2D+1)}$. Let $G$ be a spanning subgraph of $\Gamma$ with $\delta(G) \geq\left(\frac{k-1}{k}+ \gamma\right)pn$ and let $H$ be a $D$-degenerate, $k$-colourable graph on $n$ vertices with $\Delta(H) \leq \Delta$, bandwidth at most $\beta^\ast n$ and with at least ${C^{\ast}}p^{-2}$ vertices which are not contained in any triangles or four-cycles of $H$. Then $G$ contains a copy of $H$. Since trees are $1$-degenerate this implies a resilience result for trees when $p\gg (\frac{\log n}{n})^{1/3}$. This probability bound is much worse than that obtained by Balogh, Csaba, and Samotij [@balogh2011] for almost-spanning trees, and unlikely to be optimal, but it is the first resilience result for bounded degree spanning trees in $G(n,p)$. Finally, we also establish a sparse analogue of Theorem \[thm:bandwidth\] in bijumbled graphs, one of the most widely studied classes of pseudorandom graphs. A graph $\Gamma$ is $(p,\nu)$-bijumbled if for all disjoint sets $X,Y\subset V(\Gamma)$ we have $$\big\|e(X,Y)-p\|X\|\|Y\|\big\|\le\nu\sqrt{\|X\|\|Y\|}\,.$$ This goes back to an equivalent notion introduced by Thomason [@Tho87] who initiated the study of pseudorandom graphs. It is also related to the well investigated class of $(n,d,\lambda)$-graphs in that an $(n,d,\lambda)$-graph is $\big(\tfrac{d}{n},\lambda\big)$-bijumbled. Only very recently a universality result similar to Theorem \[thm:universal\] was established for bijumbled graphs in [@blowup], where it was shown that $(p,\nu)$-bijumbled graphs $G$ with $\delta(G)\ge\frac12pn$ and $\nu\ll p^{\max(4,(3\Delta+1)/2)}n$ are universal for ${\mathcal{H}}(n,\Delta)$. Our resilience result works for the same bijumbledness condition, though we do not believe it to be optimal. Local resilience results in bijumbled graphs were so far only obtained for special subgraphs $H$: Dellamonica, Kohayakawa, Marciniszyn, and Steger [@dellamonica2008] considered cycles $H$ of length $(1-o(1))n$, the results of Conlon, Fox and Zhao [@CFZ] imply resilience for $F$-factors covering $(1-o(1))n$ vertices, and Krivelevich, Lee and Sudakov [@KriLeeSud] established a resilience result for pancyclicity. Hence, previous to this work only little was known about the resilience of bijumbled (or indeed any other common notion of pseudorandom) graphs. \[thm:jumbled\] For each $\gamma >0$, $\Delta \geq 2$, and $k \geq 1$, there exists a constant $c >0$ such that the following holds for any $p>0$. Given $\nu\le cp^{\max(4,(3\Delta+1)/2)}n$, suppose $\Gamma$ is a $\big(p,\nu\big)$-bijumbled graph, $G$ is a spanning subgraph of $\Gamma$ with $\delta(G) \geq\big(\tfrac{k-1}{k}+\gamma\big)pn$, and $H$ is a $k$-colourable graph on $n$ vertices with $\Delta(H) \leq \Delta$ and bandwidth at most $c n$. Suppose further that there are at least $c^{-1}p^{-6} \nu^2n^{-1}$ vertices in $V(H)$ that are not contained in any triangles of $H$. Then $G$ contains a copy of $H$. The proofs of our results rely on sparse versions of the so-called blow-up lemma. The blow-up lemma is an important tool in extremal graph theory, proved by Komlós, Sárközy and Szemerédi [@komlos1997blow] and was for example instrumental in the proof of the bandwidth theorem and its analogue in $G(n,p)$ for constant $p$ by Huang, Lee and Sudakov [@huang2012]. However it applies only to dense graphs. Several of the earlier resilience results in sparse random graphs developed sparse blow-up type results handling special classes of graphs: Balogh, Lee and Samotij [@balogh2012corradi] proved a sparse blow-up lemma for embedding triangle factors, and in [@bottcher2013almost] a blow-up lemma for embedding almost spanning bipartite graphs in sparse graphs was used. Full versions of the blow-up lemma in sparse random graphs and pseudorandom graphs were established only very recently in [@blowup]. We will use these here. Further, we note that we actually prove somewhat stronger statements than Theorem \[thm:main\], Theorem \[thm:degenerate\], and Theorem \[thm:jumbled\] in the same sense in that a stronger statement than Theorem \[thm:bandwidth\] was proven in [@bottcher2009proof]: we allow $H$ in fact to be $(k+1)$-colourable, where the additional colour may only be assigned to very few well distributed vertices (for details see, e.g., Theorem \[thm:maink\] below). Thus, for instance, even though Theorem \[thm:main\] only implies that the local resilience of $G(n,p)$ with respect to Hamiltonicity is a.a.s. at least $\tfrac12-o(1)$ when $n$ is even, Theorem \[thm:maink\] implies it also for $n$ odd, since although the chromatic number of a Hamilton cycle is $3$, there are $3$-colourings which use the third colour only on one vertex. Organisation. {#organisation. .unnumbered} ------------- The remainder of this paper is organised as follows. In Section \[sec:preliminaries\] we introduce necessary definitions and collect some known results which we need in our proofs. Next, in Section \[sec:mainlemmas\], we outline the proof of the bandwidth theorem in sparse random graphs, Theorem \[thm:main\], and state the four technical lemmas we require. Their proofs are given in Sections \[sec:prooflemG\]–\[sec:prooflembalancing\], and the proof of Theorem \[thm:main\] is presented in Section \[sec:proofmain\]. We provide the modifications required to obtain Theorem \[thm:degenerate\] in Section \[sec:proofdegen\], and those required for Theorem \[thm:jumbled\] in Section \[sec:proofjumbled\]. Finally, Section \[sec:remarks\] contains some concluding remarks, and Appendix \[app:tools\] contains proofs of a few results which are more or less standard but which we could not find in the form we need in the literature. Preliminaries {#sec:preliminaries} ============= Throughout the paper $\log$ denotes the natural logarithm. We assume that the order $n$ of all graphs tends to infinity and therefore is sufficiently large whenever necessary. For reals $a, b >0$ and integer $k\in \mathbb N$, we use the notation $(a\pm b) = [a-b, a+b]$ and $[k] = \{1, \ldots, k\}$. Our graph-theoretic notation is standard and follows [@bollobas1998modern]. In particular, given a graph $G$ its vertex set is denoted by $V(G)$ and its edge set by $E(G)$. Let $A,B\subseteq V$ be disjoint vertex sets. We denote the number of edges between $A$ and $B$ by $e(A,B)$. For a vertex $v \in V(G)$ we write $N_G(v)$ for the neighbourhood of $v$ in $G$ and $N_G(v,A):= N_G(v) \cap A$ for the neighbourhood of $v$ restricted to $A$ in $G$. Given vertices $v_1, \ldots, v_k \in V(G)$ we denote the joint neighbourhood of $v_1, \ldots, v_k$ restricted to a set $A$ by $N_G(v_1, \ldots, v_k; A) = \bigcap_{i\in[k]} N_G(v_i, A)$. Finally, we use the notation $\deg_G(v) := \|N_G(v)\|$ and $\deg_G(v, A) := \|N_G(v,A)\|$, as well as $\deg_G(v_1, \ldots, v_k; A) := \|N_G(v_1, \ldots, v_k;A)\|$ for the degree of $v$ in $G$, the degree of $v$ restricted to $A$ in $G$ and the size of the joint neighbourhood of $v_1, \ldots, v_k$ restricted to $A$ in $G$. Finally, let $\deg_G(v) := \|N_G(v)\|$ be the degree of $v$ in $G$. For the sake of readability, we do not intend to optimise the constants in our theorems and proofs. Now we introduce some definitions and results of the regularity method as well as related tools that are essential in our proofs. In particular, we state a minimum degree version of the sparse regularity lemma (Lemma \[lem:regularitylemma\]) and the sparse blow up lemma (Lemma \[thm:blowup\]). Both lemmas use the concept of regular pairs. Let $G= (V,E)$ be a graph, ${\varepsilon}, d >0$, and $p \in (0,1]$. Moreover, let $X,Y \subseteq V$ be two disjoint nonempty sets. The $p$-density of the pair $(X,Y)$ is defined as $$d_{G,p}(X,Y) := \frac{e_G(X,Y)}{p\|X\|\|Y\|}.$$ For most of this paper, when we work with random graphs, we will be interested in the regularity concept called lower-regularity. When we work with bijumbled graphs, on the other hand, we will need the stronger concept regularity. The difference is that in the former we impose only lower bounds on $p$-densities, whereas in the latter we impose in addition upper bounds. The main reason for this difference is that our ‘regularity inheritance lemmas’ below have different requirements in random and in bijumbled graphs; we do not otherwise make use of the extra strength of ‘regular’ as opposed to ‘lower-regular’. We also need to define super-regularity, for which we require $G$ to be a subgraph of a graph $\Gamma$, which will be the random or bijumbled graph whose resilience properties we are establishing. \[def:regular\] The pair $(X,Y)$ is called $({\varepsilon},d,p)_G$-lower-regular if for every $X'\subseteq X$ and $Y'\subseteq Y$ with $\|X'\|\geq {\varepsilon}\|X\|$ and $\|Y'\|\geq {\varepsilon}\|Y\|$ we have $d_{G,p}(X',Y') \geq d- {\varepsilon}$. It is called $({\varepsilon},d,p)_G$-regular if there exists $d'\ge d$ such that for every $X'\subseteq X$ and $Y'\subseteq Y$ with $\|X'\|\geq {\varepsilon}\|X\|$ and $\|Y'\|\geq {\varepsilon}\|Y\|$ we have $d_{G,p}(X',Y') = d'\pm {\varepsilon}$. If $(X,Y)$ is either $({\varepsilon},d,p)_G$-lower-regular or $({\varepsilon},d,p)_G$-regular, and in addition we have $$\begin{aligned} \|N_G(x,Y)\| &\geq (d-{\varepsilon})\max\big(p\|Y\|,\deg_\Gamma(x,Y)/2\big)\quad\text{and}\\ \|N_G(y,X)\| &\geq (d-{\varepsilon})\max\big(p\|X\|,\deg_\Gamma(y,X)/2\big)\end{aligned}$$ for every $x \in X$ and $y \in Y$, then the pair $(X,Y)$ is called $({\varepsilon},d,p)_G$-super-regular. When we use super-regularity it will be clear from the context whether $(X,Y)$ is lower-regular or regular. Note that a regular pair is by definition lower-regular, though the converse does not hold. Furthermore, although the definition of super-regularity of $G$ contains a reference to $\Gamma$, at each place in this paper where we use super-regularity, we will see that the first term in the maximum is larger than the second. When it is clear from the context, we may omit the subscript $G$ in $({\varepsilon},d,p)_G$-regular which is used to indicate with respect to which graph a pair is regular. A direct consequence of the definition of $({\varepsilon},d,p)$-lower-regular pairs is the following proposition about the sizes of neighbourhoods in lower-regular pairs. \[prop:neighbourhood\] Let $(X,Y)$ be $({\varepsilon}, d,p)$-lower-regular. Then there are less than ${\varepsilon}\|X\|$ vertices $x\in X$ with $\|N(x,Y)\| < (d-{\varepsilon})p\|Y\|$. The next proposition asserts that small alterations of the vertex sets of an $({\varepsilon},d,p)$-(lower-)regular pair do not destroy (lower-)regularity. \[prop:subpairs3\] Let $(X,Y)$ be an $({\varepsilon},d,p)$-lower-regular pair in a graph $G$ and let $\hat{X}$ and $\hat Y$ be two subsets of $V(G)$ such that $\|X{\triangle}\hat{X}\| \leq \mu \|X\|$ and $\|Y {\triangle}\hat Y\| \leq \nu \|Y\|$ for some $0 \leq \mu, \nu \leq 1$. Then $(\hat X, \hat Y)$ is $(\hat {\varepsilon}, d, p)$-lower-regular, where $\hat {\varepsilon}:= {\varepsilon}+ 2\sqrt{\mu} + 2 \sqrt{\nu}$. Furthermore, if for any disjoint $A,A'\subset V(G)$ with $\|A\|\ge\mu\|X\|$ and $\|A'\|\ge\nu\|Y\|$ we have $e(A,A')\le (1+\mu+\nu)p\|A\|\|A'\|$, and $(X,Y)$ is $({\varepsilon},d,p)$-regular, then $(\hat X, \hat Y)$ is $(\hat {\varepsilon}, d, p)$-regular. We defer the proof of this to Appendix \[app:tools\]. In order to state the sparse regularity lemma, we need some more definitions. A partition ${\mathcal V}= \{V_i\}_{i\in\{0,\ldots,r\}}$ of the vertex set of $G$ is called an $({\varepsilon},p)_G$-regular partition of $V(G)$ if $\|V_0\|\leq {\varepsilon}\|V(G)\|$ and $(V_i,V_{i'})$ forms an $({\varepsilon},0,p)_G$-regular pair for all but at most ${\varepsilon}\binom{r}{2}$ pairs $\{i,i'\}\in \binom{[r]}{2}$. It is called an equipartition if $\|V_i\| = \|V_{i'}\|$ for every $i,i'\in[r]$. The partition ${\mathcal V}$ is called $({\varepsilon},d,p)$-(lower-)regular on a graph $R$ with vertex set $[r]$ if $(V_i, V_{i'})$ is $({\varepsilon},d,p)_G$-(lower-)regular for every $\{i,i'\} \in E(R)$. The graph $R$ is referred to as the $({\varepsilon},d,p)_G$-reduced graph of ${\mathcal V}$, the partition classes $V_i$ with $i \in [r]$ as clusters, and $V_0$ as the exceptional set. We also say that ${\mathcal V}$ is $({\varepsilon},d,p)_G$-super-regular on a graph $R'$ with vertex set $[r]$ if $(V_i, V_{i'})$ is $({\varepsilon},d,p)_G$-super-regular for every $\{i,i'\}\in E(R')$. Again, when we talk about reduced graphs or super-regularity, whether we are using lower-regularity or regularity will be clear from the context. We will however always specify whether a partition is regular or only lower-regular on $R$. Analogously to Szemeredi’s regularity lemma for dense graphs, the sparse regularity lemma, proved by Kohayakawa and Rödl [@kohayakawa1997; @kohayakawa2003], asserts the existence of an $({\varepsilon},p)$-regular partition of constant size of any sparse graph. We state a minimum degree version of this lemma, whose proof (following [@bottcher2013almost]) we defer to Appendix \[app:tools\]. \[lem:regularitylemma\] For each ${\varepsilon}>0$, each $\alpha \in [0,1]$, and $r_0\geq 1$ there exists $r_1\geq 1$ with the following property. For any $d\in[0,1]$, any $p>0$, and any $n$-vertex graph $G$ with minimum degree $\alpha p n$ such that for any disjoint $X,Y\subset V(G)$ with $\|X\|,\|Y\|\ge\tfrac{{\varepsilon}n}{r_1}$ we have $e(X,Y)\le \big(1+\tfrac{1}{1000}{\varepsilon}^2\big)p\|X\|\|Y\|$, there is an $({\varepsilon},p)_G$-regular equipartition of $V(G)$ with $({\varepsilon},d,p)_G$-reduced graph $R$ satisfying $\delta(R) \geq (\alpha-d-{\varepsilon})\|V(R)\|$ and $r_0 \leq \|V(R)\| \leq r_1$. A key ingredient in the proof of our main theorem is the so-called sparse blow up lemma developed by H[à]{}n, Kohayakawa, Person, and two of the current authors in [@blowup]. Given a subgraph $G \subseteq \Gamma =G(n,p)$ with $p \gg (\log n/n)^{1/\Delta}$ and an $n$-vertex graph $H$ with maximum degree at most $\Delta$ with vertex partitions ${\mathcal V}$ and ${\mathcal W}$, respectively, the sparse blow up lemma guarantees under certain conditions a spanning embedding of $H$ in $G$ which respects the given partitions. In order to state this lemma we need to introduce some definitions. \[def:buffer\] Let $R'$ be a graph on $r$ vertices and let $H$ be a graph with vertex partition ${\mathcal W}=\{W_i\}_{i\in[r]}$. We say that the family ${\widetilde{\mathcal{W}}}=\{{\widetilde{W}}_i\}_{i\in[r]}$ of subsets ${\widetilde{W}}_i\subseteq W_i$ is an $(\vartheta,R')$-buffer for $H$ if 1. $\|{\widetilde{W}}_i\|\geq\vartheta \|W_i\|$ for all $i\in[r]$, and 2. for each $i\in[r]$ and each $x\in{\widetilde{W}}_i$, the first and second neighbourhood of $x$ go along $R'$, i.e., for each $\{x,y\},\{y,z\}\in E(H)$ with $y\in W_j$ and $z\in W_k$ we have $\{i,j\}\in E(R')$ and $\{j,k\}\in E(R')$. Let $G$ and $H$ be graphs on $n$ vertices with partitions ${\mathcal V}=\{V_i\}_{i\in[r]}$ of $V(G)$ and ${\mathcal W}=\{W_i\}_{i\in[r]}$ of $V(H)$. We say that ${\mathcal V}$ and ${\mathcal W}$ are size-compatible if $\|V_i\|=\|W_i\|$ for all $i\in[r]$. If there exists an integer $m \geq 1$ such that $m \leq \|V_i\| \leq \kappa m$ for every $i\in [r]$, then we say that $(G,{\mathcal V})$ is $\kappa$-balanced. Given a graph $R$ on $r$ vertices, we call $(G, {\mathcal V})$ an $R$-partition if for every edge $\{x,y\}\in E(G)$ with $x \in V_i$ and $y\in V_{i'}$ we have $\{i,i'\}\in E(R)$. We will actually need a little more than just an embedding of $H$ into $G$ respecting given partitions: we will need to restrict the images of some vertices of $H$ to subsets of the clusters of $G$. The following definition encapsulates the properties we have to guarantee for the sparse blow-up lemma to obtain such an embedding. \[def:restrict\] Let ${\varepsilon},d>0$, $p \in [0,1]$, and let $R$ be a graph on $r$ vertices. Furthermore, let $G$ be a (not necessarily spanning) subgraph of $\Gamma = G(n,p)$ and let $H$ be a graph given with vertex partitions ${\mathcal V}= \{V_i\}_{i\in[r]}$ and ${\mathcal W}= \{W_i\}_{i\in[r]}$, respectively, such that $(G,{\mathcal V})$ and $(H,{\mathcal W})$ are size-compatible $R$-partitions. Let ${\mathcal I}=\{I_x\}_{x\in V(H)}$ be a collection of subsets of $V(G)$, called image restrictions, and ${\mathcal J}=\{J_x\}_{x\in V(H)}$ be a collection of subsets of $V(\Gamma)\setminus V(G)$, called restricting vertices. For each $i\in [r]$ we define $R_i\subseteq W_i$ to be the set of all vertices $x \in W_i$ for which $I_x \neq V_i$. We say that ${\mathcal I}$ and ${\mathcal J}$ are a $(\rho,\zeta,\Delta,\Delta_J)$-restriction pair if the following properties hold for each $i\in[r]$ and $x\in W_i$. 1. \[itm:restrict:numres\] We have $\|R_i\|\leq\rho\|W_i\|$. 2. \[itm:restrict:sizeIx\] If $x\in R_i$, then $I_x\subseteq \bigcap_{u\in J_x} N_\Gamma(u, V_i)$ is of size at least $\zeta(dp)^{\|J_x\|}\|V_i\|$. 3. \[itm:restrict:Jx\] If $x\in R_i$, then $\|J_x\|+\deg_H(x)\leq\Delta$ and if $x\in W_i\setminus R_i$, then $J_x=\varnothing$. 4. \[itm:restrict:DJ\] Each vertex in $V(G)$ appears in at most $\Delta_J$ of the sets of ${\mathcal J}$. 5. \[itm:restrict:sizeGa\] We have $\big\|\bigcap_{u\in J_x} N_\Gamma(u, V_i)\big\| = (p\pm{\varepsilon}p)^{\|J_x\|}\|V_i\|$. 6. \[itm:restrict:Ireg\] If $x\in R_i$, for each $xy\in E(H)$ with $y\in W_j$, $$\text{the pair }\quad\Big( V_i \cap \bigcap_{u\in J_x}N_\Gamma(u), V_j \cap \bigcap_{v\in J_y}N_\Gamma(v)\Big)\quad\text{ is $({\varepsilon},d,p)_G$-lower-regular.}$$ Suppose ${\mathcal V}$ is an $({\varepsilon},d,p)_G$-lower-regular partition of $V(G)$ with reduced graph $R$, and let $R'$ be a subgraph of $R$. We say $(G,{\mathcal V})$ has one-sided inheritance on $R'$ if for every $\{i,j\}, \{j,k\}\in E(R')$ and every $v\in V_i$ the pair $\big(N_\Gamma(v, V_j),V_k\big)$ is $({\varepsilon},d,p)_G$-lower-regular. Given a $(\vartheta,R')$-buffer ${\widetilde{\mathcal{W}}}$, we say that $(G,{\mathcal V})$ has two-sided inheritance on $R'$ for ${\widetilde{\mathcal{W}}}$ if whenever there is a triangle $w_iw_jw_k\in H$ with $w_i\in{\widetilde{W}}_i$, $w_j\in W_j$ and $w_k\in W_k$, it follows that for every $v\in V_i$ the pair $\big(N_\Gamma(v, V_j),N_\Gamma(v, V_k)\big)$ is $({\varepsilon},d,p)_G$-lower-regular. Now we can finally state the sparse blow up lemma. \[thm:blowup\] For each $\Delta$, $\Delta_{R'}$, $\Delta_J$, $\vartheta,\zeta, d>0$, $\kappa>1$ there exist ${{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{BL}$}}},\rho>0$ such that for all $r_1$ there is a ${C_{\scalebox{{0.5}}{$\mathrm{BL}$}}}$ such that for $p\geq{C_{\scalebox{{0.5}}{$\mathrm{BL}$}}}(\log n/n)^{1/\Delta}$ the random graph $\Gamma=G_{n,p}$ asymptotically almost surely satisfies the following. Let $R$ be a graph on $r\le r_1$ vertices and let $R'\subseteq R$ be a spanning subgraph with $\Delta(R')\leq \Delta_{R'}$. Let $H$ and $G\subseteq \Gamma$ be graphs given with $\kappa$-balanced, size-compatible vertex partitions ${\mathcal W}=\{W_i\}_{i\in[r]}$ and ${\mathcal V}=\{V_i\}_{i\in[r]}$ with parts of size at least $m\geq n/(\kappa r_1)$. Let ${\mathcal I}=\{I_x\}_{x\in V(H)}$ be a family of image restrictions, and ${\mathcal J}=\{J_x\}_{x\in V(H)}$ be a family of restricting vertices. Suppose that 1. \[itm:blowup:H\] $\Delta(H)\leq \Delta$, for every edge $\{x,y\}\in E(H)$ with $x\in W_i$ and $y\in W_j$ we have $\{i,j\}\in E(R)$ and ${\widetilde{\mathcal{W}}}=\{{\widetilde{W}}_i\}_{i\in[r]}$ is an $(\vartheta,R')$-buffer for $H$, 2. \[itm:blowup:G\] ${\mathcal V}$ is $({{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{BL}$}}},d,p)_G$-lower-regular on $R$, $({{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{BL}$}}},d,p)_G$-super-regular on $R'$, has one-sided inheritance on $R'$, and two-sided inheritance on $R'$ for ${\widetilde{\mathcal{W}}}$, 3. \[itm:blowup:restrict\] ${\mathcal I}$ and ${\mathcal J}$ form a $(\rho,\zeta,\Delta,\Delta_J)$-restriction pair. Then there is an embedding $\phi\colon V(H)\to V(G)$ such that $\phi(x)\in I_x$ for each $x\in H$. Observe that in the blow up lemma for dense graphs, proved by Koml[ó]{}s, S[á]{}rk[ö]{}zy, and Szemer[é]{}di [@komlos1997blow], one does not need to explicitly ask for one- and two-sided inheritance properties since they are always fulfilled by dense regular partitions. This is, however, not true in general in the sparse setting. The following two lemmas will be very useful whenever we need to choose vertices whose neighbourhoods inherit lower-regularity. \[lem:OSRIL\] For each ${{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}}, {\alpha_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}}>0$ there exist ${\varepsilon}_0 >0$ and $C >0$ such that for any $0 < {\varepsilon}< {\varepsilon}_0$ and $0 < p <1$ asymptotically almost surely $\Gamma= G(n,p)$ has the following property. For any disjoint sets $X$ and $Y$ in $V(\Gamma)$ with $\|X\|\geq C\max\big(p^{-2}, p^{-1} \log n\big)$ and $\|Y\| \geq C p^{-1} \log n$, and any subgraph $G$ of $\Gamma[X,Y]$ which is $({\varepsilon}, {\alpha_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}},p)_G$-lower-regular, there are at most $C p^{-1}\log (en/\|X\|)$ vertices $z \in V(\Gamma)$ such that $(X \cap N_{\Gamma}(z),Y)$ is not $({{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}},{\alpha_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}},p)_G$-lower-regular. \[lem:TSRIL\] For each ${{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}},{\alpha_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}}>0$ there exist ${\varepsilon}_0>0$ and $C >0$ such that for any $0<{\varepsilon}<{\varepsilon}_0$ and $0<p<1$, asymptotically almost surely $\Gamma=G_{n,p}$ has the following property. For any disjoint sets $X$ and $Y$ in $V(\Gamma)$ with $\|X\|,\|Y\|\ge C\max\{p^{-2},p^{-1}\log n\}$, and any subgraph $G$ of $\Gamma[X,Y]$ which is $({\varepsilon},{\alpha_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}},p)_G$-lower-regular, there are at most $C\max\{p^{-2},p^{-1}\log (en/\|X\|)\}$ vertices $z \in V(\Gamma)$ such that $\big(X\cap N_\Gamma(z),Y\cap N_\Gamma(z)\big)$ is not $({{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}},{\alpha_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}},p)_G$-lower-regular. We close this section with two Chernoff bounds for random variables that follow a binomial (Theorem \[thm:chernoff\]) and a hypergeometric distribution (Theorem \[thm:hypergeometric\]), respectively, and the following useful observation. Roughly speaking, it states that a.a.s. nearly all vertices in $G(n,p)$ have approximately the expected number of neighbours within large enough subsets. \[prop:chernoff\] For each ${\varepsilon}>0$ there exists a constant $C >0$ such that for every $0<p<1$ asymptotically almost surely $\Gamma=G(n,p)$ has the following properties. For any disjoint $X,Y\subset V(\Gamma)$ with $\|X\|\ge Cp^{-1}\log n$ and $\|Y\|\ge Cp^{-1}\log (en/\|X\|)$, we have $e(X,Y)=(1\pm{\varepsilon})p\|X\|\|Y\|$ and $e(X)\le 2p\|X\|^2$. Furthermore, for every $X \subseteq V(\Gamma)$ with $\|X\| \geq C p^{-1} \log n$, the number of vertices $v \in V(\Gamma)$ with $\big\|\|{N_{\Gamma}}(v,X)\| - p \|X\|\big\| > {\varepsilon}p \|X\|$ is at most $C p^{-1} \log (en/\|X\|)$. Note that in most of this paper we will use the upper bound $\log(en/\|X\|)\le\log n$ when applying this proposition, and Lemmas \[lem:OSRIL\] and \[lem:TSRIL\], valid since (in all applications) we have $\|X\|\ge e$. We will only need the full strength of these three results when proving the Lemma for $G$ (Lemma \[lem:G\]). In the proof of Proposition \[prop:chernoff\] we use the following version of Chernoff’s Inequalities (see e.g. [@janson2011random Chapter 2] for a proof). \[thm:chernoff\] Let $X$ be a random variable which is the sum of independent Bernoulli random variables. Then we have for ${\varepsilon}\leq 3/2$ $${\mathbb{P}}\big[\|X-{\mathbb{E}}[X]\| > {\varepsilon}{\mathbb{E}}[X]\big] < 2e^{-{\varepsilon}^2{\mathbb{E}}[X]/3}\,.$$ Furthermore, if $t\ge 6{\mathbb{E}}[X]$ then we have $${\mathbb{P}}\big[X\ge{\mathbb{E}}[X]+t\big]\le e^{-t}\,.$$ Since the statement of the proposition is stronger when ${\varepsilon}$ is smaller, we may assume that $0<{\varepsilon}\le 1$. We set $C'=100{\varepsilon}^{-2}$ and $C=1000C'{\varepsilon}^{-1}$. We first show that $\Gamma=G(n,p)$ a.a.s. has the following two properties. For any disjoint $A,B\subset V(\Gamma)$, with $\|A\|\ge C'p^{-1}\log n$ and $\|B\|\ge C'p^{-1}\log(en/\|A\|)$, we have $e(A,B)=\big(1\pm\tfrac{{\varepsilon}}{2}\big)p\|A\|\|B\|$. For any $A\subset V(\Gamma)$, we have $e(A)\le 4p\|A\|^2+2\|A\|\log n$, and if $\|A\|\ge C'p^{-1}\log n$ then $e(A)\le 2p\|A\|^2$. Note that these properties imply the first two conclusions of the proposition. We estimate the failure probability of the first property using Theorem \[thm:chernoff\] and the union bound. Assuming without loss of generality that $\|A\|\ge\|B\|$, this probability is at most $$\begin{aligned} \sum_{\|A\|,\|B\|\le n} \binom{n}{\|A\|}^2\cdot 2e^{-{\varepsilon}^2p\|A\|\|B\|/12}&\le 2n\sum_{\|A\|}\big(\tfrac{en}{\|A\|}\big)^{2\|A\|}e^{-{\varepsilon}^2C'\|A\|\log (en/\|A\|)/12}\\ &<2n\sum_{\|A\|}\big(\tfrac{en}{\|A\|}\big)^{-2\|A\|}\,.\end{aligned}$$ For the second property, observe that $4p\|A\|^2>7p\binom{\|A\|}{2}$, so that for any given $A$ by Theorem \[thm:chernoff\] we have $${\mathbb{P}}\big[e(A)\ge 4p\|A\|^2+2\|A\|\log n\big]\le e^{-2\|A\|\log n}=n^{-2\|A\|}\,.$$ Taking a union bound over the at most $n^{\|A\|}$ choices of $A$ given $\|A\|$, we see that the failure probability of the second property is at most $\sum_{a=1}^nn^{-a}$. Finally, the failure probability of the last property is at most $$\sum_{\|A\|\ge C'p^{-1}\log n}n^{\|A\|}\cdot 2e^{-p\binom{\|A\|}{2}/3}\le\sum_{\|A\|}2n^{\|A\|}e^{-C'\|A\|\log n/12}\le 2n^{-2}\,,$$ and since all three failure probabilities tend to zero as $n\to\infty$, we conclude that a.a.s. $G(n,p)$ enjoys both properties. Now suppose $\Gamma$ has these properties, and let $X\subset V(\Gamma)$ have size at least $Cp^{-1}\log n$. We first show that there are at most $C'p^{-1}\log (en/\|X\|)$ vertices in $\Gamma$ which have less than $(1-{\varepsilon})p\|X\|$ neighbours in $X$. If this were false, then we could choose a set $Y$ of $C'p^{-1}\log (en/\|X\|)$ vertices in $\Gamma$ which have less than $(1-{\varepsilon})p\|X\|$ neighbours in $X$. By choice of $C$ and since $\|X\|>e$, we have $(1-{\varepsilon})p\|X\|\le \big(1-\tfrac{\varepsilon}2\big)p\|X\setminus Y\|$, so we see that $e(Y,X\setminus Y)<\big(1-\tfrac{{\varepsilon}}{2}\big)p\|Y\|\|X\setminus Y\|$. This is a contradiction since $\|X\setminus Y\|\ge C'p^{-1}\log n$. Next we show that there are at most $2C'p^{-1}\log (en/\|X\|)$ vertices of $\Gamma$ which have more than $(1+{\varepsilon})p\|X\|$ neighbours in $X$. Again, if this is not the case we can let $Y$ be a set of $2C'p^{-1}\log (en/\|X\|)$ vertices of $\Gamma$ with more than $(1+{\varepsilon})p\|X\|$ neighbours in $X$. Now $e(Y)\le 4p\|Y\|^2+2\|Y\|\log n=8C'\|Y\|\log(en/\|X\|)+2\|Y\|\log n\le 10C'\|Y\|\log n$, so there are at most $\|Y\|/2$ vertices in $Y$ which have $40C'\log n$ or more neighbours in $Y$. Let $Y'\subset Y$ consist of those vertices with at most $40C'\log n$ neighbours in $Y$. For each $v\in Y'$ we have $$(1+{\varepsilon})p\|X\|\le\deg(v;X)\le \deg(v;Y)+\deg(v;X\setminus Y)\,,$$ and so, by choice of $C$, each vertex of $Y'$ has at least $\big(1+\tfrac{\varepsilon}2\big)p\|X\setminus Y\|$ neighbours in $X\setminus Y$. Since $\|Y'\|\ge C'p^{-1}\log(2en/\|X\|)$ and $\|X\setminus Y\|\ge \|X\|/2\ge C'p^{-1}\log n$, this is a contradiction. Finally, since by choice of $C$ we have $3C'p^{-1}\log n<Cp^{-1}\log n$ we conclude that all but at most $Cp^{-1}\log (en/\|X\|)$ vertices of $\Gamma$ have $(1\pm{\varepsilon})p\|X\|$ neighbours in $X$, as desired. Finally, let $N$, $m$, and $s$ be positive integers and let $S$ and $S' \subseteq S$ be two sets with $\|S\| = N$ and $\|S'\| = m$. The hypergeometric distribution is the distribution of the random variable $X$ that is defined by drawing $s$ elements of $S$ without replacement and counting how many of them belong to $S'$. It can be shown that Theorem \[thm:chernoff\] still holds in the case of hypergeometric distributions (see e.g. [@janson2011random], Chapter 2 for a proof) with ${\mathbb{E}}[X]:= ms/N$. \[thm:hypergeometric\] Let $X$ be a random variable that follows the hypergeometric distribution with parameters $N$, $m$, and $s$. Then for any ${\varepsilon}>0$ and $t\ge{\varepsilon}ms/N$ we have $${\mathbb{P}}\big[\|X - ms/N\| > t \big] < 2e^{-{\varepsilon}^2t/3}\,.$$ We require the following technical lemma, which is a consequence of the hypergeometric inequality stated in Theorem \[thm:hypergeometric\]. \[lem:hypgeo\] For each $\eta>0$ and $\Delta$ there exists $C$ such that the following holds. Let $W\subset [n]$, let $t\le 100n^\Delta$, and let $T_1,\ldots,T_t$ be subsets of $W$. For each $m\le \|X\|$ there is a set $S\subset W$ of size $m$ such that $$\|T_i\cap S\|=\tfrac{m}{\|W\|}\|T_i\|\pm \big(\eta\|T_i\|+C\log n\big)\text{ for every }i\in[t]\,.$$ Set $C=30\eta^{-2}\Delta$. Observe that for each $i$, the size of $T_i\cap S$ is hypergeometrically distributed. By Theorem \[thm:hypergeometric\], for each $i$ we have $${\mathbb{P}}\big[\|T_i\cap S\|\neq \tfrac{m}{\|W\|}\|T_i\|\pm \big(\eta\|T_i\|+C\log n\big)\big]<2e^{-\eta^2C\log n/3}<\frac{2}{n^{1+\Delta}}\,,$$ so taking the union bound over all $i\in[t]$ we conclude that the probability of failure is at most $2t/n^{1+\Delta}\le 200/n\to 0$ as $n\to\infty$, as desired. Proof overview and main lemmas {#sec:mainlemmas} ============================== Theorem \[thm:main\] is a corollary of the following more general Theorem \[thm:maink\], which we prove in Section \[sec:proofmain\]. We require one preliminary definition. \[def:zerofree\] Let $H$ be a $(k+1)$-colourable graph on $n$ vertices and let ${\mathcal L}$ be a labelling of its vertex set of bandwidth at most $\beta n$. A proper $(k+1)$- colouring $\sigma:V(H) \to \{0,\ldots,k\}$ of its vertex set is said to be $(z,\beta)$-zero-free with respect to ${\mathcal L}$ if any $z$ consecutive blocks contain at most one block with colour zero, where a block is defined as a set of the form $\{(t-1)4k\beta n +1, \ldots, t4k\beta n\}$ with some $t \in [1/(4k\beta)]$. \[thm:maink\] For each $\gamma>0$, $\Delta \geq 2$, and $k\geq 2$, there exist constants $\beta >0$, $z>0$, and $C>0$ such that the following holds asymptotically almost surely for $\Gamma = G(n,p)$ if $p\geq C\left(\frac{\log n}{n}\right)^{1/\Delta}$. Let $G$ be a spanning subgraph of $\Gamma$ with $\delta(G) \geq\left(\frac{k-1}{k}+\gamma\right)pn$ and let $H$ be a graph on $n$ vertices with $\Delta(H) \leq \Delta$ that has a labelling ${\mathcal L}$ of its vertex set of bandwidth at most $\beta n$, a $(k+1)$-colouring that is $(z,\beta)$-zero-free with respect to ${\mathcal L}$ and where the first $\sqrt{\beta} n$ vertices in ${\mathcal L}$ are not given colour zero and the first $\beta n$ vertices in ${\mathcal L}$ include $C p^{-2}$ vertices that are not contained in any triangles of $H$. Then $G$ contains a copy of $H$. Proof overview {#subsec:over} -------------- We now give a brief sketch of the proof of Theorem \[thm:maink\]. Ultimately, our goal is to apply the sparse blow-up lemma, Lemma \[thm:blowup\], to find an embedding of $H$ into $G$. Thus, the proof boils down to obtaining the required conditions. But there is a catch: this is not as such possible, as for any lower-regular partition of $G$ there can be some exceptional vertices which are ‘badly behaved’ with respect to the partition. These vertices will never satisfy the conditions of the sparse blow-up lemma, and we will have to deal with them beforehand. We will do this by ‘pre-embedding’ some vertices of $H$ to cover the exceptional vertices, and then apply the sparse blow-up lemma to complete the embedding of $H$ into $G$, using image restrictions to ensure we really obtain an embedding of $H$. Let us now fill in a few more details. We start by obtaining, in the lemma for $G$, Lemma \[lem:G\], a lower-regular partition of $G$ into parts $V_0$ and $V_{i,j}$ for $i\in[r]$ (where $r$ may be large but is bounded above by a constant) and $j\in [k]$ with several extra properties. The most important properties are that $\|V_0\|=O\big(p^{-2}\big)$, that the corresponding reduced graph, which we call $R^k_r$, has high minimum degree and contains a backbone graph, that is, contains the edge $(i,j),(i',j')$ whenever $\|i=i'\|\le 1$ and $j\neq j'$, that the lower-regular pairs $(V_{i,j},V_{i,j'})$ are not just lower-regular but super-regular, and that all vertices outside $V_0$ have inheritance properties with respect to all lower-regular pairs. In short, if the exceptional vertices $V_0$ did not exist, this partition, together with a corresponding partition of $V(H)$, would be what we need to apply the sparse blow-up lemma. Passing over for now the inconvenient existence of $V_0$, our next task is to find the corresponding partition of $V(H)$, for which we use the lemma for $H$, Lemma \[lem:H2\]. One should think of the backbone graph as consisting of copies of $K_k$ (one for each $i\in[r]$) connected in a linear order; and the high minimum degree of $R^k_r$ ensures that each $K_k$ extends to $K_{k+1}$ in $R^k_r$. The basic idea is then to split $H$ into intervals in the bandwidth order. We assign the first interval to the first $K_k$ of the backbone graph according to the given colouring of $H$, with the few vertices of colour zero assigned to a vertex extending this clique of the backbone graph to $K_{k+1}$, and so on. Using the bandwidth property and zero-freeness of the colouring one can do this in such a way as to obtain a graph homomorphism from $H$ to $R^k_r$, which is what we need. In addition, we need the number of vertices assigned to each $(i,j)\in V(R^k_r)$ to be very close to $\|V_{i,j}\|$. We cannot guarantee exact equality, but we can get very close by making further use of bandwidth, zero-freeness, and the fact that $K_k$s in $R^k_r$ extend to $K_{k+1}$s. Now we have to deal with the exceptional set $V_0$. We do this as follows. We choose a vertex $v$ in the exceptional set, and ‘pre-embed’ to it a vertex $x$ picked from the first $\sqrt{\beta}n$ vertices of ${\mathcal L}$ which is not in any triangle of $H$. Using the common neighbourhood lemma, Lemma \[lem:common\], we choose $\Delta$ neighbours of $v$ which are ‘well-behaved’ with respect to the clusters $V_{i,j}$ for some $i\in[r]$, and pre-embed the neighbours of $x$ to these vertices. The ‘well-behaved’ properties are what we need to generate image restrictions for the second neighbours of $x$ (which we will embed using the sparse blow-up lemma) satisfying the restriction pair properties. We also need to change the assignment from the Lemma for $H$ locally (up to a large but constant distance from $x$) to accommodate this: the vertex $x$, and its first and second neighbours, might have been assigned somewhere quite different previously. We repeat this until we have pre-embedded to all exceptional vertices, and let $H'$ and $G'$ be respectively the unembedded vertices of $H$ and the vertices of $G$ to which we did not pre-embed. At this point we have all the conditions we need to apply the sparse blow-up lemma to complete the embedding, except that the partitions of $H'$ and $G'$ we have do not quite have parts of matching sizes. We use the balancing lemma, Lemma \[lem:balancing\], to deal with this. The idea is simple: we take some carefully selected vertices in clusters of $G$ which are too big (compared to the assigned part of $H$) and move them to other clusters, first in order to make sure that the total number of vertices in $\bigcup_i V_{i,j}$ is correct for each $j$ (using the high minimum degree of $R^k_r$) and then (using the structure of the backbone graph) to give each cluster the correct size. At last, applying the sparse blow-up lemma, Lemma \[thm:blowup\], we complete the embedding of $H$ into $G$. We note that this proof sketch glosses over some subtleties. In particular, at the two places where ‘we choose’ vertices onto which to pre-embed, we have to be quite careful to choose vertices correctly so that this strategy can be completed and we do not destroy good properties obtained earlier. We will return to this point immediately before the proof of Theorem \[thm:maink\] in Section \[sec:proofmain\] to explain how we do this. Main lemmas ----------- In this subsection we formulate the four main lemmas that we use in the proof of Theorem \[thm:maink\] mentioned in the above overview. We defer the proofs of these lemmas to later sections. Before stating these lemmas, we need some more definitions. Let $r, k \geq 1$ and let $B^k_r$ be the backbone graph on $kr$ vertices. That is, we have $$V(B^k_r) := [r] \times [k]$$ and for every $j \neq j' \in [k]$ we have $\{(i,j),(i',j')\} \in E(B^k_r)$ if and only if $\|i-i'\|\le1$. Let $K^k_r \subseteq B^k_r$ be the spanning subgraph of $B^k_r$ that is the disjoint union of $r$ complete graphs on $k$ vertices given by the following components: the complete graph $K^k_r[\{(i,1),\ldots, (i,k)\}]$ is called the $i$-th component of $K^k_r$ for each $i\in [r]$. A vertex partition ${\mathcal V}' = \{V_{i,j}\}_{i\in[r],j\in[k]}$ is called $k$-equitable if $\big\|\|V_{i,j}\| -\|V_{i,j'}\|\big\|\leq 1$ for every $i\in [r]$ and $j,j'\in[k]$. Similarly, an integer partition $\{n_{i,j}\}_{i\in[r],j\in[k]}$ of $n$ (meaning that $n_{i,j} \in \mathbb Z_{\geq 0}$ for every $i\in [r],j\in[k]$ and $\sum_{i\in[r]j\in[k]} n_{i,j} = n$) is $k$-equitable if $\|n_{i,j}-n_{i,j'}\| \leq 1$ for every $i\in[r]$ and $j,j'\in[k]$. The lemma for $G$ says that a.a.s. $\Gamma = G(n,p)$ satisfies the following property if $p \gg (\log n/n)^{1/2}$. For any spanning subgraph $G\subset\Gamma$ with minimum degree a sufficiently large fraction of $pn$, there exists an $({\varepsilon},d,p)_G$-lower-regular vertex partition ${\mathcal V}$ of $V(G)$ whose reduced graph $R^k_r$ contains a clique factor $K^k_r$ on which the corresponding vertex sets of ${\mathcal V}$ are pairwise $({\varepsilon},d,p)$-super-regular. Furthermore, $(G,{\mathcal V})$ has one-sided and two-sided inheritance with respect to $R^k_r$, and the $\Gamma$-neighbourhoods of all vertices but the ones in the exceptional set of ${\mathcal V}$ have almost exactly their expected size in each cluster. The proof of Lemma \[lem:G\] is given in Section \[sec:prooflemG\]. \[lem:G\] For each $\gamma > 0$ and integers $k \geq 2$ and $r_0 \geq 1$ there exists $d > 0$ such that for every ${\varepsilon}\in \left(0, \frac{1}{2k}\right)$ there exist $r_1\geq 1$ and ${C^{\ast}}>0$ such that the following holds a.a.s. for $\Gamma = G(n,p)$ if $p \geq {C^{\ast}}\left(\log n/n\right)^{1/2}$. Let $G=(V,E)$ be a spanning subgraph of $\Gamma$ with $\delta(G) \geq \left(\frac{k-1}{k} + \gamma\right)pn$. Then there exists an integer $r$ with $r_0\leq kr \leq r_1$, a subset $V_0 \subseteq V$ with $\|V_0\| \leq {C^{\ast}}p^{-2}$, a $k$-equitable vertex partition ${\mathcal V}= \{{V_{i,j}}\}_{i\in[r],j\in[k]}$ of $V(G)\setminus V_0$, and a graph $R^k_r$ on the vertex set $[r] \times [k]$ with $K^k_r \subseteq B^k_r \subseteq R^k_r$, with $\delta(R^k_r) \geq \left(\frac{k-1}{k} + \frac{\gamma}{2}\right)kr$, and such that the following is true. 1. \[lemG:size\] $\frac{n}{4kr}\leq \|{V_{i,j}}\| \leq \frac{4n}{kr}$ for every $i\in[r]$ and $j\in[k]$, 2. \[lemG:regular\] ${\mathcal V}$ is $({\varepsilon},d,p)_G$-lower-regular on $R^k_r$ and $({\varepsilon},d,p)_G$-super-regular on $K^k_r$, 3. \[lemG:inheritance\] both $\big({N_{\Gamma}}(v, V_{i,j}),V_{i',j'}\big)$ and $\big({N_{\Gamma}}(v', V_{i,j}),{N_{\Gamma}}(v, V_{i',j'})\big)$ are $({\varepsilon},d,p)_G$-lower-regular pairs for every $\{(i,j),(i',j')\} \in E(R^k_r)$ and $v\in V\setminus V_0$, 4. \[lemG:gamma\] $\|{N_{\Gamma}}(v,V_{i,j})\| = (1 \pm {\varepsilon})p\|V_{i,j}\|$ for every $i \in [r]$, $j\in [k]$ and every $v \in V \setminus V_0$. Furthermore, if we replace \[lemG:inheritance\] with the weaker 1. \[lemG:inheritancep\] $\big({N_{\Gamma}}(v, V_{i,j}),V_{i',j'}\big)$ is an $({\varepsilon},d,p)_G$-lower-regular pair for every $\{(i,j),(i',j')\} \in E(R^k_r)$ and $v\in V\setminus V_0$, then we have the stronger bound $\|V_0\|\le{C^{\ast}}p^{-1}$. After Lemma \[lem:G\] has constructed a lower-regular partition ${\mathcal V}$ of $V(G)$, the second main lemma deals with the graph $H$ that we would like to find as a subgraph of $G$. More precisely, Lemma \[lem:H2\] provides a homomorphism $f$ from the graph $H$ to the reduced graph $R^k_r$ given by Lemma \[lem:G\] which has among others the following properties. The edges of $H$ are mapped to the edges of $R^k_r$, and the vast majority of the edges of $H$ are assigned to edges of the clique factor $K^k_r \subseteq R^k_r$. The number of vertices of $H$ mapped to a vertex of $R^k_r$ only differs slightly from the size of the corresponding cluster of ${\mathcal V}$. The lemma further guarantees that each of the first $\sqrt{\beta}n$ vertices of the bandwidth ordering of $V(H)$ is mapped to $(1,j)$ with $j$ being the colour that the vertex has received by the given colouring of $H$. In case $H$ is $D$-degenerate the next lemma also ensures that for every $(i,j) \in [r] \times [k]$, a constant fraction of vertices mapped to $(i,j)$ have each at most $2D$ neighbours. \[lem:H2\] Given $D, k, r \geq 1$ and $\xi, \beta > 0 $ the following holds if $\xi \leq 1/(kr)$ and $\beta \leq 10^{-10}\xi^2/(D k^4r)$. Let $H$ be a $D$-degenerate graph on $n$ vertices, let $\mathcal L$ be a labelling of its vertex set of bandwidth at most $\beta n$ and let $\sigma: V(H) \to \{0,\ldots k\}$ be a proper $(k+1)$-colouring that is $(10/\xi, \beta)$-zero-free with respect to $\mathcal L$, where the colour zero does not appear in the first $\sqrt{\beta}n$ vertices of $\mathcal{L}$. Furthermore, let $R^k_r$ be a graph on vertex set $[r] \times [k]$ with $K^k_r \subseteq B^k_r \subseteq R^k_r$ such that for every $i\in [r]$ there exists a vertex $z_i \in \big([r]\setminus\{i\}\big) \times [k]$ with $\big\{z_i, (i,j)\big\} \in E(R^k_r)$ for every $j\in [k]$. Then, given a $k$-equitable integer partition $\{m_{i,j}\}_{i\in[r],j\in[k]}$ of $n$ with $n/(10kr) \leq m_{i,j} \leq 10n/(kr)$ for every $i \in[r]$ and $j\in [k]$, there exists a mapping $f \colon V(H) \to [r]\times[k]$ and a set of special vertices $X \subseteq V(H)$ such that we have for every $i\in [r]$ and $j\in[k]$ 1. \[lemH:H1\] $m_{i,j} - \xi n \leq \|f^{-1}(i,j)\| \leq m_{i,j} + \xi n$, 2. \[lemH:H2\] $\|X\| \leq \xi n$, 3. \[lemH:H3\] $\{f(x),f(y)\} \in E(R^k_r)$ for every $\{x,y\} \in E(H)$, 4. \[lemH:H4\] $y,z\in \cup_{j'\in[k]}f^{-1}(i,j')$ for every $x\in f^{-1}(i,j)\setminus X$ and $xy,yz\in E(H)$, 5. \[lemH:H5\] $f(x) = \big(1, \sigma(x)\big)$ for every $x$ in the first $\sqrt{\beta}n$ vertices of $\mathcal{L}$, and 6. \[lemH:H6\] $\|\{x\in f^{-1}(i,j): \deg(x) \leq 2D\}\| \geq \tfrac{1}{24D} \|f^{-1}(i,j)\|$. Lemma \[lem:H2\] is a strengthened version of [@BoeTarWue Lemma 8]. The proof of [@BoeTarWue Lemma 8] is deterministic; here we use a probabilistic argument to show the existence of a function $f$ that also satisfies the additional property \[lemH:H6\], which is required for Theorem \[thm:degenerate\]. However, we still borrow ideas from the proof of [@BoeTarWue Lemma 8]. The proof of Lemma \[lem:H2\] will be given in Section \[sec:lemH\]. During the pre-embeding, we embed a vertex $x$ of $H$ onto a vertex $v$ of $V_0$, and we also embed its neighbours $N_H(x)$. This creates restrictions on the vertices of $G$ to which we can embed the second neighbours, and for application of Lemma \[thm:blowup\] we need certain conditions to be satisfied. The next lemma states that we can find vertices in $N_G(v)$, to which we will embed $N_H(x)$, satisfying these conditions. \[lem:common\] For each $d>0$, $k \geq 2$, and $\Delta \geq 2$ there exists $\alpha >0$ such that for every ${\varepsilon}^\ast \in (0,1)$ there exists ${\varepsilon}_0 >0$ such that for every $r\geq 1$ and every $0<{\varepsilon}\le{\varepsilon}_0$ there exists ${C^{\ast}}>0$ such that the following is true. If $p \geq {C^{\ast}}\left(\log n/n\right)^{1/\Delta}$, then $\Gamma = G(n,p)$ a.a.s. satisfies the following. Let $G=(V,E)$ be a (not necessarily spanning) subgraph of $\Gamma$ and $\{V_i\}_{i\in[k]}\cup \{W\}$ a vertex partition of a subset of $V$ such that the following is true for every $i,i'\in [k]$. 1. \[cnl:bal\] $\frac{n}{4kr}\le \|V_i\|\le \frac{4n}{kr}$, 2. \[cnl:Vreg\] $(V_i,V_{i'})$ is $({\varepsilon}, d, p)_G$-lower-regular, 3. \[cnl:W\] $\|W\|=10^{-10}\frac{{\varepsilon}^4 pn}{k^4r^4}$, and 4. \[cnl:Wdeg\] $\|N_G(w,V_i)\| \geq dp\|V_i\|$ for every $w \in W$. Then there exists a tuple $(w_1, \ldots, w_\Delta) \in \binom{W}{\Delta}$ such that for every $\Lambda,\Lambda^\ast\subseteq[\Delta]$, and every $i \neq i' \in [k]$ we have 1. \[cnl:Gsize\] $\|\bigcap_{j\in \Lambda} N_G(w_j,V_i)\|\geq \alpha p^{\|\Lambda\|}\|V_i\|$, 2. \[cnl:Gasizen\] $\|\bigcap_{j\in \Lambda} N_{\Gamma}(w_j)\| \le (1 + {\varepsilon}^\ast)p^{\|\Lambda\|}n$, 3. \[cnl:Gasize\] $\|\bigcap_{j\in \Lambda} N_{\Gamma}(w_j,V_i)\| = (1 \pm {\varepsilon}^\ast)p^{\|\Lambda\|}\|V_i\|$, and 4. \[cnl:Nreg\] $\big(\bigcap_{j\in \Lambda}{N_{\Gamma}}(w_j,V_i),\bigcap_{j^\ast\in \Lambda^\ast}{N_{\Gamma}}(w_{j^\ast},V_{i'})\big)$ is $({\varepsilon}^\ast, d,p)_G$-lower-regular if $\|\Lambda\|,\|\Lambda^\ast\| < \Delta$ and either $\Lambda\cap\Lambda^\ast=\varnothing$ or $\Delta\geq 3$ or both. Let $H'$ and $G'$ denote the subgraphs of $H$ and $G$ that result from removing all vertices that were used in the pre-embedding process. As a last step before finally applying the sparse blow-up lemma, the clusters in ${{ \left.\kern-\nulldelimiterspace {\mathcal V}\vphantom{\big\|} \right\|_{G'} }}$ need to be adjusted to the sizes of ${{ \left.\kern-\nulldelimiterspace W_{i,j} \vphantom{\big\|} \right\|_{H'} }}$. The next lemma states that this is possible, and that after this redistribution the regularity properties needed for Lemma \[thm:blowup\] still hold. \[lem:balancing\] For all integers $k\geq 1$, $r_1, \Delta \geq 1$, and reals $\gamma, d >0$ and $0 < {\varepsilon}< \min\{d,1/(2k)\}$ there exist $\xi >0$ and ${C^{\ast}}>0$ such that the following is true for every $p \geq {C^{\ast}}\left(\log n/n\right)^{1/2}$ and every $10\gamma^{-1}\le r \leq r_1$ provided that $n$ is large enough. Let $\Gamma$ be a graph on the vertex set $[n]$ and let $G=(V,E)\subseteq \Gamma$ be a (not necessarily spanning) subgraph with vertex partition ${\mathcal V}= \{{V_{i,j}}\}_{i\in[r],j\in[k]}$ that satisfies $n/(8kr) \leq \|{V_{i,j}}\| \leq 4n/(kr)$ for each $i\in[r]$, $j\in[k]$. Let $\{n_{i,j}\}_{i \in [r], j\in [k]}$ be an integer partition of $\sum_{i\in[r],j\in[k]} \|V_{i,j}\|$. Let $R^k_r$ be a graph on the vertex set $[r] \times [k]$ with minimum degree $\delta(R^k_r) \geq \big((k-1)/k+\gamma/2\big) kr$ such that $K^k_r \subseteq B^k_r \subseteq R^k_r$. Suppose that the partition ${\mathcal V}$ satisfies the following properties for each $i\in[r]$, each $j\neq j'\in[k]$, and each $v\in V$. 1. \[lembalancing:sizes\] We have $n_{i,j} - \xi n \leq \|V_{i,j}\| \leq n_{i,j} + \xi n$, 2. \[lembalancing:regular1\] ${\mathcal V}$ is $\big(\tfrac{{\varepsilon}}{4},d,p\big)_G$-lower-regular on $R^k_r$ and $\big(\tfrac{{\varepsilon}}{4},d,p\big)_G$-super-regular on $K^k_r$, 3. \[lembalancing:inheritance1\] both $\big({N_{\Gamma}}(v, V_{i,j}),V_{i,j'}\big)$ and $\big({N_{\Gamma}}(v, V_{i,j}),{N_{\Gamma}}(v, V_{i,j'})\big)$ are $\big(\tfrac{{\varepsilon}}{4},d,p\big)_G$-lower-regular pairs, and 4. \[lembalancing:gamma1\] we have $\|{N_{\Gamma}}(v,V_{i,j})\| = \big(1 \pm \tfrac{{\varepsilon}}{4}\big)p\|{V_{i,j}}\|$. Then, there exists a partition $\mathcal{V'}= \{V'_{i,j}\}_{i\in[r],j\in[k]}$ of $V$ such that the following properties hold for each $i\in[r]$, each $j\neq j'\in [k]$, and each $v\in V$. 1. \[lembalancing:sizesout\] We have $\|V'_{i,j}\|=n_{i,j}$, 2. \[lembalancing:symd\] We have $\|V_{i,j}{\triangle}V'_{i,j}\|\le 10^{-10}{\varepsilon}^4k^{-2}r_1^{-2} n$, 3. \[lembalancing:regular\] $\mathcal{V'}$ is $({\varepsilon},d,p)_G$-lower-regular on $R^k_r$ and $({\varepsilon},d,p)_G$-super-regular on $K^k_r$, 4. \[lembalancing:inheritance\] both $\big(N_{\Gamma}(v,V'_{i,j}), V'_{i,j'}\big)$ and $\big(N_{\Gamma}(v,V'_{i,j}), N_\Gamma(v,V'_{i,j'})\big)$ are $({\varepsilon},d,p)_G$-lower-regular pairs, and 5. \[lembalancing:gammaout\] For each $1\le s\le\Delta$ and vertices $v_1,\ldots,v_s\in[n]$ we have $$\big\|N_\Gamma(v_1,\ldots,v_s;V_{i,j}){\triangle}N_\Gamma(v_1,\ldots,v_s;V'_{i,j})\big\|\le 10^{-10}{\varepsilon}^4k^{-2}r_1^{-2}\deg_\Gamma(v_1,\ldots,v_s)+{C^{\ast}}\log n\,.$$ Furthermore, if for any two disjoint vertex sets $A,A'\subset V(\Gamma)$ with $\|A\|,\|A'\|\ge\tfrac{1}{50000kr_1}{\varepsilon}^2\xi pn$ we have $e_\Gamma(A,A')\le \big(1+\tfrac{1}{100}{\varepsilon}^2\xi\big)p\|A\|\|A'\|$, and if ‘lower-regular’ is replaced with ‘regular’ in \[lembalancing:regular1\], and \[lembalancing:inheritance1\], then we can replace ‘lower-regular’ with ‘regular’ in \[lembalancing:regular\] and \[lembalancing:inheritance\]. The lemma for G {#sec:prooflemG} =============== In this section we prove the Lemma for $G$ (Lemma \[lem:G\]), which borrows from the proof of [@bottcher2009proof Proposition 17] and from the proof of [@bottcher2013almost Lemma 9]. Our strategy is as follows. We first apply Lemma \[lem:regularitylemma\] to obtain an equitable partition of $V(G)$ within whose reduced graph we can find a backbone graph by Theorem \[thm:bandwidth\]. We let $Z_1$ be the vertices whose $\Gamma$-degrees are ‘wrong’ to this partition, or whose neighbourhoods fail to inherit lower-regularity (plus a few extra to maintain $k$-equitability), and we remove the vertices $Z_1$. Now there may be some vertices in each cluster which destroy super-regularity on the clique factor of the backbone graph. We redistribute these, and the exceptional set of the regular partition, to other clusters. Now we would like to say we are finished, but the moving of vertices may have destroyed some of the regularity inheritance, $\Gamma$-neighbourhood, and super-regularity properties we tried to obtain. However, it is easy to check that a vertex only witnesses failure of these properties if exceptionally many of its $\Gamma$-neighbours were moved from or to a cluster. We let $Z_2$ be the set of all such vertices, and remove them. We will see that $Z_2$ is so small that its removal does not significantly affect the properties we want, so that we can set $V_0=Z_1\cup Z_2$ and we are done. We first fix the constants in the proof. Given $\gamma>0$, $k\ge 2$ and $r_0\ge 1$, set $d=\tfrac{\gamma}{32}$. Let $\beta$ and $n_0$ be returned by Theorem \[thm:bandwidth\] for input $\tfrac12\gamma$, $3k$ and $k$. Let $r'_0=\max\{n_0,k/d,10k/\beta,r_0\}$. Given ${\varepsilon}\in\big(0,\tfrac{1}{2k}\big]$, let $0<{{\varepsilon}^{\ast}}\le 10^{-10}{\varepsilon}^2\gamma k^{-2}$ be small enough for both Lemmas \[lem:OSRIL\] and \[lem:TSRIL\] for input $\tfrac12{\varepsilon}$ and $d$. Let $C$ be large enough for these applications of Lemmas \[lem:OSRIL\] and \[lem:TSRIL\], and also large enough for Proposition \[prop:chernoff\] with input $\tfrac{1}{1000}\big(\tfrac{{{\varepsilon}^{\ast}}}{k}\big)^2$. Now Lemma \[lem:regularitylemma\], with input $\tfrac{1}{k}{{\varepsilon}^{\ast}}$, $\tfrac{k-1}{k}+\gamma$ and $r'_0+k$, returns $r_1$. We set ${C^{\ast}}=1000k^3r_1^5C/({{\varepsilon}^{\ast}})^2$. Given $p\ge {C^{\ast}}\big(\tfrac{\log n}{n}\big)^{1/2}$, the random graph $G(n,p)$ a.a.s. satisfies the good events of Lemmas \[lem:OSRIL\] and \[lem:TSRIL\], and Proposition \[prop:chernoff\]. We condition on $\Gamma=G(n,p)$ satisfying these good events. Given $G\subset\Gamma$ with $\delta(G)\ge\big(\tfrac{k-1}{k}+\gamma\big)pn$, we apply Lemma \[lem:regularitylemma\], with input $\tfrac{1}{k}{{\varepsilon}^{\ast}}$, $\tfrac{k-1}{k}+\gamma$, $r'_0+k$, and $d$, to $G$. We may do this because $G$ is a subgraph of $\Gamma$, and by choice of ${C^{\ast}}$ we have $Cp^{-1}\log n\le\tfrac{{{\varepsilon}^{\ast}}n}{kr_1}$, so that the condition of Lemma \[lem:regularitylemma\] is satisfied because the good event of Proposition \[prop:chernoff\] holds for $\Gamma$. The result is a $\big(\tfrac{1}{k}{{\varepsilon}^{\ast}},p\big)$-lower-regular partition of $V(G)$ into $t'\in [r'_0+k, r_1]$ equally sized clusters, with exceptional set of size at most $\tfrac{1}{k}{{\varepsilon}^{\ast}}n$, whose $({{\varepsilon}^{\ast}},d,p)$-reduced graph has minimum degree at least $\big(\tfrac{k-1}{k}+\gamma-d-\tfrac{1}{k}{{\varepsilon}^{\ast}}\big)t'$. We remove at most $k-1$ of these clusters to the exceptional set, obtaining an $({{\varepsilon}^{\ast}},p)$-lower-regular partition ${\mathcal U}$ of $V(G)$ into $kr$ equally sized clusters, where $r'_0\le kr\le r_1$, with exceptional set $U_0$ of size at most ${{\varepsilon}^{\ast}}n$, whose $({{\varepsilon}^{\ast}},d,p)$-reduced graph $R^k_r$ has minimum degree at least $\big(\tfrac{k-1}{k}+\gamma-d-\tfrac{1}{k}{{\varepsilon}^{\ast}}\big)kr-k$. By choice of $d$ and ${{\varepsilon}^{\ast}}$, and by choice of $r'_0$, we have $$\big(\tfrac{k-1}{k}+\gamma-d-\tfrac{1}{k}{{\varepsilon}^{\ast}}\big)kr-k\ge\big(\tfrac{k-1}{k}+\tfrac{\gamma}{2}\big)kr\,.$$ Observe that $B^k_r$ has bandwidth at most $2k<\beta r'_0$, and maximum degree less than $3k$. Thus Theorem \[thm:bandwidth\], with input $\tfrac{\gamma}{2}$, $3k$, and $k$, in particular states that $R^k_r$ contains a copy of $B^k_r$. We fix one such copy. We let its vertices $\{(i,j)\}_{i\in[r],j\in[k]}$ label the vertices of $R^k_r$, and similarly let the cluster of ${\mathcal U}$ corresponding to the vertex $(i,j)$ of $B^k_r$ be $U_{i,j}$ for each $i\in[r]$ and $j\in[k]$. The partition ${\mathcal U}$ is equitable, and thus in particular $k$-equitable. We now create $Z_1$ as follows. We start with all vertices $v$ of $G$ for which there are $(i,j)$ and $(i',j')$ in $V(R^k_r)$, with $\{(i,j),(i',j')\}$ an edge of $R^k_r$, such that either $\big(N_\Gamma(v,U_{i,j}),U_{i',j'}\big)$ or $\big(N_\Gamma(v,U_{i,j}),N_\Gamma(v,U_{i',j'})\big)$ is not $\big(\tfrac{1}{2}{\varepsilon},d,p\big)_G$-lower-regular. We add all vertices $v$ of $G$ for which there exists $U_{i,j}$ with $\deg_\Gamma(v,U_{i,j})\neq(1\pm{{\varepsilon}^{\ast}})p\|U_{i,j}\|$, or for which $\deg_\Gamma(v,U_0)>2{{\varepsilon}^{\ast}}p n$. Finally we add a minimum number of vertices to obtain $k$-equitability of the sets $\big\{U_{i,j}\setminus Z_1\big\}_{i\in[r],j\in[k]}$. Note that we have $\|U_{i,j}\|\ge n/(2kr_1)$ for each $i,j$, and we can estimate the number of vertices with more than $2{{\varepsilon}^{\ast}}p n$ neighbours in $U_0$ by considering a superset of $U_0$ of size ${{\varepsilon}^{\ast}}n$. It follows that for each $i,j$ we have $\log(en/\|U_{i,j}\|),\log(en/\|U_0\|)\le\log(ekr_1/{{\varepsilon}^{\ast}})$. By Lemma \[lem:OSRIL\] and Lemma \[lem:TSRIL\], and Proposition \[prop:chernoff\], we have $$\label{eq:sizeZ1} \|Z_1\|\le 4kr_1^2 C\max\big\{p^{-2},p^{-1}\log (ekr_1/{{\varepsilon}^{\ast}})\big\}\le 8k^2r_1^3Cp^{-2}/{{\varepsilon}^{\ast}}\le\tfrac{{{\varepsilon}^{\ast}}}{kr_1} n\,,$$ where the factor $k$ accounts for vertices removed to maintain $k$-equitability. We now try to obtain super-regularity on the copy of $K^k_r$ in $B^k_r$. For each $i\in[r]$ and $j\in[k]$ let $W_{i,j}$ be the vertices of $U_{i,j}\setminus Z_1$ which have less than $(d-2{{\varepsilon}^{\ast}})p\|U_{i,j'}\|$ neighbours in $U_{i,j'}$ for some $j'\neq j$. Because $(U_{i,j},U_{i,j'})$ is $({{\varepsilon}^{\ast}},d,p)$-lower-regular for each $i\in[r]$ and $j\neq j'\in[k]$, we have $\|W_{i,j}\|\le k{{\varepsilon}^{\ast}}\|U_{i,j}\|$ for each $i\in[r]$ and $j\in[k]$. Now let $W$ contain $U_0\setminus Z_1$ together with all $W_{i,j}$ for $i\in[r]$ and $j\in[k]$, and a minimum number of additional vertices from $V(G)\setminus Z_1$ to obtain $k$-equitability of the sets $\big\{U_{i,j}\setminus (Z_1\cup W)\big\}_{i\in[r],j\in[k]}$. By construction, we have $\|W\|\le{{\varepsilon}^{\ast}}n+kr\cdot k{{\varepsilon}^{\ast}}\tfrac{n}{kr}\le 2k{{\varepsilon}^{\ast}}n$. Given any $w\in W$, because $w\not\in Z_1$ we have $$\deg_\Gamma(w,U_0)\le 2{{\varepsilon}^{\ast}}pn\quad\text{and}\quad\deg_\Gamma(w,U_{i,j})\le(1+{{\varepsilon}^{\ast}})p\|U_{i,j}\|$$ for each $i\in[r]$ and $j\in[k]$. Now let us consider the edges of $G$ leaving $w$. At most $2{{\varepsilon}^{\ast}}pn$ of these go to $U_0$, and by definition at most $2dpn$ go to sets $U_{i,j}$ such that $\deg_G(w,U_{i,j})\le 2dp\|U_{i,j}\|$. Since $\deg_G(w)\ge\big(\tfrac{k-1}{k}+\gamma\big)pn$, at least $\big(\tfrac{k-1}{k}+\tfrac{\gamma}{2}\big)pn$ edges leaving $w$ go to sets $U_{i,j}$ with $\deg_G(w,U_{i,j})\ge 2dp\|U_{i,j}\|$. Since $\|U_{i,j}\|\le \tfrac{1}{kr}n$, in particular there are at least $$\frac{\big(\tfrac{k-1}{k}+\tfrac{\gamma}{2}\big)pn}{(1+{{\varepsilon}^{\ast}})p\tfrac{n}{kr}}\ge \big(\tfrac{k-1}{k}+\tfrac{\gamma}{4}\big)kr$$ sets $U_{i,j}$ with $i\in[r]$ and $j\in[k]$ such that $\deg_G(w,U_{i,j})\ge 2dp\|U_{i,j}\|$. It follows that there are at least $\tfrac{\gamma}{4}r$ indices $i\in[r]$ such that $\deg_G(w,U_{i,j})\ge 2dp\|U_{i,j}\|$ for each $j\in[k]$. We now assign to each $w\in W$ sequentially an index $c(w)\in[r]\times [k]$. For each $w$, we choose $c(w)=(i,j)$ as follows. The index $i$ is chosen minimal in $[r]$ such that $\deg_G(w,U_{i,j'})\ge 2dp\|U_{i,j'}\|$ for each $j'\in[k]$, but at most $\tfrac{100}{r}k{{\varepsilon}^{\ast}}\gamma^{-1} n$ vertices $w'\in W$ have so far been assigned $c(w')=(i,j')$ for any $j'\in[k]$. We choose $j\in[k]$ minimising the number of vertices $w'\in W$ with $c(w)=(i,j)$. Because $\|W\|\le 2k{{\varepsilon}^{\ast}}n$, this assignment is always possible. Next, for each $i\in[r]$ and $j\in[k]$, we let $V'_{i,j}$ consist of $U_{i,j}\setminus (Z_1\cup W_{i,j})$, together with all $w\in W$ such that $c(w)=(i,j)$. By construction, we have $$\|U_{i,j}{\triangle}V'_{i,j}\|\le \|Z_1\|+\|W_{i,j}\|+\tfrac{100}{r}k{{\varepsilon}^{\ast}}\gamma^{-1}n\le 1000k^2{{\varepsilon}^{\ast}}\gamma^{-1}\|U_{i,j}\|\,.$$ Finally, we let $Z_2$ be the vertices $v\in V(G)\setminus Z_1$ with $\deg_\Gamma(v,U_{i,j}{\triangle}V'_{i,j})\ge 2000k^2{{\varepsilon}^{\ast}}\gamma^{-1} p\|U_{i,j}\|$ for some $i\in[r]$ and $j\in[k]$, together with a minimum number of additional vertices of $V(G)\setminus Z_1$ to obtain $k$-equitability of the sets $V_{i,j}:=V'_{i,j}\setminus Z_2$. We set $V_0=Z_1\cup Z_2$. We claim that ${\mathcal V}=\{V_{i,j}\}_{i\in[r],j\in[k]}$ is the desired partition of $V(G)\setminus V_0$. Note that the sets $V'_{i,j}$ and $V'_{i,j'}$ differ in size by at most one for any $i\in[r]$ and $j,j'\in[k]$, by our construction of the assignment $c$. We apply Proposition \[prop:chernoff\] to estimate the number of vertices $v\in V(G)\setminus Z_1$ with $\deg_\Gamma(v,U_{i,j}{\triangle}V'_{i,j})\ge 2000k^2{{\varepsilon}^{\ast}}\gamma^{-1} p\|U_{i,j}\|$ by considering a superset of $U_{i,j}{\triangle}V'_{i,j}$ of size $1000k^2{{\varepsilon}^{\ast}}\gamma^{-1}\|U_{i,j}\|\ge {{\varepsilon}^{\ast}}n/r_1$. By Proposition \[prop:chernoff\] we thus have $$\label{eq:sizeZ2} \|Z_2\|\le r_1+Ckr_1p^{-1}\log (er_1/{{\varepsilon}^{\ast}})\le 4Ckr_1^2p^{-1}/{{\varepsilon}^{\ast}}\le\tfrac{{{\varepsilon}^{\ast}}}{kr_1}pn\,.$$ This gives $$\label{eq:symUV} \|U_{i,j}{\triangle}V_{i,j}\|\le\|U_{i,j}{\triangle}V'_{i,j}\|+\|Z_2\|\le 2000k^2{{\varepsilon}^{\ast}}\gamma^{-1}\|U_{i,j}\|\,.$$ Now given any $v\in V(G)\setminus V_0$, for each $i\in[r]$ and $j\in[k]$, because $v\not\in Z_2$ we have $\deg_\Gamma(v,U_{i,j}{\triangle}V'_{i,j})\le 2000k^2{{\varepsilon}^{\ast}}\gamma^{-1}p\|U_{i,j}\|$. We thus have $$\label{eq:symdUV} \deg_\Gamma(v,U_{i,j}{\triangle}V_{i,j})\le 2000k^2{{\varepsilon}^{\ast}}\gamma^{-1}p\|U_{i,j}\|+\|Z_2\|\le 3000k^2{{\varepsilon}^{\ast}}\gamma^{-1}p\|U_{i,j}\|\,,$$ and because $v\not\in Z_1$ we have $\deg_\Gamma(v,U_{i,j})=(1\pm{{\varepsilon}^{\ast}})p\|U_{i,j}\|$, and hence by $$\label{eq:vU} \deg_\Gamma(v,V_{i,j})=\big(1\pm 10000k^2{{\varepsilon}^{\ast}}\gamma^{-1}\big)p\|V_{i,j}\|\,.$$ Adding up and , we conclude $$\label{eq:sizeV0} \|V_0\|\le 8k^2r_1^3Cp^{-2}/{{\varepsilon}^{\ast}}+4Ckr_1^2p^{-1}/{{\varepsilon}^{\ast}}\le {C^{\ast}}p^{-2}\,,$$ as desired. The partition ${\mathcal V}=\{V_{i,j}\}_{i\in[r],j\in[k]}$ is by construction $k$-equitable, and the graph $R^k_r$ has minimum degree $\big(\tfrac{k-1}{k}+\tfrac{\gamma}{2}\big)kr$ as desired. For each $i\in[r]$ and $j\in[k]$ we have $\|U_{i,j}\|=(1\pm{{\varepsilon}^{\ast}})\frac{n}{kr}$, and so and our choice of ${{\varepsilon}^{\ast}}$ give \[lemG:size\]. Next, if $\{(i,j),(i',j')\}$ is an edge of $R^k_r$, then $G$ is $({{\varepsilon}^{\ast}},d,p)$-lower-regular on $(U_{i,j},U_{i',j'})$ by construction. By , Proposition \[prop:subpairs3\], and our choice of ${{\varepsilon}^{\ast}}$, $G$ is $({\varepsilon},d,p)$-lower-regular on $(V_{i,j},V_{i',j'})$. Given $i\in[r]$ and $j\neq j'\in[k]$, let $v$ be a vertex of $V_{i,j}$. Observe that since $v\in V_{i,j}$, either we have $v\in U_{i,j}$, in which case, since $v\not\in W$ we have $\deg_G(v,U_{i,j'})\ge (d-2{{\varepsilon}^{\ast}})p\|U_{i,j'}\|$, or $v$ is in $W$ and has $c(v)=(i,j)$, in which case $\deg_G(v,U_{i,j'})\ge dp\|U_{i,j'}\|$. By and we have $$\deg_G(v,V_{i,j'})\ge(d-2{{\varepsilon}^{\ast}})p\|U_{i,j'}\|-3000k^2{{\varepsilon}^{\ast}}\gamma^{-1}p\|U_{i,j'}\|\ge(d-{\varepsilon})p\|V_{i,j'}\|\,,$$ giving \[lemG:regular\]. If $\{(i,j),(i',j')\}\in E(R^k_r)$, then for any $v\in V(G)\setminus V_0$, since $v\not\in Z_1$, the pairs $\big(N_\Gamma(v,U_{i,j}),U_{i',j'}\big)$ and $\big(N_\Gamma(v,U_{i,j}),N_\Gamma(v,U_{i',j'})\big)$ are $\big(\tfrac12{\varepsilon},d,p\big)_G$-lower-regular. Using and , Proposition \[prop:subpairs3\] and our choice of ${{\varepsilon}^{\ast}}$, we conclude \[lemG:inheritance\]. Finally, \[lemG:gamma\] follows from and our choice of ${{\varepsilon}^{\ast}}$. Note that if we alter the definition of $Z_1$, removing the condition on $\big(N_\Gamma(v,U_{i,j}),N_\Gamma(v,U_{i',j'})\big)$, then we do not need to use Lemma \[lem:TSRIL\] and the bound in improves to $\|Z_1\|\le 8k^2r_1^3Cp^{-1}/{{\varepsilon}^{\ast}}$. Thus, if we only require \[lemG:inheritancep\], we obtain $\|V_0\|\le{C^{\ast}}p^{-1}$ as claimed. The lemma for H {#sec:lemH} =============== In this section we present the proof of Lemma \[lem:H2\]. First let us state McDiarmid’s Inequality (see e.g. [@janson2011random] for a proof) that we will use in the proof. \[lem:McDiarmid\] Let $X_1, \ldots, X_k$ be independent random variables, where $X_i$ takes values in a finite set $A_i$ for each $i\in [k]$. Suppose that a function $g: A_1 \times \ldots \times A_k \to \mathbb R$ satisfies for each $i\in [k]$ $$\sup_{x_1,\ldots, x_k, \hat x_i}\|g(x_1,x_2,\ldots,x_k)-g(x_1,x_2,\ldots, x_{i-1}, \hat{x}_i, x_{i+1}, \ldots, x_k)\| \leq c_i.$$ Then, for any ${\varepsilon}>0$, we have $${\mathbb{P}}\big[\|{\mathbb{E}}[g(X_1,\ldots, X_k)]- g(X_1,\ldots, X_k)\| \geq {\varepsilon}\big]\leq 2\exp\left\{-\frac{2{\varepsilon}^2}{\sum_{i\in[k]} c_i^2}\right\}\,.$$ The proof idea is then as follows. First, given the zero-free labelling ${\mathcal L}$ and $(k+1)$-colouring $\sigma$ of $H$, we split ${\mathcal L}$ into the blocks of the definition of zero-freeness. We partition the blocks into $r$ ‘sections’ of consecutive blocks, such that the $i$-th section contains about $\sum_{j\in[k]}m_{i,j}$ vertices, and furthermore such that the ‘boundary vertices, namely the first and last $\beta n$ vertices of each section, do not receive colour zero. Now it is easy to check that assigning the vertices of colour $j$ in the $i$-th section to $(i,j)$ for each $i\in[r]$ and $j\in[k]$, and the vertices of colour zero in the $i$-th section to $z_i$, is a graph homomorphism. However it can be very unbalanced, since different colours in $[k]$ may be used with very different frequencies in each section. To fix this, we replace $\sigma$ with a new colouring $\sigma'$, which we obtain as follows. We partition each section into ‘intervals’ of consecutive blocks, and for each interval except the last in each section, we pick a random permutation of $[k]$. We will show that there is a colouring $\sigma'$ such that all but the first few vertices of each interval are coloured according to the permutation applied to $\sigma$, with vertices of colour zero staying coloured zero. We use this colouring $\sigma'$ in place of $\sigma$ to define the mapping $f$. We let $X$ consist of all vertices whose distance is two or less to either boundary vertices, vertices near the start of an interval, or colour zero vertices. To complete the proof, we show that so few vertices receive colour zero that they do not much affect the desired conclusions. Now the mapping $f$ is in expectation balanced, and using Lemma \[lem:McDiarmid\] we can show that it is also with high probability close to balanced. It is also easy to check that, since $H$ is $D$-degenerate, in the $i$-th section of ${\mathcal L}$ there are many vertices of degree at most $2D$. In expectation these are distributed about evenly over the $\big\{(i,j)\big\}_{j\in[k]}$ by $f$, and again McDiarmid’s inequality shows that with high probability the same holds. These two observations give us \[lemH:H1\] and \[lemH:H6\], while the other four desired conclusions hold by construction. For given $D\geq 1$, set $\alpha = 1/(24D)$. Let $k, r \geq 1$ and $\xi, \beta >0$ be given, where $\xi \leq 1/(kr)$ and $\beta \leq 10^{-10}\xi^2/(D k^4r)$. Let $H$ and $K^k_r \subseteq B^k_r \subseteq R^k_r$ be graphs as in the statement of the lemma. Let $\mathcal L$ be the given labelling of $V(H)$ of bandwidth at most $\beta n$. We denote the set of the first $\sqrt{\beta}n$ vertices of $\mathcal L$ by $F$. Let $\sigma: V(H) \to \{0,\ldots k\}$ be the given proper $(k+1)$-colouring of $V(H)$ that is $(10/\xi, \beta)$-zero-free with respect to $\mathcal L$ and such that $\sigma(F) \subseteq [k]$. Also, let $z_1, \ldots, z_n$ be vertices such that $z_i \in \big([r]\setminus\{i\}\big) \times [k]$ with $\big\{z_i, (i,j)\big\} \in E(R^k_r)$ for every $i\in [r]$ and $j\in [k]$. Finally, set $b=k/\sqrt{\beta}$. Let $\{m_{i,j}\}_{i\in[r],j\in[k]}$ be the given $k$-equitable integer partition of $n$ with $n/(10kr) \leq m_{i,j} \leq 10n/(kr)$ for every $i\in[r]$ and $j\in[k]$. Let us now introduce the notation that we use in this proof. Recall that for every $t\in \big[1/(4k\beta)\big]$ the $i$-th block is defined as $$B_t:= \{(t-1)4k\beta n +1, \ldots, t4k\beta n\}.$$ Next we split the labelling $\mathcal L$ into $r$ sections, where the first and the last block of each section are zero-free. Each section is partitioned into intervals, each of which but possibly the last one consists of $b$ blocks. Since $\sigma$ is $(10/\xi, \beta)$-zero-free with respect to $\mathcal L$, we can choose indices $0 = t_0 \leq t_1 \leq \ldots \leq t_{r-1} \leq t_r = 1/(4k\beta)$ such that $B_{t_i}$ and $B_{t_{i}+1}$ are zero-free blocks for every $i\in [r]$ and $$\sum_{t=1}^{t_i}\|B_t\| \leq \sum_{t=1}^{i} \sum_{j\in[k]}m_{t,j} < 12k\beta n + \sum_{t=1}^{t_i} \|B_t\|.$$ Since $m_{i,j} \geq n/(10kr) > 12k\beta n$, indices $t_0, \ldots, t_r$ are distinct. For every $i \in [r]$ we define the $i$-th section $S_i$ as $$\bigcup_{t = t_{i-1}+1}^{t_i} B_t.$$ This means by the choice of the indices $t_0, \ldots, t_r$ that the first and last block of each section are zero-free. Since $\{m_{i,j}\}_{i\in[r],j\in[k]}$ is a $k$-equitable partition, we have in particular $$\label{eq:mijSi} \frac{1}{k} (\|S_i\|- 12k\beta n) \leq m_{i,j} \leq \frac{1}{k} \big(\|S_i\| + 12 k \beta n\big).$$ The last $\beta n$ vertices of the blocks $B_{t_i}$ and the first $\beta n$ vertices of the blocks $B_{t_{i+1}}$ are called boundary vertices of $H$. Notice that colour zero is never assigned to boundary vertices by $\sigma$. For each $i\in [r]$, we split $S_i$ into $s_i:= \left\lceil (t_i-t_{i-1}-1)/b \right\rceil$ intervals, where each of the first $(s_i-1)$ intervals is the concatenation of exactly $b$ blocks and the last interval consists of $t_i-t_{i-1}-1 - b(s_i-1) \leq b$ blocks. Therefore, for every $i\in [r]$, we have $$\label{eq:sizeSi} s_i(b-1)4k\beta n +1 \leq \|S_i\| \leq s_i b 4 k \beta n.$$ Using Equation , $b= k/\sqrt{\beta}$, and $n/(10kr)\leq m_{i,j} \leq 10n/(kr)$ we get, for every $i\in[r]$, the following bounds on $s_i$ $$\label{eq:si} \frac{1}{100rk^2\sqrt{\beta}} \leq s_i \leq \frac{10}{rk^2\sqrt{\beta}}.$$ We denote the intervals of the $i$-th section by $I_{i,1}, \ldots, I_{i,s_i}$. Let ${B^{\mathrm{sw}}}_{i,\ell}$ denote the union of the first two blocks of each interval $I_{i,\ell}$. All of these blocks but ${B^{\mathrm{sw}}}_{i,1}$ and ${B^{\mathrm{sw}}}_{i,s_i}$ will be used to switch colours within parts of $H$. Notice that we have $\|{B^{\mathrm{sw}}}_{i,\ell}\| = 8k \beta n$ and, since $\sigma$ is $(10/\xi, \beta)$-zero-free with respect to $\mathcal L$, at least one of the two blocks of ${B^{\mathrm{sw}}}_{i,\ell}$ is zero-free. We will not use ${B^{\mathrm{sw}}}_{i,1}$ and ${B^{\mathrm{sw}}}_{i,s_i}$ to switch colours because we will need that the boundary vertices do not receive colour zero. For every $i\in [r]$ and every $\ell \in \{2,\ldots, s_i-1\}$, we choose a permutation $\pi_{i,\ell}: [k] \to [k]$ uniformly at random. The next claim ensures that we can use zero-free blocks to obtain a proper colouring of the vertex set such that vertices before the switching block are coloured according to the original colouring and the colours of the vertices after the switching block are permuted as wished. A proof can be found in [@BoeTarWue]. \[claim:switching\] Let $\sigma: [n] \to \{0,\ldots, k\}$ be a proper $(k+1)$-colouring of $H$, let $B_t$ be a zero-free block and let $\pi$ be any permutation of $[k]$. Then there exists a proper $(k+1)$-colouring $\sigma'$ of $H$ with $\sigma'(x) = \sigma(x)$ for all $x\in \bigcup_{i<t} B_i$ and $$\sigma'(x) = \begin{cases} \pi(\sigma(x)) & \text{if } \sigma(x) \neq 0 \\ 0 & \text{otherwise} \end{cases}$$ for all $x\in \bigcup_{i>t}B_i$. We use Claim \[claim:switching\] to switch colours at the beginning of each interval except for the first and last interval of each section. More precisely, we switch colours within the sets ${B^{\mathrm{sw}}}_{i,\ell}$ so that the colouring of the remaining vertices in the interval $I_{i,\ell}$ matches $\pi_{i, \ell}$. Note that we can indeed use ${B^{\mathrm{sw}}}_{i,\ell}$ to do the switching since one of the two blocks in ${B^{\mathrm{sw}}}_{i,\ell}$ is zero-free. In particular, we get a proper $(k+1)$-colouring $\sigma'= \sigma'\big(\pi_{1,2}, \ldots, \pi_{r,s_r-1}\big): V(H) \to \{0, \ldots k+1\}$ of $H$ that fulfils the following. For every $x\in I_{1,1}$ we have $$\sigma'(x) = \sigma(x),$$ for each $i \in [r]$ and $\ell \in \{2,\ldots, s_i-1\}$ and every $x\in I_{i,\ell} \setminus {B^{\mathrm{sw}}}_{i,\ell}$ we have that $$\sigma'(x) = \begin{cases} \pi_{i,\ell}\big(\sigma(x)\big) & \text{if } \sigma(x) \neq 0 \\ 0 & \text{otherwise} \end{cases}$$ and for each $i\in [r]$ and every $x\in I_{i,s_{i}}\cup I_{i+1,1}$ (where $I_{r+1,1}:= \varnothing$) we have that $$\sigma'(x) = \pi_{i, s_i-1}\big(\sigma(x)\big).$$ While $\sigma'$ is well-defined on the sets ${B^{\mathrm{sw}}}_{1,2}, \ldots, {B^{\mathrm{sw}}}_{r, s_r-1}$ by Claim \[claim:switching\], the definition on these sets is rather complicated as it is depends on which of the two blocks in ${B^{\mathrm{sw}}}_{i,\ell}$ is zero-free and on the colourings before and after the switching. However, the precise definition on these sets is not important for the remainder of the proof. Hence, we omit it here. Observe that $\sigma'$ never assigns colour zero to boundary vertices. Using $\sigma'$ we now define $f= f\big(\pi_{1,2}, \ldots, \pi_{r,s_r-1}\big): V(H) \to [r] \times [k]$ as follows. For each $i\in [r]$ and $x\in S_i$ we set $$f(x):= \begin{cases} \big(i, \sigma'(x)\big) & \text{if } \sigma'(x) \neq 0 \\ z_i & \text{otherwise},\end{cases}$$ where $z_i \in \big([r]\setminus\{i\}\big) \times [k]$ is the vertex defined in the statement of the lemma. Let $X$ consist of all vertices at distance two or less from a boundary vertex of ${\mathcal L}$, from a vertex in any ${B^{\mathrm{sw}}}_{i,\ell}$, or from a colour zero vertex. We now show that $f$ and $X$ satisfy Properties \[lemH:H2\]–\[lemH:H5\] with probability 1 and Properties \[lemH:H1\] and \[lemH:H6\] with high probability. In particular, this implies that the desired $f$ and $X$ exist. We start with Property \[lemH:H1\]. For each $i\in [r]$ let $$S^\ast_i:=S_i \setminus \left( \bigcup_{ \ell \in [s_i]} {B^{\mathrm{sw}}}_{i,\ell} \cup I_{i,1} \cup I_{i,s_i} \right)$$ be the set of all vertices in $S_i$ except for the first and last interval and the first two blocks of each interval of $S_i$. We will also make use of the following restricted function $$f^\ast = f^\ast\big(\pi_{1,2},\ldots, \pi_{r,s_r}\big):= {{ \left.\kern-\nulldelimiterspace f \vphantom{\big\|} \right\|_{\bigcup_{i\in[r]} S^\ast_i } }}.$$ The basic idea of the proof of Property \[lemH:H1\] is to determine bounds on $\|{f^\ast}^{-1}(i,j)\|$ that hold with positive probability and then deduce the desired bounds on $\|f^{-1}(i,j)\|$. Since the permutations $\pi_{i,\ell}$ were chosen uniformly at random, we have by definition of $f^\ast$ that the expected number of vertices mapped to $(i,j)\in [r]\times[k]$ by $f^\ast$ is $$\begin{gathered} {\mathbb{E}}\big[\|{f^{\ast}}^{-1}(i,j)\|\big] = \frac{1}{k} \Big[ (s_i-2) (b-2) 4k \beta n -\big\|\{x \in S^\ast_{i}: \sigma(x) = 0 \}\big\| \Big] \\ + \big\|\bigcup_{\iota\in [r]\setminus \{i\}} \{x\in S^\ast_{\iota}: \sigma(x) = 0 \text{ and } z_{\iota} = (i,j) \}\big\|\,.\end{gathered}$$ In particular, the following bounds on the expected value of $\|{f^{\ast}}^{-1}(i,j)\|$ hold. $$\label{eq:fastijlow} {\mathbb{E}}\big[\|{f^{\ast}}^{-1}(i,j)\|\big] \leq (s_i-2) (b-2) 4 \beta n + \frac{\xi}{10} n$$ and $$\label{eq:fastijup} {\mathbb{E}}\big[\|{f^{\ast}}^{-1}(i,j)\|\big] \geq (1- \xi/ 10) (s_i-2) (b-2) 4 \beta n \geq (s_i-2) (b-2) 4 \beta n - \frac{\xi}{10}n.$$ If one replaced a permutation $\pi_{i,\ell}$ by some other permutation $\tilde{\pi}: [k] \to [k]$, then $\|{f^\ast}^{-1}(i,j)\|$ would change by at most $(b-2) 4k\beta n$. Hence, by McDiarmid’s Inequality (Lemma \[lem:McDiarmid\]) we have $$\begin{gathered} {\mathbb{P}}\left[\big\| (s_i-2) (b-2) 4 \beta n- \|{f^\ast}^{-1}(i,j)\|\big\| \geq \frac{\xi}{5} n \right] \overset{\eqref{eq:fastijlow}, \eqref{eq:fastijup}}\leq \\ {\mathbb{P}}\left[\big\| {\mathbb{E}}[\|{f^\ast}^{-1}(i,j)] - \|{f^\ast}^{-1}(i,j)\|\big\| \geq \frac{\xi}{10} n \right] \leq 2\exp\left\{-\frac{\xi^2 n^2}{50 (s_i-2) \big((b-2)4k\beta n\big)^2}\right\}.\end{gathered}$$ Taking the union bound over all $j\in [k]$ and using $s_i\leq 10/(rk^2\sqrt{\beta})$ and $b=k/\sqrt{\beta}$ as well as $\beta \leq 10^{-10}\xi^2/(D k^4r)$ yields $${\mathbb{P}}\left[\big\| (s_i-2) (b-2) 4 \beta n- \|{f^\ast}^{-1}(i,j)\|\big\| \geq \frac{\xi}{5} n \text{ for all } j\in [k] \right] \leq 2k \exp\left\{-\frac{\xi^2r}{8000k^2\sqrt{\beta}}\right\}\leq 2k e^{-k} < 1 .$$ Observe that $\|{f^\ast}^{-1}(i,j)\|$ is independent of the choices for $\pi_{i',\ell}$ if $i' \neq i$. Hence, with positive probability we have, for every $i\in [r]$ and $j\in [k]$, that $$(s_i-2) (b-2) 4 \beta n - \frac{\xi}{5}n \leq \|{f^\ast}^{-1}(i,j)\| \leq (s_i-2) (b-2) 4 \beta n + \frac{\xi}{5} n.$$ From the definition of $f^\ast$ it follows that $\|f^{-1}(i,j)\|\geq \|{f^\ast}^{-1}(i,j)\|$ and $$\|f^{-1}(i,j)\| \leq \|{f^\ast}^{-1}(i,j)\| + \|I_{i,1}\| + \|I_{i,s_i}\| + \sum_{\ell=2}^{s_i-1} \|{B^{\mathrm{sw}}}_{i,\ell}\|+ \Big\|\big\{x \in\hspace{-2mm} \bigcup_{\iota \in [r] \setminus \{i\}} S_{\iota}\setminus S^\ast_{\iota}: \sigma'(x) = 0 \text{ and } z_{\iota}=(i,j) \big\}\Big\|.$$ Using $s_i\leq 10/(rk^2\sqrt{\beta})$ and $b=k/\sqrt{\beta}$ and $\beta \leq 10^{-10}\xi^2/(D k^4r)$, with positive probability we have for every $i\in[r]$ and $j\in[k]$ that $$\begin{aligned} \|f^{-1}(i,j)\| &\geq \|{f^\ast}^{-1}(i,j)\| \geq (s_i-2) (b-2) 4 \beta n - \frac{\xi}{5} n\\ &\geq (s_i-2)(b-2) 4 \beta n - \frac{\xi}{5} n + \left(8 (s_i + b) \beta n - \frac{4}{5}\xi n\right)\\ &\geq s_ib4\beta n + 16 \beta n - \xi n\\ & \overset{\eqref{eq:sizeSi}}\geq \frac{1}{k}\big(\|S_i\| + 16 k \beta n\big) - \xi n \overset{\eqref{eq:mijSi}}\geq m_{i,j} - \xi n.\end{aligned}$$ On the other hand, $$\begin{aligned} \|f^{-1}(i,j)\| &\leq \|{f^\ast}^{-1}(i,j)\| + \|I_{i,1}\| + \|I_{i,s_i}\| + \sum_{\ell=2}^{s_i-1} \|{B^{\mathrm{sw}}}_{i,\ell}\| \\&\,\,\quad+ \Big\|\big\{x \in \bigcup_{\iota \in [r] \setminus \{i\}} S_{\iota}\setminus S^\ast_{\iota}: \sigma'(x) = 0 \text{ and } z_{\iota}=(i,j) \big\}\Big\|\\& \leq (s_i-2) (b-2) 4 \beta n + \frac{\xi}{5} n +8bk\beta n+ (s_i-2)8k\beta n+ \frac{\xi}{10}n \\& \leq \frac{1}{k}\big((s_i-2)(b-2)4k\beta n \big) + \xi n\\& \leq \frac{1}{k} (\|S_i\|-12k\beta n) + \xi n \overset{\eqref{eq:mijSi}}\leq m_{i,j}+\xi n,\end{aligned}$$ which shows that Property \[lemH:H1\] holds with positive probability. By definition of $X$, since $\mathcal L$ is a $\beta n$-bandwidth ordering, any vertex in $X$ is at distance at most $2\beta n$ in ${\mathcal L}$ from a boundary vertex, a vertex of some ${B^{\mathrm{sw}}}_{i,\ell}$, or from a vertex assigned colour zero. Because there are $r$ sections, the boundary vertices form $r-1$ intervals each of length $2\beta n$, and so at most $6r\beta n$ vertices of $H$ are at distance $2$ or less from a boundary vertex. There are $\sum_{i\in[r]} s_i$ intervals and hence $\sum_{i\in[r]} s_i$ switching blocks each of size $8k\beta n$. As $s_i \leq 10/(rk^2\sqrt{\beta})$ for every $i\in[r]$, there are at most $(4+8k)\beta n\cdot 10/(k^2\sqrt{\beta})$ vertices at distance 2 or less from a vertex of some switching block. Similarly, because ${\mathcal L}$ is $(10/\xi,\beta)$-zero-free, in any consecutive $10/\xi$ blocks at most one contains vertices of colour zero, and hence at most $(8+4k)\beta n$ vertices in any such $10/\xi$ consecutive blocks are at distance $2$ or less from a vertex of colour zero. Thus we have $$\|X\|\le 6r\beta n+(4+8k)\beta n\left(\tfrac{10}{k^2\sqrt{\beta}n}\right)+(8+4k)\beta n\big(\tfrac{n}{4k\beta n\cdot 10/\xi}+1\big)\le 6r\beta n+\tfrac{1}{4}\xi n+\tfrac{1}{3}\xi n\le\xi n\,,$$ which gives \[lemH:H2\]. Since $\sigma'$ is a proper colouring, and boundary vertices are not adjacent to colour zero vertices, by definition, $f$ restricted to the boundary vertices is a graph homomorphism to $B^k_r$. On the other hand, on each section $S_i$, again since $\sigma'$ is a proper colouring and since $\big\{(i,j)\big\}_{j\in[k]}\cup\{z_i\}$ forms a clique in $R^k_r$, $f$ is a graph homomorphism to $R^k_r$. Since ${\mathcal L}$ is a $\beta n$-bandwidth ordering, any edge of $H$ is either contained in a section or goes between two boundary vertices, and we conclude that $f$ is a graph homomorphism from $H$ to $R^k_r$, giving \[lemH:H3\]. Now, given $i\in[r]$ and $j\in[k]$, and $x\in f^{-1}(i,j)\setminus X$, if $\{x,y\}$ and $\{y,z\}$ are edges of $H$, then $y$ and $z$ are at distance two or less from $x$ in $H$. In particular, by definition of $X$ neither $y$ nor $z$ is either a boundary vertex, in any ${B^{\mathrm{sw}}}_{i,\ell}$, or assigned colour zero. Since boundary vertices appear in intervals of length $2\beta n$ in ${\mathcal L}$, and ${\mathcal L}$ is a $\beta n$-bandwidth ordering, it follows that $y$ and $z$ are both in $S_i$. Furthermore, suppose $x\in I_{i,\ell}$ for some $\ell$. By definition $x\not\in{B^{\mathrm{sw}}}_{i,\ell}$. Because ${B^{\mathrm{sw}}}_{i,\ell}$ and ${B^{\mathrm{sw}}}_{i,\ell+1}$ (if the latter exists) are intervals of length $8k\beta n$, both $y$ and $z$ are also in $I_{i,\ell}\setminus {B^{\mathrm{sw}}}_{i,\ell}$, and in particular both $y$ and $z$ are in $\bigcup_{j'\in[k]}f^{-1}(i,j')$, giving \[lemH:H4\]. Since $\sqrt{\beta}n \leq b4k\beta n \leq \|I_{1,1}\|$ and $\sigma'(x) \neq 0$ for each $x$ in the first $\sqrt{\beta}n$ vertices of $\mathcal{L}$, it follows directly from the definition of $f$ that $f(x) = \big(1,\sigma(x)\big)$, which shows Property \[lemH:H5\]. Finally, we show that Property \[lemH:H6\] holds with positive probability. Let $i\in[r]$ and $j\in[k]$. We define the random variable $\mathcal E_{i,j} := \|\{x \in {f^\ast}^{-1}(i,j): \deg(x) \leq 2 D\}\|$. Since $H$ is $D$-degenerate and $\mathcal L$ is a labelling of bandwidth at most $\beta n$ we have $$e\big(S_i^\ast, V(H)\big) \leq D \|S_i^\ast\| + D 4\beta n \leq D \big(1+ 1/(4D)\big) \|S_i^\ast\|.$$ Hence, it must hold that $\|\{x\in S_i^\ast: \deg(x) \geq 2D + 1\}\| (2D + 1) \leq 2 D \big(1+1/(4D)\big) \|S_i^\ast\|.$ This yields $\|\{x\in S_i^\ast: \deg(x) \leq 2D\}\| \geq \|S_i^\ast\|/(6D)$ and therefore $${\mathbb{E}}[\mathcal E_{i,j}] \geq \frac{1}{6kD} \|S_i^\ast\| \geq \frac{1}{6D} (s_i-2) (b-2)4\beta n.$$ By applying Chernoff’s Inequality (Theorem \[thm:chernoff\]) and using Equations and as well as $\alpha= 1/(24D)$ we get with positive probability $$\begin{aligned} &{\mathbb{P}}\Big[\big\|\{x \in f^{1}(i,j):\deg(x) \leq 2 D \}\big\| < \alpha \|f^{-1}(i,j)\|\Big] \overset{\ref{lemH:H1}}\leq {\mathbb{P}}\Big[\mathcal E_{i,j} < \alpha (s_i b 4 \beta n + 2\xi n)\Big]\\ &\leq {\mathbb{P}}\Big[\mathcal E_{i,j} < 2 \alpha \big((s_i-2)(b-2) 4 \beta n\big)\Big] \leq {\mathbb{P}}\Big[\mathcal E_{i,j} < \tfrac{1}{2}{\mathbb{E}}[\mathcal E_{i,j}]\Big] < 2 \exp\left\{- \frac{(s_i-2)(b-2)4\beta n}{72}\right\} < 1.\end{aligned}$$ Taking the union bound over all $i\in[r]$ and $j\in [k]$ yields that Property \[lemH:H6\] holds with positive probability. The common neighbourhood lemma ============================== In order to prove Lemma \[lem:common\] we need the following version of the Sparse Regularity Lemma, allowing for a partition equitably refining an initial partition with parts of very different sizes. Given a partition $V(G)=V_1{\mathbin{\text{\mbox{\makebox[0mm][c]{\hphantom{$\cup$}$\cdot$}$\cup$}}}}\dots{\mathbin{\text{\mbox{\makebox[0mm][c]{\hphantom{$\cup$}$\cdot$}$\cup$}}}}V_s$, we say a partition $\{V_{i,j}\}_{i\in[s],j\in[t]}$ is an equitable $({\varepsilon},p)$-regular refinement of $\{V_i\}_{i\in[s]}$ if $\|V_{i,j}\|=\|V_{i,j'}\|\pm 1$ for each $i\in[s]$ and $j,j'\in[t]$, and there are at most ${\varepsilon}s^2t^2$ pairs $(V_{i,j},V_{i',j'})$ which are not $({\varepsilon},0,p)$-regular. \[lem:SRLb\] For each ${\varepsilon}>0$ and $s\in\mathbb{N}$ there exists $t_1\geq 1$ such that the following holds. Given any graph $G$, suppose $V_1{\mathbin{\text{\mbox{\makebox[0mm][c]{\hphantom{$\cup$}$\cdot$}$\cup$}}}}\dots{\mathbin{\text{\mbox{\makebox[0mm][c]{\hphantom{$\cup$}$\cdot$}$\cup$}}}}V_s$ is a partition of $V(G)$. Suppose that $e(V_i)\le 3p\|V_i\|^2$ for each $i\in[s]$, and $e(V_i,V_{i'})\le 2p\|V_i\|\|V_{i'}\|$ for each $i\neq i'\in[s]$. Then there exist sets $V_{i,0}\subset V_i$ for each $i\in[s]$ with $\|V_{i,0}\|<{\varepsilon}\|V_i\|$, and an equitable $({\varepsilon},p)$-regular refinement $\{V_{i,j}\}_{i\in[s],j\in[t]}$ of $\{V_i\setminus V_{i,0}\}_{i\in[s]}$ for some $t\le t_1$. The proof is standard, following Scott’s method [@Scott]. We defer it to Appendix \[app:tools\]. To prove Lemma \[lem:common\], we work as follows. First, we choose a regularity parameter ${{\varepsilon}^{\ast\ast}}_0$ and apply Lemma \[lem:SRLb\] with ${{\varepsilon}^{\ast\ast}}_0$ and the initial partition $V_1\setminus W,\dots,V_k\setminus W,W$. From this partition, all we need is a part $W'\subset W$ and parts $V'_i\subset V_i\setminus W$ for each $i\in[k]$, such that each pair $(W',V'_i)$ is $({{\varepsilon}^{\ast\ast}}_0,d/2,p)$-lower-regular, which we find by averaging. We now choose our vertices $w_1,\dots,w_\Delta$ sequentially (in Claim \[claim:common\]), such that the desired \[cnl:Gsize\]–\[cnl:Nreg\] hold for all subsets of the so far chosen vertices at each stage. This is in spirit very much like the usual dense case ‘Key Lemma’ sequential embedding of vertices using regularity, but in the sparse setting here we need to work somewhat harder and use the regularity inheritance lemmas to show that we can choose vertices which give us lower-regular pairs for future embedding (rather than this being automatic from the slicing lemma, as it is in the dense case). Thus, the proof mainly amounts to showing that the number of vertices which break one of the desired properties and which we therefore cannot choose is always much smaller than $\|W'\|$. In order to show this for \[cnl:Gsize\] we need to maintain some extra properties, specifically sizes of $G$- and $\Gamma$-neighbourhoods of chosen vertices within each $V'_i$, and that these $\Gamma$-neighbourhoods of chosen vertices in each $V'_i$ form lower-regular pairs with $W'$. Note that the way we choose our various regularity parameters amounts to ensuring that, even after $\Delta-1$ successive applications of regularity inheritance lemmas, we still have sufficient regularity for our argument. Furthermore, it is important to note that the choice of ${{\varepsilon}^{\ast\ast}}_0$ does not have anything to to with ${{\varepsilon}^{\ast}}$ or ${\varepsilon}_0$, rather it affects only the returned value of $\alpha$. First we fix all constants that we need throughout the proof. Given $d>0$, $k\geq 1$, and $\Delta\geq 2$, let ${{\varepsilon}^{\ast\ast}}_{\Delta}:=8^{-\Delta}\frac{1}{(k+1)^2}\left(\frac d 8\right)^{\Delta}$. Now, for each $j=1,\dots,\Delta$ sequentially, choose ${{\varepsilon}^{\ast\ast}}_{\Delta-j}\le{{\varepsilon}^{\ast\ast}}_{\Delta-j+1}$ not larger than the ${\varepsilon}_0$ returned by Lemma \[lem:OSRIL\] for input ${{\varepsilon}^{\ast\ast}}_{\Delta-j}$ and $\tfrac{d}{2}$. Now, Lemma \[lem:SRLb\] with input ${{\varepsilon}^{\ast\ast}}_0$ and $s=k+1$ returns $t_1\ge 1$. We set $$\alpha:=\tfrac{1}{2t_1}\big(\tfrac{d}{4}\big)^\Delta\,.$$ Next, given ${{\varepsilon}^{\ast}}>0$, let ${{\varepsilon}^{\ast}}_{\Delta-1,\Delta-1}:={{\varepsilon}^{\ast}}$, and let ${{\varepsilon}^{\ast}}_{j,\Delta}={{\varepsilon}^{\ast}}_{\Delta,j}=1$ for each $1\le j\le\Delta$. For each $(j,j')\in[\Delta]^2\setminus\{(1,1)\}$ in lexicographic order sequentially, we choose $${{\varepsilon}^{\ast}}_{\Delta-j,\Delta-j'}\le\min\{{{\varepsilon}^{\ast}}_{\Delta-j+1,\Delta-j'},{{\varepsilon}^{\ast}}_{\Delta-j,\Delta-j'+1},{{\varepsilon}^{\ast}}_{\Delta-j+1,\Delta-j'+1}\}$$ not larger than the ${\varepsilon}_0$ returned by Lemma \[lem:OSRIL\] for both input ${{\varepsilon}^{\ast}}_{\Delta-j+1,\Delta-j'}$ and $d$, and for input ${{\varepsilon}^{\ast}}_{\Delta-j,\Delta-j'+1}$ and $d$, and not larger than the ${\varepsilon}_0$ returned by Lemma \[lem:TSRIL\] for input ${{\varepsilon}^{\ast}}_{\Delta-j+1,\Delta-j'+1}$ and $d$. We choose ${\varepsilon}_0$ small enough such that $(1+{\varepsilon}_0)^{\Delta} \leq 1+{{\varepsilon}^{\ast}}$ and $(1-{\varepsilon}_0)^{\Delta} \geq 1-{{\varepsilon}^{\ast}}$. Given $r\ge 1$ and ${\varepsilon}$ with $0<{\varepsilon}\le{\varepsilon}_0$, suppose that $C$ is large enough for each of these calls to Lemmas \[lem:OSRIL\] and \[lem:TSRIL\], and for Proposition \[prop:chernoff\] with input ${\varepsilon}_0$. Finally, we set $${C^{\ast}}= 10^{12}k^4t_1r^4{\varepsilon}^{-4}2^{2\Delta}C\,.$$ Given $p\ge {C^{\ast}}\big(\tfrac{\log n}{n}\big)^{1/\Delta}$, a.a.s. the good events of each of the above calls to Lemma \[lem:OSRIL\] and \[lem:TSRIL\], and to Proposition \[prop:chernoff\] and Lemma \[lem:SRLb\], occur. We condition from now on upon these events occurring for $\Gamma=G(n,p)$. Let $G=(V,E)$ be a subgraph of $\Gamma$. Suppose $\{V_i\}_{i\in[k]}$ and $W$ satisfy the conditions of the lemma. We first apply Lemma \[lem:SRLb\], with the promised input parameters ${{\varepsilon}^{\ast\ast}}_0$ and $s=k+1$, to $G[V_1\cup\dots\cup V_k\cup W]$, with input partition $\{V_i\setminus W\}_{i\in[k]}\cup\{W\}$. We can do this because $Cp^{-1}\log n<10^{-10}\frac{{\varepsilon}^4 pn}{k^4r^4}$, so that the good event of Proposition \[prop:chernoff\] guarantees that the conditions of Lemma \[lem:SRLb\] are satisfied. This returns a partition refining each set of $\{V_i\setminus W\}_{i\in[k]}\cup\{W\}$ into $1\le t\le t_1$ clusters together with a small exceptional set. Let $W'\subset W$ be a cluster which is in at most $2k{{\varepsilon}^{\ast\ast}}_0 t$ pairs with clusters in $\big(V_1\cup\dots\cup V_k\big)\setminus W$ which are not $({{\varepsilon}^{\ast\ast}}_0,p)_G$-lower-regular. Such a cluster exists by averaging. By Proposition \[prop:chernoff\] and \[cnl:bal\], at most $4(k+1){{\varepsilon}^{\ast\ast}}_0 p \tfrac{4n}{r}\|W'\|$ edges lie in the pairs between $W'$ and the $V_i$ which are not lower-regular, and by Proposition \[prop:chernoff\] and \[cnl:W\] at most $2p\|W\|\|W'\|<{{\varepsilon}^{\ast\ast}}_0 p\tfrac{n}{r}\|W'\|$ edges leaving $W'$ lie in $W$. By \[cnl:Wdeg\], for each $i\in[k]$ each $w\in W'$ has at least $dp\|V_i\|$ neighbours in $V_i$, and hence there are at least $\tfrac{dp}{2}\|V_i\|\|W'\|$ edges from $W'$ to $V_i\setminus W$ which lie in $({{\varepsilon}^{\ast\ast}}_0,p)_G$-lower-regular pairs. By averaging, for each $i\in[k]$ there exists a cluster $V'_i$ of the partition such that $(W',V_i')$ is $({{\varepsilon}^{\ast\ast}}_0, d/2, p)_G$-lower-regular. For the remainder of the proof, we will only need these $k+1$ clusters from the partition. Notice that for every $i\in[k]$ we have $$\|V_i\| \geq \|V_i'\| \geq \frac{n}{8kt_1r} \geq \frac{1}{8kt_1r} ({C^{\ast}})^{2}p^{-2}\log n \geq {C^{\ast}}p^{-2} \log n$$ and $$\label{eq:sizeW} \|W'\| \geq 10^{-11}\frac{{\varepsilon}^4 pn}{t_1k^4r^4} \geq 10^{-11}\frac{{\varepsilon}^4}{t_1k^4r^4}({C^{\ast}})^{2}p^{-1}\log n \geq {C^{\ast}}p^{-1} \log n$$ both by the choice of ${C^{\ast}}$ and $p$. We choose the $\Delta$-tuple $(w_1,\dots,w_\Delta)$ inductively, using the following claim. \[claim:common\] For each $0\le \ell\le\Delta$ there exists an $\ell$-tuple $(w_1,\ldots,w_\ell) \in \binom{W'}{\ell}$ such that the following holds. For every $\Lambda, \Lambda^\ast \subseteq [\ell]$, and every $i \neq i' \in [k]$ we have 1. \[cnl:cl:Wreg\] $\big(\bigcap_{j\in \Lambda}{N_{\Gamma}}(w_j,V_i'),W'\big)$ is $({{\varepsilon}^{\ast\ast}}_{\|\Lambda\|},\frac d 2, p)_G$-lower-regular if $\|\Lambda\|< \Delta$, 2. \[cnl:cl:NGVp\] $\|\bigcap_{j\in\Lambda} N_G(w_j,V_i')\| \geq \big(\frac{d}{4}\big)^{\|\Lambda\|} p^{\|\Lambda\|}\|V_i'\|$, 3. \[cnl:cl:NGa\] $\|\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j)\| \leq (1+{\varepsilon}_0)^{\|\Lambda\|}p^{\|\Lambda\|} n$, 4. \[cnl:cl:NGaVp\] $\|\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i')\| = (1\pm{\varepsilon}_0)^{\|\Lambda\|}p^{\|\Lambda\|} \|V_i'\|$, 5. \[cnl:cl:NGaV\] $\|\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i)\| = (1\pm{\varepsilon}_0)^{\|\Lambda\|}p^{\|\Lambda\|} \|V_i\|$, and 6. \[cnl:cl:Vreg\] $\big(\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i),\bigcap_{j^\ast\in\Lambda^\ast}{N_{\Gamma}}(w_{j^\ast},V_{i'})\big)$ is $({{\varepsilon}^{\ast}}_{\|\Lambda\|,\|\Lambda^\ast\|},d,p)_G$-lower-regular if\ $\|\Lambda\|,\|\Lambda^\ast\| < \Delta$ and either $\Delta \geq 3$ or $\Lambda \cap \Lambda^\ast = \varnothing$ or both. We prove this claim by induction on $\ell$. Recall that if $\Lambda=\emptyset$ then $\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i')$ is by definition equal to $V'_i$, and that $[0]=\emptyset$. For the base case $\ell=0$, observe that \[cnl:cl:Wreg\] follows from our choice of $W'$ and the $V_i'$. For every $i,j\in [k]$, the pair $(V_i,V_j)$ is $({\varepsilon},d,p)_G$-lower-regular by \[cnl:Vreg\], and since ${\varepsilon}\le{{\varepsilon}^{\ast}}_{0,0}$ this gives \[cnl:cl:Vreg\]. The remaining three properties \[cnl:cl:NGVp\], \[cnl:cl:NGaVp\] and \[cnl:cl:NGaV\] are tautologies for $\ell=0$. For the inductive step, suppose that for some $0\le\ell<\Delta$ there exists an $\ell$-tuple $(w_1, \ldots, w_{\ell}) \in \binom{W'}{\ell}$ satisfying \[cnl:cl:Wreg\]–\[cnl:cl:Vreg\]. We now find a vertex $w_{\ell+1} \in W'$ such that the $(\ell+1)$-tuple $(w_1, \ldots, w_{\ell+1})$ still satisfies \[cnl:cl:Wreg\]–\[cnl:cl:Vreg\]. We do this by determining, for each of these five conditions, an upper bound on the number of vertices in $W'$ that violate them and show that the sum of these upper bounds is less than $\|W'\|-\ell$. Suppose $\Lambda \subseteq [\ell]$ satisfies $\|\Lambda\| < \Delta -1$, and suppose $i \in [k]$. By the choice of $C$ and $p$ we have for every $i \in [k]$ $$\label{eq:common:sizeNGa} \big\|\bigcap_{j\in \Lambda}{N_{\Gamma}}(w_j,V_i')\big\|{ { {\overset{\mbox{\tiny{\eqref{cnl:cl:NGaVp}}}}{\ge}} } } (1-{\varepsilon}_0)^{\|\Lambda\|}p^{\|\Lambda\|}\|V_i'\| \overset{\|\Lambda\| < \Delta-1}\geq (1-{\varepsilon}_0)^{\Delta-2} p^{\Delta-2} \frac{n}{8ktr} \geq C p^{-2} \log n\,.$$ We also have $\|W'\| \geq {C^{\ast}}p^{-1} \log n$ by and $\big(\bigcap_{j\in \Lambda} {N_{\Gamma}}(w_j,V_i'), W'\big)$ is an $({{\varepsilon}^{\ast\ast}}_{\|\Lambda\|}, d/2, p)_G$-lower-regular pair by \[cnl:cl:Wreg\]. Since the good event of Lemma \[lem:OSRIL\] with input ${{\varepsilon}^{\ast\ast}}_{\|\Lambda\|+1}$ and $\tfrac{d}{2}$ occurs, there exist at most $C p^{-1} \log n$ vertices $w$ in $W'$ such that $\big(\bigcap_{j\in \Lambda}{N_{\Gamma}}(w_j,V_i') \cap {N_{\Gamma}}(w), W'\big) = \big(\bigcap_{j\in \Lambda}{N_{\Gamma}}(w_j,V_i') \cap {N_{\Gamma}}(w,V_i'),W'\big)$ is not $({{\varepsilon}^{\ast\ast}}_{\|\Lambda\|+1},\frac d 2, p)_G$-lower-regular. Summing over all possible choices of $\Lambda \subseteq [l]$ and $i \in [k]$, there are at most $2^\Delta k^2 C p^{-1} \log n$ vertices $w$ in $W'$ such that $(w_1, \ldots, w_l, w)$ does not satisfy \[cnl:cl:Wreg\]. Moving on to \[cnl:cl:NGVp\], let $\Lambda\subset[\ell]$ and $i\in[k]$ be given. We have $$\begin{aligned} \big\|\bigcap_{j\in\Lambda}N_G(w_j,V'_i)\big\|&{ { {\overset{\mbox{\tiny{\eqref{cnl:cl:NGVp}}}}{\ge}} } }\big(\tfrac{d}{4}\big)^{\|\Lambda\|}p^{\|\Lambda\|}\|V'_i\|\quad\text{and}\\ \big\|\bigcap_{j\in\Lambda}N_\Gamma(w_j,V'_i)\big\|&{ { {\overset{\mbox{\tiny{\eqref{cnl:cl:NGaVp}}}}{\le}} } }(1+{\varepsilon}_0)^{\|\Lambda\|}p^{\|\Lambda\|}\|V'_i\|\,. \end{aligned}$$ By choice of ${\varepsilon}_0$ and ${{\varepsilon}^{\ast\ast}}_{\|\Lambda\|}$, we thus have $\big\|\bigcap_{j\in\Lambda}N_G(w_j,V'_i)\big\|\ge{{\varepsilon}^{\ast\ast}}_{\|\Lambda\|}\big\|\bigcap_{j\in\Lambda}N_\Gamma(w_j,V'_i)\big\|$. Now by \[cnl:cl:Wreg\], the pair $\big(W',\bigcap_{j\in\Lambda}N_\Gamma(w_j,V'_i)\big)$ is $\big({{\varepsilon}^{\ast\ast}}_{\|\Lambda\|},\tfrac{d}{2},p\big)_G$-lower-regular, and thus the number of vertices $w\in W'$ such that $$\big\|N_G(w,V'_i)\cap\bigcap_{j\in\Lambda}N_G(w_j,V'_i)\big\|<\big(\tfrac{d}{4}\big)^{\|\Lambda\|+1}p^{\|\Lambda\|+1}\|V'_i\|$$ is at most ${{\varepsilon}^{\ast\ast}}_{\|\Lambda\|}\|W'\|\le{{\varepsilon}^{\ast\ast}}_{\Delta}\|W'\|$. Summing over the choices of $\Lambda\subset[\ell]$ and $i\in[k]$, the number of $w\in W'$ violating \[cnl:cl:NGVp\] is at most $2^\Delta k{{\varepsilon}^{\ast\ast}}_{\Delta}\|W'\|$. For \[cnl:cl:NGaVp\], given $\Lambda\subset[\ell]$ and $i\in[k]$, by \[cnl:cl:NGaVp\] we have $$\big\|\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i')\big\| = (1\pm{\varepsilon}_0)^{\|\Lambda\|}p^{\|\Lambda\|} \|V_i'\|\,,$$ and by choice of ${\varepsilon}_0$ and $p$, in particular $\big\|\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i')\big\|\ge C p^{-1}\log n$. Since the good event of Proposition \[prop:chernoff\] occurs, the number of vertices $w\in W'$ such that $\big\|{N_{\Gamma}}(w,V'_i)\cap\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i')\big\|$ is either smaller than $(1-{\varepsilon}_0)^{\|\Lambda\|+1}p^{\|\Lambda\|+1}\|V_i'\|$ or larger than $(1+{\varepsilon}_0)^{\|\Lambda\|+1}p^{\|\Lambda\|+1}\|V_i'\|$ is at most $2C p^{-1}\log n$. Summing over the choices of $\Lambda\subset[\ell]$ and of $i\in[k]$, we conclude that at most $2^{\Delta+1}kC p^{-1}\log n$ vertices of $W'$ violate \[cnl:cl:NGaVp\]. Since $n\ge \|V_i\|\ge\|V'_i\|$, the same calculation shows that a further at most $2^{\Delta+1}kC p^{-1}\log n$ vertices of $W'$ violate \[cnl:cl:NGaV\], and at most $2^{\Delta+1}kC p^{-1}\log n$ vertices of $W'$ violate \[cnl:cl:NGa\]. Finally, we come to \[cnl:cl:Vreg\]. Suppose we are given $\Lambda,\Lambda'\subset[\ell]$ and distinct $i,i'\in[k]$. Suppose that $\|\Lambda\|\le\Delta-2$ and $\|\Lambda'\|\le\Delta-1$. We wish to show that for most vertices $w\in W'$, the pair $\big({N_{\Gamma}}(w,V_i)\cap \bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i),\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V'_i)\big)$ is $\big({{\varepsilon}^{\ast}}_{\|\Lambda\|+1,\|\Lambda'\|},d,p\big)_G$-lower-regular, and furthermore, if $\Delta\ge 3$ and $\|\Lambda'\|\le\Delta-2$, that the pair $\big({N_{\Gamma}}(w,V_i)\cap \bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i),{N_{\Gamma}}(w,V_{i'})\cap\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V'_i)\big)$ is $\big({{\varepsilon}^{\ast}}_{\|\Lambda\|+1,\|\Lambda'\|+1},d,p\big)_G$-lower-regular. By \[cnl:cl:NGaV\], and by choice of ${\varepsilon}_0$, $C$ and $p$, we have $$\begin{aligned} \big\|\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i)\big\|&\ge(1-{\varepsilon}_0)^{\|\Lambda\|} p^{\|\Lambda\|}\|V_i\|\ge C p^{\|\Lambda\|-\Delta}\log n\quad\text{and}\\ \big\|\bigcap_{j\in\Lambda'}{N_{\Gamma}}(w_j,V_{i'})\big\|&\ge(1-{\varepsilon}_0)^{\|\Lambda'\|} p^{\|\Lambda'\|}\|V_{i'}\|\ge C p^{\|\Lambda'\|-\Delta}\log n\,. \end{aligned}$$ By \[cnl:cl:Vreg\], the pair $\big(\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i),\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V'_i)\big)$ is $\big({{\varepsilon}^{\ast}}_{\|\Lambda\|,\|\Lambda'\|},d,p\big)_G$-lower-regular. Since the good event of Lemma \[lem:OSRIL\] with input ${{\varepsilon}^{\ast}}_{\|\Lambda\|+1,\|\Lambda'\|}$ and $d$ occurs, there are at most $C p^{-1}\log n$ vertices $w$ of $W'$ such that $\big({N_{\Gamma}}(w,V_i)\cap\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i),\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V'_i)\big)$ is not $\big({{\varepsilon}^{\ast}}_{\|\Lambda\|+1,\|\Lambda'\|},d,p\big)_G$-lower-regular. Furthermore, if $\|\Lambda'\|\le\Delta-2$, then since the good event of Lemma \[lem:TSRIL\] with input ${{\varepsilon}^{\ast}}_{\|\Lambda\|+1,\|\Lambda'\|+1}$ and $d$ occurs, there are at most $C p^{-2}\log n$ vertices $w$ of $W'$ such that $$\Big({N_{\Gamma}}(w,V_i)\cap\bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V_i),{N_{\Gamma}}(w,V_{i'})\cap \bigcap_{j\in\Lambda}{N_{\Gamma}}(w_j,V'_i)\Big)\text{ is not }\big({{\varepsilon}^{\ast}}_{\|\Lambda\|+1,\|\Lambda'\|},d,p\big)_G\text{-lower-regular.}$$ Observe that if $\Delta=2$ the property \[cnl:cl:Vreg\] does not require this pair to be lower-regular. Summing over the choices of $\Lambda,\Lambda'\subset[\ell]$ and $i,i'\in[k]$, we conclude that if $\Delta=2$ then at most $2^{2\Delta}k^2C p^{-1}\log n$ vertices $w$ of $W'$ cause \[cnl:cl:Vreg\] to fail, while if $\Delta\ge 3$, at most $2^{2\Delta}k^2C(p^{-1}+p^{-2})\log n$ vertices $w$ of $W'$ violate \[cnl:cl:Vreg\]. Summing up, if $\Delta=2$ then at most $$\label{eq:common:bad2} 2^{\Delta}k^2C p^{-1}\log n+2^\Delta k{{\varepsilon}^{\ast\ast}}_{\Delta}\|W'\|+3\cdot 2^{\Delta+1}kC p^{-1}\log n+2^{2\Delta}k^2C p^{-1}\log n$$ vertices $w$ of $W'$ cannot be chosen as $w_{\ell+1}$. By choice of ${C^{\ast}}$ and ${{\varepsilon}^{\ast\ast}}_{\Delta}$, and by choice of $p$, this is at most $\tfrac12\|W'\|$, so that there exists a vertex of $W'$ which can be chosen as $w_{\ell+1}$, as desired. If on the other hand $\Delta\ge 3$, then at most $$\label{eq:common:bad3} 2^{\Delta}k^2C p^{-1}\log n+2^\Delta k{{\varepsilon}^{\ast\ast}}_{\Delta}\|W'\|+3\cdot 2^{\Delta+1}kC p^{-1}\log n+2^{2\Delta}k^2C (p^{-1}+p^{-2})\log n$$ vertices of $W'$ cannot be chosen as $w_{\ell+1}$. Again by choice of ${C^{\ast}}$, ${{\varepsilon}^{\ast\ast}}_{\Delta}$ and $p$, this is at most $\tfrac12\|W'\|$, and again we therefore can choose $w_{\ell+1}$ satisfying \[cnl:cl:Wreg\]–\[cnl:cl:Vreg\] as desired. Finally, let us argue why the lemma is a consequence of Claim \[claim:common\]. Let $(w_1,\ldots, w_\Delta) \in \binom{W'}{\Delta}$ be a tuple satisfying \[cnl:cl:Wreg\]–\[cnl:cl:Vreg\]. By \[cnl:cl:NGVp\], for any $\Lambda\subset[\ell]$ and $i\in[k]$ we have $$\Big\|\bigcap_{j\in\Lambda}N_G(w_j,V_i)\Big\|\ge \big(\tfrac{d}{4}\big)^{\|\Lambda\|} p^{\|\Lambda\|}\|V_i'\|\ge\big(\tfrac{d}{4}\big)^\Delta p^{\|\Lambda\|}\tfrac{\|V_i\|}{2t_1}\ge\alpha p^{\|\Lambda\|}\|V_i\|\,,$$ as required for \[cnl:Gsize\]. Properties \[cnl:Gasizen\], \[cnl:Gasize\] and \[cnl:Nreg\] are respectively \[cnl:cl:NGa\], \[cnl:cl:NGaV\] and \[cnl:cl:Vreg\], by choice of ${\varepsilon}_0$. The balancing lemma {#sec:prooflembalancing} =================== The statement of Lemma \[lem:balancing\] gives us a partition of $V(G)$ with parts $\big(V_{i,j}\big)_{i\in[r],j\in[k]}$, and a collection of ‘target integers’ $\big(n_{i.j}\big)_{i\in[r],j\in[k]}$, with each $n_{i,j}$ close to $\|V_{i,j}\|$, and with $\sum n_{i,j}=\sum \|V_{i,j}\|$. Our aim is to find a partition of $V(G)$ with parts $\big(V'_{i,j}\big)_{i\in[r],j\in[k]}$ such that $\|V'_{i,j}\|=n_{i,j}$ for each $i,j$. This partition is required to maintain similar regularity properties as the original partition, while not substantially changing common neighbourhoods of vertices. There are two steps to our proof. In a first step, we correct global imbalance, that is, we find a partition ${\widetilde{\mathcal{V}}}$ which maintains all the desired properties and which has the property that $\sum_i \|{\widetilde{V}}_{i,j}\|=\sum_i n_{i,j}$ for each $j\in[k]$. To do this, we identify some $j^\ast$ such that $\sum_i \|V_{i,j^\ast}\|>\sum_i n_{i,j^\ast}$ and $j'$ such that $\sum_i \|V_{i,j'}\|<\sum_i n_{i,j'}$. We move $\sum_i \|V_{i,j^\ast}\|- n_{i,j^\ast}$ vertices from $V_{1,j^\ast}$ to some cluster $V_{i',j'}$, maintaining the desired properties, and repeat this procedure until no global imbalance remains. In a second step, we correct local imbalance, that is, for each $i=1,\dots,r-1$ sequentially, and for each $j\in[k]$, we move vertices between ${\widetilde{V}}_{i,j}$ and ${\widetilde{V}}_{i+1,j}$, maintaining the desired properties, to obtain the partition ${\mathcal V}'$ such that $\|V'_{i,j}\|=n_{i,j}$ for each $i,j$. Observe that because ${\widetilde{\mathcal{V}}}$ is globally balanced, once we know $\|V'_{i,j}\|=n_{i,j}$ for each $i\in[r-1]$ and each $j\in[k]$ we are guaranteed that $\|V'_{r,j}\|=n_{r,j}$ for each $j\in[k]$. The proof of the lemma then comes down to showing that we can move vertices and maintain the desired properties. Because we start with a partition in which $V_{i,j}$ is very close to $n_{i,j}$ for each $i$ and $j$, the total number of vertices we move in any step is at most the sum of the differences, which is much smaller than any $n_{i,j}$. The following lemma shows that we can move any small (compared to all $n_{i,j}$) number of vertices from one part to another and maintain the desired properties. \[lem:smallmove\] For all integers $k, r_1, \Delta \geq 1$, and reals $d>0$ and $0<{\varepsilon}<1/2k$ as well as $0 < \xi < 1/(100kr_1^3)$, there exists ${C^{\ast}}>0$ such that the following holds for all sufficiently large $n$. Let $\Gamma$ be a graph on vertex set $[n]$, and let $G$ be a not necessarily spanning subgraph. Let $X,Z_1,\ldots,Z_{k-1} \subseteq V(G)$ be pairwise disjoint subsets, each of size at least $n/(16kr_1)$, such that $(X,Z_i)$ is $({\varepsilon},d,p)_G$-lower-regular for each $i$. Then for each $1\le m\le 2r_1^2\xi n$, there exists a set $S$ of $m$ vertices of $X$ with the following properties. 1. \[smallmove:degG\] For each $v\in S$ we have $\deg_G(v;Z_i)\ge(d-{\varepsilon})p\|Z_i\|$ for each $i\in[k-1]$, and 2. \[smallmove:I\] for each $1\le s\le\Delta$ and every collection of vertices $v_1,\ldots,v_s\in[n]$ we have $$\deg_\Gamma(v_1,\ldots,v_s;S)\le 100kr_1^3\xi\deg_\Gamma(v_1,\ldots,v_s; X)+\tfrac{1}{100}{C^{\ast}}\log n\,.$$ Given $k$, $r_1$, $\Delta$, $d$, $\xi$ and ${\varepsilon}$, let $C$ be returned by Lemma \[lem:hypgeo\] for input $\xi$ and $\Delta$. We set ${C^{\ast}}=100C$. Given $\Gamma$, $G$ and $X$, $Y$, $Z_1,\ldots,Z_{k-1}$, let $X'$ be the set of vertices $v\in X$ such that $\deg_G(v;Z_i)\ge(d-{\varepsilon})p\|Z_i\|$ for each $i\in[k-1]$. Because each pair $(X,Z_i)$ for $i\in[k-1]$ is $({\varepsilon},d,p)_G$-lower-regular, we have $\|X'\|\ge \|X\|-k{\varepsilon}\|X\|\ge\|X\|/2$. We now apply Lemma \[lem:hypgeo\], with input $\xi$, $\Delta$, $W=X'$ and the sets $T_i$ consisting of the sets $N_\Gamma(v_1,\ldots,v_s;X')$ for each $1\le s\le\Delta$ and $v_1,\ldots,v_s\in[n]$, to choose a set $S$ of size $m\leq 2r_1^2\xi n\leq \|X'\|$ in $X'$. We then have $$\begin{aligned} \deg_\Gamma(v_1,\ldots,v_s;S)&\le \left(\tfrac{2r_1^2\xi n}{\|X'\|}+\xi\right)\deg_\Gamma(v_1,\ldots,v_s;X')+C\log n\\ &\le 100kr_1^3\xi\deg_\Gamma(v_1,\ldots,v_s;X)+\tfrac{1}{100}{C^{\ast}}\log n\,, \end{aligned}$$ where the final inequality is by choice of ${C^{\ast}}$, and since $\|X'\|\ge\|X\|/2\ge n/(32kr_1)$. Thus the set $S$ satisfies \[smallmove:I\], and since $S\subseteq X'$ we have \[smallmove:degG\]. We now prove the balancing lemma. Given integers $k, r_1, \Delta \geq 1$ and reals $\gamma, d >0$ and $0< {\varepsilon}< \min\{d, 1/(2k)\}$, we set $$\xi =10^{-15} {\varepsilon}^4d/(k^3r_1^5).$$ Let ${C^{\ast}}_1$ be returned by Lemma \[lem:smallmove\] with input $k$, $r_1$, $\Delta$, $d$, ${\varepsilon}/4$ and $\xi$, and let ${C^{\ast}}_2$ be returned by Lemma \[lem:smallmove\] with input $k$, $r_1$, $\Delta$, $d$, $3{\varepsilon}/4$ and $\xi$. We set ${C^{\ast}}= \max\{{C^{\ast}}_1, {C^{\ast}}_2\}$. Now suppose that $p\ge{C^{\ast}}\big(\frac{\log n}{n}\big)^{1/2}$, that $10\gamma^{-1}\le r\le r_1$, and that graphs $\Gamma$ and $G$, a partition ${\mathcal V}$ of $V=V(G)$, and graphs $R^k_r$, $B^k_r$ and $K^k_r$ on $[r]\times [k]$ as in the statement of Lemma \[lem:balancing\] are given. First stage (global imbalance): We use the following algorithm. In each step where we select $S$, we make use of Lemma \[lem:smallmove\] to do so, with input $k$, $r_1$, $\Delta$, $d$, and ${\varepsilon}/4$, with $X=V_{1,j^\ast}$ and with the $Z_1,\ldots,Z_{k-1}$ being the $V_{i',j''}$ with $j''\neq j'$. We claim that the algorithm completes successfully, in other words that each of the choices is possible. and that Lemma \[lem:smallmove\] is always applicable. In each While loop, since $\sum_{i,j}\|V_{i,j}\|-n_{i,j}=0$ and since the While condition is satisfied, $j^\ast$ satisfies $\sum_{i\in[r]}\|V_{i,j^\ast}\|-n_{i,j^\ast}>0$. Observe that the While loop is run at most $k$ times, since at the end of the While loop in which we selected some $j=j^\ast$ we have $\sum_{i\in[r]}\|V_{i,j^\ast}\|-n_{i,j^\ast}=0$ and therefore we do not select $j$ as either $j^\ast$ or $j'$ in future iterations. It follows that the number of $V_{i,j}$ flagged as changed never exceeds $2k$. Now the set $V_{1,j^\ast}$ has degree at least $\big(k-1+\tfrac{\gamma k}{2}\big)r$ in $R^k_r$, and so there are at least $\gamma k r/2$ indices $i\in[r]$ such that $V_{1,j^\ast}$ is adjacent to each $V_{i,j}$ in $R^k_r$. Since $\gamma k r/2>3k$, in particular we can choose $i'$ such that $V_{1,j^\ast}$ is adjacent to each $V_{i',j}$ in $R^k_r$ and no $V_{i',j}$ is flagged as changed. It follows that each pair $(V_{1,j^\ast},V_{i',j})$ is $\big(\tfrac{{\varepsilon}}{4},d,p\big)_G$-lower-regular and thus it is possible to choose $i'$. It is possible to choose $j'$ since the While condition holds. Finally, we need to show that Lemma \[lem:smallmove\] is always applicable with the given parameters. In each application, the sets denoted $X,Z_1,\ldots,Z_{k-1}$ are parts of the partition ${\mathcal V}$ (so they were not changed by the algorithm yet). It follows that each set has size at least $n/(8kr)>n/(16kr_1)$. Since ${\mathcal V}$ is $\big(\tfrac{\varepsilon}4,d,p)$-lower-regular on $B_k^r$, the pairs $(X,Z_1),\dots,(X,Z_{k-1})$ are $\big(\tfrac{\varepsilon}4,d,p)$-lower-regular as required. Finally, by choice of $j^\ast$ we see that the sizes of the sets $S$ we select in each step are decreasing, so it is enough to show that in the first step we have $\|S\|\le r\xi n$, which follows from \[lembalancing:sizes\]. Thus Lemma \[lem:smallmove\] is applicable in each step, and we conclude that the algorithm indeed completes. We denote the resulting vertex partition by ${\widetilde{\mathcal{V}}}=\{{\widetilde{V}}_{i,j}\}_{i\in[r],j\in[k]}$. We have the following properties. 1. \[claimpVt:sizeparts\] For each $i\in[r]$ and $j\in[k]$ we have $\big\|\|{\widetilde{V}}_{i,j}\|-n_{i,j}\big\|\le 2r\xi n$, 2. \[claimpVt:regular\] ${\widetilde{\mathcal{V}}}$ is $\big(\tfrac{{\varepsilon}}{2},d,p\big)_G$-lower-regular on $R^k_r$ and $\big(\tfrac{{\varepsilon}}{2},d,p\big)_G$-super-regular on $K^k_r$, 3. \[claimpVt:NGa\] For each $i\in[r]$, $j\in[k]$ and $1\le s\le\Delta$ and $v_1,\ldots,v_s\in[n]$ we have $$\|{N_{\Gamma}}(v_1,\ldots,v_s,{\widetilde{V}}_{i,j}) {\triangle}{N_{\Gamma}}(v_1,\ldots,v_s,V_{i,j})\| \leq 100kr_1^3\xi\deg_\Gamma\big(v_1,\ldots,v_s;V(G)\big)+\tfrac{1}{100}{C^{\ast}}\log n\,.$$ Observe that vertices were removed from or added to each $V_{i,j}$ to form ${\widetilde{V}}_{i,j}$ at most once in the running of Algorithm \[alg:global\], and the number of vertices added or removed was at most $r\xi n$. Since $\|V_{i,j}\|$ satisfies \[lembalancing:sizes\], we conclude that \[claimpVt:sizeparts\] holds. Furthermore, the vertices added to or removed from $V_{i,j}$ satisfy \[smallmove:I\] and therefore \[claimpVt:NGa\] holds. Since each set $V_{i,j}$ has size at least $n/(8kr)$, we can apply Proposition \[prop:subpairs3\] with $\mu=\nu=8kr^2\xi$ to each edge of $R^k_r$, concluding that ${\widetilde{\mathcal{V}}}$ is $\big(\tfrac{{\varepsilon}}{2},d,p\big)_G$-lower-regular on $R^k_r$ since $\tfrac{{\varepsilon}}{4}+4\sqrt{8kr^2\xi}<\tfrac{{\varepsilon}}{2}$. Now for any $i\in[r]$ and $j\in[k]$, consider $v\in{\widetilde{V}}_{i,j}$. If $v\not\in V_{i,j}$, then we applied Lemma \[lem:smallmove\] to select $v$, and when we did so no $V_{i,j'}$ was flagged as changed by Algorithm \[alg:global\]. Thus by \[smallmove:degG\] we have $$\deg_G(v;{\widetilde{V}}_{i,j'})=\deg_G(v;V_{i,j'})\ge\big(d-\tfrac{{\varepsilon}}{4}\big)p\|V_{i,j'}\|=\big(d-\tfrac{{\varepsilon}}{4}\big)p\|{\widetilde{V}}_{i,j'}\|$$ for each $j'\neq j$, since $V_{i,j}$ is then flagged as changed and thus $V_{i,j'}={\widetilde{V}}_{i,j'}$ for each $j'\neq j$. If on the other hand $v\in V_{i,j}$, then by \[lembalancing:regular1\] we started with $\deg_G(v;V_{i,j'})\ge\big(d-\tfrac{{\varepsilon}}{4}\big)p\|V_{i,j'}\|$. By \[smallmove:I\] and \[lembalancing:gamma1\], we have $$\deg_G(v;{\widetilde{V}}_{i,j'})\ge \big(d-\tfrac{{\varepsilon}}{4}\big)p\|V_{i,j'}\|-\tfrac{{\varepsilon}^2}{1000kr_1}\big(1+\tfrac{{\varepsilon}}{4}\big)p\|V_{i,j'}\|-\tfrac{1}{100}{C^{\ast}}\log n \ge\big(d-\tfrac{{\varepsilon}}{2}\big)p\|{\widetilde{V}}_{i,j'}\|\,,$$ where the final inequality follows by choice of $n$ sufficiently large and since $\|{\widetilde{V}}_{i,j'}\|\le\|V_{i,j'}\|+r\xi n\le \big(1+\tfrac{{\varepsilon}d}{100}\big)\|V_{i,j'}\|$. We conclude that ${\widetilde{\mathcal{V}}}$ is $\big(\tfrac{{\varepsilon}}{2},d,p\big)$-super-regular on $K^k_r$, giving \[claimpVt:regular\]. Second stage (local imbalance): We use the following algorithm to correct the local imbalances in ${\widetilde{\mathcal{V}}}$. Again, in each step when we select $S$ we make use of Lemma \[lem:smallmove\] to do so. If we select $S$ from ${\widetilde{V}}_{i,j}$, then we use input $k$, $r_1$, $d$, $3{\varepsilon}/4$ and $\xi$ with $X={\widetilde{V}}_{i,j}$ and the sets $Z_1,\ldots,Z_{k-1}$ being ${\widetilde{V}}_{i+1,j'}$ for $j'\neq j$. If on the other hand we select $S$ from ${\widetilde{V}}_{i+1,j}$, then we use input $k$, $r_1$, $d$ and $3{\varepsilon}/4$, with $X={\widetilde{V}}_{i+1,j}$ and the sets $Z_1,\ldots,Z_{k-1}$ being ${\widetilde{V}}_{i,j'}$ for $j'\neq j$. We claim that Lemma \[lem:smallmove\] is always applicable. To see that this is true, observe first that the number of vertices which we move between any ${\widetilde{V}}_{i,j}$ and ${\widetilde{V}}_{i+1,j}$ in a given step is by \[claimpVt:sizeparts\] bounded by $2k^2r^2\xi n$. We change any given ${\widetilde{V}}_{i,j}$ at most twice in the running of the algorithm, so that in total at most $4k^2r^2\xi n$ vertices are changed. In particular, we maintain $\|{\widetilde{V}}_{i,j}\|\ge n/(16kr_1)$ throughout, and, by Proposition \[prop:subpairs3\], with input $\mu=\nu=\tfrac{4r^2\xi n}{n/(16kr_1)}<100r_1^3k\xi$, and using \[claimpVt:regular\], we maintain the property that any pair in $R^k_r$, and in particular any pair in $B^k_r$, is $\big(\tfrac{3{\varepsilon}}{4},d,p\big)$-lower-regular throughout. This shows that Lemma \[lem:smallmove\] is always applicable, and therefore the algorithm completes and returns a partition ${\mathcal V}'$. We claim that this is the desired partition. We need to check that \[lembalancing:sizesout\]—\[lembalancing:gammaout\] hold. Since for each $j\in[k]$ we have $\sum_i\|V'_{i,j}\|=\sum_i\|{\widetilde{V}}_{i,j}\|=\sum_in_{i,j}$, and since $\|V'_{i,j}\|=n_{i,j}$ for each $i\in[r-1]$ and $j\in[k]$, we conclude that $\|V'_{i,j}\|=n_{i,j}$ for all $i$ and $j$, giving \[lembalancing:sizesout\]. For the first part of \[lembalancing:regular\], we have justified that we maintain $\big(\tfrac{3{\varepsilon}}{4},d,p\big)_G$-lower-regularity on $R^k_r$ throughout the algorithm. For the second part, we need to show that for each $i\in[r]$ and $j\neq j'\in[k]$, and each $v\in V'_{i,j}$, we have $\deg_G(v;V'_{i,j'})\ge(d-{\varepsilon})p\|V'_{i,j'}\|$. If $v\in{\widetilde{V}}_{i,j}$, then by \[claimpVt:regular\] we have $\deg_G(v;{\widetilde{V}}_{i,j'})\ge\big(d-\tfrac{{\varepsilon}}{2}\big)p\|{\widetilde{V}}_{i,j'}\|$. We change ${\widetilde{V}}_{i,j'}$ at most twice to obtain $V'_{i,j'}$, both times by adding or removing vertices satisfying \[smallmove:I\]. As in the proof of Claim \[claimpVt:NGa\] above, using \[lembalancing:gamma1\] and \[claimpVt:NGa\] we obtain $\deg_G(v;{\widetilde{V}}_{i,j'})\ge(d-{\varepsilon})p\|V'_{i,j}\|$ as desired. If $v\not\in{\widetilde{V}}_{i,j}$, then it was added to the set ${\widetilde{V}}_{i,j}$ by Algorithm \[alg:local\], and ${\widetilde{V}}_{i,j'}$ was changed at most twice thereafter. Again, using \[smallmove:degG\], \[smallmove:I\], \[lembalancing:gamma1\] and \[claimpVt:NGa\] we obtain $\deg_G(v;{\widetilde{V}}_{i,j'})\ge(d-{\varepsilon})p\|V'_{i,j}\|$ as desired. Now \[lembalancing:symd\] holds since the total number of vertices moved in Algorithm \[alg:global\] is at most $k^2r\xi n$, in Algorithm \[alg:local\] at most $4k^2r^2\xi n$ vertices are changed in each cluster, and by choice of $\xi$. To see that \[lembalancing:inheritance\] holds, observe that by \[lembalancing:gamma1\], \[claimpVt:NGa\] and \[smallmove:I\] we have $$\big\|N_\Gamma(v;V_{i,j})\Delta N_\Gamma(v;V'_{i,j}) \big\|\le \tfrac{{\varepsilon}^2}{100kr_1}\deg_\Gamma\big(v;V(G)\big)+\tfrac{1}{10}{C^{\ast}}\log n\le \tfrac{{\varepsilon}^2}{50}\deg_\Gamma(v;V_{i,j})$$ where the final inequality follows by choice of $p$ and of $n$ sufficiently large. Using \[lembalancing:inheritance1\], we can apply Proposition \[prop:subpairs3\], with $\mu=\nu=\tfrac{{\varepsilon}^2}{50}$, to deduce \[lembalancing:inheritance\]. For \[lembalancing:gammaout\], observe that for any given $i\in[r]$ and $j\in[k]$ we change ${\widetilde{V}}_{i,j}$ at most twice in the running of Algorithm \[alg:local\], both times either adding or removing a set satisfying \[smallmove:I\]. By \[claimpVt:NGa\] and choice of $\xi$, we conclude that \[lembalancing:gammaout\] holds. Finally, suppose that for any two disjoint vertex sets $A,A'\subset V(\Gamma)$ with $\|A\|,\|A'\|\ge \tfrac{1}{50000kr_1}{\varepsilon}^2\xi pn$ we have $e_\Gamma(A,A')\le \big(1+\tfrac{1}{100}{\varepsilon}^2\xi\big)p\|A\|\|A'\|$. In each application of Proposition \[prop:subpairs3\] we have $\mu,\nu\ge\tfrac{1}{50}{\varepsilon}^2\xi$, and, and if we have ‘regular’ in place of ‘lower-regular’ in \[lembalancing:regular1\], and \[lembalancing:inheritance1\], we always apply Proposition \[prop:subpairs3\] to a regular pair with sets of size at least $\tfrac{{\varepsilon}}{1000kr_1}pn$, so it returns regular pairs for \[lembalancing:regular\], and \[lembalancing:inheritance\], as desired. The Bandwidth Theorem in random graphs {#sec:proofmain} ====================================== Before embarking on the proof, we first recall from the proof overview (Section \[subsec:over\]) the main ideas. Given $G$, we first use the lemma for $G$ (Lemma \[lem:G\]) to find a lower-regular partition of $V(G)$, with an extremely small exceptional set $V_0$, and whose reduced graph $R^k_r$ contains a spanning backbone graph $B^k_r$, on whose subgraph $K^k_r$ the graph $G$ is super-regular and has one- and two-sided inheritance. Given this, and $H$ together with a $(z,\beta)$-zero-free $(k+1)$-colouring, we use the lemma for $H$ (Lemma \[lem:H2\]) to find a homomorphism $f$ from $V(H)$ to $R^k_r$ almost all of whose edges are mapped to $K^k_r$ and in which approximately the ‘right’ number of vertices of $H$ are mapped to each vertex of $R^k_r$. At this point, if $V_0$ were empty, and if the ‘approximately’ were exact, we would apply the sparse blow-up lemma (Lemma \[thm:blowup\]) to obtain an embedding of $H$ into $G$. Our first aim is to deal with $V_0$. We do this one vertex at a time. Given $v\in V_0$, we choose $x\in V(H)$ from the first $\beta n$ vertices of the supplied bandwidth order ${\mathcal L}$ which is not in any triangles. We embed $x$ to $v$. We then embed the neighbours of $x$ to carefully chosen neighbours of $v$, which we obtain using Lemma \[lem:common\]. Here we use the fact that $N_H(x)$ is independent. This then fixes a clique of $K^k_r$ to which $N^2_H(x)$ must be assigned, and gives image restrictions in the corresponding parts of the lower-regular partition for these vertices. Since $N^2_H(x)$ may have been assigned by $f$ to some quite different clique in $K^k_r$, we have to adjust $f$ to match. This we can do using the fact, which follows from our assumptions on ${\mathcal L}$, that $x$ is far from vertices of colour zero. Now the idea is simply to repeat the above procedure, choosing vertices of $V(H)$ to pre-embed which are widely separated in $H$, until we pre-embedded vertices to all of $V_0$. We end up with a homomorphism $f^$ from what remains of $V(H)$ to $R^k_r$. It is easy to check that this homomorphism still maps about the right number of vertices of $H$ to each vertex of $R^k_r$, simply because $V_0$ is small. We now apply the Balancing Lemma (Lemma \[lem:balancing\]) to correct the sizes of the clusters to match $f^$, and complete the embedding of $H$ using the Sparse Blow-up Lemma (Lemma \[thm:blowup\]). There are two difficulties with this idea, the ‘subtleties’ mentioned in the proof overview (Section \[subsec:over\]). First, if $\Delta=2$ we might have $\|V_0\|\gg pn$, so that we should be worried that at some stage of the pre-embedding we choose $v\in V_0$ and discover most or all of its neighbours have already been pre-embedded to. It turns out to be easy to resolve this: we choose each $v\in V_0$ not arbitrarily, but by taking those which have least available neighbours first. We will show that this is enough to avoid the problem. More seriously, because we perform the pre-embedding sequentially, we might use up a significant fraction of $N_G(w)$ for some $w\in V(G)$ in the pre-embedding, destroying super-regularity of $G$ on $K^k_r$, or we might use up a significant fraction of some common neighbourhood which defines an image restriction for the sparse blow-up lemma. In order to avoid this, before we begin the pre-embedding we fix a set $S\subset V(G)$ whose size is a very small constant times $n$, chosen using Lemma \[lem:hypgeo\] to not have a large intersection with any $N_G(w)$ or with any $\Gamma$-common neighbourhood of at most $\Delta$ vertices of $\Gamma$ (which could define an image restriction). We perform the pre-embedding as outlined above, except that we choose our neighbours of each $v$ within $S$. This procedure is guaranteed not to use up neighbourhood sets guaranteed by super-regularity or image restriction sets, since these sets are all contained in $V\setminus V_0$ and even using up all of $S$ would not be enough to do damage. Given $\gamma>0$, $\Delta\ge 2$ and $k\ge 2$, let $d$ be returned by Lemma \[lem:G\], with input $\gamma$, $k$ and $r_0:=10\gamma^{-1}$. Let $\alpha$ be returned by Lemma \[lem:common\] with input $d$, $k$ and $\Delta$. We set $D=\Delta$, and let ${{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{BL}$}}}>0$ and $\rho>0$ be returned by Lemma \[thm:blowup\] with input $\Delta$, $\Delta_{R'}=3k$, $\Delta_J=\Delta$, $\vartheta=\tfrac{1}{100D}$, $\zeta=\tfrac14\alpha$, $d$ and $\kappa=64$. Next, putting ${{\varepsilon}^{\ast}}=\tfrac18{{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{BL}$}}}$ into Lemma \[lem:common\] returns ${\varepsilon}_0>0$. We choose ${\varepsilon}=\min\big({\varepsilon}_0,d,\tfrac{1}{4D}{{\varepsilon}^{\ast}},\tfrac{1}{100k}\big)$. Putting ${\varepsilon}$ into Lemma \[lem:G\] returns $r_1$. Next, Lemma \[lem:balancing\], for input $k$, $r_1$, $\Delta$, $\gamma$, $d$ and $8{\varepsilon}$, returns $\xi>0$. We assume without loss of generality that $\xi\le 1/(10kr_1)$, and set $\beta=10^{-12}\xi^2/(\Delta k^4r_1^2)$. Let $\mu=\tfrac{{\varepsilon}^2}{100000kr}$. Finally, suppose ${C^{\ast}}$ is large enough for each of these lemmas, for Lemma \[thm:blowup\], for Proposition \[prop:chernoff\] with input ${\varepsilon}$, and for Lemma \[lem:hypgeo\] with input ${\varepsilon}\mu^2$ and $\Delta$. We set $C=10^{10}k^2r_1^2{\varepsilon}^{-2}\xi^{-1}\Delta^{2r_1+20}\mu^{-\Delta}{C^{\ast}}$, and $z=10/\xi$. Given $p\ge C\big(\tfrac{\log n}{n}\big)^{1/\Delta}$, a.a.s. $G(n,p)$ satisfies the good events of Lemma \[thm:blowup\], Lemma \[lem:G\] and Lemma \[lem:common\], and Proposition \[prop:chernoff\], with the stated inputs. Suppose that $\Gamma=G(n,p)$ satisfies these good events. Suppose $G\subseteq \Gamma$ is any spanning subgraph with $\delta(G) \geq \big(\tfrac{k-1}{k}+\gamma\big)pn$. Let $H$ be a graph on $n$ vertices with $\Delta(H) \leq \Delta$, and $\mathcal{L}$ be a labelling of vertex set $V(H)$, of bandwidth at most $\beta n$, such that the first $\beta n$ vertices of $\mathcal L$ include $Cp^{-2}$ vertices that are not contained in any triangles of $H$, and such that there exists a $(k+1)$-colouring that is $(z, \beta)$-zero-free with respect to $\mathcal L$, and the colour zero is not assigned to the first $\sqrt{\beta}n$ vertices. Applying Lemma \[lem:G\] to $G$, with input $\gamma$, $k$, $r_0$ and ${\varepsilon}$, we obtain an integer $r$ with $10\gamma^{-1}\le kr\le r_1$, a set $V_0\subset V(G)$ with $\|V_0\|\le {C^{\ast}}p^{-2}$, a $k$-equitable partition ${\mathcal V}=\{{V_{i,j}}\}_{i\in[r],j\in[k]}$ of $V(G)\setminus V_0$, and a graph $R^k_r$ on vertex set $[r]\times[k]$ with minimum degree $\delta(R^k_r)\ge\big(\tfrac{k-1}{k}+\tfrac{\gamma}{2}\big)kr$, such that $K^k_r\subset B^k_r\subset R^k_r$, and such that 1. \[main:Gsize\] $\frac{n}{4kr}\leq \|{V_{i,j}}\| \leq \frac{4n}{kr}$ for every $i\in[r]$ and $j\in[k]$, 2. \[main:Greg\] ${\mathcal V}$ is $({\varepsilon},d,p)_G$-lower-regular on $R^k_r$ and $({\varepsilon},d,p)_G$-super-regular on $K^k_r$, 3. \[main:Ginh\] both $\big({N_{\Gamma}}(v, V_{i,j}),V_{i',j'}\big)$ and $\big({N_{\Gamma}}(v, V_{i,j}),{N_{\Gamma}}(v, V_{i',j'})\big)$ are $({\varepsilon}, d,p)_G$-lower-regular pairs for every $\{(i,j),(i',j')\} \in E(R^k_r)$ and $v\in V\setminus V_0$, and 4. \[main:Ggam\] $\|{N_{\Gamma}}(v,V_{i,j})\| = (1 \pm {\varepsilon})p\|{V_{i,j}}\|$ for every $i \in [r]$, $j\in [k]$ and every $v \in V \setminus V_0$. Given $i\in[r]$, because $\delta(R^k_r)>(k-1)r$, there exists $v\in V(R^k_r)$ adjacent to each $(i,j)$ with $j\in[k]$. This, together with our assumptions on $H$, allow us to apply Lemma \[lem:H2\] to $H$, with input $D$, $k$, $r$, $\tfrac{1}{10}\xi$ and $\beta$, and with $m_{i,j}:=\|V_{i,j}\|+\tfrac{1}{kr}\|V_0\|$ for each $i\in[r]$ and $j\in[k]$, choosing the rounding such that the $m_{i,j}$ form a $k$-equitable integer partition of $n$. Since $\Delta(H)\le\Delta$, in particular $H$ is $\Delta$-degenerate. Let $f\colon V(H) \to [r] \times [k]$ be the mapping returned by Lemma \[lem:H2\], let $W_{i,j} := f^{-1}(i,j)$, and let $X \subseteq V(H)$ be the set of special vertices returned by Lemma \[lem:H2\]. For every $i\in [r]$ and $j\in [k]$ we have 1. \[H:size\] $m_{i,j} - \tfrac{1}{10}\xi n \leq \|W_{i,j}\| \leq m_{i,j} + \tfrac{1}{10}\xi n$, 2. \[H:sizeX\] $\|X\| \leq \xi n$, 3. \[H:edge\] $\{f(x),f(y)\} \in E(R^k_r)$ for every $\{x,y\} \in E(H)$, 4. \[H:special\] $y,z\in \bigcup_{j'\in[k]}f^{-1}(i,j')$ for every $x\in f^{-1}(i,j)\setminus X$ and $xy,yz\in E(H)$, and 5. \[H:v1\] $f(x)=\big(1,\sigma(x)\big)$ for every $x$ in the first $\sqrt{\beta}n$ vertices of $\mathcal{L}$. Lemma \[lem:H2\] actually gives a little more, which we do not require for this proof. We let $F$ be the first $\beta n$ vertices of $\mathcal{L}$. By definition of $\mathcal{L}$, in $F$ there are at least $C p^{-2}$ vertices whose neighbourhood in $H$ is independent. Next, we apply Lemma \[lem:hypgeo\], with input ${\varepsilon}\mu^2$ and $\Delta$, to choose a set $S\subset V(G)$ of size $\mu n$. We let the $T_i$ of Lemma \[lem:hypgeo\] be all sets which are common neighbourhoods in $\Gamma$ of at most $\Delta$ vertices of $\Gamma$, together with the sets $V_{i,j}$ for $i\in[r]$ and $j\in[k]$. The result of Lemma \[lem:hypgeo\] is that for any $1\le\ell\le\Delta$ and vertices $u_1,\dots,u_\ell$ of $V(G)$, we have $$\label{eq:intS} \begin{split} \Big\|S\cap\bigcap_{1\le i\le\ell}N_\Gamma(u_i)\Big\|&=(1\pm{\varepsilon}\mu)\mu\Big\|\bigcap_{1\le i\le\ell}N_\Gamma(u_i)\Big\|\pm {\varepsilon}\mu p^\ell n\,,\quad \text{and}\\ \big\|S\cap V_{i,j}\big\|&\le 2\mu\|V_{i,j}\|\quad\text{for each $i\in[r]$ and $j\in[k]$,} \end{split}$$ where we use the fact $p\ge C\big(\tfrac{\log n}{n}\big)^{1/\Delta}$ and choice of $C$ to deduce ${C^{\ast}}\log n<{\varepsilon}\mu p^\Delta n$. Our next task is to create the pre-embedding that covers the vertices of $V_0$. We use the following algorithm, starting with $\phi_0$ the empty partial embedding. Set $t:=0$ Suppose this algorithm does not fail, terminating with $t=t^$. The final $\phi_{t^}$ is an embedding of some vertices of $H$ into $V(G)$ which covers $V_0$ and is contained in $V_0\cup S$. Before we specify how exactly we choose vertices at line \[line:choosenbs\], we justify that the algorithm does not fail. In other words, we need to justify that at every time $t$ there are vertices of $F$ whose neighbourhood is independent and which are not close to any vertices in ${\mathrm{dom}}(\phi_t)$, and that at every time $t$, the set $\big(N_G(v)\cap S\big)\setminus{\mathrm{im}}(\phi_t)$ is big. For the first, observe that since $\|V_0\|\le{C^{\ast}}p^{-2}$, we have ${\mathrm{dom}}(\phi_t)\le{C^{\ast}}\Delta p^{-2}$ at every step. Thus the number of vertices at distance less than $2r+20$ from ${\mathrm{dom}}(\phi_t)$ is at most $$\big(1+\Delta+\dots+\Delta^{2r+19}\big){C^{\ast}}\Delta p^{-2}< 2{C^{\ast}}\Delta^{2r+20} p^{-2}$$ which by choice of $C$ is smaller than the number of vertices in $F$ with $N_H(x)$ independent. For the second part, suppose that at some time $t$ we pick a vertex $v$ such that $\big\|\big(N_G(v)\cap S\big)\setminus{\mathrm{im}}(\phi_t)\big\|<\tfrac14\mu pn$. For each $t-\tfrac{1}{100(\Delta+1)}\mu pn\le t'<t$, we have $\big\|\big(N_G(v)\cap S\big)\setminus{\mathrm{im}}(\phi_{t'})\big\|<\tfrac3{10}\mu pn$, yet at each of these times $v$ is not picked, so that the vertex picked at each time has at most as many uncovered neighbours in $S$ as $v$. Let $Z$ be the set of vertices chosen at line \[line:choosev\] in each of these time steps. Then for each $z\in Z$ we have $\big\|\big(N_G(v)\cap S\big)\setminus{\mathrm{im}}(\phi_t)\big\|\le\tfrac3{10}\mu pn$. But since $\delta(G)>\tfrac12pn$, by and choice of ${\varepsilon}$ we have $\big\|N_G(z)\cap S\big\|\ge \tfrac25\mu pn$, so $\big\|N_G(z)\cap{\mathrm{im}}(\phi_t)\big\|\ge \tfrac1{10}\mu pn$ for each $z\in Z$. By choice of $C$, we have $\|Z\|=\tfrac{1}{100(\Delta+1)}\mu pn \ge {C^{\ast}}p^{-1}\log n$. Since $\|{\mathrm{im}}(\phi)\|\le(\Delta+1)\|V_0\|\le\tfrac{1}{100}\mu n$, by choice of $C$, this contradicts the good event of Proposition \[prop:chernoff\]. We have justified that Algorithm \[alg:pre\] completes, and indeed that at each time we reach line \[line:choosenbs\] there are at least $\tfrac14\mu pn$ vertices of $\big(N_G(v)\cap S\big)\setminus{\mathrm{im}}(\phi)$ to choose from. In order to specify how to choose these vertices, we need the following claim. \[cl:chooseW\] Given any set $Y$ of $\tfrac14\mu pn$ vertices of $V(G)$, there exists $W\subset Y$ of size at least $\tfrac{1}{8r}\mu pn$ and an index $i\in[r]$ with the following property. For each $w\in W$ and each $j\in[k]$, we have $\|N_G(w,V_{i,j})\|\ge dp\|V_{i,j}\|$. First let $Y'$ be obtained from $Y$ by removing all vertices $y\in Y$ such that either $\|N_\Gamma(y,V_0)\|\ge {\varepsilon}p n$, or for some $i\in[r]$ and $j\in[k]$ we have $\big\|N_\Gamma(y,V_{i,j})\big\|\neq (1\pm{\varepsilon})p\|V_{i,j}\|$. Because the good event of Proposition \[prop:chernoff\] occurs, the total number of vertices removed is at most $2kr {C^{\ast}}p^{-1}\log n<\tfrac12\|Y\|$, where the inequality is by choice of $C$. Now given any $y\in Y'$, if for each $i\in[r]$ there is $j\in[k]$ such that $\big\|N_G(y,V_{i,j})\big\|<dp\|V_{i,j}\|$, then, since the $\{V_{i,j}\}$ are $k$-equitable, we have $\|N_G(y)\|\le {\varepsilon}p n+dpn+(1+{\varepsilon})\tfrac{k-1}{k}pn+r<\big(\tfrac{k-1}{k}+\gamma\big)pn$, a contradiction. We conclude that for each $y\in Y'$ there exists $i\in[r]$ such that $\|N_G(y,V_{i,j})\|\ge dp\|V_{i,j}\|$ for each $j\in[k]$. We let $W$ be the vertices of $Y'$ giving a majority choice of $i$. Now at each time $t$, in line \[line:choosenbs\] of Algorithm \[alg:pre\], we choose the vertices $w_1,\dots,w_\ell$ as follows. Let $Y=\big(N_G(v_t)\cap S\big)\setminus{\mathrm{im}}(\phi_t)$. Let $i_t\in[r]$ be an index, and $W\subset Y$ be a set of size $\tfrac{1}{8r}\mu pn$, such that $\big\|N_G(w,V_{i_t,j})\big\|\ge dpn\|V_{i_t,j}\|$ for each $j\in[k]$, whose existence is guaranteed by Claim \[cl:chooseW\]. By construction, and by our choice of $\mu$, we can apply Lemma \[lem:common\] with input $d$, $k$, $\Delta$, ${{\varepsilon}^{\ast}}$, $r$ and ${\varepsilon}$, with the clusters $\big\{V_{i_t,j}\big\}_{j\in[k]}$ as the $\big\{V_i\big\}_{i\in[k]}$, and inputting a subset of $W$ of size $10^{-10}\tfrac{{\varepsilon}^4pn}{k^4r^4}$ as requried for \[cnl:W\]. This last is possible by choice of $\mu$. To verify the conditions of Lemma \[lem:common\], observe that \[cnl:bal\] follows from \[main:Gsize\], \[cnl:Vreg\] from \[main:Greg\], and \[cnl:Wdeg\] from Claim \[cl:chooseW\]. We obtain a $\Delta$-tuple of vertices in $W$ satisfying \[cnl:Gsize\]–\[cnl:Nreg\]. We let $w_1,\dots,w_\ell$ be the first $\ell$ vertices of this tuple. Let $H'=H- {\mathrm{dom}}(\phi_{t^})$. We next define image restricting vertex sets and create an updated homomorphism $f^:V(H')\to [r]\times[k]$. For each $x\in V(H)\setminus{\mathrm{dom}}(\phi_{t^})$, let $J_x=\phi_{t^}\big(N_H(x)\cap{\mathrm{dom}}(\phi_{t^})\big)$. Now, since the vertices $\{x_t\}_{t\in[t^]}$ are by construction at pairwise distance at least $2r+20$, in particular for each $y\in V(H')$ with $J_y\neq\emptyset$ the vertex $y$ is at distance two from one $x_t$, and at distance greater than $r+10$ from all others. Let $j\in [k]$ such that $f(y)=(1,j)$. Then we set $f^(y):=(i_{t},j)$. Next, for each $t\in[t^]$ and each $z\in V(H)$ at distance $3,\dots,i_t+1$ from $x_t$, we set $f^(z)$ as follows. Recall that $f(z)=(1,j)$ for some $j\in[k]$. We set $f^(z)=\big(i_t+2-{\mathrm{dist}}(x_t,z),j\big)$. Because the $\{x_t\}$ are at pairwise distance at least $2r+20$, no vertex is at distance $r+5$ or less from any two $x_t$ and $x_{t'}$, so that $f^$ is well-defined. Because $R^k_r$ contains $B^k_r$, the $f^$ we constructed so far is a graph homomorphism. Furthermore, for each $x_t$ the set of vertices $z$ at distance $i_t+1$ from $x_t$ are in the first $\sqrt{\beta}n$ vertices of ${\mathcal L}$, and so by \[H:v1\] satisfy $f^(z)=f(z)$. We complete the construction of $f^$ by setting $f^(z)=f(z)$ for each remaining $z\in V(H)\setminus{\mathrm{dom}}(\phi_{t^})$. Because $f$ is a graph homomorphism, $f^$ is also a graph homomorphism whose domain is $V(H')$. For each $i\in[r]$ and $j\in[k]$, let $W'_{i,j}$ be the set of vertices $w\in V(H')$ with $f^(w)\in V_{i,j}$, and let $X'$ consist of $X$ together with all vertices of $H'$ at distance $r+10$ or less from some $x_t$ with $t\in[t^]$. The total number of vertices $z\in V(H)$ at distance at most $r+10$ from some $x_t$ is at most $2\Delta^{r+10}\|V_0\|<\tfrac{1}{100}\xi n$. Since $W_{i,j}{\triangle}W'_{i,j}$ contains only such vertices, we have 1. \[Hp:sizeWp\] $m_{i,j}-\tfrac15\xi n\le \|W'_{i,j}\|\le m_{i,j}+\tfrac15\xi n$, 2. \[Hp:sizeX\] $\|X'\| \leq 2\xi n$, 3. \[Hp:edge\] $\{f^(x),f^(y)\} \in E(R^k_r)$ for every $\{x,y\} \in E(H')$, and 4. \[Hp:special\] $y,z\in \bigcup_{j'\in[k]}W'_{i,j'}$ for every $x\in W'_{i,j}\setminus X'$ and $xy,yz\in E(H')$. where \[Hp:sizeX\], \[Hp:edge\] and \[Hp:special\] hold by \[H:sizeX\] and definition of $X'$, by definition of $f^$, and by \[H:special\] and choice of $X'$ respectively. Furthermore, we have 1. $\frac{n}{4kr}\leq \|{V_{i,j}}\| \leq \frac{4n}{kr}$ for every $i\in[r]$ and $j\in[k]$, 2. ${\mathcal V}$ is $({\varepsilon},d,p)_G$-lower-regular on $R^k_r$ and $({\varepsilon},d,p)_G$-super-regular on $K^k_r$, 3. both $\big({N_{\Gamma}}(v, V_{i,j}),V_{i',j'}\big)$ and $\big({N_{\Gamma}}(v, V_{i,j}),{N_{\Gamma}}(v, V_{i',j'})\big)$ are $({\varepsilon}, d,p)_G$-lower-regular pairs for every $\{(i,j),(i',j')\} \in E(R^k_r)$ and $v\in V\setminus V_0$, and 4. $\|{N_{\Gamma}}(v,V_{i,j})\| = (1\pm {\varepsilon})p\|{V_{i,j}}\|$ for every $i \in [r]$, $j\in [k]$ and every $v \in V \setminus V_0$. 5. \[main:GpI\] $\big\|V_{f^(x)}\cap\bigcap_{u\in J_x}N_G(u)\big\|\ge\alpha p^{\|J_x\|}\|V_{f^(x)}\|$ for each $x\in V(H')$, 6. \[main:GpGI\] $\big\|V_{f^(x)}\cap\bigcap_{u\in J_x}N_\Gamma(u)\big\|=(1\pm{\varepsilon}^)p^{\|J_x\|}\|V_{f^(x)}\|$ for each $x\in V(H')$, and 7. \[main:GpIreg\] $\big(V_{f^(x)}\cap\bigcap_{u\in J_x}N_\Gamma(u),V_{f^(y)}\cap\bigcap_{v\in J_y}N_\Gamma(v)\big)$ is $({\varepsilon}^,d,p)_G$-lower-regular for each $xy\in E(H')$. 8. \[main:GaI\] $\big\|\bigcap_{u\in J_x}N_\Gamma(u)\big\|\le(1+{{\varepsilon}^{\ast}}) p^{\|J_x\|}n$ for each $x\in V(H')$, Properties \[main:Gsize\] to \[main:Ggam\] are repeated for convenience. Properties \[main:GpI\], \[main:GpGI\] and \[main:GaI\], are trivial when $J_x=\emptyset$, and are otherwise guaranteed by Lemma \[lem:common\]. Finally \[main:GpIreg\] follows from \[main:Greg\] when $J_x,J_y=\emptyset$, and otherwise is guaranteed by Lemma \[lem:common\]. For each $i\in[r]$ and $j\in[k]$, let $V'_{i,j}=V_{i,j}\setminus{\mathrm{im}}(\phi_{t^})$, and let ${\mathcal V}'=\{V'_{i,j}\}_{i\in[r],j\in[k]}$. Because $V_{i,j}\setminus V'_{i,j}\subset S$ for each $i\in[r]$ and $j\in[k]$, using and Proposition \[prop:subpairs3\], and our choice of $\mu$, we obtain 1. \[Gp:sizeV\] $\frac{n}{6kr}\leq \|V'_{i,j}\| \leq \frac{6n}{kr}$ for every $i\in[r]$ and $j\in[k]$, 2. \[Gp:Greg\] ${\mathcal V}'$ is $(2{\varepsilon},d,p)_G$-lower-regular on $R^k_r$ and $(2{\varepsilon},d,p)_G$-super-regular on $K^k_r$, 3. \[Gp:Ginh\] both $\big({N_{\Gamma}}(v, V'_{i,j}),V'_{i',j'}\big)$ and $\big({N_{\Gamma}}(v, V'_{i,j}),{N_{\Gamma}}(v, V'_{i',j'})\big)$ are $(2{\varepsilon}, d,p)_G$-lower-regular pairs for every $\{(i,j),(i',j')\} \in E(R^k_r)$ and $v\in V\setminus V_0$, and 4. \[Gp:GsGa\] $\|{N_{\Gamma}}(v,V'_{i,j})\| = (1 \pm 2{\varepsilon})p\|V_{i,j}\|$ for every $i \in [r]$, $j\in [k]$ and every $v \in V \setminus V_0$. 5. \[Gp:sizeI\] $\big\|V'_{f^(x)}\cap\bigcap_{u\in J_x}N_G(u)\big\|\ge\tfrac12\alpha p^{\|J_x\|}\|V'_{f^(x)}\|$, 6. \[Gp:sizeGa\] $\big\|V'_{f^(x)}\cap\bigcap_{u\in J_x}N_\Gamma(u)\big\|=(1\pm2{\varepsilon}^)p^{\|J_x\|}\|V'_{f^(x)}\|$, and 7. \[Gp:Ireg\] $\big(V'_{f^(x)}\cap\bigcap_{u\in J_x}N_\Gamma(u),V'_{f^(y)}\cap\bigcap_{v\in J_y}N_\Gamma(v)\big)$ is $(2{\varepsilon}^,d,p)_G$-lower-regular. 8. \[Gp:GaI\] $\big\|\bigcap_{u\in J_x}N_\Gamma(u)\big\|\le(1+2{{\varepsilon}^{\ast}}) p^{\|J_x\|}n$ for each $x\in V(H')$, We are now almost finished. The only remaining problem is that we do not necessarily have $\|W'_{i,j}\|=\|V'_{i,j}\|$ for each $i\in[r]$ and $j\in[k]$. Since $\|V'_{i,j}\|=\|V_{i,j}\|\pm 2\Delta^{r+10}\|V_0\|=m_{i,j}\pm 3\Delta^{r+10}\|V_0\|$, by \[Hp:sizeWp\] we have $\|V'_{i,j}\|=\|W'_{i,j}\|\pm \xi n$. We can thus apply Lemma \[lem:balancing\], with input $k$, $r_1$, $\Delta$, $\gamma$, $d$, $8{\varepsilon}$, and $r$. This gives us sets $V''_{i,j}$ with $\|V''_{i,j}\|=\|W'_{i,j}\|$ for each $i\in[r]$ and $j\in[k]$ by \[lembalancing:sizesout\]. Let ${\mathcal V}''=\{V''_{i,j}\}_{i\in[r],j\in[k]}$. Lemma \[lem:balancing\] guarantees us the following. 1. \[Gpp:sizeV\] $\frac{n}{8kr}\leq \|V''_{i,j}\| \leq \frac{8n}{kr}$ for every $i\in[r]$ and $j\in[k]$, 2. \[Gpp:Greg\] ${\mathcal V}''$ is $(4{{\varepsilon}^{\ast}},d,p)_G$-lower-regular on $R^k_r$ and $(4{{\varepsilon}^{\ast}},d,p)_G$-super-regular on $K^k_r$, 3. \[Gpp:Ginh\] both $\big({N_{\Gamma}}(v, V''_{i,j}),V''_{i',j'}\big)$ and $\big({N_{\Gamma}}(v, V''_{i,j}),{N_{\Gamma}}(v, V''_{i',j'})\big)$ are $(4{{\varepsilon}^{\ast}}, d,p)_G$-lower-regular pairs for every $\{(i,j),(i',j')\} \in E(R^k_r)$ and $v\in V\setminus V_0$, and 4. \[Gpp:GsGa\] we have $(1-4{\varepsilon})p\|V''_{i,j}\| \leq \|{N_{\Gamma}}(v,V''_{i,j})\| \leq (1 + 4{\varepsilon})p\|V''_{i,j}\|$ for every $i \in [r]$, $j\in [k]$ and every $v \in V \setminus V_0$. 5. \[Gpp:sizeI\] $\big\|V''_{f^(x)}\cap\bigcap_{u\in J_x}N_G(u)\big\|\ge\tfrac14\alpha p^{\|J_x\|}\|V''_{f^(x)}\|$, 6. \[Gpp:sizeGa\] $\big\|V''_{f^(x)}\cap\bigcap_{u\in J_x}N_\Gamma(u)\big\|=(1\pm4{\varepsilon}^)p^{\|J_x\|}\|V'_{f^(x)}\|$, and 7. \[Gpp:Ireg\] $\big(V''_{f^(x)}\cap\bigcap_{u\in J_x}N_\Gamma(u),V''_{f^(y)}\cap\bigcap_{v\in J_y}N_\Gamma(v)\big)$ is $(4{\varepsilon}^,d,p)_G$-lower-regular. Here \[Gpp:sizeV\] comes from \[Gp:sizeV\] and \[lembalancing:symd\], while \[Gpp:Greg\] comes from \[lembalancing:regular\] and choice of ${\varepsilon}$. \[Gpp:Ginh\] is guaranteed by \[lembalancing:inheritance\]. Now, each of \[Gpp:GsGa\], \[Gpp:sizeI\] and \[Gpp:sizeGa\] comes from the corresponding \[Gp:GsGa\], \[Gp:sizeI\] and \[Gp:sizeGa\] together with \[lembalancing:gammaout\]. Finally, \[Gpp:Ireg\] comes from \[Gp:Ireg\] and \[Gp:GaI\] together with Proposition \[prop:subpairs3\] and \[lembalancing:gammaout\]. For each $x\in V(H')$ with $J_x=\emptyset$, let $I_x=V''_{f^(x)}$. For each $x\in V(H')$ with $J_x\neq\emptyset$, let $I_x=V''_{f^(x)}\cap\bigcap_{u\in J_x}N_G(u)$. Now ${\mathcal W}'$ and ${\mathcal V}''$ are $\kappa$-balanced by \[Gpp:sizeV\], size-compatible by construction, partitions of respectively $V(H')$ and $V(G)\setminus{\mathrm{im}}(\phi_{t^})$, with parts of size at least $n/(\kappa r_1)$ by \[Gpp:sizeV\]. Letting $\widetilde{W}_{i,j}:=W'_{i,j}\setminus X'$, by \[Hp:sizeX\], choice of $\xi$, and \[Hp:special\], $\{\widetilde{W}_{i,j}\}_{i\in[r],j\in[k]}$ is a $\big(\vartheta,K^k_r\big)$-buffer for $H'$. Furthermore since $f^$ is a graph homomorphism from $H'$ to $R^k_r$, we have \[itm:blowup:H\]. By \[Gpp:Greg\], \[Gpp:Ginh\] and \[Gpp:GsGa\] we have \[itm:blowup:G\], with $R=R^k_r$ and $R'=K^k_r$. Finally, the pair $({\mathcal I},{\mathcal J})=\big(\{I_x\}_{x\in V(H')},\{J_x\}_{x\in V(H')}\big)$ form a $\big(\rho,\tfrac14\alpha,\Delta,\Delta\big)$-restriction pair. To see this, observe that the total number of image restricted vertices in $H'$ is at most $\Delta^2\|V_0\|<\rho\|V_{i,j}\|$ for any $i\in[r]$ and $j\in[k]$, giving \[itm:restrict:numres\]. Since for each $x\in V(H')$ we have $\|J_x\|+\deg_{H'}(x)=\deg_H(x)\le\Delta$ we have \[itm:restrict:Jx\], while \[itm:restrict:sizeIx\] follows from \[Gpp:sizeI\], and \[itm:restrict:sizeGa\] follows from \[Gpp:sizeGa\]. Finally, \[itm:restrict:Ireg\] follows from \[Gpp:Ireg\], and \[itm:restrict:DJ\] follows since $\Delta(H)\le\Delta$. Together this gives \[itm:blowup:restrict\]. Thus, by Lemma \[thm:blowup\] there exists an embedding $\phi$ of $H'$ into $G\setminus{\mathrm{im}}(\phi_{t^})$, such that $\phi(x)\in I_x$ for each $x\in V(H')$. Finally, $\phi\cup\phi_{t^}$ is an embedding of $H$ in $G$, as desired. With Theorem \[thm:maink\] in hand, we can now present the proof of Theorem \[thm:main\]. Given $\gamma$, $\Delta$, and $k$, let $\beta>0$, $z>0$, and $C >0$ be returned by Theorem \[thm:maink\] with input $\gamma$, $\Delta$, and $k$. Set $\beta^\ast := \beta/2$ and ${C^{\ast}}:= C/\beta$. Let $H$ be a $k$-colourable graph on $n$ vertices with $\Delta(H) \leq \Delta$ such that there exists a set $W$ of at least ${C^{\ast}}p^{-2}$ vertices in $V(H)$ that are not contained in any triangles of $H$ and such that there exists a labelling ${\mathcal L}$ of its vertex set of bandwidth at most $\beta^\ast n$. By the choice of ${C^{\ast}}$ we find an interval $I \subseteq {\mathcal L}$ of length $\beta n$ containing a subset $F \subseteq W$ with $\|F\| = C p^{-2}$. Now we can rearrange the labelling ${\mathcal L}$ to a labelling ${\mathcal L}'$ of bandwidth at most $2 \beta^\ast n = \beta n$ such that $F$ is contained in the first $\beta n$ vertices in ${\mathcal L}'$. Then, by Theorem \[thm:maink\] we know that $\Gamma = G(n,p)$ satisfies the following a.a.s. if $p\geq C(\log n/n)^{1/\Delta}$ and in particular if $p\geq {C^{\ast}}(\log n/n)^{1/\Delta}$. If $G$ is a spanning subgraph of $\Gamma$ with $\delta(G) \geq \big((k-1)/k+\gamma\big)pn$, then $G$ contains a copy of $H$, which finishes the proof. Lowering the probability for degenerate graphs {#sec:proofdegen} ============================================== As with Theorem \[thm:main\], we deduce Theorem \[thm:degenerate\] from the following more general statement. \[thm:degen\] For each $\gamma>0$, $\Delta \geq 2$, $D\ge 1$ and $k\geq 1$, there exist constants $\beta >0$, $z>0$, and $C>0$ such that the following holds asymptotically almost surely for $\Gamma = G(n,p)$ if $p\geq C\big(\frac{\log n}{n}\big)^{1/(2D+1)}$. Let $G$ be a spanning subgraph of $\Gamma$ with $\delta(G) \geq\big(\frac{k-1}{k}+\gamma\big)pn$ and let $H$ be a graph on $n$ vertices with $\Delta(H) \leq \Delta$ and degeneracy at most $D$, that has a labelling ${\mathcal L}$ of its vertex set of bandwidth at most $\beta n$, a $(k+1)$-colouring that is $(z,\beta)$-zero-free with respect to ${\mathcal L}$ and where the first $\sqrt{\beta} n$ vertices in ${\mathcal L}$ are not given colour zero and the first $\beta n$ vertices in ${\mathcal L}$ include $C p^{-2}$ vertices that are not in any triangles or copies of $C_4$ in $H$. Then $G$ contains a copy of $H$. The proof of Theorem \[thm:degen\] is quite similar to that of Theorem \[thm:maink\]. We provide only a sketch, highlighting the differences. The most important of these are that we do not use Lemma \[lem:common\] in the pre-embedding, and that we use a version of Lemma \[thm:blowup\] whose performance is better for degenerate graphs. In order to state this, we need the following definitions. Given an order $\tau$ on $V(H)$ and a family ${\mathcal J}$ of image restricting vertices, we define ${\pi^\tau}(x):=\|J_x\|+\big\|\{y\in N_H(x):\tau(y)<\tau(x)\}\big\|$. Now the condition on $\tau$ we need for our enhanced blow-up lemma is the following. \[def:Dpm\_bdd\_order\] Let $H$ be a graph given with buffer sets $\widetilde{\mathcal{W}}$ and a restriction pair ${\mathcal I}=\{I_i\}_{i\in[r]}$ and ${\mathcal J}=\{J_i\}_{i\in[r]}$. Let ${\widetilde{W}}=\bigcup\widetilde{\mathcal{W}}$. Let $\tau$ be an ordering of $V(H)$ and $W^e\subset V(H)$. Then $\tau$ is a $({\tilde{D}},p,m)$-bounded order* for $H$, $\widetilde{\mathcal{W}}$, ${\mathcal I}$ and ${\mathcal J}$ with exceptional set $W^e$ if the following conditions are satisfied for each $x\in V(H)$. 1. \[ord:Dx\] Define $${\tilde{D}}_x:=\begin{cases} {\tilde{D}}-2 & \text{if there is $yz\in E(H)$ with $y,z\in N_H(x)$ and $\tau(y),\tau(z)>\tau(x)$}\\ {\tilde{D}}-1 & \text{else if there is $y\in N_H(x)$ with $\tau(y)>\tau(x)$}\\ {\tilde{D}}& \text{otherwise}\,. \end{cases}$$ We have ${\pi^\tau}(x)\le {\tilde{D}}_x$, and if $x\in N({\widetilde{W}})$ even ${\pi^\tau}(x)\le {\tilde{D}}_x-1$. Finally, if $x\in{\widetilde{W}}$ we have $\deg(x)\le {\tilde{D}}$. 2. \[ord:halfD\] One of the following holds: - $x\in W^e$, - ${\pi^\tau}(x)\le \frac12 {\tilde{D}}$, - $x$ is not image restricted and every neighbour $y$ of $x$ with $\tau(y)<\tau(x)$ satisfies $\tau(x)-\tau(y)\le p^{{\pi^\tau}(x)}m$. 3. \[ord:NtX\] If $x\in N({\widetilde{W}})$ then all but at most ${\tilde{D}}-1-\max_{z\not\in W^e}{\pi^\tau}(z)$ neighbours $y$ of $x$ with $\tau(y)<\tau(x)$ satisfy $\tau(x)-\tau(y)\le p^{{\tilde{D}}} m$. To obtain the best probability bound, one should choose $\tau$ to minimise ${\tilde{D}}$. In the proof of Theorem \[thm:degen\] we will take $\tau$ to be an order witnessing $D$-degeneracy, $W^e$ will contain all image restricted vertices, and we will choose buffer sets containing vertices of degree at most $2D+1$. One can easily check that this allows us to choose ${\tilde{D}}=2D+1$. \[thm:dblow\] For all $\Delta\ge 2$, $\Delta_{R'}$, $\Delta_J$, ${\tilde{D}}$, $\alpha,\zeta, d>0$, $\kappa>1$ there exist ${\varepsilon},\rho>0$ such that for all $r_1$ there is a $C$ such that for $$p\ge C\bigg(\frac{\log n}{n}\bigg)^{1/{\tilde{D}}}$$ the random graph $\Gamma=G_{n,p}$ a.a.s. satisfies the following. Let $R$ be a graph on $r\le r_1$ vertices and let $R'\subset R$ be a spanning subgraph with $\Delta(R')\leq \Delta_{R'}$. Let $H$ and $G\subset \Gamma$ be graphs with $\kappa$-balanced, size-compatible vertex partitions ${\mathcal W}=\{W_i\}_{i\in[r]}$ and ${\mathcal V}=\{V_i\}_{i\in[r]}$, respectively, which have parts of size at least $m\ge n/(\kappa r_1)$. Let $\widetilde{\mathcal{W}}=\{{\widetilde{W}}_i\}_{i\in[r]}$ be a family of subsets of $V(H)$, ${\mathcal I}=\{I_x\}_{x\in V(H)}$ be a family of image restrictions, and ${\mathcal J}=\{J_x\}_{x\in V(H)}$ be a family of restricting vertices. Let $\tau$ be an order of $V(H)$ and $W^e\subset V(H)$ be a set of size $\|W^e\|\le{\varepsilon}p^{{\max_{x\in W^e}{\pi^\tau}(x)}}n/r_1$. Suppose that 1. \[dbul:1\] $\Delta(H)\leq \Delta$, $(H,{\mathcal W})$ is an $R$-partition, and $\widetilde{\mathcal{W}}$ is an $(\alpha,R')$-buffer for $H$, 2. \[dbul:2\] $(G,{\mathcal V})$ is an $({\varepsilon},d,p)$-lower-regular $R$-partition, which is $({\varepsilon},d,p)$-super-regular on $R'$, has one-sided inheritance on $R'$, and two-sided inheritance on $R'$ for $\widetilde{\mathcal{W}}$, 3. \[dbul:3\] ${\mathcal I}$ and ${\mathcal J}$ form a $(\rho,\zeta,\Delta,\Delta_J)$-restriction pair. 4. \[dbul:4\] $\tau$ is a $({\tilde{D}},p,{\varepsilon}n/r_1)$-bounded order for $H$, $\widetilde{\mathcal{W}}$, ${\mathcal I}$, ${\mathcal J}$ with exceptional set $W^e$. Then there is an embedding $\psi\colon V(H)\to V(G)$ such that $\psi(x)\in I_x$ for each $x\in H$. We set up constants quite similarly as in the proof of Theorem \[thm:maink\]. Specifically, given $\gamma>0$, $\Delta\ge 2$, $D$ and $k\ge 2$, let $d$ be returned by Lemma \[lem:G\], with input $\gamma$, $k$ and $r_0:=10\gamma^{-1}$. Let $\alpha=\tfrac{d}{2}$. Let ${\tilde{D}}=2D+1$. Now let ${{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{BL}$}}}>0$ and $\rho>0$ be returned by Lemma \[thm:dblow\] with input $\Delta$, $\Delta_{R'}=3k$, $\Delta_J=\Delta$, ${\tilde{D}}'$, $\vartheta=\tfrac{1}{100D}$, $\zeta=\tfrac14\alpha$, $d$ and $\kappa=64$. Let ${{\varepsilon}^{\ast}}=\tfrac18{{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{BL}$}}}$, and then Lemma \[lem:OSRIL\], for input ${{\varepsilon}^{\ast}}$ and $d$, returns ${\varepsilon}_1>0$. Let ${\varepsilon}_0>0$ be small enough both for Lemma \[lem:TSRIL\] with input ${{\varepsilon}^{\ast}}$ and $d$, and for Lemma \[lem:OSRIL\] with input ${\varepsilon}_1$ and $d$. We choose ${\varepsilon}=\min\big({\varepsilon}_0,d,\tfrac{1}{4D}{{\varepsilon}^{\ast}},\tfrac{1}{2k}\big)$. Putting ${\varepsilon}$ into Lemma \[lem:G\] returns $r_1$. Next, Lemma \[lem:balancing\], for input $k$, $r_1$, $\Delta$, $\gamma$, $d$ and $8{\varepsilon}$, returns $\xi>0$. We assume without loss of generality that $\xi\le 1/(10kr_1)$, and set $\beta=10^{-12}\xi^2/(\Delta k^4r_1^2)$. Let $\mu=\tfrac{{\varepsilon}^2}{100000kr}$. Finally, suppose ${C^{\ast}}$ is large enough for each of these lemmas, for Lemma \[thm:blowup\], for Proposition \[prop:chernoff\] with input ${\varepsilon}$, and for Lemma \[lem:hypgeo\] with input ${\varepsilon}\mu^2$ and $\Delta$. We set $C=10^{10}k^2r_1^2{\varepsilon}^{-2}\xi^{-1}\Delta^{2r_1+20}\mu^{-1}{C^{\ast}}$, and $z=10/\xi$. Given $p\ge C\big(\tfrac{\log n}{n}\big)^{1/(2D+1)}$, a.a.s. $G(n,p)$ satisfies the good events of Lemma \[thm:dblow\], Lemma \[lem:G\], Lemma \[lem:OSRIL\] and Lemma \[lem:TSRIL\], and Proposition \[prop:chernoff\], with the stated inputs. Suppose that $\Gamma=G(n,p)$ satisfies these good events. Let $G$ be a spanning subgraph of $\Gamma$ with $\delta(G)\ge\big(\tfrac{k-1}{k}+\gamma\big)pn$. Let $H$ be any graph on $n$ vertices with $\Delta(H)\le\Delta$, and let ${\mathcal L}$ be a labelling of $V(H)$ of bandwidth at most $\beta n$ whose first $\beta n$ vertices include $C p^{-2}$ vertices that are not contained in any triangles or four-cycles of $H$, and such that there exists a $(k+1)$-colouring that is $(z,\beta)$-zero-free with respect to ${\mathcal L}$, and the colour zero is not assigned to the first $\sqrt{\beta}n$ vertices. Furthermore, let $\tau$ be a $D$-degeneracy order of $V(H)$. Next, as in the proof of Theorem \[thm:maink\], we apply Lemma \[lem:G\] to $G$, obtaining a partition of $V(G)$ with the properties \[main:Gsize\]–\[main:Ggam\]. Note that if $D=1$, in place of \[main:Ginh\] we will ask only for the weaker condition 1. \[main:Ginhp\] $\big(N_\Gamma(v,V_{i,j}),V_{i',j'}\big)$ is an $({\varepsilon},d,p)_G$-lower-regular pair for every $\big\{(i,j),(i',j')\big\}\in E(R^k_r)$ and $v\in V\setminus V_0$, and thus for $D=1$ we have $\|V_0\|\le{C^{\ast}}p^{-1}$, while for $D\ge 2$ we have $\|V_0\|\le{C^{\ast}}p^{-2}$. Next, we apply Lemma \[lem:H2\] to obtain a partition of $V(H)$. We use the same inputs as in the proof of Theorem \[thm:maink\], with the exception that $D$ is now given in the statement of Theorem \[thm:degen\] rather than being set equal to $\Delta$. The result is a function $f:V(H)\to V(R^k_r)$ and a special set $X$ with the same properties \[H:size\]–\[H:v1\], and in addition 1. \[H:buf\] $\|\{x\in f^{-1}(i,j): \deg(x) \leq 2D\}\| \geq \tfrac{1}{24D} \|f^{-1}(i,j)\|$. We now continue following the proof of Theorem \[thm:maink\], using Lemma \[lem:hypgeo\] with input ${\varepsilon}\mu^2$ and $D+1$ (rather than ${\varepsilon}\mu^2$ and $\Delta$), to choose a set $S$ satisfying for each $1\le\ell\le D+1$ and vertices $u_1,\dots,u_{\ell}$ of $V(G)$. We use the same pre-embedding Algorithm \[alg:pre\], with the exception that we choose vertices at line \[line:choosenbs\] differently. As before, given $v_{t+1}\in V_0\setminus{\mathrm{im}}(\phi_t)$, we use Claim \[cl:chooseW\] to find a set $W\subset N_G(v_{t+1})$ of size at least $\tfrac{1}{8r}\mu p n$ and an index $i\in[r]$ such that for each $w\in W$ we have $\big\|N_G(w,V_{i,j})\big\|\ge dp\|V_{i,j}\|$ for each $j\in[k]$. However, rather than applying Lemma \[lem:common\], we let $w_1,\dots,w_\ell$ be distinct vertices of $W$ which satisfy \[main:GpI\]–\[main:GaI\]. We now justify that this is possible. We choose the $w_1,\dots,w_\ell$ successively. Since $x_{t+1}$ is not contained in any triangle or four-cycle of $H$, we have $\|J_x\|\le 1$ for each $x\in V(H)$, so that \[main:GpI\] is automatically satisfied. By Proposition \[prop:chernoff\], \[main:GpGI\] and \[main:GaI\] are satisfied for all but at most $2{C^{\ast}}kr_1 p^{-1}\log n$ vertices of $W$. It remains to show that we can obtain \[main:GpIreg\], which we do as follows. For $s\in[\ell]$, when we come to choose $w_s$, we insist that for any $\big\{(i,j),(i',j')\big\}\in E(R^k_r)$, the following hold. First, $\big(N_\Gamma(w_s,V_{i,j}),V_{i',j'}\big)$ is $({\varepsilon}_1,d,p)_G$-lower-regular. Second, $\big(N_\Gamma(w_s,V_{i,j}),N_\Gamma(w_s,V_{i',j'})\big)$ is $({{\varepsilon}^{\ast}},d,p)_G$-lower-regular. Third, for each $1\le t\le s-1$, $\big(N_\Gamma(w_s,V_{i,j}),N_\Gamma(w_t,V_{i',j'})\big)$ is $({{\varepsilon}^{\ast}},d,p)_G$-lower-regular. The conditions of respectively Lemma \[lem:OSRIL\], Lemma \[lem:TSRIL\], and Lemma \[lem:OSRIL\] are in each case satisfied (in the last case by choice of $w_t$) and thus in total at most $3{C^{\ast}}k^2r_1^2\max\{p^{-2},p^{-1}\log n\}$ vertices of $W$ are prohibited. Since $5{C^{\ast}}k^2r_1^2\max\{p^{-2},p^{-1}\log n\}<\tfrac{\|W\|}{2}<\ell$ by choice of $C$, at each step there is a valid choice of $w_s$. Since for each $x\in V(H')$ we have $\|J_x\|\le 1$, this construction guarantees \[main:GpIreg\]. We now return to following the proof of Theorem \[thm:maink\]. We obtain ${\mathcal V}'$ by removing the images of pre-embedded vertices, and ${\mathcal V}''$ by applying Lemma \[lem:balancing\]. Note that here \[lembalancing:gammaout\] may be trivial, that is, the error term ${C^{\ast}}\log n$ may dominate the main term when $s$ is large, but we only require it for $s=1$ to obtain \[Gpp:sizeV\]–\[Gpp:Ireg\]. Finally, we are ready to apply Lemma \[thm:dblow\] to complete the embedding. We define $({\mathcal I},{\mathcal J})$ as in the proof of Theorem \[thm:maink\]. We however let $\widetilde{W}_{i,j}$ consist of the vertices of $W'_{i,j}\setminus X$ whose degree is at most $2D$. By \[H:buf\] there are at least $\tfrac{1}{100D}\|W'_{i,j}\|$ of these, so that $\widetilde{\mathcal{W}}$ is a $(\vartheta,K^k_r)$-buffer, giving \[dbul:1\]. Now \[dbul:2\] follows from \[Gpp:Greg\] and \[Gpp:Ginh\]. Finally, $({\mathcal I},{\mathcal J})$ is a $(\rho,\tfrac14\alpha,\Delta,\Delta)$-restriction pair, giving \[dbul:3\], exactly as in the proof of Theorem \[thm:maink\]. However now we need to give an order $\tau'$ on $V(H')$ and a set $W^e\subset V(H')$. The former is simply the restriction of $\tau$ to $V(H')$, and the set $W^e$ consists of all vertices $x\in V(H)$ with $\|J_x\|>0$. We now verify the remaining conditions of Lemma \[thm:dblow\]. First, we claim $\|W^e\|\le \Delta^2\|V_0\|\le {\varepsilon}p^{\max_{x\not\in W^e}\pi^{\tau'}(x)}n/r_1$. Observe that $\pi^{\tau'}(x)\le{\pi^\tau}(x)+\|J_x\|\le D+1$. For $D=1$, we have $\|V_0\|\le{C^{\ast}}p^{-1}$, and by choice of $C$ the desired inequality follows. For $D\ge 2$, we have $\|V_0\|\le{C^{\ast}}p^{-2}$, and again by choice of $C$ we have the desired inequality. The last condition we must verify is \[dbul:4\], that $\tau'$ is a $({\tilde{D}},p,{\varepsilon}n/r_1)$-bounded order. For any vertex $x$ of $H'$, we have $\pi^{\tau'}(x)\le{\pi^\tau}(x)+1\le D+1$, and furthermore for all vertices not in $W^e$ we have $\pi^{\tau'}(x)={\pi^\tau}(x)\le D$. To verify \[ord:Dx\], first note that by construction the vertices of $\bigcup\widetilde{\mathcal{W}}$ have degree at most $2D\le{\tilde{D}}$. Further, observe that if $D=1$ then $H'$ contains no triangles, and ${\tilde{D}}=3=D+2$. Since vertices in $N\big(\bigcup\widetilde{\mathcal{W}}\big)$ are by construction not image restricted, so are not in $W^e$, this is as required for \[ord:Dx\]. If on the other hand $D\ge 2$ then ${\tilde{D}}\ge D+3$, and again the conditions of \[ord:Dx\] are met. Next, if $x\not\in W^e$ then $\pi^{\tau'}(x)\le D$, so that \[ord:halfD\] holds. Finally, observe that $\max_{z\not\in W^e}\pi^{\tau'}(z)\le D$, and vertices $x\in N\big(\bigcup\widetilde{\mathcal{W}}\big)$ by construction have $\pi^{\tau'}(x)={\pi^\tau}(x)\le D$, so that \[ord:NtX\] holds. We can thus apply Lemma \[thm:dblow\] to embed $H'$ into $G'$, completing the embedding of $H$ into $G$ as desired. The proof of Theorem \[thm:degenerate\] from Theorem \[thm:degen\] follows the deduction of Theorem \[thm:main\] from Theorem \[thm:maink\], and we omit it. The Bandwidth Theorem in bijumbled graphs {#sec:proofjumbled} ========================================= Again, Theorem \[thm:jumbled\] is a consequence of the following. \[thm:jumbledk\] For each $\gamma >0$, $\Delta \geq 2$, and $k \geq 1$, there exists a constant $c >0$ such that the following holds for any $p>0$. Given $\nu\le cp^{\max(4,(3\Delta+1)/2)}n$, suppose $\Gamma$ is a $\big(p,\nu\big)$-bijumbled graph, $G$ is a spanning subgraph of $\Gamma$ with $\delta(G) \geq\big(\tfrac{k-1}{k}+\gamma\big)pn$, and $H$ is a $k$-colourable graph on $n$ vertices with $\Delta(H) \leq \Delta$ and bandwidth at most $c n$. Suppose further that $H$ has a labelling ${\mathcal L}$ of its vertex set of bandwidth at most $\beta n$, a $(k+1)$-colouring that is $(z,\beta)$-zero-free with respect to ${\mathcal L}$, and where the first $\sqrt{\beta} n$ vertices in ${\mathcal L}$ are not given colour zero, and the first $\beta n$ vertices in ${\mathcal L}$ include $c^{-1}p^{-6} \nu^2n^{-1}$ vertices in $V(H)$ that are not contained in any triangles of $H$. Then $G$ contains a copy of $H$. The proof of Theorem \[thm:jumbledk\] is a straightforward modification of that of Theorem \[thm:maink\]. Rather than repeating the entire proof, we sketch the modifications which have to be made. Since we are working with bijumbled graphs, we need to work with regular pairs, rather than lower-regular pairs, at all times. In order to use this concept, and to work with bijumbled graphs, we need versions of Lemmas \[thm:blowup\], \[lem:OSRIL\], and \[lem:TSRIL\], and Proposition \[prop:chernoff\], which work with regular pairs and with $\Gamma$ a bijumbled graph rather than a random graph. We also need the following easy proposition, which lower bounds the possible $\nu$ for a $(p,\nu)$-jumbled graph with $p>0$. \[prop:bijn\] Suppose $\tfrac{16}{n}<p<1-\tfrac{16}{n}$. There does not exist any $(p,\nu)$-bijumbled $n$-vertex graph with $\nu\le\min\big(\sqrt{pn/32},\sqrt{(1-p)n/32}\big)$. Suppose that $\Gamma$ is a $(p,\nu)$-bijumbled graph on $n$ vertices with $p\le\tfrac12$. If $\Gamma$ contains $\tfrac12n$ vertices of degree at least $4pn$, then we have $e(\Gamma)\ge pn^2$, and letting $A,B$ be a maximum cut of $\Gamma$, by bijumbledness we have $$\tfrac12pn^2\le e(A,B)\le p\|A\|\|B\|+\nu\sqrt{\|A\|\|B\|}\le\tfrac14pn^2+\tfrac12\nu n\,,$$ and thus $\nu\ge pn/2\ge\sqrt{pn/32}$. If on the contrary $\Gamma$ contains at least $\tfrac12n$ vertices of degree less than $4pn$, then let $A$ be a set of $\tfrac{1}{8p}$ such vertices, and $B$ a set of $\tfrac{n}{4}$ vertices with no neighbours in $A$. By bijumbledness, we have $$0\ge p\|A\|\|B\|-\nu\sqrt{\|A\|\|B\|}=\tfrac{n}{32}-\nu\sqrt{n/(32p)}$$ and thus $\nu\ge \sqrt{pn/32}$. The same argument applied to $\overline{\Gamma}$ proves the $p\ge\tfrac12$ case. The following sparse blow-up lemma for jumbled graphs is proved in [@blowup]. \[thm:jblowup\] For all $\Delta\ge 2$, $\Delta_{R'}$, $\Delta_J$, $\alpha,\zeta, d>0$, $\kappa>1$ there exist ${\varepsilon},\rho>0$ such that for all $r_1$ there is a $c>0$ such that if $p>0$ and $$\beta\le cp^{\max(4,(3\Delta+1)/2)}n$$ any $(p,\beta)$-bijumbled graph $\Gamma$ on $n$ vertices satisfies the following. Let $R$ be a graph on $r\le r_1$ vertices and let $R'\subset R$ be a spanning subgraph with $\Delta(R')\leq \Delta_{R'}$. Let $H$ and $G\subset \Gamma$ be graphs given with $\kappa$-balanced, size-compatible vertex partitions ${\mathcal X}=\{X_i\}_{i\in[r]}$ and ${\mathcal V}=\{V_i\}_{i\in[r]}$, respectively, which have parts of size at least $m\ge n/(\kappa r_1)$. Let ${\widetilde{\mathcal{X}}}=\{{\widetilde{X}}_i\}_{i\in[r]}$ be a family of subsets of $V(H)$, ${\mathcal I}=\{I_x\}_{x\in V(H)}$ be a family of image restrictions, and ${\mathcal J}=\{J_x\}_{x\in V(H)}$ be a family of restricting vertices. Suppose that 1. $\Delta(H)\leq \Delta$, $(H,{\mathcal X})$ is an $R$-partition, and ${\widetilde{\mathcal{X}}}$ is an $(\alpha,R')$-buffer for $H$, 2. $(G,{\mathcal V})$ is an $({\varepsilon},d,p)$-regular $R$-partition, which is $({\varepsilon},d,p)$-super-regular on $R'$, and has one-sided inheritance on $R'$, and two-sided inheritance on $R'$ for ${\widetilde{\mathcal{X}}}$, 3. ${\mathcal I}$ and ${\mathcal J}$ form a $(\rho p^\Delta,\zeta,\Delta,\Delta_J)$-restriction pair. Then there is an embedding $\psi\colon V(H)\to V(G)$ such that $\psi(x)\in I_x$ for each $x\in H$. There are three differences between this result and Lemma \[thm:blowup\]. First, we assume a bijumbledness condition on $\Gamma$, rather than that $\Gamma$ is a typical random graph. Second, we require regular pairs in place of lower-regular pairs. Third, the number of vertices we may image restrict is much smaller. We will see that these last two restrictions do not affect our proof substantially. Next, in [@ABSS], the following regularity inheritance lemmas for bijumbled graphs are proved. \[lem:pOSRIL\] For each ${\varepsilon}',d>0$ there are ${\varepsilon},c>0$ such that for all $0<p<1$ the following holds. Let $G\subset \Gamma$ be graphs and $X,Y,Z$ be disjoint vertex sets in $V(\Gamma)$. Assume that - $(X,Z)$ is $(p,cp^{3/2}\sqrt{\|X\|\|Z\|})$-bijumbled in $\Gamma$, - $(X,Y)$ is $\big(p,cp^2(\log_2\tfrac{1}{p})^{-1/2}\sqrt{\|X\|\|Y\|}\big)$-bijumbled in $\Gamma$, and - $(X,Y)$ is $({\varepsilon},d,p)_G$-regular. Then, for all but at most at most ${\varepsilon}'\|Z\|$ vertices $z$ of $Z$, the pair $\big(N_\Gamma(z)\cap X,Y\big)$ is $({\varepsilon}',d,p)_G$-regular. \[lem:pTSRIL\] For each ${\varepsilon}',d>0$ there are ${\varepsilon},c>0$ such that for all $0<p<1$ the following holds. Let $G\subset \Gamma$ be graphs and $X,Y,Z$ be disjoint vertex sets in $V(\Gamma)$. Assume that - $(X,Z)$ is $(p,cp^{2}\sqrt{\|X\|\|Z\|})$-bijumbled in $\Gamma$, - $(Y,Z)$ is $(p,cp^3\sqrt{\|Y\|\|Z\|})$-bijumbled in $\Gamma$, - $(X,Y)$ is $(p,cp^{5/2}\big(\log_2\tfrac{1}{p}\big)^{-\frac12}\sqrt{\|X\|\|Y\|})$-bijumbled in $\Gamma$, and - $(X,Y)$ is $({\varepsilon},d,p)_G$-regular. Then, for all but at most ${\varepsilon}'\|Z\|$ vertices $z$ of $Z$, the pair $\big(N_\Gamma(z)\cap X,N_\Gamma(z)\cap Y\big)$ is $({\varepsilon}',d,p)_G$-regular. The following two lemmas, which more closely resemble Lemmas \[lem:OSRIL\] and \[lem:TSRIL\], are corollaries. \[lem:pseudOSRIL\] For each ${{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}}, {\alpha_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}}>0$ there exist ${\varepsilon}_0 >0$ and $C >0$ such that for any $0 < {\varepsilon}< {\varepsilon}_0$ and $0 < p <1$, if $\Gamma$ is any $(p,\nu)$-bijumbled graph the following holds. For any disjoint sets $X$ and $Y$ in $V(\Gamma)$ with $\|X\|\geq C p^{-3} \nu$ and $\|Y\| \geq C p^{-2} \nu$, and any subgraph $G$ of $\Gamma[X,Y]$ which is $({\varepsilon}, {\alpha_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}},p)_G$-regular, there are at most $C p^{-3}\nu^2\|X\|^{-1}$ vertices $z \in V(\Gamma)$ such that $(X \cap N_{\Gamma}(z),Y)$ is not $({{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}},{\alpha_{\scalebox{{0.5}}{$\mathrm{OSRIL}$}}},p)_G$-regular. \[lem:pseudTSRIL\] For each ${{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}},{\alpha_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}}>0$ there exist ${\varepsilon}_0>0$ and $C >0$ such that for any $0<{\varepsilon}<{\varepsilon}_0$ and $0<p<1$, if $\Gamma$ is any $(p,\nu)$-bijumbled graph the following holds. For any disjoint sets $X$ and $Y$ in $V(\Gamma)$ with $\|X\|,\|Y\|\ge Cp^{-3}\nu$, and any subgraph $G$ of $\Gamma[X,Y]$ which is $({\varepsilon},{\alpha_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}},p)_G$-regular, there are at most $Cp^{-6}\nu^2/\min\big(\|X\|,\|Y\|\big)$ vertices $z \in V(\Gamma)$ such that $\big(X\cap N_\Gamma(z),Y\cap N_\Gamma(z)\big)$ is not $({{\varepsilon}_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}},{\alpha_{\scalebox{{0.5}}{$\mathrm{TSRIL}$}}},p)_G$-regular. Note that the bijumbledness requirements of this lemma are such that if $Y$ and $Z$ are sets of size $\Theta(n)$, then $X$ must have size $\Omega\big(p^{-6}\nu^2 n^{-1}\big)$. This is where the requirement of Theorem \[thm:jumbledk\] for vertices of $H$ not in triangles comes from. Finally, we give a bijumbled graphs version of Proposition \[prop:chernoff\]. We defer its proof, which is standard, and similar to that of Proposition \[prop:chernoff\], to Appendix \[app:tools\]. \[prop:pseudchernoff\] For each ${\varepsilon}>0$ there exists a constant $C >0$ such that for every $p>0$, any graph $\Gamma$ which is $(p,\nu)$-jumbled has the following property. For any disjoint $X,Y \subseteq V(\Gamma)$ with $\|X\|,\|Y\|\ge {\varepsilon}^{-1}p^{-1}\nu$, we have $e(X,Y)=(1\pm{\varepsilon})p\|X\|\|Y\|$, and $e(X)\le 2p\|X\|^2$. Furthermore, for every $Y\subset V(\Gamma)$ with $\|Y\|\ge Cp^{-1}\nu$, the number of vertices $v \in V(\Gamma)$ with $\big\|\|{N_{\Gamma}}(v,Y)\| - p \|Y\|\big\| > {\varepsilon}p \|Y\|$ is at most $Cp^{-2}\nu^2\|Y\|^{-1}$. Now, using these lemmas, we can prove bijumbled graph versions of Lemmas \[lem:G\] and \[lem:common\], and use these to complete the proof of Theorem \[thm:jumbledk\]. All these proofs are straightforward modifications of those in the previous sections. Briefly, the modifications we make are to replace ‘lower-regular’ with ‘regular’ in all proofs, to replace applications of lemmas for random graphs with the bijumbled graph versions above, and to recalculate some error bounds. The only one of our main lemmas which changes in an important way is the following Lemma for $G$. \[lem:pseudG\] For each $\gamma > 0$ and integers $k \geq 2$ and $r_0 \geq 1$ there exists $d > 0$ such that for every ${\varepsilon}\in \left(0, \frac{1}{2k}\right)$ there exist $r_1\geq 1$ and $c,{C^{\ast}}>0$ such that the following holds for any $n$-vertex $(p,\nu)$-bijumbled graph $\Gamma$ with $\nu\le c p^3n$ and $p>0$. Let $G=(V,E)$ be a spanning subgraph of $\Gamma$ with $\delta(G) \geq \left(\frac{k-1}{k} + \gamma\right)pn$. Then there exists an integer $r$ with $r_0\leq kr \leq r_1$, a subset $V_0 \subseteq V$ with $\|V_0\| \leq {C^{\ast}}p^{-6}\nu^2 n^{-1}$, a $k$-equitable vertex partition ${\mathcal V}= \{{V_{i,j}}\}_{i\in[r],j\in[k]}$ of $V(G)\setminus V_0$, and a graph $R^k_r$ on the vertex set $[r] \times [k]$ with $K^k_r \subseteq B^k_r \subseteq R^k_r$, with $\delta(R^k_r) \geq \left(\frac{k-1}{k} + \frac{\gamma}{2}\right)kr$, and such that the following is true. 1. \[plemG:size\] $\frac{n}{4kr}\leq \|{V_{i,j}}\| \leq \frac{4n}{kr}$ for every $i\in[r]$ and $j\in[k]$, 2. \[plemG:regular\] ${\mathcal V}$ is $({\varepsilon},d,p)_G$-regular on $R^k_r$ and $({\varepsilon},d,p)_G$-super-regular on $K^k_r$, 3. \[plemG:inheritance\] both $\big({N_{\Gamma}}(v, V_{i,j}),V_{i',j'}\big)$ and $\big({N_{\Gamma}}(v', V_{i,j}),{N_{\Gamma}}(v', V_{i',j'})\big)$ are $({\varepsilon},d,p)_G$-regular pairs for every $\{(i,j),(i',j')\} \in E(R^k_r)$, $v\in V\setminus (V_0 \cup V_{i,j})$, and $v'\in V\setminus (V_0 \cup V_{i,j} \cup V_{i',j'})$, 4. \[plemG:gamma\] we have $(1-{\varepsilon})p\|{V_{i,j}}\| \leq \|{N_{\Gamma}}(v,V_{i,j})\| \leq (1 + {\varepsilon})p\|{V_{i,j}}\|$ for every $i \in [r]$, $j\in [k]$ and every $v \in V \setminus V_0$. The change here, apart from replacing ‘lower-regular’ with ‘regular’, and working in bijumbled graphs, is that $V_0$ may now be a much larger set. Nevertheless, the proof is basically the same. We begin the proof as in that of Lemma \[lem:G\], setting up the constants in the same way, with the exception that we replace Lemmas \[lem:OSRIL\] and \[lem:TSRIL\] with Lemmas \[lem:pseudOSRIL\] and \[lem:pseudTSRIL\], and Proposition \[prop:chernoff\] with Proposition \[prop:pseudchernoff\]. We require $C$ to be sufficiently large for Lemmas \[lem:pseudOSRIL\] and \[lem:pseudTSRIL\], and for Proposition \[prop:pseudchernoff\]. We define ${C^{\ast}}=100k^2r_1^3C/{{\varepsilon}^{\ast}}$ as in the proof of Lemma \[lem:G\], and set $$c=10^{-5}({{\varepsilon}^{\ast}})^3(kr_1)^{-3}({C^{\ast}})^{-1}\,.$$ We now assume $\Gamma$ is $(p,\nu)$-bijumbled rather than random, with $\nu\le cp^3n$. In particular, by choice of $c$ this implies that $$\label{eq:psG:s} 10k^2r_1^2 Cp^{-2}\nu^2n^{-1}\le{{\varepsilon}^{\ast}}pn\quad\text{and}\quad 10k^2r_1^3 C p^{-6}\nu^2n^{-1}\le{{\varepsilon}^{\ast}}n\,.$$ We obtain a regular partition, with a reduced graph containing $B^k_r$, exactly as in the proof of Lemma \[lem:G\], using Proposition \[prop:pseudchernoff\] in place of Proposition \[prop:chernoff\] to justify the use of Lemma \[lem:regularitylemma\]. The next place where we need to change things occurs in defining $Z_1$, where we replace ‘lower-regular’ with ‘regular’, and in estimating the size of $Z_1$. Using Lemmas \[lem:pseudOSRIL\] and \[lem:pseudTSRIL\], and Proposition \[prop:chernoff\] with Proposition \[prop:pseudchernoff\], we replace with $$\|Z_1\|\le kr_1^2Cp^{-6}\nu^2n^{-1}+kr_1^2Cp^{-3}\nu^2n^{-1}+2kr_1Cp^{-2}\nu^2n^{-1}\le 4kr_1^2Cp^{-6}\nu^2n^{-1} { { {\overset{\mbox{\tiny{\eqref{eq:psG:s}}}}{\le}} } }\tfrac{{{\varepsilon}^{\ast}}}{kr_1}n\,.$$ Note that the final conclusion is as in . We can now continue following the proof of Lemma \[lem:G\] until we come to estimate the size of $Z_2$, where we use Proposition \[prop:pseudchernoff\] and replace with $$\|Z_2\|\le r_1+kr_1 Cp^{-2}\nu^2n^{-1}{ { {\overset{\mbox{\tiny{\eqref{eq:psG:s}}}}{\le}} } }\tfrac{{{\varepsilon}^{\ast}}}{kr_1}pn\,.$$ Again, the final conclusion is as in . The next change we have to make is in estimating the size of $V_0$, when we replace with $$\|V_0\|\le\|Z_1\|+\|Z_2\|\le 4kr_1^2 Cp^{-6}\nu^2n^{-1}+r_1+kr_1Cp^{-2}\nu^2 n^{-1}\le {C^{\ast}}p^{-6}\nu^2n^{-1}\,.$$ Finally, we need regular pairs in \[plemG:regular\] and \[plemG:inheritance\]. We obtained regular pairs from Lemma \[lem:regularitylemma\] and in the definition of $Z_1$, so that we only need Proposition \[prop:subpairs3\] to return regular pairs. We always apply Proposition \[prop:subpairs3\] to pairs of sets of size at least $\tfrac{{{\varepsilon}^{\ast}}pn}{r_1}$, altering them by a factor ${{\varepsilon}^{\ast}}$. Now Proposition \[prop:pseudchernoff\] shows that if $X$ and $Y$ are disjoint subsets of $\Gamma$ with $\|X\|,\|Y\|\le ({{\varepsilon}^{\ast}}p)^{-1}\nu$, then $e_\Gamma(X,Y)\le (1+{{\varepsilon}^{\ast}})p\|X\|\|Y\|$, as required. By choice of $c$, we have $({{\varepsilon}^{\ast}}p)^{-1}\nu\le ({{\varepsilon}^{\ast}})^2pn/r_1$, so that the condition of Proposition \[prop:subpairs3\] to return regular pairs is satisfied. The other one of our main lemmas which requires change, Lemma \[lem:common\], only requires changing ‘lower-regular’ to ‘regular’ and replacing the random graph with a bijumbled $\Gamma$. This does require some change in the proof, as we then use the bijumbled graph versions of various lemmas, whose error bounds are different. \[lem:pseudcommon\] For each $d>0$, $k \geq 1$, and $\Delta \geq 2$ there exists $\alpha >0$ such that for every ${\varepsilon}^\ast \in (0,1)$ there exists ${\varepsilon}_0 >0$ such that for every $r\geq 1$ and every $0<{\varepsilon}\le{\varepsilon}_0$ there exists $c>0$ such that the following is true. For any $n$-vertex $(p,cp^{\Delta+1}n)$-bijumbled graph $\Gamma$ the following holds. Let $G=(V,E)$ be a (not necessarily spanning) subgraph of $\Gamma$ and $\{V_i\}_{i\in[k]}\cup \{W\}$ a vertex partition of a subset of $V$ such that the following is true for every $i,i'\in [k]$. 1. \[pcnl:bal\] $\frac{n}{4kr}\le \|V_i\|\le \frac{4n}{kr}$, 2. \[pcnl:Vreg\] $(V_i,V_{i'})$ is $({\varepsilon}, d, p)_G$-regular, 3. \[pcnl:W\] $\|W\|=\frac{{\varepsilon}pn}{16kr^2}$, and 4. \[pcnl:Wdeg\] $\|N_G(w,V_i)\| \geq dp\|V_i\|$ for every $w \in W$. Then there exists a tuple $(w_1, \ldots, w_\Delta) \in \binom{W}{\Delta}$ such that for every $\Lambda,\Lambda^\ast\subseteq[\Delta]$, and every $i \neq i' \in [k]$ we have 1. \[pcnl:Gsize\] $\|\bigcap_{j\in \Lambda} N_G(w_j,V_i)\|\geq \alpha p^{\|\Lambda\|}\|V_i\|$, 2. \[pcnl:Gasizen\] $\|\bigcap_{j\in \Lambda} N_{\Gamma}(w_j)\| \le (1 + {\varepsilon}^\ast)p^{\|\Lambda\|}n$, 3. \[pcnl:Gasize\] $ (1-{\varepsilon}^\ast)p^{\|\Lambda\|}\|V_i\| \leq \|\bigcap_{j\in \Lambda} N_{\Gamma}(w_j,V_i)\| \leq (1 + {\varepsilon}^\ast)p^{\|\Lambda\|}\|V_i\|$, and 4. \[pcnl:Nreg\] $\big(\bigcap_{j\in \Lambda}{N_{\Gamma}}(w_j,V_i),\bigcap_{j^\ast\in \Lambda^\ast}{N_{\Gamma}}(w_{j^\ast},V_{i'})\big)$ is $({\varepsilon}^\ast, d,p)_G$-regular if $\|\Lambda\|,\|\Lambda^\ast\| < \Delta$ and either $\Lambda\cap\Lambda^\ast=\varnothing$ or $\Delta\geq 3$ or both. The main modifications we make to the proof of Lemma \[lem:common\] are to replace Lemmas \[lem:OSRIL\] and \[lem:TSRIL\] with Lemmas \[lem:pseudOSRIL\] and \[lem:pseudTSRIL\], and Proposition \[prop:chernoff\] with Proposition \[prop:pseudchernoff\], and to replace all occurrences of ‘lower-regular’ with ‘regular’. We sketch the remaining modifications below. We begin the proof by setting constants as in the proof of Lemma \[lem:common\], but appealing to Lemmas \[lem:pseudOSRIL\] and \[lem:pseudTSRIL\], and Proposition \[prop:pseudchernoff\], rather than their random graph equivalents. We set $c=10^{-20}2^{-2\Delta}{\varepsilon}^5(Ct_1kr)^{-4}$. Suppose $\nu\le cp^{\Delta+2}n$, and that $\Gamma$ is an $n$-vertex $(p,\nu)$-bijumbled graph rather than a random graph. In order to apply Lemma \[lem:SRLb\] to $G$, we need to observe that its condition is satisfied by Proposition \[prop:pseudchernoff\] and because ${\varepsilon}^{-1}p^{-1}\nu<10^{-10}\tfrac{{\varepsilon}^4pn}{k^4r^4}$ by choice of $c$. The same inequality justifies further use of Proposition \[prop:pseudchernoff\] to find the desired $W'$. Estimating the size of $W'$, we replace with $$\label{eq:psizeW} \|W'\|\ge 10^{-11}\frac{{\varepsilon}^4pn}{t_1k^4r^4}\ge 10^5 Cp^{-2}\nu\,,$$ where the final inequality is by choice of $c$. We only need to change the statement of Claim \[claim:common\] by replacing ‘lower-regular’ with ‘regular’ in \[cnl:cl:Wreg\] and \[cnl:cl:Vreg\]. However we need to make rather more changes to its inductive proof. The base case remains trivial. In the induction step, we need to replace with $$\big\|\bigcap_{j\in\Lambda}N_\Gamma(w_j,V'_i)\big\|\ge (1-{\varepsilon}_0)^{\Delta-2}p^{\Delta-2}\frac{n}{8tr}\ge 10^5Cp^{-4}\nu\,,$$ where the final inequality is by choice of $c$. This, together with $\|W'\|\ge 10^5Cp^{-2}\nu$ from , justifies that we can apply Lemma \[lem:pseudOSRIL\]. We obtain that at most $2^\Delta k^2 Cp^{-3}\nu^2\tfrac{8krt_1}{n}$ vertices $w$ in $W$ violate \[cnl:cl:Wreg\]. The estimate on the number of vertices violating \[cnl:cl:NGVp\] does not change. For \[cnl:cl:NGaVp\], we need to observe that $\big\|\bigcup_{j\in\Lambda}N_\Gamma(w_j,V'_i)\big\|=(1\pm{\varepsilon}_0)^{\|\Lambda\|}p^{\|\Lambda\|}\|V'_i\|$, and in particular by choice of ${\varepsilon}_0$ and $c$ this quantity is at least $Cp^{-1}\nu$. Then Proposition \[prop:pseudchernoff\] then gives that at most $2^{\Delta+1}kCp^{-2}\nu^2\tfrac{8krt_1}{n}$ vertices destroy \[cnl:cl:NGaVp\], and the same calculation gives the same bound for the number of vertices violating \[cnl:cl:NGaV\] and \[cnl:cl:NGa\]. Finally, for \[cnl:cl:Vreg\], we need to use the inequality $(1-{\varepsilon}_0)^{\Delta-1}p^{\Delta-1}\tfrac{n}{4kr}\ge Cp^{-2}\nu$, which holds by choice of $c$, to justify that Lemmas \[lem:pseudOSRIL\] and \[lem:pseudTSRIL\] can be applied as the corresponding random graph versions are in Lemma \[lem:common\]. We obtain quite different bounds from these lemmas, however. If $\Delta=2$, then we only use Lemma \[lem:pseudOSRIL\], with an input regular pair having both sets of size at least $\tfrac{n}{4kr}$, so that the number of vertices violating \[cnl:cl:Vreg\] in this case is at most $2^{2\Delta}k^2Cp^{-3}\nu^2\tfrac{4kr}{n}$. If $\Delta\ge 3$, we use both Lemma \[lem:pseudOSRIL\] and \[lem:pseudTSRIL\]. The set playing the rôle of $X$ in Lemma \[lem:pseudOSRIL\] has size at least $(1-{\varepsilon}_0)^{\Delta-2}p^{\Delta-2}\tfrac{n}{4kr}$, while we apply Lemma \[lem:pseudTSRIL\] with both sets of the regular pair having at least this size. As a consequence, the number of vertices violating \[cnl:cl:Vreg\] is at most $2^{2\Delta+1}k^2Cp^{-6}\nu^2 (1-{\varepsilon}_0)^{2-\Delta}p^{2-\Delta}\tfrac{4kr}{n}$ for the case $\Delta\ge 3$. Putting this together, for the case $\Delta=2$ we replace with the following upper bound for the number of vertices $w\in W'$ which cannot be chosen as $w_{\ell+1}$. $$2^\Delta k^2Cp^{-3}\nu^2\tfrac{8krt_1}{n}+2^\Delta k{{\varepsilon}^{\ast\ast}}_\Delta\|W'\|+3\cdot 2^{\Delta+1} kCp^{-2}\nu^2\tfrac{8krt_1}{n}+2^{2\Delta}k^2Cp^{-3}\nu^2\tfrac{4kr}{n}$$ By choice of $c$ and ${{\varepsilon}^{\ast\ast}}_\Delta$, this quantity is at most $\tfrac12\|W'\|$, completing the induction step for $\Delta=2$. For $\Delta\ge 3$, we replace the upper bound with $$2^\Delta k^2Cp^{-3}\nu^2\tfrac{8krt_1}{n}+2^\Delta k{{\varepsilon}^{\ast\ast}}_\Delta\|W'\|+3\cdot 2^{\Delta+1} kCp^{-2}\nu^2\tfrac{8krt_1}{n}+2^{2\Delta+1}k^2Cp^{-6}\nu^2(1-{\varepsilon}_0)^{2-\Delta}p^{2-\Delta}\tfrac{4kr}{n}$$ which by choice of $c,{\varepsilon}_0$ and ${{\varepsilon}^{\ast\ast}}_\Delta$ is at most $\tfrac12\|W'\|$, completing the induction step for $\Delta\ge 3$. We conclude that the modified Claim \[claim:common\] continues to hold, and this implies the statement of Lemma \[lem:pseudcommon\] as in the proof of Lemma \[lem:common\]. The proof of Theorem \[thm:jumbledk\] is similar to that of Theorem \[thm:maink\]. Again, we sketch the modifications. We begin as in the proof of Theorem \[thm:maink\], setting up constants as there, but replacing Lemma \[lem:G\] with Lemma \[lem:pseudG\], Lemma \[lem:common\] with Lemma \[lem:pseudcommon\], Lemma \[thm:blowup\] with Lemma \[thm:jblowup\], and Proposition \[prop:chernoff\] with Proposition \[prop:pseudchernoff\]. In addition to the constants defined in the proof of Theorem \[thm:maink\] we require $0<c\le 10^{-50}{\varepsilon}^8\mu\rho\xi^2(\Delta k r_1C)^{-10}$ to be small enough for Lemmas \[lem:pseudG\] and \[lem:pseudcommon\]. Now, instead of assuming $\Gamma$ to be a typical random graph, suppose $\nu\le cp^{\max\{4,(3\Delta+1)/2\}}n$, and let $\Gamma$ be an $n$-vertex $(p,\nu)$-bijumbled graph. By Proposition \[prop:bijn\] we have $$\label{eq:jsizep} p\ge {C^{\ast}}\big(\tfrac{\log n}{n}\big)^{1/2}\,.$$ We continue following the proof of Theorem \[thm:maink\]. We now assume the first $\beta n$ vertices of ${\mathcal L}$ include $Cp^{-6}\nu^2n^{-1}$ vertices that are not contained in any triangles of $H$. We appeal to Lemma \[lem:pseudG\] rather than Lemma \[lem:G\] to obtain a partition of $V(G)$. This partition has $\|V_0\|\le {C^{\ast}}p^{-6}\nu^2n^{-1}$ (which is different to the upper bound in the proof of Theorem \[thm:maink\]), but still satisfies \[main:Gsize\] and \[main:Ggam\], and \[main:Greg\] and \[main:Ginh\] when ‘lower-regular’ is replaced by ‘regular’ in both statements. The application of Lemma \[lem:H2\] is identical. The application of Lemma \[lem:hypgeo\] is also identical, and the deduction of is still valid by . The pre-embedding is also identical, except that we replace each occurrence of ${C^{\ast}}\max\{p^{-2},p^{-1}\log n\}$ with ${C^{\ast}}p^{-6}\nu^2n^{-1}$, and that we replace the application of Proposition \[prop:chernoff\] justifying that at each visit to Line \[line:choosev\] we have at least $\tfrac14\mu p n$ choices with an application of Proposition \[prop:pseudchernoff\]. To verify the condition of the latter, and to see that this yields a contradiction we use the inequality $\|Z\|\ge\tfrac{1}{100(\Delta+1)}\mu p n\ge 2{C^{\ast}}p^{-2}\nu^2\tfrac{8r}{{\varepsilon}n}$, which holds by choice of $c$. Moving on, we justify Claim \[cl:chooseW\] by observing that $\tfrac{{\varepsilon}n}{4kr_1}\ge Cp^{-1}\nu$, which allows us to apply Proposition \[prop:pseudchernoff\] in place of Proposition \[prop:chernoff\], and that $2kr{C^{\ast}}p^{-2}\nu^2\tfrac{4kr_1}{{\varepsilon}n}\le\tfrac{\|Y\|}{2}$, both inequalities following by choice of $c$. Now Lemma \[lem:pseudcommon\], in place of Lemma \[lem:common\], finds $w_1,\dots,w_\ell$. Our construction of $f^$, and its properties, is identical, while Lemma \[lem:pseudcommon\] gives \[main:Gsize\]–\[main:GaI\], with ‘lower-regular’ replaced by ‘regular’ in \[main:Greg\], \[main:Ginh\] and \[main:GpIreg\]. The deduction of \[Gp:sizeV\]–\[Gp:GaI\] is identical, except that we use the ‘regular’ consequence of Proposition \[prop:subpairs3\]. To justify this, observe that each time we apply Proposition \[prop:subpairs3\], we apply it to a regular pair with sets of size at least $(1-{{\varepsilon}^{\ast}})p^{\Delta-1}\tfrac{n}{4kr}$ by \[main:Gsize\] and \[main:GpGI\], and we change the set sizes by a factor $(1\pm 2\mu)$, so that Proposition \[prop:pseudchernoff\] gives the required condition. To check this in turn, we need to observe that $2\mu(1-{{\varepsilon}^{\ast}})p^{\Delta-1}\tfrac{n}{4kr}\ge100\mu^{-1}p^{-1}\nu$, which follows by choice of $c$. We can thus replace ‘lower-regular’ with ‘regular’ in \[Gp:Greg\], \[Gp:Ginh\] and \[Gp:Ireg\]. Next, we still have $3\Delta^{r+10}\|V_0\|\le\tfrac{1}{10}\xi n$, so that $\|V'_{i,j}\|=\|W'_{i,j}\|\pm\xi n$ is still valid for each $i\in[r]$ and $j\in[k]$. This, together with , Proposition \[prop:pseudchernoff\], and the inequality $\tfrac{1}{50000kr_1}{\varepsilon}^2\xi pn\le 100{\varepsilon}^{-2}\xi^{-1}p^{-1}\nu$, justifies that we can apply Lemma \[lem:balancing\] to obtain \[Gpp:sizeV\]–\[Gpp:sizeGa\], with ‘lower-regular’ replaced by ‘regular’ in \[Gpp:Greg\] and \[Gpp:Ginh\]. Finally, to obtain \[Gpp:Ireg\] with ‘lower-regular’ replaced by ‘regular’, we use Proposition \[prop:subpairs3\], with the condition to output regular pairs guaranteed by the inequality $10^{-20}{\varepsilon}^4k^{-3}r_1^{-3}p^{\Delta-1}n\ge 10^{20}{\varepsilon}^{-4}k^3r_1^3Cp^{-1}\nu$, which follows by choice of $c$, and Proposition \[prop:pseudchernoff\]. Finally, we verify the conditions for Lemma \[thm:jblowup\]. The only point where we have to be careful is with the number of image restricted vertices. The total number of image restricted vertices in $H'$ is at most $\Delta^2\|V_0\|\le\Delta^2{C^{\ast}}p^{-6}\nu^2n^{-1}$, which by choice of $c$ and by \[Gpp:sizeV\] is smaller than $\rho p^\Delta\|V_{i,j}\|$ for any $i\in[r]$ and $j\in[k]$, justifying that $({\mathcal I},{\mathcal J})$ is indeed a $(\rho p^\Delta,\tfrac14\alpha,\Delta,\Delta)$-restriction pair. The remaining conditions of Lemma \[thm:jblowup\] are verified as in the proof of Theorem \[thm:maink\], and applying it we obtain an embedding $\phi$ of $H'$ into $G\setminus{\mathrm{im}}(\phi_{t^})$, so tha $\phi\cup\phi_{t^*}$ is the desired embedding of $H$ into $G$. Finally, the deduction of Theorem \[thm:jumbledk\] is essentially the same as that of Theorem \[thm:main\] from Theorem \[thm:maink\], and we omit it. Concluding remarks {#sec:remarks} ================== General spanning subgraphs -------------------------- Our main theorems place restrictions on the graphs $H$ with respect to whose containment random or pseudorandom graphs have local resilience. As was shown by Huang, Lee and Sudakov [@huang2012], such restrictions are necessary. Given ${\varepsilon}>0$, if $\Gamma$ is either a typical random graph $G(n,p)$ or a pseudorandom graph with density $p$, and $p$ is sufficiently small, then one can delete edges from $\Gamma$ in order to remove all triangles at a given vertex $v$, without deleting more than ${\varepsilon}p n$ edges at any vertex. Thus if $H$ is any graph all of whose vertices are in triangles, if $p=o(1)$ the local resilience of $\Gamma$ with respect to containment of $H$ is $o(1)$. This leads to the question: if we instead restrict $G$, requiring in addition to the conditions of Theorem \[thm:main\] that $G$ contains a positive proportion of the copies of $K_{\Delta+1}$ in $\Gamma$ at each vertex, is it true that $G$ will contain any $k$-colourable, bounded degree spanning subgraph $H$ with sublinear bandwidth without further restriction? We study this question in a forthcoming companion note to this paper, together with Schnitzer [@ABEST]. Optimality of Theorem \[thm:main\] ---------------------------------- Recall that Huang, Lee and Sudakov [@huang2012] proved that the restriction on $H$ that ${C^{\ast}}p^{-2}$ vertices should not be in triangles is necessary for all $p$. For $p$ constant, they proved a version of Theorem \[thm:main\], but the number of vertices in $H$ they require to have independent neighbourhood grows as a tower type function of $p^{-1}$, and they also require these vertices to be well-distributed in the bandwidth order, so that our result is strictly stronger than theirs. On the other hand, we do not believe that the lower bound on $p$ in Theorem \[thm:main\] is optimal. For $\Delta=2$, the statement is certainly false for $p\ll n^{-1/2}$, since then $G(n,p)$ has a.a.s. local resilience $o(1)$ with respect to containing even one triangle. It seems likely that the statement is true down to this point, a log factor improvement on our result. For $\Delta=3$, the statement as written is false for $p\ll n^{-1/3}$. Briefly, the reason for this is that in expectation a vertex is in $O\big(p^6n^3\big)$ copies of $K_4$ in $G(n,p)$, and (with some work) this implies that there is a.a.s. a subgraph of $G(n,p)$ with minimum degree very close to $pn$ and $p^{-5}n^{-1}$ vertices not in copies of $K_4$. For $p\ll n^{-1/3}$, $p^{-5}n^{-1}\gg p^{-2}$, so that we would also have to insist on many vertices of $H$ not being in copies of $K_4$ to accommodate this. Generalising this, we obtain the following conjecture. For each $\gamma >0$, $\Delta \geq 2$, and $k \geq 1$, there exist constants $\beta^\ast >0$ and ${C^{\ast}}>0$ such that the following holds asymptotically almost surely for $\Gamma = G(n,p)$ if $p \geq {C^{\ast}}n^{-2/(\Delta+2)}$. Let $G$ be a spanning subgraph of $\Gamma$ with $\delta(G) \geq\left(\frac{k-1}{k}+ \gamma\right)pn$ and let $H$ be a $k$-colourable graph on $n$ vertices with $\Delta(H) \leq \Delta$, bandwidth at most $\beta^\ast n$, there are at least ${C^{\ast}}p^{-2}$ vertices in $V(H)$ that are not contained in any triangles of $H$, and at least ${C^{\ast}}p^{-(\Delta+2)(\Delta-1)/2}n^{2-\Delta}$ vertices in $V(H)$ which are not in $K_{\Delta+1}$. Then $G$ contains a copy of $H$. This conjecture seems to be hopelessly out of reach with our current state of knowledge. We cannot even prove that $G(n,p)$ itself is universal for graphs on $\tfrac{n}{2}$ vertices with maximum degree $\Delta$. The best current result in this direction is due to Conlon, Ferber, Nenadov and Škorić [@CFNS], who show that for $\Delta\ge 3$, if $p\gg n^{-1/(\Delta-1)}\log^5 n$ then $G(n,p)$ is a.a.s. universal for graphs on $\big(1-o(1)\big)n$ vertices of maximum degree $\Delta$, finally breaking the $n^{-1/\Delta}$ barrier which is reached by several papers, but still far from the conjectured truth. It is possible that their methods could be used to prove a version of Theorem \[thm:main\] for almost-spanning $H$ in sparser random graphs, but this does not appear to be straightforward. Optimality of Theorem \[thm:degenerate\] ---------------------------------------- The ‘extra’ restriction we place in Theorem \[thm:degenerate\], of having many vertices of $H$ which are neither in triangles nor four-cycles, is an artifact of our proof. It would be possible to remove the stipulation regarding four-cycles—one can prove a version of Lemma \[lem:common\] capable of embedding vertices in a degeneracy order. However this comes at the cost of a worse lower bound on $p$. It seems likely that one would be able to obtain a result for $p\gg\big(\tfrac{\log n}{n}\big)^{1/(2D+2)}$, but we did not check the details. As with Theorem \[thm:main\], we expect that the bound $p\ge\big(\tfrac{\log n}{n}\big)^{1/(2D+1)}$ in Theorem \[thm:degenerate\] is far from the truth: again the exponent is most likely a factor of roughly $2$ too small. Again, however, proving such a statement in general seems hopeless. Nevertheless, in one interesting case we can substantially improve on Theorem \[thm:degenerate\]. Specifically, if $H$ is an $F$-factor for some fixed $F$, then we can follow the proof of Theorem \[thm:degenerate\], but set ${\tilde{D}}=D+3$. We can do this because we choose a degeneracy order on $H$ in which the copies of $F$ are segments. We obtain a version of Theorem \[thm:degenerate\] in which $H$ is required to be an $F$-factor, where $F$ is $D$-degenerate, but the lower bound on $p$ improves to $p\ge{C^{\ast}}\big(\tfrac{\log n}{n}\big)^{1/(D+3)}$. This is still not optimal, but at least the exponent is asymptotically optimal as $D$ grows, rather than being off by a factor of two in the limit. For some specific $F$ one can improve this bound further; moreover for $F$-factors one can slightly improve on Lemma \[thm:dblow\] (see the concluding remarks of [@blowup]). Optimality of Theorem \[thm:jumbled\] ------------------------------------- The requirement of ${C^{\ast}}p^{-6}\nu^2 n^{-1}$ vertices of $H$ not in triangles comes from Lemma \[lem:pTSRIL\]. This lemma is proved in [@ABSS], where it is conjectured that the bijumbledness requirement is not optimal. What exactly the optimal result should be is not clear. When $\|X\|=\|Y\|=\|Z\|=\tfrac{n}{3}$, a construction of Alon [@AlonConstr] shows that $\big(p,cp^2n\big)$-bijumbledness is necessary for some $c>0$, but in our application we are interested in $Y$ and $Z$ being of order $n$, and $X$ much smaller. We also do not believe that the bijumbledness requirement of Theorem \[thm:jumbled\] is optimal. This requirement comes from Lemma \[thm:jblowup\], and it is suggested there that the statement could still hold given only $\big(p,cp^{\Delta+C}\big)$-bijumbledness for some $C$. Such an improvement there would immediately improve the results here correspondingly. It is generally conjectured that substantial further improvement is not possible, in the strong form that it is likely that for some $C>0$ and all $\Delta$ there exists $c>0$ such that for all large $n$ an $n$-vertex $\big(p,cp^{\Delta-C}\big)$-bijumbled graph exists which does not contain $K_{\Delta+1}$ at all. Tools {#app:tools} ===== We collect in this appendix proofs of results which are more or less standard but which we could not find in the form we require in the literature. We begin by showing that small alterations to regular pairs give us regular pairs. Let $A \subseteq \hat X$ and $B \subseteq \hat Y$ such that $\|A\| \geq \hat {\varepsilon}\|\hat X\|$ and $\|B\| \geq \hat {\varepsilon}\|\hat Y\|$ be given. Define $A' := A \cap X$ and $B':= B \cap Y$ and note that $$\|A'\| \geq \|A\| - \mu \|X\| \geq \hat {\varepsilon}\|\hat X\| - \mu \|X\| \geq \hat{{\varepsilon}}(1-\mu) \|X\| - \mu \|X\| \geq \big(\hat {\varepsilon}- 2 \sqrt{\mu}\big) \|X\| \geq {\varepsilon}\|X\|$$ by the definition of $\hat{\varepsilon}$. Analogously, one can show that $\|B'\| \geq {\varepsilon}\|Y\|$. Since $(X,Y)$ is an $({\varepsilon}, d,p)$-regular pair, we know that $d_p(A',B') \geq d- {\varepsilon}$. Furthermore, we have $$\|A'\| \geq \|A\| - \mu \|X\| \geq \|A\| - \mu \frac{\|A\|}{\hat{\varepsilon}} \geq \big(1- \sqrt{\mu}\big)\|A\|$$ and by an analogous calculation we get $\|B'\| \geq \big(1- \sqrt{\nu}\big)\|B\|$. For the number of edges between $A$ and $B$ we get $$\begin{aligned} e(A,B) &\geq e(A',B') \geq (d- {\varepsilon}) p\|A'\| \|B'\| \geq (d-{\varepsilon})p \big(1-\sqrt{\mu}\big)\big(1-\sqrt{\nu}\big) \|A\| \|B\|\\ & \geq \big(d- {\varepsilon}- 2 \sqrt{\mu} - 2\sqrt{\nu}\big) p \|A\| \|B\| \geq (d-\hat{\varepsilon}) p \|A\| \|B\|.\end{aligned}$$ Therefore we have $$d_p(A,B) \geq d-\hat{\varepsilon},$$ which finishes the proof. Now suppose that $(X,Y)$ is $({\varepsilon},d,p)$-fully-regular. Let $d'$ be such that $d_p(A',B')=d'\pm{\varepsilon}$ for any $A'\subset X$ and $B'\subset Y$ with $\|A'\|\ge{\varepsilon}\|X\|$ and $\|B'\|\ge{\varepsilon}\|Y\|$. Let $A\subset\hat{X}$ and $B\subset \hat{Y}$ with $\|A\| \geq \hat {\varepsilon}\|\hat X\|$ and $\|B\| \geq \hat {\varepsilon}\|\hat Y\|$ be given. As above, we obtain $e(A,B)\ge (d'-\hat{\varepsilon}) p \|A\| \|B\|$. We also have $$\begin{aligned} e(A,B)&\le e(A',B')+e(A',B\setminus B')+e(A\setminus A', B)\\ &\le (d'+{\varepsilon})p\|A'\|\|B'\|+(1+\mu+\nu)p\|A'\|\nu\|B\|+(1+\mu+\nu)p\mu\|A\|\|B\|\\ &\le (d'+\hat{{\varepsilon}})\|A\|\|B\|\,,\end{aligned}$$ so that $(\hat{X},\hat{Y})$ is $({\varepsilon},d,p)$-fully-regular, as desired. Next, we prove the Sparse Regularity Lemma variant Lemma \[lem:SRLb\], whose proof follows [@Scott]. Given ${\varepsilon}>0$ and $s$, let $L=100s^2{\varepsilon}^{-1}$. Let $n_1=1$, and for each $j\ge 2$ let $n_j=10000{\varepsilon}^{-1}n_j2^{sn_j}$. Let $t_1=n_{1000{\varepsilon}^{-5}(L^2+16Ls^2)+1}$. We define the energy of a pair of disjoint sets $P,P'$ contained in respectively $V_i$ and $V_{i'}$ to be $${\mathcal{E}}(P,P'):=\frac{\|P\|\|P'\|\min\big(d_p(P,P')^2,2Ld_p(P.P')-L^2\big)}{\|V_i\|\|V_{i'}\|}\,.$$ Note that this quantity is convex in $d_p(P,P')$. Now given a partition ${\mathcal{P}}$ refining $\{V_i\}_{i\in[s]}$, we define the energy of ${\mathcal{P}}$ to be $${\mathcal{E}}({\mathcal{P}}):=\sum_{\{P,P'\}\subset{\mathcal{P}}}{\mathcal{E}}(P,P')\,.$$ We now construct a succession of partitions ${\mathcal{P}}_{j+1}$ for each $j\ge 1$, refining ${\mathcal{P}}_1:=\{V_i\}_{i\in[s]}$. We claim that for each $j$, the following hold. 1. \[srl:r1\] ${\mathcal{P}}_j$ partitions $V_i$ into between $n_j$ and $\big(1+\tfrac{1}{100}{\varepsilon}\big)n_j$ sets, of which the largest $n_j$ are equally sized. 2. \[srl:r2\] ${\mathcal{E}}({\mathcal{P}})\ge \tfrac{1}{1000}{\varepsilon}^5j$. We stop if ${\mathcal{P}}_j$ is $\big(\tfrac12{\varepsilon},p\big)$-regular. If not, we apply the following procedure. For each pair of ${\mathcal{P}}_j$ which is not $\big(\tfrac12{\varepsilon},0,p\big)$-regular, we take a witness of its irregularity, consisting of a subset of each side of the pair. We let ${\mathcal{P}}'_j$ be the union of the Venn diagrams of all witness sets in each part of ${\mathcal{P}}_j$. Since ${\mathcal{P}}_j$ is not $\big(\tfrac12{\varepsilon},p\big)$-regular, there are at least $\tfrac12{\varepsilon}s^2 n_j^2$ pairs which are not $\big(\tfrac12{\varepsilon},0,p\big)$-regular. By choice of $L$ and by \[srl:r1\], at least $\tfrac14{\varepsilon}s^2 n_j^2$ of these pairs have density not more than $\tfrac12L$. By the defect Cauchy-Schwarz inequality, just from refining these pairs we conclude that ${\mathcal{E}}({\mathcal{P}}'_j)\ge{\mathcal{E}}({\mathcal{P}}_j)+\tfrac{1}{1000}{\varepsilon}^5$. Note that, by convexity of ${\mathcal{E}}(P,P')$ in $d_p(P,P)$, refining the other pairs does not affect ${\mathcal{E}}(P'_j)$ negatively. We now let ${\mathcal{P}}_{j+1}$ be obtained by splitting each set of ${\mathcal{P}}'_j$ within each $V_i$ into sets of size $\tfrac{1000-{\varepsilon}}{1000n_{j+1}}\|V_i\|$ plus at most one smaller set. Again by Jensen’s inequality, we have ${\mathcal{E}}({\mathcal{P}}_{j+1})\ge{\mathcal{E}}({\mathcal{P}}'_j)$, giving \[srl:r2\]. Since ${\mathcal{P}}'_j$ partitions each $V_i$ into at most $n_j2^{sn_j}=\tfrac{1}{10000}{\varepsilon}n_{j+1}$, the total number of smaller sets is at most $\tfrac{1}{10000}{\varepsilon}n_{j+1}$. This gives \[srl:r1\]. Now observe that for any partition ${\mathcal{P}}$ refining ${\mathcal{P}}_1$, we have ${\mathcal{E}}({\mathcal{P}})\le L^2+16Ls^2$. It follows that this procedure must terminate with $j\le 1000{\varepsilon}^{-5}(L^2+16Ls^2)+1$. The final ${\mathcal{P}}_j$ is thus $\big(\tfrac12{\varepsilon},p\big)$-regular. For each $i\in[s]$, let $V_{i,0}$ consist of the union of all but the largest $n_j$ parts of ${\mathcal{P}}_j$. Let ${\mathcal{P}}$ be the partition of $\bigcup_{i\in[s]}V_i\setminus V_{i,0}$ given by ${\mathcal{P}}_j$. This is the desired equitable $({\varepsilon},p)$-regular refinement of $\{V_i\setminus V_{i,0}\}_{i\in[s]}$. Using Lemma \[lem:SRLb\] (purely in the interests of self-containment, as we could also use the results of [@kohayakawa1997]), we now prove Lemma \[lem:regularitylemma\]. Given ${\varepsilon}>0$ and $r_0$, without loss of generality we assume ${\varepsilon}\le\tfrac1{10}$. Let $t_1$ be returned by Lemma \[lem:SRLb\] for input $\tfrac{1}{1000}{\varepsilon}^2s^{-1}$ and $s=100r_0{\varepsilon}^{-1}$. Let $r_1=st_1$. Given $\alpha>0$, let $G$ be an $n$-vertex graph with minimum degree $\alpha p n$. Let $\{V_i\}_{i\in[s]}$ be an arbitrary partition of $V(G)$ into sets of as equal as possible size. By assumption, we have $e(V_i,V_{i'})\le 2p\|V_i\|\|V_{i'}\|$ for each $i\neq i'$. Furthermore, if $V_i$ is a part with $e(V_i)\ge 3p\|V_i\|^2$, then taking a maximum cut $A,A'$ of $V_i$ we have $e(A,A')\ge\tfrac32 p\|V_i\|^2$. Enlarging the smaller of $A$ and $A'$ if necessary, we have a pair of sets both of size at most $\|V_i\|$ between which there are at least $\tfrac32p\|V_i\|^2$ edges, again contradicting the assumption of Lemma \[lem:regularitylemma\]. Thus $G$ satisfies the conditions of Lemma \[lem:SRLb\] with input $\tfrac1{1000}{\varepsilon}^2s^{-1}$ and $s$. Applying that lemma, we obtain a collection $\{V_{i,0}\}_{i\in[s]}$ of sets, and an $({\varepsilon},p)$-regular partition ${\mathcal{P}}$ of $\bigcup_{i\in[s]}V_i\setminus V_{i,0}$ which partitions each $V_i\setminus V_0$ into $t\le t_1$ sets. Note that $s\le \|{\mathcal{P}}\|\le r_1$ by construction. Now let $V'_0$ be the union of the $V_{i,0}$ for $i\in[s]$, any sets $W\in{\mathcal{P}}$ that lie in more than $\tfrac14{\varepsilon}s t$ pairs which are not $(\tfrac{1}{1000}{\varepsilon},p)$-regular, and at most two vertices from each set $W\in{\mathcal{P}}$ in order that the partition of $V(G)\setminus V'_0$ induced by ${\mathcal{P}}$ is an equipartition. Because the total number of pairs which are not $(\tfrac1{1000}{\varepsilon},p)$-regular is at most $\tfrac{1}{1000}{\varepsilon}^2s^{-1}(r_0 t)^2$, the number of such sets in any given $V_i$ is at most $\tfrac{1}{100}{\varepsilon}t$, so $\|V'_{i,0}\|$ has size at most $\tfrac{1}{50}{\varepsilon}\|V_i\|$, and the number of parts of ${\mathcal{P}}$ in $V_i\setminus V'_{i,0}$ is larger than $\tfrac{t}{2}$. Thus the partition ${\mathcal{P}}'$ of $V(G)\setminus V'_0$ induced by ${\mathcal{P}}$ is an $({\varepsilon},p)$-regular equipartition of $V(G)\setminus V'_0$, and we have $\|V'_0\|\le{\varepsilon}n$. We claim that this partition ${\mathcal{P}}'$ has all the properties we require. It remains only to check that for each $d\in[0,1]$, the $d$-reduced graph of ${\mathcal{P}}'$ has minimum degree at least $(\alpha-d-{\varepsilon})t'$. Suppose that $P$ is a part of ${\mathcal{P}}'$. Now we have $e(P)\le 3p\|P\|^2$, since otherwise, as before, a maximum cut $A,A'$ of $P$ has at least $\tfrac32p\|P\|^2<\tfrac{1}{20}{\varepsilon}p\|P\|n$ edges, yielding a contradiction to the assumption on the maximum density of pairs of $G$. By construction, $P$ lies in at most $\tfrac12{\varepsilon}t'$ pairs which are not $({\varepsilon},p)$-regular, and these contain at most $(1+\tfrac{1}{10}{\varepsilon})p\|P\|\big(\tfrac12{\varepsilon}t'\|P\|\big)<\tfrac{3}{4}{\varepsilon}p \|P\|n$ edges of $G$. We conclude that at least $\alpha p \|P\|n-\tfrac{7}{8}{\varepsilon}p \|P\|n$ edges of $G$ leaving $P$ lie in $({\varepsilon},p)$-regular pairs of ${\mathcal{P}}'$. Of these, at most $dp\|P\|n$ can lie in pairs of density less than $p$, so that the remaining at least $\big(\alpha-d-\tfrac{7}{8}{\varepsilon}\big)p\|P\|n$ edges lie in $({\varepsilon},d,p)$-regular pairs. If so many edges were in less than $(\alpha-d-{\varepsilon})t'$ pairs leaving $P$, this would contradict our assumption on the maximum density of $G$, so that we conclude $P$ lies in at least $(\alpha-d-{\varepsilon})t'$ pairs which are $({\varepsilon},d,p)$-regular, as desired. Given ${\varepsilon}>0$, set $C'=100{\varepsilon}^{-2}$ and $C=200C'{\varepsilon}^{-1}$. Suppose $\Gamma$ is $(p,\nu)$-bijumbled. First, given disjoint $X,Y\subset V(\Gamma)$ with $\|X\|,\|Y\|\ge {\varepsilon}^{-1}p^{-1}\nu$, $(p,\nu)$-bijumbledness of $\Gamma$ we have $e(X,Y)=p\|X\|\|\|Y\|\pm\nu\sqrt{\|X\|\|Y\|}$, and we need only verify that $\nu\sqrt{\|X\|\|Y\|}\le{\varepsilon}p\|X\|\|Y\|$, which follows from the lower bound on $\|X\|,\|Y\|$. For the second property, let $(A,B)$ be a maximum cut of $X$. We have $e(A,B)\ge\tfrac12e(X)$, and $\|A\|\|B\|\le\tfrac14\|X\|^2$. By $(p,\nu)$-bijumbledness of $\Gamma$, we conclude $$e(X)\le 2e(A,B)\le 2p\|A\|\|B\|+2\nu\sqrt{\|A\|\|B\|}\le \tfrac12p\|X\|^2+\nu\|X\|$$ so that it is enough to verify $\nu\|X\|\le p\|X\|^2$, which duly follows from the lower bound on $\|X\|$. Now let $Y\subset V(\Gamma)$ have size at least $Cp^{-1}\nu$. We first show that there are at most $C'p^{-2}\nu^2\|Y\|^{-1}$ vertices in $\Gamma$ which have less than $(1-{\varepsilon})p\|Y\|$ neighbours in $Y$. If this were false, then we could choose a set $X$ of $C'p^{-2}\nu^2\|Y\|^{-1}$ vertices in $\Gamma$ which have less than $(1-{\varepsilon})p\|Y\|$ neighbours in $Y$. Since by choice of $C$ we have $(1-{\varepsilon})p\|Y\|\le \big(1-\tfrac{\varepsilon}2\big)p\|Y\setminus X\|$, we see that $e(X,Y\setminus X)<\big(1-\tfrac{{\varepsilon}}{2}\big)p\|X\|\|Y\setminus X\|$. Since $$\nu\sqrt{\|X\|\|Y\|}=\nu\sqrt{C'p^{-2}\nu^2}=\sqrt{C'}\nu^2p^{-1}<\tfrac{{\varepsilon}}{2}p\|X\|\|Y\setminus X\|$$ this is a contradiction to $(p,\nu)$-bijumbleness of $\Gamma$. Next we show that there are at most $2C'p^{-2}\nu^2\|Y\|^{-1}$ vertices of $\Gamma$ which have more than $(1+{\varepsilon})p\|Y\|$ neighbours in $Y$. Again, if this is not the case we can let $X$ be a set of $2C'p^{-2}\nu^2\|Y\|^{-1}$ vertices of $\Gamma$ with more than $(1+{\varepsilon})p\|Y\|$ neighbours in $Y$. If there are more than $\tfrac12\|X\|$ vertices of $X$ with more than $\tfrac12{\varepsilon}p\|Y\|$ neighbours in $X$, then we have $e(X)\ge\tfrac18{\varepsilon}p\|X\|\|Y\|$. Taking a maximum cut $A,B$ of $X$, we have $e(A,B)\ge\tfrac{1}{16}{\varepsilon}p\|X\|\|Y\|$, and by $(p,\nu)$-bijumbledness of $\Gamma$ we therefore have $$\tfrac{1}{16}{\varepsilon}p\|X\|\|Y\|\le p\|A\|\|B\|+\nu\sqrt{\|A\|\|B\|}\le\tfrac14p\|X\|^2+\tfrac12\nu\|X\|\,,$$ and since $\|X\|\le\tfrac{1}{100}{\varepsilon}\|Y\|$, we conclude $\|Y\|\le 100{\varepsilon}^{-1}p^{-1}\nu$, a contradiction to the choice of $C$. We conclude that there are $\tfrac12\|X\|$ vertices $X'$ of $X$ have at most $\tfrac12{\varepsilon}p\|Y\|$ neighbours in $X$, and hence at least $\big(1+\tfrac12{\varepsilon}\big)p\|Y\|$ neighbours in $Y\setminus X$. By $(p,\nu)$-bijumbledness of $\Gamma$ we have $$\tfrac12\|X\|\big(1+\tfrac12{\varepsilon}\big)p\|Y\|\le e(X',Y\setminus X)\le \tfrac12 p\|X\|\|Y\|+\nu\sqrt{\tfrac{1}{2}p\|X\|\|Y\|}\,,$$ from which we have ${\varepsilon}C'p^{-1}\nu^2\le 2\sqrt{C'}\nu^2p^{-1}$, a contradiction to the choice of $C'$.	{ "pile_set_name": "ArXiv" }
--- title: 'Gravitational stability and dynamical overheating of stellar disks of galaxies.' --- Gravitational stability and dynamical overheating of stellar disks of galaxies. [ Zasov A.V.$^{(1)}$, Khoperskov A.V.$^{(2)}$, Saburova A.S. $^{(1)}$]{} [ $^{(1)}$ — Sternberg Astronomical Institute Moscow State University, Moscow, Russia]{} [$^{(2)}$ — Volgograd State University, Volgograd, Russia]{} We use the marginal stability condition for galactic disks and the stellar velocity dispersion data published by different authors to place upper limits on the disk local surface density at two radial scalelengths $R=2h$. Extrapolating these estimates, we constrain the total mass of the disks and compare these estimates to those based on the photometry and color of stellar populations. The comparison reveals that the stellar disks of most of spiral galaxies in our sample cannot be substantially overheated and are therefore unlikely to have experienced a significant merging event in their history. The same conclusion applies to some, but not all of the S0 galaxies we consider. However, a substantial part of the early type galaxies do show the stellar velocity dispersion well in excess of the gravitational stability threshold suggesting a major merger event in the past. We find dynamically overheated disks among both seemingly isolated galaxies and those forming pairs. The ratio of the marginal stability disk mass estimate to the total galaxy mass within four radial scalelengths remains within a range of 0.4—0.8. We see no evidence for a noticeable running of this ratio with either the morphological type or color index. [1.Introduction]{} The majority of stars in spiral and S0 galaxies are in general concentrated in old stellar disks with characteristic half-width of several hundreds parsecs. A formation and dynamical evolution of disks are the matter of hot debates, so the analysis of disk kinematical characteristics plays a key role (see e.g. Zasov, Sil’chenko (2010) and references therein). By kinematical characteristics we mean a rotation velocity and velocity dispersion of old stars, making up the bulk of stellar disk mass, as functions of radial coordinate in a disk plane. Typical velocity dispersion of old stellar disks equals to several dozens km/s. In general, it either can reflect the velocity of turbulent motion of gas which gave the birth to the first stellar generation, or can be the result of the dynamical heating of disk during its evolution. Several possible mechanisms of equilibrium disk heating are known, the most effective are a scattering of stars on massive clouds or stochastic spirals, gravitational interaction with dark halo subunits, major and minor mergers (see Jenkins, Binney (1990); Binney, Tremaine (2008); Shapiro et al. (2003)). As far as the velocity dispersion of stars increases the efficiency of all mechanisms decreases. For initially dynamically cold disk, another mechanism capable of effectively heating a disk is the development of gravitational instability against random perturbations. It leads to the increase of stellar velocity dispersion both in the plane of a disk, and along the vertical axis (through the developing of bending instability). At the early stage of evolution during the disk mass growth, stellar velocity dispersion might remain fitted to the level of its marginal stability. If a total disk mass is constant and there exists a gas disk, forming stars with low velocity dispersion in its plane, a self-regulating process, supporting stellar disk velocity dispersion in the quasi-stationary station, is also possible (Betrin, Lin, 1987). In any case, even if gas and young stars are absent, the minimal velocity dispersion of old stars at given $R$ will be constrained by disk local marginal stability condition. Numerical simulations of models of initially weakly unstable 3D disks show the fast passage of a disk into the marginally stable state, after which the increasing of velocity dispersion practically ceases in the absence of the heating mechanisms listed above (see e.g. Khoperskov et al. (2003)). For a thin disk with parameters slowly varying with $R$ the local critical value of radial velocity dispersion of stars is defined by Toomre criterion: $c_r=c_T$, where $c_T = {3.36 G \sigma}/{\kappa}$, $\sigma$ is the disk surface density at given $R$, $\kappa$ is epicyclic frequency determined by angular velocity $\Omega$ and its radial derivative: $$\kappa =2\Omega\cdot \sqrt{(1+(R/2\Omega)\cdot(d\Omega/dR))}.$$ A finite disk thickness makes disk more stable, while non-radial perturbations have the opposite effect (they decrease the stability threshold). The stability condition taking into account both of these effects can not be expressed in analytic form. If to express the critical radial velocity dispersion as $(c_r)_{cr}=Q\cdot c_T$, then, as numerical simulations of 2D and 3D disks show, the parameter $Q$ for a wide range of $R$ lies in the interval 1.2 — 2.5 (see e.g. Bottema (1993), Khoperskov et al. (2003)). Note, that the analytically obtained local stability condition of 2D disk taking into account non-radial perturbations gives for a flat rotation curve $Q_c\approx 3$ (Polyachenko et al. (1997)). The assumption of marginally stable disks makes it possible to put constraints on its surface density: $$\label{Eq-cr-QToomre} \sigma (R) = \frac{\kappa (R) c_r (R)}{3.36 G Q(R)},$$ and hence on a total disk mass therefore. It enables to obtain relative masses of disk and spheroidal components (bulge and dark halo) (Bottema (1993), Bottema (1997), Tyurina et al. (2001), Zasov et al. (2006)) and to estimate local thickness of stellar disk (see f.e. Khoperskov et al., 2010), which allows in turn to find a volume density of gas layer in disk plane if it exists in a galaxy (Zasov, Abramova (2006), Kasparova, Zasov (2008), Zasov, Abramova (2008)). However the question of how far are the real disks from marginally stable state still remains open. There are two indirect evidences of closeness of stellar velocity dispersion to $(c_r)_{cr}$ for significant fraction of disk galaxies. First, disk mass-to-light ratios, which follow from the stability condition are in good agreement with the estimates based on stellar population evolution photometrical models (see e.g. Bottema (1997), Zasov et al. (2004)). Second, the assumption of disk marginal stability enables to explain a positive correlation of relative disks thicknesses and their relative masses within the optical borders (Zasov et al. (2002)). Note, however, that both numerical simulations (Sotnikova, Rodionov (2006), Khoperskov et al. (2010)) and the analysis of observations of edge-on galaxies (Mosenkov et al. (2010), Bizyaev, Mitronova (2009), Bizyaev (2010)) show that this correlation is rather sparse especially for galaxies with small bulges. Moreover, numerical dynamical models by far are not always compatible with the assumption of disk marginal stability (see, e.g., Zasov et al. (2008)). A difficulty of estimation of the ratio of observed velocity dispersion $c_{obs}$ to its critical value $(c_r)_{cr}$ lies first of all in the difficulty of measurement of $c_{obs}$ at sufficiently large galactocentric distances and in addition is due to a number of factors hard to be taken into account even in the axissymmetrical galaxy model. Among the latter factors are the variation of critical Toomre parameter with radius, the ratio of radial to vertical velocity dispersion and the account of asymmetric drift in the estimation of disk angular velocity especially for the early type galaxies. The main goal of this work is to find out how far the observed values of velocity dispersion at given $R$ are from the expected ones for marginally stable disks for different type galaxies. For this purpose, we analyze the velocity dispersion data, available in the literature, for a chosen galactocentric distance $R=2h$, where $h$ is the radial disk scalelenght. At this radius a disk contribution to the rotation curve is nearly maximal. At higher $R$, measurements of velocity dispersion are less reliable and also the influence of external factors dynamically heating a disk may be more important. In turn, at lower $R$ it is often hard to separate the contributions of bulge and disk stars to the measured values of $c_{obs}$, and to take into account the presence of a bar if it occurs in a galaxy. It is worth noting that, as numerical 3D N-body simulations show for different parameters of components (disk, bulge and halo), the spread of values of $Q(R)$ is minimal at $R\approx2h$. Following Khoperskov et al. (2003) here we assume $Q(2h)\approx 1.5$. [2. The sample and the method we use]{} To estimate the upper limits of disk masses corresponding to the marginal stability condition we have chosen objects for which sufficiently extensive radial distributions of stellar velocity dispersion and rotation curves were available in the literature (121 objects). Using these data we estimated the threshold value of surface density corresponding to the gravitational stability at $R=2h$ from the equation (2). Radial velocity dispersion of stars $c_r$, if it was not given in the cited literature, was determined from $c_{obs}(R)$ measured along the major axis of a galaxy: $$\label{Eq-c-obs} c_{obs}^2 = c_{\phi}^2 \sin^2 i + {c_z}^2 \cos^2 i \,,$$ which leads to: $$\label{Eq-cr-cobs} c_r=c_{obs}\cdot \left[(c_{\phi} / c_r)^2 \sin^2 i + (c_z/c_r)^2 \cos^2 i\right]^{-1/2} \,.$$ where $c_{\phi}$ and $c_z$ are azimuthal and vertical components of velocity dispersions correspondingly, and $i$ is the inclination of a disk. Radial and azimuthal velocity dispersions are interconnected by Lindblad formula $c_{\phi}/c_r=\kappa/ 2\Omega$ following from the epicyclic approximation. Direct measurements of $c_z/c_r$ ratio in galaxies show that in majority of cases it lies in the interval 0.4—0.7 (Shapiro et al. (2003)). In current work it was taken to be 1/2. In most cases rotation curve $V(R)$ has a plateau at the distance of two disk radial scalelenghts, so the $\kappa/\Omega $ ratio can be taken as $\sqrt 2$. For the increasing or decreasing rotation curves we applied the corresponding corrections. If only the stellar rotation curve $V_(R)$ was available for a galaxy, the approximate asymmetric drift correction was introduced following Neistein et al. (1999). Note, that for $c_r/V_c>0.5$ this correction is inevitably very crude. A list of galaxies with available measurements of stellar disk velocity dispersion and their observed and calculated properties is given in Table 1. For galaxies with line-of-sight velocity exceeding 800 km/s the adopted distances correspond to the Hubble constant $H_0=75$ km/s/Mpc, (except Virgo cluster galaxies for which we took $D=17 $ Mpc). For several nearby galaxies with lower line-of-sight velocity (NGC 3198, NGC 6503, IC750) we used the distance modulus given in Hyperleda database. The disk inclination $i$, color indices corrected for galactic extinction, inclination and redshift, and line-of-sight velocities were also taken from this database. For the cases when the inclinations taken from Hyperleda differed significantly from those given in the original sources referred in the Table, the preference was given to the later ones. After the estimation of surface density at $R=2h$, it is easy to find a total disk mass: $$\label{Eq-Mdisk} M_d = 2\pi h^2\, \sigma(2h)\, e^2 \,.$$ In this work a disk photometric radial scalelenght and the scalelenght of surface density are assumed to be equal. In some cases this can be quite rough approximation, but the introduced uncertainty is not so high in comparison with other error sources: a change of $h$ by $\pm$ 30% for a fixed adopted value $\sigma(2h)$ leads to the change of $M_d$ by +7%, -15%. To compare the obtained disk masses with the photometric estimates we calculated disk mass-to-light ratios in $B$ band $(M/L_B)_d$. For Sb and later type galaxies where the bulge contribution to the luminosity is low, we neglected the difference between the total and disk luminosity. For Sa—S0 galaxies we have taken into account the contribution of bulge. To do this we took a disk luminosity estimates or disk de-projected surface brightness at $R=2h$ from de-projected radial surface brightness profiles available in the literature. In the absence of such data a disk luminosity was calculated from the total luminosity using bulge-to-disk luminosity ratio $B/D$, average for given morphological type (taken from Graham (2001)): $L_{d}={L_{tot}}/{((B/D)+1)}$. Disk color index was obtained in this case from the total color index: $$(B-V)_{d}=2.5\cdot \lg\left(10^{(0.4\cdot(B-V)_{tot})}-B/T\cdot 10^{(0.4\cdot (B-V)_{bulge})}\right) \,.$$ Bulge color index $(B-V)_{bulge}$ of early type galaxy was taken to be 0.8, if the total color index of galaxy did not exceed 0.8. Otherwise (for very red galaxies), a color index of a disk was accepted to be equal to the total one. [ 3. Results]{} Surface density and disk mass estimates together with other data and the references to the data sources are listed in Table 1. It contains:\ ([1]{}) Name.\ ([2]{}) Morphological type.\ ([3]{}) Radial disk scalelength $h$ used in the paper.\ ([4]{}) Ratio of radial velocity dispersion to circular velocity at $R=2h$.\ ([5]{}) Disk surface density corresponding to the marginal stability condition at $R=2h$.\ ([6]{}) Mass-to-light ratio of a disk, marginally stable at $R=2h$ ($B$-band).\ ([7]{}) Ratio of mass of the disk, marginally stable at $R=2h$, to the total mass $M_t$ within $R=4h$, where $M_t = 4 V_c^2 h / G$ and $V_c$ is circular velocity. Cases of $M_{d}/M_t >1$ indicate a significant dynamical overheating of a disk.\ ([8]{}) Velocity dispersion and rotation velocity data sources.\ Diagrams “$(M/L_B)_{d}$—$(B-V)_0$” for stellar disks are shown in Fig. 1, where disks masses $M_d$ are obtained from the gravitational stability condition at $R=2h$.[^1] For the Galaxy the mean color index for $Sbc$ type was used (Buta et al., 1994). A straight line reproduces model relation obtained by Bell, de Jong (2001) for old stellar systems with different present-day star formation rate (SFR) using modified (bottom-light) Salpeter initial mass function (IMF). It is worth noting that the light extinction inside the galaxies that may play important role for disks with high content of dust decreases the luminosity of a galaxy and simultaneously increases its color index shifting the points practically along the theoretical correlation. \[h!\] ![Diagram “$(M/L_B)_d$—$(B-V)_0$”, plotted for (a) the entire sample of galaxies and (b) for the galaxies of types later than $S0/a$. Milky Way is marked by open circle, pair members are shown by asterisks. Straight line corresponds to model relation obtained for old stellar systems with different history of star formation by Bell, de Jong (2001). Here and in other figures the error bars correspond to the errors of velocity dispersion data used to estimate the disk masses.](pic1.jpg "fig:"){width="12cm"} ![Diagram “$(M/L_B)_d$—$(B-V)_0$”, plotted for (a) the entire sample of galaxies and (b) for the galaxies of types later than $S0/a$. Milky Way is marked by open circle, pair members are shown by asterisks. Straight line corresponds to model relation obtained for old stellar systems with different history of star formation by Bell, de Jong (2001). Here and in other figures the error bars correspond to the errors of velocity dispersion data used to estimate the disk masses.](pic2.jpg "fig:"){width="12cm"} In Fig. 1a the entire sample of galaxies with known color indices is shown; pair members are marked by asterisks. Here and in the other figures the error bars correspond to the errors of velocity dispersion used to calculate the disk local surface density. The uncertainty of parameter $Q_c$ is not included in the error bars (according to the numerical simulations it reaches at least 20%). The same diagram as in Fig. 1a, but after the exclusion of S0-a—S0 galaxies, is shown in Fig. 1b. Though the scatter of points in these diagrams is large ($\sim 0.3~dex$), it is compatible with the errors of individual mass estimates. Therefore one can conclude that there is a general agreement between $(M/L_B)_{d}$ estimates based on marginal stability condition and those based on the stellar population evolution modeling. It is remarkable that most of the galaxies, which significantly deviate from model relation have a red color $(B-V)_0>0.7$. At least half of these galaxies are above the straight line, that is most of them have disks with significant dynamical overheating. Note, that there are both paired and isolated systems among these galaxies. Pair members occur also among the late type galaxies in Fig. 1b (11 objects), but the presented statistical data are not enough to claim their systematical difference from the isolated systems. Anyway, pair, group and cluster membership is not always connected with the stellar velocity dispersion exceeding the level needed for marginal gravitational stability. It seems that only a strong gravitation perturbation can be a cause of disk dynamical overheating. This conclusion also follows, for example, from the higher mean thickness of stellar disks in interacting edge-on galaxies (Reshetnikov, Combes (1997)). The position of our Galaxy in the diagrams is shown by the open circle. For our Galaxy $(M/L_B)_d$ ratio was calculated using the local values of surface density and brightness measured for solar vicinity and accepting $h=3$ kpc. The surface brightness data in $B$-band was taken from Portinari et al. (2005).. Note, that the value of surface density in the Solar neighborhood obtained on the basis of marginal stability condition is in a good agreement with the direct estimates (Korchagin et al. (2003), Holmberg, Flynn (2004), Kuijken, Gilmore (1991)). This gives evidences of marginal stability of disk of Galaxy at least at the galactocentric distance of a few radial scalelenghts. This conclusion conforms with the idea of the absence of major merger events in the history of our Galaxy (see e.g. Wyse, 2009)). In Fig. 2a the ratio of radial velocity dispersion to circular velocity $c_r/V_c$ is correlated with morphological type of galaxies. Type $t$ is coded as de Vaucouleurs numerical type: spiral galaxies correspond to the interval $t$=1 ($Sa$) — 7 ($Sd$), higher values of $t$ represent irregular galaxies, and $t=-1...-3$ are related to S0 galaxies. The correlation between $c_r/V_c$ and type is rather weak, moreover, if to exclude S0 galaxies, it almost disappears. The same conclusion may be done for the correlation between $c_r$ and $V_c$ at the fixed radius $R=2h$ (Fig. 2b). It means that it is impossible to get a reliable estimation of disk velocity dispersion by the indirect way, from the type of a galaxy or its rotation velocity, as it was for example suggested by Bottema (1993). The conclusion that most of the galaxies with dynamically overheated disks are among the red galaxies (Fig. 1a) is confirmed by Fig. 3, where color index is compared with the relative disk mass $M_d/M_t$ within the optical radius $R=4h$. As above, here $M_d$ corresponds to disks, which are assumed to be marginally stable at least at $R=2h$, and total mass of a galaxy was calculated as $M_t = 4h\cdot V_c^2/G$. [^2] \[h!\] ![(a) Diagram “$c_r/V_c$ — morphological type”. (b) Diagram “$V_c$—$c_r$”. Galaxies with types later than Sa are marked by open circles; a straight line corresponds to the mean value of $c_r/V_c$.](crvctype.jpg "fig:"){width="12cm"} ![(a) Diagram “$c_r/V_c$ — morphological type”. (b) Diagram “$V_c$—$c_r$”. Galaxies with types later than Sa are marked by open circles; a straight line corresponds to the mean value of $c_r/V_c$.](vccr1.jpg "fig:"){width="12cm"} \[h!\] ![Diagram “$M_{d}/M_t$ — $(B-V)_0$”. A horizontal line marks $M_{d}/M_t=1$. Points above this line correspond to the galaxies with certainly overheated disks. Open circles represent the galaxies with types later than Sa. Typical errors of $M_{d}/M_t$ caused by the uncertainty of stellar velocity dispersion estimates are about 40%.](pic4.jpg "fig:"){width="12cm"} As far as $M_d$ we found is in general case the upper limit of mass of a disk, the values $ M_d/M_t \ge 1 $ obtained for some galaxies (see Fig. 3) have no physical sense and indicate unambiguously the significant overestimation of disk masses, or, in other words, the disk overheating. Almost all of these objects are lenticular galaxies with a high color indexes $(B-V)_0>0.7$. Lenticular galaxies differ from other galaxies of the sample by their distributions of both $M_{d}/M_t$ ratios (Fig. 3) and “marginal” surface densities at $R=2h$ (Fig. 4), being shifted to higher values of $M_{d}/M_t$ and densities. These differences are statistically significant: according to the Wilcoxon criterion, the level of significance $\alpha=0.99$. From the distribution of $M_{d}/M_t$ (Fig. 3) it follows that for most of galaxies disk to total mass (inside $R=4h$) ratios lay in the interval 0.4—0.8, independently of color and morphological type (if to exclude galaxies with apparently overheated disks, for which these ratios remain uncertain). \[h!\] ![Diagram “$\sigma(2h)$ — morphological type”. Pair members are marked by asterisks. Straight horizontal line separates galaxies with $\sigma(2h)=500 M_{sun}/$pc$^2$ level, which definitely evidences the overestimation of their mass.](pic3.jpg "fig:"){width="14cm"} If to consider the disky galaxies as systems with marginally stable disks, one can plot the relation $M_d$—$V_c$ (baryonic Tully-Fisher relation), connecting the most important parameter of stellar disk (a mass) with the parameter determined mostly by massive halo (rotation velocity). This diagram is shown in Fig. 5 (galaxies with $M_{d}/M_t>1$ are excluded). The relation obtained by McGaugh (2005) for a large sample of galaxies, disk masses of which were estimated from their luminosities and colors, is also shown by straight line for comparison. From this figure it follows that disk masses estimated from the marginal stability condition show the same relation as masses obtained from photometry and evolution models, which confirms the absence of systematical difference between these evaluations. [4. Conclusions]{} As it was shown above, if to assume the marginal stability of inner parts of disks of spiral galaxies, one can estimate local values of disk surface densities and as the next step — the absolute and relative values of their total masses or, strictly speaking, the upper limits of masses for stable equilibrium disks. These estimates in most cases appear to be in a good agreement with the values obtained from photometric data on the basis of stellar population evolution models (Fig. 1b, Fig. 5). It allows to admit that the disks of most spiral galaxies are close to the marginally stable state. Evidently, mass and density estimates presented in this paper for any given galaxy remain rather rough, mainly due to large errors of radial velocity dispersion $c_r$, and, to a lesser degree, of radial scalelength $h$ values. For some of the galaxies the resulting errors exceed a factor of 2 (see Fig. 1). Thus to obtain more reliable estimates for individual galaxies one have to construct dynamical models consisted of several components which reproduce the radial profiles of observed rotation velocity and velocity dispersion $c_{obs}$ of disk stars. Still, the estimation of disk masses for galaxies of different types presented in this paper allows to conclude that there is no significant systemic dynamical overheating of disks in most of galaxies, except lenticulars, where the velocity dispersion often exceeds the threshold needed for marginal stability. It follows that many spiral galaxies have not suffered a strong gravitation perturbations or mergings after the formation of bulk of their stars. This result creates a certain problem in the generally accepted theory of galaxies formation in a gravitational well of dark halo (see e.g. the discussion in Zasov, Silchenko, 2010). The conclusion of marginal stability of disks, however, may not concern their outermost regions, which require an individual analysis and often may be dynamically overheated. It is essential, that the excessively high velocity dispersion at $R=2h$ usually takes place in galaxies which, if to judge by color indexes, have a weak star formation (color index $(B-V)_0 > 0.7$). Most of them belong to $S0/a-S0$ types. For these galaxies almost total absence of young stars and gas parallel with high stellar velocity dispersion apparently resulted from at least one merger with massive satellite in their history. This event could both increase the velocity dispersion of stars and decrease the content of cold gas evidently as the result of intensive star formation burst triggered by merging. Gas partially may transform into stars, partially could be ejected from the disk due to the activity of a large number of massive stars. Nevertheless, as it follows from Figs 1a and 3, some “red” galaxies still appear to keep their disks in the state close to marginal stability despite a high color index. It supports the idea of the existence of different mechanisms responsible for the transformation of spiral galaxies into lenticulars.\ This work was supported by grants RFBR 07-02-00792, 09-02-97021. \[h!\] ![Baryonic Tully-Fisher relation between the disk mass and rotation velocity. Straight line corresponds to the relation presented by McGaugh (2005) for spiral galaxies where disks masses were found from their luminosity and color .](tf.jpg "fig:"){width="14cm"} [99]{} Zasov, A.V.; Sil’chenko, O.K., UFN 180, 4 (2010) Jenkins, A.; Binney, J., MNRAS 245, 305 (1990) Binney, J.; Tremaine, S., Galactic Dynamics, (Princeton Univ.press, Princeton, NJ, 2008) Betrin, G.; Lin, C.C., Proceedings of 10th European Regional Astronomy Meeting of the IAU, 4, 255, (Czechoslovak Academy of Sciences, 1987) Buta, R.; Mitra, S.; de Vaucouleurs, G.; Corwin, H.G., Astron.J., 107, 118 (1994) Khoperskov, A. V.; Zasov, A. V.; Tyurina, N. V., ARep. 47, 357 (2003) Bottema, R., Astron. Astrophys. 328, 517 (1997) Tyurina, N.V. ; Khoperskov, A. V.; Zasov, A. V., Astron. and Astrophys. Transactions 20, 155 (2001) Zasov, A.V.; Khoperskov, A.V., ARep. 79, 195 (2002) Zasov, A.V.; Khoperskov, A. V.; Tyurina, N. V. AstL. 30, 593 (2004) Zasov, A.V.; Khoperskov, A. V.; Tyurina, N. V., “Astrophysical Disks”. Ed. by A.M. Fridman, et al., (Springer, Dordrecht, The Netherlands, 2006), p.307 Zasov, A.V.; Abramova, O.V., ARep. 83, 976 (2006) Kasparova, A.V.; Zasov A.V., AstL. 34, 174 (2008) Zasov, A.V.; Abramova, O.V., ASP Conference Series, Ed. by Jose G. Funes, S.J.and Enrico Maria Corsini, San Francisco: Astronomical Society of the Pacific, 396, 169 (2008) Zasov, A. V.; Bizyaev, D. V.; Makarov, D. I.; Tyurina, N. V., AstL 28, 527 (2002) Sotnikova, N.Ya.; Rodionov, S.A., AstL 32, 649 (2006) Bizyaev, D.; Mitronova, S., Astrophys. J. 702, 1567 (2009) Reshetnikov, V.; Combes, F., Astron. Astrophys. 324, 80 (1997) Bell, E. F.; de Jong, R. S., Astrophys. J. 550, 212 (2001) Portinari, L.; Holmberg, J.; Flynn, C., astro-ph/0510515v1, (2005) Korchagin, V. I. et al., Astron. J. 126, 2896 (2003) Holmberg, J.; Flynn, C., MNRAS, 352, 440 (2004) Kuijken, K.; Gilmore, G., Astrophys. J 367, 9 (1991) Wyse, R. F. G., IAUS 254, 179 (2009) McGaugh, S. S., ApJ 632, 859 (2005) Shapiro, K.L.; Gerssen, J.; van der Marel, R.P., Astron.J. 126, 2707 (2003) Bershady, M.; Verheijen, M.; Andersen, D., ASP Conf. Proc. 275, 43 (2002) Bottema, R., Astron. Astrophys. 275, 16 (1993) Heraudeau, Ph.; Simien, F.; Maubon, G.; Prugniel, Ph., Astron. Astrophys. Suppl. Ser., 136, 509 (1999) Beltran, J.C.V. et al., Astron.Astrophys. 374, 394 (2001) Simien, F.; Prugniel, Ph., Astron. Astrophys. Suppl. Ser., 145, 263 (2000) Simien, F.; Prugniel, Ph., Astron. Astrophys., 384, 371 (2002) Neistein, E.; Maoz, D.; Rix, H.-W.; Tonry, J.L., Astron. J., 117, 2666 (1999) Baggett, W.E.; Baggett, S.M.; Anderson, K.S.J., Astron. J., 116, 1626 (1998) Grosbol, P. J., Astron. Astrophys. Suppl. Ser., 60, 261 (1985) Cherepashchuk, A.M.; Afanas’ev, V.L.; Zasov, A.V.; Katkov, I. Yu., ARep, 54, 578 (2010) Gerssen, J.; Kuijken, K.; Merrifield, M.R., MNRAS, 288, 618 (1997) Gerssen, J.; Kuijken, K.; Merrifield, M. R., MNRAS, 317, 545 (2000) Chilingarian, I. V.; Novikova, A. P.; Cayatte, V.; Combes, F.; Di Matteo, P.; Zasov, A. V., Astron. Astrophys., 504, 389 (2009) Herrmann, K. A.; Ciardullo, R., AIPC 804, 341 (2005) Corbelli, E., MNRAS, 342, 199 (2003) Regan, M. W.; Vogel, S. N., Astrophys. J., 434, 536 (1994) Noordermeer, E. et al., MNRAS, 384, 943 (2008) Westfall, K. B. et al., Astrophysics and Space Science Proceedings, Springer 157 (2007) Pizzella, A. et al., MNRAS, 387, 1099 (2008) Pizzella, A.; Corsini, E. M.; Vega Beltran, J. C.; Bertola, F., Astron. Astrophys., 424, 447 (2004) Mendez-Abreu, J.; Aguerri, J. A. L.; Corsini, E. M.; Simonneau, E., Astron. Astrophys., 487, 555 (2008) Simien, F.; de Vaucouleurs, G., Astrophys. J., 302, 564 (1986) Ganda K. et al., MNRAS, 367, 46 (2006) Fridman, A. M. et al., Astron. Astrophys., 430, 67 (2005) Verheijen, M. A. W.; Sancisi, R., Astron. Astrophys., 370, 765 (2001) Marquez, I.; Masegosa, J.; Moles, M.; Varela, J.; Bettoni, D.; Galletta G., Astron. Astrophys., 393, 389 (2002) Merrett H. R. et al., MNRAS, 369,120 (2006) Pohlen, M.; Trujillo, I., Astron. Astrophys., 454,759 (2006) Mathewson, D. S.; Ford, V. L.; Buchhorn, M., AJSS, 81,413 (1992) Chaves, T. A.; Irwin, J. A., Astrophys. J., 557,646 (2001) Daigle, O.; Carignan et al., MNRAS, 367,469 (2006) Noordermeer, E.; van der Hulst, J. M.; Sancisi, R.; Swaters, R. S.; van Albada, T. S., MNRAS, 376, 1513 (2007) Bremnes, T.; Binggeli, B.; Prugniel, P., Astron. Astrophys. Suppl., 141, 211 (2000) Erwin P., MNRAS, 364,283 (2005) Noordermeer, E., PhDT, Proefschrift, Rijksuniversiteit Groningen, 1 (2006) Zasov, A. V.; Moiseev, A. V.; Khoperskov, A. V.; Sidorova, E. A., AstL 52, 79 (2008) Marquez, I.; Moles, M., Astron. Astrophys. Suppl., 120, 1 (1996) Nanyao, Y.; Lu, Astrophys. J., 506, 673 (1998) McDonald, M.; Courteau, S.; Tully R. B., MNRAS, 393, 628 (2009) Kent, S. M., Astron. Astrophys. Suppl. Ser., 59, 115 (1985) D’Onofrio, M.; Capaccioli, M.; Merluzzi, P.; Zaggia, S.; Boulesteix J., Astron. Astrophys. Suppl., 134, 437 (1999) Simien, F. and Prugniel, Ph., Astron. Astrophys. Suppl. Ser., 126, 519 (1997) Bedregal, A. G.; Aragon-Salamanca, A.; Merrifield, M. R.; Milvang-Jensen B., MNRAS, 371, 1912 (2006) Simien, F. and Prugniel, Ph., Astron. Astrophys. Suppl. Ser., 122, 521 (1997) Fisher, D., Astron. J., 113, 950 (1997) Loyer, E.; Simien, F.; Michard, R.; Prugniel, Ph., Astron. Astrophys., 334, 805 (1998) Thakar, A. R.; Ryden, B. S.; Jore, K. P.; Broeils, A. H., Astrophys. J., 479, 702 (1997) Simien, F. and Prugniel, Ph., Astron. Astrophys. Suppl. Ser., 126, 15 (1997) Hinz, J. L.; Rix, H-W; Bernstein, G. M., Astron. J., 121, 683 (2001) Debattista, V. P.; Williams, T. B., Astron. J., 605, 714 (2004) Khoperskov, A.; Bizyaev, D.; Tiurina, N.; Butenko, M. Astron. Nachr. 331, 731 (2010) Fisher, D.; Illingworth, G.; Franx, M., Astron. J. 107, 160 (1994) Hunter, D.A. et al., Astrophys. J., 634, 281 (2005) Valdez-Gutierrez, M. et al., Astron. J., 124, 3157 (2002) Gavazzi, G.; Franzetti, P.; Boselli, A.; Pierini, D.; Scodeggio M., Astron. Astrophys., 361, 863 (2000) Chung, A.; Bureau, M., Astron. J., 127, 3192 (2004) Bacon, R. et al., ASP Conf. Proc., 282, 179 (2002) Pohlen, M.; Balcells, M.; Luetticke, R.; Dettmar, R.-J., Astron. Astrophys., 422, 465 (2004) Kregel, M.; van der Kruit, P. C., MNRAS, 358, 481 (2005) Kregel, M.; van der Kruit, P. C.; de Grijs R., MNRAS, 334, 646 (2002) Xilouris, E.M.; Byun, Y.I.; Kylafis, N.D.; Paleologou, E.V.; Papamastorakis, J., Astron. Astrophys., 344, 868 (1999) Wagner, S. J.; Bender, R.; Dettmar, R.-J., Astron. Astrophys., 215, 243 (1989) Polyachenko, V. L.; Polyachenko, E. V.; Strel’nikov, A. V., AstL., 23, 598 (1997) Silva, D. R. et al., Astron. J., 98, 131 (1989) Capaccioli, M.; Cappellaro, E.; Held, E. V.; Vietri, M., Astron. Astrophys., 274, 69 (1993) Mosenkov, A. V.; Sotnikova, N. Ya.; Reshetnikov, V. P., MNRAS, 401, 559 (2010) Pohlen, M.; Zaroubi, S.; Peletier, R. F.; Dettmar, R-J., MNRAS, 378, 594 (2007) Wainscoat, R. J.; Hyland, A. R.; Freeman, K. C., Astrophys. J., 348, 85 (1990) Peletier, R. F.; Balcells, M., New Astronomy, 1, 349 (1997) Barteldrees, A.; Dettmar, R.-J., Astron. Astrophys. Suppl., 103, 475 (1994) Khoperskov, A.V.; Tyurina, A.V., ARep 47, 443 (2003) Graham, A.W., Astron. J., 121, 820 (2001) Crosthwaite, L. P.; Turner, J. L.; Ho, P. T. P., Astron. J., 119, 1720 (2000) Fathi, K. et al., Astron. Astrophys., 466,905 (2007) Park, O-K. et al., ASP Conf. Proc. 230, 109 (2001) Zaritsky, D.; Hill, J. M.; Elston, R., Astron. J., 99, 1108 (1990) Herrmann, K. A.; Ciardullo, R., Astrophys. J., 705, 1686 (2009) van der Kruit, P. C.; Jimnez-Vicente, J.; Kregel, M.; Freeman, K. C., Astron. Astrophys. 379, 374 (2001) [\|c\|c\|c\|c\|c\|c\|c\|c\|c\|c\|]{} \ & [2]{}& [3]{} &[4]{}&[5]{}& [6]{}&[7]{}&[8]{}\ Name & Type& $h$ & $c_r/V_c$ & $\sigma(2h)$ &$(M/L_B)_d$&$M_d/M_t$& Ref\ & &arcsec.& & $M_{sun}/$pc$^2$& & &\ \ & [2]{}& [3]{} &[4]{}&[5]{}& [6]{}&[7]{}&[8]{}\ Name & Type& $h$ & $c_r/V_c$ & $\sigma(2h)$ &$(M/L_B)_d$&$M_d/M_t$& Ref\ & &arcsec.& & $M_{sun}/$pc$^2$& & &\ MW & SBbc & 3 \[kpc\] & 0.2 & 97.3 & 1.91 & 0.33 & [@73]\ NGC1023 & E-S0 & 61.5 & 0.38 & 225 & 4.87 & 0.61 & [@8]\ NGC1068 & Sb & 55.6 & 0.36 & 177 & 2.5 & 0.58 & [@1]\ NGC1566 & SABb & 35.7 & 0.2 & 69.7 & 0.61 & 0.32 & [@3], [@31]\ NGC2460 & Sab & 15 & 0.49 & 452 & 5.09 & 0.8 & [@1]\ NGC2549 & S0 & 20.2 & 0.5 & 354 & 9.62 & 0.81 & [@8]\ NGC2613 & Sb & 57.3 & 0.13 & 92.9 & 1.87 & 0.28 & [@3], [@32]\ NGC2768 & S0 & 44.8 & 0.6 & 130 & 6.27 & 0.97 & [@8]\ NGC2775 & Sab & 35 & 0.39 & 328 & 6 & 0.63 & [@1]\ NGC2962 & S0-a & 22.3 & 0.39 & 264 & 15.4 & 0.64 & [@6]\ NGC2985 & Sab & 30 & 0.41 & 307 & 4.69 & 0.66 & [@13]\ NGC3198 & Sc & 58 & 0.18 & 44.1 & 0.99 & 0.3 & [@3], [@33]\ NGC338 & Sab & 18.3 & 0.49 & 116 & 3.53 & 0.79 & [@11]\ NGC3489 & S0-a & 19.4 & 0.62 & 715 & 5.82 & 1.01 & [@8]\ NGC3630 & S0 & 8.3 & 0.52 & 1140 & 8.23 & 0.84 & [@6], [@7]\ NGC3949 & Sbc & 20 & 0.54 & 286 & 1.11 & 0.53 & [@19]\ NGC3982 & SABb & 10.2 & 0.4 & 666 & 2.14 & 0.65 & [@2]\ NGC4030 & Sbc & 26.2 & 0.32 & 257 & 2.73 & 0.51 & [@1]\ NGC4143 & S0 & 15.4 & 0.36 & 794 & 8.27 & 0.58 & [@7]\ NGC4203 & E-S0 & 17.2 & 0.66 & 358 & 3.76 & 1.06 & [@7]\ NGC4251 & S0 & 22.6 & 0.56 & 492 & 12.1 & 0.91 & [@8]\ NGC4419 & SBa & 19 & 0.43 & 254 & 2.46 & 0.7 & [@5]\ NGC4578 & S0 & 36.4 & 0.34 & 72.4 & 7.98 & 0.55 & [@7]\ NGC4649 & S0 & 29.6 & 0.68 & 1730 & 10.1 & 1.1 & [@8]\ NGC470 & Sb & 13.6 & 0.33 & 162 & 1.34 & 0.53 & [@5]\ NGC4753 & S0-a & 34.8 & 0.36 & 172 & 3.42 & 0.59 & [@8]\ NGC488 & Sb & 40 & 0.25 & 149 & 3.41 & 0.4 & [@12]\ NGC5247 & SABb & 40 & 0.4 & 30 & 0.41 & 0.65 & [@3]\ NGC5273 & S0 & 18.3 & 0.56 & 238 & 6.82 & 0.9 & [@7]\ NGC5440 & Sa & 15.7 & 0.65 & 414 & 16.9 & 1.05 & [@11]\ NGC5533 & Sab & 34.4 & 0.53 & 100 & 6.48 & 0.85 & [@11]\ NGC584 & S0 & 8.5 & 0.42 & 1190 & 2.1 & 0.68 & [@8]\ NGC5866 & S0-a & 21 & 0.54 & 1031 & 7.35 & 0.87 & [@8]\ NGC598\ & Sc & 319.2 & 0.44 & 120 & 1.74 & 0.81 & [@15], [@16]\ NGC6340 & S0-a & 28 & 0.48 & 293 & 12.4 & 0.98 & [@14]\ NGC6503 & Sc & 40 & 0.19 & 84 & 0.85 & 0.38 & [@3]\ NGC7469 & Sa & 8.3 & 0.54 & 171 & 0.83 & 0.88 & [@11]\ IC750 & Sab & 14.2 & 0.6 & 631 & 7.65 & 0.97 & [@4]\ NGC7743 & S0-a & 18.7 & 0.49 & 237 & 4.92 & 0.83 & [@11]\ NGC7782 & Sb & 18.3 & 0.23 & 99.2 & 2.29 & 0.38 & [@5]\ ES0 288-G25 & Sbc & 18.4 & 0.36 & 121 & 3.28 & 0.61 & [@62]\ ES0 435-G14 & Sc & 17.6 & 0.28 & 60.9 & 1.6 & 0.47 & [@62]\ ES0 435-G25 & Sc & 47.35 & 0.29 & 84.9 & 3.13 & 0.5 & [@62]\ ES0151-004 & S0 & 10.6 & 0.38 & 138 & 9.51 & 0.61 & [@59]\ ES0189-7 & SABb & 31.1 & 0.42 & 102 & 4.08 & 0.72 & [@20]\ ES0240-011 & Sc & 29.5 & 0.26 & 107 & 2.63 & 0.43 & [@59]\ ES0311-012 & S0-a & 21.4 & 0.53 & 427 & 3.23 & 0.86 & [@59]\ ES0358-g006 & S0 & 9.2 & 0.58 & 472 & 12.7 & 0.94 & [@47]\ ES0446-17 & SBb & 19.4 & 0.44 & 103 & 3.86 & 0.71 & [@20]\ ES0450-20 & SBbc & 17.3 & 0.37 & 281 & 4.36 & 0.6 & [@20]\ ES0514-10 & SABc & 27.1 & 0.61 & 95.5 & 3.49 & 0.99 & [@20]\ ES0597-036 & S0-a & 11 & 0.41 & 165 & 14 & 0.69 & [@59]\ IC1963 & S0 & 15.8 & 0.36 & 151 & 5.12 & 0.59 & [@47]\ IC4767 & S0-a & 16 & 0.44 & 141 & 11.6 & 0.7 & [@59]\ IC4937 & Sb & 21 & 0.48 & 84.5 & 8.86 & 0.81 & [@59]\ IC5096 & Sbc & 26.6 & 0.32 & 148 & 5.31 & 0.52 & [@59]\ IC5249 & SBcd & 40.1 & 0.37 & 18.6 & 1.6 & 0.59 & [@80]\ m31\* & Sb & 1548 & 0.33 & 121 & 4.1 & 0.53 & [@28]\ NGC0524 & S0-a & 19.2 & 0.61 & 1790 & 19.1 & 0.99 & [@6]\ NGC0980 & S0 & 12.75 & 0.32 & 170 & 8.07 & 0.52 & [@5]\ NGC1032 & S0-a & 26.6 & 0.61 & 322 & 14.3 & 1.03 & [@59]\ NGC1052 & E-S0 & 17.3 & 0.52 & 658 & 5.81 & 0.81 & [@8]\ NGC1167 & S0 & 24.63 & 0.47 & 258 & 15 & 0.76 & [@38], [@37]\ NGC1175 & S0-a & 12.9 & 0.49 & 269 & 8.04 & 0.79 & [@46]\ NGC128 & S0 & 24.5 & 0.23 & 141 & 7.12 & 0.38 & [@45]\ NGC1316 & S0 & 49.4 & 0.78 & 922 & 16.5 & 1.25 & [@47]\ NGC1375 & S0 & 14.8 & 0.54 & 366 & 9.84 & 0.88 & [@47]\ NGC1380 & S0 & 36.3 & 0.4 & 271 & 5.21 & 0.65 & [@47]\ NGC1380A & S0 & 19.8 & 0.45 & 104 & 6.87 & 0.74 & [@47]\ NGC1381 & S0 & 20.8 & 0.36 & 204 & 6.76 & 0.59 & [@47]\ NGC1461 & S0 & 22 & 0.44 & 403 & 18.6 & 0.81 & [@49]\ NGC1886 & Sbc & 24.3 & 0.48 & 190 & 6.1 & 0.78 & [@59]\ NGC2273 & SBa & 29.7 & 0.38 & 146 & 6.27 & 0.61 & [@38]. [@37]\ NGC2310 & S0 & 24 & 0.41 & 66.9 & 1.32 & 0.7 & [@59]\ NGC2787 & S0-a & 21.4 & 0.33 & 668 & 7.03 & 0.53 & [@11]\ NGC2964 & Sbc & 18 & 0.48 & 428 & 4.86 & 0.74 & [@24]. [@25]\ NGC3054 & Sb & 27.9 & 0.39 & 180 & 4.57 & 0.63 & [@21]\ NGC3115 & E-S0 & 25.5 & 0.5 & 2370 & 45.8 & 0.81 & [@67]\ NGC3203 & S0-a & 24.3 & 0.33 & 115 & 7.59 & 0.53 & [@59]\ NGC3245 & S0 & 15.87 & 0.48 & 881 & 9.54 & 0.78 & [@11]\ NGC3390 & Sb & 35.8 & 0.51 & 167 & 7.37 & 0.76 & [@59]\ NGC3516 & S0 & 14.21 & 0.19 & 97.5 & 5.8 & 0.31 & [@11]\ NGC3941 & S0 & 20 & 0.39 & 422 & 4.15 & 0.64 & [@49]\ NGC3957 & S0-a & 21.6 & 0.37 & 204 & 10.2 & 0.59 & [@59]\ NGC3998 & S0 & 16.3 & 0.39 & 1240 & 13.4 & 0.64 & [@49]\ NGC4036 & S0 & 20 & 0.35 & 498 & 6.43 & 0.57 & [@49]\ NGC4102 & SABb & 19 & 0.57 & 505 & 5.74 & 0.93 & [@24], [@26]\ NGC4111 & S0-a & 25.4 & 0.57 & 689 & 17 & 0.92 & [@50]\ NGC4138 & S0-a & 17.4 & 0.31 & 380 & 7.77 & 0.5 & [@51]\ NGC4150 & S0 & 12.4 & 0.57 & 400 & 4.94 & 0.93 & [@38]\ NGC4270 & S0 & 12.1 & 0.67 & 362 & 9.58 & 1.09 & [@52]\ NGC4350 & S0 & 15 & 0.48 & 742 & 7.52 & 0.77 & [@50]\ NGC4352 & S0 & 17.3 & 0.43 & 46.6 & 2.83 & 0.7 & [@7]\ NGC4449 & IB & 44.5 & 0.43 & 57.2 & 0.31 & 0.7 & [@56]. [@57]\ NGC4469 & S0-a & 32 & 0.4 & 526 & 31.85 & 0.65 & [@59]\ NGC4474 & S0 & 13.9 & 0.49 & 250 & 3.49 & 0.83 & [@7]\ NGC4594 & Sa & 62 & 0.32 & 382 & 26.9 & 0.52 & [@65]\ NGC4703 & Sb & 12.7 & 0.64 & 424 & 10.7 & 1.03 & [@59]\ NGC4710 & S0-a & 17.6 & 0.72 & 222 & 2.91 & 1.17 & [@59]\ NGC474 & S0 & 24.1 & 0.69 & 108 & 5.21 & 1.11 & [@6]\ NGC4934 & S0-a & 6.44 & 0.34 & 122 & 3.89 & 0.55 & [@53]\ NGC5529 & Sc & 29.96 & 0.35 & 190 & 3.61 & 0.6 & [@62]\ NGC5574 & E-S0 & 7.9 & 0.75 & 565 & 7.84 & 1.21 & [@7]\ NGC5678 & SABb & 21 & 0.5 & 226 & 3.04 & 0.8 & [@24], [@27]\ NGC5746 & SABb & 16.4 & 0.67 & 1140 & 2.38 & 1.3 & [@59]\ NGC5869 & S0 & 16.4 & 0.74 & 741 & 24.8 & 1.2 & [@7]\ NGC615 & Sb & 15.1 & 0.55 & 380 & 2.83 & 0.9 & [@21]\ NGC6722 & Sb & 22 & 0.39 & 161 & 4.58 & 0.66 & [@59]\ NGC6771 & S0-a & 18 & 0.54 & 328 & 18.4 & 0.8 & [@59]\ NGC6925 & Sbc & 28.2 & 0.42 & 153 & 1.82 & 0.68 & [@21]\ NGC7079 & S0 & 17.1 & 0.42 & 351 & 6.21 & 0.67 & [@54]\ NGC7123 & Sa & 17 & 0.53 & 461 & 17.3 & 0.96 & [@59]\ NGC7332 & S0 & 18.7 & 0.59 & 351 & 5.61 & 0.96 & [@55]\ NGC7457 & S0 & 12.7 & 0.62 & 449 & 5.86 & 1.06 & [@11]\ NGC7531 & SABb & 25 & 0.52 & 192 & 2.78 & 0.84 & [@21]\ NGC891 & Sb & 107 & 0.33 & 94.3 & 5.76 & 0.57 & [@62]\ PGC44931 & Sc & 22.65 & 0.36 & 84.9 & 3.34 & 0.58 & [@59]\ UGC03087 & S0 & 4.84 & 0.32 & 201 & 1.16 & 0.52 & [@11]\ UGC08823 & S0 & 4.45 & 0.44 & 274 & 3.08 & 0.75 & [@11]\ M83 & SBc & 105 & 0.5 & 240 & 1.88 & 0.91 & [@79]\ M94 & Sab & 57.2 & 0.56 & 447 & 2.66 & 0.89 & [@79]\ [^1]: Color indices are available not for all considered galaxies, so the number of points in the diagram is not equal to the number of galaxies listed in Table 1. [^2]: Note that this simple formula for a total mass, is, strictly speaking, correct for spherically symmetric distribution of density and may underestimate $M_{d}/M_t$ by 20-30% $M_{d}/M_t$ for a disk galaxy which does not possesses a massive bulge or halo.	{ "pile_set_name": "ArXiv" }
--- author: - 'B. Gaczkowski' - 'T. Preibisch' - 'T. Stanke' - 'M.G.H. Krause' - 'A. Burkert' - 'R. Diehl' - 'K. Fierlinger' - 'D. Kroell' - 'J. Ngoumou' - 'V. Roccatagliata' bibliography: - 'lupuspaper\_v11\_final\_16\_09\_2015\_literatur.bib' date: 'Received 13 May 2015; Accepted 21 September 2015' subtitle: 'APEX/LABOCA[^1], [Herschel]{}[^2] and [Planck]{} observations' title: 'Squeezed between shells? On the origin of the Lupus [I]{} molecular cloud' --- [We want to study how collisions of large-scale interstellar gas flows form and influence new dense clouds in the ISM.]{} [We performed LABOCA continuum sub-mm observations of Lupus [I]{} that provide for the first time a direct view of the densest, coldest cloud clumps and cores at high angular resolution. We complemented those by Herschel and Planck data from which we constructed column density and temperature maps. From the Herschel and LABOCA column density maps we calculated PDFs to characterize the density structure of the cloud.]{} [The northern part of Lupus [I]{} is found to have on average lower densities and higher temperatures as well as no active star formation. The center-south part harbors dozens of pre-stellar cores where density and temperature reach their maximum and minimum, respectively. Our analysis of the column density PDFs from the Herschel data show double peak profiles for all parts of the cloud which we attribute to an external compression. In those parts with active star formation, the PDF shows a power-law tail at high densities. The PDFs we calculated from our LABOCA data trace the denser parts of the cloud showing one peak and a power-law tail. With LABOCA we find 15 cores with masses between 0.07 and $1.71\,M_\odot$ and a total mass of $\approx8\,M_\odot$. The total gas and dust mass of the cloud is $\approx164\,M_\odot$ and hence $\sim5\%$ of the mass is in cores. From the Herschel and Planck data we find a total mass of $\approx174\,M_\odot$ and $\approx171\,M_\odot$, respectively.]{} [The position, orientation and elongated shape of Lupus [I]{}, the double peak PDFs and the population of pre-stellar and protostellar cores could be explained by the large-scale compression from the advancing USco H[I]{} shell and the UCL wind bubble.]{} Introduction ============ In the current picture of the dynamic interstellar medium (ISM), molecular cloud formation is attributed to collisions of large-scale flows in the ISM [see review by @mc-formation-ppvi-review-2014 and references within]. Such flows can be driven by stellar feedback processes (e.g. UV-radiation and winds) and supernovae. At the interface of the colliding flows, compression, cooling, and fragmentation of the diffuse atomic medium produces cold sheets and filaments that later may become molecular and self-gravitating and dominate the appearance of the ISM as observed today [see review by @filaments-sf-ppvi-review-andre-2014]. In this picture the fast formation (and dispersion) of molecular clouds and the often simultaneous onset of star formation within [see @rapid-formation-mcs-hartmann-2001; @mc-formation-stars-vazquez-semadeni-2007; @clumps-mhd-mc-formation-banerjee-2009; @filaments-in-simulations-mc-formation-gomez-vazquez-semadeni-2014] appears plausible. One example of such a large-scale flow is an expanding shell or super-shell around e.g. an OB-association or in general driven by multiple stellar feedback of a star cluster or association [see review by @supershells-mc-formation-review-dawson-2013]. Molecular clouds may then either form inside the wall of such a shell [@mcs-in-supershells-dawson-2011] or at the interface region when two such shells collide with each other. The latter has been recently investigated by [@gmc-formation-colliding-shells-dawson-2015] for a young giant molecular cloud (GMC) at the interface of two colliding super-shells. From the comparison of CO observations with high-resolution 3D hydrodynamical simulations they found that the GMC assembled into its current form by the action of the shells. The Scorpius-Centaurus OB-association [@sco-cen-blaauw-1964; @sco-cen-hipparcos-dezeeuw-1999; @usco-full-population-preibisch-2002; @sco-cen-sfhb-preibisch-mamajek-2008 Sco-Cen] is the closest site of recent massive star formation to us and it consists of three sub-groups with different ages and well known stellar populations down to $2\,M_\odot$ [@sco-cen-stars-hipparcos-debruijne-1999]. The oldest one is the the Upper Centaurus-Lupus (UCL) sub-group with an age of $\sim 17$ Myr harboring 66 B-stars. With an age of $\sim 15$ Myr the Lower Centaurus Crux (LCC) sub-group is somewhat younger and contains 42 B-stars. The youngest sub-group is Upper-Scorpius (USco) with an age of $\sim 5$ Myr and consisting of 49 B-stars. The feedback of the numerous massive stars in Sco-Cen probably cleared the inner region of the association from diffuse matter creating expanding loop-like H[I]{} shells around each of the sub-groups of the association [@sco-cen-HI-degeus-1992]. At the edge of the USco shell several dense molecular clouds with very young ($\leq 1-2$ Myr) stellar populations are found. Of those the most prominent ones are the Lupus [I]{} cloud (near the western edge of the shell) and the $\rho$ Oph cloud (near the eastern edge). The Lupus [I]{} molecular cloud complex [for an overview see @lupus-clouds-sfhb] is found at a distance of 150 pc and consists of a $\approx2.6\degr\times0.6\degr$ (corresponds to a physical size of $\approx6.8\times1.6$ pc) main filament extending in a north to south direction (Galactic coordinates) and a ring-like structure of $\sim0.6\degr$ in diameter west of the main filament [towards UCL; see @lupus-clouds-13co(2-1)-Tothill-2009]. Recently [@lupus-submm-polarimetry-matthews-2014] noted also two smaller secondary filaments of which one is about half a degree long and runs perpendicular to the main filament and seems to connect with it in the south. The other one is about a degree long and lies south-west of the main filament extending from the southern end of the main filament to the ring-like structure. In this work we concentrate our analysis only on the main filament commonly seen in all observations. We will refer to it as Lupus [I]{} or the Lupus [I]{} filament. The cloud is found between the USco and the UCL sub-groups at a location where the expanding USco shell appears to interact with a bubble currently powered by the winds of the remaining B-stars of UCL. With its close distance Lupus [I]{} represents a good candidate where we can study how such a collision process may form and influence new dense clouds in the ISM. Lupus [I]{} has been mapped as part of several large surveys like the [Herschel]{} Gould-belt survey [@herschel-gould-belt-andree-2010; @lupus-clouds-herschel-rygl-2013] and the [Spitzer]{} Legacy Program ’From molecular clouds to planet-forming disks’ [c2d; @lupus-clouds-spitzer-c2d-chapman-2007; @lupus-clouds-spitzer-c2d-merin-2008]. These near-infrared to far-infrared surveys revealed the population of young stellar objects (YSOs) within the cloud showing that it is dominated by pre-stellar and protostellar cores indicating an on-going star formation event. [@lupus-clouds-herschel-rygl-2013] found that the star formation rate (SFR) is increasing over the past $0.5-1.5$ Myr and [@lupus-clouds-spitzer-c2d-merin-2008] estimated a SFR of $\rm 4.3\,M_\odot\,Myr^{-1}$ for Lupus [I]{} from their [Spitzer]{} data. Extinction maps of Lupus [I]{} have been created by various authors using different methods. [@lupus-extinction-star-counts-cambresy-1999] created an extinction map based on optical star counts with a resolution of a few arc-minutes. Using 2MASS data [@lupus-clouds-spitzer-c2d-chapman-2007] created a visual extinction map with a $2\arcmin$ resolution. Also from 2MASS data a wide field extinction map of Lupus [I]{} has been presented by [@lupus-2mass-extinction-maps-lombardi-2008]. It allowed extinction measurements down to $A_K=0.05$ mag, but had a resolution of $3\arcmin$. [@lupus-clouds-spitzer-c2d-merin-2008] created extinction maps from their [Spitzer]{} data by estimating the visual extinction towards each source classified as a background star, based on their spectral energy distribution (SED) from 1.25 to $24\,\mu$m. Their maps had a resolution of $2-5\arcmin$. [@lupus-clouds-herschel-rygl-2013] created a column density map of dust emission in Lupus [I]{} from their [Herschel]{} data by making a modified black body fit for each pixel in the four bands from $160-500\,\mu$m. The resulting resolution was $36\arcsec$. Here we present a far-infrared and sub-mm analysis of the Lupus [I]{} cloud based on newly obtained APEX/LABOCA sub-mm continuum data at $870\,\mu$m and complementing those with [Herschel]{} and [Planck]{} archival data. In Sec. \[sec:observations\] we briefly describe the observations used and their data reduction process. Sec. \[sec:methods\] gives an overview of the methods used to create column density maps and probability distribution functions (PDFs) from the observations. We present and discuss our results in Sec. \[sec:results\]. In Sec. \[sec:surrounding\] we consider the surroundings of Lupus [I]{} and the influence of large-scale processes on the cloud. Finally, we summarize and conclude our paper in Sec. \[sec:summary\]. Observations and data reduction {#sec:observations} =============================== LABOCA data ----------- The sub-mm continuum observations of Lupus [I]{} were performed with the APEX 12-m telescope located in the Chilean Atacama desert [@apex]. We used the LArge APEX BOlometer CAmera [LABOCA, @laboca] which operates in the atmospheric window at $870\,\mu$m (345 GHz). The angular resolution is 19.2(HPBW), and the total field of view is 11.4. At the distance of Lupus [I]{} (150 pc) the angular resolution corresponds to a spatial scale of $\sim2800$AU ($\sim0.01$pc). This is sufficient to resolve the structure of molecular cores. The LABOCA observations of Lupus [I]{} were obtained in Max-Planck and ESO Periods 91 and 92, on 24th March, 31st August, 12th, 13th September 2013, and 31st March 2014 (PI: B. Gaczkowski). They were performed in the on-the-fly (OTF) mode, scanning perpendicular as well as along the filament’s major axis with a random position angle to the axis for each scan to reduce striping effects and improve the sampling. The total observing time was 11.3 hours and the weather conditions good to average. Data reduction included standard steps for sub-millimeter bolometer data, using the [BoA]{} software package [@boa]. First, data were converted from instrumental count values to a Jansky scale using a standard conversion factor, then a flatfield correction (derived from scans of bright, compact sources) was applied. Corrections for atmospheric opacity were derived from skydips taken every 1-2 hours, and finally residual correction factors were determined from observations of planets and secondary calibrator sources. Data at the turning points of the map were flagged, as well as spikes. The data from the individual bolometers were then corrected for slow amplitude drifts due to instrumental effects and atmosphere by subtracting low-order polynomials from the time-stream data. Short timescale sky brightness variations (sky-noise) was corrected by removing the correlated (over a large number of bolometers) signal in an iterative fashion. Then the time-stream data were converted into sky-brightness maps for each scan, and finally all scans combined into one map. The removal of the correlated sky variations is known to filter out also astronomical emission from extended sources, usually leading to negative artifacts surrounding bright emission structures. In order to recover some of the extended emission, an iterative source modeling procedure was applied, using the result from the previous iteration to construct an input model for the following iteration. To construct the model, the previous map was smoothed, all pixels below a pixel value of 0 set to 0, and the map smoothed again. In the following iteration, the model is subtracted from the time-stream data before de-spiking, baseline subtraction, and sky-noise removal, and added back to the data stream afterwards, before the new map is created. This procedure effectively injects artificial flux into the mapped area and is prone to create runaway high surface brightness areas especially in regions without significant astronomical signal and low signal-to-noise. For that reason, the model image was set entirely to 0 in areas of the map with poor coverage, below a certain threshold in rms (basically the map edges), and the procedure was stopped after 25 iterations. Typically, structures on scales of the order of $3-4\arcmin$ can be recovered, larger scales get more and more filtered out [see simulations done by @chamaeleon-laboca-belloche-2011]. The gridding was done with a cell size of $6.1\arcsec$ and the map was smoothed with a Gaussian kernel of size $9\arcsec$ (FWHM). The angular resolution of the final map is $21.2\arcsec$ (HPBW) and the rms noise level is 23 mJy/$21.2\arcsec$-beam (hereafter the notation per beam in context of LABOCA will mean per $21.2\arcsec$-beam). The resulting map is shown in Fig. \[img:laboca-map\]. In order to determine the total sub-mm flux of the entire LABOCA map, we integrated all pixel values above the $3\sigma$ noise level of 69 mJy/beam. This yielded a total flux value of 476 Jy. Our peak intensity in the map is $I^\mathrm{max}_\nu=1.37$ Jy/beam. ![Lupus [I]{} LABOCA $870\,\mu$m map with square-root intensity scaling in units of Jy/$21.2\arcsec$-beam. The positions of the 15 cores that were found by [Clumpfind]{} are marked with white crosses. The white polygon marks the common area within which the mass of the cloud was calculated from the LABOCA and [Herschel]{} column density maps (see Section \[sec:total-mass-lupus\]).[]{data-label="img:laboca-map"}](Lupus_I_Laboca_map_poly_cores_gimp.png){width="40.00000%"} Herschel archival data ---------------------- The Lupus [I]{} cloud complex was observed by the [Herschel]{} far-infrared observatory [@herschel] in January 2011 as part of the Gould-belt survey [@herschel-gould-belt-andree-2010; @lupus-clouds-herschel-rygl-2013]. The Photodetector Array Camera and Spectrometer [PACS, @pacs] and the Spectral and Photometric Imaging Receiver [SPIRE, @spire] were used to map an area of $\approx2\degr \times 2.3\degr$. We retrieved these data from the [Herschel]{} science archive and reduced them using HIPE v12.1 [@hipe] for the calibration and the deglitching of the Level 0 PACS and SPIRE data. The Level 1 data of both instruments were then used to produce the final maps with the Scanamorphos package v24 [@scanamorphos /parallel and /galactic options were switched on in both cases to preserve extended structures in complex, bright Galactic fields]. The pixel-sizes for the five maps at 70, 160, 250, 350 and 500$\mu$m were chosen as $3.2\arcsec$, $4.5\arcsec$, $6\arcsec$, $8\arcsec$, and $11.5\arcsec$, respectively. The absolute calibration of the [Herschel]{} data was done following the approach described in [@dust-temperature-herschel-planck-bernard-2010]. Using the [Planck]{} and the IRAS data of the same field as the [Herschel]{} data one calculates the expected fluxes to be observed by [Herschel]{} at each band and from that computes the zero-level offsets (see Table \[tbl:herschel-offsets\]). Those are then added to the [Herschel]{} maps to create the final mosaics. [r c]{} Band & Offset\ $[\,\mu{\rm m}]$ & MJy $\rm sr^{-1}$\ 70 & 5.68954\ 160 & 14.3011\ 250 & 8.11490\ 350 & 4.06405\ 500 & 1.68754\ \[tbl:herschel-offsets\] For the dust properties and temperatures considered in our analysis the color corrections for both PACS and SPIRE are of order 1-2% and hence negligible. Planck archival data -------------------- We retrieved the all-sky maps of the High-Frequency-Receiver (HFI) in the 353, 545, and 857 GHz bands (corresponding to the FIR and sub-mm wavelengths of 850, 550, and $350\,\mu$m) from the [Planck]{} legacy archive (release PR1 21.03.2013). From these data cubes in HEALPix[^3] format we extracted for each wavelength separately the area covering the Lupus [I]{} region and created gnomonic projected maps. The resolution in each of the three bands is $5\arcmin$ and the pixel size was chosen as $1.7\arcmin$. Data analysis {#sec:methods} ============= Column density and temperature maps of Lupus [I]{} -------------------------------------------------- From our three data sets of LABOCA, [Herschel]{}, and [Planck]{} we calculated column density maps to characterize the cloud in a multi-wavelength approach. Additionally, we constructed a temperature map from the [Herschel]{} data. ### Column density and temperature map from [Herschel]{} SED fit with all SPIRE bands The standard and often practiced way to derive column density and temperature maps from the [Herschel]{} data is to fit an SED to the observed fluxes of the [Herschel]{} bands for each pixel of the maps [see e.g. @herschel-chamaeleon-sed-fit-oliveira-2014; @herschel-planck-column-density-maps-lombardi-2014; @herschel-physical-dust-properties-massive-sf-regions-battersby-2014]. Here we fit the SED to the three SPIRE bands 250, 350, and $500\,\mu$m. We do not include the two PACS 70 and $160\,\mu$m bands, because those observations were corrupted by stray moonlight and hence are not reliable for an analysis of large-scale structures. Because Lupus [I]{} is optically thin to the dust emission at the considered densities and wavelengths we can model its emission as a modified black body. We assume that the long wavelength emission ($\lambda\ge 250\,\mu$m) of a pixel \[i,j\] comes from a unique species of grains being all at the same equilibrium temperature, and having a power-law wavelength dependent opacity. If $L_{\nu}^{i,j}(\lambda)$ is the monochromatic luminosity of pixel \[i,j\] at wavelength $\lambda$, then it can be expressed as $$L^{i,j}_{\nu}(\lambda) = M^{i,j}_\mathrm{d}\times\kappa(\lambda_0)\left(\frac{\lambda_{0}}{\lambda}\right)^\beta\times4\pi\,B_\nu(\lambda,T^{i,j}_\mathrm{d})$$ where $\kappa$ is the dust opacity, $\beta$ the emissivity index, $B_\nu(\lambda,T_\mathrm{d})$ the black body spectral flux density for a dust temperature $T_\mathrm{d}$. The two free parameters $M^{i,j}_\mathrm{d}$ and $T^{i,j}_\mathrm{d}$ are the dust mass and temperature per pixel of the material along the line-of-sight. Here we adopt a typical $\kappa(\lambda_0)=5.91\,{\rm cm^2\,g^{-1}}$ with $\lambda_0 = 350\,\mu$m for dust grains with thin ice mantels and gas densities $<10^5$cm$^{-3}$ [@dust-opacities-for-ps-cores-ossenkopf-1994]. Using the available information about $\beta$ from the [Planck]{} data, we found that the emissivity index within the Lupus [I]{} cloud lies between $\sim1.6$ and $\sim1.7$. Therefore we fixed its value to $\beta=1.65$ as an average value within the cloud. Considering the low resolution of the $\beta$ map of $30\arcmin$ this is a reasonable approximation. In this way we also limit the number of free parameters in the fit making it more stable. After convolving the 250 and $350\,\mu$m maps to the resolution of the $500\,\mu$m band using the kernels from [@convolution-kernels-aniano-2011], the modified black body fit was performed pixel-by-pixel. From the dust mass $M^{i,j}_\mathrm{d}$ in each pixel the total column density for both dust and gas was then calculated as $$N^\mathrm{H_{SED}}_\mathrm{H_2} = \frac{M_{i,j}\,R}{\mu_\mathrm{H_2}\,m_H}$$ where $R = 100$ is the gas-to-dust mass ratio, $\mu_\mathrm{H_2}=2.8$ the molecular weight per hydrogen molecule, and $m_\mathrm{H}$ the hydrogen atom mass. The resulting dust temperature values $T^{i,j}_\mathrm{d}$ from the fit at each pixel give the temperature map of Lupus [I]{} (shown in Fig. \[img:herschel-temp-map\]) at the resolution of the $500\,\mu$m band, i.e. $\rm FWHM_{500}=36\arcsec$. Since the composition of the dust grains and their density is unknown, the choice of a particular dust model and thus a specific opacity law introduces an uncertainty of the dust opacity and thus of the resulting column density. In order to estimate it, we took different $\kappa(\lambda_0)$ values at $350\,\mu$m for grains with and without ice mantles and the three initial gas densities of $<10^5$, $10^5$, and $10^6$cm$^{-3}$ in the online table of [@dust-opacities-for-ps-cores-ossenkopf-1994]. They vary between $3.64\,{\rm cm^2\,g^{-1}}$ (MRN[^4] without ice mantles and $\rho<10^5$cm$^{-3}$) and $11.3\,{\rm cm^2\,g^{-1}}$ (MRN with thin ice mantles and $\rho=10^6$cm$^{-3}$). Hence, our chosen value of $\kappa(\lambda_0)$ might still vary by a factor of about 2. The statistical error on both the final column density map and the temperature map consists of errors of the calibration, the photometry, and the SED fitting process. We conservatively estimated the sum of these uncertainties to be $\sim20\%$ for both maps. ### Column density map from [Herschel]{} SPIRE $250\,\mu$m map To obtain a column density map with the resolution of the SPIRE $250\,\mu$m band (i.e. $\rm FWHM_{250}=18\arcsec$), we followed the technique described in [@column-density-maps-herschel-juvela-2013]. Using the intensity map of the SPIRE $250\,\mu$m band and the previously calculated temperature map (at the lower resolution) to compute the Planck function $B_\nu(T_\mathrm{d})$, it is possible to gain another factor of two in resolution compared to the SED fitting (to all three SPIRE bands) case. In this way the column density simply is $$N^\mathrm{H_{250}}_\mathrm{H_2} = \frac{I_{250\,\mu{\rm m}}^{i,j}\,R}{B_\nu(250\,\mu{\rm m},T_\mathrm{d}^{i,j})\,\kappa_{250\,\mu{\rm m}}\,\mu_\mathrm{H_2}\,m_H} \label{eq:NH2}$$ with $I_{250\,\mu{\rm m}}^{i,j}$ the intensity of the SPIRE $250\,\mu$m band. We used this higher resolution [Herschel]{} column density map for our further analysis of Lupus [I]{}. ### Column density map from the LABOCA data {#sec:col-dens-map-Laboca} Lupus [I]{} is sufficiently far away from ionizing UV sources to exclude significant amounts of free-free emission. We may therefore assume the sub-mm fluxes to be entirely due to thermal dust emission. To determine whether the cloud is optically thin in this wavelength regime one can look at the peak intensity in the LABOCA map, which is $I^{\rm max}_\nu=1.37$ Jy/beam. The corresponding optical depth can be calculated via formula (3) of [@atlasgal]. $$\tau_{870\,\mu\mathrm{m}}=-\ln\left[ 1-\frac{I_\nu}{\Omega\,B_\nu(T_\mathrm{d})}\right]$$ where $\Omega$ is the beam solid angle. In the case of Lupus [I]{} this yields values of $\tau_{870\,\mu{\rm m}}\le0.01$ for temperatures $T_\mathrm{d}\ge10$ K. Hence the cloud is clearly optically thin in this wavelength regime. Therefore, the $870\,\mu$m intensities are directly proportional to the column densities of the interstellar dust and the line-of-sight extinction. They can be converted to the beam-averaged hydrogen molecule column density via equation (\[eq:NH2\]). We again assume a gas-to-dust mass ratio of $R = 100$ and for consistency extrapolate our dust model used for the [Herschel]{} analysis to the LABOCA wavelength, which yields a value of $\kappa_{870\,\mu{\rm m}} = 1.32\,{\rm cm}^2\,{\rm g}^{-1}$. Because the assumption of a constant dust temperature throughout the cloud is not valid, we used the temperature map from the [Herschel]{} analysis to calculate the black body function at each pixel of the map. In this way one obtains a column density map at the original resolution of the LABOCA map (i.e. $\rm FWHM_{870}=21.2\arcsec$). ### Column density and temperature map from the [Planck]{} data The thermal emission of interstellar dust over the whole sky was captured by the HFI-instrument of [Planck]{} at its 6 available wavelengths between $350\,\mu$m and 3mm. Together with the IRAS $100\,\mu$m data the emission can be well modeled by a modified black body. The details of the model and the fitting procedure are described in [@planck-2013-sky-thermal-dust-emission-model]. The resulting maps of temperature and optical depth (at $850\,\mu$m) can be downloaded from the [Planck]{} archive[^5]. We extracted the part covering Lupus [I]{} from both maps to get gnomonic projected maps. The column density map was then computed from the optical depth map using the relation $$N_\mathrm{H_2} = \frac{5.8\times10^{21}\,{\rm cm^{-2}}}{2}\times0.497\times10^4\,\tau_{\rm 353\,GHz}$$ suggested in [@planck-2013-sky-thermal-dust-emission-model] for regions with higher column density than the diffuse ISM, i.e. molecular clouds. Column density PDFs ------------------- Probability distribution functions (PDFs hereafter) of the column density are a widely used way to characterize the evolution and state of molecular clouds. Both simulators [see e.g. @pdf-simulations-turbulent-clouds-federrath-klessen-2012; @pdf-column-density-simu-ward-2014; @pdf-column-density-evolution-numerical-girichidis-2014] and observers [see e.g. @pdf-column-density-kainulainen-2009; @pdf-column-density-kainulainen-2011; @herschel-column-density-pdf-taurus-palmeirim-2013; @herschel-hii-compression-pdf-tremblin-2014; @pdf-column-density-schneider-2015; @pdf-serpens-core-roccatagliata-2015] use them in their studies and they are a good tool to compare both simulations and observations with each other. As Lupus [I]{} is at near distance to us and lies at a high Galactic latitude, far away from the Galactic plane, we expect the contamination of the map by overlaying foreground or background emission to be small. We normalize the PDFs to the dimensionless $\eta=\ln( N_\mathrm{H_2}/\langle N_\mathrm{H_2}\rangle)$ which gives the opportunity to compare different regions of different column density, as well as the same cloud, but observed with different instruments. For similar observations of the Orion clouds @pdf-column-density-orion-schneider-2013 showed that the effect of varying the resolution of the maps within $18\arcsec$ and $36\arcsec$ did not affect the main features of the PDFs (shape, width, etc.) significantly. The Lupus [I]{} cloud is much closer, yielding a much finer physical resolution and the tests we performed on our maps smoothing them with a grid of Gaussians with different FWHM and rebinning them to several different pixel sizes confirmed their findings. If the difference, however, is as big as e.g. the one in angular resolution between [Herschel]{} and [Planck]{}, it is not longer possible to maintain the features of the PDF unchanged when degrading the [Herschel]{} map by a factor of $\approx8$ (from $500\,\mu$m) to $\approx17$ (from $250\,\mu$m). But in our case both the angular resolution and the pixel size of the [Herschel]{} and the LABOCA column density map are almost the same. Therefore, both maps have sufficiently large numbers of pixels to have a high statistic. The basic shape of the PDF in the low column density regime can be described by a lognormal distribution. In more complicated cases the PDF might also be the combination of two lognormals: $$\label{eq_fit} p(\eta)=\frac{\epsilon_1}{\sqrt{2\pi\sigma_1^2}}\exp\left(\frac{-(\eta-\mu_1)^2}{2\sigma_1^2}\right)+\frac{\epsilon_2}{\sqrt{2\pi\sigma_2^2}}\exp\left(\frac{-(\eta-\mu_2)^2}{2\sigma_2^2}\right)$$ where $\epsilon_i$ is the norm, $\mu_i$ and $\sigma_i$ are the mean logarithmic column density and dispersion of each lognormal. In the high column density regime one often finds a deviation from the lognormal. This can be modeled with a power-law of slope $s$ which is equivalent to the slope of a spherical density profile $\rho(r) \propto r^{-\alpha}$ with $\alpha = -2/s + 1$ [see @sf-efficiency-in-turbulent-magnetized-clouds-federrath-klessen-2013; @pdf-column-density-orion-schneider-2013]. Using a least-square method (Levenberg Marquardt algorithm; Poisson weighting)[^6] we derived the characteristic values of the distributions by fitting the PDFs with either one lognormal to the distribution around the single peak and where applicable a power-law to the high density tail, or with two lognormals to the distributions around the first and second peak, respectively, and a possible power-law tail. We checked the robustness of the fits by performing it on four different binsizes for the histograms (0.05, 0.1, 0.15, 0.2) and looking at the variation of the resulting fit parameters which was of the order $2\%-8\%$. We took the final parameters from the fits for a binsize of 0.1 [see @pdf-column-density-schneider-2015] and conservatively adopted an error of 10% for all fit parameters. The results are summarized in Table \[tbl:PDF-fit-parameters\]. [c c c c c c c c c]{} Region & $\sigma_1$ & $\rm N_{{H_2},peak1}$ & $\sigma_2$ & $\rm N_{{H_2},peak2}$ & $<\rm N_\mathrm{H_2}>$ & $\rm N_{H_2,dev}$ & s & $\alpha$\ & & \[$\times10^{21}\,{\rm cm^{-2}}$\] & & \[$\times10^{21}\,{\rm cm^{-2}}$\] & \[$\times10^{21}\,{\rm cm^{-2}}$\] & \[$\times10^{21}\,{\rm cm^{-2}}$\] & &\ Whole cloud & 0.56 & 0.866 & 0.26 & 2.57 & 1.44 & 3.60 & -2.55 & 1.78\ Center-south & 0.46 & 0.934 & 0.35 & 2.71 & 1.72 & 4.50 & -2.43 & 1.82\ north & 0.55 & 0.794 & 0.18 & 2.23 & 1.10 & – & – & –\ \[tbl:PDF-fit-parameters\] Results {#sec:results} ======= Column density and temperature maps {#sec:results:Column density and temperature maps} ----------------------------------- The [Herschel]{} column density map of Lupus [I]{} is shown in Fig. \[img:Herschel\_coldens\_high\]. The average column density is <$\rm N_\mathrm{H_2}$> $\rm =1.44\times10^{21}\,cm^{-2}$. The map reveals two distinct regions within the filament (see also Fig. \[img:herschel-col-dens-map-3d\]). The north where the column densities are lowest (<$\rm N_\mathrm{H_2}$> $\rm =1.10\times10^{21}\,cm^{-2}$) and just one dense core can be seen [@lupus-clouds-herschel-rygl-2013] and the center-south part where several cores with high column densities above $\rm 10^{22}\,cm^{-2}$ are found and the average column density (<$\rm N_\mathrm{H_2}$> $\rm =1.72\times10^{21}\,cm^{-2}$) is about 60% higher than in the northern part. Comparing our map to the one derived by [@lupus-clouds-herschel-rygl-2013] we note that our values are lower by a factor of about $2-2.5$. This can be explained by the use of a different dust model (opacity and dust spectral index) and the inclusion of the PACS $160\,\mu$m band in their work. The extinction maps created from 2MASS and [Spitzer]{} data [see @lupus-clouds-spitzer-c2d-chapman-2007; @lupus-2mass-extinction-maps-lombardi-2008; @lupus-clouds-spitzer-c2d-merin-2008] have a resolution that is lower by a factor of $\sim7-30$. Nevertheless, between those maps and our column density map is a clear resemblance. The column density map of Lupus [I]{} we obtained from LABOCA (see Fig. \[img:Laboca\_coldens\]) is dominated by the dozen cores in the center-south part and the denser dust in the central part of the filament. The average column density is <$\rm N_\mathrm{H_2}$> $\rm =1.47\times10^{21}\,cm^{-2}$. The [Planck]{} column density map of Lupus [I]{} with a resolution of $5\arcmin$ reveals the basic structure of the cloud which agrees well with the column density maps from LABOCA and [Herschel]{}. This is consistent with what is shown by the other two maps when smoothed to the resolution of [Planck]{} and thus smoothing out the highest column density peaks. The temperature map obtained from the [Herschel]{} SED fit (see Fig. \[img:herschel-temp-map\]) shows an anti-correlation of the temperature with the column density. The densest parts are the coldest and the less dense the material the warmer it becomes. Also one sees again a difference between the northern and the center-south part of the cloud. In the north dust temperatures between $\approx15-23$ K with a mean and median of $\approx19$ K are found. But only in the dense pre-stellar core and slightly north of it the temperature drops down to $\approx15$ K. The maximum temperature in the center-south part is $\approx22$ K at the edges of the filament. The inner part of the filament has an average temperature of $\approx17$ K. But it drops down to even 11 K in the densest cores. From the histogram of the dust temperatures (see Fig. \[img:herschel-temp-histo\]) one can see that most of the dust in the north (orange histogram) ranges between $\approx18-20$ K whereas in the center-south part (green) it is between $\approx16-18$ K. This means that the dust in the center-south is on average 2 K colder than in the north. ![3D-surface plot of the [Herschel]{} $250\,\mu$m column density map. One can clearly identify the cores in the center-south of Lupus [I]{} and see the on average lower column density in the northern part of the cloud. The red dashed boxes mark the two regions we distinguish in this work.[]{data-label="img:herschel-col-dens-map-3d"}](Lupus_I_Herschel_NH2_3D_with_region_labels.png){width="50.00000%"} Column density PDFs ------------------- We derived PDFs of the Lupus [I]{} cloud from the [Herschel]{} and LABOCA column density maps for the two distinct parts of the cloud (north and center-south; see Fig. \[img:herschel-col-dens-map-3d\]), as well as for the whole cloud. They are shown in Fig. \[img:PDFs\]. For a correct interpretation of the PDFs one has to consider the completeness limit of the underlying column density map. Here we adopted the lowest closed contour as such a limit which is $1.1\times10^{21}$ and $2\times10^{21}\,{\rm cm^{-2}}$ for the [Herschel]{} and LABOCA map, respectively. It is marked by the vertical dashed line in the plots of Fig. \[img:PDFs\]. Looking at the entire cloud the [Herschel]{} PDF is very complex with clear deviations from a simple lognormal distribution which would be expected for a cloud that is dominated by isothermal, hydrodynamic turbulence [e.g. @gravitational-collaps-in-mc-klessen-2000]. The distribution shows two peaks in the low column density regime and a power-law tail in the high density end. The first peak at $\rm N_{{H_2},peak1}=8.66\times10^{20}\,{\rm cm^{-2}}$ falls below our completeness limit, so it might be just reflecting the drop of observational sensitivity and will not be used for interpretation. But the distribution above the limit can be well represented by a fit of two lognormals around the first and second peak which is at $\rm N_{{H_2},peak2}=2.57\times10^{21}\,{\rm cm^{-2}}$. The width of the first component is $\sigma_1=0.56$, but could also be broader due to the possible underestimate of the column densities below the completeness limit. Nevertheless, it is more than twice as broad as the width of the second lognormal ($\sigma_2=0.26$) and larger than in other nearby clouds like Maddalena, Auriga [@pdf-column-density-orion-schneider-2013] or Aquila [@pdf-column-density-schneider-2015]. This can be a sign of broadening by turbulence and external compressive forcing [see @turbulence-ism-forcing-federrath-2010; @pdf-hii-turbulence-shock-simulation-tremblin-2012; @sf-efficiency-in-turbulent-magnetized-clouds-federrath-klessen-2013]. The power-law tail follows a slope of $\rm s=-2.55$ or corresponding $\alpha=1.78$ for a spherical density distribution. The deviation from the lognormal into the power-law occurs near $\rm N_{{H_2},dev}=3.6\times10^{21}\,{\rm cm^{-2}}$ which is slightly below the recent values by [@pdf-column-density-schneider-2015] who found the transition into the power-law tail to be at $A_V\sim4-5$ mag[^7] in their investigation of four low-mass and high-mass star forming regions. But it is higher than the result of [@pdf-serpens-core-roccatagliata-2015] who find the deviation to occur at $A_V\sim1.2-2$ mag in the Serpens Core region. From the [Herschel]{} data the center-south part of the cloud shows a PDF very similar to that of the entire cloud. Here again two peaks are present and the distribution above the completeness limit can be fitted by a double lognormal and a power-law tail. The width of the first fitted lognormal is narrower, the width of the second broader than those of the entire cloud with values of $\sigma_1=0.46$ and $\sigma_2=0.35$. The slope of the power-law is slightly flatter with $\rm s=-2.43$, but leading to a comparable exponent $\alpha=1.82$. Also in the northern part of the cloud the double peak lognormal profile of the PDF from [Herschel]{} is found again. But now there is no excess above the second lognormal at higher densities. The peak position of the first lognormal is lower ($\rm N_{{H_2},peak1}=7.94\times10^{20}\,{\rm cm^{-2}}$) compared to the center-south and the whole cloud, but with similar width ($\sigma_1=0.55$). The peak position of the second lognormal ($\rm N_{{H_2},peak2}=2.23\times10^{21}\,{\rm cm^{-2}}$) is lower than that of the whole cloud and its width ($\sigma_2=0.18$) is approximately half as broad. Although it is not possible to strictly constrain the shape of the PDF below the completeness limit, one can say that the distribution above shows not just two components (lognormal and a power-law) but at least three because of the second peak that appears before the deviation into the power-law. Such a shape could be attributed to an initial turbulent cloud which later was compressed by an external driving agent. Such an agent can either be an ionization front, an expanding shell driven by winds or/and supernovae or colliding flows. The initial lognormal form of the PDF of the cloud develops a second component caused by the dense compressed gas. This behavior was found in simulations by [@pdf-hii-turbulence-shock-simulation-tremblin-2012] of an initial turbulent medium that is ionized and heated by an ionization source. They concluded that a double peak is present when the ionized gas pressure dominates over the turbulent ram pressure of the cloud. [@pdf-double-peaks-colliding-flows-matsumoto-2015] studied the evolution of turbulent molecular clouds swept by a colliding flow. They found that the PDF exhibits two peaks in the case when the Mach numbers of the initial turbulence and the colliding flow are of the same order $M=10$ (model HT10F10; see their Fig. 11b) and the line of sight is perpendicular to the colliding flow. Then the low column density peak represents the colliding flow and the higher column density peak the sheet cloud, respectively. Observationally, double peak PDFs have been reported and studied recently by [@pdf-rosette-cluster-formation-Schneider-2012], [@pdf-auriga-california-herschel-harvey-2013], and [@herschel-hii-compression-pdf-tremblin-2014] for several nearby clouds exposed to an ionization source. Areas of the cloud close to the ionization front indeed showed the predicted second peak in the PDF due to the compression induced by the expansion of the ionized gas into the molecular cloud. In the case of Lupus [I]{} the source of the compression is very likely the expanding H[I]{} shell around the USco sub-group of Sco-Cen (possibly together with a supernova that exploded within USco) and the wind bubble of the remaining B-stars of UCL pushing from the eastern and western side of the cloud, respectively. This scenario will be discussed in more detail in Sec. \[sec:surrounding\]. Lognormal PDFs indicate shock waves [@shocks-ism-imf-kevlahan-2009] or turbulence [e.g. @structure-flows-ism-vazquez-1994; @turbulence-ism-forcing-federrath-2010]. @pdf-hii-turbulence-shock-simulation-tremblin-2012 have shown in a simulation of Stroemgren spheres advancing into turbulent regions that the region compressed by the ionization front also has a lognormal density PDF, but shifted to higher density due to the compression. A PDF of a region that includes compressed and undisturbed parts of the cloud will thus show a double peak PDF. The power-law tail that is seen in our PDFs could be explained by active star formation and the transition to a gravity dominated density regime represented by the star forming cores. This can be shown by comparing the PDFs of the two regions in the north and the center-south. In the north where there is almost no star formation and just one pre-stellar core and $\approx20$ unbound cores can be found [@lupus-clouds-herschel-rygl-2013], the PDF shows no power-law tail in the high density regime. In the center-south instead where almost all the star formation activity takes place with plenty of very young bound cores, the PDF shows the power-law tail very clearly. Also numerical studies have shown that the PDF for an actively star forming region develops a clear deviation from the lognormal in form of a power-law tail [@pdf-evolution-gravity-turbulence-ballesteros-2011; @pdf-column-density-evolution-numerical-girichidis-2014]. With LABOCA the PDF of Lupus [I]{} can be modeled as one lognormal with a power-law tail at high densities ($\rm N_{{H_2},dev}=6.0\times10^{21}\,{\rm cm^{-2}}$). The peak of the lognormal distribution lies at $\rm N_{{H_2},peak}=1.43\times10^{21}\,{\rm cm^{-2}}$ which is below the completeness limit of the LABOCA column density map. The width of the lognormal is broad ($\sigma=0.56$), but uncertain due to the peak position and thus possibly even broader. The power-law slope ($s=-2.66$, $\alpha=1.75$) is very similar to the one of the [Herschel]{} PDF indicating the star formation activity in the cloud. The PDF of the center-south region shows the same behavior as the one for the whole cloud. The lognormal part is slightly broader ($\sigma=0.60$), but the power-law slope ($s=-2.59$, $\alpha=1.77$) and the deviation point ($6.0\times10^{21}\,{\rm cm^{-2}}$) are almost the same compared to the whole cloud. In the northern part only the now narrower lognormal is seen instead ($\sigma=0.52$), indicating the lack of star formation activity and the dominance by the turbulence induced trough the compression. In all three cases the positions of the peak are very close to each other. As was already seen in the column density map from LABOCA, its PDF shows only the dense cores and the intercore medium. The instrument is not sensitive to all material that can still be probed with [Herschel]{}. With LABOCA we thus sample more the second lognormal of the [Herschel]{} PDF and the power-law tail. Core distribution {#sec:core-distribution} ----------------- For the core analysis in the LABOCA map we used the [Clumpfind]{} package [@clumpfind]. It decomposes the emission of the map into a set of clumps or cores contouring the data at given threshold levels. The results can be found in Tab. \[tbl:laboca-cores\] and the distribution of the cores (represented by the white crosses) is shown in Fig. \[img:laboca-map\]. The algorithm identified 15 different cores with masses between 0.07 and $1.71\,M_\odot$. Their total mass is $8.37\,M_\odot$ and their total flux is 25.39 Jy. This corresponds to $\approx5\%$ of the total mass of Lupus [I]{} (see Sec. \[sec:total-mass-lupus\]) and $\approx5\%$ of the total flux of the LABOCA map, respectively. For the computation of the core masses, we derived their temperatures from the [Herschel]{} SPIRE SED fit temperature map as a mean temperature within the ellipse representing the core. With this temperature the mass of the cores was calculated following [@atlasgal] $$M=\frac{d^2 F_\nu R}{B_\nu(T_\mathrm{d}) \kappa_\nu}$$ where $d$ is the distance to Lupus [I]{}, $F_\nu$ the total flux of the core, $R=100$ the dust-to-gas ratio, and $\kappa_\nu = 1.32\,{\rm cm}^2\,{\rm g}^{-1}$ (see Sec. \[sec:col-dens-map-Laboca\]). Most interestingly, all but one core are found in the center-south part of Lupus [I]{}. This confirms our interpretation of the column density maps that is the part where active star formation is going. The distribution of the cores from the [Herschel]{} data [@lupus-clouds-herschel-rygl-2013] agrees with this picture and the core distribution from our LABOCA map. The northern part of Lupus [I]{} is mainly populated by unbound cores and just one pre-stellar core wheras the center-south is dominated by pre-stellar cores. Many of those coincide with our LABOCA cores. However, a direct assignment is not easily possible since the [Herschel]{} coordinates can only be estimated from the source images in the paper of [@lupus-clouds-herschel-rygl-2013] and in some cases several [Herschel]{} sources seem to be on the position of one LABOCA source or vice-versa. [@lupus-clouds-spec-cores-Benedettini-2012] found eight dense cores in Lupus [I]{} using high-density molecular tracers at 3 and 12 mm with the Mopra telescope (red diamonds in Figure \[img:c2d-ysos-on-herschel\]). Seven of those have one or more counterparts in our LABOCA map. The matches are given in Table \[tbl:laboca-cores\]. They classify five of their cores as very young protostars or pre-stellar cores (Lup1 C1-C3, C5, and C8) and the remaining three also as very likely to be protostellar or pre-stellar. For three of their cores (Lup1 C4, C6, and C7) they calculated a kinetic temperature of $\approx12$ K. This agrees within 20% with the core temperatures derived from the [Herschel]{} SPIRE SED fit temperature map. From the 17 young stellar objects found in Lupus [I]{} by the [Spitzer]{} c2d near-infrared survey [@lupus-clouds-spitzer-c2d-chapman-2007; @lupus-clouds-spitzer-c2d-merin-2008] 11 lie within the boundaries of the LABOCA map (blue boxes in Figure \[img:c2d-ysos-on-herschel\]). But only one object (IRAS 15398–3359) clearly matches one of our LABOCA cores (\#1). This is the long known low-mass class 0 protostar IRAS 15398–3359 (inside the B228 core) which has a molecular outflow [see recently e.g. @lupusI-iras-outflow-oya-2014; @outflows-in-low-mass-ps-Dunham-2014]. Two other objects of the c2d survey are close, but offset by $\approx0.5\arcmin$ to the center of our cores \#3 and \#5, respectively. Nevertheless, each of those two objects still lies within the boundaries of the corresponding ellipse representing the LABOCA core (see Table \[tbl:laboca-cores\]). Most of the LABOCA cores are potentially at a very early evolutionary stage, i.e. without a protostar inside to heat it and eventually destroy the surrounding dust envelope to allow near-infrared radiation to escape. In fact the only cores with a $70\,\mu$m counterpart, which is a good proxy for a protostar inside a core, are number \#1 and \#3. Therefore, we do not expect to find many [Spitzer]{} counterparts. ![[Herschel]{} column density map of the center-south region with contour levels of 1.1, 4, 7, and $10\times10^{21}\,{\rm cm^{-2}}$. The different symbols mark the center positions of the cores found with LABOCA and of objects found in other studies as discussed in Section \[sec:core-distribution\].[]{data-label="img:c2d-ysos-on-herschel"}](Lupus_I_cores_All_YSOs_on_HerschelColDens_Center-south-gimp.png){width="35.00000%"} [@lupus-clouds-13co-cores-vilas-boas-2000] found 15 (14) condensations in $\rm C^{18}O$ ($\rm^{13}CO$) with the 15 m SEST telescope. Eight cores lie within our LABOCA map (orange crosses in Figure \[img:c2d-ysos-on-herschel\]) and of those three (Lu7, Lu10, B228) coincide with LABOCA detected cores (\#12, \#5, and \#1). Their $\rm C^{18}O$ excitation temperatures are 9, 11, and 10 K, respectively. This is $\approx20-40\%$ lower than the temperatures from the [Herschel]{} map. For cores Lu7 and Lu10 they calculated a mass from $\rm C^{18}O$ of 11.7 and 3.1 $M_\odot$, respectively. This is a factor of $\approx20$ and $\approx7$ higher than the LABOCA masses. But the sizes of their condensations were on average larger than at least three times their beam size of $48\arcsec$. Figure \[img:c2d-ysos-on-herschel\] shows the [Herschel]{} column density map of the center-south region with contour levels of 1.1, 4, 7, and $10\times10^{21}\,{\rm cm^{-2}}$. Overploted are the LABOCA cores found in this study (green circles), as well as the above mentioned cores and YSOs found by [@lupus-clouds-spec-cores-Benedettini-2012] (red diamonds), [@lupus-clouds-13co-cores-vilas-boas-2000] (orange crosses), and the [Spitzer]{} c2d survey (blue boxes), respectively. Our findings confirm that Lupus [I]{} harbors a population of very young cores that are forming stars just now and probably have ages below 1 Myr. This could support the idea of an external shock agent, like the USco shell, sweaping up the cloud and triggering the simultaneous formation of new stars within the cloud. [r c c c c c c c c c]{} \# & Ra & Dec & FWHM$_{\rm x}$ & FWHM$_{\rm y}$ & I$_{\rm max}$ & F$_{\rm tot}$ & M$_{\rm core}$ & T$_{\rm core}$ & Counterpart\ & & & $[\arcsec]$ & $[\arcsec]$ & \[Jy/beam\] & \[Jy\] & \[M$_\odot$\] & \[K\] &\ 1 & 15:43:01.68 & $-$34:09:08.9 & 58.22 & 44.38 & 1.37 & 5.61 & 1.71 & 14.0 & Lup1 C4\ 2 & 15:44:59.85 & $-$34:17:08.8 & 62.51 & 44.09 & 0.51 & 3.97 & 1.59 & 12.0 & Lup1 C6\ 3 & 15:45:13.58 & $-$34:17:08.8 & 46.56 & 33.45 & 0.33 & 1.48 & 0.40 & 15.0 & Lup1 C7\ 4 & 15:45:16.02 & $-$34:16:56.6 & 35.31 & 34.95 & 0.31 & 1.00 & 0.27 & 15.0 & Lup1 C7\ 5 & 15:45:03.80 & $-$34:17:57.3 & 57.22 & 42.24 & 0.28 & 1.41 & 0.46 & 13.5 & Lup1 C6\ 6 & 15:42:52.86 & $-$34:08:02.1 & 96.42 & 49.94 & 0.28 & 3.85 & 1.34 & 13.0 & Lup1 C3\ 7 & 15:42:43.57 & $-$34:08:14.2 & 70.62 & 44.38 & 0.25 & 1.64 & 0.57 & 13.0 & Lup1 C3\ 8 & 15:42:44.55 & $-$34:08:32.5 & 30.55 & 20.85 & 0.25 & 0.48 & 0.17 & 13.0 & Lup1 C3\ 9 & 15:42:44.06 & $-$34:08:44.6 & 36.82 & 19.26 & 0.24 & 0.49 & 0.17 & 13.0 & Lup1 C3\ 10 & 15:42:50.42 & $-$34:08:38.5 & 53.28 & 30.34 & 0.22 & 1.05 & 0.36 & 13.0 & Lup1 C3\ 11 & 15:45:25.10 & $-$34:24:01.8 & 42.46 & 69.57 & 0.22 & 1.63 & 0.53 & 13.5 & Lup1 C8\ 12 & 15:42:23.02 & $-$34:09:33.2 & 68.97 & 46.08 & 0.21 & 1.72 & 0.53 & 14.0 & Lup1 C1\ 13 & 15:39:09.92 & $-$33:25:30.9 & 27.24 & 27.70 & 0.20 & 0.34 & 0.07 & 17.0 & –\ 14 & 15:45:55.75 & $-$34:29:23.7 & 28.10 & 26.28 & 0.20 & 0.29 & 0.08 & 14.5 & –\ 15 & 15:42:47.90 & $-$33:53:15.3 & 43.80 & 19.59 & 0.19 & 0.43 & 0.12 & 15.0 & Lup1 C2\ \[tbl:laboca-cores\] Total mass estimates of Lupus [I]{} {#sec:total-mass-lupus} ----------------------------------- To derive the total mass of the cloud from the three different column density maps we defined a polygon around the filament (delineated in Fig. \[img:laboca-map\] on the LABOCA map) to derive the total mass always in the same area ($\rm \approx1\,deg^2$). The total gas and dust mass was then calculated via the formula $$\rm M^{\rm tot}_{Lupus~{\rm I}}=\sum\,N_\mathrm{H_2}\,\mu_\mathrm{H_2}\,m_H\,A_p$$ with $A_p$ the area of a pixel in $\rm cm^2$. As the common lower level we chose $\rm N_\mathrm{H_2}>10^{21}\,cm^{-2}$ which corresponds to $\rm A_V>1$ mag. The resulting total mass for Lupus [I]{} is $\rm M^{\rm tot}_{Lupus~{\rm I}}\approx171\,M_\odot$, $\approx174\,M_\odot$, and $\approx164\,M_\odot$ for the [Planck]{}, [Herschel]{}, and LABOCA data. This means that the total masses calculated from the three data sets agree with each other. Comparing the total mass of the cloud to the total mass in cores from LABOCA ($\approx8\,M_\odot$) one sees that only about 5% of the mass is concentrated in the densest condensations. Most of the dust and gas is in more diffuse components. The most recent literature value for the total mass of Lupus [I]{} is from the [Herschel]{} data by [@lupus-clouds-herschel-rygl-2013]. They calculated the total mass in a much bigger area than this work ($\approx 4.5\,{\rm deg}^2$ compared to our $\approx 1\,{\rm deg}^2$) and used a different method to create their column density map as mentioned already in Sec. \[sec:results:Column density and temperature maps\]. Therefore, a difference of about a factor $4-5$ arises between their value of $M=830\,M_\odot$ (for $A_V>2$ mag) and our finding. Other literature values cover a wide range of total masses for Lupus [I]{}, depending on the tracer used and the size of the area that was considered. From the [Spitzer]{} c2d near-infrared extinction maps [@lupus-clouds-spitzer-c2d-merin-2008] determined a total mass of $479\,M_\odot$ for $A_V>3$. Various CO measurements [e.g. @lupus-clouds-13co(1-0)-Tachihara-1996; @lupus-clouds-c18o-Hara-1999; @lupus-clouds-13co(2-1)-Tothill-2009] yielded values of $\approx280 - 880\,M_\odot$. Direct comparisons with our values are not always possible since all maps cover different parts of the Lupus [I]{} cloud complex and the material is not homogeneously distributed to allow scaling the mass with the area. But we note that our values agree with most of the literature values within a factor of 2-3 which is expected considering the uncertainties in the choice of the dust model, the dust-to-gas ratio and the CO-to-H$_2$ conversion factors. The surroundings of Lupus [I]{} and the interaction with USco and UCL {#sec:surrounding} ===================================================================== Best suited for looking at the dust surroundings of Lupus [I]{} are the [Planck]{} data (see Fig. \[img:lupus-usco-xray-planck\] and Fig. \[img:lupus-usco-xray-planck-contours\]). These observations at 350, 550, and $850\,\mu$m cover the whole sky at a resolution of $5\arcmin$. Lupus [I]{} lies on the eastern edge (Galactic coordinates) of a ring-like dust ridge (labeled in Fig. \[img:lupus-usco-xray-planck-contours\]) that extends from about $b=+10\degr$ to $b=+25\degr$ in Galactic latitude with a center at about $l=+345\degr$, $b=17.5\degr$. The ridge is $\sim5\degr$ wide which corresponds to $\sim13$ pc at the distance of Lupus [I]{}. Besides Lupus [I]{} it consists of several small molecular clouds extending north of Lupus [I]{} and then bending towards the east connecting with the $\rho$ Ophiuchus molecular cloud on the opposite site of Lupus [I]{}. Further west of Lupus [I]{} no dust emission is seen in the [Planck]{} maps. The same is true for the inside of the dust ridge. Between $\rho$ Ophiuchus and Lupus [I]{} a roundish dust void is seen. However, these two dust voids on either side of Lupus [I]{} are filled with hot X-ray gas which can be seen with ROSAT (left in Fig. \[img:lupus-usco-xray-planck\]). Fig. \[img:lupus-usco-xray-planck-contours\] shows the dust emission seen by [Planck]{} in $350\,\mu$m. Overlayed in white are the contours of the ROSAT diffuse X-ray emission at 3/4 keV[^8]. Inside USco the contours follow the edge of the dust ridge indicating that the hot X-ray gas might be in contact with the cold dust. On the western side of Lupus [I]{} the contours mark the outline of the second roundish X-ray emission seen in ROSAT. The cyan dots mark the remaining early B-type stars of UCL. Both contours seem to wrap around Lupus [I]{} what might be a sign that the cloud is embedded in hot ISM. These dust voids and the observed X-ray gas might be explained by the cumulative feedback of the massive stars in the USco and UCL sub-groups of Sco-Cen. Their creation has been interpreted in a scenario of propagating molecular cloud formation and triggered cloud collapse and star formation taking place within the last 17 Myr in the UCL and USco sub-groups of Scorpius-Centaurus [see @sco-cen-HI-degeus-1992; @ob-stars-triggered-sf-preibisch-zinnecker-2007]. The expansion of the UCL H[I]{} shell that started $\sim10$ Myr ago and was driven by winds of the massive stars and supernovae explosions has probably cleared out almost all the dust and molecular material west of Lupus [I]{}. But the observed X-ray gas on that side is probably not related to the supernova explosions, because the X-ray luminosity of supernova-heated superbubbles dims on a timescale of typically less than a million years via expansion losses and mixing with entrained gas [@feedback-superbubbles-xrays-krause-2014]. The shell today has a radius of $\sim110$ pc [as seen from H[I]{} data; @sco-cen-HI-degeus-1992], generally consistent [compare, e.g. @superbubble-evolution-disk-galaxies-baumgartner-breitschwerdt-2013; @superbubbles-vishniac-krause-2013] with the inferred age of the stellar group, about 17 million years. Therefore, this X-ray gas is probably currently heated by the remaining B-stars of UCL. Six of those stars south-west of Lupus [I]{} ($325\degr<l<335\degr$ and $10\degr<b<15\degr$; see Fig. \[img:lupus-usco-xray-planck-contours\]) lie at positions that favor them as likely being the sources of this wind bubble. The USco H[I]{} shell has started its expansion $\sim5$ Myr ago powered by the winds of the OB-stars and quite possibly a recent supernova explosion, about 1.5 Myr ago, as suggested by the detection of the 1.8 MeV gamma ray line towards the USco sub-group and the detection of the pulsar PSR J1932+1059 [@sco-cen-al26-diehl-2010 and references therein]. What now forms the before mentioned dust ridge is probably the remaining material of the USco parental molecular cloud that was swept up by the advancing USco shell leaving the dust void filled with X-ray gas. From simulations, we expect large scale oscillations of the hot, X-ray emitting gas [@feedback-superbubbles-xrays-krause-2014], because the energy source is never exactly symmetric. It seems like the UCL wind bubble could be colliding with the USco H[I]{} shell right at the position of Lupus [I]{} squeezing it in-between. This wind bubble might have provided a counter-pressure to the expanding USco shell and thus favored this position for an additional compression of the shell material. In this way a new molecular cloud could have been created there and it might explain why we do not see more very young star forming clouds (except $\rho$ Oph) distributed within the wall of the USco shell. With its distance of 150 pc the cloud also is neither in the foreground of the B-stars of UCL that have an average distance of $\sim140$ pc [@sco-cen-hipparcos-dezeeuw-1999] nor in front of USco [$\sim145$ pc; @usco-full-population-preibisch-2002]. To rule out shadow effects completely, however, a more comprehensive analysis of the X-ray data would be required. From our preliminary analysis of the GASS H[I]{} data [@gass-hi-mcclure-griffiths-2009; @gass-hi-kalberla-2010] we see that Lupus [I]{} is indeed embedded within the expanding shell around USco (Kroell et al., in prep.). The shell has an expansion velocity of $\approx7\,{\rm km\,s^{-1}}$ and a thickness of $\approx6$ pc. We estimate the current outer radius to be $\approx36$ pc (both inner and outer radius of the USco shell from our H[I]{} model are delineated in Fig. \[img:lupus-usco-xray-planck\]). These findings complement previous suggestions of an interaction of Lupus [I]{} (and $\rho$ Oph) with the two sub-groups of Sco-Cen that were originally proposed by [@sco-cen-HI-degeus-1992]. From the spatial and velocity structure of $\rm ^{12}CO$ observations [@lupus-clouds-12co-Tachihara-2001] also found evidence for an interaction. [@lupus-clouds-13co(2-1)-Tothill-2009] in their analysis of $\rm ^{13}CO$ and CO(4-3) found enhanced line widths at the western end of Lupus [I]{} and a velocity gradient across the filament, i.e. in the direction of the USco shell expansion, of $\rm \sim1\,km\,s^{-1}$ which they conclude to be consistent with a dynamical interaction between Lupus [I]{} and the USco H[I]{} shell. [@lupus-submm-polarimetry-matthews-2014] have found that the large-scale magnetic field is perpendicular to the Lupus [I]{} filament, i.e. points in the direction of the USco shell expansion. This might have favored the accumulation of cold, dense atomic gas along the field lines and promote faster molecule formation [@rapid-formation-mcs-hartmann-2001; @mc-formation-mhd-vazquez-semadeni-2011]. All these are hints towards the scenario that Lupus [I]{} was affected by large-scale external compressing forces coming from the expansion of the USco H[I]{} shell and the UCL wind bubble. This might explain its position, orientation and elongated shape, the appearance of a double peak in the PDF and the large population of very young pre-stellar cores that are seen with both LABOCA and [Herschel]{}. ![image](ScoCen-Vergleich-XrayRosatNEW-Planck850-eps-converted-to.pdf){width="\linewidth"} ![image](UCL_B1-3_stars_Planck_Xray-eps-converted-to.pdf){width="\linewidth"} Summary and Conclusions {#sec:summary} ======================= From our LABOCA observations of Lupus [I]{} as well as from archival [Herschel]{} and [Planck]{} data we created column density maps of the cloud. In addition we calculated a temperature map from an SED fit using the three SPIRE bands of [Herschel]{}. All maps suggest that the cloud can be divided into two distinct regions. The northern part that has on average lower densities and higher temperatures as well as no active star formation and the center-south part with dozens of pre-stellar cores where density and temperature reach their maximum and minimum, respectively. From the LABOCA and [Herschel]{} maps we derived column density PDFs for the entire cloud as well as for the two above mentioned regions separately. The [Herschel]{} PDF of Lupus [I]{} showed a double peak profile with a power-law tail. The power-law can be attributed to the star formation activity in the center-south part of the cloud since it disappears in the PDF of the northern part. However, the double peak profile is conserved throughout the cloud and possibly arises due to the large-scale compressions from the cumulative massive star feedback of the Sco-Cen sub-groups. Such a sign for compression in the PDF was previously found in both observations and simulations of advancing ionization fronts, also supported by simulations of colliding turbulent flows. With LABOCA we probe the denser parts of Lupus [I]{} and find one lognormal and the power-law tail for the whole cloud and the center-south. The PDF of the northern part shows only the lognormal behavior. The distribution of the 15 cores that were found in our LABOCA map confirms that only the center-south part of Lupus [I]{} is actively forming stars whereas the north is quiescent. The cores have masses between 0.07 and $1.71\,M_\odot$ and a total mass of $\approx8\,M_\odot$. As the cloud has a total gas and dust mass of $\approx164\,M_\odot$ (from LABOCA for $\rm N_\mathrm{H_2}>10^{21}\,cm^{-2}$) this means that $\sim5\%$ of the mass is in cores. All of those cores are pre-stellar or protostellar confirming that we are witnessing a large star formation event [see e.g. @lupus-clouds-spec-cores-Benedettini-2012; @lupus-clouds-herschel-rygl-2013]. The total mass from [Herschel]{} and [Planck]{} for $\rm N_\mathrm{H_2}>10^{21}\,cm^{-2}$ was $\approx174\,M_\odot$ and $\approx171\,M_\odot$, respectively. We argue that the main driving agents in the formation process of Lupus [I]{} are the advancing USco H[I]{} shell, which most likely contains the cloud, and the UCL wind bubble interacting with the USco shell and hence squeezing Lupus [I]{} in-between these two shells in a complex manner. The age of the population of young stellar objects suggests a compression event $\sim1-2$ Myr ago, which could be the interaction between the two shells and possibly a supernova explosion as suggested by the gamma ray data [@sco-cen-al26-diehl-2010]. This large-scale compressions might be the reason for the position, orientation and elongated shape of Lupus [I]{}, as well as the double peak PDFs and the population of very young pre-stellar cores we found. In future work we will analyze our newly performed high-resolution APEX CO line observations to study the kinematics of the Lupus [I]{} cloud. We will also compare our observational findings with dedicated numerical simulations and search for signs in the theoretical PDF that will allow us to distinguish between different scenarios of molecular cloud formation. We would like to thank our anonymous referee for his constructive comments which helped to improve this paper. This work was supported by funding from Deutsche Forschungsgemeinschaft under DFG project number PR 569/10-1 in the context of the Priority Program 1573 “Physics of the Interstellar Medium”. Additional support came from funds from the Munich Cluster of Excellence “Origin and Structure of the Universe”. We thank Jean-Philippe Bernard for computing the [Planck]{} offsets for the [Herschel]{} maps. The Herschel spacecraft was designed, built, tested, and launched under a contract to ESA managed by the Herschel/Planck Project team by an industrial consortium under the overall responsibility of the prime contractor Thales Alenia Space (Cannes), and including Astrium (Friedrichshafen) responsible for the payload module and for system testing at spacecraft level, Thales Alenia Space (Turin) responsible for the service module, and Astrium (Toulouse) responsible for the telescope, with in excess of a hundred subcontractors. Based on observations obtained with [Planck]{} (<http://www.esa.int/Planck>), an ESA science mission with instruments and contributions directly funded by ESA Member States, NASA, and Canada. HIPE is a joint development by the [Herschel]{} Science Ground Segment Consortium, consisting of ESA, the NASA [Herschel]{} Science Center, and the HIFI, PACS and SPIRE consortia. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France. We acknowledge the use of NASA’s [SkyView]{} facility (http://skyview.gsfc.nasa.gov) located at NASA Goddard Space Flight Center. [^1]: The Atacama Pathfinder Experiment (APEX) is a collaboration between the Max-Planck-Institut für Radioastronomie (MPIfR), the European Southern Observatory (ESO), and the Onsala Space Observatory (OSO). [^2]: [Herschel]{} is an ESA space observatory with science instruments provided by European-led Principal Investigator consortia and with important participation from NASA. [^3]: <http://healpix.jpl.nasa.gov/html/intro.htm> [^4]: Mathis-Rumpl-Nordsieck size distribution of interstellar grains [@dust-opacity-mrn-1977] [^5]: <http://irsa.ipac.caltech.edu/data/Planck/release_1/all-sky-maps/maps/HFI_CompMap_ThermalDustModel_2048_R1.20.fits> [^6]: [MPCURVEFIT]{} routine for IDL [^7]: $\rm N(H_2)/A_v=0.94\times10^{21}\,cm^{-2}\,mag^{-1}$ [@nh2-to-av-factor-bohlin-1978]. [^8]: The very strong X-ray emission seen near the position of the B2IV star south-west of Lupus [I]{} ($\kappa$Cen; $l=326.872\degr, b=14.754\degr$) is neither related to the star nor to Sco-Cen, but most likely caused by the Quasar \[VV2006\] J150255.2$-$415430.	{ "pile_set_name": "ArXiv" }
--- abstract: 'The young massive stellar object G35.20-0.74 was observed in the mid-infrared using T-ReCS on Gemini South. Previous observations have shown that the near infrared emission has a fan-like morphology that is consistent with emission from the northern lobe of a bipolar radio jet known to be associated with this source. Mid-infrared observations presented in this paper show a monopolar jet-like morphology as well, and it is argued that the mid-infrared emission observed is dominated by thermal continuum emission from dust. The mid-infrared emission nearest the central stellar source is believed to be directly heated dust on the walls of the outflow cavity. The hydroxyl, water, and methanol masers associated with G35.20-0.74 are spatially located along these mid-infrared cavity walls. Narrow jet or outflow cavities such as this may also be the locations of the linear distribution of methanol masers that are found associated with massive young stellar objects. The fact that G35.20-0.74 has mid-infrared emission that is dominated by the outflow, rather than disk emission, is a caution to those that consider mid-infrared emission from young stellar objects as only coming from circumstellar disks.' author: - 'James M. De Buizer' title: 'The Remarkable Mid-Infrared Jet of Massive Young Stellar Object G35.20-0.74' --- Introduction ============ While molecular outflows are an apparently ubiquitous phenomena in regions of high-mass star formation (Shepherd and Churchwell 1996; Zhang et al. 2001; Beuther et al. 2002), clear examples of individual young massive stars with well-defined bipolar jets are relatively few. One such example is G35.20-0.74, a massive star formation region containing an early B-type star surrounded by an ultracompact HII region (UCHII region) with a bipolar jet-like radio structure. Water, OH, and methanol masers (Forster & Caswell 1989, Hutawaraforn & Cohen 1999) are all associated with this massive young stellar object, though the nature of their relationship with respect to G35.20-0.74 is still unclear. This jet and outflow have recently been observed in the near-infrared (NIR) by Fuller, Zijlstra, & Williams (2001) and in the mid-infrared (MIR) at low spatial resolution ($\sim$1$\arcsec$) by De Buizer et al. (2005). The nature and structure of the thermal infrared emission seen from this outflow in these two studies has prompted this high spatial resolution ($\sim$0.35$\arcsec$ at 11.7$\micron$) MIR follow-up study. In this Letter, I will explore the characteristics of the MIR emission in G35.20-0.74 and relationships between the detailed morphologies of the infrared emission, masers, and radio continuum emission. Observations ============ Observations of G35.20-0.74 were carried out at Gemini South on the night of 2005 July 12 through patchy clouds. Imaging was performed with the Thermal-Region Camera and Spectrograph (T-ReCS) using the Si-5 ($\lambda$$_c$=11.7$\micron$, $\Delta\lambda$=1.1$\micron$) filter and Qa filter $\lambda$$_c$=18.3$\micron$, $\Delta\lambda$=1.6$\micron$). T-ReCS utilizes a Raytheon 320$\times$240 pixel Si:As IBC array which is optimized for use in the 7–26 $\mu$m wavelength range. The pixel scale is 0.089$\arcsec$/pixel, yielding a field of view of 28$\farcs$8$\times$21$\farcs$6. Co-added frames were saved every 10 sec, and the telescope was nodded every 30 sec. The co-added frames were examined individually during the data reduction process and those plagued by clouds (i.e., showing high and/or variable background or decreased source flux) were discarded. In both filters the northern and southern radio jet regions were imaged, yielding final mosaicked images with effective field of views of approximately 28$\arcsec$$\times$35$\arcsec$. The 8$\arcsec$ overlapping region was used to register the northern and sourthern images of each mosaic. However, the individual images were cropped before they were mosaicked so the final mosaic has the same signal to noise across the whole image. The final effective exposure times for the mosaics presented here are therefore 140 sec at 11.7$\micron$ and 180 sec at 18.3$\micron$. Flux calibration was achieved by observing the MIR standard star HD 169916 ($\lambda$ Sgr) at similar airmasses to the G35.20-0.74 observations. The assumed flux densities for HD 169916 were taken to be 22.29 Jy at 11.7$\micron$ and 9.24 Jy at 18.3$\micron$. These assumed standard star flux densities were found by convolving the spectral irradiance templates of the stars from Cohen et al. (1999) with the given T-ReCS filter transmission profile. Derived flux densities for the entire region of G35.20-0.74 are measured to be 3.06$\pm$0.09 Jy at 11.7 $\micron$ and 46.87$\pm$3.66 Jy at 18.3 $\micron$ using a square aperture of 6$\farcs$9$\times$13$\farcs$5. These flux densities are quoted with their 1-$\sigma$ total error, which is a quadrature addition of the statistical variation from the aperture photometry (due to the standard deviation of the background array noise) and the flux calibration error. The flux calibration error was found from the standard deviation of the variation of the standard star flux density in each co-added nod position. Since the science and standard star images were cleaned of any effects from clouds, the quoted errors are thought to be robust, however the errors were calculated using all available data. Comparisons to the lower angular resolution observations by De Buizer et al. (2005) show the values derived here to be consistent with those derived in that previous work. Discussion ========== The mosaicked images at 11.7 and 18.3 $\micron$ are presented in Figure 1. These images are cropped to show only the parts of the field that have MIR sources. No significant MIR emission was detected outside these cropped areas at a 3-$\sigma$ upper limit of 41 mJy$\cdot$arcsec$^{-2}$ at 11.7 $\micron$ and 283 mJy$\cdot$arcsec$^{-2}$ at 18.3 $\micron$. Sources 1, 2, and 10 are seen at 11.7 $\micron$ but not at 18.3 $\micron$, and source 3 is seen at 18.3 $\micron$, but not 11.7 $\micron$. Source 4 is marginally detected at 18.3 $\micron$. The remaining sources are detected at both wavelengths and are mostly knots of emission associated with the MIR monopolar jet of G35.20-0.74. The origin of Figure 1 is the expected location of the outflow source itself. This source is a B2.6 star (as derived from the 8.5GHz flux density of Gibb et al. 2003 and using the method described in De Buizer et al. 2005) that can be seen as a UC HII region in the radio and has been dubbed G35.2N. Relations to radio continuum and NIR emission --------------------------------------------- The MIR images were registered with respect to the NIR K and L$^\prime$ images of Fuller et al. (2001). Very accurate relative astrometry ($<$0.15$\arcsec$) was achieved because of the presence of three compact MIR sources (1, 2, and 4) that are also present in the K and/or L$^\prime$ images. The absolute astrometry of the NIR images (and, consequently, the MIR images) comes from matching up NIR point sources with their optical counterparts found in the USNO B1.0 astrometric catalog. The estimates of the 1-$\sigma$ absolute uncertainty in these coordinates are 0.3$\arcsec$ for Right Ascension and 0.1$\arcsec$ for Declination. Figure 2a shows the K emission (thick white contours) overlaid on the 11.7 $\micron$ image, and Figure 2b shows the L$^\prime$ emission (thin white contours) overlaid on the 18.3 $\micron$ image. The L$^\prime$ emission from the jet looks very similar to what is seen in the MIR. The K emission appears to be dominated more by the material in the north, further along the outflow axis, with very little emission down near the outflow source itself. The convex structure seen in the NIR breaks up into separate MIR components (knots 5 and 8), and therefore is probably not a bow-shock as implied by Fuller et al. (2001). The L$^\prime$ images of Fuller et al. (2001) show what they claim is weak NIR emission from the southern jet of G35.20-0.74 (see Figure 2b). Interestingly, this emission is extremely weak at K, bright at L$^\prime$, not detected at 11.7 $\micron$, but present at 18.3 $\micron$ (Source 3 in Figure 1). The MIR images were also registered with respect to the high-resolution 8.5 GHz radio continuum images of Gibb et al. (2003) in Figure 2a (thin white contours), and with the low-resolution 15 GHz radio continuum image of Heaton & Little (1988) in Figure 2b (thick white contours). The 1-$\sigma$ relative astrometric error between the MIR and radio continuum images is estimated to be 0.34$\arcsec$ in Right Ascension and 0.18$\arcsec$ in Declination. The MIR and NIR images and contours shown in Figure 2 have been shifted +0$\fs$023 (+0.35$\arcsec$) in Right Ascension to place G35.N on the infrared outflow axis (this is approximately the estimated 1-$\sigma$ astrometric uncertainty). In Figure 2b it can be seen that the overall extent of the northern radio lobe is comparable to the MIR emission. There is also considerable MIR emission coming from the central radio continuum emitting region near the outflow source, however there is no MIR emission from the southern radio peaks. G35.2N is also the location of one of the two millimeter peaks (Gibb et al. 2003) in this region (crosses in Figure 2b). The two northernmost radio knots lie close to, but are not exactly coincident with MIR sources 6 and 7. For these knots, the radio and the MIR may be tracing slightly different emitting regions within the knots themselves. Nature of the mid-infrared emission ----------------------------------- MIR emission from outflows has been detected previously (e.g., Noriega-Crespo 2004), however these outflows have been claimed to be dominated by shock lines of H$_2$ contained within the filters used. For the observations presented here there are no H$_2$ lines within the bandpass of either the 11.3 or 18.3 $\micron$ filters. There is a possibility that there may be some contribution to the emission at 11.7 $\micron$ because of PAH emission, however this is not a concern at 18.3 $\micron$. The steeply rising spectral slope from L$^\prime$ to 18.3 $\micron$ of the narrow, elongated infrared emission coincident with and immediately north of the position of G35.2N demonstrates that the infrared emission here is dominated by longer wavelength continuum emission. Therefore the nature of the infrared emission is concluded to be dominantly continuum dust emission from the outflow cavity walls. This cavity was created by the molecular outflow punching a hole in the dense molecular material surrounding young stellar source at the center of G35.20-0.74. The central source is mostly likely directly heating the walls of this cavity. The northern lobe of the outflow was found to be slightly blueshifted towards Earth (i.e., in CO by Gibb et al. 2003; in CI by Little, Kelly, & Murphy 1998). Given this fortuitous geometry, we can see directly into the outflow cavity itself due to the outflow clearing away the material along our line of sight. The sources further north of G35.20-0.74, namely sources 5-9, are expected to be knots of dust either in the outflow itself of clumps of pre-existing material that are being impinged upon by the outflow. Source 6 lies 19200 AU from G35.2N and is still at an estimated dust color temperature of 112 K. This is based on the 11.7 and 18.3 $\micron$ flux densities of this source and neglecting the possible effects of silicate absorption (see De Buizer et al. 2005 for method and limitations). What is heating the dust this far out? Smaller dust grains can be heated out to farther distances than large dust grains. The typical size range of interstellar grains is believed to be 0.003-10 $\micron$, and typical grain compositions include smooth astronomical silicates, graphite, and silicon carbide (Laor & Draine 1993, Draine & Lee 1984). For the following I will use the equation for dust temperature given by Sellgren, Werner, & Dinerstein (1983) and the ultraviolet and infrared emissivities of Draine & Lee (1984). Assuming the dust is made up of smooth astronomical silicates, dust with a lower limit size of 0.003 $\micron$ can be heated to 112K only out to $\sim$16000 AU by a B2.6 star. If the dust is made of graphite, one can heat out to the distance of source 6 with grains having a typical size of 0.005 $\micron$, still near the lower limit size. However, if silicon carbide is the assumed composition of the dust, then one can get heating out much farther than source 6, namely $\sim$52000 AU at the 0.003 $\micron$ lower limit size. There is a possibility of some contribution from shock heating, although Fuller et al. 2001 claim no detection of shock-excited H$_2$ in the region. Beaming of the MIR emission along the outflow axis, rather than the isotropic emission assumed in the above calculations, could also help in heating grains farther out. Interestingly, the MIR luminosity derived from the dust color temperature gives an estimated value of 1.6$\times$10$^3$ L$_{\sun}$. Assuming the MIR luminosity is all the luminosity of the source (an obvious underestimate) and calculating a spectral type from that bolometric luminosity using the method of De Buizer et al. (2005) gives a value of $\sim$B3, consistent with the radio derived spectral type. In summary, all of the dust, even as far out as source 6, can indeed be heated directly by G35.2N, depending on dust composition and size (as well as beaming), though we cannot rule out contributions from other possible heating mechanisms. As discussed in the previous section, MIR source 3, coincident with NIR emission from the presumed infrared southern counterjet, does not have a smoothly increasing spectral slope typical of dust continuum emission, but instead is only present at L$^\prime$ and 18.3 $\micron$. This implies that the emission in this southern source is dominated by line emission of some kind. The usual suspects are: 1) H$_2$ emission from shocks, however, Fuller et al. (2001) claim no detection of H$_2$ in the region; 2) PAH emission from the photo-dissociation region of the outflow interface with the molecular cloud, however the L$^\prime$ and 18.3 $\micron$ filters do not encompass any PAH features; 3) \[FeII\] emission from shocks. This last one may be a possibility since \[FeII\] lines are found in both filter bandpasses, but spectroscopy will be needed to know for sure. Masers ------ The positions of the OH masers from Hutawarakorn & Cohen (1999) and water masers from Forster & Caswell (1989) are shown in Figure 2c (asterisks and X symbols, respectively). The positions of these maser sources with respect to the radio continuum emission were taken from Gibb et al. (2003). There are also methanol masers plotted in Figure 2c (large crosses), that were found by A.G. Gibb (private communication), and have positions known to 0.2$\arcsec$ with respect to the radio continuum. Interestingly, the combined distribution of the masers is in a V-shape, with its apex near the G35.2N source itself. From their positions with respect to the near and MIR emission, the masers appear to be tracing the outflow cavity walls. In the case of the OH and water masers, they may be excited to emit here by the slower oblique shocks created by the outflow on the cavity walls. Methanol masers on the other hand are believed to be excited by MIR radiation (i.e., Sobolev & Deguchi 1994; Sobolev, Cragg, & Godfrey 1997). The copious amount of MIR emission associated with this outflow source implies that at least the pumping mechanism needed for the generation of methanol masers exists on such outflow cavity walls. Given the narrow opening angle of this outflow cavity as seen in the MIR and the fact that methanol masers are often found in linear distributions (e.g. Norris et al 1993), it is possible that methanol masers in general are associated with the cavities walls of outflowing young massive stars. Previous observations by De Buizer (2003) found the majority of linear methanol maser distributions in that sample were found at position angles similar to H$_2$ emission expected to trace outflows from the host YSOs. Furthermore, other sources like NGC 7538 IRS 1 (De Buizer & Minier 2005) have been observed in the MIR to have outflow-like MIR emission at the same position angle as other outflow indicators (CO in the case of NGC 7538 IRS1), as well as having a linear methanol maser distribution at a similar position angle as well. All of this may point to a scenario where the linearly distributed methanol masers may be associated more generally with outflows of massive YSOs, and not with circumstellar disks as was previously thought (i.e. Norris et al 1993). Implications for mid-infrared bright YSOs ----------------------------------------- There is a tendency to think that if MIR emission is detected in a young stellar object (especially if it appears elongated in its morphology) that it is emission from a circumstellar disk (i.e. Lada & Lada 2003). Recent observations and theories (e.g., Miroshnichenko et al. 1999) have shown that, at least for the more massive YSOs, the dominant source of MIR emission comes from an accretion envelope. These accretion envelopes may be elongated as well (De Buizer, Osorio, & Calvet 2005). The observations presented here show that we must take in to account another source of elongated MIR emission. On a 3-4m telescope (De Buizer et al. 2005), or with the resolution of the Spitzer Space Telescope, the MIR emission from this G35.20-0.74 could be misinterpreted as a circumstellar disk. Only recently with the increase of MIR instruments on 8-10m telescopes have we begun to resolve source like G35.20-0.74 with such detail as to discern their true MIR emission as being outflow-related (e.g., NGC7538 IRS 1; De Buizer & Minier 2005). Therefore one must be cautious when trying to infer disk properties from unresolved or partially resolved sources since the outflow cavities of YSOs can be the dominant source of MIR emission. On the other hand, the mere presence of such outflow cavities with small opening angles implies the presence of collimating accretion disks at the bases of the outflows. So while MIR emission may not always be a direct tracer of circumstellar disks and their properties, the presence of such emission in the near circumstellar environment of YSOs may still indirectly indicate the presence of a disk. Based on observations obtained at the Gemini Observatory, which is operated by AURA, Inc., under a cooperative agreement with the NSF on behalf of the Gemini partnership: the NSF (U.S.), the PPARC (U.K.), the NRCC (Canada), CONICYT (Chile), the ARC (Australia), CNPq (Brazil) and CONICET (Argentina). I would also like to thank James Radomski for his illuminating discussions which helped in the writing of this article. Beuther, H., Schilke, P., Sridharan, T. K., Menten, K. M., Walmsley, C. M., & Wyrowski, F. 2002, , 383, 892 Cohen, M., Walker, R. G., Carter, B., Hammersley, P., Kidger, M., & Noguchi, K. 1999, , 117, 1864 De Buizer, J. M., Radomski, J. T., Telesco, C. M., & Pi[ñ]{}a, R. K. 2005, , 156, 179 De Buizer, J. M., & Minier, V. 2005, , 628, L151 De Buizer, J. M. 2003, , 341, 277 Forster, J. R., & Caswell, J. L. 1989, , 213, 339 Fuller, G. A., Zijlstra, A. A., & Williams, S. J. 2001, , 555, L125 Gibb, A. G., Hoare, M. G., Little, L. T., & Wright, M. C. H. 2003, , 339, 1011 Heaton, B. D., & Little, L. T. 1988, , 195, 193 Hutawarakorn, B., & Cohen, R. J. 1999, , 303, 845 Lada, C. J., & Lada, E. A. 2003, , 41, 57 Little, L. T., Kelly, M. L., & Murphy, B. T. 1998, , 294, 105 Miroshnichenko, A., Ivezi[ć]{} , [Ž]{}., Vinkovi[ć]{} , D., & Elitzur, M. 1999, , 520, L115 Noriega-Crespo, A., et al. 2004, , 154, 352 Norris, R. P., Whiteoak, J. B., Caswell, J. L., Wieringa, M. H., & Gough, R. G. 1993, , 412, 222 Sellgren, K., Werner, M. W., & Dinerstein, H. L. 1983, , 271, L13 Shepherd, D. S., & Churchwell, E. 1996, , 457, 267 Sobolev, A. M., & Deguchi, S. 1994, , 291, 569 Sobolev, A. M., Cragg, D. M., & Godfrey, P. D. 1997, , 324, 211 Zhang, Q., Hunter T. R., Brand, J., Sridharan, T. K., Molinari, S., Kramer, M. A., & Cesaroni, R. 2001, , 552, 167	{ "pile_set_name": "ArXiv" }
[ Irreducibility of the Cayley-Menger determinant,]{}\ \ \ [Mowaffaq Hajja$^{1}$, Mostafa Hayajneh$^2$, Bach Nguyen$^3$, and Shadi Shaqaqha$^4$]{}\ \ (1): P. O. Box 1, Philadelphia University, 19392, Amman, Jordan\ [email protected]\ \ (2), (4): Yarmouk University, Irbid, Jordan\ [email protected], [email protected]\ \ (3): Louisiana State University, Baton Rouge, LA, USA\ [email protected] > [Abstract.]{} [If $S$ is a given regular $n$-simplex, $n \ge 2$, of edge length $a$, then the distances $a_1$, $\cdots$, $a_{n+1}$ of an arbitrary point in its affine hull to its vertices are related by the fairly known elegant relation $\phi_{n+1} (a,a_1,\cdots,a_{n+1})=0$, where $$\phi = \phi_t (x, x_1,\cdots,x_{n+1}) = \left( x^2+x_1^2+\cdots+x_{n+1}^2\right)^2 > - t\left( x^4+x_1^4+\cdots+x_{n+1}^4\right).$$ The natural question whether this is essentially the only relation is answered positively by M. Hajja, M. Hayajneh, B. Nguyen, and Sh. Shaqaqha in a recently submitted paper entitled [Distances from the vertices of a regular simplex]{}. In that paper, the authors made use of the irreducibility of the polynomial $\phi $ in the case when $n \ge 2$, $t=n+1$, $x= a \ne 0$, and $k = {\mathbb{R}}$, but supplied no proof, promising to do so in another paper that is turning out to be this one. It is thus the main aim of this paper to establish that irreducibility. In fact, we treat the irreducibility of $\phi$ without restrictions on $t$, $x$, $a$, and $k$. As a by-product, we obtain new proofs of results pertaining to the irreducibility of the general Cayley-Menger determinant that are more general than those established by C. D’Andrea and M. Sombra in [Sib. J. Math. [46]{}, 71–76.]{} ]{} [Keywords:]{} Cayley-Menger determinant; circumscriptible simplex; discriminant; homogeneous polynomial; irreducible polynomial; isodynamic simplex; isogonic simplex; orthocentric simplex; Pompeiu’s theorem; pre-kites; quadratic polynomial; regular simplex; symmetric polynomial; tetra-isogonic simplex; volume of a simplex Introduction {#111} ============ Let $S=[A_1, \cdots, A_{n+1}]$, $n \ge 2$, be a regular $n$-simplex of edge length $a$, and let $B$ be an arbitrary point in its affine hull. It is fairly well known that the distances $a_1,\cdots,a_{n+1}$ from $B$ to the vertices of $S$ satisfy the elegant relation $$\begin{aligned} \label{f-a} \left(a^2+a_1^2+\cdots+a_{n+1}^2\right)^2 &=& (n+1)\left(a^4+a_1^4+\cdots+a_{n+1}^4\right);\end{aligned}$$ see [@Bentin-1]. The natural question whether this is (essentially) the only relation was posed and answered in the affirmative in [@Shadi]. However, the proof rests heavily on a result whose proof was not included therein, but was postponed for another paper, namely this one. This result states that the polynomial obtained from (\[f-a\]) by thinking of $a_1,\cdots,a_{n+1}$ as indeterminates is irreducible over ${\mathbb{R}}$ when $a \ne 0$ and $n \ge 2$. The main goal of this paper is to prove this, and to actually give a complete treatment of the irreducibility of the more general polynomial $$\begin{aligned} \label{g} g &=& \left(a^2+x_1^2+\cdots+x_{n+1}^2\right)^2 - t\left(a^4+x_1^4+\cdots+x_{n+1}^4\right) \in k[x_1,\cdots,x_{n+1}]\end{aligned}$$ for any field $k$, any $t \in k$, and any $a \in k$ (including the case $a=0$). It turns out that if ${\mbox{char~}}k \ne 2$, then $g$ is irreducible except in the two cases $(n,a,t)=(2,0,2)$ and $(n,a,t)=(2,0,3)$. When $(n,a,t)=(2,0,2)$, $g$ is the Heron polynomial with the well known factorization given in (\[uuu\]). When $(n,a,t)=(2,0,3)$, $g$ is reducible if and only if $k$ contains a primitive third root of unity, in which case $g$ factors as in (\[vvvv\]). If ${\mbox{char~}}k = 2$, then $g$ reduces to the polynomial $(1-t)(a+x_1+\cdots+x_{n+1})^4.$ The cases when $n=0$ and $n=1$ are ignored because they have no geometric significance. Also, the case $n=1$ turned out to be rather lengthy and complex, but it gives rise to interesting number theoretic aspects. We do not include these cases in this paper. The polynomial obtained from $g$ by replacing $n+1$ with $n$ and $t$ with $n$ is essentially the Cayley-Menger determinant of a special $n$-simplex, named a prekite, that was introduced and studied in [@prekites], and calculated in Theorem 4.1 there. If we put $a=0$ in $g$, then the instances $t=0$, $t=n$, $t=n-1$, and $t=n-2$ are very closely related to the Cayley-Menger determinants, with respect to certain parametrizations, of orthocentric, isodynamic, circumscriptible, and tetra-isogonic $n$-simplices; see Theorem 5.2 of [@impurity]. Thus the results of this paper would help give complete factorizations of the Cayley-Menger determinants of the aforementioned five special families of $n$-simplices. This raises the question whether the Cayley-Menger determinant $M$ of a general $n$-simplex is irreducible. It is proved in Theorem \[CM\] that the answer is affirmative except in the single case $n=2$, in which case $M$ is the Heron polynomial given in (\[uuu\]). This result has been obtained earlier in [@Siberian] when $k$ is ${\mathbb{R}}$ or ${\mathbb{C}}$. We should note, however, that the result in [@Siberian] does not imply the results in this paper, and does not give any information about the irreducibility of the Cayley-Menger determinant of any of the five special families mentioned earlier. This is because the Cayley-Menger determinants of these special families are obtained from the general Cayley-Menger determinant $M$ given in (\[CMCM\]) by replacing each $x_{ij}^2$, $1 \le i < j \le n+1$, by quantities like $x_i+x_j$, $x_i x_j$, $(x_i+x_j)^2$, $x_i^2+x_ix_j+x_j^2$; see [@impurity Theorem 5.1, p. 54]. Such substitutions obviously do not preserve irreducibility. Furthermore, our treatment is not confined to the cases when $k = {\mathbb{R}}$ or ${\mathbb{C}}$. The paper is organized as follows. Section \[222\] contains preliminary facts that we shall need, and freely use, in the sequel. These are quite few, and come from the elementary theories of symmetric and of homogeneous polynomials. Section \[333\] contains proofs of the main results, namely Theorems \[DDD\], \[CCDD\], and \[CM\]. These proofs make use of Lemmas \[AAA\], \[BBB\], and \[CCC\]. Preliminaries {#222} ============= In this section, we put together definitions and simple facts pertaining to symmetric and homogeneous polynomials that we shall freely use throughout the paper. For ease of reference, we also include a simple theorem about the factorization of quadratics. Let $A$ be any commutative ring with identity, and let $B=A[x_1,\cdots,x_n]$ be the polynomial ring in the $n$ indeterminates $x_1,\cdots,x_n$. Then the symmetric group ${\mathcal{S}}_n$ acts as permutations on the indices of $x_1,\cdots,x_n$, and hence as $A$-automorphisms on $B$. A polynomial in $B$ is called [symmetric]{} if it is invariant under the action of (every element of) ${\mathcal{S}}_n$. It is well known that the ring $T=A[x_1,\cdots,x_n]^{{\mathcal{S}}_n}$ of symmetric polynomials is given by $$\begin{aligned} \label{symsym} T=A[x_1,\cdots,x_n]^{{\mathcal{S}}_n} &=&A[e_1,\cdots,e_n],\end{aligned}$$ where $e_1, \cdots, e_n$ are the so-called [elementary symmetric polynomials]{} defined by $$e_j = \sum_{1 \le i_1 < i_2 < \cdots < i_j \le n} x_{i_1} x_{i_2} \cdots x_{i_j};$$ see [@Lang Theorem 6.1, p. 191]. A non-zero polynomial $\Phi$ in $B$ is said to be [homogeneous]{} if all of its terms are of the same (total) degree $d$. More precisely, if $$\Phi (\lambda x_1, \cdots, \lambda x_n) = \lambda^d \Phi (x_1, \cdots, x_n)$$ for any variable $\lambda$. It is clear that every non-zero polynomial $\Phi \in B$ of (total) degree $n \ge 0$ can be written uniquely in the form $$\Phi = h^{(0)} + h^{(1)} + \cdots + h^{(n)},$$ where $h^{(j)}$ is either 0 or homogeneous of degree $j$, and $h^{(n)}$ is not zero. The last term $h^{(n)}$ is called the [leading homogeneous component]{} of $h$, and will be denoted by $\Phi^$. Notice that $$\begin{aligned} \deg \Phi^ = \deg \Phi. \label{degh}\end{aligned}$$ If $A$ is an integral domain, then it is easy to see that $$\begin{aligned} (\Phi \Psi)^* = \Phi^* \Psi^. \label{fgh}\end{aligned}$$ Then it follows from (\[degh\]) and (\[fgh\]) that if $f$ is reducible, then $f^$ is reducible. It also follows that $$\begin{aligned} \label{hhh} \mbox{every factor of a homogeneous polynomial is homogeneous};\end{aligned}$$ see [@Walker Theorem 10.5, p. 28]. To see that this is not true if $A$ was not an integral domain, consider the factorization $x^2 = (x-2)(x+2)$ in ${\mathbb{Z}}_4[x]$. We end this section by proving a simple property of quadratic polynomials that we shall use later. \[Q\] Let $A$ be an integral domain with ${\mbox{char~}}A \ne 2$, and let $A[x]$ be the polynomial ring over $A$ in the indeterminate $x$. Let $H =ax^2+bx+c \in A[x]$, with $a \ne 0$, and let $\Delta = b^2 - 4 ac$ be its discriminant. Then - If $H$ is reducible, then $\Delta$ is a square (in $A$). - If $\Delta$ is a square, and if $2a$ is a unit, then $H$ is reducible. - If $H$ is a square, then $\Delta = 0$. - If $\Delta = 0$, and $a$ is a square and $2a$ is a unit, then $H$ is a square. [Proof.]{} To prove (i), suppose that $H$ is reducible. Then $H=(U x + V)(ux+v)$ for some $U, V, u, v \in A$. Multiplying out and equating coefficients, we obtain $Uu=a$, $Uv+Vu=b$, and $Vv=c$. Therefore $\Delta = b^2 - 4ac = (Uv+Vu)^2 - 4 (Uu)(Vv) =(Uv-Vu)^2$, which is a square, as desired. To prove (ii), suppose that $\Delta = b^2 - 4ac$ is a square, say $\Delta = \delta^2$, and that $a$ is a unit. Letting $\alpha = \left(-b+\delta\right)\left(2a\right)^{-1}$ and $\beta = \left(-b-\delta\right)\left(2a\right)^{-1}$, we can easily see that $H=a(x-\alpha)(x-\beta),$ and hence is reducible, as desired. To prove (iii), suppose that $H$ is a square, say $H=(ux+v)^2$, where $u, v \in A$. Then $a = u^2$, $b=2uv$, and $c=v^2$. Thus $b^2-4ac=0$. To prove (iv), suppose that $2a$ is a unit and that $a$ is a square, say $a=u^2$, and $b^2-4ac=0$. Then it is easy to check that $H=\left(ux+ b (2u)^{-1} \right)^2,$ a square. $\Box$ [ The assumption that $2a$ is a unit in (ii) and (iv) above cannot be replaced by the weaker assumption that $a$ is a unit. For (ii), consider the example $A = {\mathbb{Z}}[\sqrt{5}]$ and $H=x^2+\sqrt{5} x + 1$. Then $a=\Delta = 1$, but $H$ is irreducible. In fact, if $H=(x+s)(x+t)$, then $st=1$, $s+t=\sqrt{5}$, and $(s-t)^2 = (s+t)^2-4st=1,$ and hence $s-t = \pm 1$. In view of the fact that $s+t=\sqrt{5}$, we obtain $2s=\sqrt{5}\pm 1$, contradicting the fact that $s \in A$. For (iv), let $D = \{ f(x) = c_0+c_1 x+\cdots +c_n x^n\in {\mathbb{Z}}[x] : c_1 \mbox{~is even}\},$ and let $H = x^2 + 2x + 1$. Then $\Delta = 0$, and $H$ is not a square in $D$.]{} The main results {#333} ================ In this section, we establish, in Theorems \[CCDD\], \[DDD\], and \[CM\], the main irreducibility theorems of this paper. Lemmas \[AAA\], \[BBB\], and \[CCC\] are needed in the proofs. Lemma \[CBA\] disposes of the case when $t=0$ in (\[g\]). \[CBA\] Let $k$ be a field, and let $g=c_0 + c_1x_1^2+\cdots + c_n x_n^2 \in k[x_1,\cdots, x_n]$, where $n \ge 1$ and $c_1,\cdots,c_n \in k \setminus \{0\}$. For $0 \le i \le n$, let $t_i = c_i/c_n$, and let $h=t_0 + t_1x_1^2+\cdots + t_n x_n^2.$ Then $h$ (and hence $g$) is reducible if and only if $-t_j$ is a square for all $j$, $0 \le j \le n-1$ and if - either ${\mbox{char~}}k =2$, in which case $$\begin{aligned} h&=&\left(\sqrt{t_0} + \sqrt{t_1} x_1+ \cdots + \sqrt{t_{n-1}} x_{n-1} + x_{n} \right)^2,\end{aligned}$$ - or ${\mbox{char~}}k \ne 2$ and $n=1$, in which case, $$\begin{aligned} h&=&\left(x_1 - \sqrt{-t_0}\right)\left(x_1 + \sqrt{-t_0}\right), \end{aligned}$$ - or ${\mbox{char~}}k \ne 2$, $n=2$, and $t_0 = 0$, in which case $$\begin{aligned} h&=&\left(x_2 - \sqrt{-t_1}x_1 \right)\left(x_2 + \sqrt{-t_1} x_1\right). \end{aligned}$$ [Proof.]{} Suppose that $h$ is reducible. Then any factorization of $h$ must be of the form $$\begin{aligned} \label{Must-1} h &=&(a_0 + a_1 x_1+ \cdots+a_nx_n)(b_0 + b_1 x_1+ \cdots+b_nx_n).\end{aligned}$$ We may also assume that $a_n=b_n=1,$ since $a_n b_n = 1$. Comparing coefficients of $x_nx_i$, $1 \le i \le n-1$, and coefficients of $x_n$, we see that $b_i=-a_i$ for $0 \le i \le n-1$. Therefore (\[Must-1\]) can be rewritten as $$\begin{aligned} \label{Must-2} h &=&(a_0 + a_1 x_1+ \cdots+a_{n-1} x_{n-1}+ x_n)(-a_0 -a_1 x_1- \cdots -a_{n-1} x_{n-1} + x_n).\end{aligned}$$ Comparing coefficients of $x_i^2$, $1 \le i \le n-1$, and the constant terms, we see that $-t_i = a_i^2$ for $0 \le i \le n-1$. Thus $- t_i$ is a square for all $i$, $0 \le i \le n-1.$ Of course, $t_n = 1$. If char ($k$) = 2, then $h$ factors into $$\begin{aligned} h &=&(a_0 + a_1 x_1+ \cdots+a_{n-1} x_{n-1}+ x_n)^2.\end{aligned}$$ If char ($k$) $\ne 2$, and $n > 2$, then comparing the coefficients of $x_1x_2$ (in (\[Must-2\])), we obtain $-2 a_1 a_2 = 0$, contradicting the assumptions that $a_1^2 = -t_1 \ne 0$, $a_2^2 = -t_2 \ne 0$, and $2 \ne 0$. If (char ($k$) $\ne 2$, and) $n = 1$, then $h$ factors into $h = (a_0 + x_1)(-a_0 + x_1)$. If char ($k$) $\ne 2$, and $n = 2$, then comparing the coefficients of $x_1$, we obtain $-2 a_0 a_1 = 0$, and hence $a_0 = 0$, since $a_1^2 = -t_1 \ne 0$ and $2 \ne 0$. Therefore $h$ factors into $h = (a_1 x_1 + x_2)(-a_1 x_1 + x_2).$ This completes the proof. $\Box$ It follows, for example, that the polynomial $x_1^2 + \cdots + x_n^2$ is reducible over the field ${\mathbb{C}}$ of complex numbers if and only if $n=1$ or $n=2$. In particular, $x^2+y^2+z^2$ is irreducible over ${\mathbb{C}}$, a fact that appears as a problem in [@AMATYC]. In view of the lemma above, we may exclude the case $t=0$ in (\[g\]). Also, it is obvious that if ${\mbox{char~}}k = 2$, then $g = (1-t) (a+x_1+\cdots+x_n)^4$. That is why the lemmas and theorems below assume that $t\ne 0$ and ${\mbox{char~}}k \ne 2$. \[AAA\] Let $k$ be a field with ${\mbox{char~}}k \ne 2$, and let $$\begin{aligned} f &=& (x_1^2+\cdots+x_n^2)^2-t(x_1^4+ \cdots +x_n^4) \in k[x_1,\cdots,x_n], \end{aligned}$$ where $t \in k$ and $t \ne 0$, and where $n \ge 3$. Then $f$ has a factor that is symmetric in two of the variables if and only if $n=3$ and $t=2$. In this case, $$\begin{aligned} f &=& (x_1^2+x_2^2+x_3^2)^2-2(x_1^4+ x_2^4+ x_3^4)\nonumber \\&=& (x_1+x_2+x_3)(-x_1+x_2+x_3)(x_1-x_2+x_3)(x_1+x_2-x_3). \label{uuu} \end{aligned}$$ [Proof.]{} We shall freely use the fact that factors of a homogeneous polynomial are homogeneous; see (\[hhh\]). Let $$\begin{aligned} &&y = x_{n-1}, ~ z = x_n,~ u = y+z,~v=yz, \label{yzuv}\\ &&S_2=x_1^2+\cdots+x_{n-2}^2,~S_4=x_1^4+\cdots+x_{n-2}^4,\label{S2S4}\\ &&R=k[x_1,\cdots,x_n] = k[x_1,\cdots,x_{n-2},y,z],~ R_0=k[x_1,\cdots,x_{n-2},u,v]. \label{RR}\end{aligned}$$ Clearly, $R$ and $R_0$ are polynomial rings over $k$ in the $n$ respective variables, and $R_0$ is the subring of $R$ consisting of the $(y,z)$-symmetric polynomials, i.e., polynomials that are symmetric in the variables $y$ and $z$; see (\[symsym\]). Using the identities $$\begin{aligned} y^2+z^2&=&u^2-2v,\\ y^4+z^4&=&(y^2+z^2)^2-2v^2~=~(u^2-2v)^2-2v^2, $$ and the definition $$f=(S_2+y^2+z^2)^2-t(S_4+y^4+z^4),$$ we obtain (after few lines of computations, or by Maple) $$\begin{aligned} \label{fSuv} f&=&(4-2t)v^2 -4 ((1-t)u^2+S_2)v +S_2^2 + (1-t) u^4 + 2S_2 u^2 -t S_4.\end{aligned}$$ Letting $F \in R_0$ denote the right hand side of (\[fSuv\]), i.e., $$\begin{aligned} F&=& (4-2t)v^2 -4 ((1-t)u^2+S_2)v +S_2^2 + (1-t) u^4 + 2S_2 u^2 -t S_4, \label{F}\end{aligned}$$ we see that $f$ has a $(y,z)$-symmetric factor if and only if $F$ is reducible in $R_0$. From now on, we will be working in $R_0$. Suppose that $F$ is reducible. If $t=2$, then $F$ simplifies into $$\begin{aligned} F_0&=& -4 (-u^2+S_2)v +S_2^2 - u^4 + 2S_2 u^2 -2 S_4 \\ &=& 4 (u^2-S_2)v - (u^2-S_2)^2 + 2S_2^2 -2 S_4.\end{aligned}$$ Since $F_0$ is linear in $v$, it follows that every factor of $F_0$ is a factor of $(u^2-S_2)$ and of $-(u^2-S_2)^2+2S_2^2 -2 S_4$. But any factor of $u^2-S_2$ must contain $u$, while $2S_2^2 -2 S_4$ does not. Therefore $F_0$ is reducible if and only if $2S_2^2 -2 S_4 = 0$, i.e., if and only if $n=3$, in which case $S_4=x_1^4=(x_1^2)^2=S_2^2$. In this case $F_0$ is as given in (\[uuu\]). If $t \ne 2$, then $F$ is a quadratic in $v$ with coefficients in $k[x_1,\cdots,x_{n-2},u]$. Also, its leading coefficient $(4-2t)$ is a unit. By (i) and (ii) of Lemma \[Q\], it is reducible if and only if its discriminant $\Delta$ in $v$ is a square in $k[x_1,\cdots,x_{n-2},u]$. Using Maple, we find that $$\begin{aligned} \label{D} \Delta &=& 8t((t-1)u^4-2S_2u^2+(2-t)S_4+S_2^2).\end{aligned}$$ If $t=1$, then $\Delta = 8(-2S_2 u^2 + S_4 + S_2^2)$. By (iii) of Lemma \[Q\], this cannot be a square, since its discriminant in $u$ is $8^3S_2(S_4 + S_2^2) \ne 0$. Thus suppose that $$t \ne 1.$$ Then for $\Delta$ to be a square, we must have $$\begin{aligned} \label{DD} \Delta &=& 8t((t-1)u^4-2S_2u^2+(2-t)S_4+S_2^2) ~=~ (\alpha u^2 + \beta u + \gamma)^2, \end{aligned}$$ where $\alpha$, $\beta$, and $\gamma$ belong to $k[x_1,\cdots,x_{n-2}]$. Equating the coefficients of $u^4$ and $u^3$ in (\[DD\]), we see that $\alpha^2 = 8t (t-1) \ne 0$ and $2 \alpha \beta = 0$. Therefore $\beta =0$, and we have $$\begin{aligned} \label{DDDD} \Delta &=& 8t((t-1)u^4-2S_2u^2+(2-t)S_4+S_2^2) ~=~ (\alpha u^2 + \gamma)^2. \end{aligned}$$ Equating coefficients in (\[DDDD\]), we find that $$\alpha^2 = 8t(t-1),~~\gamma^2 = 8t ((2-t)S_4+S_2^2),~~2 \alpha \gamma = 8t (-2S_2).$$ It follows that $(t-1) ((2-t)S_4+S_2^2) = (-S_2)^2.$ This simplifies into $(2-t)((t-1)S_4 - S_2^2) = 0.$ Since $t \ne 2$, it follows that $S_2^2$ and $S_4$ are linearly dependent over $k$. This happens if and only if $n=3$, in which case $S_4=x_1^4=(x_1^2)^2=S_2^2$, and $t-1=1$. This is the case treated earlier. Thus $\Delta$ is not a square, a contradiction. This completes the proof. $\Box$ \[BBB\] Let $k$ be a field with ${\mbox{char~}}k \ne 2$, and let $$\begin{aligned} f &=& (x_1^2+\cdots+x_n^2)^2-t(x_1^4+ \cdots +x_n^4) \in k[x_1,\cdots,x_n], \end{aligned}$$ where $t \in k$ and $t \ne 0$, and where $n \ge 3$. Then $f$ has a linear factor if and only if $n=3$ and $t=2$. In this case, $f$ factors as in (\[uuu\]). [Proof.]{} If $(n,t)=(3,2)$, then $f$ factors as in (\[uuu\]), and we are done. So we assume that $(n,t)\ne (3,2)$, that $f$ has a linear factor, say $g=a_1x_1+\cdots+a_nx_n$, and we seek a contradiction. Clearly we may assume that $a_1 \ne 0$. By Lemma \[AAA\], $g$ is not symmetric in the variables $x_2$ and $x_3$. Therefore $a_2 \ne a_3$. If $g_1$ is the polynomial obtained from $g$ by interchanging $x_2$ and $x_3$, then $g_1$ is not an associate of $g$, and it is also a factor of $f$. Therefore $gg_1$ is a $(x_2,x_3)$-symmetric factor of $f$ (of degree 2). This contradicts Lemma \[AAA\]. $\Box$ \[CCC\] Let $k$ be a field with ${\mbox{char~}}k \ne 2$, and let $$\begin{aligned} f &=& (x_1^2+x_2^2+x_3^2)^2-t(x_1^4+x_2^4+x_3^4) \in k[x_1,x_2,x_3], \end{aligned}$$ where $t \in k$ and $t \ne 0$. If $t=2$, then $f$ factors as in (\[uuu\]). If $t \ne 2$, then $f$ is reducible if and only if $t=3$ and $k$ contains a primitive third root $\omega$ of 1. In this case, $$\begin{aligned} f&=&(x_1^2+x_2^2+x_3^2)^2 - 3(x_1^4+x_2^4+x_3^4) \nonumber \\ &=& (-2)(x_1^2+\omega x_2^2 + \omega^2 x_3^2)(x_1^2+\omega^2 x_2^2 + \omega x_3^2). \label{vvvv}\end{aligned}$$ Also, the factors in the right hand sides of (\[vvvv\]) are irreducible. [Proof.]{} There is nothing to prove in the case when $t=2$, since this is covered in Lemma \[BBB\]. Thus we assume that $t \ne 2$. Let $x_1, x_2, x_3$ be renamed as $x, y, z$, and suppose that $f$ is reducible and that $g$ is an irreducible factor of $f$. Since $t \ne 2$, Lemma \[BBB\] implies that $g$ is not linear. Thus $g$ is an irreducible quadratic. Thus $$\begin{aligned} g&=&ax^2+by^2+cz^2+\alpha yz + \beta zx + \gamma xy,\end{aligned}$$ where $a, b, c, \alpha, \beta, \gamma$ are in $k$. Also, $$\begin{aligned} f&=& (x^2+y^2+z^2)^2 - t(x^4+y^4+z^4)\\ &=& (1-t)(x^4+y^4+z^4)+2(x^2y^2+y^2z^2+z^2x^2).\end{aligned}$$ Letting $s$ be the permutation $s = (x \mapsto y \mapsto z \mapsto x)$, we see that $f$ is divisible by $g$, $s(g)$, and $s^2 (g)$. Since $\deg (g s(g) s^2(g)) = 6 > \deg f$, and since $g, s(g), s^2(g)$ are irreducible, it follows that two (and hence all) of the polynomials $g$, $s(g)$, and $s^2 (g)$ are associates (i.e., constant multiples of each other). Thus $g = \lambda s(g)$ for some $\lambda \in k$. Since $$\begin{aligned} s(g)&=&ay^2+bz^2+cx^2+\alpha zx + \beta xy + \gamma yz,\end{aligned}$$ it follows that $$b = \lambda a,~ c = \lambda b,~a = \lambda c,~ \beta = \lambda \alpha,~\gamma = \lambda \beta,~ \alpha = \lambda \gamma.$$ Thus $$a = \lambda^3 a,~\alpha = \lambda^3 \alpha.$$ If both $a$ and $\alpha$ are zero, then $a=b=c=\alpha = \beta = \gamma=0$, and $g=0$, a contradiction. Thus either $a \ne 0$ or $\alpha \ne 0$. In both cases $ \lambda^3 = 1$. If $\lambda = 1$, then $a=b=c$ and $\alpha = \beta = \gamma$, and hence $g$ is symmetric, contradicting Lemma \[AAA\]. Thus $\lambda \ne 1$ and $\lambda^3 = 1$. Therefore $\lambda$ is a primitive third root of 1 (and ${\mbox{char~}}k$ cannot be 3). Thus $g$ is of the form $$g = a (x^2+\lambda y^2+ \lambda^2 z^2) + \alpha (yz+\lambda zx+ \lambda^2 xy),$$ where $\lambda$ is a primitive third root of 1. Applying the permutation $y \mapsto z \mapsto y$ to $g$, we obtain another factor $$h = a (x^2+\lambda^2 y^2+ \lambda z^2) + \alpha (yz+\lambda^2 zx+ \lambda xy),$$ of $f$ that is not associate of $g$. Therefore $gh$ divides $f$, and has the same degree as $f$. Therefore $f = c gh$, where $c \in k$. Equating the coefficients of $x^3z$ and $x^2yz$ in the identity $f=cgh$, we obtain $-a\alpha c = 0$ and $2a\alpha c - \alpha^2 c = 0$. Thus $\alpha =0$ and $$g = a (x^2+\lambda y^2+ \lambda^2 z^2),~~ h = a (x^2+\lambda^2 y^2+ \lambda z^2).$$ Therefore $$\begin{aligned} gh &=& a^2 [(x^4+y^4+z^4) - (x^2y^2+y^2z^2+z^2x^2)].\end{aligned}$$ But $$\begin{aligned} f &=& (1-t) (x^4+y^4+z^4) +2(x^2y^2+y^2z^2+z^2x^2).\end{aligned}$$ Therefore it follows from $f = c gh$ that $(1-t)= ca^2$ and $-a^2c = 2$. Hence $1-t= -2$ and $t=3$. Therefore $$\begin{aligned} f&=&(x^2+y^2+z^2)^2-3(x^4+y^4+z^4) \\ &=&(-2)(x^2+\lambda y^2 + \lambda^2 z^2)(x^2+\lambda^2 y^2 + \lambda z^2),\end{aligned}$$ as desired. The two factors in (\[vvvv\]) are irreducible because $f$ has no linear factors, by Lemma \[BBB\]. This completes the proof. $\Box$ \[CCDD\] Let $k$ be a field with ${\mbox{char~}}k \ne 2$, and let $a$ be a non-zero element of $k$. Let $$\begin{aligned} g &=& (a^2 + x_1^2+\cdots+x_n^2)^2-t(a^4 +x_1^4+ \cdots +x_n^4) \in k[x_1,\cdots,x_n], \end{aligned}$$ where $t \in k$ and $t \ne 0$, and where $n \ge 3$. Then $g$ is irreducible. [Proof.]{} We start with the case $n=3$, and we rename $x_1$, $x_2$, $x_3$ as $x$, $y$, $z$. Thus we are to prove that the polynomial $$g = (a^2 + x^2 + y^2 + z^2)^2 - t(a^4 + x^4 + y^4 + z^4),~~a\ne 0, ~t \ne 0, ~{\mbox{char~}}k \ne 2,$$ is irreducible. The general case will follow easily as shown later. Suppose that $g$ is reducible, and that $g = \alpha \beta$, where $\alpha$ and $\beta$ are non-constant polynomials in $k[x,y,z]$. Let $G$, $A$, and $B$ be the leading homogeneous components of $g$, $\alpha$, and $\beta$, respectively. Then $$G=(x^2+y^2+z^2)^2-t(x^4+y^4+z^4),$$ and $G = AB$, and $A$ and $B$ are non-constant. Therefore $G$ is reducible. By Lemma \[CCC\], we have the following two cases: $$\begin{aligned} G&=&(x^2+y^2+z^2)^2-2(x^4+y^4+z^4)\nonumber \\ &=&(x+y+z)(-x+y+z)(x-y+z)(x+y-z), \label{GUV}\end{aligned}$$ and hence $$\begin{aligned} g&=&(a^2+x^2+y^2+z^2)^2-2(a^4+x^4+y^4+z^4).\end{aligned}$$ Suppose that one of the factors $\alpha$ or $\beta$, say $\alpha$, is linear. Then we may assume that either $\alpha=x+y+z+c$ or $\alpha=x+y-z+c$, where $c \in k$. In the first case, we plug $(x,y,z)=(x,-x,-c)$ in $g = \alpha \beta$, and in the second case, we plug $(x,y,z)=(x,-x,c)$. In both cases, we obtain $$\begin{aligned} 0&=&(a^2+2x^2+c^2)^2-2(a^4+2x^4+c^4) \\&=& 4x^2(a^2+c^2) +(-a^4-c^4+2a^2c^2)\\ &=& 4x^2(a^2+c^2)-(a^2-c^2)^2.\end{aligned}$$ Therefore $a^2+c^2=a^2-c^2=0$, and hence $a=c=0$, contradicting the assumption that $a \ne 0$. If $\alpha$ and $\beta$ are quadratic irreducible, then $A$ and $B$ are quadratic. Also $G=AB$ by (\[fgh\]). By (\[GUV\]), we may assume that $A$ or $B$, say $A$, is $(x+y+z)(-x+y+z)$. Thus $$\begin{aligned} \alpha&=&(x+y+z)(-x+y+z)+L,\end{aligned}$$ where $L$ is a linear polynomial. By applying the permutation $\sigma : x \mapsto y \mapsto x$ and then $\tau : x \mapsto z \mapsto x$, we see that the two (irreducible) polynomials $$\begin{aligned} \alpha_1=(x+y+z)(x-y+z)+\sigma (L) \mbox{~~and~~} \alpha_2=(x+y+z)(x+y-z)+ \tau (L)\end{aligned}$$ are also factors of $g$. Since the coefficients of $x^2$ in the polynomials $\alpha_1$ and $\alpha_2$ are the same (and equal 1), and since they are not equal, it follows that $\alpha_1$ and $\alpha_2$ are not associates. By considering the coefficients of $y^2$ and $z^2$, we see that no two of the polynomials $\alpha$, $\alpha_1$ and $\alpha_2$ are associates. Therefore $\alpha \alpha_1 \alpha_2$ divides $g$, a contradiction since $\deg g = 4 < 6$. Therefore $g$ cannot be reducible. $$\begin{aligned} G&=&(x^2+y^2+z^2)^2-3(x^4+y^4+z^4)\\ &=&(-2)(x^2+\omega y^2+ \omega^2 z^2)(x^2+\omega^2 y^2+ \omega z^2),\end{aligned}$$ where $\omega \in k$ is a primitive third root of 1, and where the quadratics on the right hand side are irreducible. Therefore $A$ and $B$ are the polynomials $$u(x^2+\omega y^2+\omega^2 z^2) \mbox{~and~} v(x^2+\omega^2 y^2+\omega z^2),$$ where $$u, v \in k,~uv=-2.$$ Hence $\alpha$ and $\beta$ are the polynomials $$u(x^2+\omega y^2+\omega^2 z^2 +L) \mbox{~and~} v(x^2+\omega^2 y^2+\omega z^2 +K),$$ where $L$ and $K$ are linear polynomials. Also, $$\begin{aligned} g&=&(a^2+x^2+y^2+z^2)^2-3(a^4+x^4+y^4+z^4) \nonumber\\ &=&(-2)(x^2+\omega y^2+ \omega^2 z^2+L)(x^2+\omega^2 y^2+ \omega z^2+K). \label{Kh1}\end{aligned}$$ Let $L_0$ and $K_0$ be the constant terms of $L$ and $K$, respectively. Plugging $x=y=z=0$ in (\[Kh1\]), we obtain $-2 L_0 K_0 = -2 a^4$, and hence $$\begin{aligned} L_0 K_0 &\ne& 0. \label{Kh2}\end{aligned}$$ Applying the permutation $\sigma : x \mapsto y \mapsto x$, we obtain $$\begin{aligned} g&=&(a^2+x^2+y^2+z^2)^2-3(a^4+x^4+y^4+z^4)\\ &=&(-2)(y^2+\omega x^2+ \omega^2 z^2+ \sigma (L))(y^2+\omega^2 x^2+ \omega z^2+\sigma (K)),\\ &=&(-2)(x^2+\omega^2 y^2+ \omega z^2+ \omega^2 \sigma (L))(x^2+\omega y^2+ \omega^2 z^2+\omega \sigma (K)).\end{aligned}$$ Since $k[x,y,z]$ is a unique factorization domain, it follows that $$\begin{aligned} \sigma (L) = \omega K,~ \sigma (K) = \omega^2 L. \label{Kh3}\end{aligned}$$ Similarly, if $\tau$ is the transposition $x \mapsto z \mapsto x$, then $$\begin{aligned} \tau (L) = \omega^2 K,~ \tau (K) = \omega L. \label{Kh4}\end{aligned}$$ Observing that the constant terms of $L$ and $K$ are unchanged under permutations on $x, y, z$, and using (\[Kh3\]) and (\[Kh4\]), we obtain $$\begin{aligned} L_0 = \omega K_0,~L_0 = \omega^2 K_0.\end{aligned}$$ It follows that $L_0 = K_0 =0$, contradicting (\[Kh2\]). Thus we have proved the case $n=3$. If $n \ge 3$, we let $h$ be obtained from $g$ by putting $x_j = 0$ for all $j \ge 4$. Since $h$ is irreducible by the case $n=3$, it follows that $g$ is irreducible. In fact, if $g=\alpha \beta$, and if $A$ and $B$ are obtained from $\alpha$ and $\beta$ by plugging $x_j = 0$ for all $j \ge 4$, then either $A$ is zero or $A$ is homogeneous of the same degree as $\alpha$. Since $h=AB$, and $h \ne 0$, it follows that $A\ne 0$ and therefore $h$ is reducible, a contradiction. This completes the proof. $\Box$ \[DDD\] Let $k$ be a field with ${\mbox{char~}}k \ne 2$, and let $$\begin{aligned} f &=& (x_1^2+\cdots+x_n^2)^2-t(x_1^4+ \cdots +x_n^4) \in k[x_1,\cdots,x_n], \end{aligned}$$ where $t \in k$ and $t \ne 0$, and where $n \ge 3$. Then $f$ is reducible in and only in the following two cases: - $n=3$ and $t=2$, in which case $f$ is as given in (\[uuu\]). - $n=3$, $t=3$, and $k$ contains a primitive third root $\omega$ of unity, in which case $f$ is as given in (\[vvvv\]). [Proof.]{} The case when $n=3$ was completely treated in Lemmas \[BBB\] and \[CCC\]. So we assume that $n \ge 4$. For any polynomial $F \in k[x_1,\cdots,x_n]$, let $F^* \in k[x_1,x_2,x_3]$ be the polynomial obtained from $F$ by putting $x_4=1$ and $x_j = 0$ for all $j \ge 5$. Then $f^$ is the case $n=3$ and $a=1$ of Theorem \[CCDD\], and is hence irreducible. To show that $f$ is irreducible, we suppose that $f = gh$, and we show that $g$ or $h$ is a constant. Since $f^=g^h^$, and since $f^$ is irreducible, it follows that $g^$ or $h^$, say $g^$, is a constant. Since $f^* \ne 0$, it follows that $g^$ is a non-zero constant. Therefore $$4 = \deg (f^) = \deg (g^) + \deg (h^)= 0 + \deg (h^).$$ Hence $\deg (h^) = 4$. Since $\deg (h) \ge \deg (h^)$, it follows that $\deg (h) = 4$. Therefore $4 = \deg (f) = \deg (g) + \deg (h) = \deg (g) + 4$, and hence $g$ is a constant. Thus $f$ is irreducible, as desired. $\Box$ \[CM\] Let $k$ be a field with ${\mbox{char~}}k \ne 2$, and let $R=k[x_{ij} : 1 \le i < j \le n+1]$, $n \ge 2$, be the polynomial ring over $k$ in the set $X=\{x_{ij} : 1 \le i < j \le n+1\}$ of $(n+1)n/2$ indeterminates. For $1 \le i \le j \le n+1$, let us make the convention that $x_{i,j} = x_{j,i}$ and $x_{j,j} = 0$. Let $M$ be the Cayley-Menger determinant in $X$, i.e., $M$ is the $(n+2)\times (n+2)$ determinant whose entries $c_{i,j}$, $0 \le i, j \le n+1$, are given by $$\begin{aligned} c_{i,j} &=& \left\{ \begin{array}{lcl} 0 &\mbox{if}& i=j,\\ 1 &\mbox{if}& i \ne j \mbox{~and~} ij=0,\\ x_{i,j}^2 & \mbox{otherwise.} \end{array}\right. \end{aligned}$$ Thus $$\begin{aligned} \label{CMCM} M&=& \left\| \begin{array}{cccccc} 0&1&1&1& \cdots\cdots & 1\\ 1&0&x_{1,2}^2&x_{1,3}^2& \cdots\cdots&x_{1,n+1}^2\\ 1&x_{2,1}^2&0&x_{2,3}^2& \cdots\cdots&x_{2,n+1}^2\\ 1&x_{3,1}^2&x_{3,2}^2& 0& \cdots\cdots&x_{3,n+1}^2\\ \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots \\ \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots \\ \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots & \cdots \cdots \\ 1&x_{n+1,1}^2&x_{n+1,2}^2& x_{n+1,3}^2& \cdots \cdots & 0 \end{array} \right\|.\end{aligned}$$ Then 1. $M$ is homogeneous of homogeneity degree $2n$, 2. $M$ is reducible if and only if $n=2$. In this case, if we let $$x_{1,2} = z,~x_{2,3} = x,~x_{1,3}=y,$$ then $$\begin{aligned} M&=& \left\| \begin{array}{cccccc} 0&1&1&1\\ 1&0&z^2&y^2\\ 1&z^2&0&x^2\\ 1&y^2&x^2& 0 \end{array} \right\| \nonumber \\ &=& -(x + y + z) (-x+y+z) (x - y + z) (x+y-z). \label{Heron-M}\end{aligned}$$ [Proof.]{} (i) Let $M_1$ be the matrix obtained from $M$ by replacing each $x_{i,j}$ in $M$ by $\lambda x_{i,j}$, let $M_2$ be obtained from $M_1$ by pulling out $\lambda^2$ as a common factor in each row of $M_1$ except the upper most one, and let $M_3$ be obtained from $M_2$ by pulling out $1/\lambda^2$ as a common factor in the left most column. Then $M_3=M$, and therefore $$M_1 = \lambda^{2(n+1)} M_2 = \lambda^{2(n+1)-2} M_3 = \lambda^{2n} M.$$ Hence $M$ is either 0 or homogeneous of homogeneity degree $2n$. To see that $M \ne 0$, it is easy to see that, and it follows from Lemma 3.3 of [@impurity], the determinant obtained from $M$ by plugging $x_{i,j}=1$ for every $x_{i,j}$ is $(-1)^{n-1} (n+1) \ne 0$. This completes the proof of (i). \(ii) The case $n=2$, i.e., the identity (\[Heron-M\]), is easy to check. So we assume that $$n\ge 3.$$ Let $V_0$, $V_1$, and $V$ be the sets of indeterminates defined by $$\begin{aligned} V_1 = \{ x_{i,j} : 1 \le i < j \le n\},~V_0 = \{ x_{n+1,j} : 1 \le j \le n\},~ V = V_0 \cup V_1, \label{VV0V1}\end{aligned}$$ and let $R_0$ and $R$ be the polynomial rings $$\begin{aligned} R_0 = k[V_0],~R = k[V]. \label{R0R}\end{aligned}$$ For any non-zero $\phi \in R$, let the total degree of $\phi$ be denoted by $\deg (\phi)$, and let the degree of $\phi$ in the indeterminates $V_0$ be denoted by $\deg_0 (\phi)$. Since the indeterminates $V_0$ appear only in the last row and last column of $M$, it follows that $$\begin{aligned} \deg_0 (M) &\le& 4. \label{delta}\end{aligned}$$ Now suppose that $M$ is reducible, say $$\begin{aligned} M &=& fg, \label{Mfg}\end{aligned}$$ where $f$ and $g$ are homogeneous polynomials in $R$ of degrees $t, s \ge 1$. We are to arrive at a contradiction. For any $\phi \in R$, define $\phi^ \in R_0 [x]$ to be the polynomial obtained from $\phi$ by replacing every $x_{i,j}$ and $x_{j,i}$, $1 \le i < j \le n$, by $x$. Then $$\begin{aligned} M^* &=& f^g^. \label{Mfgstar}\end{aligned}$$ By [@prekites Theorem 4.1], $M^$ is the Cayley-Menger determinant of the $n$-pre-kite $$PK[n;x;x_{n+1,1},\cdots,x_{n+1,n}]$$ and is given by $$\begin{aligned} \label{Mstar=H} M^&=& (-x^2)^{n-2} H, \mbox{~where~} \nonumber\\ H &=& n(x^4+x_{n+1,1}^4+\cdots+x_{n+1,n}^4) -(x^2+x_{n+1,1}^2+\cdots+x_{n+1,n}^2)^2.\end{aligned}$$ By Theorem \[DDD\], $H$ is irreducible. Since $M^* = f^* g^$, it follows that $H$ divides one of the polynomials $f^$ and $g^$, say $g^$. Therefore $$\begin{aligned} \deg_0 (g) = \deg_0 (g^) \ge \deg_0 (H) = 4,\end{aligned}$$ and hence $$\begin{aligned} 4 \ge \deg_0 (M) = \deg_0 (f) + \deg_0 (g) \ge \deg_0 (f) + 4.\end{aligned}$$ Therefore $\deg_0 (f) = 0$, and hence $f$ does not contain any of the variables $x_{i,n+1}$, $1 \le i \le n$. Thus $f$ must contain at least one of the other variables, say $x_{1,2}$. Let $F$ be obtained from $f$ by replacing $x_{1,2}$ by $x_{1,n+1}$. Since $M$ is symmetric, and since $f$ divides $M$, it follows that $F$ divides $M$, and therefore $M=FG$, for some $G \in R$. Thus $M^=F^G^=(-x^2)^{n-2}H$. We now show that this is a contradiction. Since $H$ is irreducible, it follows that either $H$ divides $F^$ or $G^$. The first case is impossible since $F^$ does not contain $x_{2,n+1}$. The second case implies that $F^$ divides $(-x^2)^{n-2}$, which is again impossible because $F^$ contains $x_{1,n+1}$. This completes the proof. $\Box$ Remarks {#555} ======= In the mathematical literature, there have been some polynomials which have attracted special attention, either for the elegance, or for their popping up in diverse, seemingly unrelated, contexts. The polynomial $x^3+y^3+z^3-3xyz$ is the favorite polynomial alluded to in the title of [@favorite], and it is the subject of Remark 4 (p. 193) of [@EM-ec]. Remark 5 (p. 194) of [@EM-ec] is devoted to the polynomial $x^3-(a^2-b^2-c^2)x+2abc$. The Newton polynomial $x^3-(a^2+b^2+c^2)x+2abc$, too, has been a source of fascination, and has appeared in [@OMT], and is the subject of [@NewtonDE]. The polynomials $f$ and $g$ defined in (\[f-a\]) and (\[g\]) above also have their shares. The special case $$\begin{aligned} h=(a^2+x^2+y^2+z^2)^2-3(a^4+x^4+y^4+z^4)\end{aligned}$$ which describes the relation among the distances of the vertices of an equilateral triangle of side length $a$ to an arbitrary point in its plane is the tool that solves a problem that has appeared very frequently. The problem challenges readers to find $a$ given $x, y, z$ (or to find $z$ given $a$, $x$, $y$). The three known numbers are usually given to be $3,4,5$, and the fourth unknown number is found to be some irrational number that is not very pleasant, namely $\sqrt{25+12\sqrt{3}}$; see [@500 Problem 327, p. 162)], [@Wagon Chapter 1, Problem 27, p. 8; solution, p. 83–84]. A similar problem is solved in [@Graham §55, p. 34], where the given numbers are $80$, $100$, and $150$, and where the answer is again not pleasant. However, the problem appears in [@CMJ], where the given numbers are 5, 7, and $8$, and where the fourth number is found to be 3. One wonders whether there are integral quadruplets other than $(3, 5, 7, 8)$. This naturally leads to the Diophantine equation $$\begin{aligned} \label{Sastry} (w^2+x^2+y^2+z^2)^2-3(w^4+x^4+y^4+z^4)&=&0,\end{aligned}$$ which is studied in [@Sastry], where many (but not all) of its integer solutions are found. It would be an interesting challenge to try to find [all]{} the integer solutions of (\[Sastry\]), and even of the more general (\[g\]). [99]{} S. Abu-Saymeh and M. Hajja, Equicevian points on the altitude of a triangle, [Elem. Math.]{} [67]{} (2012), 187–195. S. Abu-Saymeh, M. Hajja, and M. Hayajneh, The open mouth theorem, or the scissors lemma, for orthocentric tetrahedra, [J. Geom.]{} [103]{} (2012), 1–16. E. J. Barbeau, M. S. Klamkin, and W. O. J. Moser, [Five Hundred Mathematical Challenges]{}, Spectrum Series, MAA, Washington, D. C., 1995. J. Bentin, Regular simplicial distances, [Math. Gaz.]{} [79]{} (1995), 106. C. D’Andrea and M. Sombra, The Cayley-Menger determinant is irreducible for $n \ge 4$, Sib. J. Math. [46]{}, 71–76. L. A. Graham, [Ingenious Mathematical Problems and Methods]{}, Dover, N. Y. 1959. M. Hajja and M. Hayajneh, Impurity of the corner angles in certain special families of simplices, [J. Geom.]{} 105* (2016), 539–560. M. Hajja, M. Hayajneh, and I. Hammoudeh, Pre-kites: Simplices having a regular facet, preprint. M. Hajja, M. Hayajneh, B. Nguyen, and Sh. Shaqaqha, Distances from the vertices of a regular simplex, preprint. M. Hajja and J. Sondow, Newton’s quadrilateral and his equation $x^3-(a^2+b^2+c^2)x + 2abc$, preprint. J. D. E. Konhauser, D. Velleman, and S. Wagon, [Which Way Did the Bicycle Go? ... and Other Intriguing Mathematical Mysteries]{}, The Dolciani Math. Expositions, No. 18, MAA, Washington, D. C., 1996. S. Lang, [Algebra]{}, Third Edition, Addison-Wesley Publishing Company, Reading, Massachusetts, 1993. D. MacHale, My favourite polynomial, [Math. Gaz.]{}, [75]{} (1991), 157–165. S. Rabinowitz, Problem N-3, [AMATYC Review]{} [9]{} (Spring 1988), 71. K. R. S. Sastry, Natural number solutions to $3(p^4+q^4+r^4+s^4)=(p^2+q^2+r^2+s^2)^2$, [Mathematics and Computer Education]{}, Winter 2000; 34, 1. N. Schaumberger, Problem 187, [College Math. J.]{} [12]{} (1981), 155; solution by Howard Eves, ibid, [13]{} (1982), 276–283. R. J. Walker, [Algebraic Curves]{}, Springer-Verlag, New York, 1978.	{ "pile_set_name": "ArXiv" }
--- abstract: \| Bishop’s informal set theory is briefly discussed and compared to Lawvere’s Elementary Theory of the Category of Sets (ETCS). We then present a constructive and predicative version of ETCS, whose standard model is based on the constructive type theory of Martin-L[ö]{}f. The theory, CETCS, provides a structuralist foundation for constructive mathematics in the style of Bishop. 03B15, 03G30, 18B05, 18B25. author: - 'Erik Palmgren[^1]' date: 'October 23, 2009. Revised: August 11, 2010 and July 14, 2011' title: \| Constructivist and Structuralist Foundations:\ Bishop’s and Lawvere’s Theories of Sets --- Introduction ============ Errett Bishop’s book [Foundations of Constructive Analysis]{} from 1967 contains a chapter on set theory. This set theory, apart from being informal, is quite unlike any of the theories of Zermelo–Fraenkel or G[ö]{}del–Bernays, which are derived from the iterative concept of set. > “A set is not an entity which has an ideal existence: a set exists only when it has been defined. To define a set we prescribe, at least implicitly, what we (the constructing intelligence) must do in order to construct an element of the set, and what we must do to show that two elements are equal.” (Bishop 1967, p. 2) We find a similar explanation of what a set is also in the type theory of Martin-L[ö]{}f (1984). Both explanations are aligned to Cantor’s early explanation of sets from 1882 in the respect that they mention conditions for equality of elements explicitly. See Tait (2000) for a discussion. Bishop (1967, p. 74) emphasizes that two elements may not be compared unless they belong to some common set. This indicates a type-theoretic attitude to the foundations. Bishop’s version of set theory has, despite its constructiveness, a more abstract character than e.g. ZF set theory in that it does not concern coding issues for basic mathematical objects. It defines a subset of a set $X$ to be a pair $(A,i_A)$ where $i_A:A \to X$ is function so that $a = b$ if, and only if, $i(a)= i(b)$. An element $x\in X $ is a member of the subset if $x=i_A(a)$ for some $a \in A$. That the subset $(A,i_A)$ is included in another subset $(B,i_B)$ of $X$ is defined by requirement that there is a function $f:A \to B$ so that $i_A= i_B \circ f$, i.e. that the diagram $$\label{bishincl} \bfig \Vtriangle[A`B`X;f`i_A`i_B] \efig$$ commutes. The subsets are equal in case $f$ is a bijection. Unions and intersection are only defined when the involved sets are subsets of the same underlying set. These and other features of Bishop’s set theory are remarkably reminiscent of Lawvere’s [Elementary Theory of the Category of Sets]{} (ETCS) introduced in 1964. ETCS is obtained by singling out category-theoretic universal properties of various set construction in such a way that they become invariant under isomorphism; see (McLarty 2004) and the introduction (McLarty 2005) to (Lawvere 2005), the full version of the 1964 paper. This invariance is of course fundamental for a [structuralist foundation]{}. ETCS is an elementary theory in the sense that it uses classical first order logic as a basis, and make no special assumption on existence of second order or higher order objects. The theory is equivalent to the axioms of a well-pointed topos with the axiom of choice (McLarty 2004, MacLane 1998). It should be emphasized that ETCS was introduced to give an immediate axiomatization of sets, while the Lawvere-Tierney elementary theory of a topos was intended to give axioms for sheaves of sets over an arbitrary topological space. Bishop (1970a, 1970b) considered various versions of G[ö]{}del’s system T as a possible foundation for his set theory. At the basis of the interpretation is a system of computable functions and functionals, which in effect are the core operations of certain modern programming languages. Full-fledged systems suitable for the formalization of constructive mathematics in the style of Bishop emerged later with the constructive type theory of Martin-L[ö]{}f (1975) and the constructive set theories CST (Myhill 1975) and CZF (Aczel 1978). Of these, the type-theoretic system is the more fundamental from a constructive semantical point of view, since it describes explicitly how the computation of functions are carried out. Indeed, the mentioned set-theoretic system, CZF, can be justified on the grounds of Martin-L[ö]{}f’s type theory (MLTT) as shown by Aczel (1978) by a model construction. In MLTT the explanation of when elements of a set (type) are equal halts at the level of definitional equality. There are no quotient constructions, so it is customary to consider a type together with an equivalence relation, as a set-like object, a so-called [setoid]{}. This gives two possible conceptions of constructive sets based on the formal theories CZF and MLTT, namely iterative sets (sets as trees) and setoids respectively. In this paper we present a constructive version of ETCS, called CETCS, which is obtained abstracting on category-theoretic properties of CZF sets and of setoids in a universe in MLTT. A first requirement on CETCS is of course that we use intuitionistic first order logic instead of the customary classical logic. CETCS has however the property that by adding the law of excluded middle and the axiom of choice (AC), we get a theory equivalent to ETCS. Furthermore the theories of Aczel–Myhill and Martin-L[ö]{}f are (generalized) predicative, so that power set principles are not valid. Thus a constructive ETCS cannot be obtained by adding axioms to the elementary theory of toposes. In Moerdijk and Palmgren (2000, 2002) a notion of predicative topos was introduced taking the setoids of MLTT with a hierarchy of universes as a standard model. Other variants of predicative toposes have been introduced and studied (van den Berg 2005); see also Maietti (2005) and Awodey and Warren (2005). A drawback of the category of setoids, as opposed categories of sets, is that there is no canonical choice of pullbacks (Sect. 6, Hofmann 1994). This makes the formulation of some axioms a bit less concise, but also more general. We emphasize that ETCS does not deal with the set-class distinction or replacement axioms. ETCS with replacement has however been considered (Osius 1974, McLarty 2004). A constructive treatment of the set-class distinction was given by Joyal and Moerdijk (1995) by the introduction of notion of a small map. Predicatively acceptable versions of this were developed in (Moerdijk and Palmgren 2002) and (Moerdijk and van den Berg 2008). It seems rather straightforward to extend CETCS to include axioms for small maps along those lines. Another possible extension of CETCS is to add inductively defined subsets. We leave these investigations for another occasion. A feature of CETCS is that it introduces a constructive version of well-pointedness. Shulman (2010) gives a definition of this notion which works for weaker categories. An outline of the paper is a follows: In Section 2 a standard first-order logic definition of categories is given. We present in Section 3 some notation regarding relations and subobjects for categories where products are not supposed to be chosen. The axioms of ETCS and CETCS are presented in parallel and compared in Section 4. In Section 5 some elementary set-theoretic consequence are drawn from CETCS, which indicates its usefulness for Bishop style constructive mathematics. It is shown that CETCS together with the axiom of choice and classical logic gives the original ETCS. The relation of CETCS to standard category theory notions is given in Section 6 and Section 7. This can part can be skipped by the reader that is not particularly interested in categorical logic. Section 7 contains a technical contribution which shows how a “functor free” formulation of locally cartesian closed categories (LCCCs) can be employed in categorical logic. ### Acknowledgment {#acknowledgment .unnumbered} The main results of this article were obtained while the author was a fellow of the Swedish Collegium for Advanced Study, January through June 2009. Many thanks go to the Collegium and its principal Professor Bj[ö]{}rn Wittrock for the opportunity to work in this stimulating research environment. The author is grateful to Institut Mittag-Leffler for support, and to the anonymous referee for helpful remarks. Elementary Categories ===================== We shall take care to formulate all the axioms so that they may be easily cast in many sorted first-order (intuitionistic) logic. Following the notation of Mac Lane (1998), a category ${\cal C}$ is specified by an algebraic signature consisting of three collections ${\cal C}_0,{\cal C}_1,{\cal C}_2$ (for objects, mappings (or arrows), composable mappings) and six functions ${{\sf id}}: {\cal C}_0 \to {\cal C}_1$, ${{\sf dom}}, {{\sf cod}}: {\cal C}_1 \to {\cal C}_0$, ${{\sf comp}}: {\cal C}_2 \to {\cal C}_1$, ${{\sf fst}},{{\sf snd}}: {\cal C}_2 \to {\cal C}_1$. The intention is that ${{\sf dom}}$ gives the domain of the mapping while ${{\sf cod}}$ gives its codomain. The collection ${\cal C}_2$ is supposed to consist of composable mappings $$\cdot \to^f \cdot \to^g \cdot$$ and ${{\sf fst}}$ gives the first of these mappings while ${{\sf snd}}$ gives the second mapping. Then ${{\sf comp}}$ is the composition operation. The axioms for a category are then briefly as follows, where variables ranges are $x \in {\cal C}_0$, $f,g,h,k,\ell \in {\cal C}_1$, $u,v \in {\cal C}_2$: (K1) ${{\sf dom}}({{\sf id}}_x) = x$, (K2) ${{\sf cod}}({{\sf id}}_x) = x$, (K3) ${{\sf dom}}({{\sf comp}}(u)) = {{\sf dom}}({{\sf fst}}(u))$, (K4) ${{\sf cod}}({{\sf comp}}(u)) = {{\sf cod}}({{\sf snd}}(u))$ and - ${{\sf fst}}(u) = {{\sf fst}}(v), {{\sf snd}}(u) = {{\sf snd}}(v) \Longrightarrow u=v$ - ${{\sf dom}}(f) = {{\sf cod}}(g) \Longrightarrow (\exists u:{\cal C}_2)\, ({{\sf snd}}(u) = f\, \&\, {{\sf fst}}(u) = g)$ We introduce abbreviations: for mappings $f,g,h$ write $h \equiv g \circ f$ for $(\exists u \in {\cal C}_2)[{{\sf fst}}(u)=f\, \&\, {{\sf snd}}(u) = g \, \&\, {{\sf comp}}(u) = h],$ that is, the diagram $$\bfig \Atriangle/<-`>`>/[\cdot`\cdot`\cdot;f`g`h] \efig$$ is composable and commutes. Write $k \circ h \equiv g \circ f$ if there is a mapping $m$ so that $m \equiv g \circ f$ and $m \equiv k \circ h$, that is, the following diagram composes and commutes $$\bfig \square[\cdot`\cdot`\cdot`\cdot; f`h`g`k] \efig$$ In terms of these abbreviations we can express the monoid laws: (K7) $f \equiv f \circ ({{\sf id}}_{{{\sf dom}}(f)})$, (K8) $f \equiv ({{\sf id}}_{{{\sf cod}}(f)}) \circ f$, and (K9) if $k \equiv f \circ g$ and $\ell \equiv g \circ h$ then $k \circ h \equiv f \circ \ell$. $f:a \to b$ and $a \to^f b$ are abbreviations for the conjunction ${{\sf dom}}\, f = a \,\&\, {{\sf cod}}\, f = b$. We shall often omit $\circ$ and write $h\equiv gf$ for $h \equiv g \circ f$. Moreover $\equiv$ is often replaced by $=$ when there is no danger of confusion. Subobjects and Relations ======================== We may define the notion of an $n$-ary relation in any category. Recall that a mapping $f:A \to B$ is [monic]{} or [is a mono]{} if for any mappings $h,k: U\to A$ with $fh = fk$ it holds that $h=k$. We write in this case $f: A {\to/ >->/}B$. This notion can be generalized to several mappings. A sequence of mappings $r_1: R \to X_1, \ldots, r_n: R \to X_n$ are [jointly monic,]{} if for any $f,g:U \to R$ $$r_1 f = r_1 g, \ldots, r_n f = r_n g \Longrightarrow f= g.$$ In this case we write $(r_1,\ldots,r_n): R {\to/ >->/}(X_1,\ldots,X_n)$. We regard this as an [$n$-ary relation between the objects $X_1,\ldots,X_n$.]{} In particular, a [binary relation between $X_1$ and $X_2$]{} is a pair of mappings $r_1: R \to X_1$ and $r_2: R \to X_2$ which are jointly monic. Another particular case is: if the category has a terminal object ${{\bf 1}}$, a $0$-ary relation $(): R {\to/ >->/}()$ means that the unique map $R \to {{\bf 1}}$ is a mono. Consider a category ${\cal C}$ with a terminal object ${{\bf 1}}$. An [element]{} of an object $A$ of ${\cal C}$ is a mapping $x:{{\bf 1}}\to A$. For a monic $m: M \to X$ and element $x$ of $X$ write $x {\, \epsilon \,}m$ if $(\exists a:{{\bf 1}}\to M) ma =x$. We say that [$x$ is a member of $m$.]{} More generally, if $(m_1,\ldots, m_n): M {\to/ >->/}(X_1,\ldots,X_n)$ and $(x_1,\ldots, x_n) : {{\bf 1}}{\to/ >->/}(X_1,\ldots,X_n)$ we write $(x_1,\ldots,x_n) {\, \epsilon \,}(m_1,\ldots, m_n)$ if there is $a:{{\bf 1}}\to M$ so that $m_i a = x_i$ for all $i=1,\ldots,n$. To simplify notation we often write $x \in X$ and $(x_1,\ldots,x_n) \in (X_1,\ldots,X_n)$ for $x : {{\bf 1}}\to X$ and $(x_1,\ldots,x_n) : {{\bf 1}}\to (X_1,\ldots,X_n)$, respectively. Note the difference between the signs $\in$ (elementhood) and ${\, \epsilon \,}$ (membership). We shall be interested in categories where there is no canonical construction for products, but where it is merely assumed that they exist. Recall that an [$n$-ary product diagram]{} in a category is a sequence of mappings $X \to_{p_i} X_i$ ($i=1,\ldots,n$) so that for any sequence of mappings $C \to_{p_i} X_i$ ($i=1,\ldots,n$) there is a unique $h:C \to X$ such that $f_i \equiv h p_i$ for all $i=1,\ldots,n$. We write $$h \equiv \langle f_1,\ldots,f_n \rangle_{\bar{p}}$$ when $f_i \equiv h p_i$ for all $i=1,\ldots,n$, where $\bar{p}=p_1,\ldots,p_n$. It is convenient to drop the subscripts $\bar{p}$ when the product diagrams are obvious from the context. \[prop31\] Suppose that $X \to_{p_i} X_i$ ($i=1,\ldots,n$) is a product diagram. If $(r_1,\ldots,r_n) : R \to (X_1,\ldots,X_n)$, $r':R \to X$ and $r' \equiv \langle r_1,\ldots,r_n \rangle_{\bar{p}}$, then $r'$ is monic iff $(r_1,\ldots,r_n)$ are jointly monic. Moreover, for $(x_1,\ldots,x_n) \in (X_1,\ldots,X_n)$, $x' \in X$ with $x' \equiv \langle x_1,\ldots,x_n \rangle_{\bar{p}}$, we have $$x' {\, \epsilon \,}r' \Longleftrightarrow (x_1,\ldots,x_n) {\, \epsilon \,}(r_1,\ldots,r_n). \; {\Box}$$ A binary relation $f=(\xi,\upsilon) : R {\to/ >->/}(X,Y)$ is a [partial function]{} in case $\xi$ is mono. It is a [total function]{} in case $\xi$ is iso. A relation $$f=(\xi_1,\ldots,\xi_n,\upsilon) : R {\to/ >->/}(X_1,\ldots,X_n,Y)$$ is a [partial function of $n$ variables]{} if $(\xi_1,\ldots,\xi_n) : R {\to/ >->/}(X_1,\ldots,X_n)$. We write $$f : (X_1,\ldots,X_n) {\rightharpoondown}Y.$$ It is [ total function of $n$ variables]{} if $R\to^{\xi} X_i$ ($i=1,\ldots,n$) is a product diagram. We write $$f : (X_1,\ldots,X_n) \to Y.$$ For $x_1 \in X_1,\ldots, x_n \in X_n$ and $y \in Y$ we write $$f(x_1,\ldots,x_n) \equiv y$$ in case $(x_1,\ldots,x_n,y) {\, \epsilon \,}f$. Axioms of ETCS and CETCS ======================== Lawvere’s theory ETCS (Lawvere 2005) has eight axioms: (L1) finite roots exist, (L2) the exponential of any pair of objects exist, (L3) there is a Dedekind-Peano object, (L4) the terminal object is separating, (L5) axiom of choice, (L6) every object not isomorphic to an initial object contains an element, (L7) Each element of a sum is a member of one of its injections, (L8) there is an object with more than one element. We present a constructive version of ETCS, called CETCS, and some extensions, by laying down axioms for a category $\cal C$. (It should be evident that the following axioms may be formulated in first-order logic in a language with ${\cal C}_0, {\cal C}_1, {\cal C}_2$ as sorts and the function symbols ${{\sf id}}, {{\sf dom}}, {{\sf cod}}, {{\sf comp}}, {{\sf fst}},{{\sf snd}}$ as indicated in Section 2.) Lawvere’s (L1) says that the category is [bicartesian,]{} i.e. both cartesian and cocartesian. Recall that $\cal C$ is [cartesian]{} if the conditions (C1) – (C3) are satisfied: (C1) There is a terminal object ${{\bf 1}}$ in $\cal C$. (C2) Binary products exist: For any pair of objects $A$ and $B$ there exists an object $P$ and two mappings $$A \to/<-/^p P \to^q B$$ which are such that if $A \to/<-/^f X \to^g B$ then there exists a unique $h: X \to P$ so that $p h \equiv f$ and $q h \equiv g$. (C3) Equalizers exist: For any parallel pair of mappings $A \two^f_g B$ there exists a mapping $e:E \to A$ so that $f e \equiv g e$ and such that whenever $h:X \to A$ satisfies $f h \equiv g h$ then there exists a unique $k: X \to E$ with $e k \equiv h$. A category $\cal C$ is [cocartesian]{} if it satisfies (D1) – (D3), which are the categorical duals of (C1) – (C3). (D1) There is an initial object ${{\bf 0}}$ in $\cal C$. (D2) Binary sums exist: For any pair of objects $A,B$ there is a diagram $$\label{sumdia} A \to^i S \to/<-/^j B$$ such that if $A \to^f T \to/<-/^g B$ then there is a unique $h :S \to T$ with $h i \equiv f$ and $h j \equiv g$. (D3) Coequalizers exist: For any parallel pair of mappings $A \two^f_g B$ there exists a mapping $q:B \to Q$ so that $q f \equiv q g$ and such that whenever $h:B \to Y$ satisfies $h f \equiv h g$ then there exists a unique $k: Q \to E$ with $k q \equiv h$. The axiom (L2) of ETCS says together with (L1) that the category is cartesian closed. Instead, we take for an axiom the following ($\Pi$) which, together with cartesianess and axiom (G) below, states that the category is locally cartesian closed. (This axiom is a theorem of ETCS.) ($\Pi$) Dependent products exist: For any mappings $Y \to^g X \to^f I $ there exists a commutative diagram $$\label{defpid} \bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}`g`] \square(550,0)[P`F`X`I;\pi_1`\pi_2`{{\varphi}}`f] \efig$$ where the square is a pullback, and which is such that for any element $i \in I$ and any partial function $\psi=(\xi,\upsilon) : R {\rightharpoondown}(X,Y)$ such that - for all $(x,y) \in (X,Y)$, $(x,y) {\, \epsilon \,}\psi$ implies $g y \equiv x$ and $f x \equiv i,$ - if $f x \equiv i$, then there is $y \in Y$ with $(x,y) {\, \epsilon \,}\psi$, then there is a unique $s \in F$ so that ${{\varphi}}s = i$ and for all $(x,y) \in (X,Y)$, $$\label{equival1} (s,x,y) {\, \epsilon \,}\alpha \Longleftrightarrow (x,y) {\, \epsilon \,}\psi.$$ Here $\alpha=(\pi_1,\pi_2,{{\rm ev}}): P {\to/ >->/}(F,X,Y)$. A diagram (\[defpid\]) satisfying these properties is called a [universal dependent product diagram]{} or shortly a [universal $\Pi$-diagram]{} for $Y \to^g X \to^f I$. The third axiom (L3) of ETCS says, in now common terminology, that there exists a [natural numbers object]{} (NNO). A category $\cal C$ has an NNO if there is a sequence of mappings (the NNO) ${{\bf 1}}\to^0 N \to^s N$ so that for any other sequence of mappings ${{\bf 1}}\to^b A \to^h A$ there is a unique $f: N \to A$ with $f 0 \equiv b$ and $f S \equiv h f$. Axiom (L4) states in modern terminology that ${{\bf 1}}$ is a separating object, i.e. as in Proposition \[prop41\]. We consider instead a stronger axiom (G) which is a theorem of ETCS. A mapping $f:A \to B$ of $\cal C$ is [onto]{} if for any $y \in B$ there exists an $x \in A$ so that $y \equiv f x$. Our axiom is \(G) Any mapping which is both onto and mono, is an isomorphism. The fifth axiom (L5) of ETCS states the axiom of choice in peculiar way; see Section 5.2. A more standard way is to first define an object $P$ of $\cal C$ to be a [choice object,]{} if for any onto $f:A \to P$ there is a $g:P \to A$ with $f g = {{\sf id}}_P$. The [axiom of choice]{} (AC) says that every object is a choice object. This is a far too strong assumption in a constructive setting. There is a constructively acceptable weakening which accords well with Bishop’s distinction of operations and functions, the [presentation axiom]{} (Aczel 1978): (PA) For any object $A$ there is an onto mapping $P \to A$ where $P$ is a choice object. Axiom (L6) of ETCS says in contrapositive form: if an object has no elements then it is an initial object. We take instead \(I) The object ${{\bf 0}}$ has no elements. This together with (G) implies (L6). The Axiom (L7) of ETCS is [each element of a sum is a member of one of its injections.]{} We adopt this axiom unaltered but call it the [disjunction principle (DP)]{} as it connects sums to disjunctions: (DP) In a sum diagram $A \to^i S \to/<-/^j B$: for any $z\in S$, $z {\, \epsilon \,}i$ or $z {\, \epsilon \,}j$. The final axiom (L8) of ETCS states that there exists object with at least two elements. We state this as (NT, Non-triviality) For any sum diagram ${{\bf 1}}\to^x S \to/<-/^y {{\bf 1}}$ it holds that $x \ne y$. There are two further axioms that we shall consider, which are in fact theorems of ETCS. (Fct) Factorization. Any mapping $f$ can be factored as $f \equiv i e$ where $i$ is mono and $e$ is onto. (Eff) All equivalence relations are effective. For each equivalence relation $(r_1,r_2) : R {\to/ >->/}(X,X)$ there is some mapping $e: X \to E$ so that $$(x_1,x_2) {\, \epsilon \,}(r_1,r_2) \Longleftrightarrow e x_1 \equiv e x_2$$ for all $(x_1,x_2) \in (X,X)$. In summary, the theory CETCS consists of the axioms (C1 – C3), (D1- D3), ($\Pi$), (G), (PA), (I), (DP), (NT), (Fct) and (Eff). Observe that it is a finitely axiomatized theory just as ETCS. We do not know whether this set of axioms is optimal. [Note that it is not assumed that the (co)products or (co)equalizers are given as functions of their data. The axiom (G) is in the terminology of Johnstone (2002) that [${{\bf 1}}$ generates $\cal C$.]{} It entails that one can “reason using elements” as the two following results exemplify. This gives a substantial simplification of the internal logic.]{} \[prop41\] Let $\cal C$ be a cartesian category which satisfies (G). Then - For any pair of mappings $f,g: A \to B$, $f=g$ whenever $(\forall x\in A)(f x = g x)$. - A mapping $f:A \to B$ is monic if and only if $(\forall x,y\in A)(f x = f y \Rightarrow x = y)$. [Proof.]{} (b) follows easily from (a). To prove the non-trivial direction of (a): assume that $(\forall x\in A)(f x = g x)$. Construct an equalizer $E \to^e A \two^f_g B$ of $f$ and $g$. Then $e$ is monic. By the assumption and the equalizing property it is also easy to see it is onto. Hence by (G) $e$ is an isomorphism. Since $f e = g e$ we get $f=g$. ${\Box}$ Define an element-wise inclusion relation for monos $m:M \to X $ and $n:N \to X$ $$m {\, \dot{\subseteq} \,}n \Longleftrightarrow_{\rm def} (\forall x \in X)(x {\, \epsilon \,}m \Rightarrow x {\, \epsilon \,}n)$$ The standard inclusion relation in a category is given by $m \le n\Longleftrightarrow_{\rm def} (\exists f:M \to N)(m = n f)$. Compare diagram (\[bishincl\]). Their correspondence is given by: \[elts2\] Let $\cal C$ be a cartesian category which satisfies (G). Then for all monos $m:M \to X $ and $n:N \to X,$ $$m {\, \dot{\subseteq} \,}n \Longleftrightarrow m \le n.$$ [Proof.]{} ($\Leftarrow$) This is straightforward in any category with a terminal object. ($\Rightarrow$) Suppose that $m:M {\to/ >->/}X $ and $n:N {\to/ >->/}X$ satisfies $m {\, \dot{\subseteq} \,}n$. Form a pullback square $$\bfig \square[P`N`M`X;p`q`n`m ] \efig$$ To prove $m \le n$ it is evidently enough to show that $q$ is an isomorphism. Now $q$ is the pullback of a mono, so it is a mono as well. By (G) it is sufficient to show that $q$ is onto. Let $y \in M$. Thus $m y {\, \epsilon \,}m$ and by assumption also $m y {\, \epsilon \,}n$. There is thus $t\in N$ with $m y = n t$. Hence by the pullback square there is a unique $u\in P$ so that $q u =y$ and $p u = t$. In particular, this shows that $q$ is onto. ${\Box}$ Functions as a graphs and as morphisms can be characterized as follows. \[fungraph\] Let $\cal C$ be a cartesian category which satisfies (G). Let $r=(r_1,r_2):R \to (X,Y)$ be a relation. Then - $r$ is a partial function if and only if $$\label{pfcon} (\forall x\in X)(\forall y,z \in Y)[(x,y) {\, \epsilon \,}r\; \&\; (x,z) {\, \epsilon \,}r \Rightarrow y=z].$$ - $r$ is a total function if and only if $$\label{pfcon2} (\forall x \in X)(\exists ! y \in Y)(x,y) {\, \epsilon \,}r.$$ - [(Unique Choice)]{} If $(\forall x \in X)(\exists ! y \in Y)(x,y) {\, \epsilon \,}r$, then there is $f:X \to Y$ with $$(\forall x \in X)(x,f x) {\, \epsilon \,}r.$$ [Proof.]{} (a): by definition $r$ is a partial function if and only if $r_1$ is mono. By Proposition \[prop41\], $r$ is thus a partial function precisely when $$(\forall s,t \in R)[r_1 s= r_1 t \Rightarrow s=t].$$ This is easily seen to be equivalent to (\[pfcon\]). (b, $\Rightarrow$): Suppose $r$ is a total function. Then $r_1$ is iso. For $x \in X$, we have $(x,y) {\, \epsilon \,}r$ with $y= r_2 r^{-1}_1 x)$. By (a) it follows that $y$ is unique. (b, $\Leftarrow$): Suppose (\[pfcon2\]) holds. By (a) $r_1$ is mono. For each $x \in X$ there is some $t \in R$ and $y=r_2 t$ so that $(x,y) {\, \epsilon \,}r$. Thus $r_1$ is onto, and by (G) $r_1$ is iso. (c): This is clear from (b, $\Leftarrow$) since then $r_1$ is invertible, and we may take $f=r_2 r^{-1}_1$: for $x \in X$, $x= r_1 r^{-1}_1 x$ and $f x = r_2 r^{-1}_1 x$ so $(x,fx) {\, \epsilon \,}r$. ${\Box}$ Let $\cal C$ be a cartesian category which satisfies (G). Then a commutative diagram $$\bfig \square[P`A`B`C;\pi_1`\pi_2`f`g] \efig$$ is a pullback diagram if, and only if, $$\label{pbchar} (\forall x\in A)(\forall y \in B)[fx = gy \Longrightarrow (\exists ! t \in P) x = \pi_1 t \; \&\; y = \pi_2 t].$$ [Proof.]{} ($\Rightarrow$) Immediate. ($\Leftarrow$): Assume (\[pbchar\]). It follows that $\pi_1$ and $\pi_2$ are jointly monic. Suppose there is given a commutative square $$\bfig \square[Q`A`B`C;q_1`q_2`f`g]. \efig$$ Form the pullback $$\bfig \square[R`Q`P`A\times B;h`k`\langle q_1,q_2 \rangle`\langle \pi_1,\pi_2 \rangle]. \efig$$ Clearly $h$ is mono, since it is a pullback of a mono. By (\[pbchar\]) $$(\forall s \in Q)(\exists ! t \in P) [q_1 s= \pi_1 t \; \&\; q_2 s = \pi_2 t].$$ but this implies that $h$ is onto. Hence $h$ is iso by (G). Thus $m=kh^{-1}: Q \to P$ satisfies $\pi_i m =q_i$ for $i=1,2$, and is the desired map. It is unique since $\pi_1$ and $\pi_2$ are joint monic. ${\Box}$ Basic Set-theoretic Consequences ================================ We mention some easy consequences of the axioms. \[qprop\] [(Quotient sets.)]{} Suppose that the bicartesian category $\cal C$ satisfies (G). For any equivalence relation $r=_{\rm def } (r_1,r_2) : R {\to/ >->/}(X,X)$ there exists a mapping $q: X \to Q$ so that for all $(x_1,x_2) \in (X,X)$ $$\label{rqequiv} (x_1,x_2) {\, \epsilon \,}r \Longrightarrow q x_1 = q x_2$$ and if $f: X \to Y$ is any mapping with $$\label{rqequiv2} (x_1,x_2) {\, \epsilon \,}r \Longrightarrow f x_1 = f x_2.$$ then there exists a unique $h:Q\to Y$ with $h q = f$. In case the category also satisfies (Eff) it follows that (\[rqequiv\]) is an equivalence. [Proof.]{} Construct a coequalizer diagram $$R \two^{r_1}_{r_2} X \to^q Q.$$ Since the diagram commutes, the implication (\[rqequiv\]) holds. Let $f:X \to Y$ be any mapping satisfying the implication (\[rqequiv2\]). Thus for any $t \in R$, $f r_1 t = f r_2 t$. Thus by Proposition \[prop41\] (a) we have $f r_1 = f r_2$ and since $q$ is a coequalizer, there is a unique $h:Q\to Y$ with $h q = f$. From Axiom (Eff) it follows that there is some $e: X \to E$ such that $$\label{rqeuiv3} (x_1,x_2) {\, \epsilon \,}r \Longleftrightarrow e x_1 = e x_2$$ for all $(x_1,x_2) \in (X,X)$. Thus $e r_1 = e r_2$. Let $e': Q \to E$ be the unique mapping so that $e' q = e$. Thus if $q x_1 = q x_2$, it follows that $e x_1 = e x_2$ and hence $(x_1, x_2) {\, \epsilon \,}r$ by (\[rqeuiv3\]). ${\Box}$ [(Induction.)]{} Assume that $\cal C$ is a cartesian category which satisfies (G) and (NNO). Let $r: R {\to/ >->/}N$. Suppose that $0 {\, \epsilon \,}r$ and that for each $n\in N$, $n {\, \epsilon \,}r$ implies $S n {\, \epsilon \,}r$. Then for all $n \in N$, $n {\, \epsilon \,}r$. [Proof.]{} Since $0 {\, \epsilon \,}r$, there is $z: {{\bf 1}}\to R$ with $0 \equiv r z$. Form a pullback square $$\bfig \square[P`R`R`N;q`p`r`S \circ r] \efig$$ As $r$ is mono, so is $p$. We claim that $p$ is onto. Let $u : {{\bf 1}}\to R$. Thus $r u {\, \epsilon \,}r$. Hence by assumption $ S r u {\, \epsilon \,}r$. There is thus a map $v: {{\bf 1}}\to R$ with $S r u = r v$. By the pullback property there is $x: {{\bf 1}}\to P$ so that $p x = u$ and $q x = v$. In particular $p$ is onto. By (G) $p$ is an isomorphism. Let $p^{-1}$ be its inverse. Thus $q p^{-1} : R \to R$. By the property of the natural number object there is a unique $f:N \to R$ with $f 0 = z$ and $f S = q p^{-1} f$. Now $(r f) 0 = 0$ and $$(r f) S = r q p^{-1} f = S r f.$$ But ${{\sf id}}_N$ instead of $r \circ f$ also satisfies these two equations. Thus $r f = {{\sf id}}$. Thus for any $n\in N$, $r f n = n$, and hence $n {\, \epsilon \,}r$. ${\Box}$ [(Exponential objects.)]{} Assume that $\cal C$ is a cartesian category that satisfies (G) and ($\Pi$). Then for any objects $X$ and $Y$ there is an object $E$ and a total function $e: (E,X) \to Y$ such that for every morphism $f: X \to Y$ there is a unique $s \in E$ such that for $x \in X$ and $y \in Y$: $$e(s,x) \equiv y \Longleftrightarrow f \circ x \equiv y$$ [(Dependent choices.)]{} Assume that [ ]{}$\cal C$ is a cartesian category that satisfies (G), ($\Pi$), (Fct) and (PA). Then for any object $X$, any total relation $r=(r_1,r_2):R {\to/ >->/}(X,X)$ and any $x \in X$ there is a morphism $f:N \to X$ with $f 0 = x$ and for all $n \in N$ $$\label{dcrelation} (f n, f \circ S n) {\, \epsilon \,}r.$$ [Proof.]{} (Sketch) Take a projective cover $p:P \to X$ of $X$. Since $r$ is total, we have thus for each $u \in P$ some $v \in P$ with $(p u, p v) {\, \epsilon \,}r$. As $P$ is a choice object, there is a morphism $g: P \to P$ with $(p u, p g u) {\, \epsilon \,}r$ for all $u \in P$. Let $x \in X$. Then $p \circ w \equiv x$ for some $w \in P$. Now ${{\bf 1}}\to^0 N \to^S N$ is a natural numbers object, so there is $h:N \to P$ with $h 0 = w$ and $h S = g h$. Now it is easy to check by induction that $f =_{\rm def} p h$ satisfies (\[dcrelation\]). ${\Box}$ Constructing New Relations -------------------------- We review some of the possibilities to construct relations in a bicartesian category satisfying the axioms (G), ($\Pi$), (DP), (Fct) and (I). On any object $X$ the identity mapping gives a universally true relation ${\sf t}_X = {{\sf id}}_X : X \to X$, i.e. for all $x\in X$ $$x {\, \epsilon \,}{\sf t}_X.$$ The unique mapping from the initial object ${\sf f}_X: {{\bf 0}}\to X$ gives an universally false relation, i.e. for all $x \in X$, $$\lnot (x {\, \epsilon \,}{\sf f}_X).$$ If $E \to^e X \two^g_h Y$ is an equalizer diagram, then for $x \in X$ $$x {\, \epsilon \,}e \Longleftrightarrow g x = h x.$$ Given a relation $r=(r_1,\ldots,r_n): R {\to/ >->/}(X_1,\ldots,X_n)$ we can extend it with a variable. Let $Y$ be a object and let $R \to/<-/^p R'\to^q Y$ be a product diagram. The extended relation $$r'= (r_1 p,\ldots,r_n p,q) : R' {\to/ >->/}(X_1,\ldots,X_n, Y)$$ satisfies, for all $(x_1,\ldots,x_n,y) \in (X_1,\ldots,X_n,Y)$ that $$(x_1,\ldots,x_n,y) {\, \epsilon \,}r' \Longleftrightarrow (x_1,\ldots,x_n) {\, \epsilon \,}r.$$ If $\sigma: \{1,\ldots,n\}\to \{1,\ldots,n\}$ is a permutation then $$r_{\sigma}= (r_{\sigma(1)},\ldots,r_{\sigma(n)}): R {\to/ >->/}(X_{\sigma(1)},\ldots,X_{\sigma(n)})$$ satisfies for all $(x_1,\ldots,x_n) \in (X_{\sigma(1)},\ldots,X_{\sigma(n)})$ $$(x_1,\ldots,x_n) \in r_{\sigma} \Longleftrightarrow (x_{\sigma^{-1}(1)},\ldots,x_{\sigma^{-1}(n)}) \in r.$$ The following lemma is standard If in the universal $\Pi$-diagram $$\bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}`g`] \square(550,0)[P`F`X`I;\pi_1`\pi_2`{{\varphi}}`f] \efig$$ the mapping $g$ is mono, then so is ${{\varphi}}$. ${\Box}$ Relations can be combined using the logical operations ($\land$, $\lor$, $\Rightarrow$) and quantifiers ($\forall$,$\exists$) over fixed objects: \[elemcomp\] Let $\cal C$ be a bicartesian category satisfying the axioms (G), ($\Pi$), (DP), (Fct) and (I). Let $r=(r_1,\ldots,r_n): R {\to/ >->/}(X_1,\ldots,X_n)$ and $s=(s_1,\ldots,s_n): S {\to/ >->/}(X_1,\ldots,X_n)$. Then exists $(r \land s),(r \lor s), (r \Rightarrow s): R {\to/ >->/}(X_1,\ldots,X_n)$ so that for all $x=(x_1,\ldots,x_n)\in (X_1,\ldots,X_n)$ - $x{\, \epsilon \,}(r \land s)$ if and only if $x {\, \epsilon \,}r$ and $x {\, \epsilon \,}s$, - $x{\, \epsilon \,}(r \lor s)$ if and only if $x {\, \epsilon \,}r$ or $x {\, \epsilon \,}s$, - $x{\, \epsilon \,}(r \Rightarrow s)$ if and only if $x {\, \epsilon \,}r$ implies $x {\, \epsilon \,}s$, Moreover, if $m: M {\to/ >->/}(X_1,\ldots,X_n,Y)$ then there is $\forall(m): A {\to/ >->/}(X_1,\ldots,X_n)$ and $\exists(m): E {\to/ >->/}(X_1,\ldots,X_n)$ so that for all $x=(x_1,\ldots,x_n) \in (X_1,\ldots,X_n)$ - $x{\, \epsilon \,}\forall(m)$ if and only if for all $y\in Y$, $(x_1,\ldots,x_n,y) {\, \epsilon \,}m$, - $x{\, \epsilon \,}\exists(m)$ if and only if for some $y\in Y$, $(x_1,\ldots,x_n,y) {\, \epsilon \,}m$. [Proof.]{} By Proposition \[prop31\] it is enough to prove (a) – (c) for the case when $n=1$; write $X=X_1$, $r=r_1$, $s=s_1$. As for (a): form the pullback square $$\bfig\square[P`S`R`X;q`p`s`r]\efig.$$ The diagonal, call it $(r \land s)$ is a mono. It is straightforward by the pullback property that the equivalence in (a) holds. As for (b): form a sum diagram $R \to^i U \to/<-/^j S$. Let $f: U \to X$ be the unique mapping with $r = f i$ and $s = f j$. Let $U \to^e I \to^m X$ be a factorization of $f$ as an onto mapping followed by a mono (Fct). We claim that $(r\lor s)=_{\rm def} m$ satisfies the equivalence in (b). Suppose that $x \in X$ satisfies $x {\, \epsilon \,}r$. Then $x = r t$ for some $t \in R$. Thus $x = f i t = m e i t$, and hence $x {\, \epsilon \,}m$. Similarly $x {\, \epsilon \,}s$ implies $x {\, \epsilon \,}m$. Suppose on the other hand that $x {\, \epsilon \,}m$. Now, $e$ is onto so there is some $u \in U$ with $x = f u$. Then by Axiom (DP) we have $u= i t$ for some $t \in R$, in which case $x {\, \epsilon \,}r$, or we have $u= j v$ for some $v \in S$, in which case $x {\, \epsilon \,}s$. As for (c): Form the pullback $$\label{casec1} \bfig \square/ >->` >->` >->` >->/[Q`S`R`X;q`p`s`r] \efig$$ Axiom ($\Pi$) yields for $Q \to^p R \to^r X$ a universal $\Pi$-diagram $$\label{casec2} \bfig \qtriangle/<-`->`/[Q``;{{\rm ev}}`p`] \square(550,0)[P`F`R`X;\pi_1`\pi_2`{{\varphi}}`r] \efig$$ We claim that $(r \Rightarrow s) =_{\rm def} {{\varphi}}$ makes the equivalence in (c) true. Let $x \in X$. To prove ($\Rightarrow$) assume that $x {\, \epsilon \,}{{\varphi}}$. Thus $x = {{\varphi}}u$ for some $u \in F$. Suppose $x {\, \epsilon \,}r$. Thus $x= r v$ for some $v \in R$. By the pullback in (\[casec2\]) there is $w:{{\bf 1}}\to w$ so that $\pi_2 w = v$ and $\pi_1 w = u$. We have further by the diagrams $$x= r v = r \pi_2 w = r p \,{{\rm ev}}\, w = s q \,{{\rm ev}}\, w.$$ This shows $x {\, \epsilon \,}s$. As for the converse ($\Leftarrow$) suppose the implication $$x {\, \epsilon \,}r \Rightarrow x {\, \epsilon \,}s$$ holds. We aim to show $x {\, \epsilon \,}{{\varphi}}$ using the properties of the universal $\Pi$-diagram. Form a pullback diagram $$\label{casec3} \bfig \square/ >->` >->` >->` >->/[T`Q`{{\bf 1}}`X;t``r p`x] \efig$$ Then $\psi= (p t,t):T {\to/ >->/}(R,Q)$ is evidently a partial function since $p$ and $t$ are monic. If $\psi(u) \equiv v$, then there is $w \in T$, so that $u= p t w$ and $v=t w$, and hence $u = p v$ and $r u = r p v =x$. This verifies condition (a) of ($\Pi$). To verify condition (b) of ($\Pi$), assume that $r u = x$. Thus $x {\, \epsilon \,}r$, and so by the implication above $x {\, \epsilon \,}s$, i.e. $s w = x$, for some $w$. By the pullback (\[casec1\]) there is $v: {{\bf 1}}\to Q$ with $u = p v$ and $w = q v$. Thus $r p v = x$. But, by the pullback (\[casec3\]) there is $z: {{\bf 1}}\to T$ with $t z = v$. Now $(p t z,t z) = (u,v)$, i.e. $\psi(u) \equiv v$. According to ($\Pi$), there is now some $k \in F$ with ${{\varphi}}k =x$. Thus $x {\, \epsilon \,}{{\varphi}}$. As for (d): Suppose $m=(m_1,m_2): M {\to/ >->/}(X,Y)$. Using (Fct) factor $m_1$ into a onto mapping followed by a mono $M \to^e I \to^i X$. Let $\exists(m) = i$. Thus using the fact that $e$ is onto $$x {\, \epsilon \,}\exists(m) \Leftrightarrow (\exists t\in I) x= i t \Leftrightarrow (\exists s \in M) x = i e s \Leftrightarrow (\exists s \in M) x = m_1 s.$$ The latter implies that $(x, m_2 s) {\, \epsilon \,}(m_1,m_2)$. Clearly $m_2 s \in Y$. Conversely, suppose that for some $y \in Y$ we have $(x,y) {\, \epsilon \,}(m_1,m_2)$. Thus for some $s \in M$ it holds that $x= m_1 s$ and $y= m_2 s$ and we have $x {\, \epsilon \,}\exists(m)$. As for (e): Suppose $m=(m_1, m_2): M {\to/ >->/}(X,Y)$. First construct a product diagram $X \to/<-/^p U \to^q Y$. Then let $m' \equiv \langle m_1,m_2 \rangle_{p,q}$. Use ($\Pi$) to obtain a universal $\Pi$-diagram $$\label{casee} \bfig \qtriangle/<-`->`/[M``;{{\rm ev}}`m'`] \square(550,0)[P`F`U`X;\pi_1`\pi_2`{{\varphi}}`p] \efig$$ We let $\forall(m) = {{\varphi}}$. Suppose $x \in X$. To prove (e, $\Rightarrow$) suppose $x {\, \epsilon \,}{{\varphi}}$ and $y \in Y$. Thus $x = {{\varphi}}f$ for some $f \in F$ and moreover there is $u \in U$ with $x= p u$ and $y = q u$. By the pullback in (\[casee\]) we get $w \in P$ so that $u = \pi_2 w$ and $f = \pi_1 w$. From the triangle of (\[casee\]) it follows that $ m' \,{{\rm ev}}\, w = \pi_2 w $. Hence $u {\, \epsilon \,}m'$ and thus $(x,y) {\, \epsilon \,}m$. To prove (e, $\Leftarrow$) let $x \in X$ be fixed and suppose that for all $y \in Y$, $(x,y) {\, \epsilon \,}m$. Let $n: N \to M$ be the pullback of $x$ along $m_1$: $$\label{casee2} \bfig\square[N`M`{{\bf 1}}`X;n``m_1`x]\efig$$ Then $(m' n, n) : N {\to/ >->/}(U,M)$ is a partial function since both $m'$ and $n$ are mono. As for condition (a): if $(u,v) {\, \epsilon \,}(m' n, n)$ then $u = m' n t$ and $v= n t$ for some $t \in N$. Clearly, $m' v= u$ and $p u =m_1 n t = x$. Regarding condition (b): Suppose that $u\in U$ satisfies $p u = x$. Let $y = q u$. By the first assumption $(x,y) {\, \epsilon \,}m$. Thus for some $s \in M$, $x= m_1 s$ and $y = m_2 s$. By construction of $m'$ we have $m' s = u$. Since $x= m_1 s$, the pullback (\[casee2\]) gives a unique $t\in N$ with $s = n t$. Thus $t$ is a witness to $(u,s) {\, \epsilon \,}(m' n, n)$. Since conditions (a) – (b) are now verified, ($\Pi$) gives $f \in F$ satisfying, in particular, ${{\varphi}}f = x$. Hence $x {\, \epsilon \,}{{\varphi}}$. ${\Box}$ Decidable Relations and Classical Logic --------------------------------------- Let $\cal C$ be a CETCS category. Construct a two element set using the sum axiom ${{\bf 1}}\to^f {{\bf 2}}\to/<-/^t{{\bf 1}}$. If $r: P {\to/ >->/}X$ is decidable, i.e. for all $x \in X$, $$\mbox{$x {\, \epsilon \,}r$ or $\lnot x{\, \epsilon \,}r$},$$ then we can construct $\chi_r: X \to {{\bf 2}}$ so that for all $x \in X$ $$\mbox{$x {\, \epsilon \,}r \land \chi_r(x) = t$ or $(\lnot x{\, \epsilon \,}r) \land \chi_r(x) = f$},$$ It follows that $\chi_r$ is the unique map $X \to {{\bf 2}}$ such that $x {\, \epsilon \,}r$ iff $\chi_r(x) =t$. Thus ${{\bf 1}}\to^t {{\bf 2}}$ classifies decidable relations. In case we take the axioms of CETCS with classical logic every relation is decidable, and hence ${{\bf 1}}\to^t {{\bf 2}}$ is a full subobject classifier for the category. In this case $\cal C$ is a topos. The Lawvere’s choice axiom (L5) states: If $f:A \to B$ is mapping and $A$ contains at least one element, then there is a mapping $g:B \to A$ so that $fgf=f$. In CETCS with classical logic (AC) and (L5) are equivalent. ETCS and CETCS +PEM + AC have the same theorems. Correspondence to Standard Categorical Formulations =================================================== \[lm61\] Let $\cal C$ be a cartesian category which satisfies (G). Then a pullback of an onto mapping is again an onto mapping. We recall some basic definitions from (Johnstone 2002): A sequence of mappings $A \to^e I \to^m B$ is an [image factorization]{} of $f:A \to B$ if $f \equiv m \circ e$ and $m$ is a mono, and whenever $f \equiv m' \circ e'$ where $m':I' \to B$ is mono then there is some $t:I \to I'$ with $m \equiv m' \circ t$. Such an $m$ is a called an [image of $f$]{}. A morphism $f$ is a [cover]{} if whenever it can be factored as $f \equiv m \circ g$ where $m$ mono, then $m$ be must an isomorphism. \[imfactcov\] In any category, if $A \to^e I {\to/ >->/}^i B$ is an image factorization of $f:A \to B$, then $e$ is a cover. \[ontomono\] Let $\cal C$ be a cartesian category satisfying (G). If $f: X \to Y$ is factored as $X \to^g I {\to/ >->/}^i Y$, where $g$ is onto, then it is an image factorization. [Proof.]{} Suppose therefore that $X \to^h J {\to/ >->/}^j Y$ is another factoring of $f$. It is sufficient to show that $i \le j$ as subobjects of $Y$. By Proposition \[elts2\] it is equivalent to prove $i {\, \dot{\subseteq} \,}j$. Suppose that $y \in Y$ satisfies $y {\, \epsilon \,}i $. Then $y= i t$ for some $t \in I$. Now $g$ is onto, so there is $x \in X$ with $g x = t$. Now $y= i g x = f x = j (h x)$. Hence $y {\, \epsilon \,}j$. Thus we have $i {\, \dot{\subseteq} \,}j$. ${\Box}$ \[lm66\] Suppose that $\cal C$ is a cartesian category that satisfies (G). Then - every onto mapping is a cover, - if $\cal C$ in addition satisfies (Fct), then every cover is onto. [Proof.]{} (a): If $f:A\to B$ is onto, then $A\to^f B \to ^{{{\sf id}}}B$ is an image factorization, so by Theorem \[ontomono\] and Proposition \[imfactcov\] $f$ is a cover. (b): Let $f:A\to B$ be a cover. By (Fct) take a factorization $A\to^e I \to^i B$ of $f$ where $e$ is onto and $i$ is mono. Now since $f$ is a cover, $i$ is an isomorphism. Hence $f$ is onto as well. ${\Box}$ In standard category-theoretic terms (Johnstone 2002) various combinations of the CETCS axioms can be characterized by the following theorems. First recall that a [regular category]{} is a category with finite limits, which has image factorization and where covers are preserved by pullbacks. \[regchar\] Let $\cal C$ be a cartesian category satisfying (G). Then $\cal C$ satisfies (Fct) if, and only if, $\cal C$ is a regular category where the terminal object is projective. [Proof.]{} ($\Rightarrow$) According to Theorem \[ontomono\] $\cal C$ has image factorizations. By Lemma \[lm66\] onto morphisms are the same as covers. Thus by Lemma \[lm61\] covers are preserved by pullbacks. This shows that $\cal C$ is regular. If $A \to {{\bf 1}}$ is a cover then it is onto, and hence ${{\bf 1}}$ is a choice object. Since the category is regular, it follows that ${{\bf 1}}$ is projective. ($\Leftarrow$) Suppose ${{\bf 1}}$ is projective. Hence any cover is onto. Thus by regularity, any morphism can be factored as an onto morphism followed by a mono. This gives (Fct). ${\Box}$ \[lccchar\] Let $\cal C$ be a cartesian category satisfying (G). Then $\cal C$ is locally cartesian closed if and only if $\cal C$ satisfies the axiom ($\Pi$). [Proof.]{} See Section 7. ${\Box}$ \[balance\] In a CETCS category $\cal C$ every epi is onto; consequently $\cal C$ is balanced. [Proof.]{} Let $f:A \to B$ be an epi. Form the sum ${{\bf 1}}\to^i S \to/<-/^j B$. Let $m: M {\to/ >->/}B$ be a subobject so that $y {\, \epsilon \,}m$ iff $(\exists x\in A) f x = y$. Then form the sum ${{\bf 1}}\to^r K \to/<-/^s M$ and let $k: K \to S$ be the unique mapping so that $k r = i$ and $k s = j m$. Define, using Section 5.1, an equivalence relation $(r_1,r_2) : R \to (S,S)$ by $$(u,z) {\, \epsilon \,}(r_1,r_2) \Longleftrightarrow ((\exists w \in K) kw=u \Leftrightarrow (\exists w \in K) kw=z).$$ By Proposition \[qprop\] let $q: S \to Q$ be such that $$(u,z) {\, \epsilon \,}(r_1,r_2) \Longleftrightarrow qu = qz.$$ Let $g:B \to S$ be given by $g = i \circ !_B$ and $h=j:B \to S$. It is straightforward to check that for all $x\in A$, $qgfx = qhfx$. Thus $qgf=qhf$, and since $f$ is epi, $qg=qh$. For each $y\in B$ we have, since $(\exists w \in K) kw=gy$ is true, that $$(\exists w \in K) kw = hy.$$ By (DP) and disjointness of sums we must have $w = st$ for some $t \in M$. Hence $jmt = kst = kw=hy = jy$. Since $j$ is mono, $mt=y$. Thus $y {\, \epsilon \,}m$, that is $(\exists x \in A) fx=y$. The last statement follows by Axiom (G). ${\Box}$ Let $\cal C$ be a category. Then $\cal C$ satisfies CETCS if, and only if, $\cal C$ has the following properties - it is locally cartesian closed, - it is a pretopos, - it has NNO, - its terminal object is projective and generates $\cal C$, - ${{\bf 0}}\not\cong {{\bf 1}}$, - it satisfies the disjunction property, - it has enough projectives. [Proof.]{} ($\Rightarrow$): (i) follows from Theorem \[lccchar\]. Properties (iii),(vi),(vii) are axioms of CETCS. (iv) follows from Theorem \[regchar\]. (v) is clear by (I). By (Lemma 1.5.13 –14, Johnstone 2002) every locally cartesian closed which is cocartesian and balanced (Lemma \[balance\]) is a pretopos. ($\Leftarrow$) It is known that in a locally cartesian closed pretopos with NNO has all coequalizers (Remark 2.8, Moerdijk and Palmgren 2000). Using Theorem \[lccchar\] we get axiom $(\Pi)$. Axioms (G), (NNO), (PA) and (DP) are given. (I) follows easily from (v) using uniqueness of mappings. In a pretopos the pullback object of $x$ and $y$ in a sum diagram ${{\bf 1}}\to^x S \to/<-/^y {{\bf 1}}$ will be ${{\bf 0}}$, so (NT) follows from (I). In pretopos every map can be factored as a cover followed by a mono. But using that ${{\bf 1}}$ is projective we can show that covers are onto, so (Fct) is verified. In a pretopos all equivalence relations are effective, so (Eff) follows. ${\Box}$ Functor-free Formulation of LCCCs ================================= The standard way (Johnstone 2002) of defining a locally cartesian category $\cal C$ is to say that it is a cartesian category so that pullbacks along a mapping $f:X \to Y$ induces a functor $f^: {\cal C}/Y \to {\cal C}/X$ and that this functor has a right adjoint $\Pi_f : {\cal C}/X \to {\cal C}/Y$. These functors must, in particular, be defined on the objects of the slice categories. This means that the pullback object must be possible to construct as a function of mappings $g: A \to Y$ and $f:X \to Y$. This can be forced if one assumes the full axiom of choice in the meta-theory of $\cal C$, but is not possible if we only use intuitionistic logic. Makkai (1996) has developed a theory of functors — anafunctors — by which one can avoid such uses of choice. In (Palmgren 2008) we showed how LCCCs could be formulated replacing $f^$ and $\Pi_f$ by the appropriate anafunctors, so that $\Pi_f$ is the right adjoint of $f^$. We here extract what is the existence condition for such $\Pi_f$ and formulate it without functors. Thus a functor-free formulation of LCCC will be given in Definition \[funfree\]. [A $\Pi$-diagram]{} for $Y \to^g X \to^f I$ is a commutative diagram of the form $$\label{pidiag} \bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}`g`] \square(550,0)[P`F`X`I;\pi_1`\pi_2`{{\varphi}}`f] \efig$$ where the square on the right is a pullback diagram. The object $F$ is called the [parameter object]{} of the diagram. If we have a second $\Pi$-diagram for $Y \to^g X \to^f I$ $$\label{pidiag1b} \bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}'`g`] \square(550,0)[P'`F'`X`I;\pi'_1`\pi'_2`{{\varphi}}'`f] \efig$$ we say that a mapping $t: F' \to F$ is a [$\Pi$-diagram morphism]{} from the second diagram to the first diagram if ${{\varphi}}t \equiv {{\varphi}}'$ and the unique map $s:P' \to P$ such that $\pi_2 s \equiv \pi'_2$ and $\pi_1 s \equiv t \pi'_1$ also satisfies ${{\rm ev}}s \equiv {{\rm ev}}'$. $$\label{pidm} \xymatrix{ && {P'} \ar@{->}[rr]\ar@{->}[dll] \ar@{->}[ddr] \ar@{.>}[drr] && {F'}\ar@{->}[ddr] \ar@{.>}[drr] && \\ {Y}\ar@{->}[drrr]&&&& {P} \ar@{->}[rr] \ar@{->}[llll] \ar@{->}[dl] && {F}\ar@{->}[dl] \\ &&& {X} \ar@{->}[rr] && {I} & \\ }$$ It is easily seen that the $\Pi$-diagrams and $\Pi$-diagram morphisms over fixed mappings $Y \to^g X \to^f I$ forms a category. A [universal $\Pi$-diagram for $Y \to^g X \to^f I$:]{} is a $\Pi$-diagram $$\label{pidiag2} \bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}`g`] \square(550,0)[P`F`X`I;\pi_1`\pi_2`{{\varphi}}`f] \efig$$ which is such that for any other $\Pi$-diagram $$\label{pidiag3} \bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}'`g`] \square(550,0)[P'`F'`X`I;\pi'_1`\pi'_2`{{\varphi}}'`f] \efig$$ there is a unique mapping $n: F' \to F$ so that ${{\varphi}}' \equiv {{\varphi}}n$ and that the unique mapping $m: P' \to P$, with $n \pi'_1 \equiv \pi_1 m$ and $\pi'_2 \equiv \pi_2 m$, satisfies ${{\rm ev}}' \equiv \,{{\rm ev}}\, m$. \[funfree\] [ A cartesian category is [locally cartesian closed,]{} if it satisfies the [generalized exponential axiom]{} or the [$\Pi$-axiom]{}: for every composable pair of maps $Y \to^g X \to^f I$ there is an universal exponential diagram as in (\[pidiag2\]). That is, the category of $\Pi$-diagrams over $Y \to^g X \to^f I$ has a terminal object.]{} Characterization of Universal $\Pi$-diagrams -------------------------------------------- We have the following characterization of $\Pi$-diagrams where the parameter object is $F=1$. \[lem21\] Consider a cartesian category satisfying (G). Let $Y \to^g X \to^f I$ be morphisms and let $i \in I$ be an element. For a pair of morphisms $\psi=(r_1,r_2):R \to (X,Y)$ the diagram $$\label{defpid01} \bfig \qtriangle/<-`->`/[Y``;r_2`g`] \square(550,0)[R`1`X`I;`r_1`i`f] \efig$$ is a $\Pi$-diagram if and only if - $\psi$ is a partial function (i.e. $r_1$ is mono) - $(\forall x \in X)[f x = i \Longrightarrow (\exists y \in Y)(x,y) {\, \epsilon \,}\psi]$ - $(\forall x \in X)(\forall y \in Y)[(x,y) {\, \epsilon \,}\psi \Longrightarrow f x =i \land g y =x]$ [Proof.]{} ($\Rightarrow$) Suppose (\[defpid01\]) is a $\Pi$-diagram. Since $i$ is mono, the pullback diagram entails that $r_1$ is mono. Hence $\psi$ is a partial function. Property (A2) follows by the pullback property. (A3) follows since the whole diagram is commutative. ($\Leftarrow$) Suppose that (A1) – (A3) are satisfied. By (A3) it follows that the entire diagram commutes. (A1) and (A2) together yields that the square is a pullback. ${\Box}$ \[lem22\] Consider two $\Pi$-diagrams in a cartesian category satisfying (G). $$\label{defpid02} \bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}'`g`] \square(550,0)[P'`F'`X`I;\pi_1'`\pi_2'`{{\varphi}}'`f] \qtriangle(1500,0)/<-`->`/[Y``;{{\rm ev}}`g`] \square(2050,0)[P`F`X`I;\pi_1`\pi_2`{{\varphi}}`f] \efig$$ Let $\chi : F' \to F$ be such that ${{\varphi}}\chi ={{\varphi}}'$. There is a unique $\kappa: P' \to P$ so that $\pi_1 \kappa = \chi \pi'_1$ and $\pi_2 \kappa = \pi'_2$. For this $\kappa$ it holds that ${{\rm ev}}\kappa = {{\rm ev}}'$ if and only if for all $v \in F'$, $x \in X$ and $y \in Y$ $$(v,x,y) {\, \epsilon \,}(\pi'_1,\pi'_2,{{\rm ev}}') \Longleftrightarrow (\chi v, x, y) {\, \epsilon \,}(\pi_1, \pi_2, {{\rm ev}}).$$ [Proof.]{} ($\Leftarrow$): Assume the equivalence. Let $t \in P'$ be arbitrary. We prove ${{\rm ev}}\kappa t = {{\rm ev}}' t$. Clearly $(\pi'_1 t,\pi'_2 t,{{\rm ev}}' t) {\, \epsilon \,}(\pi'_1,\pi'_2,{{\rm ev}}') $, so by the equivalence $(\chi \pi'_1 t,\pi'_2 t,{{\rm ev}}' t) {\, \epsilon \,}(\pi_1, \pi_2, {{\rm ev}})$. Thus there is a $u \in P$ with $\chi \pi'_1 t = \pi_1 u$, $\pi'_2 t = \pi_2 u$ and ${{\rm ev}}' t= {{\rm ev}}u$. Now we have $\pi_1 u = \chi \pi'_1 t = \pi_1 \kappa t $ and $\pi_2 u = \pi'_2 t = \pi_2 \kappa t$. By the pullback property, $\pi_1$ and $\pi_2$ are jointly mono, so $u= \kappa t$. Thus ${{\rm ev}}' t= {{\rm ev}}\kappa t$. ($\Rightarrow$): Assume ${{\rm ev}}\kappa= {{\rm ev}}'$. Suppose $(v,x,y) {\, \epsilon \,}(\pi'_1,\pi'_2,{{\rm ev}}') $. Thus for some $t \in P'$, it holds that $v= \pi'_1 t$, $x=\pi'_2 t$ and $y={{\rm ev}}' t$. Hence $x= \pi_2 \kappa t$, $y= {{\rm ev}}\kappa t$. Finally $\pi_1 \kappa =\chi \pi'_1$ gives $\chi v = \pi_1 \kappa t$, so that $(\chi v, x, y) {\, \epsilon \,}(\pi_1, \pi_2, {{\rm ev}})$. For the converse, assume $(\chi v, x, y) {\, \epsilon \,}(\pi_1, \pi_2, {{\rm ev}})$. Thus $\chi v= \pi_1 s$, $x=\pi_2 s$ and $y={{\rm ev}}s$ for some $s \in P$. Then $$f \pi_2 s = {{\varphi}}\pi_1 s = {{\varphi}}\chi v = {{\varphi}}' v.$$ Thus there is a unique $t \in P'$ with $\pi'_2 t = \pi_2 s$ and $\pi'_1 t=v$. We have then $\pi'_2 t=x$, so to prove $(v,x,y) {\, \epsilon \,}(\pi'_1,\pi'_2,{{\rm ev}}')$ it suffices to show $y={{\rm ev}}' t$. Now ${{\rm ev}}' t = {{\rm ev}}\kappa t$. We have $\pi_1 \kappa t = \chi \pi'_1 t = \chi v = \pi_1 s $ and $\pi_2 \kappa t = \pi'_2 t = \pi_2 s$. By the pullback property $\pi_1$ and $\pi_2$ are jointly mono, so $\kappa t =s$. Hence $y= {{\rm ev}}s = {{\rm ev}}\kappa t = {{\rm ev}}' t$ as desired. ${\Box}$ \[th74\] Let $\cal C$ be a cartesian category satisfying (G). Let $Y \to^g X \to^f I$ be fixed morphisms. Suppose that the $\Pi$-diagram $$\label{diag74} \bfig \qtriangle(0,0)/<-`->`/[Y``;{{\rm ev}}`g`] \square(550,0)[P`F`X`I;\pi_1`\pi_2`{{\varphi}}`f] \efig$$ is universal for $Y \to^g X \to^f I$. Then for every $i \in I$ and for every pair of morphisms $\psi=(r_1,r_2):R \to (X,Y)$ satisfying (A1) – (A3), there is a unique $v \in F$ with ${{\varphi}}v =i$ such that for all $x \in X$ and $y \in Y$ $$\label{equ74} (x,y) {\, \epsilon \,}\psi \Longleftrightarrow (v, x, y) {\, \epsilon \,}\alpha.$$ Here $\alpha=(\pi_1, \pi_2, {{\rm ev}}): P \to (F,X,Y)$. [Proof.]{} By Lemma \[lem21\] (\[defpid01\]) is a $\Pi$-diagram. Since (\[diag74\]) is a universal diagram, there is a map $v:1 \to F$ such that ${{\varphi}}v = i$ and for all $u \in 1$, $x \in X$ and $y\in Y$, $$(u,x,y) {\, \epsilon \,}(!,r_1,r_2) \Longleftrightarrow (vu,x,y) {\, \epsilon \,}\alpha$$ (by Lemma \[lem22\]). But $vu=v$ and $(u,x,y) {\, \epsilon \,}(!,r_1,r_2)$ is equivalent to $(x,y) {\, \epsilon \,}\psi$, so (\[equ74\]) is proved. ${\Box}$ There is a converse \[th75\] Let $\cal C$ be a cartesian category satisfying (G). Let $Y \to^g X \to^f I$ be fixed morphisms. Consider the $\Pi$-diagram $$\label{pidiag1} \bfig \qtriangle(0,0)/<-`->`/[Y``;{{\rm ev}}`g`] \square(550,0)[P`F`X`I;\pi_1`\pi_2`{{\varphi}}`f] \efig$$ and let $\alpha=(\pi_1, \pi_2, {{\rm ev}}): P \to (F,X,Y)$. Suppose that for every $i \in I$ and for every pair of morphisms $\psi=(r_1,r_2):R \to (X,Y)$ satisfying (A1) – (A3), there is a unique $v \in F$ with ${{\varphi}}v =i$ such that for all $x \in X$ and $y \in Y$ $$(x,y) {\, \epsilon \,}\psi \Longleftrightarrow (v, x, y) {\, \epsilon \,}\alpha.$$ Then (\[pidiag1\]) is universal for $Y \to^g X \to^f I$. [Proof.]{} Let $$\bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}'`g`] \square(550,0)[P'`F'`X`I;\pi'_1`\pi'_2`{{\varphi}}'`f] \efig$$ be an arbitrary $\Pi$-diagram. For $v' \in F'$ form the pullback $$\label{qpb1} \bfig \square(0,0)[Q`1`P`F';!`q`v'`\pi'_1] \efig$$ Then the composed diagram $$\label{qpb2} \bfig \qtriangle/<-`->`/[Y``;{{\rm ev}}'q`g`] \square(550,0)[Q`1`X`I;!`\pi'_2q`{{\varphi}}' v'`f] \efig$$ is, obviously, again a $\Pi$-diagram. For $x \in X$ and $y \in Y$ we then have $$(x,y) {\, \epsilon \,}(\pi'_2 q, {{\rm ev}}'q) \Longleftrightarrow (v',x,y) {\, \epsilon \,}(\pi'_1,\pi'_2,{{\rm ev}}').$$ Indeed, suppose $x= \pi'_2 q u$ and $y= {{\rm ev}}' q u$ for some $u\in Q$. We have by (\[qpb1\]) $v'= v' ! u = \pi'_1 q u$. Hence $(v',x,y) {\, \epsilon \,}(\pi'_1,\pi'_2,{{\rm ev}}')$. Conversely, suppose $v'= \pi'_1t$, $x= \pi'_2 t$ and $y= {{\rm ev}}' t$ for some $t \in P'$. From $v'= \pi'_1t$ it follows by (\[qpb1\]) that there is a unique $s \in Q$ with $t= q s$. Thus $x= \pi'_2 q s$ and $y= {{\rm ev}}' q s$ and hence $(x,y) {\, \epsilon \,}(\pi'_2 q, {{\rm ev}}'q)$. Now (\[qpb2\]) is a $\Pi$-diagram so $\psi = (\pi'_2 q, {{\rm ev}}'q): Q \to (X,Y)$ satisfies (A1) – (A3) for $i= {{\varphi}}' v'$ (by Lemma \[lem22\]). Hence by assumption we have that there is a unique $v \in F$ with ${{\varphi}}v= {{\varphi}}' v'$ and for all $x \in X$ and $y \in Y$ $$(v,x,y) {\, \epsilon \,}(\pi_1, \pi_2, {{\rm ev}}) \Longleftrightarrow (x,y) {\, \epsilon \,}\psi.$$ In conclusion, for every $v' \in F'$ there is a unique $v \in F$ such that ${{\varphi}}v= {{\varphi}}' v'$ and $$\label{eqv75} (\forall x \in X)(\forall y \in Y)[(v,x,y) {\, \epsilon \,}(\pi_1, \pi_2, {{\rm ev}}) \Longleftrightarrow (v',x,y) {\, \epsilon \,}(\pi'_1, \pi'_2, {{\rm ev}}')].$$ By unique choice (Proposition \[fungraph\]) and Theorem \[elemcomp\] there is $\chi:F' \to F$ so that for all $v'\in F$ it holds that ${{\varphi}}\chi v'= {{\varphi}}' v'$ and $$\label{star35} (\forall x \in X)(\forall y \in Y)[(\chi v',x,y) {\, \epsilon \,}(\pi_1, \pi_2, {{\rm ev}}) \Longleftrightarrow (v',x,y) {\, \epsilon \,}(\pi'_1, \pi'_2, {{\rm ev}}')].$$ Hence according to Lemma \[lem22\] the unique map $\kappa:P' \to P$ satisfying $\pi_1 \kappa = \chi \pi'_1$ and $\pi_2 \kappa = \pi'_2$ also satisfies ${{\rm ev}}\kappa = {{\rm ev}}'$. To finish the proof we have to show that $\chi$ is unique. Suppose that $\theta: F' \to F$ satisfies ${{\varphi}}\theta = {{\varphi}}'$ and that $\lambda:P' \to P$ is the unique map with $\pi_1 \lambda = \theta \pi'_1$ and $\pi_2 \lambda = \pi'_2$, and that this $\lambda$ satisfies ${{\rm ev}}\lambda = {{\rm ev}}'$. Then by Lemma \[lem22\], it holds for all $v' \in F$ $$(\forall x \in X)(\forall y \in Y)[(\theta v',x,y) {\, \epsilon \,}(\pi_1, \pi_2, {{\rm ev}}) \Longleftrightarrow (v',x,y) {\, \epsilon \,}(\pi'_1, \pi'_2, {{\rm ev}}')].$$ Thus by the uniqueness in (\[eqv75\]) for all $v' \in F'$, $\theta v' = \chi v'$. Hence $\theta = \chi$. ${\Box}$ [(Theorem \[lccchar\])]{} Let $\cal C$ be a cartesian category satisfying (G). Then $\cal C$ is locally cartesian closed if and only if $\cal C$ satisfies the axiom ($\Pi$). [Proof.]{} ($\Rightarrow$) This is Theorem \[th74\]. ($\Leftarrow$) Suppose axiom ($\Pi$) holds. By Theorem \[th75\] this says that every $Y \to^g X \to^f I$ has a universal $\Pi$-diagram. ${\Box}$ [99]{} P. Aczel. The type theoretic interpretation of constructive set theory. [Logic Colloquium ’77 (Proc. Conf., Wroclaw, 1977),]{} pp. 55–66, Stud. Logic Foundations Math., 96, North-Holland, Amsterdam-New York, 1978. S. Awodey and M.A. Warren. Predicative algebraic set theory. [Theory and Applications of Categories]{} 15 (2005/06), No. 1, pp.1 – 39. B. van den Berg. Inductive types and exact completions. [Annals of Pure and Applied Logic]{} 134 (2005), pp. 95 – 121. B. van den Berg and I. Moerdijk. Aspects of predicative Algebraic Set Theory I: Exact completion. [Annals of Pure and Applied Logic]{} 156 (2008), pp. 123 – 159. E. Bishop. [Foundations of Constructive Analysis.]{} McGraw-Hill 1967. E. Bishop. Mathematics as a numerical language. In: [Intuitionism and Proof Theory.]{} North-Holland, Amsterdam 1970, pp. 53 – 71. E. Bishop. [Compiling Mathematics into Algol.]{} Unpublished text for a seminar. 1970. M. Hofmann. On the interpretation of type theory in locally cartesian closed categories. [Proc. Computer Science Logic ’94, Kazimierz, Poland.]{} Jerzy Tiuryn and Leszek Pacholski, eds. Lecture Notes in Computer Science, Vol. 933, Springer 1994. P.T. Johnstone. [Sketches of an Elephant: A Topos Theory Compendium,]{} vol 1–2. Oxford University Press 2002. A. Joyal and I. Moerdijk. [Algebraic Set Theory.]{} Cambridge University Press 1995. F.W. Lawvere. An Elementary Theory of the Category of Sets (Long Version).[Theory and Applications of Categories]{} 11 (2005), 7 – 35. F.W. Lawvere and R. Rosebrugh. [Sets for Mathematics.]{} Cambridge University Press 2003. S. Mac Lane. [Categories for the Working Mathematician.]{} 2nd edition. Springer 1998. M.E. Maietti. Modular correspondence between dependent type theories and categories including pretopoi and topoi. [Math. Structures Comput. Sci.]{} 15 (2005), no. 6, 1089–1149. M.E. Maietti and G. Sambin. Toward a minimalist foundation for constructive mathematics. In: L. Crosilla and P. Schuster (eds.) [From sets and types to topology and analysis]{}, 91–114, Oxford Logic Guides, 48, Oxford Univ. Press, Oxford, 2005. P. Martin-L[ö]{}f. An intuitionistic theory of types: predicative part. In: H.E. Rose and J. Shepherdson (eds.) [Logic Colloquium ’73,]{} North-Holland, Amsterdam, 1975. C. McLarty. Exploring categorical structuralism. [Philosophia Mathematica]{} 12(2004), 37 – 53. C. McLarty. ETCS and Philosophy of Mathematics. [Theory and Applications of Categories]{} 11 (2005), 2 – 4. M. Makkai. Avoiding the axiom of choice in general category theory. [Journal of Pure and Applied Algebra]{} 108 (1996), 109 – 173. I. Moerdijk and E. Palmgren. Well-founded trees in categories. [Annals of Pure and Applied Logic]{} 104 (2000), 189 – 218. I. Moerdijk and E. Palmgren. Type Theories, Toposes and Constructive Set Theory: Predicative Aspects of AST. [Annals of Pure and Applied Logic]{} 114 (2002), 155 – 201. J. Myhill. Constructive set theory. [Journal of Symbolic Logic]{} 40 (1975), no. 3, 347 – 382. G. Osius. Categorical set theory: a characterization of the category of sets. [Journal of Pure and Applied Algebra]{} 4 (1974), 79 – 119. E. Palmgren. Locally cartesian closed categories without chosen constructions. [Theory and Applications of Category Theory]{} 20 (2008), 5 – 17. M. A. Shulman. Stack semantics and the comparison of material and structural set theories. ArXiv:1004.3802v1. April 2010. W.W. Tait. Cantor’s Grundlagen and the paradoxes of set theory. In: [Between Logic and Intuition: Essays in Honor of Charles Parsons]{} (ed. G. Sher and R. Tieszen), 269 – 290. Cambridge University Press 2000. (Also reprinted in W.W. Tait, [The Provenance of Pure Reason: Essays in the Philosophy of Mathematics and its History.*]{} Oxford University Press 2005.) Erik Palmgren Swedish Collegium for Advance Study, Uppsala, and Institut Mittag-Leffler, Djursholm, and Department of Mathematics, Uppsala University, Sweden. [^1]: Research supported by Swedish Collegium for Advanced Study, Uppsala, Swedish Research Council (VR) and Institut Mittag-Leffler, Djursholm, Sweden.	{ "pile_set_name": "ArXiv" }
--- abstract: 'The Jovian dust streams are high-speed bursts of submicron-sized particles traveling in the same direction from a source in the Jovian system. Since their discovery in 1992, they have been observed by three spacecraft: Ulysses, Galileo and Cassini. The source of the Jovian dust streams is dust from Io’s volcanoes. The charged and traveling dust stream particles have particular signatures in frequency space and in real space. The frequency-transformed Galileo dust stream measurements show different signatures, varying orbit-to-orbit during Galileo’s first 29 orbits around Jupiter. Time-frequency analysis demonstrates that Io is a localized source of charged dust particles. Aspects of the particles’ dynamics can be seen in the December 2000 joint Galileo-Cassini dust stream measurements. To match the travel times, the smallest dust particles could have the following range of parameters: radius: 6 nm, density: 1.35–1.75 g/, sulfur charging conditions, which produce dust stream speeds: 220$\backslash$450 (Galileo$\backslash$Cassini) and charge potentials: 5.5$\backslash$6.3 V (Galileo$\backslash$Cassini).' author: - Amara Graps - Eberhard Grün - Harald Krüger - 'Mihá[ł]{}y Horányi' - 'H[å]{}kan Svedhem' bibliography: - 'amthesis.bib' title: Io Revealed in the Jovian Dust Streams --- Overview ======== The Jovian dust streams are high-speed collimated streams of submicron-sized particles traveling in the same direction from a source in the Jovian system. They were discovered in March 1992 by the cosmic dust detector instrument onboard the Ulysses spacecraft, when the spacecraft was just past its closest approach to Jupiter. Observations of the Jovian dust stream phenomena continued in the next nine years. A second spacecraft, Galileo, now in orbit around Jupiter, is equipped with an identical dust detector instrument to Ulysses’ dust instrument. Before and since the Galileo spacecraft’s arrival in the Jupiter system in December 1995, investigators recorded more dust stream observations. In July and August 2000, a third spacecraft with a dust detector (combined with a chemical analyzer), Cassini, traveling on its way to Saturn, recorded more high-speed streams of submicron-sized particles from the Jovian system. The many years-long successful Jovian dust streams observations reached a pinnacle on December 30, 2000, when both the Cassini and Galileo dust detectors accomplished a coordinated set of measurements of the Jovian dust streams inside and outside of Jupiter’s magnetosphere. Indirect methods applied by previous researchers have pointed to Io being the simplest explanation for the question of the origin of the Jovian dust streams. We first show by methods that Io is the source of the Jovian dust streams. To address the issue of identifying Io directly in the Galileo dust detector data, we apply time-frequency analysis, in particular, Fourier methods, to the Galileo dust data. Additional frequency signatures, such as amplitude modulation, also emerge from the time-frequency analysis. The second part of this paper focuses on the dust streams dynamics. Here, we apply a detailed Jovian particles and fields model to simulate a dust stream particle’s trajectory as the particle moves from Io’s orbit through Jupiter’s magnetosphere and beyond. Through the model, we show one possible set of parameters that match the travel times seen in the December 30, 2000 Galileo-Cassini joint dust stream measurements. Io’s Frequency Fingerprint ========================== In order to find Io’s frequency ‘fingerprint’ in the Galileo dust detector data, we followed the following steps: 1) we transformed the Galileo dust detector data into frequency space via periodograms, 2) we noted frequency patterns such as amplitude modulations, 3) we compared the frequency-transformed Galileo data with synthetic data, and, 4) we noted spacecraft effects such as Doppler shifts. Time-Frequency Analysis via Periodograms ---------------------------------------- The “classic" or Schuster periodogram [@Brett:88] is conventionally defined as the modulus-squared of the discrete Fourier transform. If the input time series contains a periodic feature, then the periodogram can be calculated for any frequency and it displays the presence of a sinusoid near one frequency value as a distinct peak in the spectrum. The Lomb-Scargle periodogram applied here is a slightly modified version of the classic periodogram giving a simpler statistical behavior [@Scargle:82]. Our best Galileo dust dataset for detecting Io’s frequency fingerprint emerged from the earlier Galileo orbits around Jupiter, because the spacecraft orbital geometry of the first years of the Galileo mission favored higher fluxes of dust stream particles. Therefore, by combining two years of the early data, we gained a higher signal-to-noise dataset. In Fig. \[periodogram\], we show a Lomb-Scargle periodogram for the first two years, 1996-1997, of Galileo dust impact rate data. This particular periodogram is the best example from the Galileo dust detector data showing, with high confidence: Io’s frequency signature, Jupiter’s frequency signatures, and amplitude modulation effects. The periodogram shows the following frequency signatures. 1. A strong peak near the origin, 2. An asymmetric peak: maximum at 0.6 day$^{-1}$, center at 0.7 $\pm$ 0.2 day$^{-1}$, 3. An asymmetric peak: center at 1.7 $\pm$ 0.2 day$^{-1}$, 4. A tall peak: center at 2.4 $\pm$ 0.1 day$^{-1}$, 5. A peak: center at 3.1 $\pm$ 0.1 day$^{-1}$, 6. Harmonics of the previous three peaks, and 7. Progressively smaller and less-defined peaks. Amplitude Modulation in Frequency Space --------------------------------------- Frequencies in frequency space can interact in numerous ways. We interpret the frequency peaks, seen in Fig. \[periodogram\], to be the result of Io’s frequency of orbital rotation, Jupiter’s magnetic field frequency of rotation, and an interaction between these two frequencies called amplitude modulation (AM). The simplest case of AM is a sinusoid modulating the amplitude of a carrier signal, which is itself a sinusoid. Then the carrier signal is broken down in frequency space into several sinusoidal oscillations: $x\approx \sin (\omega _0t)+\sin (\omega _0t)\sin (\Omega t)$, which can be converted to sums of frequencies using a trigonometric identity for sine products. The result is a signature in frequency space that displays a carrier frequency: $\omega_0$ with side frequencies (“modulation products”): ($\omega_0+\Omega$) and ($\omega_0-\Omega$). The process of amplitude modulation applies to Io’s orbital and Jupiter’s rotational frequencies in the following way. Jupiter’s rotation period is 9.8 hours corresponding to a frequency of 2.4 rotations per day. Io’s orbital period (and rotation period) is 1.8 days, corresponding to a frequency of 0.6 rotations per day. If the dust originates from Io, and the dust flux is modulated by Jupiter, then the spectrum in frequency space would appear like the spectrum in Fig. \[periodogram\], where the modulation products (sidelobes) at Jupiter’s frequency at full and half-rotations are due to Jupiter’s frequency modulating Io’s frequency of orbital rotation. The frequency difference between Jupiter’s rotational frequency and each of the sidelobes is the same frequency as Io’s frequency of orbital rotation. In addition, if one or both of the original signals have broad spectra, then the spectrum of the modulation products will be broadened to the same extent (since the modulation process is linear). Therefore, the spread of Io’s peak is repeated in the same way for the sidelobes of Jupiter’s frequency peaks (full and half-rotation). Modulation products appeared in about 20% of the first 29 frequency-transformed Galileo orbits. Galileo’s orbit E4 was the first orbit for which a hint of the amplitude modulation appeared, then in orbit G8, the signature was unmistakable. Other Galileo spacecraft orbits for which one can clearly see the modulation products are: C10, E18, and G29. Periodograms of each of the individual orbits can be seen in @Graps:2001. Synthesizing the Frequency-transformed Dust Stream Data ------------------------------------------------------- The main physical processes behind the frequency-transformed data can become clearer when one synthesizes data and compares the synthesized data to the real data. We have synthesized a spectrum, using typical periods from Galileo’s dust impact rate data, a Jupiter rotation and half rotation period, and an Io orbital rotational period, which show in the real data. We added some Gaussian-distributed noise, and after transforming the synthesized data into frequency space with an FFT, frequency peaks appear with their modulation products, which are in the same locations as those in the 1996-1997 Galileo dust detector periodogram. Figures of the synthesized time series and frequency-transformed data can be seen in Fig. \[synthetic\]. Spacecraft Effects in Frequency Space ------------------------------------- Several Galileo spacecraft orbital characteristics can be identified in the frequency-transformed data. The first effect is at the origin. For each orbit, the dust instrument receives more dust impacts while in the inner Jovian system than while in the outer Jovian system, which in frequency space, results in a peak at the origin. The second spacecraft effect is a Doppler effect between Galileo and Io, as the “observer" and the “source." In frequency space, the result is that both the Io and the Jupiter peaks can be smeared by Doppler shifts. The Io frequency peak Doppler shift is to shorter periods. This asymmetry appears in the modulation products, as well. A table of the Doppler shift trends can be seen in @Graps:2001. August-September 2000 Dust Storm -------------------------------- In August and September 2000, both Cassini (travelling by at $\sim$1 A.U. from Jupiter), and Galileo (in an orbit carried to $\sim$250 from Jupiter, where =7.134 10$^{9}$ cm is the equatorial radius of Jupiter) detected increases in the rate of dust impacts, which were approximately 100 times their nominal impact rates. In the frequency-transformed data from both spacecraft, Io’s frequency signature swamped all other frequency signatures in the Galileo data, which was noteworthy, because both spacecraft were located far from Jupiter, outside of Jupiter’s magnetosphere. Galileo detected the ‘dust storm’ earlier during the perijove portion of its G28 orbit, during Days $\sim$218–240. After Day 240, Galileo’s impact rate decreased, however, Cassini observed high impact rates particularly on $\sim$Day 251 and $\sim$Day 266. (See the periodogram of the dust impact rate in Fig. \[storm\]). Cassini’s dust detector’s observational geometry is very different from Galileo’s, however, by accident, Cassini captured the dust storm approximately 1-2 weeks after Galileo detected the dust storm. Conclusions of Io’s Frequency Fingerprint ----------------------------------------- Frequency analysis via Fourier techniques of the Galileo dust data provides our first direct evidence of an Io dust source. The presence of Io’s rotation frequency argues that Io is a localized source of charged dust particles because charged dust from diffuse sources would couple to Jupiter’s magnetic field and appear in frequency space with Jupiter’s rotation frequency and its harmonics. A confirmation of Io’s role as a localized charged dust source arises through the modulation effects. The Galileo dust detector periodogram data shows variabilities, orbit-to-orbit, even for some orbits which share similar orbital geometry. This orbit-to-orbit variability is a clue that, either the intervening medium or Io itself, i.e. its volcanoes, is a source of the variability. Jovian Dust Stream Dynamics =========================== To describe the dynamics of the dust streams, we apply a detailed Jovian particles and fields model to simulate a dust stream particle’s trajectory as the particle moves from Io’s orbit through Jupiter’s magnetosphere and beyond. For the model, one needs to assume approximations for the following: 1) Jupiter’s magnetic field, 2) Jupiter’s plasma, 3) dust particle density, 4) dust particle optical property, 5) charging processes, and 6) forces. Jovian Dust Stream Model Details -------------------------------- We approximate Jupiter’s magnetic field by implementing Connerney’s $\chem O_4$ or $\chem O_6$ model [@Conn:81a; @Conn:93], which is a quadrupole expansion of the planet’s internal field. Additionally, we hinge Connerney’s current sheet described in @Conn:81b, to approximate Jupiter’s magnetodisk. Connerney’s sheet implementation considers the magnetodisk as a perturbation to Jupiter’s internal field. Jupiter’s plasma is approximated using a plasma model which is a fit to the Voyager 1 and 2 cold plasma measurements, described in @Bag:89. We assumed a constant mixing ratio of 50% between single ionized oxygen and sulfur ions. For the dust particle density, we use density values 1.35–2.0 g/. The dust particle’s optical properties are manifested via $Q_{pr}$, which affects the particle’s dynamics through the radiation pressure force. For this work, a $Q_{pr}$ value for the dust particle is calculated based on the particle’s size, and following the curve in Burns et al.’s classic paper: @Burns:79. The charge of the dust particles is approximated by summing over the currents: photoelectron emission, ion and electron collection, and secondary electron emission. The charge of the particle, which varies in time, is integrated simultaneously with the particle’s acceleration. Here, we model the acceleration by considering the following forces: Jupiter’s gravitational force, the light pressure force, the Lorentz force, and the solar gravitational force [@Hor:97]. We have neglected the neutral gas and plasma (Coulomb) drag forces on the dust particle because the time-scales used in these runs are short (hours to a day), compared to the time-scales over which those forces have an effect. More details of the Jovian dust stream model can be found in M. Horányi’s papers, such as @Hor:97, and in @Graps:2001. Dynamics Results ---------------- A window of particle sizes exists for which dust particles can escape from traveling in Keplerian orbits in Jupiter’s magnetosphere. For small dust particles, their motion is as plasma ions and electrons, which gyrate about Jupiter’s magnetic field lines. For large dust particles, their motion is governed by gravity. From Io’s orbital location, the window of dust particle sizes for escaping particles is approximately radius 5 nm to 35 nm. This particle size is strongly dependent on the charging assumptions, especially the secondary electron emission material assumption. In our numerical experiments thus far, we found that smaller dust particles can be ejected when the impacting ion or electron energy is lower than the energy of other impacting ions or electrons in the secondary electron emission process. For example, a particle with sulfur properties can be ejected from Io’s location with a minimum size 4 nm, versus a particle with silicate material properties, which can be ejected from Io’s location with a minimum size 6 nm. As the dust stream particle moves through the Jovian magnetosphere, equilibrium potential is rarely reached. Therefore, as the particle moves outward, it continues to collect charges, which further accelerate the particle. We now think that the Jovian dust stream particles move faster (at least 400 km/sec) than previously assumed in the earlier work presented by @Zook:96, which suggested dust speeds, at least 200 km/sec. Joint Measurements: December 2000 from Galileo & Cassini -------------------------------------------------------- On December 30, 2000, the Cassini spacecraft closely flew by Jupiter, providing a simultaneous two-spacecraft measurement (Cassini-Galileo) of particles from one collimated stream from the Jovian dust streams. Particles in a stream were detected by Galileo, as the spacecraft was orbiting inside of the Jovian magnetosphere close to Ganymede (8–12 Jovian radii), and then particles in the stream traveled to Cassini, as Cassini flew by Jupiter at approximately 140 Jovian radii. Figure \[joint\] shows the dust impact rate data for the dual dust stream measurements, the gold line denotes the Galileo rates and the green line indicates the Cassini rates. We assumed that the same dust stream at each spacecraft began where the black horizontal line marks the midpoint of the peak rise in impact rate. The travel time between the two black-marked peaks is approximately 7 hours. One goal of the dynamical modeling was to match this travel time. From preliminary modeling, Fig. \[phase100\] shows the result of one possible trajectory of a Jovian dust streams particle released near Io’s orbit. Here, the smallest dust particles could have the following range of parameters: size: 6 nanometers, density: 1.35–1.75 g/, initial charge potential: 1–4 V, secondary electron emission yield: 3.0, dependent on a maximum electron energy 300 eV, and a photoelectron emission yield: 0.1–1.0, which produce dust particle speeds: 220$\backslash$450 (Galileo$\backslash$Cassini) and charge potentials: 5.5$\backslash$6.3 V (Galileo$\backslash$Cassini). Smaller and larger particles than 6 nanometers result in the wrong direction towards Cassini, and with travel times that are either too fast or too slow. Synopsis ======== Our work from frequency analysis shows Io as the dominant source of the Jovian dust streams. The variability seen in the frequency analysis shows that we might be able to use dust stream measurements to monitor Io’s volcanoes’ plume activity. Our charging and dynamics modeling, (more details in @Graps:2001), shows that the dust streams’ equilibrium potential is rarely reached in the Jovian magnetosphere, and that the Jovian dust stream particles travel faster than found previously in @Zook:96. In our preliminary analysis and modeling of the Galileo-Cassini dust stream measurements, we show that one set of conditions, which can match the travel times, gives a dust streams speed at Galileo of 220 , with a charge potential: 5.5 V, and a dust streams speed at Cassini of 450 , with a charge potential: 6.3 V. Acknowledgements {#acknowledgements .unnumbered} ================ The authors gratefully acknowledge the hard work of the Galileo and Cassini Dust Science Teams. Funding provided by the Deutsches Zentrum für Luft-und Raumfahrt E.V. (DLR), and the Deutsche Forschungsgemeinschaft (DFG).	{ "pile_set_name": "ArXiv" }
--- author: - 'Y. Wang' - 'F. Primas' - 'C.Charbonnel' - 'M. Van der Swaelmen' - 'G. Bono' - 'W. Chantereau' - 'G. Zhao' subtitle: 'II. Analysis and results of NGC104, NGC6121, and NGC6809 [^1] [^2]' title: Sodium abundances of AGB and RGB stars in Galactic globular clusters --- [We investigate the Na abundance distribution of asymptotic giant branch (AGB) stars in Galactic globular clusters (GCs) and its possible dependence on GC global properties, especially age and metallicity.]{} [We analyze high-resolution spectra of a large sample of AGB and red giant branch (RGB) stars in the Galactic GCs NGC104, NGC6121, and NGC6809 obtained with FLAMES/GIRAFFE at ESO/VLT, and determine their Na abundances. This is the first time that the AGB stars in NGC6809 are targeted. Moreover, to investigate the dependence of AGB Na abundance dispersion on GC parameters, we compare the AGB \[Na/H\] distributions of a total of nine GCs, with five determined by ourselves with homogeneous method and four from literature, covering a wide range of GC parameters.]{} [NGC104 and NGC6809 have comparable AGB and RGB Na abundance distributions revealed by the K$-$S test, while NGC6121 shows a lack of very Na-rich AGB stars. By analyzing all nine GCs, we find that the Na abundances and multiple populations of AGB stars form complex picture. In some GCs, AGB stars have similar Na abundances and/or second-population fractions as their RGB counterparts, while some GCs do not have Na-rich second-population AGB stars, and various cases exist between the two extremes. In addition, the fitted relations between fractions of the AGB second population and GC global parameters show that the AGB second-population fraction slightly anticorrelates with GC central concentration, while no robust dependency can be confirmed with other GC parameters.]{} [Current data roughly support the prediction of the fast-rotating massive star (FRMS) scenario. However, considering the weak observational and theoretical trends where scatter and exceptions exist, the fraction of second-population AGB stars can be affected by more than one or two factors, and may even be a result of stochasticity.]{} Introduction {#section:introduction} ============ Galactic globular clusters (GCs) have been the subject of a variety of stellar evolution studies; first, because for a long time they were thought to consist of a single stellar population (i.e., coeval and sharing the same initial chemical properties) thus making them the ideal stellar laboratory, and second, because of their more recently discovered intriguing complexity of being inhabited by multiple stellar populations, a feature that has turned out to be common to most Galactic globular clusters. This multiplicity has been identified based on the appearance of multimodal sequences in different regions (e.g., main sequence (MS), sub-giant branch (SGB), red giant branch (RGB) and horizontal branch (HB)) of GC color-magnitude diagrams (CMD; e.g., [@Piotto2012; @Milone2012b; @Piotto2015; @Milone2015a; @Milone2015b; @Nardiello2015b]) that were associated to the variations in He and light element (e.g., C, N, and O) abundances in their initial chemical composition (see e.g., [@Milone2012b; @Chantereau2015] and references therein). With the advent of multi-object spectrographs mounted on 8$-$10m-class telescopes, detailed chemical abundance analyses have also uncovered specific features $-$ elemental (anti-)correlations $-$ between the light element pairs C$-$N, O$-$Na, Mg$-$Al (e.g., [@Carretta2016] for a recent review). These are commonly interpreted as a signature of the existence of at least two stellar populations: a first-population (1P) of GC stars displaying Na and O abundances consistent with that of halo field stars of similar metallicity; and a second-population (2P) of GC stars characterized by Na overabundances and O deficiencies. Although the O$-$Na pair is probably the most documented one in terms of data, a similar picture is also derived from the other pairs, Mg$-$Al (e.g., [@Carretta2014a; @Carretta2014b]) and C$-$N (e.g., [@Carretta2005; @Pancino2010]). A wealth of observational data has been collected and analyzed for a respectable number of Galactic GCs at different evolutionary phases, for example, from MS and SGB to RGB and HB (see [@Wang2016], here after PaperI, for a more detailed summary). However, asymptotic giant branch (AGB) stars have rarely been targeted in a systematic way until very recently, due to their paucity in GCs (a result of their short lifetime) and inefficient RGB/AGB separation criteria. Recently, several studies have focused on GC AGB stars, mainly stimulated by the claim by @Campbell2013 [hereafter C13] that no Na-rich 2P AGB stars exists in NGC6752. This striking finding was challenged by @Lapenna2016 who re-observed the 20 AGB stars of the C13 sample at higher resolution with ESO-VLT/UVES, and found that both 1P and 2P stars populate the AGB of NGC 6752, with only stars with extreme Na enhancement missing. The presence of 2P AGB stars in this GC is also supported by @Gruyters2017 who claimed a photometric split on the AGB sequence using Str[ö]{}mgrem photometry. Other GC AGB stars have also been scrutinized. @Johnson2015 studied 35 AGB stars in 47Tuc (NGC104) and found that the AGB and RGB samples of 47Tuc have nearly identical \[Na/Fe\] dispersions, with only a small fraction ($\lesssim 20\%$) of Na-rich stars that may fail to ascend the AGB. @GarciaHernandez2015 showed that 2P AGB stars exist in metal-poor GCs with a study of Mg and Al abundances in 44 AGB stars from four metal-poor GCs (M13, M5, M3 and M2). In PaperI, we looked at NGC2808 and also found that its AGB and RGB stars share similar Na abundance dispersions. Moreover, we found more Na-rich 2P stars in the AGB sample than in the RGB one. The multiple populations in AGB stars in NGC2808 was also confirmed by @Marino2017 who carried out a study combining spectroscopy and photometry. They also looked at NGC6121 (M4) and found it hosts two main populations in agreement with the finding by @Lardo2017 that AGB stars show broadened distribution in close analogy with their RGB counterparts in the $C_{UBI}$[^3]$- V$ diagram. We note, however, that their conclusion on NGC6121 contradicts the result of @MacLean2016 who found that the AGB is populated by Na-poor and O-rich stars (from the analysis of 15 AGB and 106 RGB stars). NGC6266 (M62) was also found to have only 1P AGB stars by @Lapenna2015, but their conclusion may be affected by the small number statistics of their sample (6 AGB and 13 RGB stars). It is now largely accepted that GCs experienced self-enrichment during their early evolution, and that 2P stars formed out of the Na-rich, O-poor ashes of high-temperature-burning hydrogen ejected by more massive 1P stars and diluted with interstellar gas (e.g., [@PrantzosCharbonnel2006; @Prantzos2007]). However, the nature of the polluters remains highly debated, as well as the mode and timeline of the formation of 2P stars. Among the most commonly-invoked scenarios, one finds fast-rotating massive stars (FRMS, with initial masses above $\sim25\,\mathrm{M_{\sun}}$; [@MaederMeynet2006; @PrantzosCharbonnel2006; @Decressin2007b; @Decressin2007a; @Krause2013]), massive AGB stars (with initial masses of $\sim6-11\,\mathrm{M_{\sun}}$; [@Ventura2001; @DErcole2010; @Ventura2011; @Ventura2013]) and supermassive stars ($\sim10^{4}\,\mathrm{M_{\sun}}$; [@Denissenkov2014; @Denissenkov2015]). Other possible polluters have also been explored, like massive stars in close binaries ($10-20\,\mathrm{M_{\sun}}$; [@deMink2009; @Izzard2013]), FRMS paired with AGB stars [@Sills2010] or with high-mass interactive binaries [@Bastian2013; @Cassisi2014a]. So far, none of the proposed scenarios have been able to reconcile all aspects of the formation and evolution of GCs with the spectroscopic and photometric complexity exhibited by these systems, nor with the new constraints coming from extragalactic young massive clusters that have masses similar to the initial mass postulated for GCs within the self-enrichment framework (e.g., [@Bastian2015; @Renzini2015; @Krause2016; @Charbonnel2016EAS]). One key feature to pay attention to is how the various scenarios differ from one another. The origin and amount of He enrichment predicted for 2P stars is one such example, which has important consequences on the way the various sequences of the CMDs can be populated (e.g., [@DAntona2010; @Chantereau2015; @Chantereau2016]). Interestingly enough, the different theoretical predictions for the coupling between He and Na enrichments in the initial composition of 2P stars are expected to differentially affect the extent of the Na dispersion today among RGB and AGB stars in individual GCs, in proportions that depend on their age and metallicity [@CCWC2016]. In the original FRMS framework, 2P low-mass stars are predicted to be born with large and correlated spreads in both He and Na abundances [@Decressin2007b]. Since the lifetime and the fate of stars strongly depend on their initial He content, the FRMS scenario predicts that, above a certain threshold, or cutoff, of initial He and Na abundance, 2P stars do miss the AGB (so-called AGB-manqué) and evolve directly towards the white dwarf stage after central He burning [@Charbonnel2013; @Chantereau2015]. This provides in principle a nice explanation for the lack of Na-rich AGB stars observed in NGC6752 by C13. @CCWC2016 have also shown that within the original FRMS scenario, the maximum Na content expected on the AGB is a (weak) function of both the metallicity and the age of GCs. Namely, at a given metallicity, younger clusters are expected to host AGB stars exhibiting a larger Na spread than older clusters, and at a given age, higher Na dispersion along the AGB is predicted in metal-poor GCs than in the metal-rich ones. Additionally, mass loss along the RGB has been shown to strongly impact the evolution of low-mass stars on the AGB, and therefore to modify the theoretical Na cut on the AGB ([@CCWC2016]; see also [@Cassisi2014b]): the higher the mass loss, the stronger the trends with age and metallicity. However, the situation might be much more complex, as revealed by the derivation of the helium variations between 1P and 2P stars in several GCs by multiwavelength photometry of multiple sequences, which turn out to be much lower than predicted by both the original FRMS and AGB scenario [@Anderson2009; @diCriscienzo2010; @Pasquini2011; @Milone2012; @Milone2012b; @Milone2013; @Piotto2013; @Marino2014; @Larsen2015; @Milone2015b; @Nardiello2015a; @Nardiello2015b]. It is therefore fundamental to test model predictions with observations of AGB and RGB stars in GCs spanning a large range in age and metal content. This is necessary to probe the degree of stochasticity lying behind the broad variety of chemical patterns observed in GCs (e.g., [@Bastian2015]). Considering the current limited sample of GC AGB stars with accurately determined Na (and O) abundances, we carried out a systematic observational campaign of four GCs (NGC104, NGC2808, NGC6121 and NGC6809). We have already presented our first results of NGC2808 in PaperI. Here, we report and discuss our results of the other three GCs, investigating whether the presence of Na-rich stars on AGB is dependent on metallicity and/or other GC parameters. The paper is organized as follows. In Sects. \[section:obsreduction\] and \[section:stellarparametersabundances\], we describe the observations and detail the analysis of the data for our sample of GCs. In Sect. \[section:OtherGCs\], we present the re-analysis of C13 data for NGC6752 and we show other four GCs from the literature. In Sect. \[section:discussion\], we compare the behavior of Na along the AGB and the RGB for the full GC sample (ours plus literature); we discuss also the possible correlations between the corresponding fractions of 1P and 2P stars and the GC global properties, and compare with the theoretical predictions of the original FRMS scenario. A summary and concluding remarks close the paper in Sect. \[section:summary\]. -------------- --------- ------- ----------------- ------------------------------- ------------------------------- ------------------------------- -- Instrument Setup R $\lambda$-range Exp.Time (s) Exp.Time (s) Exp.Time (s) (nm) NGC104 (47Tuc) NGC6121 (M4) NGC6809 (M55) GIRAFFE HR13 22500 612.0$-$640.5 2$\times$1200 1$\times$2700 2$\times$3600 HR15 19300 660.7$-$696.5 2$\times$1200 1$\times$1800 2$\times$2770 HR19 14000 774.5$-$833.5 2$\times$2700 2$\times$2700 2$\times$2770 UVES$-$fibre Red 580 47000 480$-$680 2$\times$3600 + 4$\times$1200 3$\times$2700 + 1$\times$1800 2$\times$3600 + 4$\times$2770 -------------- --------- ------- ----------------- ------------------------------- ------------------------------- ------------------------------- -- --------- ---------------- ------------ ------------- ---------------------------------- ---------------------------------- ---------------------------------- GC Observed stars AGB star RGB star RV $\sigma_\mathrm{RV}$ RV$_\mathrm{Harris}$ members members ($\mathrm{km}\,\mathrm{s}^{-1}$) ($\mathrm{km}\,\mathrm{s}^{-1}$) ($\mathrm{km}\,\mathrm{s}^{-1}$) NGC104 94 46 + 4 40 + 4 -17.3 9.9 -18.0 NGC6121 95 17 + 2 63 + 5 70.4 3.3 70.7 NGC6809 110 23 + 1 (1) 74 + 10 (6) 173.6 3.7 174.7 NGC2808 100 30 + 3 38 + 2 104.6 8.0 101.6 --------- ---------------- ------------ ------------- ---------------------------------- ---------------------------------- ---------------------------------- Observation and data reduction {#section:obsreduction} ============================== As already mentioned in PaperI, we selected our targets in NGC104, NGC6121, and NGC6809 from the Johnson-Morgan photometric database which is part of the project described in @Stetson2000 [@Stetson2005] and covers a magnitude range of about three magnitudes for each GC. To distinguish AGB from RGB stars, we used several CMDs with different combinations of color indices and magnitudes. We found that in the CMDs of, for example, (U-I)$-$U, (U-I)$-$I, and (B-I)$-$V, AGB and RGB stars can be separated efficiently, similarly to @GarciaHernandez2015. Figure \[CMD\] shows the location of the member stars in the CMDs of (B-I)$-$V. All our spectra were obtained with the high-resolution multi-object spectrograph FLAMES, mounted on ESO/VLT-UT2 [@Pasquini2003], taking advantage of GIRAFFE (HR 13, HR 15, and HR 19) for the majority of our sample stars and used the UVES fibres (Red 580) for the brightest objects of each cluster. Table \[obslog\] summarizes the most relevant details of our observational campaign. Data reduction followed standard procedures and was carried out as described in PaperI. The final co-added spectra have signal-to-noise ratios (S/N) ranging between 50 and 400 for the GIRAFFE spectra and 50 and 230 for the UVES spectra, depending on the magnitude of the star. We identified non-cluster-member stars based on the derived stellar radial velocities and removed them from further analysis. The number of stars (initially observed and later confirmed as cluster members) are listed in Table \[basic1\] for each of the three GCs, together with their derived mean barycentric radial velocities, whereas Table \[basic2\] lists the most relevant information of only the member stars, that is, their evolutionary phases (AGB/RGB), the instrument used for collecting the spectrum (GIRAFFE/UVES), and their coordinates, photometry and barycentric radial velocities. ![Photometric CMDs (left) and $\log g - \log T_{\rm eff}$ distributions (right) of the cluster member stars. Red circles and blue squares represent AGB and RGB stars, respectively, while the GIRAFFE and UVES samples can be distinguished by open and filled symbols, respectively. The same symbols are used throughout the paper.[]{data-label="CMD"}](CMD_NGC104.eps "fig:"){width="44.00000%"} ![Photometric CMDs (left) and $\log g - \log T_{\rm eff}$ distributions (right) of the cluster member stars. Red circles and blue squares represent AGB and RGB stars, respectively, while the GIRAFFE and UVES samples can be distinguished by open and filled symbols, respectively. The same symbols are used throughout the paper.[]{data-label="CMD"}](CMD_NGC6121.eps "fig:"){width="44.00000%"} ![Photometric CMDs (left) and $\log g - \log T_{\rm eff}$ distributions (right) of the cluster member stars. Red circles and blue squares represent AGB and RGB stars, respectively, while the GIRAFFE and UVES samples can be distinguished by open and filled symbols, respectively. The same symbols are used throughout the paper.[]{data-label="CMD"}](CMD_NGC6809.eps "fig:"){width="44.00000%"} NGC Star ID$^{a}$ Evol. Ph. Instrument RA (J2000) Dec (J2000) $B$ $e_{B}$ $V$ $e_{V}$ $I$ $e_{I}$ $J_{2MASS}$ $e_{J}$ $H_{2MASS}$ $e_{H}$ $K_{2MASS}$ $e_{K}$ RV($\mathrm{km\,s}^{-1}$) --------- --------------- ----------- ------------ ------------- -------------- --------- --------- --------- --------- --------- --------- ------------- --------- ------------- --------- ------------- --------- --------------------------- -- -- 104 AGB58283 AGB GIRAFFE 00 22 54.05 -72 05 17.00 13.802 0.0034 12.579 0.0023 11.355 0.0035 10.414 0.024 9.699 0.025 9.573 0.025 -15.25 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ 6121 AGB30561 AGB GIRAFFE 16 23 16.75 -26 34 28.00 13.170 0.0010 11.732 0.0011 10.093 0.0033 8.851 0.029 8.038 0.034 7.832 0.023 72.04 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ 6809 AGB246868 AGB GIRAFFE 19 39 38.65 -30 48 32.90 14.582 0.0028 13.757 0.0009 12.714 0.0015 11.965 0.023 11.455 0.026 11.359 0.026 171.88 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ Stellar parameters and abundance analysis of our sample stars {#section:stellarparametersabundances} ============================================================= Effective temperature and surface gravity {#section:Teff-logg} ----------------------------------------- Optical B, V, and I magnitudes are available for all our stars from the photometric database mentioned in Sect. \[section:obsreduction\]. By cross-matching the coordinates of our targets with the 2MASS catalog [@Skrutskie2006] we have also compiled the corresponding J, H, and K infrared magnitudes. The foreground reddenings and distance moduli adopted for the three GCs are listed in Table \[reddening-distance\]. We note that for the distance modulus of NGC104, @Bono2008 took an average of the values derived from the tip of the red giant branch and the RR Lyrae methods. For NGC6809, by combining information from photometry (V magnitude and B-V) and period data of 13 RR Lyrae stars [@Olech1999], we computed the absolute distance modulus using the dual band metal-dependent Period$-$Wesenheit (PWZ) relation recently derived by @Marconi2015, which is almost metallicity independent in B and V. For NGC104 and NGC6809, we adopted the @Cardelli1989 relations[^4], as in PaperI. NGC6121, instead, is more peculiar because it is located in the Galactic plane behind the Sco-Oph cloud complex, and a non-standard reddening law should be applied ([@Hendricks2012], and references therein). For this GC, we thus adopted the recent reddening law and distance modulus reported by @Hendricks2012 who used a combination of broadband near-infrared and optical Johnson-Cousins photometry to study the dust properties in the line of sight to this cluster. GC E(B-V) $\delta$E(B-V) (m-M)$_V$ --------- ------------ ---------------- ------------- NGC104 0.04$^{a}$ 0.028$^{d}$ 13.40$^{a}$ NGC6121 0.37$^{b}$ 0.200$^{b}$ 11.28$^{b}$ NGC6809 0.11$^{c}$ 0.027$^{d}$ 13.61 : Foreground reddenings, differential reddenings and distance moduli of sample GCs.[]{data-label="reddening-distance"} As in PaperI, we derived the stellar effective temperatures ($T_\mathrm{eff}$) using the @RamirezMelendez2005 photometric calibrations for giants and adopting five de-reddened color indices, that is, $(B-V)_{0}$, $(V-I)_{0}$, $(V-J)_{0}$, $(V-H)_{0}$, and $(V-K)_{0}$. The variations among these temperature scales are smaller than or comparable to the error of the mean temperature of the five scales. So we took the mean value of the temperatures derived from the five color indices as our final $T_\mathrm{eff}$. The surface gravities $\log g$ were derived from first principles, that is, by using effective temperatures, bolometric corrections (taken from [@Alonso1999]) and stellar masses. For the latter, test runs with our stellar evolution code showed that stellar masses on the RGB have a slight dependence on the metallicity and age of the cluster, while these differences become negligible on the AGB. Therefore, we assumed $m=0.61\,M_{\odot}$ for the AGB stars in all three clusters, whereas we differentiated the values for the RGB stars with $m_\mathrm{RGB,NGC104}=0.91\,M_{\odot}$, $m_\mathrm{RGB,NGC6121}=0.87\,M_{\odot}$ and $m_\mathrm{RGB,NGC6809}=0.81\,M_{\odot}$. The right panels of Fig. \[CMD\] show the $\log T_\mathrm{eff} - \log g$ distributions of the member stars. It is worth mentioning that all three GCs suffer from some differential reddening. In the case of NGC104 and NGC6809 the differential reddenings are small and comparable in magnitude to the errors on the derived reddenings (cf. Table \[reddening-distance\] and references therein). If they are taken into account as one extra source of uncertainty, the typical errors on their final effective temperatures and gravities become of the order of $\pm 70\,\mathrm{K}$ (NGC104), $\pm 80\,\mathrm{K}$ (NGC6809) and $\pm 0.06$ (for both clusters), respectively. The case of NGC6121 is however more complex because of its much larger differential reddening ($\sim 0.20\,mag$, [@Hendricks2012]) . The errors on $T_\mathrm{eff}$ and $\log g$ could reach $\Delta T_\mathrm{eff} \sim \pm 260\,\mathrm{K}$ and $\Delta \log g \sim \pm 0.12\,\mathrm{dex}$ if the reported differential reddening is taken into account at face value, a significant difference from $\Delta T_\mathrm{eff} \sim \pm 50\,\mathrm{K}$ and $\Delta \log g \sim \pm 0.03\,\mathrm{dex}$ derived by accounting for only the intrinsic error on the reddening. However, a range of differential reddening values has been proposed for this cluster, for example, from $\sim 0.05\,mag$ by @Cudworth1990 and $\sim 0.10\,mag$ by @Monelli2013 [@Lardo2017] up to $\sim 0.25\,mag$ by @Mucciarelli2011. Considering the complexity and uncertainties in the reddening of NGC6121, we empirically took the median and decided to consider $\Delta T_\mathrm{eff} \sim \pm 150\,\mathrm{K}$ and $\Delta \log g \sim \pm 0.08\,\mathrm{dex}$ as representative of our analytical uncertainties. Metallicity and microturbulent velocity {#section:Metallicity} --------------------------------------- Metallicity (\[Fe/H\]) and microturbulence ($\xi_{\rm t}$) were determined as in PaperI, where a detailed description of our methodology is provided. In short, metallicities were derived by measuring the equivalent widths (EWs) of both and unblended lines, restricting our selection to lines with EWs between $20\,\mathrm{m}$Åand $120\,\mathrm{m}$Å. For the computation of the abundances, we used 1D LTE spherical MARCS model atmospheres [@Gustafsson2008], the LTE stellar line analysis programme MOOG ([@Sneden1973], 2014 version) and we assumed a solar iron abundance of $\log\epsilon\mathrm{(Fe)}_{\odot}=7.50$ [@Asplund2009]. Because of their known inter-dependencies, all stellar parameters ($T_\mathrm{eff}$, $\log g$, \[Fe/H\], $\xi_\mathrm{t}$) were derived iteratively and following standard procedures. Since standard LTE analyses of lines tend to underestimate the true iron abundance, we applied non-LTE (NLTE) corrections to all our LTE values ([@Lind2012], and references therein). Our final stellar parameters are summarized in Table \[stellarpar\], while Table \[metal\] lists the mean metallicities of the AGB and RGB samples in the three GCs. For convenience, we have added to this table also the values derived in PaperI for NGC2808. We note that our RGB results agree well with those derived by @Carretta2009a within the associated errors (cf. two rightmost columns of Table \[metal\]). A more detailed comparison with the literature forms part of Sect.4. Finally, as our overall metallicity of each GC, we chose to use the average value of \[/H\]$_\mathrm{NLTE}$ and \[/H\], that is, \[Fe/H\]$_\mathrm{NGC\,104} = -0.82$dex, \[Fe/H\]$_\mathrm{NGC\,6121} = -1.14$dex, and \[Fe/H\]$_\mathrm{NGC\,6809} = -1.86$dex. --------- ----------- ----------- ------------------ ------------------------------ ---------- ------------------------- ----------------------- ------------------ ------------------ ------------------ ------------------------ NGC Star ID Evol. Ph. $T_\mathrm{eff}$ $\sigma T_\mathrm{eff}~^{a}$ $\log g$ $\xi_{\rm t}$ \[/H\]$_\mathrm{LTE}$ rms\_lines \[/H\] rms\_lines \[/H\]$_\mathrm{NLTE}$ ($\mathrm{K}$) ($\mathrm{K}$) ($\mathrm{km\,s}^{-1}$) ($\mathrm{dex}$) ($\mathrm{dex}$) ($\mathrm{dex}$) ($\mathrm{dex}$) ($\mathrm{dex}$) 104 AGB58283 AGB 4291 39.3 1.23 1.60 -0.85 0.12 -0.89 0.15 -0.82 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ 6121 AGB30561 AGB 4433 25.3 1.28 1.75 -1.29 0.10 -1.26 0.02 -1.24 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ 6809 AGB246868 AGB 4964 25.7 1.86 1.16 -2.00 0.15 -1.89 0.01 -1.91 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ --------- ----------- ----------- ------------------ ------------------------------ ---------- ------------------------- ----------------------- ------------------ ------------------ ------------------ ------------------------ ------------ ----------------------- ------------------------ ----------------------------- ---------------- ---------------- ---------------------------------- ------------------- GC/sample \[/H\]$_\mathrm{LTE}$ \[/H\]$_\mathrm{NLTE}$ rms(\[/H\]$_\mathrm{NLTE}$) \[/H\] rms(\[/H\]) \[/H\]$_\mathrm{LTE}$ (Carretta) \[/H\] (Carretta) $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ NGC104 AGB sample -0.82 -0.79 0.07 -0.91 0.13 RGB sample -0.77 -0.75 0.08 -0.82 0.15 -0.743$\pm$0.047 -0.769$\pm$0.081 NGC6121 AGB sample -1.21 -1.15 0.08 -1.22 0.09 RGB sample -1.15 -1.11 0.08 -1.15 0.06 -1.200$\pm$0.043 -1.197$\pm$0.082 NGC6809 AGB sample -2.03 -1.93 0.07 -1.91 0.04 RGB sample -1.92 -1.85 0.06 -1.84 0.06 -1.967$\pm$0.041 -1.933$\pm$0.093 NGC2808 AGB sample -1.19 -1.14 0.09 -1.14 0.10 RGB sample -1.12 -1.08 0.07 -1.09 0.07 -1.100$\pm$0.059 -1.160$\pm$0.089 ------------ ----------------------- ------------------------ ----------------------------- ---------------- ---------------- ---------------------------------- ------------------- Sodium abundance {#section:Na} ---------------- Our stellar Na abundances were derived via spectrum synthesis of the Na doublet at 6154$-$6160Å, using MOOG and MARCS spherical model atmospheres interpolated to match our derived stellar parameters. A solar sodium abundance of $\log\epsilon(\mathrm{Na})_{\odot}=6.24$ [@Asplund2009] was adopted throughout the analysis. As already done for NGC2808 (cf. PaperI) we took the average of the abundances derived from both doublet lines as our final Na abundance. Similarly to iron, Na abundances determined from neutral lines (as in our case) are also affected by the NLTE effect and were therefore corrected accordingly. We list the Na abundances derived for the individual stars in Table \[Na\_star\], along with the NLTE-corrected values based on the grids computed by @Lind2011. Table \[Na\_GC\] is similar to Table \[metal\], now summarizing the average Na abundances of our three GCs. For convenience, it also reports the results from our previous analysis on NGC2808. --------- ----------- ----------- ------------------------- -------------------------- ------------------------- NGC Star ID Evol. Ph. \[Na/H\]$_\mathrm{LTE}$ \[Na/H\]$_\mathrm{NLTE}$ \[Na/\]$_\mathrm{NLTE}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ 104 AGB58283 AGB -0.10 -0.21 0.61 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ 6121 AGB30561 AGB -0.79 -0.85 0.39 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ 6809 AGB246868 AGB -1.45 -1.53 0.38 $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ $\dots$ --------- ----------- ----------- ------------------------- -------------------------- ------------------------- ------------ ------------------------- ---------------- -------------------------- -------------------------- ------------------------- ------------------------------------ ----------------------- GC \[Na/H\]$_\mathrm{LTE}$ $\delta$NLTE \[Na/H\]$_\mathrm{NLTE}$ $\sigma_\mathrm{[Na/H]}$ \[Na/\]$_\mathrm{NLTE}$ $\sigma_\mathrm{[Na/\ion{Fe}{i}]}$ IQR$_\mathrm{[Na/H]}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ NGC104 AGB sample -0.42 -0.07 -0.49 0.19 0.30 0.20 0.285 RGB sample -0.30 -0.09 -0.39 0.17 0.36 0.19 0.330 NGC6121 AGB sample -0.90 -0.07 -0.97 0.17 0.19 0.16 0.300 RGB sample -0.75 -0.08 -0.82 0.20 0.29 0.19 0.250 NGC6809 AGB sample -1.60 -0.07 -1.66 0.13 0.27 0.15 0.220 RGB sample -1.56 -0.07 -1.62 0.17 0.23 0.16 0.250 NGC2808 AGB sample -1.00 -0.06 -1.06 0.13 0.09 0.16 0.160 RGB sample -0.98 -0.06 -1.04 0.21 0.05 0.20 0.390 ------------ ------------------------- ---------------- -------------------------- -------------------------- ------------------------- ------------------------------------ ----------------------- We note that in the two most metal-rich GCs in our sample, the Na doublet was saturated or approaching saturation in 27 stars of NGC104 and 4 stars of NGC6121, respectively. These stars were dropped from any further discussion. Error analysis {#section:Error} -------------- We estimated the uncertainties in our derived abundances, following the procedures described in PaperI and considering errors of both random and systematic nature. As random measurement uncertainty, we considered $\sigma/\sqrt{\mathrm{N}}$, where $\sigma$ is the line-to-line dispersion and $\mathrm{N}$ is the number of lines measured. A correction according to a t-distribution was applied to the Na and abundances considering the limited number of lines present in our spectra (cf. PaperI for more details). For the systematic uncertainty, we selected a total of six stars per GC: Four stars observed with GIRAFFE $-$ one cool/hot in each AGB/RGB sample; and two stars observed with UVES-fibre $-$ one cool AGB and one cool RGB, as this sub-sample includes only cool stars. With these 18 stars, we tested the effect of varying stellar parameters and EWs (or other key parameters of the analysis) by their associated errors on the derived abundances. Typical total uncertainties, combining both random and systematic sources of error, are summarized in Table \[Error\] for the GIRAFFE and UVES samples of each GC. ---------------- ---------------- ---------------- ---------------- ---------------- \[/H\] \[/H\] \[Na/H\] \[Na/\] $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ $\mathrm{dex}$ NGC104 GIRAFFE sample $\pm 0.07$ $\pm 0.12$ $\pm 0.13$ $\pm 0.13$ UVES sample $\pm 0.07$ $\pm 0.13$ $-$ $-$ NGC6121 GIRAFFE sample $\pm 0.13$ $\pm 0.14$ $\pm 0.16$ $\pm 0.11$ UVES sample $\pm 0.09$ $\pm 0.18$ $\pm 0.16$ $\pm 0.10$ NGC6809 GIRAFFE sample $\pm 0.11$ $\pm 0.07$ $\pm 0.13$ $\pm 0.12$ UVES sample $\pm 0.09$ $\pm 0.05$ $\pm 0.13$ $\pm 0.10$ ---------------- ---------------- ---------------- ---------------- ---------------- : Total uncertainties of derived Fe and Na abundances.[]{data-label="Error"} Stars observed with both GIRAFFE and UVES-fibre ----------------------------------------------- In the case of NGC6809, we were able to optimize our fibre configurations by swapping targets between fibre types. We thus observed seven objects (one AGB and six RGB) with both GIRAFFE-Medusa and UVES-fibre, which allowed us to test whether or not any zero-point abundance difference exists between the two different sets of spectra. We found the following average differences (UVES-fibre minus GIRAFFE): $\Delta$\[/H\]$_\mathrm{NLTE}\,=\,+0.02\pm0.04\,\mathrm{dex}$, $\Delta$\[/H\]$\,=\,-0.03\pm0.07\,\mathrm{dex}$, $\Delta$\[Na/H\]$_\mathrm{NLTE}\,=\,-0.01\pm0.05\,\mathrm{dex}$, and $\Delta$\[Na/Fe\]$_\mathrm{NLTE}\,=\,-0.03\pm0.06\,\mathrm{dex}$. These results show that, despite a lower resolution and smaller spectral coverage, the majority of our spectra (indeed observed with GIRAFFE) match very well the results obtained from the analysis of the UVES-fibre counterparts. We highlight that in the following Sections, although it does not make any significant difference, we have considered the UVES-fibre results for these seven stars. Observed Na abundance distribution along the RGB and AGB {#section:abund_results} -------------------------------------------------------- ![Abundance distributions of our complete (AGB $+$ RGB) sample in NGC104, NGC6121, and NGC6809 (top, middle, and bottom) [Left:]{} \[Na/H\]$_\mathrm{NLTE}-T_\mathrm{eff}$; [right:]{} \[Na/\]$_\mathrm{NLTE}-T_\mathrm{eff}$. Symbols are the same as in Fig. \[CMD\]. The error bars correspond to our estimates for the GIRAFFE sample. []{data-label="Na_distr"}](Na-Teff_distr2_NGC104.eps "fig:"){width="46.00000%"} ![Abundance distributions of our complete (AGB $+$ RGB) sample in NGC104, NGC6121, and NGC6809 (top, middle, and bottom) [Left:]{} \[Na/H\]$_\mathrm{NLTE}-T_\mathrm{eff}$; [right:]{} \[Na/\]$_\mathrm{NLTE}-T_\mathrm{eff}$. Symbols are the same as in Fig. \[CMD\]. The error bars correspond to our estimates for the GIRAFFE sample. []{data-label="Na_distr"}](Na-Teff_distr2_NGC6121.eps "fig:"){width="46.00000%"} ![Abundance distributions of our complete (AGB $+$ RGB) sample in NGC104, NGC6121, and NGC6809 (top, middle, and bottom) [Left:]{} \[Na/H\]$_\mathrm{NLTE}-T_\mathrm{eff}$; [right:]{} \[Na/\]$_\mathrm{NLTE}-T_\mathrm{eff}$. Symbols are the same as in Fig. \[CMD\]. The error bars correspond to our estimates for the GIRAFFE sample. []{data-label="Na_distr"}](Na-Teff_distr2_NGC6809.eps "fig:"){width="46.00000%"} Our final Na abundance distributions are shown in Fig. \[Na\_distr\] as a function of effective temperature for AGB and RGB stars in NGC104, NGC6121, and NGC6809. Here, we present them in the form of NLTE \[Na/H\] and \[Na/Fe\] (where Fe now only refers to ) ratios, as listed in Table \[Na\_star\]. However, as pointed out by @Campbell2017, the \[/H\] can be affected by the $T_\mathrm{eff}$ scale, which further influences the degree of $-$ discrepancy that has been found especially in AGB stars by several works [@Ivans2001; @Lapenna2014; @Lapenna2015; @Lapenna2016; @Mucciarelli2015] and also by us (see also Sect. \[section:NGC6752\]). Any uncertainty in the determination of the iron abundance will affect the accuracy of the \[Na/Fe\] ratios, while \[Na/H\] is very robust, as also detailed by @Campbell2017 who show that parameter variations between studies (caused by using different methods, tools, input data, etc.) have little effect on the derived \[Na/H\]. This was also our conclusion from PaperI, which led us to base our discussion of the Na distribution in the RGB and AGB samples on the \[Na/H\] abundance indicator only. For each individual cluster we perform a two-sample Kolmogorov-Smirnov (K-S) test to estimate the similarity of the \[Na/H\] distributions in the AGB and RGB samples. All numbers and conclusions discussed below are roughly confirmed by the dispersions ($\sigma$) and the interquartile range (IQR) values of the distributions reported in Table \[Na\_GC\]. For each cluster, we also provide a critical summary of how our derived Na abundances compare to other abundance studies of similar data quality. ### NGC104 {#subsubsection:NGC104} NGC104 (40 AGB, 27 RGB) is the GC in which the AGB and RGB samples are the closest in terms of Na abundance distributions, with the AGB stars spanning an even slightly larger range of Na abundances. The K-S test, with D=0.294 and p-value=0.101 derived, indicates that the AGB and RGB samples share the same \[Na/H\] distribution at 95$\%$ significance level. Our result for this cluster is in very good agreement with @Johnson2015 who found nearly identical \[Na/Fe\] dispersion in their AGB sample (35 stars) to that in the RGB sample (113 stars) analyzed by @Cordero2014, following the same methodology of @Johnson2015. We carried out a detailed check for 12 AGB and 13 RGB stars in common with our sample (identified by coordinates cross-matching with angular distance < 0.3”) and found mean differences (in the sense of our result minus theirs) of $-78\pm60$K in $T_\mathrm{eff}$, $-0.07\pm0.22$ in $\log g$, $-0.25\pm0.17$kms$^{-1}$ in $\xi_\mathrm{t}$, and $-0.12\pm0.10$dex in \[Fe/H\]. We assign these differences mainly to the different methods employed to derive the stellar parameters $T_\mathrm{eff}$ and $\log g$ (photometry in the present study, spectroscopy in @Johnson2015 and @Cordero2014). However, the negligible difference in \[Na/H\] ($0.02\pm0.12$dex; after having checked that their conclusions for \[Na/Fe\] hold also for the \[Na/H\] ratio) confirms the agreement found in the Na abundance distribution. ### NGC6121 {#subsubsection:NGC6121} We find that in NGC6121 (19 AGB, 63 RGB), the AGB stars occupy the bottom two-thirds of the \[Na/H\] distribution of the RGB ones. There is an actual difference of 0.26dex between the maximum \[Na/H\] values of the AGB and RGB samples, indicating a lack of very Na-rich AGB stars in this cluster. The D=0.408 and p-value=0.011 from the K-S test confirms that the two populations do not share the same Na distribution. Our results for this GC agree with those of @MacLean2016 who studied a sample of 106 RGB and 15 AGB stars in NGC6121. In their analysis, they derived a difference of 0.4dex between the maximum \[Na/H\] values of their AGB and RGB samples (compared to our value of 0.26dex, we have more Na-rich AGB stars than @MacLean2016). Although the authors do not provide the coordinates of their sample stars, we were able to find 29 RGB stars in common, by investigating the overlap existing between their sample and @Marino2008 and applying our cross-matching criterion with an angular distance $\leq$ 0.3”. The mean differences in the derived stellar parameters and Na abundance for this RGB subsample were found to be (in the sense of our result minus theirs) $-142\pm78$K in $T_\mathrm{eff}$, $-0.05\pm0.13$ in $\log g$, $-0.04\pm0.23$kms$^{-1}$ in $\xi_\mathrm{t}$, $-0.01\pm0.09$dex in \[Fe/H\], and $0.04\pm0.12$dex in \[Na/H\]. Except for the offset in $T_\mathrm{eff}$, we consider this to be a very good agreement. Besides, our derived abundances also agree quite well with the results of @Marino2017 who studied 17 AGB stars in this cluster. By comparing 14 common stars, differences of $-0.01\pm0.05$dex in \[Fe/H\] and $0.00\pm0.07$dex in \[Na/H\]$_\mathrm{LTE}$ were found between the two studies (our result minus theirs), while the differences in $T_\mathrm{eff}$, $\log g$, and $\xi_\mathrm{t}$ were, respectively, $-69\pm54$K, $-0.09\pm0.17$, and $-0.22\pm0.11$kms$^{-1}$. ### NGC6809 {#subsubsection:NGC6809} In NGC6809 (23 AGB, 77 RGB), we find that the Na abundances of the RGB sample spread more evenly while the AGB stars tend to be more concentrated. The difference between the maximum \[Na/H\] values reached by each sample amounts to 0.12dex (the AGB sample reaching lower values). The difference is only half of the one observed in NGC6121, and the two-sided K-S test does not provide any strong evidence for the two distributions to be different, with D=0.190 and p-value=0.498. This is the first time that AGB stars are targeted and analyzed for their Na abundances in this cluster. We have however 27 RGB stars in common with the @Carretta2009a sample, for which we found mean differences of $52\pm27$K in $T_\mathrm{eff}$, $-0.01\pm0.01$ in $\log g$, $-0.16\pm0.38$kms$^{-1}$ in $\xi_\mathrm{t}$, $0.01\pm0.05$dex in \[/H\], $0.10\pm0.08$dex in \[/H\], and $0.00\pm0.19$dex in \[Na/H\]. Within the errors, our results are consistent with those derived by @Carretta2009a. ### NGC2808 {#subsubsection:NGC2808} For completeness, we recall here the main results of PaperI: the Na abundances of the 33 AGB and 40 RGB stars we analyzed in NGC2808 can be considered to follow the same distribution (K-S test gives D=0.268 and p-value=0.137) although a difference of 0.21dex was found between the maximum \[Na/H\] values of the AGB and RGB samples, with the AGB maximum value being lower. Other clusters {#section:OtherGCs} ============== Re-analysis of C13 data for NGC6752 {#section:NGC6752} ----------------------------------- To enlarge the number of clusters studied self-consistently, we decided to reanalyze the publicly available data of C13 following our analytical methods. We adopted the reddening E(B-V)=0.04mag and visual distance modulus (m-M)$_V$=13.24mag from @Gratton2003, and assumed stellar masses of $m_\mathrm{AGB}=0.61\,M_{\odot}$ and $m_\mathrm{RGB}=0.83\,M_{\odot}$, following the theoretical predictions of stellar evolution models [assuming an age of 12.5 Gyr, @Chantereau2015]. Using the methods described in Sects. \[section:Teff-logg\] and \[section:Metallicity\], we derived the stellar parameters and the metallicities for a total of 44 individual stars (20 AGB and 24 RGB stars). Overall, we find a good agreement with the C13 result on the effective temperature with a mean difference of $\Delta\,T_\mathrm{eff} = -24\pm51\,\mathrm{K}$ (ours minus C13), while the mean differences on the gravity and microturbulent velocity are $\Delta\,\log g = -0.14\pm0.07$ and $\Delta\,\xi_\mathrm{t} = -0.30\pm0.25\,\mathrm{km\,s}^{-1}$, respectively. We derive the following LTE Fe abundances: \[/H\]$_\mathrm{AGB} = -1.85\pm0.02$dex ($\sigma = 0.06$dex), \[/H\]$_\mathrm{AGB} = -1.65\pm0.06$dex ($\sigma = 0.04$dex), \[/H\]$_\mathrm{RGB} = -1.66\pm0.02$dex ($\sigma = 0.07$dex), and \[/H\]$_\mathrm{RGB} = -1.55\pm0.06$dex ($\sigma = 0.05$dex), and find for the AGB stars a very good agreement with @Lapenna2016, who re-observed the 20 AGB stars of C13 at higher resolution with ESO-VLT/UVES (\[/H\]$_\mathrm{AGB} = -1.80\pm0.01$dex and \[/H\]$_\mathrm{AGB} = -1.58\pm0.01$dex). After applying NLTE corrections to the abundances [@Bergemann2012; @Lind2012], we derived the average values of \[/H\]$_\mathrm{NLTE,AGB} = -1.76\pm0.04$dex and \[/H\]$_\mathrm{NLTE,RGB} = -1.60\pm0.04$dex, which bring the $-$ values for the RGB stars into agreement (within the errors) and almost halves the $-$ difference for the AGB stars. Compared to the value assumed by C13 for all their sample stars (\[Fe/H\]=$-$1.54dex), our derived metallicities are slightly lower and point to a 0.1dex difference between AGB and RGB stars. ![Comparison of the Na abundance (NLTE) of NGC6752 sample between our derived results and those of C13. Red circles represent AGB stars, and blue squares represent RGB stars[]{data-label="comp_Campbell2013"}](Comp-Na_NGC6752_Campbell2013.eps){width="46.00000%"} Considering the weakness of the $6154-60$Å Na lines in a non-negligible number of spectra of the C13 sample, we followed their choice and derived the Na abundances from the EWs of the Na doublet at $5682-5688$Å[^5]. We apply the same NLTE corrections [@Lind2011] to the derived Na abundances as for the other GCs and find a mean difference (ours minus C13) of $-0.16\pm0.06$dex in \[Na/H\] and of $-0.03\pm0.13$dex in \[Na/Fe\]. We compare the derived Na abundances from the two studies in Fig. \[comp\_Campbell2013\]. The differences in \[Na/H\] appear to be systematic (likely due to the combination of different stellar parameters and adopted solar abundances), while the effect on \[Na/Fe\] is influenced mainly by the adopted Fe abundance (overall cluster metallicity in the case of C13, individual stellar metallicities in our case; see also [@Lapenna2016]). The remaining offsets are likely to come from the NLTE correction applied to the Na abundance (C13 used the [@Gratton1999] values). Figure \[Na\_NGC6752\] shows all results of this re-analysis (left column), with a direct comparison to the values published by C13 (middle column). For AGB stars, our \[Na/H\] distribution agrees quite well with that of C13, while this is not the case for \[Na/Fe\]. Testing the assumption on the metallicities made by the two analyses shows that we can fully reproduce C13 results as soon as we use an overall metallicity value for the cluster. Both our \[Na/H\] and \[Na/\] distributions derived for the AGB sample agree well with those derived by @Lapenna2016 from higher-resolution spectral data, as shown in the right column of Fig. \[Na\_NGC6752\] (labeled as L16; the RGB stars are from our re-analysis and are shown only to aid the comparison), which supports our abundance results. As in Sect. \[section:abund\_results\], we take \[Na/H\] as the Na abundance indicator and find that there is a significant lack of Na-rich AGB stars in the sample of NGC6752 (2P AGB stars account for $\sim$15%, cf. Sect. \[section:discussion\]). However, @Lapenna2016 claimed that both 1P and 2P stars populate the AGB of NGC6752 with $\sim$65% of AGB stars belonging to 2P based on the \[O/Fe\]-\[Na/Fe\] distribution. Only stars with extreme Na enhancements are claimed to be missing in their AGB sample. We note, however, that this conclusion was derived based on \[Na/Fe\] ratios. The $-$ discrepancy (especially their different behaviors in AGB and RGB stars) may affect the relative Na abundance distributions between AGB and RGB stars (e.g., Fig. \[Na\_NGC6752\], left column) so that the \[Na/Fe\] indicator carries a larger uncertainty compared to the \[Na/H\] (Sect. \[section:abund\_results\]). Furthermore, we note that our derived fraction of 2P-AGB stars ($\sim$15%) is lower than the one predicted by @Cassisi2014b [$\sim$50%] based on simulations of the horizontal branch. While the specific fraction numbers of different stellar populations depend on the adopted separation criteria (see also Sect. \[section:1Pvs2P\]), the sampling could also affect the results. Although we believe that this GC deserves further scrutiny, for the sake of a coherent discussion, we use our derived abundance ratios in the following Sections. ![Na abundance (NLTE) distributions of the sample of C13 in NGC6752. The left-column panels show our re-analysis results, middle-column panels show the Na abundance distribution from C13, and the right-column panels show the Na abundance of the AGB sample derived by @Lapenna2016 together with the RGB stars from our re-analysis. For convenience of comparison, the data from C13 and @Lapenna2016 are shifted systematically to our scale with a constant which is equal to the mean difference between their results and ours, while the dispersions derived in each study are kept. Red circles and blue squares represent AGB and RGB stars, respectively. []{data-label="Na_NGC6752"}](comp_NaH-NaFe_NGC6752_3studies.eps){width="46.00000%"} Literature data for four additional GCs {#section:literature-GCs} --------------------------------------- In the literature, one finds four other GCs whose AGB stars have been targeted for their Na abundances through moderate- and high-resolution spectroscopic observations: NGC5904 [@Ivans2001; @Lai2011], NGC5986 [@Johnson2017], NGC6205 [@Johnson2012], and NGC6266 [@Lapenna2015]. The Na abundance patterns observed in these clusters are shown in Fig. \[NaH\_5GCs-literature\], where the data points have already been adjusted to be on the same solar abundance and NLTE-correction scales as our own data set; for NGC5904 the abundance data from @Lai2011 have been unified to the system of @Ivans2001 according to the common star between the two studies, while for the common star we adopt the data derived by @Ivans2001. [[NGC6205]{}]{} (M13; \[Fe/H\] = $-$1.57, @Johnson2012) seems to be relatively devoid of Na-normal AGB stars, with a difference of $\sim$0.4dex between the \[Na/H\] minimum values in the RGB and the AGB samples. But, as suggested by @GarciaHernandez2015, due to the difficulty of distinguishing AGB from RGB stars at the bright end of the giant branch, some misclassification might have occurred, which could slightly favor the RGB sample. Notwithstanding, the presence of a large fraction of Na-rich AGB stars in NGC6205 is clear. [[NGC5986]{}]{} (\[Fe/H\] = $-$1.54, @Johnson2017) shows comparable \[Na/H\] spreads in AGB and RGB star samples with the maximum \[Na/H\] value of AGB stars being 0.14dex lower than that of RGB stars. The sample of AGB stars is, however, rather limited. [[NGC5904]{}]{} (M5; \[Fe/H\] = $-$1.22, @Ivans2001 [@Lai2011]) shows a paucity of very Na-rich AGB stars compared to RGB stars, with a difference of 0.25dex between the maximum \[Na/H\] value of the AGB and RGB samples. [[NGC6266]{}]{} (M62; \[Fe/H\] = $-$1.05, @Lapenna2015) makes itself distinct by showing no Na-rich AGB star in the small sample analyzed by @Lapenna2015. However, in our opinion, no firm conclusion can be drawn because the AGB phase may not be sufficiently sampled. Discussion for the full sample of GCs with Na abundance determination on the AGB {#section:discussion} ================================================================================ ![image](NaHnlte_our-4GCs_distr.eps){width="94.00000%"} ![image](NaHnlte_distr_5GCs-literature.eps){width="96.00000%"} Comparison criteria {#subsection:criteria} ------------------- For the reasons previously described, we use the \[Na/H\]$_\mathrm{NLTE}$ data to discuss the differences and similarities in terms of Na abundances in the nine GCs for which both the RGB and AGB have been studied. The abundance determination is self-consistent for five GCs (NGC6809, NGC6121, NGC2808, NGC104, and NGC6752; this paper and Paper I). For the four other GCs (NGC6205, NGC5986, NGC5904, and NGC6266), we use the data from the original papers that we modified to have consistent Na reference solar abundance and NLTE corrections (Sect. \[section:literature-GCs\]). We use the same definition as in PaperI to distinguish Na-normal and Na-rich stars (often called 1P and 2P stars in the literature; see e.g., [@PrantzosCharbonnel2006; @Carretta2009a]), the latter ones being defined as those having \[Na/H\] higher than \[Na/H\]$_\mathrm{cri}$=\[Na/H\]$_\mathrm{min}$+0.3dex, where $ $\[Na/H\]$_\mathrm{min}$ is the minimum Na value derived for the RGB+AGB sample in a given cluster and 0.3dex is about one third of the \[Na/H\] spread. Table \[9GCs\_comp\] gathers the values of \[Na/H\]$_\mathrm{cri}$, of the fraction of Na-rich RGB and AGB 2P stars (f$_\mathrm{2P,RGB}$ and f$_\mathrm{2P,AGB}$ respectively), and of the Na spreads $\Delta$\[Na/H\] for the RGB and AGB subsamples in each GC; we have also collected in the table important cluster characteristics that are relevant for the discussion. Na abundance distributions among the RGB and AGB samples {#section:comparisonall} -------------------------------------------------------- We gather all the \[Na/H\]$_\mathrm{NLTE}$ data for the nine GCs as a function of stellar effective temperature (Figs. \[NaH\_4GCs\] and \[NaH\_5GCs-literature\]) and in the form of continuous histograms (Figs. \[Na\_hist\_5GCs\] and \[Na\_hist\_4GCs\], where every star is represented by a Gaussian profile with a weight of one and standard deviation equalling the uncertainty on the measurement) for RGB and AGB samples. In all cases, the Na dispersion observed for RGB and AGB stars does not depend on the effective temperature (Figs. \[NaH\_4GCs\] and \[NaH\_5GCs-literature\]) or the brightness of the stars (not shown here). This means that there is no in situ evolution effect that modifies the Na abundance inside the GC evolved stars we observe today. A quick look at these figures shows that in the majority of the GCs under scrutiny (eight out of nine, the only exception being NGC104), the Na spread is smaller among AGB stars than among RGB stars (see the actual dispersion numbers in Table \[9GCs\_comp\]). More specifically, the maximum \[Na/H\] values derived for the AGB stars are lower than the ones derived for the RGB stars. In three out of these eight GCs (NGC6809, NGC6205, and NGC5986), the maximum values for Na on the RGB and the AGB are however marginally consistent considering the errors. Three clusters (NGC2808, NGC5904, NGC6121) clearly lack the most Na-rich AGB star. Finally, the last two clusters (NGC6752 and NGC6266[^6]) stand out by showing (almost) no Na-rich AGB stars (i.e., with \[Na/H\] higher than \[Na/H\]$_\mathrm{cri}$) in the samples analyzed so far. ![Continuous histograms of \[Na/H\] of the AGB (red) and RGB (blue) samples for the five GCs we have analyzed in a self-consistent way (whose \[Fe/H\] decrease from top to bottom).[]{data-label="Na_hist_5GCs"}](Histogram-continuous_5GCs_NaH.eps){width="38.00000%"} ![Same as Fig. \[Na\_hist\_5GCs\] but for literature data.[]{data-label="Na_hist_4GCs"}](Histogram-continuous_4GCs_NaH.eps){width="38.00000%"} --------------------------------- ----------- ----------- ----------- ----------- ----------- ----------- ----------- ----------- ----------- NGC6809 NGC6121 NGC2808 NGC104 NGC6752 NGC6205 NGC5986 NGC5904 NGC6266 M55 M4 47Tuc M13 M5 M62 $ $\[Na/H\]$_\mathrm{cri}$ -1.70 -0.97 -1.05 -0.50 -1.65 -1.50 -1.46 -1.35 -1.05 $\Delta$\[Na/H\]$_\mathrm{AGB}$ 0.48 0.53 0.54 0.70 0.36 0.62 0.61 0.51 0.30 f$_\mathrm{2P,AGB}$ 65$\pm$11 53$\pm$11 55$\pm$9 40$\pm$8 15$\pm$11 87$\pm$13 57$\pm$18 67$\pm$13 0$\pm$23 $\Delta$\[Na/H\]$_\mathrm{RGB}$ 0.70 0.75 0.75 0.57 0.91 1.12 0.76 0.68 1.08 f$_\mathrm{2P,RGB}$ 66$\pm$6 76$\pm$6 52$\pm$8 67$\pm10$ 75$\pm$11 76$\pm$5 72$\pm$12 69$\pm$9 62$\pm$14 $ $\[Fe/H\] -1.86 -1.14 -1.11 -0.82 -1.60 -1.57 -1.54 -1.22 -1.05 Age 13.00 11.50 11.00 11.75 12.50 12.00 12.25 11.50 11.60 $\pm$0.25 $\pm$0.38 $\pm$0.38 $\pm$0.25 $\pm$0.25 $\pm$0.38 $\pm$0.75 $\pm$0.25 $\pm$0.60 M$_\mathrm{V}$ -7.57 -7.19 -9.39 -9.42 -7.73 -8.55 -8.44 -8.81 -9.18 mass 0.269 0.195 1.420 1.500 0.317 0.775 0.599 0.857 1.220 HBR 0.87 -0.06 -0.49 -0.99 1.00 0.97 0.97 0.31 0.32 r$_\mathrm{h}$ 4.46 2.34 2.12 3.65 2.72 3.34 3.18 4.60 2.47 r$_\mathrm{t}$ 25.10 20.79 43.42 56.10 64.39 56.40 31.83 61.96 18.00 ellipticity 0.02 0.00 0.12 0.09 0.04 0.11 0.06 0.14 0.01 c.c. 0.93 1.65 1.56 2.07 2.50 1.53 1.23 1.73 1.71 $\sigma_\mathrm{v}$ 4.0 4.0 13.4 11.0 4.9 7.1 $-$ 5.5 14.3 --------------------------------- ----------- ----------- ----------- ----------- ----------- ----------- ----------- ----------- ----------- The continuous histograms of the \[Na/H\] (Figs. \[Na\_hist\_5GCs\] and \[Na\_hist\_4GCs\]) indicate that the Na abundance distributions of the RGB and AGB stars cannot be described by one type of profile (single- or double-peak) and confirm that they vary from cluster to cluster. According to the current data, both AGB and RGB samples of NGC104 and NGC6205 are bimodal; those of NGC6121 and NGC6809 are unimodal; for NGC2808 and NGC6752, their RGB and AGB samples appear to be bimodal and unimodal, respectively; NGC6266 has a unimodal AGB sample and trimodal RGB sample; NGC5904 shows bimodal distribution in AGB sample but a broadened profile in RGB sample which indicates that three closely located peaks may exist; while NGC5986 shows trimodal distributions in both AGB and RGB samples. Separations can clearly be found between the main peaks of AGB and RGB samples of NGC6121, NGC6752, NGC6266, and NGC5986, with the ones of AGB always having lower \[Na/H\] values than RGB, and the main peaks in NGC6752 separate the most. Fractions of 1P and 2P stars {#section:1Pvs2P} ---------------------------- The number ratio between 1P and 2P stars has been extensively used to constrain the models that aim at explaining the chemical properties of the stellar populations in GCs (e.g., [@PrantzosCharbonnel2006; @Carretta2010; @Decressin2010; @SchaererCharbonnel2011; @Charbonnel2014; @Larsen2014; @BastianLardo2015; @KhalajBaumgardt2015]). As a general concept, 1P and 2P refer respectively to GC stars that present chemical abundances similar or different to those of field stars of similar metallicity. This can be seen in Fig. \[GCvsfield\] where we compare the Na data for RGB and AGB stars that we use in the discussion (including our determinations and data from the literature) with the Na data in Galactic field stars gathered and homogenized by @Carretta2013. Using the same definition of \[Na/H\]$_\mathrm{cri}$ as in PaperI (see Sect. \[subsection:criteria\]), we calculate the fraction of 2P stars (f$_\mathrm{2P}$) both on the RGB and AGB in each individual GC (Table \[9GCs\_comp\]). Our results agree with the finding of @Carretta2009a that the 2P RGB component is present in all clusters, with a fraction between $\sim 50\%$ and $75\%$. For the AGB, we confirm previous abundance studies that revealed a complex picture, with some clusters being almost devoid of 2P AGB stars (NGC6752, NGC6266), some having similar 2P fractions on the AGB to on the RGB (NGC2808, NGC5904, and NGC6809), and all the possible intermediate cases between these two extreme behaviors. Overall, the fraction of 2P AGB stars varies between $\sim 0\%$ and $87\%$ (these extreme values corresponding to NGC6266 and NGC6205, respectively). ![image](NaH-FeH_9GCs_RGB_field-stars.eps){width="47.00000%"} ![image](NaH-FeH_9GCs_AGB_field-stars.eps){width="47.00000%"} Obviously, the actual numbers one obtains for the fractions of 1P and 2P stars in a given GC depend on the adopted distinction criteria. For example, our conclusion that NGC6121 hosts 2P AGB stars seems at first sight to contradict the claim by @MacLean2016 that there is no 2P AGB star in this cluster. However, @MacLean2016 derived the population separation point (PSP) for their NGC6121 sample (\[Na/O\]=$-$0.16dex) by identifying a minimum in the \[Na/O\] distribution between the two subpopulations in the RGB sample of @Marino2008, and they found that the abundance distribution was consistent with all the AGB stars being of 1P, while 45% of the RGB stars belong to 2P. However, if the criterion by @Carretta2009a to distinguish 1P and 2P stars (i.e., \[Na/Fe\]$_\mathrm{cri} =\,$\[Na/Fe\]$_\mathrm{min}\,+\,0.3$dex) is applied, one finds 6 out of 15 AGB stars and 65 out of 106 RGB stars belonging to 2P, which accounts for 40% and 61% of their AGB and RGB sample, respectively. Although it is still slightly different from the fractions found for our own sample (53% AGB and 76% RGB stars belonging to 2P), it is possible to identify some 2P AGB stars from the Na abundance distributions of both samples (ours and [@MacLean2016]) when the same 1P$-$2P separation criterion is adopted. Moreover, from the \[O/Fe\]$-$\[Na/Fe\] distribution of the AGB stars studied by @Marino2017 we can infer that the Na-rich/O-poor stars account for 47%, which is close to our 2P AGB fraction of 53%. The photometric studies of @Marino2017 and @Lardo2017 also support the claim that NGC6121 hosts multiple populations on AGB. Thus we believe that this cluster should have more than one stellar population along its AGB. Dependencies of the AGB population fraction and theoretical considerations {#subsection:F2ptheory} -------------------------------------------------------------------------- ### Dependencies of the AGB population fraction on the GC parameters {#subsubsection:dependency} We investigate possible dependencies between the fraction of 2P AGB stars (f$_\mathrm{2P,AGB}$) we derived and the GC global parameters listed in Table \[9GCs\_comp\]. We consider \[Fe/H\], age, absolute V magnitude (M$_\mathrm{V}$), mass, HB morphology index (HBR), half-light radius (r$_\mathrm{h}$), tidal radius (r$_\mathrm{t}$), ellipticity, central concentration (c.c.), and central velocity dispersion ($\sigma_\mathrm{v}$). Figure \[relations\_f2P\_GCs\] shows the data points and the linear fits derived by least square fitting with the errors taken into account. The Pearson correlation coefficients for each set of data points are also listed as a reference. When we consider only the five GCs that we have analyzed in a consistent manner (black points and solid lines), we obtain weak positive correlations between f$_\mathrm{2P,AGB}$ and half-light radius; and anticorrelations with GC age (weak), tidal radius, and central concentration. When taking all nine clusters together into account (dashed lines), weak positive relations exist between f$_\mathrm{2P,AGB}$ and half-light radius and ellipticity; and negative relations are shown between f$_\mathrm{2P,AGB}$ and \[Fe/H\] (weak), central concentration, and central velocity dispersion (weak). If the two extreme GCs showing almost no 2P AGB stars (NGC6752 and NGC6266) are disregarded (dash-dotted lines), f$_\mathrm{2P,AGB}$ correlates positively with GC age, HBR, and half-light radius (weak); while the dependencies on \[Fe/H\], M$_\mathrm{V}$, mass, central concentration, and central velocity dispersion are negative. Overall, only the anticorrelation between f$_\mathrm{2P,AGB}$ and central concentration agrees among the fits of all three subsamples. Except for central concentration, the inconsistent results of the three sets of fits indicate that there is no robust dependency for f$_\mathrm{2P,AGB}$ on any other global GC properties. ![Relations between the fraction of 2P AGB stars (f$_\mathrm{2P,AGB}$) and the selected GC global parameters. The data points were all derived using the criteria defined in Sect. \[subsection:criteria\]. Filled and open circles are used, respectively, for the five GCs we analyzed self-consistently (NGC6809, NGC2808, NGC6121, NGC104, and NGC6752) and for those we took from the literature (NGC6205, NGC5986, NGC5904, and NGC6266). The associated error bars are computed from Beta distribution. The lines are the linear fits derived by least square fitting (with the errors taken into account); the solid and dashed lines are fitted from our own five GCs and all the nine GCs, respectively, while the dash-dotted ones represent the fits disregarding the two most scattered points of NGC6266 and NGC6752. The Pearson correlation coefficients considering the five (R5), nine (R9), and seven (R7) data points, respectively, are listed at the bottom of each panel (except for the very bottom right panel, which reports the five (R5), eight (R8), and six (R6) data points, as no $\sigma_\mathrm{v}$ data is available for NGC5986). []{data-label="relations_f2P_GCs"}](Relations_F2pAGB-GCpar_9GCs.eps){width="45.00000%"} ### Theoretical considerations {#subsubsection:theoretical} ![Ratio of the \[Na/Fe\] spreads on the AGB and RGB as a function of GC age. Predictions of @CCWC2016 for the original FRMS scenario are shown for different values of metallicity (colors) and for two values of the $\eta$ parameter adopted for Reimers mass loss along the RGB (0.5 and 0.65, solid and dotted lines respectively). The values derived in the present analysis are shown for the nine GCs (blue, red, and black stars, with the colors of the symbols indicating the theoretical track that has the closest \[Fe/H\] from that of each individual GC). []{data-label="NaFe_Obs-Model"}](NaFe_AGB-to-RGB_Obs-Model.eps){width="45.00000%"} @Charbonnel2013 proposed that the lack of sodium-rich AGB stars in NGC6752 could be due to a correlation between Na and He enrichment in the initial mixture of the low-mass stars we observe today, as predicted in the original FRMS scenario [@Decressin2007a]. They showed that in this framework, 2P stars born with an initial He abundance above a cutoff value (and consequently with an initial Na abundance above a cutoff value) are predicted to miss the AGB and evolve directly towards the white dwarf stage after central He burning, because of the impact of helium on stellar evolution. @CCWC2016 showed that when one assumes the same initial Na$-$He correlation for all GCs within the FRMS framework as well as standard mass-loss rates on the RGB, the maximum Na content expected for 2P stars on the AGB is a function of both the metallicity and the age of GCs. At a given \[Fe/H\], younger clusters are expected to host AGB stars exhibiting a larger Na spread than older clusters; and, at a given age, higher Na dispersion along the AGB is predicted in metal-poor GCs than in the metal-rich ones. This is depicted in Fig. \[NaFe\_Obs-Model\] where we show the model predictions for the ratio between the Na spread on the AGB and that on the RGB as a function of GC age and for different metallicities. Among the nine currently available GCs, we have two pairs of GCs with similar metallicities, that is, NGC2808 and NGC6266, and NGC2808 and NGC6121, with the relatively younger GC in each pair always having larger \[Na/H\] dispersion and 2P-AGB fraction. NGC5986, NGC6205, and NGC6752 have comparable metallicities and slightly different ages, but still the relatively younger GC has more Na-rich 2P AGB stars. Besides, NGC5904, NGC6121, and NGC6266 have almost the same age but slightly different metallicities, with the most metal-poor one showing the largest number of 2P AGB stars. In other words, we have five GCs that lie in the domain where the AGB stars are expected to present large Na abundance dispersions within the FRMS framework. These are NGC104 and NGC2808 that are relatively young and metal-rich, NGC6205 and NGC5986 that have medium age and metallicity, and NGC6809 which is old and metal-poor. On the other hand, according to the FRMS prediction and based on their metallicities and ages, NGC5904, NGC6121, NGC6752, and NGC6266 are expected to have their AGB stars showing smaller Na abundance dispersions compared to their RGB counterparts. Thus, and to the very first order, the observations of these subgroups of GCs are consistent with the trends with age and metallicity predicted by the FRMS scenario. However, there are several important issues that might blur the general trends one tries to identify (see also the discussion in @CCWC2016). First, @Charbonnel2013 and @CCWC2016 showed that the maximum Na abundance expected on the AGB for a given age and metallicity strongly depends on the mass-loss rate adopted on the RGB (see also [@Cassisi2014b]), especially for the oldest GCs and the most metal-rich ones. This is clear from Fig. \[NaFe\_Obs-Model\] where we show the predictions for the ratio of the Na spreads on the AGB and RGB for two values (0.5 and 0.65) of the $\eta$ parameter adopted in Reimers prescription used by @CCWC2016 to compute the mass-loss rate on the RGB. The trends with age and metallicity become stronger when higher mass-loss rates are considered. Additionally, the comparison between the predictions and the observed data for the nine GCs plotted in Fig. \[NaFe\_Obs-Model\] supports variations of the RGB mass-loss efficiency from one GC to another within the range derived for $\eta$ by @McDonaldZijlstra2015 [median and maximum values of 0.477 and 0.65 respectively] to explain the variations of the horizontal branch morphology for 56 Galactic GCs. Second, both the observed and theoretical trends with age and metallicity are very weak (and we showed that in the observational case they depend on the considered subsamples), and scatters and exceptions exist in the data as can be seen in Fig. \[relations\_f2P\_GCs\]. This indicates to some extent that the fraction of 2P AGB stars is not simply affected by one single parameter/factor. Mass loss is certainly an important player as discussed above. Third, although the ages we use here are taken from a single source (except for NGC6266), they were not derived for the \[Fe/H\] values reported in the present paper, which might also introduce some confusion in the derivation of the trends with age and metallicity. Finally, the initial Na$-$He correlation and its dependency with cluster properties is an important parameter to predict the theoretical value of the maximum Na expected on the AGB for a given GC. Unfortunately, this correlation is not constrained yet, which forced @CCWC2016 to assume a similar relation over the metallicity range they investigated. To date, the only GCs for which non-negligible He enrichment were estimated using both photometry and spectroscopic measurements are NGC2808 [@Milone2015b] and NGC6266 [@Milone2015]. However, their Na distributions on the AGB are radically different and more similar to other GCs (NGC6121 and NGC6752, respectively) for which very modest He variations have been estimated ([@Villanova2012] for NGC6121, [@Milone2013] for NGC6752, and [@Nardiello2015a] for both GCs). Therefore, the theoretical interpretation of the data is not straightforward, bearing also in mind that the above-mentioned comparisons depend on the chosen abundance data set (cf. abundance discrepencies between different studies mentioned earlier on). Overall, and as already pointed out in a more general context by @Bastian2015, the broad variety of chemical patterns certainly reveals a high degree of stochasticity that is challenging our understanding of the formation and evolution of GCs and of their stellar populations. Summary {#section:summary} ======= After the claim by C13 that no Na-rich AGB stars were present in NGC6752, several studies were recently devoted to the determination of the Na content of AGB stars in GCs. In this work, we present the first analysis for NGC6809, together with that of a large stellar sample in NGC104 and NGC6121, and a re-analysis of NGC6752. This is complementary to our previous study of NGC2808 (PaperI). For these five GCs, we use a photometric method to derive effective temperature and surface gravity of the sample stars. Equivalent widths of unblended iron lines are measured to determine \[Fe/H\], and the Na abundances are derived via spectrum synthesis, assuming local thermodynamic equilibrium (LTE). Non-LTE corrections are then applied to both and Na abundances, following the prescriptions by @Lind2011, @Bergemann2012, and @Lind2012. We provide a large set of stellar parameters and Na and Fe abundances derived self-consistently for the sample of 254 RGB stars and 145 AGB stars in five GCs. We compare our results with those of the literature, and we include in our discussion the only four other GCs (NGC5904, NGC5986, NGC6205, and NGC6266) for which Na has been studied along the AGB. The total sample of nine clusters covers a large range of \[Fe/H\] ($-$1.86 to $-$0.82), age (11 to 13 Gyr), and global properties. We chose to use the \[Na/H\]$_\mathrm{NLTE}$ data to avoid the influence of the FeI-FeII discrepancy that has different behaviors in RGB and AGB stars, thus hampering the analysis. In addition, we define as Na-rich stars, or 2P stars, the objects that have a \[Na/H\] value above \[Na/H\]$_\mathrm{cri}$ =\[Na/H\]$_\mathrm{min}$+0.3dex, where \[Na/H\]$_\mathrm{min}$ is the minimum Na value derived for the sample (RGB + AGB) in a given cluster and 0.3dex is about one third of the \[Na/H\] spread. This definition agrees well with the general concept that 2P GC stars have higher Na abundance than field stars of similar metallicity. Although different criteria as well as abundance indicators (e.g., \[Na/Fe\] combined with O abundances, @Lapenna2016) can obviously lead to different evaluations of the ratios between 1P and 2P stars, we chose this approach to carry out a consistent analysis of the Na abundance for our large sample of RGB and AGB stars in the nine GCs. Our analysis confirms the complex picture of the Na abundances in the AGB stellar populations in Galactic GCs. Eight out of nine GCs under scrutiny exhibit a lower \[Na/H\] spread on the AGB than on the RGB, with the maximum \[Na/H\] abundance being consistently lower on the AGB than on the RGB. The only exception is the most metal-rich cluster NGC104, whose AGB stars show slightly larger Na spread than its RGB ones. Two clusters, NGC6752 and NGC6266, clearly stand out by showing (almost) no Na-rich (2P) AGB stars among the sample stars analyzed so far. We note, however, that uncertainties still exist since [i.]{} different conclusions have been reported for NGC6752 thus demanding further homogeneous studies in both AGB and RGB stars; and [ii.]{} no firm conclusion can be drawn for NGC6266, due to the paucity of stars scrutinized so far. We then find all possible behaviors between these two extremes. In NGC6121, NGC5904, and NGC2808, the AGB stars occupy about the bottom two thirds of the Na distribution of the RGB stars, with a difference of 0.26, 0.25, and 0.21 dex respectively between the maximum \[Na/H\] value of the AGB and RGB samples, indicating a deficit of very Na-rich AGB stars in these clusters. In the case of NGC5986, instead, the AGB and RGB samples show comparable spreads in \[Na/H\] with the maximum Na abundance of AGB stars being 0.14dex lower than their RGB counterparts, but the AGB sample is still rather limited, thus weakening any drawn conclusion. In NGC6809 the Na abundance of the RGB sample spreads more evenly while the AGB stars tend to be more concentrated, and a 0.12dex lower \[Na/H\]$_\mathrm{max}$ present in the AGB sample compared to that of the RGB one. Finally, NGC6205 seems to have very few 1P AGB stars, and the maximum Na values for its RGB and AGB components are marginally consistent within the observational errors. Linear fits between the fraction of Na-rich 2P AGB stars and the GC parameters reveal that the AGB 2P fraction slightly anticorrelates with GC central concentration, with no conclusive results on possible trends with other GC parameters since they depend on the considered subsamples. By checking the AGB populations of pairs/subgroups of GCs and the trend of AGB Na abundance distributions of the nine GCs, we find that the current data roughly support the theoretical prediction of the original FRMS scenario according to which the initial Na and He abundances were correlated in the original mixture of the GC stars we observe today. However and as underlined in the discussion, this cannot be considered as a strong conclusion. Indeed, the predictions for the evolution of the stellar models along the AGB strongly depend on the mass loss on the RGB, and the initial Na$-$He correlation and its possible dependency with cluster properties is not sufficiently constrained yet. The fact that both the observed and the theoretical trends with age, metallicity, and other global GC properties are mostly very weak, and that scatter and exceptions exist indicate that the fraction of 2P AGB stars is affected by more than one or two factors and is probably subject to stochasticity. YW acknowledges the support from the European Southern Observatory, via its ESO Studentship programme. This work was partly funded by the National Natural Science Foundation of China under grants 1233004 and 11390371, as well as the Strategic Priority Research Program The Emergence of Cosmological Structures of the Chinese Academy of Sciences, Grant No. XDB09000000. CC and WC acknowledge support from the Swiss National Science Foundation (FNS) for the project 200020-159543 Multiple stellar populations in massive star clusters - Formation, evolution, dynamics, impact on galactic evolution. We are indebted to Peter Stetson for kindly providing us with accurate Johnson-Morgan photometry. We thank Simon Campbell and Karin Lind for useful discussions, the latter also for giving us access to her NLTE correction grids. We thank the International Space Science Institute (ISSI, Bern, CH) for welcoming the activities of ISSI Team 271 Massive star clusters across the Hubble Time (2013 - 2016). Finally, we thank the anonymous referees for their useful comments. This work has made use of the VALD database, operated at Uppsala University, the Institute of Astronomy RAS in Moscow, and the University of Vienna. [108]{} natexlab\#1[\#1]{} , A., [Arribas]{}, S., & [Mart[í]{}nez-Roger]{}, C. 1999, , 140, 261 , J., [Piotto]{}, G., [King]{}, I. R., [Bedin]{}, L. R., & [Guhathakurta]{}, P. 2009, , 697, L58 , M., [Grevesse]{}, N., [Sauval]{}, A. J., & [Scott]{}, P. 2009, , 47, 481 , N., [Cabrera-Ziri]{}, I., & [Salaris]{}, M. 2015, , 449, 3333 , N., [Lamers]{}, H. J. G. L. M., [de Mink]{}, S. E., [et al.]{} 2013, , 436, 2398 , N. & [Lardo]{}, C. 2015, , 453, 357 , M., [Lind]{}, K., [Collet]{}, R., [Magic]{}, Z., & [Asplund]{}, M. 2012, , 427, 27 , C., [Campos]{}, F., & [Kepler]{}, S. O. 2013, , 435, 263 , G., [Stetson]{}, P. B., [Sanna]{}, N., [et al.]{} 2008, , 686, L87 , J., [Lorimer]{}, D. R., [Turk]{}, P. J., [et al.]{} 2011, , 742, 51 , S. W., [D’Orazi]{}, V., [Yong]{}, D., [et al.]{} 2013, , 498, 198 , S. W., [MacLean]{}, B. T., [D’Orazi]{}, V., [et al.]{} 2017, ArXiv e-prints \[\] , J. A., [Clayton]{}, G. C., & [Mathis]{}, J. S. 1989, , 345, 245 , E. 2013, , 557, A128 , E. 2014, , 795, L28 , E. 2016, ArXiv e-prints \[\] , E., [Bragaglia]{}, A., [Gratton]{}, R. G., [et al.]{} 2014, , 564, A60 , E., [Bragaglia]{}, A., [Gratton]{}, R. G., [et al.]{} 2006, , 450, 523 , E., [Bragaglia]{}, A., [Gratton]{}, R. G., [et al.]{} 2009, , 505, 117 , E., [Bragaglia]{}, A., [Gratton]{}, R. G., [et al.]{} 2010, , 516, A55 , E., [Gratton]{}, R. G., [Lucatello]{}, S., [Bragaglia]{}, A., & [Bonifacio]{}, P. 2005, , 433, 597 , S. & [Salaris]{}, M. 2014, , 563, A10 , S., [Salaris]{}, M., [Pietrinferni]{}, A., [Vink]{}, J. S., & [Monelli]{}, M. 2014, , 571, A81 , W., [Charbonnel]{}, C., & [Decressin]{}, T. 2015, , 578, A117 , W., [Charbonnel]{}, C., & [Meynet]{}, G. 2016, , 592, A111 , C. 2016, in EAS Publications Series, Vol. 80, EAS Publications Series, ed. E. [Moraux]{}, Y. [Lebreton]{}, & C. [Charbonnel]{}, 177–226 , C. & [Chantereau]{}, W. 2016, , 586, A21 , C., [Chantereau]{}, W., [Decressin]{}, T., [Meynet]{}, G., & [Schaerer]{}, D. 2013, , 557, L17 , C., [Chantereau]{}, W., [Krause]{}, M., [Primas]{}, F., & [Wang]{}, Y. 2014, , 569, L6 , M. J., [Pilachowski]{}, C. A., [Johnson]{}, C. I., [et al.]{} 2014, , 780, 94 , K. M. & [Rees]{}, R. 1990, , 99, 1491 , F., [Ventura]{}, P., [Caloi]{}, V., [et al.]{} 2010, , 715, L63 , S. E., [Pols]{}, O. R., [Langer]{}, N., & [Izzard]{}, R. G. 2009, , 507, L1 , T., [Baumgardt]{}, H., [Charbonnel]{}, C., & [Kroupa]{}, P. 2010, , 516, A73 , T., [Charbonnel]{}, C., & [Meynet]{}, G. 2007, , 475, 859 , T., [Meynet]{}, G., [Charbonnel]{}, C., [Prantzos]{}, N., & [Ekstr[ö]{}m]{}, S. 2007, , 464, 1029 , P. A. & [Hartwick]{}, F. D. A. 2014, , 437, L21 , P. A., [VandenBerg]{}, D. A., [Hartwick]{}, F. D. A., [et al.]{} 2015, , 448, 3314 , A., [D’Antona]{}, F., [Ventura]{}, P., [Vesperini]{}, E., & [McMillan]{}, S. L. W. 2010, , 407, 854 , M., [Ventura]{}, P., [D’Antona]{}, F., [Milone]{}, A., & [Piotto]{}, G. 2010, , 408, 999 , D. A., [M[é]{}sz[á]{}ros]{}, S., [Monelli]{}, M., [et al.]{} 2015, , 815, L4 , R. G., [Bragaglia]{}, A., [Carretta]{}, E., [et al.]{} 2003, , 408, 529 , R. G., [Carretta]{}, E., [Eriksson]{}, K., & [Gustafsson]{}, B. 1999, , 350, 955 , P., [Casagrande]{}, L., [Milone]{}, A. P., [et al.]{} 2017, , 603, A37 , B., [Edvardsson]{}, B., [Eriksson]{}, K., [et al.]{} 2008, , 486, 951 , W. E. 1996, , 112, 1487 , B., [Stetson]{}, P. B., [VandenBerg]{}, D. A., & [Dall’Ora]{}, M. 2012, , 144, 25 , I. I., [Kraft]{}, R. P., [Sneden]{}, C., [et al.]{} 2001, , 122, 1438 , R. G., [de Mink]{}, S. E., [Pols]{}, O. R., [et al.]{} 2013, , 84, 171 , C. I., [Caldwell]{}, N., [Rich]{}, R. M., [et al.]{} 2017, , 842, 24 , C. I., [McDonald]{}, I., [Pilachowski]{}, C. A., [et al.]{} 2015, , 149, 71 , C. I. & [Pilachowski]{}, C. A. 2012, , 754, L38 , P. & [Baumgardt]{}, H. 2015, , 452, 924 , M., [Charbonnel]{}, C., [Decressin]{}, T., [Meynet]{}, G., & [Prantzos]{}, N. 2013, , 552, A121 , M. G. H., [Charbonnel]{}, C., [Bastian]{}, N., & [Diehl]{}, R. 2016, , 587, A53 , D. K., [Smith]{}, G. H., [Bolte]{}, M., [et al.]{} 2011, , 141, 62 , E., [Lardo]{}, C., [Mucciarelli]{}, A., [et al.]{} 2016, , 826, L1 , E., [Mucciarelli]{}, A., [Ferraro]{}, F. R., [et al.]{} 2015, , 813, 97 , E., [Mucciarelli]{}, A., [Lanzoni]{}, B., [et al.]{} 2014, , 797, 124 , C., [Salaris]{}, M., [Savino]{}, A., [et al.]{} 2017, , 466, 3507 , S. S., [Baumgardt]{}, H., [Bastian]{}, N., [et al.]{} 2015, , 804, 71 , S. S., [Brodie]{}, J. P., [Forbes]{}, D. A., & [Strader]{}, J. 2014, , 565, A98 , K., [Asplund]{}, M., [Barklem]{}, P. S., & [Belyaev]{}, A. K. 2011, , 528, A103 , K., [Bergemann]{}, M., & [Asplund]{}, M. 2012, , 427, 50 , A. D. & [van den Bergh]{}, S. 2005, , 360, 631 , B. T., [Campbell]{}, S. W., [De Silva]{}, G. M., [et al.]{} 2016, , 460, L69 , A. & [Meynet]{}, G. 2006, , 448, L37 , M., [Coppola]{}, G., [Bono]{}, G., [et al.]{} 2015, , 808, 50 , A. F., [Milone]{}, A. P., [Przybilla]{}, N., [et al.]{} 2014, , 437, 1609 , A. F., [Milone]{}, A. P., [Yong]{}, D., [et al.]{} 2017, , 843, 66 , A. F., [Villanova]{}, S., [Piotto]{}, G., [et al.]{} 2008, , 490, 625 , I. & [Zijlstra]{}, A. A. 2015, , 448, 502 , A. P. 2015, , 446, 1672 , A. P., [Marino]{}, A. F., [Piotto]{}, G., [et al.]{} 2013, , 767, 120 , A. P., [Marino]{}, A. F., [Piotto]{}, G., [et al.]{} 2015, , 447, 927 , A. P., [Marino]{}, A. F., [Piotto]{}, G., [et al.]{} 2015, , 808, 51 , A. P., [Piotto]{}, G., [Bedin]{}, L. R., [et al.]{} 2012, , 537, A77 , A. P., [Piotto]{}, G., [Bedin]{}, L. R., [et al.]{} 2012, , 744, 58 , M., [Milone]{}, A. P., [Stetson]{}, P. B., [et al.]{} 2013, , 431, 2126 , A., [Lapenna]{}, E., [Massari]{}, D., [Ferraro]{}, F. R., & [Lanzoni]{}, B. 2015, , 801, 69 , A., [Salaris]{}, M., [Lovisi]{}, L., [et al.]{} 2011, , 412, 81 , D., [Milone]{}, A. P., [Piotto]{}, G., [et al.]{} 2015, , 573, A70 , D., [Piotto]{}, G., [Milone]{}, A. P., [et al.]{} 2015, , 451, 312 , A., [Kaluzny]{}, J., [Thompson]{}, I. B., [et al.]{} 1999, , 118, 442 , E., [Rejkuba]{}, M., [Zoccali]{}, M., & [Carrera]{}, R. 2010, , 524, A44 , L., [Alonso]{}, J., [Avila]{}, G., [et al.]{} 2003, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 4841, Instrument Design and Performance for Optical/Infrared Ground-based Telescopes, ed. M. [Iye]{} & A. F. M. [Moorwood]{}, 1682–1693 , L., [Mauas]{}, P., [K[ä]{}ufl]{}, H. U., & [Cacciari]{}, C. 2011, , 531, A35 , G., [Milone]{}, A. P., [Anderson]{}, J., [et al.]{} 2012, , 760, 39 , G., [Milone]{}, A. P., [Bedin]{}, L. R., [et al.]{} 2015, , 149, 91 , G., [Milone]{}, A. P., [Marino]{}, A. F., [et al.]{} 2013, , 775, 15 , N. & [Charbonnel]{}, C. 2006, , 458, 135 , N., [Charbonnel]{}, C., & [Iliadis]{}, C. 2007, , 470, 179 , I. & [Mel[é]{}ndez]{}, J. 2005, , 626, 465 , A., [D’Antona]{}, F., [Cassisi]{}, S., [et al.]{} 2015, , 454, 4197 , P., [Hilker]{}, M., & [Richtler]{}, T. 1999, , 350, 476 , J. C., [Courteau]{}, S., [Graves]{}, G., & [Schiavon]{}, R. P. 2014, , 210, 10 , D. & [Charbonnel]{}, C. 2011, , 413, 2297 , A. & [Glebbeek]{}, E. 2010, , 407, 277 , M. F., [Cutri]{}, R. M., [Stiening]{}, R., [et al.]{} 2006, , 131, 1163 , C. A. 1973, PhD thesis, THE UNIVERSITY OF TEXAS AT AUSTIN. , P. B. 2000, , 112, 925 , P. B. 2005, , 117, 563 , D. A., [Brogaard]{}, K., [Leaman]{}, R., & [Casagrande]{}, L. 2013, , 775, 134 , P. & [D’Antona]{}, F. 2011, , 410, 2760 , P., [D’Antona]{}, F., [Mazzitelli]{}, I., & [Gratton]{}, R. 2001, , 550, L65 , P., [Di Criscienzo]{}, M., [Carini]{}, R., & [D’Antona]{}, F. 2013, , 431, 3642 , S., [Geisler]{}, D., [Piotto]{}, G., & [Gratton]{}, R. G. 2012, , 748, 62 , Y., [Primas]{}, F., [Charbonnel]{}, C., [et al.]{} 2016, , 592, A66 [^1]: Based on observations made with ESO telescopes at the La Silla Paranal Observatory under programme ID 093.D-0818(A). [^2]: Full Tables 3, 5, 7 are only available in electronic form at the CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or via http://cdsweb.u-strasbg.fr/cgi-bin/qcat?J/A+A/. [^3]: $C_{UBI} = (U - B) - (B - I)$ [^4]: A(B) = 4.145E(B-V); A(V) = 3.1E(B-V); A(I) = 1.485E(B-V); A(J) = 0.874E(B-V); A(H) = 0.589E(B-V); A(K) = 0.353E(B-V). [^5]: The choice of the Na doublet has only a negligible impact on the derived abundances ($-$0.027dex, $\sigma$=0.073dex, [[bluer $-$ redder]{}]{} doublet). [^6]: Larger uncertanity may exist for this cluster considering the paucity of stars that have been studied so far [@Lapenna2015].	{ "pile_set_name": "ArXiv" }
--- abstract: 'We consider retrieving a specific temporal segment, or moment, from a video given a natural language text description. Methods designed to retrieve whole video clips with natural language determine what occurs in a video but not when. To address this issue, we propose the Moment Context Network ([MCN]{}) which effectively localizes natural language queries in videos by integrating local and global video features over time. A key obstacle to training our [MCN]{} model is that current video datasets do not include pairs of localized video segments and referring expressions, or text descriptions which uniquely identify a corresponding moment. Therefore, we collect the Distinct Describable Moments ([DiDeMo]{}) dataset which consists of over 10,000 unedited, personal videos in diverse visual settings with pairs of localized video segments and referring expressions. We demonstrate that [MCN]{} outperforms several baseline methods and believe that our initial results together with the release of [DiDeMo]{} will inspire further research on localizing video moments with natural language.' author: - \| Lisa Anne Hendricks$^{1\thanks{Work done at Adobe Research during LAH's summer internship}}$, Oliver Wang$^2$, Eli Shechtman$^2$, Josef Sivic$^{2,3\footnotemark[1]}$, Trevor Darrell$^1$, Bryan Russell$^2$\ $^1$UC Berkeley, $^2$Adobe Research, $^3$INRIA\ [<https://people.eecs.berkeley.edu/~lisa_anne/didemo.html>]{} title: Localizing Moments in Video with Natural Language --- Acknowledgements {#acknowledgements .unnumbered} ================ TD was supported by DARPA, AFRL, DoD MURI award N000141110688, NSF awards IIS-1427425, IIS-1212798, and the Berkeley AI Research (BAIR) Lab. Supplemental {#supplemental .unnumbered} ============	{ "pile_set_name": "ArXiv" }
--- abstract: \| Using heat kernel estimates, we prove the pathwise uniqueness for strong solutions of irregular stochastic differential equation driven by a family of Markov process, whose generator is a non-local and non-symmetric Lévy type operator. Due to the extra term $1_{[0,\sigma(X_{s-},z)]}(r)$ in multiplicative noise, we need to derive some new regularity results for the generator and use a trick of mixing $L_1$ and $L_2$-estimates by Kurtz and Protter [@Ku-Po]. [[Keywords and Phrases:]{} Heat kernel estimates, non-local operator, irregular SDE, pathwise uniqueness]{} address: - \| Longjie Xie: School of Mathematics and Statistics, Jiangsu Normal University, Xuzhou, Jiangsu 221000, P.R.China\ Email: [email protected] - \| Lihu Xu: 1. Department of Mathematics, Faculty of Science and Technology, University of Macau, Av. Padre Tomás Pereira, Taipa Macau, China. 2. UM Zhuhai Research Institute, Zhuhai, 519080, China\ Email: [email protected], [email protected] author: - '[Longjie Xie]{} and [Lihu Xu]{}' title: Irregular Stochastic differential equations driven by a family of Markov processes --- [^1] Introduction ============ Nowadays, much attentions have been paid to the non-local operators and their corresponding pure jump processes, as these processes are more realistic models for many practice applications. Consider the following non-local and non-symmetric Lévy type operator: $$\begin{aligned} \sL f(x):=\sL_\nu^\sigma f(x)+b(x)\cdot\nabla f(x),\quad\forall f\in C^\infty_0(\mR^d), \label{oper}\end{aligned}$$ where $b(x)$ is a measurable function and $$\begin{aligned} \sL^\sigma_{\nu} f(x):=\int_{\mR^d}\big[f(x+z)-f(x)-1_{\{\|z\|\leq 1\}}z\cdot\nabla f(x)\big]\sigma(x,z)\nu(\dif z).\end{aligned}$$ Here, $\nu$ is a Lévy measure on $\mR^d$ satisfying $$\int_{\mR^d\setminus \{0\}}(\|z\|^2\wedge 1)\nu(\dif z)<\infty,$$ and $\sigma:\mR^d\times\mR^d\rightarrow\mR$ is measurable. The operator $\sL$ is a non-local version of the classical second order elliptic operator with non-divergence form and has been intensely studied in the last decade by people in the community of analysis and PDEs, see [@E-I-K; @Ko] and the references therein. While from the probability point of view, it is known via the martingale method (see[@M-P]) that under certain assumptions on $\nu, \sigma$ and $b$, there exists a Markov process $X_t$ with $\sL$ as its generator, and the measure $\nu$ describes the jumps of $X_t$. It is natural to ask wether one can construct $X_t$ via the Itô’s calculus so that we can have another look at $\sL$ from the view of stochastic differential equations (SDEs). However, the classical SDE driven by pure jump Lévy process is not very suitable (see more discussions in [@Xie Section 1]). Its connection to SDE was found very recently. To specify the SDE that we are going to study, let $\cN(\dif z,\dif r,\dif t)$ be a Poisson random measure on $\mR^d\times[0,\infty)\times[0,\infty)$ with intensity measure $\nu(\dif z)\dif r\dif t$, and $\tilde \cN(\dif z,\dif r,\dif t):=\cN(\dif z,\dif r,\dif t)-\nu(\dif z)\dif r\dif t$ is the compensated Poisson random measure. Then, the Markov process $X_t$ corresponding to $\sL$ should satisfy the following SDE: $$\begin{aligned} \dif X_t&=\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}\!\!1_{[0,\sigma(X_{s-},z)]}(r)z\tilde \cN(\dif z,\dif r,\dif t)\no\\ &\quad+\int_0^{\infty}\!\!\!\!\int_{\|z\|> 1}\!\!1_{[0,\sigma(X_{s-},z)]}(r)z \cN(\dif z,\dif r,\dif t)+b(X_t)\dif t, \ \ \ \ X_0=x\in\mR^d. \label{sde2}\end{aligned}$$ In fact, noticing that for a function $f$ and any $r>0$, $$\begin{aligned} f\big(x+1_{[0,\sigma(x,z)]}(r)z\big)-f(x)=1_{[0,\sigma(x,z)]}(r)\big[f(x+z)-f(x)\big],\end{aligned}$$ an application of Itô’s formula shows that the generator of the solution to SDE (\[sde2\]) is given exactly by (\[oper\]). Note that the driven noise is a Markov process but not necessarily Lévy type [@Kurz2]. This makes such kind of SDEs more interesting and are worthy of study. Under the conditions that $b$ is bounded and global Lipschitz continuous, and $\sigma$ is bounded with $$\begin{aligned} \int_{\mR^d}\|\sigma(x,z)-\sigma(y,z)\|\cdot\|z\|\nu(\dif z)\leq C_1\|x-y\|, \ \ \ \ \forall \ \ x, y \in \mR^d, \label{kur}\end{aligned}$$ and some other assumptions, Kurtz [@Kurz2] showed the existence and uniqueness of strong solution to SDE , see also [@Im-Wi; @Ku-Po] for related results and applications. Our aim in this paper is to prove that SDE admits a unique strong solution under some weak assumptions on the coefficients $\sigma$ and $b$ as well as the jump measure $\nu$, from which we can see the regularization effect of such kind of noises on the deterministic systems. The irregular SDEs driven by pure jump noises have been extensively studied in the past several decades. Note that when $d=1$ and $L_t$ is a symmetric $\alpha$-stable process with $\alpha<1$, Tanaka, Tsuchiya and Watanabe [@Ta-Ts-Wa] showed that if $b$ is bounded and $\beta$-Hölder continuous with $\alpha+\beta<1$, SDE $$\begin{aligned} \dif X_t=\dif L_t+b(X_t)\dif t,\quad X_0=x\in\mR^d \label{levy}\end{aligned}$$ may not have pathwise uniqueness strong solutions. When $\alpha\geq 1$, $b$ is bounded and $\beta$-Hölder continuous with $\beta>1-\frac{\alpha}{2},$ it was proved by Priola [@Pri] that there exists a unique strong solution $X_t(x)$ to SDE (\[levy\]) for each $x\in\mR^d$. Recently, Zhang [@Zh00] obtained the pathwise uniqueness to SDE (\[levy\]) when $\alpha>1$ and $b$ is a bounded function in some local Sobolev space. See also [@B-B-C; @Ch-So-Zh; @Pri2] for related results. We also would like to mention the paper [@M-X] where SDEs driven by multiplicative Lévy noise with Lipschitz diffusion coefficient and Hölder drift was considered. For the study of irregular SDEs driven by Brownian motion, we refer readers to [@Fa-Lu-Th; @Fe-Fl-2; @Fl-Gu-Pr; @Kr-Ro; @M-N-P-Z; @W; @Wa; @XZ; @Zh3; @Zh1]. Let us compare our results with the literatures above. To prove the uniqueness of strong solution, we shall follow a well known strategy [@Ch-So-Zh; @Kr-Ro; @Pri; @Zh1], which is to derive a new SDE with better coefficients by Zvonkin’s transformation and get the uniqueness of the original equation from the new one. The crucial point of this approach is to study the regularity of transformation equations which vary with different SDEs. There are several new aspects that we would like to stress for SDE as the following. First of all, our main tool for studying the transformation equation (see below) is the heat kernel (also called fundamental solution) of the operator $\sL_\nu^\sigma$, it seems the first time to use heat kernel estimates to study the pathwise uniqueness irregular SDEs, see [@C-K-K; @K-S] for the study of weak uniqueness of SDEs with Lévy noise by using heat kernels. Secondly, all the above works are for singular SDEs driven by Brownian motions or additive Lévy noises, in the latter case, one only needs to study the operator $\sL_0$ defined by $$\sL_0 f(x):=\int_{\mR^d}\Big[f(x+z)-f(x)-1_{\{\|z\|\leq 1\}}z\cdot\nabla f(x)\Big]\nu(\dif z),\quad\forall f\in C_0^\infty(\mR^d).$$ The analysis relies on the nice property of $\sL_0$ and the $C^2$ smoothing effect of its semigroup. However, we study the multiplicative noise and the semigroup generated by $\sL_{\nu}^\sigma$ only has $C^{\alpha+\beta}$ regularity with $\alpha+\beta<2$ (see Remark \[ree\]), we need to use more delicate analysis and interpolation theorems to fit our less regularity property into the frame of Zvonkin’s argument. We mention that in [@Xie], the first author consider the same SDE with critical case $\alpha=1$ and $b$ in Hölder spaces, here we shall consider $\alpha\in(1,2)$ but with a more irregular drift term $b$ in fractional Sobolev spaces, and the proof in this paper is more involved. Lastly, when proving the Krylov-type estimate and performing Zvonkin transformation, we needs to solve a semi-linear elliptic equation and the resolvent equation of $\sL_\nu^\sigma$ in the framework of Sobolev space. Because a well developed elliptic equation theory as in [@Kr-Ro; @Zh3; @Zh1] is not available for $\sL_\nu^\sigma$, we derive a generalized Itô’s formula for Hölder functions and solve the corresponding integral equation in Sobolev spaces. Another novelty in our analysis is the technique for handling the extra term $1_{[0,\sigma(X_{s-},z)]}(r)$ when we prove the pathwise uniqueness in the last section. The usual $L_2$-estimate in the known literatures is not applicable. Fortunately we can use a trick of mixing $L_1$ and $L_2$-estimate as the replacement [@Kurz2; @Ku-Po]. Due to the irregularity of $b$ and $\sigma$, it is much more complicated than [@Kurz2] to apply this trick. Finally, we mention that studying the unique strong solution of SDE with irregular coefficients not only has its own interests but also helps to better understand the nonlocal operator $\sL$ ([@Im-Wi]). Another motivation for studying SDE is because of the special noise. As mentioned above, the driven noise is a Markov process but not necessarily Lévy type. This has been found very useful in applications, for instance, Markov type noise plays a crucial role as the control when proving Freidlin-Wentzell type large deviation for Lévy type SDEs via weak convergence approach [@BDM11; @BCD13; @ZhZh15]. The organization of the paper is as the following. Section 2 gives the main result with some comments and comparisons with known literatures. Sections 3 and 4 are both preparation sections, the former for some estimates of heat kernel of $\sL_\nu^\sigma$ and the latter for the regularity of the corresponding semigroup $\cT_t$. Krylov’s estimate and Zvonkin’s transformation are studied in the 5th section and applied in the last section to prove the strong uniqueness of SDE . Throughout this paper, we use the following convention: $C$ with or without subscripts will denote a positive constant, whose value may change in different places, and whose dependence on parameters can be traced from calculations. We would like to gratefully thank Professors Rengming Song and Xicheng Zhang for very helpful discussions. Main result =========== We assume that for all $x\in\mR^d$, $$\begin{aligned} \sigma(x,z)=\sigma(x,-z),\quad\forall z\in\mR^d, \label{s2}\end{aligned}$$ and that there exists a function $\tilde\kappa$ such that $$\begin{aligned} \nu(\dif z)=\frac{\tilde\kappa(z)}{\|z\|^{d+\alpha}}\dif z, \quad \tilde\kappa(z)=\tilde\kappa(-z),\quad \kappa_0\leq \tilde\kappa(z)\leq \kappa_1,\label{nu}\end{aligned}$$ with $\alpha\in(1,2)$ and $\kappa_0, \kappa_1$ are two positive constants. The symmetric in $z$ of $\sigma$ and $\tilde\kappa$ is a common assumption in the literature, see [@Ca-Si]. As a result, we can also write $\sL_\nu^\sigma$ as $$\begin{aligned} \sL^\kappa_{\alpha}\varphi(x)&=\text{p.v.}\int_{\mR^d}\big[\varphi(x+z)-\varphi(x)\big]\sigma(x,z)\nu(\dif z)\no\\ &=\text{p.v.}\int_{\mR^d}\big[\varphi(x+z)-\varphi(x)\big]\frac{\kappa(x,z)}{\|z\|^{d+\alpha}}\dif z,\label{op}\end{aligned}$$ where $$\begin{aligned} \kappa(x,z)=\sigma(x,z)\tilde\kappa(z). \label{kappa}\end{aligned}$$ The operator $\sL^\kappa_{\alpha}$ is a non-local and non-symmetric operator, which can be seen as a generalization of variable coefficients fractional Laplacian operator. For brevity, we set $B_n:=\{x\in\mR^d:\|x\|\leq n\}$. Our main result is: \[main\] Let (\[s2\]) hold and the Lévy measure $\nu$ satisfies (\[nu\]). Suppose that for any $n\in\mN$: 1. There exists a function $\zeta\in L^{q}(B_n)$ with $q>d/\alpha$, such that for almost all $x,y\in B_n$, $$\begin{aligned} \int_{\mR^d}\|\sigma(x,z)-\sigma(y,z)\|(\|z\|\wedge1)\nu(\dif z)\leq \|x-y\|\Big(\zeta(x)+\zeta(y)\Big),\label{a1}\end{aligned}$$ and for some constants $k^n_0, k^n_1>0$, $\beta\in(0,1)$ and $C_n>0$, $$\begin{aligned} k^n_0\leq \sigma(x,z)\leq k^n_1,\quad\|\sigma(x,z)-\sigma(y,z)\|\leq C_n\|x-y\|^{\beta},\quad \forall x,y\in B_n,\,\,\forall z\in\mR^d.\label{ho}\end{aligned}$$ 2. For some $\theta\in (1-\frac{\alpha}{2},1)$ and $p>2d/\alpha$, $$\begin{aligned} \int_{B_n}\!\int_{B_n}\frac{\|b(x)-b(y)\|^p}{\|x-y\|^{d+\theta p}}\dif x\dif y<+\infty,\end{aligned}$$ and it holds $$\begin{aligned} \sup_{x\in B_n}\|b(x)\|<\infty.\end{aligned}$$ Then, for each $x\in\mR^d$, there exists an stopping time $\varsigma(x)$ (called the explosion time) and a unique strong solution $X_t(x)$ to SDE (\[sde2\]) such that $$\begin{aligned} \lim_{t\uparrow\varsigma(x)}X_t(x)=\infty,\quad a.s.. \label{xt}\end{aligned}$$ Let us make some comments on the assumptions and give an example for our result with a comparison with known literatures. (1). It is clear that the assumption (\[a1\]) is a generalization of condition (\[kur\]) in [@Kurz2]. For an very interesting example of $\sigma$, we can take $$\sigma(x,z)=K(z)+\tilde\sigma(x)\|z\|^{\gamma} \ \ {\rm for} \ \ \|z\| \le 1, \ \ \ \ \ \sigma(x,z)=K(z)+\tilde\sigma(x) \ \ {\rm for} \ \ \|z\|>1,$$ with $0<K_1\leq K(z)\leq K_2$, $\gamma>\alpha-1$ and $\nabla\tilde\sigma\in L^q_{loc}(\mR^d)$ with $q>d/\alpha$, where $\nabla$ denotes the weak derivative. Since we assume $\alpha>1$, our theorem can cover the regime $q \in (d/\alpha, d]$. However, for the following SDE driven by multiplicative Brownian motion [@Zh3]: $$\dif X_t=\sigma(X_t)\dif W_t+b(X_t)\dif t,\quad X_0=x\in\mR^d,$$ one has to assume that $\nabla\sigma\in L^q(\mR^d)$ with $q>d$. Here, the main trick is that $\sigma$ appears in the term $1_{[0,\sigma(X_{s-},z)]}(r)$. When $\nabla\tilde\sigma\in L^q_{loc}(\mR^d)$ with $q>d$, we can also have the Hölder continuity in (\[ho\]) by the Sobolev embedding theorem. (2). For interesting examples of irregular drift coefficient $b$, we can take $b(x)=1_A(x)$ for a measurable set $A \in \mR^d$, see [@Zh00 Remark 1.2] for details. (3). The conditions (\[s2\]), (\[nu\]) and (\[ho\]) are assumed so that we can use the results obtained in [@Ch-Zh]. Under (\[s2\]), (\[nu\]) and the global assumptions $$\begin{aligned} 0<\tilde k_0\leq \sigma(x,z)\leq \tilde k_1,\,\,\,\|\sigma(x,z)-\sigma(y,z)\|\leq C_0\|x-y\|^{\beta},\quad \forall x,y\in \mR^d,\,\,\forall z\in\mR^d, \label{s1}\end{aligned}$$ it was proved that there exists a unique fundamental solution $p(t,x,y)$ for $\sL^\kappa_\alpha$, see [@Ch-Zh Theorem 1.1]. Here, we only need the local boundness and the local Hölder continuity of $\sigma$ in (\[ho\]) thanks to the stopping time technique. Furthermore, we shall prove better regularities of $p(t,x,y)$ (see Theorem \[seen\]) than obtained in [@Ch-Zh], which seem to be new and have independent interests. Heat kernel estimates ===================== We briefly recall the construction of the heat kernel $p(t,x,y)$ for operator $\sL^\kappa_{\alpha}$ in [@Ch-Zh], from which we derive some important estimates of $p(t,x,y)$ (see Theorem \[seen\] below) for further use in next sections. From now on, we assume that (\[s2\]), (\[nu\]) and (\[s1\]) always hold. First of all, in view of (\[s2\]) (\[nu\]) and (\[op\]), we can also write $$\begin{aligned} \sL^\kappa_{\alpha} \varphi(x)=\frac{1}{2}\int_{\mR^d}\delta_\varphi(x,z)\frac{\kappa(x,z)}{\|z\|^{d+\alpha}}\dif z,\quad\forall \varphi\in C^\infty_0(\mR^d), \label{opp}\end{aligned}$$ where $$\delta_\varphi(x,z):=\varphi(x+z)+\varphi(x-z)-2\varphi(x).$$ In order to reflect the dependence of $\kappa$ with respect to $x$, we shall also use $\sL_\alpha^{\kappa,x}$ instead of $\sL_\alpha^\kappa$. To shorten the notation, we set for $\gamma, \beta\in \mR$, $$\varrho_\gamma^\beta(t,x):=t^{\frac{\gamma}{\alpha}}\big(\|x\|^{\beta}\wedge 1\big)\big(\|x\|+t^{1/\alpha}\big)^{-d-\alpha}.$$ The following 3-P type inequalities shall be used below from time to time. (i). If $\gamma_1+\beta_1>0$ and $\gamma_2+\beta_2>0$, then there exists a constant $C_1>0$ such that for all $t\geq 0$ and $x,y\in\mR^d$, $$\begin{aligned} &\int_0^t\!\!\!\int_{\mR^d}\varrho_{\gamma_1}^{\beta_1}(t-s,x-z)\varrho_{\gamma_2}^{\beta_2}(s,z-y)\dif z\dif s\no\\ &\leq C_1\Big(\varrho^{0}_{\gamma_1+\gamma_2+\beta_1+\beta_2}+\varrho_{\gamma_1+\gamma_2+\beta_2}^{\beta_1}+ \varrho^{\beta_2}_{\gamma_1+\gamma_2+\beta_1}\Big)(t,x-y). \label{3p}\end{aligned}$$ (ii). For all $\beta_1,\beta_2\in[0,\alpha]$ and $\gamma_1,\gamma_2\in\mR$, there exists a constant $C_2>0$ such that for any $t\geq 0$ and $x\in\mR^d$, $$\begin{aligned} &\int_{\mR^d}\varrho_{\gamma_1}^{\beta_1}(t,x-z)\varrho_{\gamma_2}^{\beta_2}(t,z)\dif z\no\\ &\leq C_2\Big(\varrho^{0}_{\gamma_1+\gamma_2+\beta_1+\beta_2-\alpha}+\varrho_{\gamma_1+\gamma_2+\beta_2-\alpha}^{\beta_1}+ \varrho^{\beta_2}_{\gamma_1+\gamma_2+\beta_1-\alpha}\Big)(t,x). \label{3p1}\end{aligned}$$ The first inequality is given by [@Ch-Zh Lemma 2.1 (iii)], while the second one can be proved entirely by the same arguments as [@Ch-Zh Lemma 2.1 (ii)], the details are omitted. Let $p_{\alpha}(t,x)$ denote the heat kernel of operator $\Delta^{\frac{\alpha}{2}}$ (or equivalently, the transition density of $d$-dimensional symmetric $\alpha$-stable process). It is well known that there exists a constant $C_0$ such that $$\begin{aligned} C_0^{-1}\varrho_\alpha^0(t,x)\leq p_{\alpha}(t,x)\leq C_0\varrho_\alpha^0(t,x), \label{k0}\end{aligned}$$ and for every $k\in\mN$, it holds for some $C_k$ that $$\begin{aligned} \|\nabla^k p_{\alpha}(t,x)\|\leq C_k\varrho_{\alpha-k}^0(t,x). \label{kk}\end{aligned}$$ Set for $z\in\mR^d$, $$\begin{aligned} \delta_{p_\alpha}(t,x;z):=p_\alpha(t,x+z)+p_\alpha(t,x-z)-2p_\alpha(t,x).\end{aligned}$$ It was shown by [@Ch-Zh Lemma 2.2] that that there exists a constant $C>0$ such that $$\begin{aligned} \|\delta_{p_\alpha}(t,x;z)\|\leq C\Big((t^{-\frac{2}{\alpha}}\|z\|^2)\wedge 1\Big)\Big(\varrho^0_\alpha(t,x\pm z)+\varrho^0_\alpha(t,x)\Big).\label{de}\end{aligned}$$ With this estimate in hand and following the same ideas as in the proof of [@Ch-Zh Theorem 2.4], we can derive the fractional derivative estimate of $p_{\alpha}(t,x)$. For completeness, we sketch the details here. \[fe\] For any $0<\gamma<2$, there exists a constant $C_\gamma$ such that $$\begin{aligned} \|\Delta^{\frac{\gamma}{2}}p_{\alpha}(t,x)\|\leq C_\gamma\varrho^0_{\alpha-\gamma}(t,x). \label{e1}\end{aligned}$$ We may assume that $t\leq 1$, since the general case follows by the Chapman-Kolmogorov equation. By the definition of fractional Laplacian and as (\[opp\]), we can write for $0<\gamma<2$, $$\begin{aligned} \Delta^{\frac{\gamma}{2}}p_{\alpha}(t,x)=\frac{c_{d,\gamma}}{2}\int_{\mR^d}\delta_{p_\alpha}(t,x;z)\frac{1}{\|z\|^{d+\gamma}}\dif z.\label{22}\end{aligned}$$ Consequently, we have by (\[de\]) $$\begin{aligned} \|\Delta^{\frac{\gamma}{2}}p_\alpha(t,x)\|&\leq C_{d,\gamma}\int_{\mR^d}\|\delta_{p_\alpha}(t,x;z)\|\cdot\|z\|^{-d-\gamma}\dif z\\ &\leq C_1\varrho^0_\alpha(t,x)\int_{\mR^d}\Big((t^{-\frac{2}{\alpha}}\|z\|^2)\wedge 1\Big)\|z\|^{-d-\gamma}\dif z\\ &\quad+C_1\!\!\int_{\mR^d}\Big((t^{-\frac{2}{\alpha}}\|z\|^2)\wedge 1\Big)\varrho^0_\alpha(t,x\pm z)\|z\|^{-d-\gamma}\dif z=:I_1+I_2.\end{aligned}$$ For $I_1$, by the assumption that $\gamma<2$, one can check easily that $$\begin{aligned} I_1&=C_1\varrho^0_{\alpha-2}(t,x)\!\int_{\|z\|\leq t^{1/\alpha}}\|z\|^{2-d-\gamma}\dif z+C_1\varrho^0_{\alpha}(t,x)\!\int_{\|z\|> t^{1/\alpha}}\|z\|^{-d-\gamma}\dif z\leq C_2\varrho^0_{\alpha-\gamma}(t,x).\end{aligned}$$ As for the second term, similarly we write $$\begin{aligned} I_2&=C_1t^{-\frac{2}{\alpha}}\!\!\int_{\|z\|\leq t^{1/\alpha}}\varrho^0_\alpha(t,x\pm z)\|z\|^{2-d-\gamma}\dif z+C_1\!\!\int_{\|z\|> t^{1/\alpha}}\varrho^0_\alpha(t,x\pm z)\|z\|^{-d-\gamma}\dif z=:I_{21}+I_{22}.\end{aligned}$$ We further control $I_{21}$ by $$\begin{aligned} I_{21}&\leq C_3\varrho^0_{\alpha-2}(t,x)\!\!\int_{\|z\|\leq t^{1/\alpha}}\|z\|^{2-d-\gamma}\dif z\leq C_4\varrho^0_{\alpha-\gamma}(t,x).\end{aligned}$$ For $I_{22}$, if $\|x\|\leq 2t^{1/\alpha}$, then $$\begin{aligned} I_{22}&\leq C_3t^{-\frac{d}{\alpha}}\!\int_{\|z\|> t^{1/\alpha}}\|z\|^{-d-\gamma}\dif z\leq C_3t^{-\frac{d+\gamma}{\alpha}}\leq C_4\varrho^0_{\alpha-\gamma}(t,x).\end{aligned}$$ If $\|x\|> 2t^{1/\alpha}$, we can deduce that $$\begin{aligned} I_{22}&\leq C_1\left(\int_{\frac{\|x\|}{2}>\|z\|> t^{1/\alpha}}+\int_{\|z\|> \frac{\|x\|}{2}}\right)\varrho^0_\alpha(t,x\pm z)\|z\|^{-d-\gamma}\dif z\\ &\leq C_2t\int_{\frac{\|x\|}{2}>\|z\|> t^{1/\alpha}}\big(\|x\pm z\|+t^{1/\alpha}\big)^{-d-\alpha}\|z\|^{-d-\gamma}\dif z+C_2\|x\|^{-d-\gamma}\int_{\|z\|> \frac{\|x\|}{2}} \varrho^0_\alpha(t,x\pm z)\dif z\\ &\leq C_3\varrho^0_{\alpha}(t,x)\int_{\|z\|> t^{1/\alpha}}\|z\|^{-d-\gamma}\dif z+C_3\|x\|^{-d-\gamma}\leq C_4\varrho^0_{\alpha-\gamma}(t,x).\end{aligned}$$ Combing the above computations, we get (\[e1\]). Now we fix $y\in\mR^d$, consider the freezing operator $$\sL_{\alpha}^{\kappa,y}f(x):=\text{p.v.}\int_{\mR^d}[f(x+z)-f(x)]\frac{\kappa(y,z)}{\|z\|^{d+\alpha}}\dif z,$$ where $\kappa$ is given by (\[kappa\]). It is known that there exists a symmetric $\alpha$-stable like process corresponding to $\sL_{\alpha}^{\kappa,y}$. Let $p_y(t,x)$ be the heat kernel of operator $\sL_{\alpha}^{\kappa,y}$. Since $\kappa$ is uniformly bounded, it follows from [@Ch-Ku Theorem 1.1] that for some constant $C_0$ independent of $y$, $$\begin{aligned} C_0^{-1}\varrho_\alpha^0(t,x)\leq p_{y}(t,x)\leq C_0\varrho_\alpha^0(t,x). \label{p0}\end{aligned}$$ Moreover, if we set $$\hat\kappa(y,z):=\kappa(y,z)-\frac{\tilde k_0\kappa_0}{2},$$ where $\tilde k_0, \kappa_0$ are the constants in (\[s1\]) and (\[nu\]), respectively, and let $\hat p_y(t,x)$ be the heat kernel of operator $\sL_{\alpha}^{\hat\kappa,y}$, by the construction of Lévy process, we can write $$\begin{aligned} p_y(t,x)=\int_{\mR^d}p_{\alpha}(\tfrac{\tilde k_0\kappa_0}{2}t,x-z)\hat p_y(t,z)\dif z, \label{p00}\end{aligned}$$ see also [@Ch-Zh (2.23)]. The advantage of (\[p00\]) is that we can derive certain estimates for $p_y(t,x)$ by using properties of $p_\alpha(t,x)$. As an easy result, we have the following fractional derivative estimate of $p_y(t,x)$ and the Hölder continuity of $\nabla p_y(t,x)$. Here and below, both operators $\Delta^{\frac{\gamma}{2}}$ and $\nabla$ are acted with respect to the variable $x$. For any $0<\gamma<2$, it holds $$\begin{aligned} \|\Delta^{\frac{\gamma}{2}}p_{y}(t,x)\|\leq C_\gamma\varrho^0_{\alpha-\gamma}(t,x). \label{e2}\end{aligned}$$ and for any $\vartheta\in(0,1)$, $t>0$ and all $x,x',y\in\mR^d$, $$\begin{aligned} \|\nabla p_y(t,x)-\nabla p_y(t,x')\|\leq C_\vartheta\|x-x'\|^\vartheta\varrho_{\alpha-1-\vartheta}^0(t,\tilde x). \label{p0h}\end{aligned}$$ where $C_\gamma, C_\vartheta$ are positive constants independent of $y$, and $\tilde x$ is the one of the two points $x$ and $x'$ which is nearer to zero point. It is enough to prove the estimates with $t\in(0,1)$. The first assertion can be verified by using the Fubini’s theorem, (\[k0\]), (\[e1\]), (\[p0\]), (\[p00\]) (\[3p1\]) and easy computations. As for the second inequality, without lose of generality, we may assume that $\|x\|\leq \|x'\|$. In view of (\[kk\]), we know that when $\|x-x'\|\geq t^{\frac{1}{\alpha}}/2$, $$\begin{aligned} \|\nabla p_\alpha(t,x)-\nabla p_\alpha(t,x')\|&\leq C_1\|x-x'\|^\vartheta\Big(\varrho_{\alpha-1-\vartheta}^0(t,x)+\varrho_{\alpha-1-\vartheta}^0(t,x')\Big)\\ &\leq C_1\|x-x'\|^\vartheta\varrho_{\alpha-1-\vartheta}^0(t,x).\end{aligned}$$ While for $\|x-x'\|< t^{\frac{1}{\alpha}}/2$, we have by the mean value theorem that for some $\varepsilon\in[0,1]$, $$\begin{aligned} \|\nabla p_\alpha(t,x)-\nabla p_\alpha(t,x')\|&\leq C_2\|x-x'\|\varrho_{\alpha-2}^0\big(t,x+\varepsilon(x'-x)\big)\\ &\leq C_2\|x-x'\|\varrho_{\alpha-2}^0(t,x)\leq C_2\|x-x'\|^\vartheta\varrho_{\alpha-1-\vartheta}^0(t,x).\end{aligned}$$ The desired estimate (\[p0h\]) follows by (\[p00\]), (\[p0\]) and (\[3p1\]). Let $\beta$ be the Hölder index in (\[s1\]). Below, we always suppose that $\alpha+\beta<2$. This is assumed just to simplify the proof and is in fact not an restriction at all. Indeed, since we also assumed that $\sigma$ is bounded, (\[s1\]) still holds true for any $\beta'<\beta$. Hence, it is enough to study the pathwise uniqueness of SDE (\[sde2\]) when $\beta<2-\alpha$. We show the following estimate. \[as\] Under (\[s1\]), we have for $\gamma\in(0,2-\alpha)$ and all $x\in\mR^d$, $$\begin{aligned} \bigg\|\int_{\mR^d}\Delta^{\frac{\alpha+\gamma}{2}}p_y(t,x-y)\dif y\bigg\|\leq C_{d,\alpha,\gamma}t^{\frac{\beta-\gamma}{\alpha}-1},\label{00}\end{aligned}$$ and for any $\vartheta\in(0,1)$ and $x, x'\in\mR^d$, $$\begin{aligned} \bigg\|\int_{\mR^d}\Big[\nabla p_y(t,x-y)-\nabla p_y(t,x'-y)\Big]\dif y\bigg\|\leq C_{d,\vartheta}\|x-x'\|^\vartheta t^{\frac{\beta-\vartheta-1}{\alpha}},\label{000}\end{aligned}$$ where $C_{d,\alpha,\gamma}, C_{d,\vartheta}$ are positive constants. Since $\hat p_y(t,x)$ is a density function of Markov process, we have for any $\xi\in\mR^d$, $$\int_{\mR^d}\hat p_\xi(t,x-y)\dif y=1.$$ Combing this with (\[22\]), (\[p00\]) and using the Fubini’s theorem, it is easily checked that $$\begin{aligned} \int_{\mR^d}\!\!\int_{\mR^d}\delta_{p_\xi}(t,x-y;z)\frac{c_{d,\alpha,\gamma}}{\|z\|^{d+\alpha+\gamma}}\dif z\dif y&=\int_{\mR^d}\Delta^{\frac{\alpha+\gamma}{2}}p_\xi(t,x-y)\dif y\\ &=\int_{\mR^d}\!\!\int_{\mR^d}\Delta^{\frac{\alpha+\gamma}{2}}p_\alpha(\tfrac{\tilde k_0\kappa_0}{2}t,x-y-z)\hat p_{\xi}(t,z)\dif z\dif y\\ &\!\!\!\stackrel{y+z=\tilde z}{=}\int_{\mR^d}\Delta^{\frac{\alpha+\gamma}{2}}p_\alpha(\tfrac{\tilde k_0\kappa_0}{2}t,x-z)\dif z=0,\quad \forall \xi\in\mR^d.\end{aligned}$$ As a result, we can write $$\begin{aligned} \sE_1:=\int_{\mR^d}\Delta^{\frac{\alpha+\gamma}{2}}p_y(t,x-y)\dif y&=\int_{\mR^d}\!\!\int_{\mR^d}\delta_{p_y}(t,x-y;z)\frac{c_{d,\alpha,\gamma}}{\|z\|^{d+\alpha+\gamma}}\dif z\dif y\no\\ &=\int_{\mR^d}\!\!\int_{\mR^d}\big[\delta_{p_y}(t,x-y;z)-\delta_{p_\xi}(t,x-y;z)\big]\Big\|_{\xi=x}\frac{c_{d,\alpha,\gamma}}{\|z\|^{d+\alpha+\gamma}}\dif z\dif y.\label{ii}\end{aligned}$$ By the proof of [@Ch-Zh Theorem 2.5], we know that for any $0<\gamma'<\alpha$, there exists a $C_{\gamma'}$ such that $$\begin{aligned} \big[\delta_{p_y}(t,x-y;z)-\delta_{p_\xi}(t,x-y;z)\big]\Big\|_{\xi=x}&\!\leq C_{\gamma'}\big(\|x-y\|^\beta\wedge 1\big)\Big((t^{-\frac{2}{\alpha}}\|z\|^2)\wedge 1\Big)\\ &\times\Big(\big(\varrho^0_\alpha+\varrho^{\gamma'}_{\alpha-\gamma'}\big)(t,x-y\pm z)+\big(\varrho^0_\alpha+\varrho^{\gamma'}_{\alpha-\gamma'}\big)(t,x-y)\Big).\end{aligned}$$ Taking this into (\[ii\]), choosing $\gamma'$ such that $\alpha+\gamma+\gamma'<2$ and arguing entirely the same as in the proof of Lemma \[fe\], we find that $$\begin{aligned} \sE_1\leq C_{d,\alpha,\gamma}\!\int_{\mR^d}\varrho^{\beta}_{-\gamma}(t,x-y)\dif y\leq C_{d,\alpha,\gamma}t^{\frac{\beta-\gamma}{\alpha}-1}.\end{aligned}$$ Hence, (\[00\]) is true. We proceed to prove (\[000\]). Using (\[p00\]) again, we write $$\begin{aligned} \nabla p_y(t,x-y)-\nabla p_{y}(t,x'-y)=\int_{\mR^d}\sK_{\nabla p_{\alpha}}(t;x,x';y,z)\hat p_y(t,z)\dif z,\end{aligned}$$ where $$\sK_{\nabla p_{\alpha}}(t;x,x';y,z):=\nabla p_{\alpha}\big(\tfrac{\tilde k_0\kappa_0}{2}t,x-y-z\big)-\nabla p_{\alpha}\big(\tfrac{\tilde k_0\kappa_0}{2}t,x'-y-z\big).$$ Let $\tilde x$ be the one of the two points $x$ and $x'$ which is nearer to $y+z$. Then, we know from the proof of (\[p0h\]) that for any $\vartheta\in(0,1)$, $$\|\sK_{\nabla p_{\alpha}}(t;x,x';y,z)\|\leq C_\vartheta\|x-x'\|^\vartheta\varrho_{\alpha-1-\vartheta}^0\big(t,\tilde x-y-z\big).$$ We may argue as in (\[ii\]) to deduce that $$\begin{aligned} \sE_2&:=\int_{\mR^d}\Big[\nabla p_y(t,x-y)-\nabla p_y(t,x'-y)\Big]\dif y\\ &=\int_{\mR^d}\!\int_{\mR^d}\sK_{\nabla p_{\alpha}}(t;x,x';y,z)\big[\hat p_y(t,z)-\hat p_{\xi}(t,z)\big]\Big\|_{\xi=\tilde x}\dif z\dif y.\end{aligned}$$ Thanks to [@Ch-Zh Theorem 2.5], we know that for any $0<\gamma'<\alpha$, there exists a $C_{\gamma'}$ such that $$\begin{aligned} \big[\hat p_y(t,z)-\hat p_{\xi}(t,z)\big]\Big\|_{\xi_=\tilde x}\leq C_{\gamma'}\big(\|\tilde x-y\|^{\beta}\wedge1\big)\Big(\varrho^{0}_{\alpha}+\varrho^{\gamma'}_{\alpha-\gamma'}\Big)(t,z),\end{aligned}$$ which yields by (\[3p1\]) that $$\begin{aligned} \sE_2&\leq C_\vartheta\|x-x'\|^\vartheta\int_{\mR^d}\!\int_{\mR^d}\varrho^0_{\alpha-1-\vartheta}(t,\tilde x-y-z)\Big(\varrho^{0}_{\alpha}+\varrho^{\gamma'}_{\alpha-\gamma'}\Big)(t,z)\dif z\cdot\big(\|\tilde x-y\|^{\beta}\wedge1\big)\dif y\\ &\leq C_\vartheta\|x-x'\|^\vartheta\int_{\mR^d}\Big(\varrho^{\beta}_{\alpha-1-\vartheta}+\varrho^{\gamma'+\beta}_{\alpha-1-\vartheta-\gamma'}\Big)\big(t,\tilde x-y\big)\dif y\leq C_\vartheta\|x-x'\|^\vartheta t^{\frac{\beta-\vartheta-1}{\alpha}}.\end{aligned}$$ The proof is finished. Now, the Levi’s parametrix method suggests that the fundamental solution $p(t,x,y)$ of $\sL^{\kappa,x}_{\alpha}$ should be of the form $$\begin{aligned} p(t,x,y)=p_0(t,x,y)+\int_0^t\!\!\!\int_{\mR^d}p_0(t-s,x,z)q(s,z,y)\dif z\dif s, \label{heat}\end{aligned}$$ where $p_0(t,x,y):=p_y(t,x-y)$ and $q(t,x,y)$ satisfies the integral equation $$q(t,x,y)=q_0(t,x,y)+\int_0^t\!\!\!\int_{\mR^d}q_0(t-s,x,z)q(s,z,y)\dif z\dif s$$ with $$q_0(t,x,y):=\big(\sL^{\kappa,x}_{\alpha}-\sL_{\alpha}^{\kappa,y}\big)p_0(t,x,y).$$ The following lemma collects some estimates that we shall use below, whose proof can be found in [@Ch-Zh]. The following statements hold: 1. ([@Ch-Zh Theorem 3.1]) There exist constants $C_1, C_2$ such that for all $t\geq 0$ and $x,y\in\mR^d$, $$\begin{aligned} \|q(t,x,y)\|\leq C_1\big(\varrho_0^{\beta}+\varrho_\beta^0\big)(t,x-y), \label{q1}\end{aligned}$$ and for any $\gamma<\beta$, $t\geq 0$ and every $x,x',y\in\mR^d$, $$\begin{aligned} \|q(t,x,y)-q(t,x',y)\|\leq C_2\Big(\|x-x'\|^{\beta-\gamma}\wedge 1\Big)\Big(\big(\varrho^0_{\gamma}&+\varrho^{\beta}_{\gamma-\beta}\big)(t,x-y)\no\\ &\quad+\big(\varrho^0_{\gamma}+\varrho^{\beta}_{\gamma-\beta}\big)(t,x'-y)\Big), \label{q2}\end{aligned}$$ where $\beta$ is the constant in (\[ho\]). 2. ([@Ch-Zh Theorem 1.1]) It hold for all $t>0$ and $x\in\mR^d$, $$\begin{aligned} \int_{\mR^d}p(t,x,y)\dif y=1, \label{11}\end{aligned}$$ and there exists a constant $C_3$ such that $$\begin{aligned} \|\nabla p(t,x,y)\|\leq C_3\varrho^0_{\alpha-1}(t,x-y). \label{na}\end{aligned}$$ In [@Ch-Zh], it was also shown that for a constant $C>0$, $$\|\Delta^{\frac{\alpha}{2}}p(t,x,y)\|\leq C\varrho^0_{0}(t,x-y),$$ in which the main point is to handle the singularity caused by the integral with respect to $s$. To study the strong solution of equation , we need to prove more delicate estimates and (\[ph\]) as below, the proof is quite involved. \[seen\] Suppose (\[s1\]) holds true. Then, there exist constants $C_{d,\alpha,\gamma}, C_\vartheta>0$ such that for any $0\leq \gamma<\beta$, $$\begin{aligned} \|\Delta^{\frac{\alpha+\gamma}{2}}p(t,x,y)\|\leq C_{d,\alpha,\gamma}\varrho^0_{-\gamma}(t,x-y). \label{es}\end{aligned}$$ and for any $\vartheta\in(0,\alpha+\beta-1)$, $t>0$ and all $x,x',y\in\mR^d$, $$\begin{aligned} \|\nabla p(t,x,y)-\nabla p(t,x',y)\|\leq C_\vartheta\|x-x'\|^\vartheta\varrho_{\alpha-1-\vartheta}^0(t,\tilde x-y), \label{ph}\end{aligned}$$ where $\tilde x$ is the one of the two points $x$ and $x'$ which is nearer to $y$. Still, we only consider the case when $t\leq 1$. For brevity, we set $$\sS(t,x,y):=\int_0^t\!\!\!\int_{\mR^d}p_0(t-s,x,z)q(s,z,y)\dif z\dif s.$$ By Fubini’s theorem, we can write $$\begin{aligned} \big\|\Delta^{\frac{\alpha+\gamma}{2}}\sS(t,x,y)\big\|&\leq \Bigg\|\!\int_{\frac{t}{2}}^t\!\!\!\int_{\mR^d}\Delta^{\frac{\alpha+\gamma}{2}}p_0(t-s,x,z)\Big(q(s,z,y)-q(s,x,y)\Big)\dif z\dif s\Bigg\|\\ &\quad+\Bigg\|\!\int_{\frac{t}{2}}^t\!\!\!\int_{\mR^d}\Delta^{\frac{\alpha+\gamma}{2}}p_0(t-s,x,z)\dif zq(s,x,y)\dif s\Bigg\|\\ &\quad+\Bigg\|\!\int^{\frac{t}{2}}_0\!\!\!\int_{\mR^d}\Delta^{\frac{\alpha+\gamma}{2}}p_0(t-s,x,z)q(s,z,y)\dif z\dif s\Bigg\|\\ &=:\sC_1(t,x,y)+\sC_2(t,x,y)+\sC_3(t,x,y).\end{aligned}$$ For $\gamma<\beta$, we choose a $\gamma'>0$ such that $\gamma+\gamma'<\beta$, and by (\[e2\]), (\[q2\]) and (\[3p\]), we have $$\begin{aligned} \sC_1(t,x,y)&\leq C_1\!\!\int_{\frac{t}{2}}^t\!\!\!\int_{\mR^d}\varrho^{\beta-\gamma'}_{-\gamma}(t-s,x-z)\Big(\varrho^0_{\gamma'}+\varrho^{\beta}_{\gamma'-\beta}\Big)(s,z-y)\dif z\dif s\\ &\quad+C_1\!\!\int_{\frac{t}{2}}^t\!\!\!\int_{\mR^d}\varrho^{\beta-\gamma'}_{-\gamma}(t-s,x-z)\dif z\Big(\varrho^0_{\gamma'}+\varrho^{\beta}_{\gamma'-\beta}\Big)(s,x-y)\dif s\\ &\leq C_1\!\!\int_{0}^t\!\!\!\int_{\mR^d}\varrho^{\beta-\gamma'}_{-\gamma}(t-s,x-z)\Big(\varrho^0_{\gamma'}+\varrho^{\beta}_{\gamma'-\beta}\Big)(s,z-y)\dif z\dif s\\ &\quad+C_2\!\!\int_{\frac{t}{2}}^t(t-s)^{\frac{\beta-\gamma-\gamma'}{\alpha}-1}\Big(\varrho^0_{\gamma'}+\varrho^{\beta}_{\gamma'-\beta}\Big)(s,x-y)\dif s\\ &\leq C_3\Big(\varrho^{0}_{\beta-\gamma}+\varrho^{\beta-\gamma'}_{\gamma'-\gamma}+\varrho^{\beta}_{-\gamma}\Big)(t,x-y)\\ &\quad+C_3\!\!\int_{\frac{t}{2}}^t(t-s)^{\frac{\beta-\gamma-\gamma'}{\alpha}-1}\Big(\varrho^0_{\gamma'}+\varrho^{\beta}_{\gamma'-\beta}\Big)(s,x-y)\dif s\leq C_4\varrho^0_{-\gamma}(t,x-y).\end{aligned}$$ Thanks to (\[00\]) and taken into account of (\[q1\]), it holds $$\begin{aligned} \sC_2(t,x,y)\leq C_1\!\!\int^t_{\frac{t}{2}}(t-s)^{\frac{\beta-\gamma}{\alpha}-1}\Big(\varrho_0^{\beta}+\varrho_\beta^0\Big)(s,x-y)\dif s\leq C_2\varrho^0_{-\gamma}(t,x-y).\end{aligned}$$ Finally, we have by (\[e2\]), (\[q1\]) and (\[3p\]) that for any $\gamma{'\!'}>0$, $$\begin{aligned} \sC_3(t,x,y)&\leq C_1\!\!\int^{\frac{t}{2}}_0\!\!\!\int_{\mR^d}\varrho^0_{-\gamma}(t-s,x-z)\Big(\varrho_0^{\beta}+\varrho_\beta^0\Big)(s,z-y)\dif z\dif s\\ &\leq C_2t^{-\frac{\gamma+\gamma{'\!'}}{\alpha}}\!\int^{t}_0\!\!\!\int_{\mR^d}\varrho^0_{\gamma{'\!'}}(t-s,x-z)\Big(\varrho_0^{\beta}+\varrho_\beta^0\Big)(s,z-y)\dif z\dif s\leq C_3\varrho^0_{-\gamma}(t,x-y).\end{aligned}$$ Based on the above estimates, we thus get (\[es\]) by (\[e2\]) and (\[heat\]). Next, we proceed to prove (\[ph\]). As in the proof of Lemma \[as\], we set $$\sK_{\nabla p_0}(t;x,x';y):=\nabla p_0(t,x,y)-\nabla p_0(t,x',y).$$ Then, estimate (\[p0h\]) yields that for any $\vartheta\in(0,1)$, $$\|\sK_{\nabla p_0}(t;x,x';y)\|\leq C_\vartheta\|x-x'\|^\vartheta\varrho_{\alpha-1-\vartheta}^0(t,\tilde x-y).$$ As above, we can write $$\begin{aligned} \big\|\nabla\sS(t,x,y)-\nabla\sS(t,x',y)\big\|&\leq \Bigg\|\int_{\frac{t}{2}}^t\!\!\!\int_{\mR^d}\sK_{\nabla p_0}(t-s;x,x';z)\Big(q(s,z,y)-q(s,\tilde x,y)\Big)\dif z\dif s\Bigg\|\\ &\quad+\Bigg\|\int_{\frac{t}{2}}^t\!\!\!\int_{\mR^d}\sK_{\nabla p_0}(t-s;x,x';z)\dif zq(s,\tilde x,y)\dif s\Bigg\|\\ &\quad+\Bigg\|\!\int^{\frac{t}{2}}_0\!\!\!\int_{\mR^d}\sK_{\nabla p_0}(t-s;x,x';z)q(s,z,y)\dif z\dif s\Bigg\|\\ &=:\sD_1(t,x,x',y)+\sD_2(t,x,x',y)+\sD_3(t,x,x',y).\end{aligned}$$ For $\vartheta<\alpha+\beta-1$, choose a $\gamma'$ such that $\vartheta+\gamma'<\alpha+\beta-1$, we follow the same procedure as in the estimate of $\sC_1(t,x,y)$ to derive that $$\begin{aligned} \sD_1(t,x,x',y)&\leq C_\vartheta\|x-x'\|^\vartheta\!\!\int_{\frac{t}{2}}^t\!\!\!\int_{\mR^d}\varrho^{\beta-\gamma'}_{\alpha-1-\vartheta}(t-s,\tilde x-z)\Big(\varrho^0_{\gamma'}+\varrho^{\beta}_{\gamma'-\beta}\Big)(s,z-y)\dif z\dif s\\ &\quad+C_\vartheta\|x-x'\|^\vartheta\!\!\int_{\frac{t}{2}}^t\!\!\!\int_{\mR^d}\varrho^{\beta-\gamma'}_{\alpha-1-\vartheta}(t-s,\tilde x-z)\dif z\Big(\varrho^0_{\gamma'}+\varrho^{\beta}_{\gamma'-\beta}\Big)(s,x-y)\dif s\\ &\leq C_\vartheta\|x-x'\|^\vartheta\varrho_{\alpha-1-\vartheta}^0(t,\tilde x-y).\end{aligned}$$ As for $\sD_2(t,x,x',y)$ and $\sD_3(t,x,x',y)$, we may use (\[000\]) and (\[p0h\]) respectively, and argue the same way as estimating $\sC_2(t,x,y)$ and $\sC_3(t,x,y)$ to get that $$\sD_2(t,x,x',y)+\sD_3(t,x,x',y)\leq C_\vartheta\|x-x'\|^\vartheta\varrho_{\alpha-1-\vartheta}^0(t,\tilde x-y),$$ which in turn yields (\[ph\]). Smoothing properties of the semigroup ===================================== Let $\cT_t$ be the semigroup corresponding to $\sL^\kappa_{\alpha}$, that is, $$\cT_tf(x):=\int_{\mR^d}p(t,x,y)f(y)\dif y,\quad\forall f\in\cB_b(\mR^d).$$ We shall use heat kernel estimates obtained in the last section to derive some space regularities of $\cT_t$, which has its own independent interest and will be used to study Krylov-type estimate and Zvonkin’s transformation in the next section. Let us first introduce some notations. Let $p\geq1$ and $\\|\cdot\\|_p$ denote the norm in $L^p(\mR^d)$. For $0<\gamma<2$, define the Bessel potential space $\mH^{\gamma}_p:=\mH^{\gamma}_p(\mR^d)$ by $$\mH^{\gamma}_p(\mR^d):=\big\{f\in L^p(\mR^d): \Delta^{\frac{\gamma}{2}}f\in L^p(\mR^d)\big\}$$ with norm $$\\|f\\|_{\gamma,p}:=\\|f\\|_p+\\|\Delta^{\frac{\gamma}{2}}f\\|_p.$$ In fact, this space can also be defined to be the complete space of $C^{\infty}_0(\mR^d)$ under the norm $$\\|\sF^{-1}\big((1+\|\cdot\|^{\gamma})(\sF f)\big)\\|_p<\infty,\quad \forall f\in C^{\infty}_0(\mR^d),$$ where $\sF$ (resp. $\sF^{-1}$) denotes the Fourier transform (resp. the Fourier inverse transform). By Sobolev’s embedding theorem, if $\gamma-\frac{d}{p}>0$ is not an integer, then ([@Tri p. 206, (16)]) $$\begin{aligned} \mH^{\gamma}_p\hookrightarrow C^{\gamma-\frac{d}{p}}_b(\mR^d),\label{emb}\end{aligned}$$ where for some $\gamma>0$, $C^{\gamma}_b(\mR^d)$ is the usual Hölder space with norm $$\\|f\\|_{C^{\gamma}_b}:=\sum_{i=1}^{[\gamma]}\\|\nabla^i f(x)\\|_\infty+\big[\nabla^{[\gamma]}f\big]_{\gamma-[\gamma]},$$ here, $[\gamma]$ denotes the integer part of $\gamma$, and for a function $f$ on $\mR^d$ and $\vartheta\in(0,1)$, $$\begin{aligned} [f]_{\vartheta}:=\sup_{x\neq y}\frac{\|f(x)-f(y)\|}{\|x-y\|^{\vartheta}}. \label{hol}\end{aligned}$$ Noticing that for $n\in \mN$, $\mH^{n}_p$ is just the usual Sobolev space with equivalent norm ([@St p. 135, Theorem 3]) $$\\|f\\|_{n,p}=\\|f\\|_p+\\|\nabla^nf\\|_{p},$$ here and below, $\nabla$ denotes the weak derivative of $f$. While for $0<\gamma\neq$ integer, the fractional Sobolev space $\mW^{\gamma}_p$ is defined by $$\begin{aligned} \\|f\\|_{\mW^{\gamma}_p}:=\\|f\\|_p+\sum_{k=0}^{[\gamma]}\Bigg(\int_{\mR^d}\!\!\int_{\mR^d}\frac{\|\nabla^kf(x)-\nabla^kf(y)\|^p}{\|x-y\|^{d+(\gamma-[\gamma])p}}\dif x\dif y\Bigg)^{1/p}<\infty.\label{fs}\end{aligned}$$ The relation between $\mH^{\gamma}_p$ and $\mW^{\gamma}_p$ is that (cf. [@Tri p. 190]): for $\gamma>0$, $\eps\in(0,\gamma)$ and $p>1$, $$\begin{aligned} \mH^{\gamma+\eps}_p\hookrightarrow\mW^{\gamma}_p\hookrightarrow\mH^{\gamma-\eps}_p.\label{re}\end{aligned}$$ Moreover, the following relationship can be found in [@Tri p. 185]: for $p>1$, $\gamma_1\neq\gamma_2$ and $\vartheta\in(0,1)$, $$\begin{aligned} [\mH^{\gamma_1}_p,\mH^{\gamma_2}_p]_\vartheta=\mH^{\gamma_1+\vartheta(\gamma_2-\gamma_1)}_p, \label{fw}\end{aligned}$$ where $[A, B]_{\vartheta}$ denotes the complex interpolation space between two Banach space $A$ and $B$. Recall the following complex interpolation result ([@Tri p. 59, Theorem (a)]). \[inter\] Let $A_i\subseteq B_i$, $i=0,1$ be Banach spaces and $\sT: A_i\rightarrow B_i$, $i=0,1$ be a bounded linear operator. For any $\theta\in(0,1)$, we have $$\\|\sT\\|_{A_{\theta}\rightarrow B_{\theta}}\leq \\|\sT\\|_{A_0\rightarrow B_0}^{1-\theta}\\|\sT\\|_{A_1\rightarrow B_1}^{\theta},$$ where $A_{\theta}:=[A_0, A_1]_{\theta}$, $B_{\theta}:=[B_0, B_1]_{\theta}$, and $\\|\sT\\|_{A_{\theta}\rightarrow B_{\theta}}$ denotes the operator norm of $\sT$ mapping $A_{\theta}$ to $B_{\theta}$. Given a locally integrable function $f$ on $\mR^d$, the Hardy-Littlewood maximal function of $f$ is defined by $$\cM f(x):=\sup_{0<r<\infty}\frac{1}{\|B_r\|}\int_{B_r}\|f(x+y)\|\dif y,$$ where $\|B_r\|$ denotes the Lebesgue measure of ball $B_r$. The following well known result can be found in [@St p. 5, Theorem 1] and [@Zh3]. \(i) For $p\in(1,\infty]$ and all $f\in L^p(\mR^d)$, there exists a constant $C_{d,p}>0$ such that $$\begin{aligned} \\|\cM f\\|_p\leq C_{d,p}\\|f\\|_p. \label{mf}\end{aligned}$$ (ii) For every $f\in \mH^1_p$, there is a constant $C_d>0$ such that for a.e. $x,y\in\mR^d$, $$\begin{aligned} \|f(x)-f(y)\|\leq C_{d}\|x-y\|\Big(\cM\|\nabla f\|(x)+\cM\|\nabla f\|(y)\Big). \label{w11}\end{aligned}$$ Now, we proceed to study the regularities of the semigroup $\cT_t$. There exist constants $C_1, C_2$ such that for any $\vartheta\in(0,1)$, $$\begin{aligned} \\|\nabla \cT_tf\\|_{\infty}\leq C_{1}t^{\frac{\vartheta-1}{\alpha}}[f]_\vartheta ,\quad \forall \ f \in C_0^\infty(\mR^d), \label{33}\end{aligned}$$ and for $\vartheta\in(0,1)$, $\vartheta'\in(0,\alpha+\beta-1)$, $$\begin{aligned} [\nabla \cT_tf]_{\vartheta'}\leq C_{2}t^{\frac{\vartheta-\vartheta'-1}{\alpha}}[f]_\vartheta,\quad \forall \ f \in C_0^\infty(\mR^d),\label{34}\end{aligned}$$ where $[\cdot]$ is defined by (\[hol\]). Using (\[11\]), we find that $$\int_{\mR^d}\nabla p(t,x,y)\dif y=0.$$ Thus, we can write $$\begin{aligned} \nabla \cT_tf(x)=\int_{\mR^d}\nabla p(t,x,y)\big[f(y)-f(x)\big]\dif y. \label{cha}\end{aligned}$$ Thus, by (\[na\]), it is easy to find that for any $\vartheta\in(0,1)$, $$\\|\nabla \cT_tf\\|_{\infty}\leq C_1t^{-\frac{1}{\alpha}} [f]_\vartheta\!\!\int_{\mR^d}\frac{\|x-y\|^\vartheta}{(\|x-y\|+t^{1/\alpha})^{d+\alpha}}\dif y\leq C_1t^{\frac{\vartheta-1}{\alpha}}[f]_\vartheta.$$ To prove (\[34\]), we write $$\begin{aligned} \nabla \cT_tf(x)-\nabla \cT_tf(x')&=\int_{\mR^d}\Big(\nabla p(t,x,y)-\nabla p(t,x',y)\Big)\big(f(y)-f(\tilde x)\big)\dif y,\end{aligned}$$ where $\tilde x$ is the one of the two points $x$ and $x'$ which is nearer to $y$. Taking into account of (\[ph\]) we arrive that for $0<\vartheta'<\alpha+\beta-1$, $$\begin{aligned} \nabla \cT_tf(x)-\nabla \cT_tf(x')&\leq C_2\|x-x'\|^{\vartheta'}[f]_\vartheta\!\int_{\mR^d}\varrho_{\alpha-1-\vartheta'}^\vartheta(t,\tilde x-y)\dif y\no\\ &\leq C_2\|x-x'\|^{\vartheta'}t^{\frac{\vartheta-\vartheta'-1}{\alpha}}[f]_\vartheta,\end{aligned}$$ which in turn implies the desired result. Let $\theta\in(0,1)$ and $\gamma+\theta<\alpha+\beta$ hold, then for every $p>1$, $$\begin{aligned} \\|\cT_tf\\|_{\gamma+\theta,p}\leq C_{\gamma,p}t^{-\gamma/\alpha}\\|f\\|_{\theta,p}, \ \ \ \ \forall \ f \in \mH^{\theta}_p, \label{tt}\end{aligned}$$ where $C_{\gamma,p}>0$ is a constant. Thanks to a standard approximation argument, we only need to prove the estimate for $f\in C^{\infty}_0(\mR^d)$. Observe $$\begin{aligned} \\|T_tf\\|^p_{p}&=\int_{\mR^d}\Bigg(\int_{\mR^d}p(t,x,y)f(y)\dif y\Bigg)^p\dif x\\ &=\int_{\mR^d}\Bigg(\int_{\mR^d}p^{1/q}(t,x,y)p^{1/p}(t,x,y)f(y)\dif y\Bigg)^p\dif x\\ &\leq \int_{\mR^d}\Bigg(\int_{\mR^d}p(t,x,y)\dif y\Bigg)^{p/q}\Bigg(\int_{\mR^d}p(t,x,y)f^p(y)\dif y\Bigg)\dif x,\end{aligned}$$ thus $$\\|\cT_tf\\|_{p}\leq \\|f\\|_p.$$ Hence, $\cT_t$ is a contraction semigroup and we may assume $t<1$ below. By Fubini’s theorem and , for $\beta_1<\alpha+\beta$ we have $$\begin{aligned} \Delta^{\frac{\beta_{1}}{2}} \cT_tf(x)&=\int_{\mR^d}\Delta^{\frac{\beta_{1}}{2}}p(t,x,y)f(y)\dif y\\ &\leq C_1t^{-\beta_1/\alpha}\int_{\mR^d}\varrho_{\alpha}^0(t,x-y)f(y)\dif y,\end{aligned}$$ which implies that for $\beta_1<\alpha+\beta$, $$\\|\cT_tf\\|_{\beta_1,p}\leq C_1t^{-\beta_1/\alpha}\\|f\\|_p.$$ Let $\beta_2$ be such that $1+\beta_2<\alpha+\beta$, as in (\[cha\]), we write $$\begin{aligned} \Delta^{\frac{1+\beta_2}{2}} \cT_tf(x)&=\int_{\mR^d}\Delta^{\frac{1+\beta_2}{2}}p(t,x,y)\big[f(y)-f(x)\big]\dif y.\end{aligned}$$ Then, it follows by (\[es\]) and (\[w11\]) that $$\begin{aligned} \|\Delta^{\frac{1+\beta_2}{2}} \cT_tf(x)\|&\leq \int_{\mR^d}\|\Delta^{\frac{1+\beta_2}{2}}p(t,x,y)\|\cdot\|x-y\|\Big(\cM\|\nabla f\|(x)+\cM\|\nabla f\|(y)\Big)\dif y\\ &\leq C\!\int_{\mR^d}\varrho_{\alpha-1-\beta_2}^0(t,x-y)\cdot\|x-y\|\Big(\cM\|\nabla f\|(x)+\cM\|\nabla f\|(y)\Big)\dif y\\ &=C\cM\|\nabla f\|(x)\!\int_{\mR^d}\varrho_{\alpha-1-\beta_2}^0(t,y)\|y\|\dif y+C\!\int_{\mR^d}\varrho_{\alpha-1-\beta_2}^0(t,y)\|y\|\cdot\cM\|\nabla f\|(x-y)\dif y.\end{aligned}$$ Since $\alpha>1$, one can check easily that $$\begin{aligned} \int_{\mR^d}\varrho_{\alpha-1-\beta_2}^0(t,y)\|y\|\dif y= \left(\int_{\|y\| \le t^{1/\alpha}}+\int_{\|y\|>t^{1/\alpha}}\right)\varrho_{\alpha-1-\beta_2}^0(t,y)\|y\|\dif y\leq t^{-\frac{\beta_2}{\alpha}},\end{aligned}$$ which, together with (\[mf\]) and Minkovski’s inequality, yields $$\begin{aligned} \\|\Delta^{\frac{1+\beta_2}{2}} \cT_tf\\|_p&\leq Ct^{-\frac{\beta_2}{\alpha}}\\|\cM\|\nabla f\|\\|_p+C\bigg\\|\int_{\mR^d}\varrho_{\alpha-1-\beta_2}^0(t,y)\|y\|\cM\|\nabla f\|(x-y)\dif y\bigg\\|_p\\ &\leq Ct^{-\frac{\beta_2}{\alpha}}\\|\nabla f\\|_p+C\!\!\int_{\mR^d}\varrho_{\alpha-1-\beta_2}^0(t,y)\|y\|\cdot\\|\cM\|\nabla f\|(x-y)\\|_p\dif y\\ &\leq Ct^{-\frac{\beta_2}{\alpha}}\\|\nabla f\\|_p.\end{aligned}$$ Hence, we get $$\\|\cT_tf\\|_{1+\beta_2,p}\leq Ct^{-\frac{\beta_2}{\alpha}}\\|f\\|_{1,p}.$$ By the interpolation (\[fw\]), for $\theta\in(0,1)$ $$\mH^{\theta}_p=[L^p,\mH^{1}_p]_{\theta},\quad\mH^{\beta_1+(1+\beta_2-\beta_1)\theta}_p=[\mH^{\beta_1}_p,\mH^{1+\beta_2}_p]_{\theta},$$ and Lemma \[inter\], we can derive that for $\gamma=(1-\theta)\beta_1+\theta\beta_2<\alpha+\beta-\theta$, $$\\|\cT_tf\\|_{\gamma+\theta,p}\leq C t^{-\frac{\gamma}{\alpha}}\\|f\\|_{\theta,p}.$$ The proof is finished. \[ree\] The proof in [@Pri] and [@Zh00] relies heavily on the symmetry of $\Delta^{\frac{\alpha}{2}}$ and the smooth properties at least up to second order of its heat kernel $p_{\alpha}(t,x)$. However, the operator $\sL_{\alpha}^\kappa$ considered here is non-symmetric and its heat kernel has no more regularity than ‘$\alpha+\beta$’-order, as we have seen in Lemma \[seen\]. Krylov-type estimate and Zvonkin’s transformation ================================================= This section consists of two parts, one is to obtain a Krylov-type estimate for the strong solution of SDE , while the other is to transforms SDE (\[sde2\]) into a new one with better drift coefficient by Zvonkin’s transformation. The regularity of the semigroup $\cT_t$ obtained in the last section will play an important role in the two subsections below. Krylov’s estimate ----------------- Let $X_t(x)$ be a strong solution to SDE (\[sde2\]). Usually, the Itô’s formula is performed for functions $f\in C^2_b(\mR^d)$. However, this is too strong for our latter use. Notice that by (\[s1\]), $\sL_{\alpha}^\kappa f$ is meaningful for any $f\in C^{\gamma}_b(\mR^d)$ as long as $\gamma>\alpha$. Indeed, we have by (\[nu\]) that $$\begin{aligned} \sL^\kappa_{\alpha}f(x)&\leq C_{d,\alpha}\!\int_{\|z\|\leq 1}\!\!\int_0^1\!\big\|\nabla f(x+rz)-\nabla f(x)\big\|\dif r\frac{\dif z}{\|z\|^{d+\alpha-1}}+C_{d,\alpha}\\|f\\|_{\infty}\\ &\leq C_{d,\alpha}\!\int_{\|z\|\leq 1}\frac{\dif z}{\|z\|^{d+\alpha-\gamma}}\\|f\\|_{\gamma}+C_{d,\alpha}\\|f\\|_{\infty}<\infty.\end{aligned}$$ We first show that Itô’s formula holds for $f(X_t)$ when $f\in C^{\gamma}_b(\mR^d)$ with $\gamma>\alpha$. \[ito\] Suppose (\[nu\]) and (\[s1\]) hold. Let $X_t$ satisfy (\[sde2\]) and $f\in C^{\gamma}_b(\mR^d)$ with $\gamma>\alpha$. Then, we have $$\begin{aligned} f(X_t)-f(x)-\int_0^t\!\sL f(X_s)\dif s=\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\mR^d}\big[f\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(r)z\big)-f(X_s)\big]\tilde \cN(\dif z\times\dif r\times \dif s),\end{aligned}$$ where $\sL=\sL_\alpha^\kappa+b\cdot\nabla$. Let $\rho\in C^{\infty}_0(\mR^d)$ such that $\int_{\mR^d}\rho(x)\dif x=1$. Define $\rho_n(x):=n^d\rho(nx)$, and $$\begin{aligned} f_n(x):=\int_{\mR^d}f(y)\rho_n(x-y)\dif y.\label{mo}\end{aligned}$$ Hence, we have $f_n\in C^2_b(\mR^d)$ with $\\|f_n\\|_{C^\gamma_b}\leq \\|f\\|_{C^\gamma_b}$, and $\\|f_n-f\\|_{C^{\gamma'}_b}\rightarrow0$ for every $\gamma'<\gamma$. By using Itô’s formula for $f_n(X_t)$, we get $$\begin{aligned} f_n(X_t)-f_n(x)-\int_0^t\!\sL f_n(X_s)\dif s=\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\mR^d}\big[f_n\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(r)z\big)-f_n(X_{s-})\big]\tilde \cN(\dif z\times\dif r\times \dif s).\end{aligned}$$ Now we are going to pass the limits on the both sides of the above equality. It is easy to see that for every $\omega$ and $x\in\mR^d$, $$f_n(X_t)-f_n(x)\rightarrow f(X_t)-f(x),\quad \text{as}\,\,n\rightarrow\infty.$$ Since $$\begin{aligned} \|f_n(x+z)-f_n(x)-z\cdot\nabla f_n(x)\|\leq C\|z\|^{\gamma}\\|f_n\\|_{C^\gamma_b}\leq C\|z\|^{\gamma}\\|f\\|_{C^\gamma_b},\end{aligned}$$ we can get by dominated convergence theorem that for every $\omega$, $$\int_0^t\!\sL f_n(X_s)\dif s\rightarrow\int_0^t\!\sL f(X_s)\dif s,\quad \text{as}\,\,n\rightarrow\infty.$$ Finally, by the isometry formula, we have $$\begin{aligned} &\mE\bigg\|\!\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\mR^d}\Big[f_n\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(r)z\big)-f_n(X_s)\\ &\qquad-f\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(r)z\big)+f(X_{s-})\Big]\tilde \cN(\dif z\times\dif r\times \dif s)\bigg\|^2\\ &=\mE\int_0^t\!\!\!\int_{\mR^d}\!\!\int_0^{\infty}1_{[0,\sigma(X_s,z)]}(r)\big\|f_n(X_s+z)-f_n(X_s) -f(X_s+z)+f(X_s)\big\|^2\dif r\nu(\dif z)\dif s\\ &\leq C\!\!\int_0^t\!\!\!\int_{\mR^d}\mE\big\|f_n(X_s+z)-f_n(X_s) -f(X_s+z)+f(X_s)\big\|^2\nu(\dif z)\dif s\rightarrow0,\quad \text{as}\,\,n\rightarrow\infty,\end{aligned}$$ where in the last step we have used the fact that $\sigma$ is bounded, $\\|f_n\\|_{C^\gamma_b}\leq \\|f\\|_{C^\gamma_b}$ and the dominated convergence theorem again. The proof is finished. We need the following results about the semi-linear elliptic PDE to prove the Krylov’s estimate. \[p1\] Let $\lambda,\bar k>0$. For any $f\in C_b^\infty(\mR^d)$, there exists a unique classical solution $u\in C_b^{\alpha+\theta}(\mR^d)$ with $\theta<\beta$ to equation $$\begin{aligned} \lambda u-\sL^\kappa_\alpha u-\bar k\|\nabla u\|=f, \label{pide3}\end{aligned}$$ which also satisfies the following integral equation: $$\begin{aligned} u(x)=\int_0^\infty\!\e^{-\lambda t}\cT_t\big(\bar k\|\nabla u\|+f\big)(x)\dif t.\label{u}\end{aligned}$$ Moreover, for $\lambda$ big enough, we have for any $1<\gamma<\alpha$, $$\begin{aligned} \\|u\\|_{\gamma,p}\leq C\\|f\\|_p.\label{ess}\end{aligned}$$ Let us first construct the solution of via Picard’s iteration argument. Set $u_0\equiv0$ and for $n\in \mN$, define $u_n$ recursively by $$u_n(x):=\int_0^\infty\!\e^{-\lambda t}\cT_t\big(\bar k\|\nabla u_{n-1}\|+f\big)(x)\dif t.$$ In view of (\[na\]), it is easy to check that $u_1\in C^1_b(\mR^d)$, and $u_2$ is thus well defined, and so on. We further write that $$\begin{aligned} \nabla u_1(x)-\nabla u_1(y)&=\int_0^\infty\!\e^{-\lambda t}\big[\nabla \cT_tf(x)-\nabla \cT_tf(y)\big]\dif t\\ &=\left(\int_0^{\|x-y\|^{\alpha}}+\int^{\infty}_{\|x-y\|^{\alpha}}\right)\e^{-\lambda t}\big[\nabla \cT_tf(x)-\nabla \cT_tf(y)\big]\dif t=:I_1+I_2.\end{aligned}$$ For $I_1$, we have by (\[33\]) that for $0<\vartheta<2-\alpha$, $$\begin{aligned} I_1\leq C\!\!\int_0^{\|x-y\|^{\alpha}}\!t^{\frac{\vartheta-1}{\alpha}}\dif t \cdot[f]_\vartheta\leq C\|x-y\|^{\alpha+\vartheta-1}[f]_\vartheta.\end{aligned}$$ Meanwhile, as a direct result of (\[34\]), we have for $\vartheta\in(0,\beta)$ and $\alpha+\vartheta-1<\vartheta'<\alpha+\beta-1$, $$\begin{aligned} I_2\leq C\|x-y\|^{\vartheta'} [f]_\vartheta\!\int^{\infty}_{\|x-y\|^{\alpha}}t^{\frac{\vartheta-\vartheta'-1}{\alpha}}\dif t\leq C\|x-y\|^{\alpha+\vartheta-1} [f]_\vartheta.\end{aligned}$$ Consequently, we get that $u_1\in C_b^{\alpha+\vartheta}(\mR^d)$ if $f\in C_b^{\vartheta}(\mR^d)$. Noticing that $u_1\in C_b^{\alpha+\vartheta}(\mR^d)$ implies that $\|\nabla u_1\|\in C_b^{\vartheta}(\mR^d)$ since $$\big\|\|\nabla u_1\|(x)-\|\nabla u_1\|(y)\big\|\leq \|\nabla u_1(x)-\nabla u_1(y)\|.$$ Repeating the above argument, we have for every $n\in\mN$ and $\vartheta\in(0,\beta)$, $$u_n\in C_b^{\alpha+\vartheta}(\mR^d).$$ Moreover, since $$\begin{aligned} u_n(x)-u_m(x)&=\int_0^\infty\!\e^{-\lambda t}\cT_t\big(\bar k\|\nabla u_{n-1}\|-k\|\nabla u_{m-1}\|\big)(x)\dif t\\ &\leq \bar k\!\!\int_0^\infty\!\e^{-\lambda t}\cT_t\big\|\nabla u_{n-1}-\nabla u_{m-1}\big\|(x)\dif t,\end{aligned}$$ we further have that for $\vartheta'\!'$ with $\vartheta<\vartheta'\!'<\alpha+\vartheta-1$, $$\begin{aligned} \\|u_n-u_m\\|_{C_b^{\alpha+\vartheta}}&\leq \bar k\!\!\int_0^\infty\!\e^{-\lambda t}\\|\cT_t\big(\|\nabla u_{n-1}-\nabla u_{m-1}\|\big)\\|_{C_b^{\alpha+\vartheta}}\dif t\\ &\leq C_{\bar k}\!\!\int_0^\infty\!\e^{-\lambda t}\Big(1+t^{\frac{\vartheta-1}{\alpha}}+t^{\frac{\vartheta'\!'-\vartheta}{\alpha}-1}\Big)\dif t\cdot\\|u_{n-1}-u_{m-1}\\|_{C_b^{1+\vartheta'\!'}}\\ &\leq C_{\bar k}\lambda^{-\frac{\vartheta'\!'-\vartheta}{\alpha}}\\|u_{n-1}-u_{m-1}\\|_{C_b^{\alpha+\vartheta}},\end{aligned}$$ where we have also used the fact that $\lambda>1$. This means that $u_n$ is Cauchy sequence in $C_b^{\alpha+\vartheta}(\mR^d)$. Thus, there exists a limit function $u\in C_b^{\alpha+\vartheta}(\mR^d)$ with $\vartheta\in(0,\beta)$ satisfying (\[u\]). Since by [@Ch-Zh (1.7)], for every $f\in C^{\vartheta}_b(\mR^d)$, $$\p t\cT_tf(x)=\sL^\kappa_{\alpha}\cT_tf(x),$$ we have by integral by part, $$\begin{aligned} \sL^\kappa_{\alpha}u(x)&=\int_0^\infty\!\e^{-\lambda t}\sL^\kappa_{\alpha}\cT_t\big(\bar k\|\nabla u\|+f\big)(x)\dif t\\ &=\int_0^\infty\!\e^{-\lambda t}\p t\cT_t\big(\bar k\|\nabla u\|+f\big)(x)\dif t\\ &=-\bar k\|\nabla u\|(x)-f(x)+\lambda u(x),\end{aligned}$$ which means that $u$ satisfies PIDE (\[pide3\]). Moreover, we have by (\[tt\]) that $$\begin{aligned} \\|u\\|_{\gamma,p}&\leq \int_0^\infty\!\e^{-\lambda t}\\|\cT_t\big(\bar k\|\nabla u\|+f\big)\\|_{\gamma,p}\dif t\\ &\leq C_{\bar k}\!\!\int_0^\infty\!\e^{-\lambda t}t^{-\frac{\gamma}{\alpha}}\dif t\Big(\\|\nabla u\\|_{p}+\\|f\\|_{p}\Big)\\ &\leq C_{\bar k}\lambda^{\frac{\gamma}{\alpha}-1}\Big(\\|u\\|_{1,p}+\\|f\\|_{p}\Big).\end{aligned}$$ Choosing $\lambda$ big enough such that $C_{\bar k}\lambda^{\frac{\gamma}{\alpha}-1}<1$, we can get (\[ess\]). The whole proof is finished. Now, we can give the Krylov’s estimate for strong solutions of SDE (\[sde2\]). Let $X_t$ be a strong solution of SDE (\[sde2\]). Then, for any $T>0$, there exist a constant $C_T$ such that for any $f\in L^p(\mR^d)$ with $p>d/\alpha$, we have $$\begin{aligned} \mE\Bigg(\int_0^T\!\!f(X_s)\dif s\Bigg)\leq C_T \\|f\\|_p. \label{kry1}\end{aligned}$$ We first suppose that $f\in C^{\infty}_0(\mR^{d})$. By Lemma \[p1\], there exists a solution $u\in C_b^{\alpha+\vartheta}(\mR^d)$ with $0<\vartheta<\beta$ to equation (\[pide3\]), which is given by (\[u\]). According to Lemma \[ito\], we can use Itô’s formula to get for any $t>0$, $$\begin{aligned} u(X_t)&=u(x)+\int_0^t\!\!\sL^\kappa_{\alpha}u(X_s)+b\cdot\nabla u(X_s)\dif s\\ &\quad+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\mR^d}\big[u\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(r)z\big)-u(X_{s-})\big]\tilde \cN(\dif z\times\dif r\times \dif s).\end{aligned}$$ By (\[pide3\]) and take $\bar k$ big enough such that $\bar k>\\|b\\|_{\infty}$, we have for any $T>0$, $$\begin{aligned} \int_0^T\!\!\sL^\kappa_{\alpha}u(X_s)+b\cdot\nabla u(X_s)\dif s&=\lambda\int_0^T\!u(X_s)\dif s+\int_0^T\!\!\Big(b\cdot\nabla u(X_s)-\bar k\|\nabla u\|(X_s)\Big)\dif s-\int_0^T\!\!f(X_s)\dif s\\ &\leq \lambda\int_0^T\!u(X_s)\dif s-\int_0^T\!\!f(X_s)\dif s.\end{aligned}$$ Consequently, it follows that $$\begin{aligned} \mE\Bigg(\int_0^T\!\!f(X_s)\dif s\Bigg)\leq \lambda\mE\Bigg(\int_0^T\!\|u(X_s)\|\dif s\Bigg)+2\\|u\\|_{\infty}\leq C_{\lambda,T}\\|u\\|_{\infty}.\end{aligned}$$ Since $p>d/\alpha$, there exists a $\gamma<\alpha$ such that $p>d/\gamma$. Now, using (\[ess\]), we conclude that $$\begin{aligned} \mE\Bigg(\int_0^T\!\!f(X_s)\dif s\Bigg)\leq C_{\lambda,T}\\|u\\|_{\infty}\leq C_{\lambda,T}\\|u\\|_{\gamma,p}\leq C_{\lambda,T}\\|f\\|_{p},\end{aligned}$$ where the second inequality is due to the Sobolev embedding (\[emb\]). By a standard density argument as in [@Zh00], we get the desired result for general $f\in L^p(\mR^d)$. Zvonkin’s transformation ------------------------ Now, we follow the idea in [@Kr-Ro; @Zh1; @Zh00] to transform SDE (\[sde2\]) into a new one with better coefficients. Unlike [@Kr-Ro; @Zh1; @Zh00], we do not have well developed elliptic equation theory to solve the following equation in Bessel potential space (or Sobolev space): $$\begin{aligned} \lambda u(x)-\sL^\kappa_{\alpha} u(x)-b\cdot\nabla u(x)=b(x). \label{pide1}\end{aligned}$$ The point is that the operator $\sL^\kappa_{\alpha}$ is non-symmetric and has no comparison with operator $\Delta^{\frac{\alpha}{2}}$. To be more precisely, even if we know $\Delta^{\frac{\alpha}{2}}u\in L^p(\mR^d)$ for some $p>1$, it is not easy to claim that $\sL^\kappa_{\alpha}u\in L^p(\mR^d)$, and vice versa. The authors in [@Kr-Ro; @Zh1] encountered with classical local seconder order differential operator $$\begin{aligned} \sL_2:=\frac{1}{2}\sum_{i,j=1}^da_{ij}(x)\frac{\p^2}{\p{x_i}\p{x_j}},\end{aligned}$$ which has been well studied, it is clear that if $\big(a_{ij}(x)\big)$ is bounded, then $\Delta u\in L^p(\mR^d)$ implies $\sL_2 u\in L^p(\mR^d)$ for every $p>1$. While [@Zh00] only needs to handle the symmetric operator $\Delta^{\frac{\alpha}{2}}$. Therefore, some new ways are required in our case. First, we note that the elliptic equation can still be solved in the framework of Hölder space. The following result can be proved as in Lemma \[p1\], we omit the details here. \[int\] Assume that for some $\vartheta\in(0,\beta)$, $b\in C^\vartheta_b(\mR^d)$ . Then, there exists a classical solution $u\in C_b^{\alpha+\vartheta}(\mR^d)$ to (\[pide1\]). Meanwhile, $u$ also satisfies the integral equation: $$\begin{aligned} u(x)=\int_0^\infty\!\e^{-\lambda t}\cT_t\big(b\cdot\nabla u+b\big)(x)\dif t. \label{g1}\end{aligned}$$ Despite that we can not solve the elliptic equation (\[pide1\]) in Bessel potential space (or Sobolev space), we can solve the integral equation (\[g1\]) in this framework thanks to the Bessel regularity of $\cT_t$ obtained in Section 4. \[equ\] Let $1<\gamma<\alpha$. Suppose that for some $p>\frac{d}{\gamma}$ and $0<\theta\in(1-\gamma+\frac{d}{p},1)$, $$b\in L^{\infty}(\mR^d)\cap\mW^{\theta}_p.$$ Then, for $\lambda$ big enough there exists a function $u\in \mH^{\gamma+\theta}_p$ satisfying the integral equation (\[g1\]). Moreover, we have $$\begin{aligned} \\|u\\|_{\gamma+\theta,p}\leq C_1\\|b\\|_{\mW^\theta_p} \label{esa}\end{aligned}$$ and $$\begin{aligned} \\|u\\|_{\infty}+\\|\nabla u\\|_{\infty}\leq \frac{1}{2}. \label{es2}\end{aligned}$$ We only show the priori estimate (\[esa\]). Then ,the existence of solutions follows by the standard continuity method. Since $1<\gamma<\alpha$, choose $\eps\in(0,\alpha-\gamma)$, we have by (\[tt\]) and (\[re\]) that $$\begin{aligned} \\|u\\|_{\gamma+\theta,p}&\leq \int_0^\infty\!\e^{-\lambda t}\\|\cT_{t}(b\cdot\nabla u+b)\\|_{\gamma+\theta,p}\dif t\\ &\leq C_1\!\int_0^\infty\!\e^{-\lambda t}t^{-\frac{\gamma+\eps}{\alpha}}\dif t\cdot\Big(\\|b\cdot\nabla u\\|_{\theta-\eps,p}+\\|b\\|_{\theta-\eps,p}\Big)\\ &\leq C_1\lambda^{\frac{\gamma+\eps}{\alpha}-1}\cdot\Big(\\|b\cdot\nabla u\\|_{\mW^\theta_p}+\\|b\\|_{\mW^\theta_p}\Big).\end{aligned}$$ In view of (\[fs\]), (\[emb\]), and thanks to the condition that $\gamma>1$ and $\gamma+\theta-1-\tfrac{d}{p}>0$, we can know $$\begin{aligned} \\|b\cdot\nabla u\\|_{\mW^\theta_p}&\leq \\|b\\|_\infty\\|\nabla u\\|_p+\\|b\\|_{\mW^\theta_p}\\|\nabla u\\|_{\infty}+\\|b\\|_{\infty}\\|\nabla u\\|_{\mW^\theta_p}\\ &\leq \\|b\\|_\infty\\|u\\|_{1,p}+\\|b\\|_{\mW^\theta_p}\\|u\\|_{\gamma+\theta,p}+ \\|b\\|_{\infty}\\|u\\|_{\mW^{1+\theta}_p}\\ &\leq \Big(\\|b\\|_{\mW^\theta_p}+\\|b\\|_{\infty}\Big)\\|u\\|_{\gamma+\theta,p},\end{aligned}$$ where we also used (\[re\]) in the last inequality. It then follows that $$\\|u\\|_{\gamma+\theta,p}\leq C_1\lambda^{\frac{\gamma+\eps}{\alpha}-1}\Big(\\|b\\|_{\mW^\theta_p}+\\|b\\|_{\infty}\Big)\\|u\\|_{\gamma+\theta,p}+C_1\lambda^{\frac{\gamma+\eps}{\alpha}-1}\\|b\\|_{\mW^\theta_p}.$$ Hence, we can choose $\lambda_1$ big enough such that $$C_1\lambda^{\frac{\gamma+\eps}{\alpha}-1}_1\Big(\\|b\\|_{\mW^\theta_p}+\\|b\\|_{\infty}\Big)<\frac{1}{2},$$ which means (\[esa\]) is true. Moreover, we can take $\lambda\geq \lambda_1$ such that $$\lambda^{\frac{\gamma+\eps}{\alpha}-1}\\|b\\|_{\mW^\theta_p}<\frac{1}{8},$$ and then get $$\\|u\\|_{\infty}+\\|\nabla u\\|_{\infty}\leq2\\|u\\|_{\gamma+\theta,p}\leq\frac{1}{2}.$$ The proof is finished. In the sequel, we assume that $b\in L^{\infty}(\mR^d)\cap\mW^{\theta}_p$ with $$\begin{aligned} \theta\in(1-\frac{\alpha}{2},1),\quad p>\frac{2d}{\alpha}. \label{index}\end{aligned}$$ Notice that we can always choose a $1<\gamma<\alpha$ such that $$\theta>1-\gamma+\frac{d}{p}\quad\text{and}\quad p>\frac{d}{\gamma}.$$ Hence, according to Theorem \[equ\], for $\lambda$ big enough we can get a function $u\in \mH^{\gamma+\theta}_p$ satisfying the integral equation (\[g1\]). Define $$\Phi(x):=x+u(x).$$ In view of (\[es2\]), we also have $$\frac{1}{2}\|x-y\|\leq\big\|\Phi(x)-\Phi(y)\big\|\leq \frac{3}{2}\|x-y\|,$$ which implies that the map $x\rightarrow\Phi(x)$ forms a $C^1$-diffeomorphism and $$\begin{aligned} \frac{1}{2}\leq \\|\nabla\Phi\\|_{\infty},\\|\nabla\Phi^{-1}\\|_{\infty}\leq 2, \label{upd}\end{aligned}$$ where $\Phi^{-1}(\cdot)$ is the inverse function of $\Phi(\cdot)$. We prove the following Zvonkin’s transformation. \[zvon\] Let $\Phi(x)$ be defined as above and $X_t$ solve SDE (\[sde2\]). Then, $Y_t:=\Phi(X_t)$ satisfies the following SDE: $$\begin{aligned} Y_t&=\Phi(x)+\int_0^t\tilde b(Y_s)\dif s+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}\tilde g(Y_{s-},z)1_{[0,\tilde\sigma(Y_{s-},z)]}(r)\tilde \cN(\dif z\times\dif r\times \dif s)\no\\ &\quad+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|> 1}\tilde g(Y_{s-},z)1_{[0,\tilde\sigma(Y_{s-},z)]}(r)\cN(\dif z\times\dif r\times \dif s), \label{sde3}\end{aligned}$$ where $$\begin{aligned} \tilde b(x)=\lambda u\big(\Phi^{-1}(x)\big)-\int_{\|z\|>1}\!\big[u\big(\Phi^{-1}(x)+z\big)-u\big(\Phi^{-1}(x)\big)\big]\sigma\big(\Phi^{-1}(x),z\big)\nu(\dif z)\end{aligned}$$ and $$\begin{aligned} \tilde g(x,z):=\Phi\big(\Phi^{-1}(x)+z\big)-x,\quad \tilde\sigma(x,z):=\sigma\big(\Phi^{-1}(x),z\big).\end{aligned}$$ Let $b_n$ be the mollifying approximation of $b$ defined as in (\[mo\]). Then, it is obvious that $$\begin{aligned} \\|b_n\\|_{\infty}\leq \\|b\\|_{\infty},\quad \\|b_n-b\\|_{\mW^{\vartheta}_p}\rightarrow0,\,\,\text{as}\,\,n\rightarrow\infty. \label{bn}\end{aligned}$$ Meanwhile, $b_n\in C_b^\vartheta(\mR^d)$ for any $\vartheta\in(0,\beta)$ and we may assume $$\begin{aligned} b_n\rightarrow b,\,\,a.e.,\,\,\text{as}\,\,n\rightarrow\infty. \label{bnn}\end{aligned}$$ Let $u_n\in C_b^{\alpha+\vartheta}(\mR^d)$ be the classical solution to the elliptic equation (\[pide1\]) with $b$ replaced by $b_n$. According to Lemma \[int\], we know that $u_n$ also satisfies the integral equation $$\begin{aligned} u_n(x)=\int_0^\infty\!\e^{-\lambda t}\cT_t\big(b_n\cdot\nabla u_n+b_n\big)(x)\dif t.\end{aligned}$$ We proceed to show that $$\begin{aligned} \\|u_n-u\\|_{\gamma+\vartheta,p}\rightarrow0,\,\,\text{as}\,\,n\rightarrow\infty. \label{unn}\end{aligned}$$ In fact, write $$\begin{aligned} u_n(x)-u(x)=\int_0^\infty\!\e^{-\lambda t}\cT_t\Big(b_n(\nabla u_n-\nabla u)+(b_n-b)\cdot\nabla u+(b_n-b)\Big)(x)\dif t.\end{aligned}$$ As in the proof of Theorem \[equ\], we have $$\begin{aligned} \\|u_n-u\\|_{\gamma+\theta,p}\leq C_1\lambda^{\frac{\gamma+\eps}{\alpha}-1}\cdot\Big(\\|b_n(\nabla u_n-\nabla u)\\|_{\mW^\theta_p}+\\|(b_n-b)\cdot\nabla u\\|_{\mW^\theta_p}+\\|b_n-b\\|_{\mW^\theta_p}\Big).\end{aligned}$$ At the same time, we can also get by (\[bn\]) that $$\begin{aligned} \\|b_n(\nabla u_n-\nabla u)\\|_{\mW^\theta_p}\leq \Big(\\|b_n\\|_{\mW^\theta_p}+\\|b_n\\|_{\infty}\Big)\\|u_n-u\\|_{\gamma+\theta,p}\leq \Big(\\|b\\|_{\mW^\theta_p}+\\|b\\|_{\infty}\Big)\\|u_n-u\\|_{\gamma+\theta,p}.\end{aligned}$$ Hence, $$\begin{aligned} \\|u_n-u\\|_{\gamma+\theta,p}&\leq C_\lambda\Big(\\|(b_n-b)\cdot\nabla u\\|_{\mW^\theta_p}+\\|b_n-b\\|_{\mW^\theta_p}\Big)\\ &\leq C_\lambda\\|b_n-b\\|_{\mW^\theta_p}+C_\lambda\!\!\int_{\mR^d}\!\!\int_{\mR^d}\frac{\|\nabla u(x)-\nabla u(y)\|^p}{\|x-y\|^{d+\vartheta p}}\|b_n(y)-b(y)\|\dif x\dif y\rightarrow0,\,\,\text{as}\,\,n\rightarrow\infty,\end{aligned}$$ where we have used (\[bn\]), (\[bnn\]) and the dominated convergence theorem. Now, we define $$\Phi_n(x):=x+u_n(x).$$ By Lemma \[ito\] and recalling that $u_n$ satisfies (\[pide1\]), we can use Itô’s formula for $u_n$ to get $$\begin{aligned} u_n(X_t)&=u_n(x)+\int_0^t\!\Big(\lambda u_n+\big(b-b_n\big)\cdot\nabla u_n-b_n\Big)(X_s)\dif s\\ &\quad-\int_0^t\!\!\!\int_0^{\infty}\!\!\!\int_{\|z\|>1}\big[u_n\big(X_s+1_{[0,\sigma(X_s,z)]}(v)z\big)-u_n(X_s)\big]\nu(\dif z)\dif r\dif s\\ &\quad+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|> 1}\big[u_n\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(v)z\big)-u_n(X_{s-})\big] \cN(\dif z\times\dif r\times \dif s)\\ &\quad+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}\big[u_n\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(v)z\big)-u_n(X_{s-})\big]\tilde \cN(\dif z\times\dif r\times \dif s).\end{aligned}$$ Adding this with SDE and noticing that $$\Phi_n\big(x+y\big)-\Phi_n(x)=u_n(x+y)-u_n(x)+y,$$ $$\begin{aligned} \Phi_n\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(r)z\big)-\Phi_n(X_{s-})=1_{[0,\sigma(X_{s-},z)]}(r)\left[\Phi_n\big(X_{s-}+z\big)-\Phi_n(X_{s-})\right],\\end{aligned}$$ we obtain $$\begin{aligned} Y^n_t&:=\Phi_n(X_t)=\Phi_n(x)+\int_0^t\!\lambda u_n(X_s)\dif s+\int_0^t\!\Big(\big(b-b_n\big)\cdot\nabla u_n+\big(b-b_n\big)\Big)(X_s)\dif s\\ &\quad-\int_0^t\!\!\!\int_0^{\infty}\!\!\!\int_{\|z\|>1}\big[u_n\big(X_s+1_{[0,\sigma(X_s,z)]}(r)z\big)-u_n(X_s)\big]\nu(\dif z)\dif r\dif s\\ &\quad+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|> 1}\big[\Phi_n\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(r)z\big)-\Phi_n(X_{s-})\big] \cN(\dif z\times\dif r\times \dif s)\\ &\quad+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}\big[\Phi_n\big(X_{s-}+1_{[0,\sigma(X_{s-},z)]}(r)z\big)-\Phi_n(X_{s-})\big]\tilde \cN(\dif z\times\dif r\times \dif s)\\ &=\Phi_n(x)+\int_0^t\!\lambda u_n(X_s)\dif s+\int_0^t\!\Big(\big(b-b_n\big)\cdot\nabla u_n+\big(b-b_n\big)\Big)(X_s)\dif s\\ &\quad-\int_0^t\!\!\!\int_{\|z\|>1}\big[u_n(X_s+z)-u_n(X_s)\big]\sigma(X_s,z)\nu(\dif z)\dif s\\ &\quad+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|> 1}1_{[0,\sigma(X_{s-},z)]}(r)\big[\Phi_n(X_{s-}+z)-\Phi_n(X_{s-})\big] \cN(\dif z\times\dif r\times \dif s)\\ &\quad+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}1_{[0,\sigma(X_{s-},z)]}(r)\big[\Phi_n(X_{s-}+z)-\Phi_n(X_{s-})\big]\tilde \cN(\dif z\times\dif r\times \dif s)\\ &=:\Phi_n(x)+\cI_1+\cI_2+\cI_3+\cI_4.\end{aligned}$$ Now we are going to take limits for the above equality. First of all, it is easy to see that $$\lim_{n\rightarrow\infty}Y^n_t=\Phi(X_t)=Y_t.$$ By the dominated convergence theorem and (\[unn\]), we also have $$\begin{aligned} \cI_2+\cI_3\rightarrow&-\!\int_0^t\!\!\!\int_{\|z\|>1}\big[u(X_s+z)-u(X_s)\big]\sigma(X_s,z)\nu(\dif z)\dif s\\ &+\int_0^t\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|> 1}1_{[0,\sigma(X_{s-},z)]}(r)\big[\Phi(X_{s-}+z)-\Phi(X_{s-})\big] \cN(\dif z\times\dif r\times \dif s).\end{aligned}$$ As for $\cI_4$, it follows from (\[upd\]), (\[unn\]) and the dominated convergence theorem that $$\begin{aligned} &\mE\bigg\|\int_0^t\!\!\!\int_0^{\infty}\!\!\!\int_{\|z\|\leq1}\!\!1_{[0,\sigma(X_{s-},z)]}(r)\Big[\Phi^n(X_{s-}+z)-\Phi^n(X_{s-})\\ &\qquad\qquad\qquad\qquad\qquad-\Phi(X_{s-}+z)+\Phi(X_{s-})\Big]\tilde{\cN}(\dif z\times\dif r\times \dif s)\bigg\|^2\\ &=\mE\int_0^t\!\!\!\int_0^{\infty}\!\!\!\int_{\|z\|\leq1}\!\!1_{[0,\sigma(X_s,z)]}(r)\Big\|\Phi^n_s(X_s+z)-\Phi^n_s(X_s)-\Phi_s(X_s+z)+\Phi_s(X_s)\Big\|^2\nu(\dif z)\dif r\dif s\\ &=\mE\int_0^t\!\!\!\int_{\|z\|\leq1}\Big\|\Phi^n_s(X_s+z)-\Phi^n_s(X_s)-\Phi_s(X_s+z)+\Phi_s(X_s)\Big\|^2\sigma(X_s,z)\nu(\dif z)\dif s\\ &\rightarrow0,\quad\text{as}\,\, n\rightarrow\infty.\end{aligned}$$ Finally, Krylov’s estimate (\[kry1\]) yields that $$\mE\Bigg(\!\int_0^tb(X_s)-b_n(X_s)\dif s\Bigg)\leq C\\|b-b_n\\|_p\rightarrow0,$$ which in turn implies by (\[unn\]) that $$\begin{aligned} \cI_1&\rightarrow \lambda\!\!\int_0^t u(X_s)\dif s,\quad\text{as}\,\, n\rightarrow\infty.\end{aligned}$$ Combing the above calculations, and noticing that $X_s=\Phi^{-1}(Y_s)$, we get the desired result. At the end of this section, we collect some properties of the new coefficients. For a function $f$ on $\mR^d$, set $$\cJ_zf(x):=f(x+z)-f(x).$$ Suppose that: 1. The global condition (\[s1\]) holds true and (\[a1\]) is satisfied for almost all $x,y\in\mR^d$ with $\zeta\in L^q(\mR^d)$, $q>d/\alpha$. 2. For $\theta, p$ satisfying (\[index\]), $$b\in L^{\infty}(\mR^d)\cap\mW^{\theta}_p.$$ Then, we have: Under [(H$\sigma'$)]{}-[(Hb$'$)]{}, there exist constants $C_1, C_2$ such that for a.e. $x,y\in\mR^d$, $$\begin{aligned} \|\tilde b(x)-\tilde b(y)\|\leq C_1\|x-y\|\cdot\Big(1+\zeta\big(\Phi^{-1}(x)\big)+\zeta\big(\Phi^{-1}(y)\big)\Big)\label{b}\end{aligned}$$ and $$\begin{aligned} \|\tilde g(x,z)-\tilde g(y,z)\|\leq C_2\|x-y\|\cdot\Big(\cM\|\nabla\cJ_zu\|(\Phi^{-1}(x))+\cM\|\nabla\cJ_zu\|(\Phi^{-1}(y))\Big).\label{g}\end{aligned}$$ Moreover, for any $p>1$ and $\gamma\in(1,2)$, it holds for all $f\in\mH^{\gamma}_p$ that $$\begin{aligned} \\|\cJ_zf\\|_{1,p}\leq C_{p,d,\gamma}\|z\|^{\gamma-1}\\|f\\|_{\gamma,p},\label{jz}\end{aligned}$$ where $C_{p,d,\gamma}$ is a positive constant. Recall the definition of $\tilde b$ and $\tilde g$ in Lemma \[zvon\]. Since $\sigma$ is bounded and thanks to (\[es2\]), (\[upd\]), (\[a1\]), we get $$\begin{aligned} \|\tilde b(x)-\tilde b(y)\|&\leq \lambda\big\|u\big(\Phi^{-1}(x)\big)-u\big(\Phi^{-1}(y)\big)\big\|+\int_{\|z\|>1}\!\big\|u\big(\Phi^{-1}(x)+z\big)-u\big(\Phi^{-1}(y)+z\big)\big\|\nu(\dif z)\\ &\quad+\int_{\|z\|>1}\!\big\|u\big(\Phi^{-1}(x)\big)-u\big(\Phi^{-1}(y)\big)\big\|\nu(\dif z)+\int_{\|z\|>1}\!\big\|\sigma\big(\Phi^{-1}(x),z\big)-\sigma\big(\Phi^{-1}(y),z\big)\big\|\nu(\dif z)\\ &\leq C_\lambda\|x-y\|+C_0\|x-y\|\Big(\zeta\big(\Phi^{-1}(x)\big)+\zeta\big(\Phi^{-1}(y)\big)\Big),\end{aligned}$$ which gives (\[b\]). By (\[w11\]), further have $$\begin{aligned} \|\tilde g(x,z)-\tilde g(y,z)\|&=\big\|\Phi\big(\Phi^{-1}(x)+z\big)-\Phi\big(\Phi^{-1}(x)\big)-\Phi\big(\Phi^{-1}(y)+z\big)+\Phi\big(\Phi^{-1}(y)\big)\big\|\\ &=\big\|u\big(\Phi^{-1}(x)+z\big)-u\big(\Phi^{-1}(x)\big)-u\big(\Phi^{-1}(y)+z\big)+u\big(\Phi^{-1}(y)\big)\big\|\\ &=\big\|\big(\cJ_zu\big)(\Phi^{-1}(x))-\big(\cJ_zu\big)(\Phi^{-1}(y))\big\|\\ &\leq C_2\|x-y\|\cdot\Big(\cM\|\nabla\cJ_zu\|(\Phi^{-1}(x))+\cM\|\nabla\cJ_zu\|(\Phi^{-1}(y))\Big).\end{aligned}$$ As for (\[jz\]), it was proved by [@Zh00 Lemma 2.3]. Proof of the main result ======================== Now, we are ready to give the proof of our main result. Comparing with the usual SDEs driven by Brownian motion or pure jump Lévy processes, for SDEs of the form (\[sde2\]) or (\[sde3\]), some new tricks are needed to handle the term $1_{[0,\sigma(X_s,z)]}(r)$, as we shall see below. The point is that we have to use $L^1$-estimate as well as $L^2$-estimate to deduce the pathwise uniqueness. The proof will consist of two steps.\ [Step 1:]{} We assume that [(H$\sigma'$)]{}-[(Hb$'$)]{} hold. It was shown in [@M-P Proposition 3] that under these conditions, there exists a martingale solution to operator $\sL$. Meanwhile, it is known that the martingale solution for $\sL$ is equivalent to the weak solution to SDE (\[sde2\]), see [@Kurz2 Lemma 2.1]. Thus, the existence and uniqueness of weak solution hold for SDE (\[sde2\]). Thus, it suffices to show the pathwise uniqueness. Let $X_t$ and $\hat X_t$ be two strong solutions for SDE (\[sde2\]) both starting from $x\in\mR^d$, and set $$Y_t:=\Phi(X_t),\quad \hat Y_t:=\Phi(\hat X_t).$$ Since the uniqueness if a local property, as the argument in [@Ik-Wa Theorem IV. 9.1] and [@Zh00], we only need to prove by Lemma \[zvon\] that $$\begin{aligned} Z_t\equiv0,\quad\forall t\geq 0, \label{66}\end{aligned}$$ where $Z_t$ is given by $$\begin{aligned} Z_{t\wedge\tau_1}&=\int_0^{t\wedge\tau_1}\!\big[\tilde b(Y_s)-\tilde b(\hat Y_s)\big]\dif s +\!\int_0^{t\wedge\tau_1}\!\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}\Big[\tilde g(Y_{s-},z)1_{[0,\tilde\sigma(Y_{s-},z)]}(r)\\ &\quad\qquad\quad\qquad\quad\qquad\quad-\tilde g(\hat Y_{s-}, z)1_{[0,\tilde\sigma(\hat Y_{s-},z)]}(r)\Big]\tilde \cN(\dif z\times\dif r\times \dif s)=:\cJ^{t\wedge\tau_1}_1+\cJ^{t\wedge\tau_1}_2.\end{aligned}$$ Set $$A_1(t):=\int_0^t\Big(1+\zeta(X_s)+\zeta(\hat X_s)\Big)\dif s,$$ then following by an approximation argument as in [@Zh3; @Zh00], it is easy to see by (\[b\]) that for almost all $\omega$ and every stopping time $\eta$, $$\begin{aligned} \sup_{t\in[0,\eta]}\left\|\cJ^{t\wedge\tau_1}_1\right\|\leq C_1\!\int_0^{\tau_1\wedge\eta}\|Z_s\|\cdot\Big(1+\zeta(X_s)+\zeta(\hat X_s)\Big)\dif s=C_1\!\int_0^{\tau_1\wedge\eta}\|Z_s\| \dif A_1(s).\end{aligned}$$ As for the second term, write $$\begin{aligned} \cJ_2^{t\wedge\tau_1}&=\int_0^{t\wedge\tau_1}\!\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}1_{[0,\tilde\sigma(Y_{s-},z)\wedge\tilde\sigma(\hat Y_{s-},z)]}(r)\Big[\tilde g(Y_{s-},z)1_{[0,\tilde\sigma(Y_{s-},z)]}(r)\\ &\qquad\qquad\qquad\qquad\qquad\qquad\qquad-\tilde g(\hat Y_{s-}, z)1_{[0,\tilde\sigma(\hat Y_{s-},z)]}(r)\Big]\tilde \cN(\dif z\times\dif r\times \dif s)\\ &\quad+\int_0^{t\wedge\tau_1}\!\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}1_{[\tilde\sigma(Y_{s-},z)\vee\tilde\sigma(\hat Y_{s-},z),\infty]}(r)\Big[\tilde g(Y_{s-},z)1_{[0,\tilde\sigma(Y_{s-},z)]}(r)\\ &\qquad\qquad\qquad\qquad\qquad\qquad\qquad-\tilde g(\hat Y_{s-}, z)1_{[0,\tilde\sigma(\hat Y_{s-},z)]}(r)\Big]\tilde \cN(\dif z\times\dif r\times \dif s)\\ &\quad+\int_0^{t\wedge\tau_1}\!\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}1_{[\tilde\sigma(Y_{s-},z)\wedge\tilde\sigma(\hat Y_{s-},z),\tilde\sigma(Y_{s-},z)\vee\tilde\sigma(\hat Y_{s-},z)]}(r)\Big[\tilde g(Y_{s-},z)1_{[0,\tilde\sigma(Y_{s-},z)]}(r)\\ &\qquad\qquad\qquad\qquad\qquad\qquad\qquad-\tilde g(\hat Y_{s-}, z)1_{[0,\tilde\sigma(\hat Y_{s-},z)]}(r)\Big]\tilde \cN(\dif z\times\dif r\times \dif s)\\ &=:\cJ_{21}^{t\wedge\tau_1}+\cJ_{22}^{t\wedge\tau_1}+\cJ_{23}^{t\wedge\tau_1}.\end{aligned}$$ We proceed to estimate each component. First, for $\cJ_{21}^{t\wedge\tau_1}$, we use the Doob’s $L^2$-maximal inequality to deduce that for any stopping time $\eta$, $$\begin{aligned} \mE\left[\sup_{t\in[0,\eta]}\|\cJ_{21}^{t\wedge\tau_1}\|\right]&\leq \mE\Bigg(\int_0^{\tau_1\wedge\eta}\!\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}1_{[0,\tilde\sigma(Y_s,z)\wedge\tilde\sigma(\hat Y_s,z)]}(r)\big\|\tilde g(Y_s,z)-\tilde g(\hat Y_s, z)\big\|^2\dif r\nu(\dif z) \dif s\Bigg)^{\frac{1}{2}}\\ &=\mE\Bigg(\int_0^{\tau_1\wedge\eta}\!\!\!\!\int_{\|z\|\leq 1}\big[\tilde\sigma(Y_s,z)\wedge\tilde\sigma(\hat Y_s,z)\big]\cdot\big\|\tilde g(Y_s,z)-\tilde g(\hat Y_s, z)\big\|^2\nu(\dif z) \dif s\Bigg)^{\frac{1}{2}}.\end{aligned}$$ Thus, if we set $$A_2(t):=\int_0^t\!\!\!\int_{\|z\|\leq 1}\!\Big(\cM\|\nabla\cJ_zu\|(X_s)+\cM\|\nabla\cJ_zu\|(\hat X_s)\Big)^2\nu(\dif z) \dif s,$$ we further have by the fact that $\tilde\sigma$ is bounded and (\[g\]) that $$\begin{aligned} \mE\left[\sup_{t\in[0,\eta]}\|\cJ_{21}^{t\wedge\tau_1}\|\right]&\leq C_2\mE\Bigg(\int_0^{\tau_1\wedge\eta}\|Z_s\|^2\!\int_{\|z\|\leq 1}\!\Big(\cM\|\nabla\cJ_zu\|(X_s)+\cM\|\nabla\cJ_zu\|(\hat X_s)\Big)^2\nu(\dif z) \dif s\Bigg)^{\frac{1}{2}}\\ &=C_2\mE\Bigg(\int_0^{\tau_1\wedge\eta}\|Z_s\|^2\dif A_2(s)\Bigg)^{\frac{1}{2}}.\end{aligned}$$ Next, it is easy to see that for any $t\geq 0$, $$\cJ_{22}^{t\wedge\tau_1}\equiv0.$$ Finally, we use the $L^1$-estimate (see [@Kurz3 P$_{174}$] or [@Kurz2 P$_{157}$]) to control the third term by $$\begin{aligned} \mE\left[\sup_{t\in[0,\eta]}\|\cJ_{23}^{t\wedge\tau_1}\|\right]&\leq 2\mE\!\int_0^{\tau_1\wedge\eta}\!\!\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}1_{[\tilde\sigma(Y_s,z)\wedge\tilde\sigma(\hat Y_s,z),\tilde\sigma(Y_s,z)\vee\tilde\sigma(\hat Y_s,z)]}(r)\\ &\qquad\qquad\qquad\times\big\|\tilde g(Y_s,z)1_{[0,\tilde\sigma(Y_s,z)]}(r)-\tilde g(\hat Y_s, z)1_{[0,\tilde\sigma(\hat Y_s,z)]}(r)\big\|\nu(\dif z)\dif r \dif s\\ &\leq2\mE\!\int_0^{\tau_1\wedge\eta}\!\!\!\int_{\|z\|\leq 1}\|\tilde\sigma(Y_s,z)-\tilde\sigma(\hat Y_s,z)\|\cdot\Big(\|\tilde g(Y_s,z)\|+\|\tilde g(\hat Y_s,z)\|\Big)\nu(\dif z)\dif s.\end{aligned}$$ Since $$\|\tilde g(x,z)\|=\big\|\Phi\big(\Phi^{-1}(x)+z\big)-\Phi\big(\Phi^{-1}(x)\big)\big\|\leq \frac{3}{2}\|z\|,$$ and taking into account of (\[a1\]), we get $$\begin{aligned} \mE\left[\sup_{t\in[0,\eta]}\|\cJ_{23}^{t\wedge\tau_1}\|\right]&\leq C_3\mE\!\int_0^{\tau_1\wedge\eta}\!\!\!\int_{\|z\|\leq 1}\|\tilde\sigma(Y_s,z)-\tilde\sigma(\hat Y_s,z)\|\cdot\|z\|\nu(\dif z)\dif s\\ &\leq C_3\mE\!\int_0^{\tau_1\wedge\eta}\|Z_s\|\Big(\zeta(X_s)+\zeta(\hat X_s)\Big)\dif s\leq C_3\mE\!\int_0^{\tau_1\wedge\eta}\|Z_s\|\dif A_1(s).\end{aligned}$$ Combing the above computations, and set $$A(t):=A_1(t)+A_2(t),$$ we arrive at that for any stopping time $\eta$, there exists a constant $C_0$ such that $$\begin{aligned} \mE\left[\sup_{t\in[0,\eta]}\|Z_{t\wedge\tau_1}\|\right]\leq C_0\mE\!\int_0^{\tau_1\wedge\eta}\|Z_s\|\dif A(s)+C_0\mE\Bigg(\int_0^{\tau_1\wedge\eta}\|Z_s\|^2\dif A(s)\Bigg)^{\frac{1}{2}}. \label{ee}\end{aligned}$$ By our assumption that $\zeta\in L^{q}(\mR^d)$ with $q>d/\alpha$ and the Krylov estimate (\[kry1\]), we find that $$\mE A_1(t)\leq t+C\\|\zeta\\|_q<\infty.$$ Meanwhile, since $p>2d/\alpha$, using the Fubini’s theorem, Krylov estimate, Minkovski’s inequality and taken into account of (\[mf\]), we can get $$\begin{aligned} \mE A_2(t)&=\int_{\|z\|\leq 1}\!\mE\!\int_0^t\Big(\cM\|\nabla\cJ_zu\|(X_s)+\cM\|\nabla\cJ_zu\|(\hat X_s)\Big)^2\dif s\nu(\dif z)\\ &\leq C\!\int_{\|z\|\leq 1}\!\\|(\cM\|\nabla\cJ_zu\|)^2\\|_{p/2}\nu(\dif z)= C\!\int_{\|z\|\leq 1}\!\\|\cM\|\nabla\cJ_zu\|\\|_{p}^2\nu(\dif z)\\ &\leq C\!\int_{\|z\|\leq 1}\\|\nabla\cJ_zu\\|_{p}^2\nu(\dif z)\leq C\!\int_{\|z\|\leq 1}\\|\cJ_zu\\|_{1,p}^2\nu(\dif z).\end{aligned}$$ Recall our assumption that $\theta>1-\tfrac{\alpha}{2}$. Hence, we can choose a $\gamma<\alpha$ such that $$2(\gamma+\theta-1)>\alpha.$$ Consequently, it follows from (\[jz\]) and (\[nu\]) that $$\begin{aligned} \mE A_2(t)\leq C\\|u\\|_{\gamma+\theta,p}^2\!\int_{\|z\|\leq 1}\|z\|^{2(\gamma+\theta-1)}\nu(\dif z)<\infty.\end{aligned}$$ Therefore, $t\mapsto A(t)$ is a continuous strictly increasing process. Define for $t\geq 0$ the stopping time $$\eta_t:=\inf\{s\geq 0: A(s)\geq t\}.$$ Then, it is clear that $\eta_t$ is the inverse of $t\mapsto A(t)$. Since $A(t)\geq t$, we further have $\eta_t\leq t$. Taking $\eta_t$ into (\[ee\]), we have by the change of variable $$\begin{aligned} &\mE\left[\sup_{s\in[0,t]}\|Z_{\tau_1\wedge\eta_s}\|\right]=\mE\left[\sup_{s\in[0,\eta_t]}\|Z_{s\wedge\tau_1}\|\right]\\ &\leq C_0\mE\!\int_0^{\tau_1\wedge\eta_t}\|Z_s\|\dif A(s)+C_0\mE\Bigg(\int_0^{\tau_1\wedge\eta_t}\|Z_s\|^2\dif A(s)\Bigg)^{\frac{1}{2}}\\ &\leq C_0\mE\!\int_0^{\eta_t}\|Z_{\tau_1\wedge s}\|\dif A(s)+C_0\mE\Bigg(\int_0^{\eta_t}\|Z_{\tau_1\wedge s}\|^2\dif A(s)\Bigg)^{\frac{1}{2}}\\ &=C_0\mE\!\int_0^{t}\|Z_{\tau_1\wedge\eta_s}\|\dif s+C_0\mE\Bigg(\int_0^{t}\|Z_{\tau_1\wedge\eta_s}\|^2\dif s\Bigg)^{\frac{1}{2}}\leq C_0\big(t+\sqrt{t}\big)\mE\left[\sup_{s\in[0,t]}\|Z_{\tau_1\wedge\eta_s}\|\right].\end{aligned}$$ Now, take $t_0$ small enough such that $$C_0\big(t_0+\sqrt{t_0}\big)<1,$$ it holds that for almost all $\omega$, $$\sup_{s\in[0,\eta_{t_{0}}]}\|Z_{s\wedge\tau_1}\|=\sup_{s\in[0,t_0]}\|Z_{\tau_1\wedge\eta_s}\|=0.$$ In particular, $$Z_{\eta_{t_0}\wedge\tau_1}=0,\quad a.s..$$ The same arguments as above and write for $t>t_0$, $$\begin{aligned} Z_{t\wedge\tau_1}&=\int_{\eta_{t_0}\wedge\tau_1}^{t\wedge\tau_1}\!\big[\tilde b(Y_s)-\tilde b(\hat Y_s)\big]\dif s+\int_{\eta_{t_0}\wedge\tau_1}^{t\wedge\tau_1}\!\!\int_0^{\infty}\!\!\!\!\int_{\|z\|\leq 1}\Big[\tilde g(Y_{s-},z)1_{[0,\tilde\sigma(Y_{s-},z)]}(r)\\ &\quad\qquad\quad\qquad\quad\qquad\quad\qquad\quad\qquad-\tilde g(\hat Y_{s-}, z)1_{[0,\tilde\sigma(\hat Y_{s-},z)]}(r)\Big]\tilde \cN(\dif z\times\dif r\times \dif s),\end{aligned}$$ we can get $$\begin{aligned} &\mE\left[\sup_{s\in[t_0,t]}\|Z_{\tau_1\wedge\eta_s}\|\right]=\mE\left[\sup_{s\in[\eta_{t_0},\eta_t]}\|Z_{s\wedge\tau_1}\|\right]\\ &\leq C_0\mE\!\int_{\eta_{t_0}\wedge\tau_1}^{\tau_1\wedge\eta_t}\|Z_s\|\dif A(s)+C_0\mE\Bigg(\int_{\eta_{t_0}\wedge\tau_1}^{\tau_1\wedge\eta_t}\|Z_s\|^2\dif A(s)\Bigg)^{\frac{1}{2}}\\ &\leq C_0\mE\!\int_{t_0}^{t}\|Z_{\tau_1\wedge\eta_s}\|\dif s+C_0\mE\Bigg(\int_{t_0}^{t}\|Z_{\tau_1\wedge\eta_s}\|^2\dif s\Bigg)^{\frac{1}{2}}\\ &\leq C_0\Big[(t-t_0)+\sqrt{(t-t_0)}\Big]\mE\left[\sup_{s\in[t_0,t]}\|Z_{\tau_1\wedge\eta_s}\|\right].\end{aligned}$$ Hence, for for almost all $\omega$, $$\sup_{s\in[0,\eta_{2t_{0}}]}\|Z_{s\wedge\tau_1}\|=0.$$ Repeating the above arguments, we may get for any $k>0$, $$\sup_{s\in[0,\eta_{kt_{0}}]}\|Z_{s\wedge\tau_1}\|=0.$$ Noticing that $\eta_t$ is also strictly increasing, we have for all $t\geq 0$, $$Z_{t\wedge\tau_1}=0,\quad a.s..$$ Thus, (\[66\]) is proven.\ [Step 2:]{} Assume now that $\sigma$ and $b$ satisfy [(H$\sigma$)]{}-[(Hb)]{}. For each $n\in\mN$, let $\chi_n(x)\in[0,1]$ be a nonnegative smooth function in $\mR^d$ with $\chi_n(x)=1$ for all $x\in B_n$ and $\chi_n(x)=0$ for all $x\notin B_{n+1}$. Let $$b_n(x):=\chi_n(x)b(x),\quad \zeta_n(x):=\chi_{n+1}(x)\zeta(x)+\cM\|\nabla\chi_{n+1}\|(x),$$ and $$\sigma_n(x,z):=1+\chi_{n+1}(x)\sigma(x,z)\big(\|z\|\wedge 1\big)+\Big(1-\chi_n(x)\big(\|z\|\wedge 1\big)\Big)\left(1+\sup_{x\in B_{n+2}}\|\sigma(x,z)\|\right)\mI_{d\times d}.$$ Then, one can check easily that (\[s1\]) holds and $b_n$ satisfies [(Hb$'$)]{}. Meanwhile, $$\begin{aligned} &\int_{\mR^d}\|\sigma_n(x,z)-\sigma_n(y,z)\|(\|z\|\wedge1)\nu(\dif z)\leq C\!\!\int_{\mR^d}\|\chi_{n+1}(x)-\chi_{n+1}(y)\|\big(\|z\|^2\wedge1\big)\nu(\dif z)\\ &\quad\leq C\!\!\int_{\mR^d}\|\sigma(x,z)-\sigma(y,z)\|(\|z\|\wedge1)\nu(\dif z)\Big(\chi_{n+1}(x)\wedge\chi_{n+1}(y)\Big)\\ &\quad\leq C\|x-y\|\Big(\cM\|\nabla\chi_{n+1}\|(x)+\cM\|\nabla\chi_{n+1}\|(y)\Big)+C\|x-y\|\Big(\chi_{n+1}(x)\zeta(x)+\chi_{n+1}(y)\zeta(y)\Big).\end{aligned}$$ It is obvious that $\zeta_n\in L^q(\mR^d)$ with $q>d/\alpha$, hence [(H$\sigma'$)]{} is also true. Therefore, for each $x\in\mR^d$, there exist a unique strong solution $X_t^n(x)$ to SDE (\[sde2\]) with coefficients $\sigma_n$ and $b_n$. For $n\geq k$, define $$\varsigma_{n,k}(x):=\inf\{t\geq0: \|X_t^n(x)\|\geq k\}\wedge n.$$ By the uniqueness of the strong solution, we have $$\mP\Big(X_t^n(x)=X_t^k(x),\,\forall t\in[0,\varsigma_{n,k}(x))\Big)=1,$$ which implies that for $n\geq k$, $$\varsigma_{k,k}(x)\leq \varsigma_{n,k}(x)\leq \varsigma_{n,n}(x),\quad a.s..$$ Hence, if we let $\varsigma_k(x):=\varsigma_{k,k}(x)$, then $\varsigma_k(x)$ is an increasing sequence of stopping times and for $n\geq k$, $$\mP\Big(X_t^n(x)=X_t^k(x),\,\forall t\in[0,\varsigma_{k}(x))\Big)=1.$$ Now, for each $k\in\mN$, we can define $X_t(x):=X^k_t(x)$ for $t<\varsigma_k(x)$ and $\varsigma(x):=\lim_{k\rightarrow\infty}\varsigma_k(x)$. It is clear that $X_t(x)$ is the unique strong solution to SDE (\[sde2\]) up to the explosion time $\varsigma(x)$ and (\[xt\]) holds. [10]{} Budhiraja A., Dupuis P. and Maroulas V.: Variational representations for continuous time processes. Ann. Inst. Henri. Poincar. Probab. Stat., [[47]{}]{} (2011), 725–747. Budhiraja A., Chen J. and Dupuis P.: Large deviations for stochastic partial differential equations driven by a Poisson random measure. Stochastic Process. Appl., [[123]{}]{} (2013), 523–560. Bass R. F., Burdzy K. and Chen Z.: Stochastic differential equations driven by stable processes for which pathwise uniqueness fails. Stoch. Proc. Appl., [[111]{}]{} (2004), 1–15. Caffarelli L. and Silvestre L.: The Evans-Krylov theorem for nonlocal fully nonlinear equations. Ann. Math., [174]{} (2011), 1163–1187. Chen Z., Kim P. and Kumagai T.: Global heat kernel estimates for symmetric jump processes. Trans. Amer. Math. Soc., [363]{} (2011), 5021–5055. Chen Z. and Kumagai T.: Heat kernel estimates for stable-like processes on $d$-sets. Stochastic Process. Appl., [[108]{}]{} (2003), 27–62. Chen Z., Song R. and Zhang X.: Stochastic flows for Lévy processes with Hölder drift. http://arxiv.org/abs/1501.04758. Chen Z. and Zhang X.: Heat kernel and analyticity of non-symmetric jump diffusion semigroups. Prob. Theory and Related Fields, [156]{} (2015), 1–46. Eidelman S. D., Ivasyshen S. D. and Kochubei A. N.: Analytic Methods in the Theory of Differential and Pseudo-differential Equations of Parabolic Type. Birkhauser, Basel (2004). Fang S., Luo D. and Thalmaierb A.: Stochastic differential equations with coefficients in Sobolev spaces. J. Funct. Anal., [259]{} (2010), 1129–1168. Fedrizzi E. and Flandoli F.: Hölder Flow and Differentiability for SDEs with Nonregular Drift. Sto. Ana. and App., [[31]{}]{} (2013), 708–736. Flandoli F., Gubinelli M. and Priola E.: Well-posedness of the transport equation by stochastic perturbation. Invent. Math., [180]{}(1) (2010), 1–53. Ikeda N. and Watanabe S.: Stochastic Differential Equations and Diffusion Processes, 2nd edition. North-Holland, Kodansha, 1989. Imkellera P. and Willrich N.: Solutions of martingale problems for Lévy-type operators with discontinuous coefficients and related SDEs. Sto. Pro. App., [[126]{}]{} (2016), 703–734. Kim P. and Song R.: Stable process with singular drift. Stoch. Proc. Appl., [124]{} (2014), 2479–2516. Kochubei A. N.: Parabolic pseudodifferential equations, hypersingular integrals and Markov processes. Math. USSR Izv., [33]{} (1989), 233–259. \[translation from Izv. Akad. Nauk SSSR Ser. Mat., [52]{} (1988), 909–934.\] Krylov N. V. and Röckner M.: Strong solutions of stochastic equations with singular time dependent drift. Probab. Theory Related Fields, [[131]{}]{}(2) (2005), 154–196. Kurtz T. G.: Martingale Problems, Particles and Filters. http://www.math.wisc.edu/$\sim$kurtz/Lectures/ill06pst.pdf. Kurtz T. G.: Equivalence of Stochastic Equations and Martingale Problems. Stochastic Analysis, (2010), 113–130. Kurtz T. G. and Protter P. E.: Weak convergence of stochastic integrals and differential equations. II. Infinite-dimensional case. In Probabilistic models for nonlinear partial differential equations (Montecatini Terme, 1995), volume 1627 of Lecture Notes in Math., pages 197–285. Springer, Berlin, 1996. Menoukeu P. O., Meyer B. T., Nilssen T., Proske F. and Zhang T.: A variational approach to the construction and Malliavin differentiability of strong solutions of SDEs. Math. Ann., [[357]{}]{} (2013), 761–799. Mikulevcius R. and Pragarauskas H.: On the Cauchy problem for integro-differential operators in Sobolev classes and the martingale problem. J. Diff. Eq., [256]{} (2014), 1581–1626. Mikulevcius R. and Xu F.: On the rate of converge of strong Euler approximation for SDEs driven by Levy processes. https://arxiv.org/pdf/1608.02303.pdf. Priola E.: Pathwise uniqueness for singular SDEs driven by stable processes. Osaka Journal of Mathematics, [49]{} (2012), 421–447. Priola E.: Stochastic flow for SDEs with jumps and irregular drift term. http://arXiv:1405.2575v1. Stein E. M.: Singular integrals and differentiability properties of functions. Princeton Mathematical Series 30, Princeton University Press, Princeton, NJ, 1970. Tanaka H., Tsuchiya M. and Watanabe S.: Perturbation of drift-type for Lévy processes. J. Math. Kyoto Univ., [14]{} (1974), 73–92. Triebel H.: Interpolation Theory, Function Spaces, Differeential Operators. North-Holland Publishing Company, Amsterdam, 1978. Wang F. Y.: Gradient Estimates and Applications for SDEs in Hilbert Space with Multiplicative Noise and Dini Continuous Drift. J. Diff. Eq., [[3]{}]{} (2016), 2792–2829. Wang F. Y.: Integrability Conditions for SDEs and Semi-Linear SPDEs. http://arxiv.org/pdf/1510.02183.pdf. Xie L.: Singular SDEs with critical non-local and non-symmetric Lévy type generator. Stoch. Proc. App., (2017), http://dx.doi.org/10.1016/j.spa.2017.03.014. Xie L. and Zhang X.: Sobolev differentiable flows of SDEs with local Sobolev and super-linear growth coefficients. Ann. Prob., [44]{}(6) (2016), 3661–3687. Zhai J. and Zhang T.: Large deviations for 2-D stochastic Navier-Stokes equations driven by multiplicative Lévy noises. Bernoulli, [21]{} (2015), 2351–2392. Zhang X.: Strong solutions of SDEs with singular drift and Sobolev diffusion coefficients. Stoch. Proc. Appl., [[115]{}]{} (2005), 1805–1818. Zhang X.: Stochastic homemomorphism flows of SDEs with singular drifts and Sobolev diffusion coefficients. Electron. J. Probab., [[16]{}]{} (2011), 1096–1116. Zhang X.: Stochastic differential equations with Sobolev drifts and driven by $\alpha$-stable processes. Ann. Inst. H. Poincare Probab. Statist., [49]{} (2013), 915–931. [^1]: The first author is supported by the Project Funded by PAPD of Jiangsu Higher Education Institutions. The second author is partially supported by the grants NSFC No 11571390, Macao S.A.R. FDCT/030/2016/A1 and University of Macau MYRG2016-00025-FST	{ "pile_set_name": "ArXiv" }
--- abstract: 'Spallative nucleosynthesis is thought to be the only process capable of producing significant amount of Beryllium (Be) in the universe. Therefore, both energetic particles (EPs) and nuclei to be spalled (most efficiently C, N and O nuclei in this case) are required, which indicates that supernovae (SNe) may be directly involved in the synthesis of the Be nuclei observed in the halo stars of the Galaxy. We apply current knowledge relating to supernova remnant (SNR) evolution and particle shock acceleration to calculate the total Be yield associated with a SN explosion in the interstellar medium, focusing on the first stages of Galactic chemical evolution (i.e. when metallicity $0\le Z\le 10^{-2}$). We show that dynamical aspects must be taken into account carefully, and present analytical calculations of the spallation reactions induced by the EPs accelerated at both the forward and the reverse shocks following the SN explosion. Our results show that the production of Be in the early Galaxy is still poorly understood, and probably implies either selective acceleration processes (greatly favouring CNO acceleration), reconsideration of the observational data (notably the O vs Fe correlation), or even new energy sources.' author: - Etienne Parizot - Luke Drury date: 'Recieved date; accepted date' subtitle: 'I. Analytical Estimates' title: Spallative Nucleosynthesis in Supernova Remnants --- Introduction ============ There has recently been considerable interest in the Galactic evolution of the abundances of the light elements Li, Be and B (Feltzing & Gustafsson 1994; Reeves 1994; Cassé et al. 1995; Fields et al. 1994,1995; Ramaty et al. 1996,1997; Vangioni-Flam et al. 1998). The Be abundance is particularly interesting because this element is thought to be produced exclusively by spallation reactions involving collisions between nuclei of the CNO group of elements and protons or alpha particles at energies greater than about $30\,\rm MeV$ per nucleon (MeV/n). Thus the evolution of the Be abundance contains information about the particle acceleration and cosmic ray history of the Galaxy. The evolution of the Be abundance, and indeed the evolution of all elemental abundances, has to be deduced from observations of the fossil abundances preserved in the oldest halo stars. Advances in spectroscopy over the last decade have greatly improved the quality of the data available (Duncan et al. 1992,1997; Edvardsson et al. 1994; Gilmore et al. 1992; Kiselman & Carlsson 1996; Molaro et al. 1997; Ryan et al. 1994) and the main result is easily summarized : in old halo stars of low metallicity, the ratio of the Be abundance to the Iron (Fe) abundance appears constant, that is to say the Be abundance rises [linearly]{} with the Fe abundance. This has been a surprising result. Naively one had expected that, because Be is a secondary product produced from the primary CNO nuclei, its abundance should vary [quadratically]{} as a function of the primary abundances at low metallicities. Indeed, considering that the cosmic rays (CRs) responsible for the Be production are somehow related to the explosion of supernovae (SNe) in the Galaxy, it is natural to assume that their flux is proportional to the SN rate, $\d N_{\mathrm{SN}}/\d t$. Now since the number of CNO nuclei present in the Galaxy at time $t$ is proportional to the total number of SN having already exploded, $N_{\mathrm{SN}}(t)$, the Be production rate has to be proportional to $N_{\mathrm{SN}}\d N_{\mathrm{SN}}/\d t$. Therefore, the integrated amount of Be grows as $N_{\mathrm{SN}}^{2}$, that is quadratically with respect to the ambient metallicity (C,N,O or Fe, assumed to be more or less proportional to one another). The above reasoning, however, relies on two basic assumptions that need not be fulfilled : i) the CRs recently accelerated interact with all the CNO nuclei already produced and dispersed in the entire Galaxy and ii) the CRs are made of the ambient material, dominated by H and He nuclei. Instead, it might be i) that the proton rich CRs recently accelerated interact predominantly with the freshly synthesized CNO nuclei near the explosion site and ii) that a significant fraction of the CNO rich SN ejecta are also accelerated. In both cases, a linear growth of the Be abundance with respect to Fe or O would arise very naturally, since the number of Be-producing spallation reactions induced by each individual supernova would be directly linked to its local, individual CNO supply, independently of the accumulated amount of CNO in the Galaxy. In fact, as emphasised by Ramaty et al. (1997), the simplest explanation of the observational data is to assume that each core-collapse supernova produces on average 0.1 of Fe, one to few of the CNO elements and $2.8\times10^{-8}\,\Msun$ of Be, [with no metallicity dependence]{}. Clearly if this is the case and the production of Be is directly linked to that of the main primary elements, the observed linear relation between Be and Fe will be reproduced whatever the complications of infall, mixing and outflow required by the Galactic evolution models. On the other hand, although the simplest explanation of the data is clearly to suppose a primary behaviour for the Be production, it is possible that this could be an artifact of the evolutionary models (as argued, e.g., by Casuso & Beckman 1997). Most work in this area has attempted to deduce information about cosmic ray (or other accelerated particle populations) in the early galaxy by working backwards from the abundance observations. While perfectly legitimate, our feeling is that the observational errors and the uncertainties relating to Galactic evolution in general make this a very difficult task. We have chosen to approach the problem from the other direction and ask what currently favoured models for particle acceleration in supernova remnants (SNRs) imply for light element production. This is in the general spirit of recent calculations of the $\pi^0$-decay gamma-ray luminosity of SNRs (Drury et al. 1994) and the detailed chemical composition of SNR shock accelerated particles (Ellison, Drury & Meyer, 1997) where we look for potentially observable consequences of theoretical models for cosmic ray production in SNRs. Interestingly enough, the study of particle acceleration in SNRs suggests that both alternatives to the naive scenario mentioned above do occur in practice, as demonstrated qualitatively in Sect. \[PartAccInSNRs\]. The first of these alternatives, namely the local interaction of newly accelerated cosmic rays in the vicinity of SN explosion sites, has already been called upon by Feltzing & Gustaffson (1994), as well as the second, the acceleration of enriched ejecta through a SN reverse shock, by Ramaty et al. (1997). However, no careful calculations have yet been done, taking the dynamics of the process into account, notably the dilution of SN ejecta and the adiabatic losses. Yet we show below that they have a crucial influence on the total amount of Be produced, and that a time-dependent treatment is required. Indeed, the evolution of a SNR is essentially a dynamical problem in which the acceleration rate as well as the chemical composition inside the remnant are functions of time. The results of the full calculation of both processes and the discussion of their implications for the chemical evolution of the Galaxy will be found in an associated paper (Parizot & Drury 1999). Here we present simple analytical calculations which provide an accurate understanding of the dynamics of light element production in SNRs and elucidates the role and influence of the different parameters, notably the ambient density. Although Li and B are also produced in the processes under study, we shall choose here Be as our ‘typical’ light element, because nuclear spallation of CNO is thought to be its only production mechanism, while Li is also (and actually mainly) produced through $\alpha + \alpha$ reactions, $^{7}$Li may be produced partly in AGB stars (Abia et al. 1993), and $^{11}$B neutrino spallation may be important as seems to be required by chemical evolution analysis (Vangioni-Flam et al. 1996). In order to compare our results with the observations, we simply note that, as emphasized in Ramaty et al. (1997), the data relating to the Galactic Be evolution as a function of \[Fe/H\] indicate that $\sim 1.6\,10^{-6}$ nuclei of Be must be produced in the early Galaxy for each Fe nucleus. Therefore, if Be production is indeed induced, directly or indirectly, by SNe explosions, and since the average SN yield in Fe is thought to be $\sim 0.11\,\Msun$, each supernova must lead to an average production of $\sim 3.8\,10^{48}$ nuclei (or $\sim 2.8\,10^{-8}\,\Msun$) of Be, with an uncertainty of about a factor of 2 (Ramaty et al. 1997). We adopt this value as the ‘standard needed number’ of Be per supernova explosion. To state this again in a different way, for an average SN yield in CNO of, say, $\sim 1\,\Msun$, the required spallation rate per CNO atom is $\sim 3\,10^{-8}$. Particle acceleration in SNRs {#PartAccInSNRs} ============================= It is generally believed that cosmic ray production in SNRs occurs through the process of diffusive shock acceleration operating at the strong shock waves generated by the interaction between the ejecta from the supernova explosion and the surrounding medium. Significant effort has been put into developing dynamical models of SNR evolution which incorporate, at varying levels of detail, this basic acceleration and injection process (one of the major advantages of shock acceleration is that it does not require a separate injection process). Qualitatively the main features can be crudely summarised as follows. In a core collapse SN the collapse releases roughly the gravitational binding energy of a neutron star, some $10^{53}\,\rm erg$, but most of this is radiated away in neutrinos. About $\ESN = 10^{51}\,\rm erg$ is transferred, by processes which are still somewhat obscure, to the outer layers of the progenitor star which are then ejected at velocities of a few percent of the speed of light. Initially the explosion energy is almost entirely in the form of kinetic energy of these fast-moving ejecta. As the ejecta interact with the surrounding circumstellar and interstellar material they drive a strong shock ahead into the surrounding medium. The region of very hot high pressure shocked material behind this forward shock also drives a weaker shock backwards into the ejecta giving rise to a characteristic forward reverse shock pair separated by a rather unstable contact discontinuity. This initial phase of the remnant evolution lasts until the amount of ambient matter swept up by the remnant is roughly equal to the original ejecta mass. At this so-called sweep-up time, , the energy flux through the shocks is at its highest, the expansion of the remnant begins to slow down, and a significant part of the explosion energy has been converted from kinetic energy associated with the bulk expansion to thermal (and non-thermal) energy associated with microscopic degrees of freedom of the system. The remnant now enters the second, and main, phase of its evolution in which there is rough equipartition between the microscopic and macroscopic energy densities. The evolution in this phase is approximately self-similar and resembles the exact solution obtained by Sedov for a strong point explosion in a cold gas. It is important to realise that the approximate equality of the energy associated with the macroscopic and microscopic degrees of freedom in the Sedov-like phase is not a static equilibrium but is generated dynamically by two competing processes. As long as the remnant is compact the energy density, and thus pressure, of the microscopic degrees of freedom is very much greater than that of the external medium. This strong pressure gradient drives an expansion of the remnant which adiabatically reduces the microscopic degrees of freedom of the medium inside the remnant and converts the energy back into bulk kinetic energy of expansion. At the same time the strong shock which marks the boundary of the remnant converts this macroscopic kinetic energy of expansion back into microscopic internal form. Thus there is a continuous recycling of the original explosion energy between the micro and macro scales. This continues until either the external pressure is no longer negligible compared to the internal, or the time-scales become so long that radiative cooling becomes important. The time scales for the conversion of kinetic energy to internal energy and vice versa are roughly equal and of order the dynamical time scale of the remnant which is of order the age of the remnant, hence the approximately self-similar evolution. In terms of particle acceleration the theory assumes that strong collisionless shocks in a tenuous plasma automatically and inevitably generate an approximately power law distribution of accelerated particles which connects smoothly to the shock-heated particle distribution at ‘thermal’ energies and extends up to a maximum energy constrained by the shock size, speed, age and magnetic field. The acceleration mechanism is a variant of Fermi acceleration based on scattering from magnetic field structures on both sides of the shock. A key point is that these scattering structures are not those responsible for general scattering on the ISM, but strongly amplified local structures generated in a non-linear bootstrap process by the accelerated particles themselves. As long as the shock is strong it will be associated with strong magnetic turbulence which drives the effective local diffusion coefficient down to values close to the Bohm value. As pointed out by Achterberg et al. (1994) the extreme sharpness of the radio rims of some shell type SNRs can be interpreted as observational evidence for this type of effect. The source of free energy for the wave excitation is of course the strong gradient in the energetic particle distribution at the edge. Thus in the interior of the remnant, where the gradients are absent or much weaker, we do not expect such low values of the diffusion coefficient. The net effect is that the edge of the remnant, as far as the accelerated particles are concerned, is both a self-generated diffusion barrier and a source of freshly accelerated particles. Except at the very highest energies the particles produced at the shock are convected with the post-shock flow into the interior of the remnant and effectively trapped there until the shock weakens to the point where the self-generated wave field around the shock can no longer be sustained. At this point the diffusion barrier collapses and the trapped particle population is free to diffuse out into the general ISM. In terms of bulk energetics, the total energy of the accelerated particle population is low during the first ballistic phase of the expansion (because little of the explosion energy has been processed through the shocks) but rises rapidly as $t\approx\tSW$. During the sedov-like phase the total energy in accelerated particles is roughly constant at a significant fraction of the explosion energy (0.1 to 0.5 typically). However, this is because of the dynamic recycling described above. Any individual particle is subject to adiabatic losses on the dynamical time-scale of the remnant, while the energy lost this way goes into driving the shock and thus generating new particles, distributed over the whole energy spectrum. Spallation reactions within SNRs ================================ Qualitative overview -------------------- We now turn to the production of Li, Be and B (LiBeB) by spallation reactions within a SNR. As emphasized above, there are two obvious mechanisms. One is the irradiation of the CNO ejecta by accelerated protons and alphas. It is clear that the fresh CNO nuclei produced by the SN will, for the lifetime of the SNR, be exposed to a flux of energetic particles (EPs) very much higher than the average interstellar flux, and this must lead to some spallation production of light elements. This process starts at about with a very intense radiation field and continues with an intensity decreasing roughly as $R^{-3}\propto t^{-6/5}$ (where $R$ is the radius of the SNR) until the remnant dies. The second process is that some of the CNO nuclei from the ejecta are accelerated, either by the reverse shock in its brief powerful phase at $t\approx\tSW$ or by some of this material managing to get ahead of the forward shock. This later possibility is not impossible, but seems unlikely to be as important as acceleration by the reverse shock. Calculations of the Raleigh-Taylor instability of the contact discontinuity do suggest that some fast-moving blobs of ejecta can punch through the forward shock at about (Jun & Norman 1996), and in addition Ramaty and coworkers have suggested that fast moving dust grains could condense in the ejecta at $t<\tSW$ and then penetrate through into the region ahead of the main shock. In all these pictures acceleration of CNO nuclei takes place only at about and the energy deposited in these accelerated particles is certainly less than the explosion energy , although it might optimistically reach some significant fraction of that value (say $\la 10\%$). Crucially the accelerated CNO nuclei are then confined to the interior of the SNR and will thus be adiabatically cooled on a rather rapid time-scale, initially of order . Evaluation of the first process (forward shock) ----------------------------------------------- From the above arguments, it is clear that SNe do induce some Be production. Now the question is : how much? Let us first consider the irradiation of the ejecta by particles (H and He nuclei) accelerated at the forward shock during the Sedov-like phase – process 1. We have already indicated that detailed studies of acceleration in SNRs show that the fraction of the explosion energy given to the EPs is roughly constant during the Sedov-like phase and of order 0.1 to 0.5 or so. Let $\theta_{1}$ be that fraction. Since the EPs are distributed more or less uniformly throughout the interior of the remnant, the energy density can be estimated as $$\EdCR \approx \frac{3\theta_{1}\ESN}{4\pi R^{3}} \label{EdCR}$$ where $R$ is the radius of the remnant and is the explosion energy. To derive a spallation rate from this we need to assume some form for the spectrum of the accelerated particles. Shock acceleration suggests that the distribution function should be close to the test-particle form $f(p)\propto p^{-4}$ and extend from an injection momentum close to ‘thermal’ values to a cut-off momentum at about $p_{\mathrm{max}} = 10^{5}\,{\rm GeV}/c$. The spallation rate per target CNO atom to produce a Be atom is then obtained by integrating the cross sections $$\nu_{\rm spall} = \int_{\pth}^{p_{\mathrm{max}}} \sigma v f(p) \d p,$$ with the normalisation $\int E(p)f(p)4\pi p^{2}\d p = \EdCR$. Looking at graphs of the spallation cross-sections for Be (as given, e.g., in Ramaty et al. 1997), it is clear that these cross-sections can be well approximated as zero below a threshold at about 30–40 MeV/n and a constant value $\sigma_0\simeq 5\times 10^{-27}\,\mathrm{cm}^{2}$ above it. One then obtains roughly : $$\nu_{\rm spall} \approx \frac{\sigma_{0}}{mc} \EdCR \frac{1-\ln(\pth/mc)}{1 + \ln(p_{\rm max}/mc)} \simeq 0.2 \sigma_0 c \frac{\EdCR}{m c^2},$$ where $\pth \simeq mc/5$ is the momentum corresponding to the spallation threshold and $m$ refers to the proton mass. Fortunately, for this form of the spectrum the upper cut-off and the spallation threshold only enter logarithmically. A softer spectrum would lead to higher spallation yields and a stronger dependence on the spallation threshold. Using Eq. (\[EdCR\]) and the adiabatic expansion law for the forward shock radius, $R = R_{\mathrm{SW}}(t/\tSW)^{2/5}$, we can now estimate the total fraction of the CNO nuclei which will be converted to Be during the Sedov-like phase as : $$\begin{aligned} \phi_{1} &=& \int_{\tSW}^{\tend} 0.2 \sigma_0 c \frac{\theta_{1}\ESN}{mc^{2}} \frac{3}{4\pi R^3} \d t\\ &=& 0.2 \sigma_0 c \frac{\theta_{1}\ESN}{mc^{2}} \frac{3}{4\pi \RSW^3} \int_{\tSW}^{\tend} \left(\frac{t}{\tSW}\right)^{-6/5} dt, \label{phi1Interm}\end{aligned}$$ or $$\phi_{1} = \sigma_0 c \frac{\theta_{1}\ESN}{mc^{2}} \frac{\rho_{0}}{\Mej} \tSW \left[1 - \left(\frac{\tSW}{\tend}\right)^{1/5}\right] \label{phi1}$$ where as usual $\rho_{0}$ denotes the density of the ambient medium into which the SNR is expanding and is the total mass of the SNR ejecta. We now recall that the sweep-up time is given in terms of the SN parameters and the ambient number density, $n_{0} \approx \rho_{0}/m$, as $$\tSW = \frac{n_{0}^{-1/3}}{v_{\mathrm{ej}}} \left(\frac{3}{4\pi}\frac{M_{\mathrm{ej}}}{m}\right)^{1/3},$$ where $v_{\mathrm{ej}}\approx (2\ESN/\Mej)^{1/2}$ is the velocity of the ejecta, or numerically : $$\tSW = (1.4\,10^{3}\,\mathrm{yr}) \left(\frac{M_{\mathrm{ej}}}{10\Msun}\right)^{\hspace{-2pt}\frac{5}{6}} \hspace{-4pt} \left(\frac{E_{\mathrm{SN}}}{10^{51}\mathrm{erg}}\right)^{\hspace{-2pt}-\frac{1}{2}} \hspace{-4pt} \left(\frac{n_{0}}{1\mathrm{cm}^{-3}}\right)^{\hspace{-2pt}-\frac{1}{3}}. \label{SweepUpTime}$$ Replacing in Eq. (\[SweepUpTime\]) and using canonical values of $\ESN = 10^{51}\,\rm erg$ and $\Mej = 10\,\Msun$, we finally get : $$\phi_{1} \simeq 4\times10^{-10}\,\theta_{1} \left(\frac{n_{0}}{1\,\mathrm{cm}^{-3}}\right)^{2/3} \left[1 - \left(\frac{\tSW}{\tend}\right)^{1/5}\right]. \label{phi1Num}$$ Clearly this falls short of the value of order $10^{-8}$ required to explain the observations, even for values of $\theta_{1}$ as high as $0.5$. It might seem from Eq. (\[phi1Num\]) that very high ambient densities could help to make the spallation yields closer to the needed value. This is however not the case. First, the above estimate does not take energy losses into account, while both ionisation and adiabatic losses act to lower the genuine production rates. Second, and more significantly, the ratio $\tSW/\tend$ (and a fortiori its fifth root) becomes very close to 1 in dense environments, lowering $\phi_{1}$ quite notably (see Fig. \[BeYield\]). In fact, it turns out that there is no Sedov-like phase at all in media with densities of order $10^{4}\,\mathrm{cm}^{-3}$, the physical reason being that the radiative losses then act on a much shorter time-scale, eventually shorter than the sweep-up time. Evaluation of the second process (reverse shock) {#SecondProcess} ------------------------------------------------ Let us now turn to the second process, namely the spallation of energetic CNO nuclei accelerated at the reverse shock from the SN ejecta and interacting within the SNR with swept-up ambient material. We have argued above that this reverse shock acceleration is only plausible at times around and certainly the amount of energy transferred to CNO nuclei cannot be more than a fraction of . Let $\theta_2$ be the fraction of the explosion energy that goes into accelerating the ejecta at or around , and $\theta_{\mathrm{CNO}}$ the fraction of that energy that is indeed transferred to CNO nuclei. These particles are then confined to the interior of the remnant where they undergo spallation reactions as well as adiabatic losses. Let us again assume that the spectrum is of the form $f(p)\propto p^{-4}$. Then the production rate of Be atoms per unit volume is approximately $$0.2 {n_{0} \sigma_0\over m c} {\EdCR\over 14} \label{ProdRate2}$$ where now refers to the accelerated CNO nuclei, the factor 14 comes from the mean number of nucleons per CNO nucleus and the factor 0.2, as before, from the $f(p)\propto p^{-4}$ spectral shape (assuming the same upper cut-off position, but this only enters logarithmically). Integrating over the remnant volume, we obtain the spallation rate at : $$\frac{\d\mathcal{N}_{\mathrm{Be}}}{\d t} \approx 0.2 \sigma_{0}c \frac{\theta_{\mathrm{CNO}}\theta_{2}E_{\mathrm{SN}}}{14 m c^{2}}. \label{ProdRate2bis}$$ Now the adiabatic losses need to be evaluated rather carefully. It is generally argued that they act so that the momentum of the particles scales as the inverse of the linear dimensions of the volume occupied. Accordingly, in the expanding spherical SNR the EPs should lose momentum at a rate $\dot{p}/p = - \dot{R}/R$, reminiscent, incidentally, of the way photons behave in the expanding universe. In our case, however, the situation is complicated by the fact that the EPs do not push directly against the ‘walls’ limiting the volume of confinement, which move at the expansion velocity, $V = \dot{R}$, but are reflected off the diffusion barrier consisting of magnetic waves and turbulence at rest with respect to the downstream flow, and thus expanding at velocity $\frac{3}{4}\dot{R}$. To see how this influences the actual adiabatic loss rate, it is safer to go back to basic physical laws. Adiabatic losses must arise because the EPs are more or less isotropised within the SNR and therefore participate to the pressure. Now this pressure, $P$, works positively while the remnant expands, implying an energy loss rate equal to the power contributed, given by : $$\frac{\d U}{\d t} = - \int\hspace{-4pt}\int_{\mathcal{S}}\vec{F}\cdot\vec{v} = - \int\hspace{-4pt}\int_{\mathcal{S}}P\d S\times\frac{3}{4}\dot{R} = - 3\pi R^{2}\dot{R}P, \label{KinEnTh}$$ where $U = \frac{4}{3}\pi R^{3}\epsilon$ is the total kinetic energy of the particles. Considering that $P = \frac{2}{3}\epsilon$ in the non-relativistic limit (NR) and $P = \frac{1}{3}\epsilon$ in the ultra-relativistic limit (UR), Eq. (\[KinEnTh\]) can be re-writen as : $$\begin{split} \frac{\d\epsilon}{\d t} = -\frac{9}{4}P\frac{\dot{R}}{R} &= - \frac{3}{2}\epsilon\frac{\dot{R}}{R}\quad(\mathrm{NR})\\ &= - \frac{3}{4}\epsilon\frac{\dot{R}}{R}\quad(\mathrm{UR}). \end{split}$$ Finally, dividing both sides by the space density of the EPs and noting that $E = p^{2}/2m$ in the NR limit, and $E = pc$ in the UR limit, we obtain the momentum loss rate for individual particles, valid in any velocity range : $$\frac{\dot{p}}{p} = -\frac{3}{4}\frac{\dot{R}}{R}. \label{AdiabLossRate}$$ From this one deduces that at the time when the remnant has expanded to radius $R$, only those particles whose [initial]{} momenta at were more than $(R/\RSW)^{3/4}\pth$ are still above the spallation threshold. For a $p^{-4}$ distribution function the integral number spectrum decreases as $p^{-1}$ and thus the number of accelerated nuclei still capable of spallation reactions decreases as $R^{-3/4}\propto t^{-3/10}$. For a softer accelerated spectrum the effect would be even stronger because there are proportionally fewer particles at high initial momenta. This being established, we can integrate Eq. (\[ProdRate2bis\]) over time, to obtain the total production of Be atoms : $$\mathcal{N}_{\mathrm{Be}} = 0.2 n \sigma_{0} c \frac{\theta_{\mathrm{CNO}}\theta_{2}\ESN}{14 mc^{2}} \int_{\tSW}^{\tend} \left(\frac{t}{\tSW}\right)^{-3/10}\,\d t \label{phi2Interm}$$ that is : $$\mathcal{N}_{\mathrm{Be}} = \frac{2}{7} n \sigma_{0} c \frac{\theta_{\mathrm{CNO}}\theta_{2}\ESN}{14 mc^{2}} \tSW \left[\left(\frac{\tend}{\tSW}\right)^{7/10} - 1\right].$$ Dividing by the total number of CNO nuclei in the ejecta, $N_{\mathrm{ej,CNO}} = (\theta_{\mathrm{CNO}}/14) N_{\mathrm{ej,tot}} \simeq (\theta_{\mathrm{CNO}} M_{\mathrm{ej}}/14 m)$, we get the final result : $$\phi_{2} = \sigma_{0}c \frac{\theta_{2}E_{\mathrm{SN}}}{m c^{2}}\frac{\rho_{0}}{M_{\mathrm{ej}}}\tSW \frac{2}{7}\left(\frac{\tend}{\tSW}\right)^{7/10} \left[1 - \left(\frac{\tSW}{\tend}\right)^{7/10}\right]. \label{phi2}$$ Note that we assumed that the mass fraction of CNO in the ejecta is the same as the energy fraction of CNO in the EPs (which was the original meaning of $\theta_{\mathrm{CNO}}$). Considering that all nuclear species have the same spectrum in MeV/n, and thus a total energy proportional to their mass number, this simply means that the acceleration process is not chemically selective, in the sense that the composition of the EPs is just the same as that of the material passing through the shock. Numerically, again with $\ESN=10^{51}\,\rm erg$ and $\Mej = 10\,\Msun$, we finally obtain : $$\begin{split} \phi_2 \simeq 1\times 10^{-10}&\theta_{2} \left(n_0\over 1\,\rm cm^{-3}\right)^{2/3}\\ &\times\left(\frac{\tend}{\tSW}\right)^{7/10} \left[1 - \left(\frac{\tSW}{\tend}\right)^{7/10}\right]. \label{phi2Num} \end{split}$$ Relative contribution of the two processes {#RelativeContribution} ------------------------------------------ It is worth emphasizing the similarity between expressions (\[phi1\]) and (\[phi2\]) that we obtained for the spallation rates per CNO nuclei by the two processes considered here. This formal analogy allows us to write down their relative contributions straightforwardly : $$\frac{\phi_{2}}{\phi_{1}} = \frac{\theta_{2}}{\theta_{1}} \times \frac{2}{7}\left(\tend/\tSW\right)^{7/10} \frac{\left[1 - \left(\tSW/\tend\right)^{7/10}\right]} {\left[1 - \left(\tSW/\tend\right)^{1/5}\right]}. \label{phi2/phi1}$$ As is often the case, this similarity is not fortuitous and has a physical meaning. The two processes may indeed be regarded as ‘dual’ processes, the first consisting of the irradiation of the SN ejecta by the ambient medium, and the second of the ambient medium by the SN ejecta. The ‘symmetry’ is only broken by the dynamical aspect of the processes. First, of course, the energy imparted to the EPs in both cases needs not be the same, for it depends on the acceleration efficiency as well as the total energy of the shock involved (forward or reverse). This is expressed by the expected ratio $\theta_{2}/\theta_{1}$. And secondly, in the first process one has to fight against the dilution of the ejecta – integration of $(t/\tSW)^{-6/5}$, see Eq. (\[phi1Interm\]) – while in the second process one fights against the adiabatic losses – integration of $(t/\tSW)^{-3/10}$, see Eq. (\[phi2Interm\]). This is expressed by the last factor in Eq. (\[phi2/phi1\]). Clearly the latter decrease of the production rates is the least dramatic, and the reverse shock process must dominate the LiBeB production in supernova remnants. However, this conclusion still depends on the genuine efficiency of reverse shock acceleration, and once the relative acceleration efficiency $\theta_{2}/\theta_{1}$ is given, the weight of the first process relative to the second still depends on the total duration of the Sedov-like phase, appearing numerically in Eq. (\[phi2/phi1\]) through the ratio /, which in turn depends on the ambient density, $n_{0}$. The expression of as a function of the parameters has been given in Eq. (\[SweepUpTime\]), so we are left with the evaluation of the time, , when the magnetic turbulence collapses and the EPs leave the SNR, putting an end to Be production. We argued above that should correspond to the end the Sedov-like phase, when the shock induced by the SN explosion becomes radiative, that is when the cooling time of the post-shock gas becomes of the same order as the dynamical time. In principle, the cooling rate can be derived from the so-called cooling function, $\Lambda(T) (\mathrm{erg~cm}^{3}\mathrm{s}^{-1}$), which depends on the physical properties of the post-shock material, notably on its temperature, $T$, and metallicity, $Z$ : $$\tau_{\mathrm{cool}}\approx\frac{\frac{3}{2}k_{\mathrm{B}}T}{n\Lambda(T)}, \label{tauCool}$$ where $n$ is the post-shock density, equal to $4n_{0}$ if the compression ratio is that of an ideal strong shock (nonlinear effects probably act to increase the compression ratio to values larger than 4). As for the dynamical time, we simply write $$\tau_{\mathrm{dyn}} \approx \frac{\dot{R}}{R} \approx \frac{5}{2}\,t. \label{tauDyn}$$ To obtain , we then need to solve the following equation in the variable $t$, obtained by equating $\tau_{\mathrm{cool}}$ and $\tau_{\mathrm{dyn}}$ given above : $$t \approx \frac{3k_{\mathrm{B}}T}{20n_{0}\Lambda(T)}, \label{tEndEq}$$ where it should be clear that the right hand side also depends on time through the temperature, $T$, and thus indirectly through the cooling function too. In the non-radiative SNR expansion phase, the function $T(t)$ is obtained directly from the hydrodynamical jump conditions at the shock discontinuity : $$T \approx \frac{3m}{8k_{\mathrm{B}}}V^{2},$$ or numerically : $$T \approx (2\times 10^{5}\,\mathrm{K}) \left(\frac{E_{\mathrm{SN}}}{10^{51}\mathrm{erg}}\right)^{\hspace{-2pt}\frac{2}{5}} \hspace{-4pt} \left(\frac{n_{0}}{1\mathrm{cm}^{-3}}\right)^{\hspace{-2pt}-\frac{2}{5}} \hspace{-4pt} \left(\frac{t}{10^{5}yr}\right)^{\hspace{-2pt}-\frac{6}{5}}. \label{Temperature}$$ To solve Eq. (\[tEndEq\]), we still need to know the cooling function $\Lambda(T)$. In the range of temperatures corresponding to the end of the Sedov-like phase, $10^{5}\,\mathrm{K}\la T\la 10^{7}\,\mathrm{K}$, it happens to depend significantly on metallicity, with differences up to two orders of magnitude for metallicities from $Z = 0$ to $Z = 2 Z_{\odot}$ (Böhringer & Hensler 1989). Because we focus on Be production in the early Galaxy, we adopt the cooling function corresponding to zero metallicity, represented in Fig. \[CoolingFunction\] (adapted from Böhringer & Hensler 1989), which holds for values of $Z$ up to $\sim 10^{-2}Z_{\odot}$. For high enough ambient densities, the shock will become radiative early in the SNR evolution, when the temperature is still very high, say above $T\ga 2\,10^{6}$ K. In this case, the cooling function is dominated by Bremsstrahlung emission and can be written analytically as : $$\Lambda_{\mathrm{Br}}(T) \approx (2.4\,10^{-23}\,\mathrm{erg~cm}^{3}\mathrm{s}^{-1}) \left(\frac{T}{10^{8}\,\mathrm{K}}\right)^{1/2}. \label{LambdaBr}$$ Substituting from (\[Temperature\]) and (\[LambdaBr\]) in Eq. (\[tEndEq\]) and solving for $t$, we find : $$\tend = (1.1\,10^{5}\,\mathrm{yr}) \left(\frac{E_{\mathrm{SN}}}{10^{51}\mathrm{erg}}\right)^{1/8} \left(\frac{n_{0}}{1\mathrm{cm}^{-3}}\right)^{-3/4}. \label{tEnd}$$ To check the consistency of our assumption $\Lambda \approx \Lambda_{\mathrm{Br}}$ (i.e. $T\ga 2\,10^{6}$ K), let us now report Eq. (\[tEnd\]) in (\[Temperature\]) and write down the temperature $T_{\mathrm{End}}$ at the end of the Sedov-like phase : $$T_{\mathrm{End}} \approx (2\times 10^{5}\,\mathrm{K}) \left(\frac{E_{\mathrm{SN}}}{10^{51}\mathrm{erg}}\right)^{1/4} \left(\frac{n_{0}}{1\mathrm{cm}^{-3}}\right)^{1/2},$$ which means that the above analytical treatment is valid only for ambient densities greater than about $100\,\mathrm{cm}^{-3}$. For lower densities, we must solve Eq. (\[tEndEq\]) graphically. First, we invert Eq. (\[Temperature\]) to express $t$ as a function of temperature, then we plot the function $f(T)\equiv 3k_{\mathrm{B}}T/(20n_{0}t)$ on the same graph as $\Lambda$ (see Fig. \[CoolingFunction\] for an example), find the value of $T$ at intersection, and finally convert this value into the sought time making use again of Eq. (\[Temperature\]). The results, showing as a function of the ambient density, are shown in Fig. \[TimeScales\]. We now have all the ingredients to plot the efficiency ratio of the two processes calculated above. Figure \[Phi2/Phi1\] shows the ratio $\phi_{2}/\phi_{1}$ given in Eq. (\[phi2/phi1\]) as a function of the ambient density, assuming that $\theta_{1} = \theta_{2}$. Two different values of the ejected mass have been used, corresponding to different progenitor masses ($\sim 10-40\,\Msun$). It can be seen that low densities are more favourable to the reverse shock acceleration process. This is due to $\tend/\tSW$ being larger, implying a larger dilution of the ejecta (process 1 less efficient) and smaller adiabatic losses, which indeed decrease as $t^{-1}$ (process 2 more efficient). The part of the plot corresponding to $\phi_{2}/\phi_{1}\le 1$ is not physical, because it requires $\tend\le\tSW$, which simply means that the Sedov-like phase no longer exists and the whole calculation becomes groundless. Note however that in Fig. \[Phi2/Phi1\] the energy imparted to the EPs has been assumed equal for both processes, which is most certainly not the case. Actually, if $\theta_{2}/\theta_{1} = 0.1$ (e.g. $\theta_{1} = 10\%$ and $\theta_{1} = 1\%$), then process 1 is found to dominate Be production during the Sedov-like phase, regardless of the ambient density. Spallation reactions after the Sedov-like phase =============================================== At the end of the Sedov-like phase, the EPs are no longer confined and leave the SNR to diffuse across the Galaxy. At the stage of chemical evolution we are considering here, there are no or few metals in the interstellar medium (ISM), so that energetic protons and $\alpha$ particles accelerated at the forward shock will not produce any significant amount of Be after (although Li production will still be going on through $\alpha + \alpha$ reactions). In the case of the second process, however, the EPs contain CNO nuclei which just cannot avoid being spalled while interacting with the ambient H and He nuclei at rest in the Galaxy. This may be regarded as a third process for Be production, which lasts until either the EPs are slowed down by Coulombian interactions to subnuclear energies (i.e. below the spallation thresholds) or they simply diffuse out of the Galaxy. Since the confinement time of cosmic rays in the early Galaxy is virtually unknown, we shall assume here that the Galaxy acts as a thick target for the EPs leaving the SNR, an assumption which actually provides us with an upper limit on the spallation yields. Unlike the first two processes evaluated above, this third process is essentially independent of dynamics. Thus, time-dependent calculations are no longer needed and, from this stage on, the calculations made by Ramaty et al. (1997) or any steady-state calculation is perfectly valid. In particular, the ambient density has no influence on light element production, since a greater number of reactions per second, as would result from a greater density, implies an equal increase of both the spallation rates and the energy loss rate. Once integrated over time, both effects cancel out exactly, and in fact, given the energy spectrum of the EPs, the efficiency of Be production (and Li, and B), expressed as the number of nuclei produced per erg injected in the form of EPs, depends only on their chemical composition. Results are shown in Fig. \[Be/erg\] for different values of the source abondance ratios, $\mathrm{H}/\mathrm{He}$ and $(\mathrm{H}+\mathrm{He})/(\mathrm{C}+\mathrm{O})$, allowing one to derive the spallation efficiency for any composition. Two-steps processes (such as $^{12}\mathrm{C} + ^{1}\mathrm{H} \longrightarrow ^{10}\mathrm{B}$ followed by $^{10}\mathrm{B} + ^{1}\mathrm{H} \longrightarrow ^{9}\mathrm{Be}$) have been taken into account. Test runs show good agreement with the results of Ramaty et al. (1997). As can be seen on Fig. \[Be/erg\], pure Carbon and Oxygen have a production efficiency of about $0.22\,\mathrm{nuclei}/\mathrm{erg}$, while this efficiency decreases by at least a factor of 10 for compositions with hundred times more H and He than metals (or about ten times more by mass). According to models of explosions for SN with low metallicity progenitors, the average $(\mathrm{H}+\mathrm{He})/(\mathrm{C}+\mathrm{O})$ ratio among the EPs should indeed be expected to be $\ga 200$, unless selective acceleration occurs to enhance the abundance of the metals. As a consequence, efficiencies greater than $\sim 10^{-2}\,\mathrm{nuclei}/\mathrm{erg}$ should not be expected, so that a production of $\sim 4\,10^{48}$ atoms of Be requires an energy of $\sim 4\,10^{50}$ erg to be imparted to the EPs. This seems very unlikely considering that the total energy available in the reverse shock (the source of the EPs) should be of order one tenth of the SN explosion energy, not to mention the acceleration efficiency. Moreover, a significant fraction of the energy originally imparted to the EPs has been lost during the Sedov-like phase of the SNR evolution through adiabatic losses. To evaluate the ‘surviving’ fraction of energy, it sufficies to go back to Eq. (\[AdiabLossRate\]), which indicates that when the radius of the shock is multiplied by a factor $\eta$, the momentum $p$ of all the particles is multiplied by a factor $\eta^{-3/4}$. It is worthwhile noting that, because of their specific momentum dependence, adiabatic losses do not modify the shape of the EP energy spectrum. In our case, $f(p)\propto p^{-4}$, so that when all momenta $p$ are divided by a factor $\zeta$, the distribution function $f(p)$ is divided by the same factor $\zeta$. To see that, the easiest way is to work out the number of particules between momenta $p$ and $p + \d p$ after the momentum scaling. This number writes $\d N^{\prime} = f^{\prime}(p)4\pi p^{2}\d p$, where $f^{\prime}(p)$ is the new distribution function. Now $\d N^{\prime}$ must be equal to the number of particles that had momentum between $\zeta p$ and $\zeta(p + \d p)$, which is, by definition, $\d N = f(\zeta p)4\pi(\zeta p)^{2}\zeta\d p$. Equating $\d N$ and $\d N^{\prime}$ yields the result $f^{\prime}(p) = \zeta^{-1}f(p)$. Putting all pieces together, we find that when the shock radius $R$ is multiplied by a factor $\eta$, the distribution function and, thus, the total energy of the EPs are multiplied by $\eta^{-3/4}$. Now considering that $R$ increases as $t^{2/5}$ during the Sedov-like phase, we find that the total energy of the EPs decreases as $t^{-3/10}$. Note that this is nothing but an other way to work out the decrease of the spallation rates for our second process during the Sedov-like phase (cf. Sect. \[SecondProcess\]). Finally, we find that a fraction $(\tend/\tSW)^{-3/10}$ of the initial energy imparted to the EPs is still available for spallation at the end of the Sedov-like phase. This factor is plotted on Fig. \[EnergyReduction\], as a function of the ambient density. It can be seen that for $n_{0} = 1\,\mathrm{cm}^{-3}$, the energy available to power our third process of light element production has been reduced by adiabatic losses to not more than one third of its initial value, and less than one half for densities up to $100\,\mathrm{cm}^{-3}$. Clearly, high densities are favoured (energetically) because they tend to shorten the Sedov-like phase, and therefore merely avoid the adiabatic losses. Discussion ========== Since light element production in the interstellar medium obviously requires a lot of energy in the form of supernuclear particles (i.e. with energies above the nuclear thresholds) as well as metals (especially C and O), it is quite natural to consider SNe as possible sources of the LiBeB observed in halo stars. We have analysed in detail the spallation nucleosynthesis induced by a SN explosion on the basis of known physics and theoretical results relating to particle shock acceleration. Two major processes can be identified, depending on whether the ISM or the ejecta are accelerated, respectively at the forward and reverse shocks. In the first case, the EPs consist mostly of protons and alpha particles and must therefore interact with C and O nuclei, which are much more numerous within the SNR than in the surrounding medium (especially at early stages of Galactic evolution). The process will thus last as long as the EPs stay confined in the SNR, i.e. approximately during the Sedov-like phase, but not more. In the second case, freshly synthesized CNO nuclei are accelerated, and Be production occurs through interaction with ambient H and He nuclei. The process is then divided into two, one stretching over the Sedov-like phase, with the particles suffering adiabatic losses, and the other one occuring outside the remnant, with only Coulombian losses playing a role. We have calculated the total Be production in these three processes, taking the dynamics of the SNR evolution into account (dilution of the ejecta by metal-poor material and adiabatic losses). The results are shown in Fig. \[BeYield\] for processes 1 and 2 (from Eqs. (\[phi1Num\]) and (\[phi2Num\])). We find that with canonical values of $\theta_{1} = 0.1$, $\theta_{2} = 0.01$, $M_{\mathrm{ej}} = 10\,\Msun$ and a mean ambient density $n_{0} = 10\,\mathrm{cm}^{-3}$, the fraction of freshly synthesized CNO nuclei spalled into Be in these processes is $\phi_{1} \sim 8\,10^{-11}$ and $\phi_{2} \sim 3\,10^{-11}$, respectively, which is very much less than the value ‘required’ by the observations, discussed in the introduction ($\phi_{\mathrm{obs}}\sim 3\,10^{-8}$). Even allowing for unreasonably high values of the acceleration efficiency, $\theta_{1}\sim \theta_{2}\la 1$, the total Be production by processes 1 and 2 would still be more than one order of magnitude below the observed value. As suggested by Fig. \[BeYield\] and our analytical study, higher densities improve the situation. However, even with $n_{0} = 10^{3}\,\mathrm{cm}^{-3}$ and acceleration efficiencies equal to 1, the Be yield is still unsufficient. Moreover, it should be noted that our calculations did not consider Coulombian energy losses (because they are negligeable as compared to adiabatic losses for usual densities), which become important as the density increases and therefore make the Be yield smaller. Finally, since we are trying to account for the mean abundance of Be in halo stars, as compared to Fe, we have to evaluate the Be production for an ambient density corresponding to the mean density encountered around explosion sites in the early Galaxy, which is very unlikely to be as high as $10^{3}\,\mathrm{cm}^{-3}$. It could even be argued that although the gas density might have been higher in the past than it is now (hence our ‘canonical value’ $n_{0} = 10\,\mathrm{cm}^{-3}$), the actual mean density about SN explosion sites could be lower than $1\,\mathrm{cm}^{-3}$, because most SNe may explode within superbubble interiors, where the density is much less than in the mean ISM. Thus, our conclusion is that processes 1 and 2 both fail in accounting for the Be observed in metal-poor stars in the halo of our Galaxy. Concerning the third process, adopting canonical values for the parameters again leads to unsufficient Be production, as noted in the previous section. While higher densities improve the situation by avoiding the adiabatic energy losses, one should nevertheless expect at least half of the EP energy to be lost in this way, for any reasonable density. This means that even if 10% of the explosion energy is imparted to EPs accelerated at the reverse shock, which is certainly a generous upper limit, the required number of $\sim 4\,10^{48}$ nuclei of Be per SN implies a spallation efficiency of $\sim 0.1$ nucleus/erg. Now Fig. \[Be/erg\] shows that this requires an EP composition in which at least one particle out of ten is a CNO nucleus. In other words, the ejected mass of CNO must be of the order of that of H and He together. None of the SN explosion models published so far can reproduce such a requirement, and so there is clearly a problem with Be production in the early Galaxy. The results presented here are in fact interesting in many regards. First, they show that it is definitely very difficult to account for the amount of Be found in halo stars. Consequently, we feel that the main problem to be addressed in this field of research is probably not the chemical evolution of Be (and Li and B) in the Galaxy, as given by the ensemble of the data points in the abundance vs metallicity diagrams (e.g. whether Be is proportional to Fe or to its square) but, to begin with, the position of any of these points. Are we able to describe in some detail one process which could explain the amount of Be (relative to Fe) present in any of the stars in which it is observed? The answer, we are afraid, seems to be no at this stage. It is however instructive to ask why the processes investigated here have failed. Concerning process 1 (acceleration of ISM, interaction with fresh CNO within the SNR), the main reason is that the CNO rich ejecta are ‘too much diluted’ by the swept-up material as the SNR expands, so that the spallation efficiency is too low (or the available energy is too small). However, it seems rather hard to think of any region in the Galaxy where the concentration in CNO is higher than inside a SNR during the Sedov-like phase (especially in the first stages of chemical evolution)! So the conclusion that process 1 cannot work, even with a 100% acceleration efficiency, seems to rule out any other process based on the acceleration of the ISM, initially devoided of metals. The other solution is then of course to accelerate CNO nuclei themselves, which provides the maximum possible spallation efficiency, independently of the ambient metallicity. Every energetic CNO will lead to the production of as much Be as possible given the spallation cross sections and the energy loss rates. The latter cannot physically be smaller than the Coulombian loss rate in a neutral medium, and this leads to the efficiency plotted in Fig. \[Be/erg\]. Unfortunately, a significant amount of the CNO rich ejecta of an isolated SN can only be accelerated at the reverse shock at a time around the sweep-up time, . This means that i) the total amount of energy available is smaller than the explosion energy (probably of order 10%, i.e. $\sim 10^{50}$ erg), and ii) the accelerated nuclei will suffer adiabatic losses during the Sedov-like phase, reducing their energy by a factor of 2 or 3. As shown above, this makes process 2-3 incapable of producing enough Be, as long as the EPs have a composition reflecting that of the SN ejecta. This suggest that a solution to the problem could be that the reverse shock accelerates preferentially CNO nuclei rather than H and He. For example, recent calculations have shown that such a selective acceleration arises naturally if the metals are mostly condensed in grains (Ellison et al. 1997). The proposition by Ramaty et al. (1997) that grains condense in the ejecta before being accelerated could then help to increase the abundance of CNO in the EPs. However, we have to keep in mind that any selective process called upon must be very efficient indeed, since as we indicated above, the data require that the EP composition be as rich as one CNO nuclei out of ten EPs, which means that CNO nuclei must be accelerated at least ten times more efficiently than H and He. This would have to be increased by another factor of ten if the energy initially imparted to the EPs by the acceleration process were only a factor 2 or 3 lower (i.e. $\sim 3\,10^{49}$ erg, which is more reasonable from the point of view of particle acceleration theory). Clearly, more work is needed in this field before one can safely invoke a solution in terms of selective acceleration. As can be seen, playing with the composition to increase the spallation efficiency has its own limits, and in any case, Fig. \[Be/erg\] gives an unescapable upper limit, obtained with pure Carbon and Oxygen (at least for the canonical spectrum considered here - other spectra were also investigated, as in Ramaty et al. (1997), leaving the main conclusions unchanged). This would then suggest that another source of energy should be sought. However, the constraint that it should be more energetic than SNe is rather strong. Another interesting line of investigations could be the study of the collective effects of SNe. Most of the massive stars and SN progenitors are believed to be born (and indeed observed, Melnik & Efremov 1995) in associations, and their joint explosions lead to the formation of superbubbles which may provide a very favourable environment for particle acceleration (Bykov & Fleishman 1992). Parizot et al. (1998) have proposed that these superbubbles could be the source of most of the CNO-rich EPs, and Parizot & Knoedlseder (1998) further investigated the gamma-ray lines induced by such an energetic component. The most interesting features of a scenario in which Be-producing EPs are accelerated in superbubbles is that i) when a new SN explodes, the CNO nuclei ejected by the previous SNe are accelerated at the forward shock, instead of the reverse shock in the case of an isolated SN, which implies a greater energy, and ii) no significant adiabatic losses occur, because of the dimensions and low expansion velocity of the superbubble. This makes the superbubble scenario very appealing, and it will be investigated in detail in a forthcoming paper. However that may be, we should also keep in mind that when we say that a process does not produce enough Be, it always means that it does not produce enough Be as compared to Fe. Now it could also be that SN explosion models actually produce too much Fe. The point is that Be is compared to Fe in the observations, while it has no direct physical link with it. Indeed, Be is not made out of Fe, but of C and O. So to be really conclusive, the studies of spallative nucleosynthesis should compare theoretical Be/O yields to the corresponding abundance ratio in metal-poor stars. Unfortunately, the data are much more patchy for Be as a function of \[O/H\] than as a function of \[Fe/H\], especially in very low metallicity stars. The usually assumed proportionality between O and Fe could turn out to be only approximate, as recent observational works possibly indicate (Israelian et al. 1998; Boesgaard et al. 1998; these observations, however, still ?? need to be confirmed by an independent method, all the more that they come into conflict with several theoretical and observational results; cf. Vangioni-Flam et al. 1998b). We shall address this question in greater detail in the attending paper (Parizot and Drury, 1999, Paper II). Finally, we wish to stress that the calculations presented in this paper rely on a careful account of the dynamics of the problem. More generally, time-dependent calculations are required to properly evaluate the spallation processes in environments where compositions and energy densities are evolving. In particular, as argued in Parizot (1998), no variation with density can be obtained with a stationary model, since an increase in the density induces an equivalent and cancelling increase in the spallation rates and the energy loss rates. By contrast, we have shown that all three of the processes considered here are more efficient at higher density – a result which could not have been found otherwise. Detailed, numerical time-dependent calculations will be presented in paper II, with conclusions similar to those demonstrated here. This work was supported by the TMR programme of the European Union under contract FMRX-CT98-0168. It was initiated during a visit by LD to the Service d’Astrophysique, CEA Saclay, whose hospitality is gratefully acknowledged. We wish to thank M. Cassé and Elisabeth Vangioni-Flam for stimulating discussions of these and related topics. Abia C., Isern J., Canal R., 1993, A&A, 275, 96 Achterberg A., Blandford R. D., Reynolds S. P., 1994, A&A, 281, 220 Boesgaard A. M., King J. R., Deliyannis C. P., Vogt S. S., 1998, submitted Böhringer H., Hensler G., 1989, A&A, 215, 147 Bykov A. M., Fleishman G. D., 1992, MNRAS, 255, 269 Cassé M., Lehoucq R., Vangioni-Flam E., 1995, Nat, 374, 337 Casuso E., Beckman J. E., 1997, ApJ, 475, 155 Drury L. O’C., Aharonian F. A., Völk H. J., 1994, A&A, 287, 959 Ellison D., Drury L. O’C., Meyer J.-P., 1997, ApJ, 487, 197 Edvardsson B., Gustafsson B., Johansson S. G., et al., 1994, A&A, 290, 176 Feltzing S., Gustafsson B., 1994, ApJ, 423, 68 Fields B. D., Olive K. A., Schramm D. N., 1995, ApJ, 435, 185 Fields B. D., Olive K. A., Schramm D. N., 1995, ApJ, 439, 854 Gilmore G., Gustafsson B., Edvardsson B., Nissen P. E., 1992, Nat, 357, 379 Israelian G., García-López R. J., Rebolo R., 1998, submitted to ApJ Jun B.-I., Norman M. L., 1996, ApJ, 472, 245 Kiselman D., Carlsson M., 1996, A&A, 311, 680 Melnik A. M., Efremov Yu. N., 1995, Astron. Lett., 21, 10 Molaro P., Bonifacio P., Castelli F., Pasquini L., 1997, A&A, 319, 593 Parizot E. M. G., 1998, submitted to A&A Parizot E., Drury L., 1999, in preparation (Paper II) Parizot E., Knödlseder J., 1998, in: The transparent Universe, Proceedings of the 3rd INTEGRAL Workshop (Taormina) Parizot E., Cassé M., Vangioni-Flam E., 1998, A&A, 328, 107 Ramaty R., Kozlovsky B., Lingenfelter R. E., 1996, ApJ, 456, 525 Ramaty R., Kozlovsky B., Lingenfelter R. E., Reeves H., 1996, ApJ, 488, 730 Reeves H., 1994, Review of Modern Physics, 66, 193 Ryan S., Norris I., Bessel M., Deliyannis C., 1994, ApJ, 388, 184 Vangioni-Flam E., Cassé M., Fields B. D., Olive K. A., , 1996, ApJ, 468, 199 Vangioni-Flam E., Ramaty R., Olive K. A., Cassé M., 1998, A&A, 337, 714 Vangioni-Flam E., Cassé M., Cayrel R., et al., 1998b, submitted to ApJ	{ "pile_set_name": "ArXiv" }
--- abstract: 'The [Gaia]{} H-R diagram shows the presence of apparently young stars at high tangential velocities. Using a simple analytical model, I show that these stars are likely to be blue stragglers. Once normalized to red giant stars, the fraction of nearby halo blue stragglers is of order 20 percent, and remarkably close to that measured in dwarf galaxies. Motivated by this similarity, I apply to field blue stragglers scaling relations inferred from blue stragglers in dwarf galaxies. Doing this for the Milky Way halo returns an average stellar density of $(3.4 \pm 0.7) \times 10^{-5} M_{\odot}/\rm{pc}^3$ and a dark matter density of $\simeq 0.006^{+0.005}_{-0.003}\,M_{\odot}/\rm{pc}^3 \simeq 0.22^{+0.20}_{-0.10}\,\rm{GeV}/\rm{cm}^3$ within 2 kpc from the Sun. These values compare favorably to other determinations available in the literature but are based on an independent set of assumptions. A few considerations of this methodology are discussed, most notably that the correlation between the dark matter halo core density and stellar mass seen in dwarf galaxies seems to hold also for the nearby Milky Way halo.' author: - Luca Casagrande bibliography: - 'luca\_v2.bib' title: Connecting the local stellar halo and its dark matter density to dwarf galaxies via blue stragglers --- Introduction {#sec:intro} ============ Recent results have shown that the close binary fraction ($P\lesssim 10^{4}$ days and $a\lesssim 10$ AU) of solar-type stars is anticorrelated with metallicity [@mo; @elb], implying that most solar-type stars with \[Fe/H\]$<-1$ will interact with a close binary. While the implications of this results are manyfold, here I focus on the rate of blue-straggler stars (BSS). First observed by [@sandage53] in the globular cluster M3 as an apparent extension of the classical main sequence, BSS are now believed to be the product of mass transfer and/or merger in close binaries or multiple-star systems [e.g. @knigge; @santana]. The metallicity anticorrelation of close binaries implies that the fraction of BSS is expected to increase at decreasing metallicity [@wyse20]. This means that the blue stragglers population should become more prominent when moving to increasingly metal-poor and old populations (so that BSS can be readily identified populating the left-hand side of the turnoff), such as moving from the thin to the thick disk and halo. Here I use data from [Gaia]{} DR2 [@gaiaDR2] and simple analytical considerations to show that this identification is indeed possible. The fraction of BSS found at high tangential velocities is in excellent agreement with that measured in dwarf spheroidal and ultrafaint dwarf galaxies. For these stellar systems, the fraction of BSS is proportional to their total stellar mass [@santana], which then correlates to their dark matter halo core density. Here I show that by applying these relations to the BSS identified in the Galactic stellar halo, it is possible to obtain measurements of the local stellar and dark matter density that are in agreement with those derived by other means. This result supports the assumption of applying to the halo of a spiral galaxy like the Milky Way scaling relations for BSS in dwarf galaxies. Thus, field BSS might be able to provide a new diagnostic to study the stellar and dark matter density in the halo. A number of methodologies have been developed over the years to measure these two densities, which are central, e.g., to guide direct dark matter detection experiments, to understand the formation of the Milky Way and to place it in the cosmological context with other, similar-mass galaxies [e.g., @read14; @deason; @desalas]. However, these measurements are far from definitive. For the stellar density in the halo, a wide range of local normalizations have been reported in the literature. Local dynamical dark matter density measurements are strongly affected by the imperfect knowledge of the baryonic contribution, and in spite of the data from Gaia, different analyses still return dissimilar results [e.g., @s18; @desalas]. The use of blue stragglers to derive stellar and dark matter densities is explored here by developing a methodology that is based on assumptions largely different from those used by other methods. Blue Straggler selection {#sec:data} ======================== From [Gaia]{} DR2 I retrieve all stars satisfying conditions (1), (2) and (3) of [@arenou], and with parallax errors below 10 percent, totaling $67.7$ million objects. These requirements are also used in [@gaiaHR] to study the fine structure of the Hertzsprung-Russell diagram, removing most of the artifacts while still allowing us to see the imprint of genuine binaries. Avoiding selection against binaries might be relevant given that [@ps00] concluded that a significant fraction of BSS are binaries. For this same reason I do not impose a threshold on the Renormalized Unit Weight Error (RUWE, technical note GAIA-C3-TN-LU-LL-124-01), and note that the present sample has a median RUWE of $1.0$, with 97% of the stars having RUWE$<1.4$. BSS are identified in a fashion similar to [@santana] and normalized to the number of red giant branch (RGB) stars identified in a similar range of absolute magnitudes. The advantage of using the number count of stars selected in a similar range of intrinsic luminosities is that their ratio is largely insensitive to selection effects stemming from the Malmquist bias [@malmquist]. BSS and RGB stars are defined as stars falling within the blue and red boxes shown in the color-magnitude diagram of Figure \[fig:video\]. These boxes are obtained following [@Gaia_evans] to convert $gr$ magnitudes from [@santana] into the [Gaia]{} photometric system, plus small zero-points shifts to optimize these boxes with the actual position occupied by BSS and RGB stars in the [Gaia]{} H-R diagram. For the RGB box, the boundary at cool temperatures is extended to colors redder than in [@santana], to account for a range of metallicities and ages in the Galactic disk that is much larger than that encountered in dwarf galaxies or globular clusters. This extension is, however, irrelevant when dealing with stars at high tangential velocities, which occupy the leftmost position on the RGB box (see also animation associated with Figure \[fig:video\]). [Gaia]{} colors and magnitudes of all stars have been corrected for reddening using a rescaled version of the [@sfd] map as described in [@kunder], with reddening coefficients from [@cv18]. [animation]{}[video.mp4]{} The video associated with Figure \[fig:video\] shows how the ratio of the number of stars falling into the BSS and RGB boxes varies when selecting stars with different tangential velocities. As $V_{\rm{TAN}}$ increases, the dominant stellar population changes from the thin to the thick disk, until the ratio remains constant at a value of $0.2-0.3$ for tangential velocities that are typical of halo stars. The box used to identify BSS suffers from contamination from main-sequence stars populating this region of the H-R diagram. This effect is very strong in a young stellar population like the thin disk [say, $V_{\rm{TAN}}<40{\rm{\,km\,s}^{-1}}$, see, e.g., @gaiaHR] and decreases when moving toward older populations like the thick disk and halo. Figure \[fig:video\] is obtained using all stars without any restriction on their Galactic latitude $b$, or height above the Galactic plane $Z$. Another source of uncertainty is due to the arbitrary definition of the BSS and RGB boxes, as well as to the reddening corrections applied. To account for all these uncertainties, I run 10,000 Monte Carlo realizations where each time I randomly changed the boundaries of the BSS and RGB boxes by up to several hundreds of mag, reddening by 20 percent, let the width of the boxcar vary anywhere between $10$ and $60{\rm{\,km\,s}^{-1}}$, and considered only stars with heights above the Galactic plane varying in the range $0<Z<400$ pc. The results are shown in the left panel of Figure \[fig:mcanal\]. At low $V_{\rm{TAN}}$ the ratio of stars falling into the BSS and RGB boxes varies quite substantially, and this is largely driven by the adopted cuts in $Z$. The higher stars are above the Galactic plane, the lower is the contamination from young thin disk stars that otherwise would fall into the BSS box. Therefore, the ratio of stars into the BSS and RGB boxes decreases. I verified that very similar results are obtained if doing a cut in the projected height above the plane, i.e. Galactic latitude $b$, instead of $Z$. Remarkably, the trend reverses for tangential velocities around $200{\rm{\,km\,s}^{-1}}$ and the ratio becomes nearly constant above $\sim 300{\rm{\,km\,s}^{-1}}$, with very low scatter independently of the set of parameters of each Monte Carlo realization. Analytical model ================ The number of main-sequence stars that contaminate the BSS box relative to the RGB one can be modeled generating a synthetic stellar population, but this is beyond the scope of this analysis. Instead, simple analytical considerations suffice to understand the trend. The number of main-sequence stars that fall into the BSS box at any given time is given by $$N_{\rm{BSS}} = \int_{M_1}^{M_2} \phi(m) \textrm{d}m$$ where $\phi(m)$ is the Initial Mass Function (IMF), and $M_1$ and $M_2$ are minimum and maximum stellar mass in the BSS box. For stars above roughly a solar mass, the IMF slope of [@salpeter] can be safely used, i.e. $\phi(m)\propto m^{-2.35}$. Similarly, the number of stars that are on the RGB phase at a given time is given by all stars massive enough to have evolved off the main sequence, i.e. with masses higher than the time-dependent turnoff mass $M(t)$. The latter can be readily derived from the relationship between main-sequence lifetime and stellar mass [e.g. @kippe]: $$\label{eq:kippe} \frac{M(t)}{M_{\odot}}\sim\left( \frac{t}{t_{\odot}} \right)^{-1/3}.$$ This implies that at any given time the following ratio $R'$ holds between newly formed stars that fall in the BSS and RGB selection boxes: $$\label{eq:diff} R'=\frac{\displaystyle\int_{M_1}^{M_2} \phi(m) \textrm{d}m}{\displaystyle\int_{M(t)}^{\infty} \phi(m) \textrm{d}m} = \frac{M_1^{-1.35}-M_2^{-1.35}}{(t/t_{\odot})^{0.45}}.\vspace{0.1cm}$$ For practical purposes, the power law of the IMF implies that the upper limit of integration at the denominator goes quickly toward a negligible contribution, be it a few tens of solar masses or infinity. If one wishes instead to derive the ratio $R$ between all stars formed until a given time that fall onto the BSS and RGB boxes, it suffices to integrate over $t$, and to correct for the stars that have evolved off the selection boxes. For the BSS box, this implies removing all stars that have evolved off the main sequence, i.e. with masses in the range $M(t)$ to $M_2$. For the RGB box, one should correct for all stars that have left the giant branch. At a given age, the RGB spans a mass range that is of order of a few percent of the turnoff mass, $p \sim 1.1-1.03$. $$R=\frac{\displaystyle\int_0^t \left[ \displaystyle\int_{M_1}^{M_{2}}\phi(m)\textrm{d}m - \displaystyle\int_{M(t)}^{M_{2}}\phi(m)\textrm{d}m \right]\,\textrm{d}t }{\displaystyle\int_0^t \displaystyle\int_{M(t)}^{p\,M(t)} \phi(m) \textrm{d}m\,\textrm{d}t}$$ $$\label{eq:int} =\frac{1.45\,M_1^{-1.35}\,- (t/t_{\odot})^{0.45}}{(1-p^{-1.35})\,(t/t_{\odot})^{0.45}}.$$ Here the integration over time corresponds to assuming a constant star-formation history, which suffices to describe most of the evolution of the Galactic disk, at least over the past $\sim 8$ Gyr [e.g., @snaith] The dependence of equation \[eq:int\] with time is shown in the left panel of Figure \[fig:mcanal\], where I have used isochrones for an informed guess on $M_1$. I have adopted the MIST isochrones [@choi] with $-2.0\le \rm{[Fe/H]} \le 0.5$ [roughly the range of metallicities covered by stars with $V_{\rm{TAN}}<200{\rm{\,km\,s}^{-1}}$, e.g., @sahl] and with ages spanning over the entire grid of MIST isochrones. I have then identified all isochrone masses that fall in the BSS box of Figure \[fig:video\], obtaining the values $M_1=1.0$ and $M_2=1.8$ for the 10th and 90th percentiles, respectively. Stars with $V_{\rm{TAN}}>200{\rm{\,km\,s}^{-1}}$ typically have ages older than $\sim 10$ Gyr and belong to the kinematically hot tail of the thick disk and to the stellar halo [e.g. @helmi; @dimatteo; @sahl]. The simple model of equation \[eq:int\] is sufficient to inform that at these old ages there will not be residual main-sequence stars falling into the BSS box. This reinforces the interpretation that the majority of stars identified in the BSS box with $V_{\rm{TAN}}\gtrsim 200{\rm{\,km\,s}^{-1}}$ are genuine blue stragglers. The constant ratio of BSS at high $V_{\rm{TAN}}$ can also be qualitatively understood. I assume $0.8 M_{\odot}$ as the typical stellar mass for halo (sub)giant stars [e.g. @vdb; @eps], and I further assume that if a star occupies the BSS box, it must roughly be $>1 M_{\odot}$ (as supported by the isochrones check done above). The number of BSS with final mass $>1 M_{\odot}$ is given by a fraction of all possible combinations of mass for stars between $M_a=0.2$ and $M_b=0.8 M_{\odot}$, and which are in a binary system. Adopting the [@kroupa01] IMF (whose broken power law is more appropriate than the Salpeter one for masses below $M_K=0.5 M_{\odot}$) leads to: $$R = f_c f_m \frac{\displaystyle\int_{t_2}^t \displaystyle\int_{M_a}^{M_b}\phi(m)\textrm{d}m\,\textrm{d}t}{\displaystyle\int_{t_1}^t \displaystyle\int_{M(t)}^{p\,M(t)} \phi(m)\textrm{d}m\,\textrm{d}t}=$$ $$1.43 f_c f_m t_{\odot}^{0.43} \frac{\left(t-t_2\right)}{\left(t^{1.43}-t_1^{1.43}\right)} \times$$ $$\label{eq:kroupa} \frac{\displaystyle\frac{4}{15}\left(M_a^{-0.3}-M_K^{-0.3}\right)+\displaystyle\frac{2}{65}\left(M_K^{-1.3}-M_b^{-1.3}\right)}{\displaystyle\frac{2}{65} (1-p^{-1.3})}.$$ Here the number of RGB stars is given by the number of objects in the appropriate mass range that formed at least $t_1=9$ Gyr ago, where this age corresponds to a turnoff mass of $\sim 0.8 M_{\odot}$ (equation \[eq:kippe\]). Moving to the number of BSS, I do not make any assumption when their mergers occur (which in fact could happen at times more recent than $t_1$), nor their formation channels (e.g., binary interaction when both components are still on the main sequence, but also from interaction with an evolved primary with a mass somewhat higher than $M_b$). There is a degree of stochasticity in the time at which each binary merger will occur, besides the effect of stellar lifetime in the range $M_a$ to $M_b$, and of stragglers in the range $M_1$ to $M_2$ (which will then evolve toward the giant phase). Modeling these effects goes beyond the analytical formulation presented here. The net effect of decreasing the number of BSS at any given time can be expressed as a delay, where the number of BSS will start to increase linearly from $t_2>t_1$. Choosing, e.g., $t_2=10$ Gyr produces a smooth slope similar to what is seen at ${V_{\rm{TAN}}}$ between $200$ and $300{\rm{\,km\,s}^{-1}}$. For $t_2=t_1$ the rise is much steeper, although irrelevant for the qualitative sake of this discussion. Only a certain fraction of all possible mass combinations in the range $0.2-0.8 M_{\odot}$ will result in a sum $>1 M_{\odot}$, and this is accounted for by $f_c$. This correction factor is determined numerically, by generating a distribution of masses according to the adopted IMF, and for a given fraction of binaries $f_b$ computing how many will have a total mass $>1 M_{\odot}$. This returns $f_c \sim 0.18 f_b$, where $f_b$ can be taken from observations. The only free parameter is thus $f_m$, which is the fraction of binaries undergoing mass transfer/merger. This can be determined by requiring the plateau at old ages of Equation \[eq:kroupa\] to match that observed in the halo (Figure \[fig:mcanal\]). Decent agreement is obtained if $f_c f_m= 0.18 f_b f_m = 0.018$. Adopting $f_b=0.5$ for the fraction of metal-poor close binaries [@mo] implies $f_m=0.2$, i.e. 20 percent of close binaries (or equivalently 10 percent of stars) will undergo some sort of mass transfer and/or merger. Note that different values of binary fraction will vary the percentage of binaries undergoing mass transfer (e.g., $f_b=0.4$ implies $f_m=0.25$), but the total fraction of stars ($f_b f_m$) remains unchanged at 10 percent. The purpose of this analytical formulation is simply to show that with a few basic assumptions on the IMF and stellar lifetimes, it is possible to qualitatively describe the trend seen in the BSS-to-RGB ratio of Figure \[fig:video\] and \[fig:mcanal\]. At low ${V_{\rm{TAN}}}$ (i.e., young and intermediate age stellar populations) the trend reflects the number of main-sequence versus red giant branch stars. The flattening seen at high ${V_{\rm{TAN}}}$ (old populations) can instead be described assuming that stars in the BSS box are created by stellar mergers with a set of reasonable parameters. Building a volume complete sample of halo blue stragglers {#sec:norma} ========================================================= Membership to the stellar halo based only on $V_{\rm{TAN}}$ is rather approximate. [@gaiaHR] have revealed that stars with $V_{\rm{TAN}}>200{\rm{\,km\,s}^{-1}}$ fall along two well-defined sequences separated by roughly $0.1$ magnitude in color, dubbed the red and blue sequence. At $V_{\rm{TAN}}=200{\rm{\,km\,s}^{-1}}$ there is still a clear contribution of thick disk stars falling onto the red sequence, and the stellar halo becomes clearly dominant only above $V_{\rm{TAN}}=250-300{\rm{\,km\,s}^{-1}}$ [see Figure 5 in @sahl]. The blue sequence has been speculated to be formed by stars accreted by one (or more) massive dwarf galaxy, whereas the red sequence likely comprises the tail of the thick disk, kinematically heated by the accretion event [e.g., @helmi; @haywood; @dimatteo; @sahl; @koppe; @mye]. The BSS-to-RGB ratio in the left panel of Figure \[fig:mcanal\] bottoms off and reverses between 200 and $300{\rm{\,km\,s}^{-1}}$, after which it stabilizes to a mean (and median) value of $0.20$ with a standard deviation of $0.03$. This value is remarkably robust and well within the range measured by [@santana] in dwarf galaxies with no recent star formation ($0.29$ with a standard deviation of $0.17$), with selection boxes similar to those adopted here[^1]. Whether or not the local halo is formed by one or more disrupted dwarf galaxies, it might not come as a surprise that the density of BSS in low metallicity, low density environments, such as the Galactic halo and dwarf galaxies, is similar [see, e.g., @momany07]. [@santana] report a correlation between the number of BSS in a dwarf galaxy and the total stellar mass of the system (their equation 5). Motivated by the constant BSS-to-RGB ratio found at high tangential velocities and its similarity to that measured in dwarf galaxies, I use the aforementioned correlation to test whether it returns a sensible estimate for the stellar mass in the local halo. To successfully doing so, it is crucial to correctly assign BSS to the halo. This is not trivial to do purely based on $V_{\rm{TAN}}$ because of contamination from the thick disk above $200{\rm{\,km\,s}^{-1}}$, and the fact that the halo extends below this velocity. The blue sequence of the [Gaia]{} H-R diagram offers a way out. I define an RGB to be a member of the [Gaia]{} blue sequence (bRGB) if it falls on the green box of Figure \[fig:video\]. This box is contained within the RGB box, and its boundaries on the right-hand side have been defined selecting the midpoint where the Gaia red and blue sequences are most separated in the H-R diagram of high ${V_{\rm{TAN}}}$ stars (in a fashion similar to @sahl. This is best appreciated in the animation of Figure \[fig:video\]). This region encompasses mostly old, metal-poor red giants from the halo, although younger and more metal-rich red giant stars from the Galactic disk can contaminate it, especially at low ${V_{\rm{TAN}}}$. I use metallicities from the SkyMapper photometric survey [@c19] to identify halo stars purely from their chemistry, and I estimate the fraction of missing halo stars when cutting at a given ${V_{\rm{TAN}}}$ and height $Z$. With this correction, I then derive a volume complete number of halo bRGB, and use the constant BSS-to-RGB ratio of Figure \[fig:mcanal\] to estimate the number of halo blue stragglers, and total stellar mass within the same volume through equation 5 of [@santana]. It can be seen from the animation of Figure \[fig:video\] that at the highest ${V_{\rm{TAN}}}$ (where the halo sample is the cleanest, with all RGB stars virtually on the blue sequence) the BSS-to-RGB ratio remains constant at $0.2$. This indicates that contamination from thick disk stars affects equally BSS and RGB stars (not unexpectedly, since they have similar intrinsic luminosities and hence probe similar distances), and their ratio is thus a robust quantity. Below, all these steps are explained in detail with a case study, using a set of fixed parameters. This procedure is then generalized using 100,000 Monte Carlo realizations, where the parameters adopted in the case study are changed within a reasonable range. [Gaia]{} DR2 parallaxes have a typical precision of order $0.04$ mas for stars brighter than $G\sim15$ (which is appropriate for the bulk of this sample). This, together with the requirement of 10 percent precision in parallaxes (Section \[sec:data\]) limits completeness to parallaxes $\gtrsim0.4$ mas (or distances closer than about $2.5$ kpc). This estimate, however, does not account for the fact that stars with different intrinsic luminosities will be complete to different distances. The distribution of $G$ magnitudes for bRGB shows the typical power law of a uniform, volume complete sample up to $G\simeq14.1$ (Figure \[fig:3pan\]a). The green box of Figure \[fig:video\] requires a complete sampling of bRGB stars to be sensitive to $M_G\sim2.5$, thus implying that $D=10^{\frac{14.1-2.5}{5}-2}\sim 2$ kpc is the farthest distance at which the sample is complete. At bright magnitudes [Gaia]{} DR2 is complete down to $G\sim7$ , which together with the bright limit of $M_G\sim0.6$ for the bRGB box, translates to a distance completeness of $\sim 190$ pc. This limit is of no concern since I only select stars with $\|Z\|>400$ pc, to avoid regions heavily affected by reddening and crowding, as well as strong contamination from the disk (this cut in $Z$ eliminates all bRGB within $\sim 10^{\circ}$ from the Galactic plane). There are 1824 bRGB satisfying these criteria. SkyMapper provides \[Fe/H\] for some 9 million stars in the southern sky, with no selection other than having good photometry, in a color range that well encompasses the RGB box, and Galactic latitudes $\|b\|>5^{\circ}$ [@c19]. All stars from the SkyMapper sample are in [Gaia]{} DR2: I apply the quality flags and reddening corrections described in Section \[sec:data\] and identify SkyMapper members of the [Gaia]{} blue sequence with $\|Z\|>400$ pc using the same green selection box of Figure \[fig:video\]. Also for this sample restricting to distances closer than $2$ kpc is appropriate (the distribution of $G$ magnitudes for the SkyMapper sample peaks at a value similar to that of the [Gaia]{} sample). Figure \[fig:3pan\]c shows that within the bRGB box there is a considerable fraction of metal rich giants, as well as metal-poor stars with ${V_{\rm{TAN}}}<250{\rm{\,km\,s}^{-1}}$. I classify a star as halo if its \[Fe/H\]$<-1$, and I define the following correction for the fraction of missing halo stars: $$f_{ma}=1-\frac{n_{{V_{\rm{TAN}}}>250}^{\|Z\|>400}}{n_{[Fe/H]<-1}^{\|Z\|>400}},$$ where the numerator and denominator are the number of bRGB stars 400 pc above the Galactic plane with tangential velocities above $250{\rm{\,km\,s}^{-1}}$ and metallicities below $-1$, respectively. I find that $f_{ma}\sim 0.6$, i.e., within $2$ kpc about 60 percent of halo stars are lost when cutting at ${V_{\rm{TAN}}}>250{\rm{\,km\,s}^{-1}}$ and $\|Z\|>400$pc. Thus, the complete number of halo bRGB is of order 4500. The choice of using \[Fe/H\]$<-1$ to chemically assign stars to the halo is arbitrary, and the transition from the thick disc and halo is not clear-cut [e.g., @rl08; @ruchti10; @sahl]. Nevertheless, from Figure \[fig:3pan\]c it is clear that at the highest ${V_{\rm{TAN}}}$ (where the fraction of genuine halo stars is the highest) the bulk of stars has \[Fe/H\]$<-1$. Also, the adopted choice is consistent with the literature, where the broad metallicity distribution of the halo is found to become prominent below $-1$ [e.g., @rn91; @an13], whereas at high ${V_{\rm{TAN}}}$ the thick disk peaks at $-0.7$ [@sahl]. It must be noted that the completeness of the samples is partly decreased by the quality cuts described in Section \[sec:data\]. To assess their effect, I query the [Gaia]{} archive requiring only parallaxes better than 10 percent. This results in a sample of about $72.5$ million objects, i.e. about 7 percent larger than the one used in Section \[sec:data\]. This is consistent with the order 10 percent effect found by [@bb] when introducing quality cuts on a sample with parallaxes better than 20 percent. If I only consider stars in a color range broadly consistent with the location of the of BSS ($0<G_{{\rm BP}}-G_{{\rm RP}}<0.8$) and RGB ($0.7<G_{{\rm BP}}-G_{{\rm RP}}<1.3$) boxes, the mean and median difference of the two samples as a function of Galactic latitude is 6 percent, with a scatter of 4 percent. This check is to ensure the absence of significant trends with latitude, due to the fact that the quality flag [phot\_bp\_rp\_excess\_factor]{} is sensitive to increasing stellar crowding toward the plane of the Galaxy [@Gaia_evans]. I thus increase by 6 percent the number of halo bRGB, and convert those into the expected number of halo BSS using a fraction of 0.20 from Figure \[fig:mcanal\]. Using equation 5 from [@santana], I estimate a total halo stellar mass of $0.9\times10^6M_{\odot}$ within $\sim2$ kpc from the Sun. Accounting for the volume of a missing spherical segment of height $\pm Z$: $$V = \frac{\pi Z}{3}(6D^2-2Z^2),$$ where $D=2$ kpc and $Z=0.4$ kpc, returns a local stellar halo density $\rho= 3.4 \times 10^{-5} M_{\odot}/\rm{pc}^3$. The procedure outlined above is repeated 100,000 times, varying each time with Gaussian random errors the boundaries of the bRGB box (Figure \[fig:video\]), reddening by 20 percent, parallaxes within their quoted errors, completeness correction by $6\pm4$ percent, and imposing different cuts in $200<\|Z{\rm(pc)}\|<400$ and $200<{V_{\rm{TAN}}}({\rm{\,km\,s}^{-1}})<350$. A correction for the fraction of missing halo bRGB is determined each time, by similarly perturbing for the SkyMapper sample reddening, parallaxes, and metallicities by $0.2$ dex. The mode of the distribution of $G$ magnitudes is determined for the [Gaia]{} and the SkyMapper sample, and the brightest of the two is used to find the farthest distance at which both samples are complete given the faintest $M_G$ of the bRGB box used. The correction for the fraction of missing halo bRGB is applied to derive the actual number of halo bRGB, which is then converted into a number of BSS using a ratio of $0.20\pm0.03$. From the procedure described above, I obtain the following value for the average stellar halo density within $2$ kpc from the Sun (Figure \[fig:rho\]): $$\rho=(3.4 \pm 0.7) \times 10^{-5} M_{\odot}/\rm{pc}^3.$$ While the correction for the fraction of missing halo bRGB stars is a strong function of ${V_{\rm{TAN}}}$, $\rho$ is remarkably flat, as one would expect after applying a proper completeness correction. A wide range of density normalizations have been found in the literature [$3-15\,\times 10^{-5} M_{\odot}/\rm{pc}^3$ @morrison93; @fj98; @gould98; @dj10]. The value derived here compares more favorably to low normalizations, although the very different values obtained by different authors over the years highlight how difficult it is to derive a definitive measurement. It must be noted that the relation of [@santana] is not exactly linear between the number of BSS and stellar mass. In other words, the stellar mass of a dwarf galaxy with $n$ blue stragglers is different from the mass of $m$ dwarf galaxies, each containing $i$ blue stragglers $\sum i=n$. In the extreme (and unrealistic) case that each blue staggler comes from a different dwarf galaxy, the difference with respect to assuming all from the same dwarf amounts to $1-n^{-0.11}$, or about 50 percent for $n \sim 10^3$. Recent evidence suggests that the local Galactic halo is the result of two or few massive mergers [e.g., @mye], in which case the difference reduces to the order of some percent. However, we do not know the total number of blue stragglers, of which only a fraction is observed within 2 kpc. Fortunately, Figure \[fig:fit\] shows that the relation of [@santana] can be well approximated with a linear function within its uncertainties. Adopting the linear form of Figure \[fig:fit\], the inferred stellar halo density increases at the percent level only. It should be pointed out that the relation of [@santana] is calibrated between $\sim10^3$ and $\sim10^6 M_{\odot}$, and here it has been applied within this range. The comparison of the linear function in Figure \[fig:fit\] extends up to $10^9 M_{\odot}$ [the total stellar mass of the halo; e.g., @deason]. While it must be explored whether the adopted linear relation holds to this regime, the point here is that while different linear functions will change somewhat the stellar mass derived, the effect is within the quoted uncertainties. Scaling relations {#sec:bho} ================= The good agreement of the local baryon density with estimates from the literature from the previous section warrants further investigation of whether scaling relations derived from BSS in dwarf galaxies can be applied to the Milky Way halo. Figure \[fig:scaling\] shows the volume density of blue stragglers as a function of stellar density for both dwarf galaxies (solid blue line) and globular clusters (solid red line) from [@santana]. Stellar densities have been calculated within a half-light radius, using $0.5\,L_{V}/L_{\odot,V}$ [from @munoz18], and assuming a stellar mass-to-light ratio of $M_{\star}/L_{V}=1.5 M_{\odot}/L_{\odot,V}$. This value is appropriate for both dwarf galaxies and globular clusters in this sample [e.g., @woo08; @bau]. I have used half-light radii from Sérsic profiles $R_{h,s}$ since in [@munoz18] those are available for both dwarf galaxies and globular clusters, and note here that differences are negligible for systems having half-light radii from exponential or Plummer profiles. I multiply by $2/3$ the number of BSS in [@santana] since those are counted up to two half-light radii[^2]. The density of BSS steeply correlates with stellar density in both dwarf galaxies and globular clusters. For dwarf galaxies, the intercept of the relation with the density of halo BSS determined in the previous section returns a stellar density of $(3.1 \pm 0.5) \times 10^{-5} M_{\odot}/\rm{pc}^3$. The good agreement with the value previously determined is not unexpected, both ultimately depending on the same set of data. However, it must be noted that the choice of comparing the density of halo BSS to that within a half-light radius of dwarf galaxies is arbitrary. If stellar densities were to be computed within two half-light radii, the stellar density inferred for the halo would change to $(2.6 \pm 0.6) \times 10^{-5} M_{\odot}/\rm{pc}^3$. Not unexpectedly, the largest source of systematic uncertainty is the adopted stellar mass-to-light ratio, where a change of $\pm 0.5 M_{\odot}/L_{\odot,V}$ affects stellar densities by $\pm 1.0 \times 10^{-5} M_{\odot}/\rm{pc}^3$. More interestingly, the density of BSS is found to correlate with the dark matter density of dwarf galaxies, here computed again within a half-light radius. This can be understood from the dark matter scaling laws in late-type and dwarf spheroidal galaxies, where the dark matter halo core density correlates with the absolute magnitude of a galaxy, i.e., roughly stellar mass content [@kf16 see also Figure \[fig:smd\]]. Here stellar mass content is traced by BSS. The dark matter density has been estimated using dynamical mass-to-light ratios $M/L_{V}$ reported in [@munoz18], from which $M_{\rm{DM}}=(M_{\rm{dyn}}-M_{\star})=(M/L_{V}-M_{\star}/L_{V})L_{V}/2$, where the factor of 2 follows from the choice of working at half-light radius (i.e. in the following $M_{\rm{DM}}$, $M_{\rm{dyn}}$ and $M_{\star}$ are all computed within half-light radius). The contraction of the dark matter halo due to the addition of stars can be corrected by adiabatically expanding the half-light radius. Assuming circular orbits and angular momentum conservation, this gives $$R_{h,s}^{'}=\frac{M_{\star}+M_{\rm{DM}}}{M_{\rm{DM}}} R_{h,s}$$ [e.g., @blu; @for], which has virtually no effect since for this sample of dwarf galaxies $M_{\star}$ is orders of magnitude smaller than $M_{\rm{DM}}$. The approach used here averages the dark matter density over a half-light radius, which is appropriate if (dwarf) galaxies –as it seems– have cored profiles [e.g., @serra10; @read16; @li20]. The derived $M_{\rm{DM}}$ does not account for the effect of tidal stripping of the halos of dwarf galaxies. Nevertheless, this simple methodology returns dark matter densities that typically agree to within a few tens of percent with the values derived from the detailed modeling of [@read19] for the same galaxies in their sample. To account for uncertainties, I have repeated the above procedure to derive stellar and dark matter densities using half-light radii, stellar and dynamical masses from [@mcc12]. Differences with respect to the values obtained using mass-to-light ratios from [@munoz18] are typically of few tens of percent. I then generate a million realizations building each time a sample that randomly mixes data from [@mcc12] and [@munoz18], and I perturb them by a Gaussian of width equal to half of their differences. The intercept with the local density of halo BSS (also perturb within its uncertainties) returns a median dark matter density: $$\rho_{DM}= 0.0058^{+0.0053}_{-0.0026}\, M_{\odot}/\rm{pc}^3$$ $$=0.22^{+0.20}_{-0.10}\,\rm{GeV}/\rm{cm}^3.$$ The mean density is instead $0.0076 M_{\odot}/\rm{pc}^3 = 0.29\,\rm{GeV}/\rm{cm}^3$. In can also be appreciated from Figure \[fig:scaling\] that varying the stellar mass-to-light ratio by $\pm 0.5 M_{\odot}/L_{\odot,V}$ has a negligible impact on the inferred dark matter density. The median (mean) value of $\rho_{DM}$ determined here is in overall good agreement with those recently reported in the literature, which are in the range $0.005-0.013\,M_{\odot}/\rm{pc}^3$ [$\simeq 0.2-0.5\,\rm{GeV}/\rm{cm}^3$, see, e.g., @swe12; @z13; @bt12; @br13; @mckee15; @pmm17; @s18; @ds19]. These methods are based on dynamically modeling the rotation curve or the vertical motion of stars, and in either case a number of assumptions are needed. One of the most important ones is the contribution of baryonic matter to the local dynamical mass, which is nontrivial to determine and strongly correlates with the inferred dark matter density [e.g., @flynn06; @s18]. The use of BSS proposed here is largely independent of the baryonic content (Figure \[fig:scaling\], right panel), and the mass estimators used to infer dynamical masses within the half-light radius of dwarf galaxies are believed to be accurate [@camp17; @gosa]. The choice of applying a scaling relation inferred from dwarf galaxies to estimate the local dark matter halo is motivated by the similar BSS-to-RGB ratio measured in the local halo compared to dwarf galaxies. Admittedly, however, in dwarf galaxies these scaling relations are estimated at half-light radii, whereas here they are applied to field stars as a whole. Discussion and Conclusions {#sec:end} ========================== The correlation between stellar mass density and volume density of BSS shown in Figure \[fig:scaling\] can be readily understood from the findings of [@santana] i.e., the number count of BSS increases with the stellar mass content of dwarf galaxies, whereas it stays constant in globular clusters. With increasing stellar mass, globular clusters are typically more compact. This means that the volume density of BSS increases with stellar density, moving from the bottom left to the top right of the red line. For dwarf galaxies, the pathway is opposite. With increasing stellar mass content, the number of BSS in dwarf galaxies increases, and so do half-light radii. This leads to a decrease of both stellar mass density and BSS volume density with increasing galaxy mass, i.e. moving from the top right to the bottom left of the blue curve. Because the stellar mass content of dwarf galaxies correlates positively with their dark matter content, the same trend still holds when dark matter density is used instead of stellar density. Whether the correlation between the density of BSS and that of dark matter is indicative of a connection between baryons and dynamics is something worth contemplating [e.g., @sancisi; @stacy], but beyond the scope of this paper. The constant number of BSS as a function of cluster mass ($M_{\star}$) translates into a decreasing number of BSS per unit mass ($\propto 1/M_{\star}$), whereas the number of BSS per unit mass stays roughly constant in dwarf galaxies ($\propto M_{\star}^{\alpha} \times 1/M_{\star} \sim \rm{const}$), where $\alpha \sim 1$ (see Figure \[fig:fit\]). The number of BSS per unit mass can be interpreted either as a measure of formation or disruption efficiency of BSS. For example, if BSS are the product of close binaries, it could be argued that in denser stellar systems close binaries are less likely to form, or that closer binaries are more easily disrupted. While addressing these questions is beyond the scope of this paper [see, e.g., @momany15 for a review], I note that the trends discussed here using BSS can be traced in the stellar mass versus density relation (Figure \[fig:smd\]). This relation is equivalent to the more popular absolute magnitude vs. half-light radius relation, and shows how the stellar mass of a system has a positive correlation with stellar density in globular clusters, and a negative correlation with dark matter density in dwarf galaxies. BSS are thus tracing these scaling relations, and here I have applied them to the Milky Way halo. Further investigations are needed to confirm the use BSS as a proxy of baryon and dark matter density in the halo, as this technique could be very powerful, e.g., at measuring these quantities across the Milky Way halo on the same scale as in external galaxies. Remarkably, applying to the local Milky Way halo scaling relations inferred for BSS in dwarf galaxies is able to return both a stellar density and dark matter density that are in overall good agreement with other determinations in the literature. As discussed in the paper, BSS thus seem to trace stellar mass in low density, low metallicity environments regardless if in dwarf galaxies or in the halo. When it comes to dark matter, the correlation between the dark matter core density and stellar mass seen in dwarf galaxies returns a meaningful result also for the local halo. It thus seems that the nearby halo of a bright spiral like the Milky Way can lie on some of the scaling laws for dwarf galaxies. Whether this holds universally, or because the nearby halo is largely formed by disrupted dwarf galaxies remains to be seen. I thank Rosemary Wise for an inspiring colloquium on blue stragglers that set this work in motion, and Ken Freeman, Chris Flynn, Helmut Jerjen, Thomas Nordlander, and Aldo Serenelli for a reading of the manuscript and comments. Enrico di Teodoro, Ashley Ruiter, and Ivo Seitenzahl are acknowledged for helpful discussions, and Felipe Santana for useful correspondence. I thank an anonymous referee for constructive criticism, which has improved the paper. Funding for this work has been provided by the ARC Future Fellowship FT160100402. Parts of this research were conducted by the ARC Centre of Excellence ASTRO 3D, through project number CE170100013. This work has made use of data from the European Space Agency (ESA) mission [Gaia]{} (<https://www.cosmos.esa.int/gaia>), processed by the [Gaia]{} Data Processing and Analysis Consortium (DPAC, <https://www.cosmos.esa.int/web/gaia/dpac/consortium>). Funding for the DPAC has been provided by national institutions, in particular the institutions participating in the [Gaia]{} Multilateral Agreement. [^1]: As in Santana et al. (2013), I use the BSS-to-RGB ratio ($R$) at the face value. The actual value will be slightly higher, $R/(1-R)$ under the assumption that a fraction $R$ of RGB stars are in fact evolved BSS. At the same time, the measured $R$ is an upper limit, since the adopted box does not extend to the tip of the RGB. [^2]: Assuming for simplicity an exponential profile $e^{-kx}$ where $k$ is an integer, there is $3/4$ of light within two half-light radii, and $2/3$ of this $3/4$ is within a half-light.	{ "pile_set_name": "ArXiv" }
--- author: - \| \ \ \ \ \ \ Masoud Khalkhali\ Mathematics Department, University of Western Ontario\ London ON, Canada date: title: Very Basic Noncommutative Geometry --- Introduction ============ One of the major advances of science in the 20th century was the discovery of a mathematical formulation of quantum mechanics by Heisenberg in 1925 [@hei]. From a mathematical point of view, transition from classical mechanics to quantum mechanics amounts to, among other things, passing from the [commutative algebra]{} of [classical observables]{} to the [noncommutative algebra]{} of [quantum mechanical observables]{}. Recall that in classical mechanics an observable (e.g. energy, position, momentum, etc.) is a function on a manifold called the phase space of the system. Immediately after Heisenberg’s work, ensuing papers by Dirac [@dir] and Born-Heisenberg-Jordan [@bhj], made it clear that a quantum mechanical observable is a (selfadjoint) operator on a Hilbert space called the state space of the system. Thus the commutative algebra of functions on a space is replaced by the noncommutative algebra of operators on a Hilbert space. A little more than fifty years after these developments, Alain Connes realized that a similar procedure can in fact be applied to areas of mathematics where the classical notions of space (e.g. measure space, locally compact space, or a smooth space) looses its applicability and pertinence and can be replaced by a new idea of space, represented by a noncommutative algebra. Connes’ theory, which is generally known as [noncommutative geometry]{}, is a rapidly growing new area of mathematics that interacts with and contributes to many disciplines in mathematics and physics. For a recent survey see Connes’ article [@ac00]. Examples of such interactions and contributions include: theory of operator algebras, index theory of elliptic operators, algebraic and differential topology, number theory, standard model of elementary particles, quantum Hall effect, renormalization in quantum field theory, and string theory. (For a description of these relations in more details see the report below.) To understand the basic ideas of noncommutative geometry one should perhaps first come to grips with the idea of a [noncommutative space]{}. The inadequacy of the classical notions of space manifests itself for example when one deals with highly singular bad quotients"; spaces such as the quotient of a nice space by the ergodic action of a group or the space of leaves of a foliation. In all these examples the quotient space is typically ill behaved even as a topological space. For example it may fail to be even Hausdorff, or have enough open sets, let alone being a reasonably smooth space. The unitary dual of a noncompact (Lie) group, except when the group is abelian or almost abelian, is another example of an ill behaved space. One of Connes’ key observations is that in all these situations one can attach a noncommutative algebra, through a [noncommutative quotient construction]{}, that captures most of the information. Examples of this noncommutative quotient construction include crossed product by action of a group, or by action of a groupoid. In general the noncommutative quotient is the groupoid algebra of a topological groupoid. Noncommutative geometry has as its limiting case the classical geometry, but geometry expressed in algebraic terms. Thus to understand its relation with classical geometry one should first understand one of the most important ideas of mathematics which can be expressed as a [duality]{} between commutative algebra and geometry. This is by no means a new observation or a new trend. To the contrary, this duality has always existed and been utilized in mathematics and its applications. The earliest example is perhaps the use of numbers in counting! It is, however, the case that throughout the history each new generation of mathematicians find new ways of formulating this principle and at the same time broaden its scope. Just to mention a few highlights of this rich history we mention Descartes (analytic geometry), Hilbert (affine varieties and commutative algebras), Gelfand-Naimark (locally compact spaces and commutative $C^$-algebras), Grothendieck (Schemes and topos theory), and Connes (noncommutative geometry). A key idea here is the well-known relation between a space and the commutative algebra of functions on that space. More precisely there is a duality between certain categories of geometric spaces and categories of algebras representing those spaces. Noncommutative geometry builds on, and vastly extends, this fundamental duality between geometry and commutative algebras. For example, by a celebrated theorem of Gelfand and Naimark [@gelnai] one knows that the category of locally compact Hausdorff spaces is equivalent to the dual of the category of commutative $C^$-algebras. Thus one can think of not necessarily commutative $C^$-algebras as the dual of a category of [noncommutative locally compact spaces]{}. What makes this a successful proposal is first of all a rich supply of examples and secondly the possibility of extending many of the topological and geometric invariants to this new class of spaces. Let us briefly recall a few other examples from a long list of results in mathematics that put in duality certain categories of geometric objects with a corresponding category of algebraic objects. To wit, Hilbert’s Nullstellensatz states that the category of algebraic varieties over an algebraically closed field is equivalent to the dual of the category of finitely generated commutative algebras without nilpotent elements (so called reduced algebras). This is a perfect analogue of the Gelfand-Naimark theorem in the world of commutative algebras. Similarly, the Serre-Swan theorem states that the category of vector bundles over a compact Hausdorff space (resp. affine algebraic variety) X is equivalent to the category of finitely generated projective modules over the algebra of continuous functions (resp. regular functions) on X. Thus a pervasive idea in noncommutative geometry is to treat (certain classes) of noncommutative algebras as noncommutative spaces and try to extend tools of geometry, topology, and analysis to this new setting. It should be emphasized, however, that, as a rule, this extension is never straightforward and always involve surprises and new phenomena. For example the theory of the flow of weights and the corresponding modular automorphism group in von Neumann algebras has no counterpart in classical measure theory, though the theory of von Neumann algebras is generally regarded as noncommutative measure theory. Similarly the extension of de Rham homology for manifolds to cyclic cohomology for noncommutative algebras was not straightforward and needed some highly nontrivial considerations. Of all the topological invariants for spaces, topological $K$-theory has the most straightforward extension to the noncommutative realm. Recall that topological $K$-theory classifies vector bundles on a topological space. Using the above mentioned Serre-Swan theorem, it is natural to define, for a not necessarily commutative ring $A$, $K_0(A)$ as the group defined by the semi-group of isomorphism classes of finite projective $A$-modules. The definition of $K_1(A)$ follows the same pattern as in the commutative case, provided $A$ is a Banach algebra and the main theorem of topological $K$-theory, the Bott periodicity theorem, extends to all Banach algebras. The situation is much less clear for $K$-homology, the theory dual to $K$-theory. By the work of Atiyah, Brown-Douglas-Fillmore, and Kasparov, one can say, roughly speaking, that $K$-homology cycles on a space $X$ are represented by abstract elliptic operators on $X$ and while $K$-theory classifies vector bundles on $X$, $K$-homology classifies the abstract elliptic operators on $X$. The pairing between $K$-theory and $K$-homology takes the form $<[D], [E]>=$ the Fredholm index of the elliptic operator $D$ with coefficients in the vector bundle $E$. Now one good thing about this way of formulating $K$-homology is that it almost immediately extends to noncommutative $C^$-algebras. The two theories are unified in a single theory called $KK$-theory due to G. Kasparov. Cyclic cohomology was discovered by Connes in 1981 [@ac81; @ac85] as the right noncommutative analogue of de Rham homology of currents and as a target space for noncommutative Chern character maps from both $K$-theory and $K$-homology. One of the main motivations of Connes seems to be transverse index theory on foliated spaces. Cyclic cohomology can be used to identify the $K$-theoretic index of transversally elliptic operators which lie in the $K$-theory of the noncommutative algebra of the foliation. The formalism of cyclic cohomology and Chern-Connes character maps form an indispensable part of noncommutative geometry. In a different direction, cyclic homology also appeared in the 1983 work of Tsygan [@tsy] and was used, independently, also by Loday and Quillen [@lq] in their study of the Lie algebra homology of the Lie algebra of stable matrices over an associative algebra. We won’t pursue this aspect of cyclic homology in these notes. A very interesting recent develpoment in cyclic cohomology theory is the [Hopf-cyclic cohomology]{} of Hopf algebras and Hopf module (co)algebras in general. Motivated by the original work of Connes and Moscovici [@cm1; @cm2] this theory is now extended and elaborated on by several authors [@ak1; @ak2; @hkrs1; @hkrs2; @kr1; @kr2; @kr3; @kr4]. There are also very interesting relations between cocycles for Hopf-cyclic cohomology theory of the Connes-Moscovici Hopf algebra $\mathcal{H}_1$ and operations on spaces of modular forms and modular Hecke algebras [@cm3; @cm4], and spaces of $\mathbb{Q}$-lattices [@cma]. We will say nothing about these develpments in these notes. Neither we shall discuss the approach of Cuntz and Quillen to cyclic cohomology theory and their cellebrated proof of excision property for periodic (bivariant) cyclic cohomology [@cu; @cst; @cq1; @cq2; @cq3]. The following dictionary" illustrates noncommutative analogues of some of the classical theories and concepts originally conceived for spaces. In these notes we deal only with a few items of this dictionary. For a much fuller account and explanations, as well as applications of noncommutative geometry, the reader should consult Connes’ beautiful book [@acb]. $$ [commutative]{} [noncommutative]{} -------------------------- ---------------------------------- measure space von Neumann algebra locally compact space $C^\ast$ - algebra vector bundle finite projective module complex variable operator on a Hilbert space real variable sefadjoint operator infinitesimal compact operator range of a function spectrum of an operator $K$-theory $K$-theory vector field derivation integral trace closed de Rham current cyclic cocycle de Rham complex Hochschild homology de Rham cohomology cyclic homolgy Chern character Chern-Connes character Chern-Weil thoery noncommutative Chern-Weil thoery elliptic operator $K$-cycle spin Riemannian manifold spectral triple index theorem local index formula group, Lie algebra Hopf algebra, quantum group symmetry action of Hopf algebra $$ Noncommutative geometry is already a vast subject. These notes are just meant to be an introduction to a few aspects of this fascinating enterprize. To get a much better sense of the beauty and depth of the subject the reader should consult Connes’ magnificent book [@acb] or his recent survey [@ac00] and references therein. Meanwhile, to give a sense of the state of the subject at the present time, its relation with other fields of mathematics, and its most pressing issues, we reproduce here part of the text of the final report prepared by the organizers of a conference on noncommutative geometry in 2003 [^1]:\ \ This conjecture, in its simplest form, is formulated for any locally compact topological group. There are more general Baum-Connes conjectures with coefficients for groups acting on C\-algebras, for groupoid C\-algebras, etc., that for the sake of brevity we don’t consider here. In a nutshell the Baum-Connes conjecture predicts that the K-theory of the group C\-algebra of a given topological group is isomorphic, via an explicit map called the Baum-Connes map, to an appropriately defined K-homology of the classifying space of the group. In other words invariants of groups defined through noncommutative geometric tools coincide with invariants defined through classical algebraic topology tools. The Novikov conjecture on the homotopy invariance of higher signatures of non-simply connected manifolds is a consequence of the Baum-Connes conjecture (the relevant group here is the fundamental group of the manifold). Major advances were made in this problem in the past seven years by Higson-Kasparov, Lafforgue, Nest-Echterhoff-Chabert, Yu, Puschnigg and others.\ [2. Cyclic cohomology and KK-theory]{}\ A major discovery made by Alain Connes in 1981, and independently by Boris Tsygan in 1983, was the discovery of cyclic cohomology as the right noncommutative analogue of de Rham homology and a natural target for a Chern character map from K-theory and K-homology. Coupled with K-theory, K-homology and KK-theory, the formalism of cyclic cohomology fully extends many aspects of classical differential topology like Chern-Weil theory to noncommutative spaces. It is an indispensable tool in noncommutative geometry. In recent years Joachim Cuntz and Dan Quillen have formulated an alternative powerful new approach to cyclic homology theories which brings with it many new insights as well as a successful resolution of an old open problem in this area, namely establishing the excision property of periodic cyclic cohomology.* For applications of noncommutative geometry to problems of index theory, e.g. index theory on foliated spaces, it is necessary to extend the formalism of cyclic cohomology to a bivariant cyclic theory for topological algebras and to extend Connes’s Chern character to a fully bivariant setting. The most general approach to this problem is due to Joachim Cuntz. In fact the approach of Cuntz made it possible to extend the domain (and definition) of KK-theory to very general categories of topological algebras (rather than just C\-algebras). The fruitfulness of this idea manifests itself in the V. Lafforgue’s proof of the Baum-Connes conjecture for groups with property T, where the extension of KK functor to Banach algebras plays an important role. A new trend in cyclic cohomology theory is the study of the cyclic cohomology of Hopf algebras and quantum groups. Many noncommutative spaces, such as quantum spheres and quantum homogeneous spaces, admit a quantum group of symmetries. A remarkable discovery of Connes and Moscovici in the past few years is the fact that diverse structures, such as the space of leaves of a (codimension one) foliation or the space of modular forms, have a unified quantum symmetry. In their study of transversally elliptic operators on foliated manifolds Connes and Moscovici came up with a new noncommutative and non-cocommutative Hopf algebra denoted by $\mathcal{H}_n$ (the Connes-Moscovici Hopf algebra). $\mathcal{H}_n$ acts on the transverse foliation algebra of codimension $n$ foliations and thus appears as the quantized symmetries of a foliation. They noticed that if one extends the noncommutative Chern-Weil theory of Connes from group and Lie algebra actions to actions of Hopf algebras, then the characteristic classes defined via the local index formula are in the image of this new characteristic map. This extension of Chern-Weil theory involved the introduction of cyclic cohomology for Hopf algebras.\ [3. Index theory and noncommutative geometry]{}\ The index theorem of Atiyah and Singer and its various generalizations and ramifications are at the core of noncommutative geometry and its applications. A modern abstract index theorem in the noncommutative setting is the local index formula of Connes and Moscovici. A key ingredient of such an abstract index formula is the idea of an spectral triple due to Connes. Broadly speaking, and neglecting the parity, a spectral triple $(A, H, D)$ consists of an algebra $A$ acting by bounded operators on the Hilbert space $H$ and a self-adjoint operator $D$ on $H$. This data must satisfy certain regularity properties which constitute an abstraction of basic elliptic estimates for elliptic PDE’s acting on sections of vector bundles on compact manifolds. The local index formula replaces the old non-local Chern-Connes cocycle by a new Chern character form $Ch (A, H, D)$ of the given spectral triple in the cyclic complex of the algebra $A$. It is a local formula in the sense that the cochain $Ch (A, H, D)$ depends, in the classical case, only on the germ of the heat kernel of $D$ along the diagonal and in particular is independent of smooth perturbations. This makes the formula extremely attractive for practical calculations. The challenge now is to apply this formula to diverse situations beyond the cases considered so far, namely transversally elliptic operators on foliations (Connes and Moscovici) and the Dirac operator on quantum $SU_2$ (Connes).\ [4. Noncommutative geometry and number theory]{}\ Current applications and connections of noncommutative geometry to number theory can be divided into four categories. (1) The work of Bost and Connes, where they construct a noncommutative dynamical system $(B, \sigma_t)$ with partition function the Riemann zeta function $\zeta (\beta )$, where $\beta$ is the inverse temperature. They show that at the pole $\beta =1$ there is an spontaneous symmetry breaking. The symmetry group of this system is the group of idéles which is isomorphic to the Galois group $Gal (Q^{ab}/Q)$. This gives a natural interpretation of the zeta function as the partition function of a quantum statistical mechanical system. In particular the class field theory isomorphism appears very naturally in this context. This approach has been extended to the Dedekind zeta function of an arbitrary number field by Cohen, Harari-Leichtnam, and Arledge-Raeburn-Laca. All these results concern abelian extensions of number fields and their generalization to non-abelian extensions is still lacking. (2) The work of Connes on the Riemann hypothesis. It starts by producing a conjectural trace formula which refines the Arthur-Selberg trace formula. The main result of this theory states that this trace formula is valid if and only if the Riemann hypothesis is satisfied by all $L$-functions with Grössencharakter on the given number field $k$. (3) The work of Connes and Moscovici on quantum symmetries of the modular Hecke algebras $\mathcal{A}(\Gamma)$ where they show that this algebra admits a natural action of the transverse Hopf algebra $\mathcal{H}_1$. Here $\Gamma$ is a congruence subgroup of $SL (2, Z)$ and the algebra $\mathcal{A}(\Gamma)$ is the crossed product of the algebra of modular forms of level $\Gamma $ by the action of the Hecke operators. The action of the generators $X, Y$ and $\delta_n$ of $\mathcal{H}_1$ corresponds to the Ramanujan operator, to the weight or number operator, and to the action of certain group cocycles on $GL^+ (2, Q)$, respectively. What is very surprising is that the same Hopf algebra $\mathcal{H}_1$ also acts naturally on the (noncommutative) transverse space of codimension one foliations. (4) Relations with arithmetic algebraic geometry and Arakelov theory. This is currently being pursued by Consani, Deninger, Manin, Marcolli and others.\ [5. Deformation quantization and quantum geometry* ]{}\ The noncommutative algebras that appear in noncommutative geometry usually are obtained either as the result of a process called noncommutative quotient construction or by deformation quantization of some algebra of functions on a classical space. These two constructions are not mutually exclusive. The starting point of deformation quantization is an algebra of functions on a Poisson manifold where the Poisson structure gives the infinitesimal direction of quantization. The existence of deformation quantizations for all Poisson manifolds was finally settled by M. Kontsevich in 1997 after a series of partial results for symplectic manifolds. The algebra of pseudodifferential operators on a manifold is a deformation quantization of the algebra of classical symbols on the cosphere bundle of the manifold. This simple observation is the beginning of an approach to the proof of the index theorem, and its many generalizations by Elliott-Natsume-Nest and Nest-Tsygan, using cyclic cohomology theory. The same can be said about Connes’s groupoid approach to index theorems. In a different direction, quantum geometry also consists of the study of noncommutative metric spaces and noncommutative complex structures."\ Let us now briefly describe the contents of these notes. In Section 2 we describe some of the fundamental algebra-geometry correspondences at work in mathematics. The most basic ones for noncommutative geometry are the Gelfand-Naimark and the Serre-Swan theorems. In Section 3 we describe the noncommutative quotient construction and give several examples. This is one of the most universal methods of constructing noncommutative spaces directly related to classical geometric examples. Section 4 is devoted to cyclic cohomology and its various definitions. In Section 5 we define the Chern-Connes character map, or the noncommutative Chern character map, from $K$-theory to cyclic cohomology. In an effort to make these notes as self contained as possible, we have added three appendices covering very basic material on $C^$-algebras, projective modules, and category theory language. These notes are partly based on series of lectures I gave at the Fields Institute in Toronto, Canada, in Fall 2002 and at the Institute for Advanced Studies in Physics and Mathematics (IPM), Tehran, Iran, in Spring 2004. I also used part of these notes in my lectures at the [second annual spring institute and workshop on noncommutative geometry]{} in Spring 2004, Vanderbilt University, USA. It is a great pleasure to thank the organizers of this event, Alain Connes (director), to whom I owe much more than I can adequately express, Dietmar Bisch, Bruce Hughes, Gennady Kasparov, and Guoliang Yu. I would also like to thank Reza Khosrovshahi the director of the mathematics division of IPM in Tehran whose encouragement and support was instrumental in bringing these notes to existence. Some examples of geometry-algebra correspondence ================================================= We give several examples of geometry-commutative algebra correspondences. They all put into correspondence, or duality, certain categories of geometric objects with a category of algebraic objects. Presumably, the more one knows about these relations the better one is prepared to pursue noncommutative geometry. Locally compact spaces and commutative $C^$-algebras ----------------------------------------------------- In functional analysis the celebrated [Gelfand-Naimark Theorem]{} [@gelnai] states that the category of locally compact Hausdorff spaces is anti-equivalent to the category of commutative $C^$-algebras: $$\{\text{\bf locally compact Hausdorff spaces}\} \simeq \{\text{\bf commutative $C^$-algebras} \}^{op}.$$ Let $\mathcal{S}$ be the category whose objects are locally compact Hausdorff spaces and whose morphisms are continuous and [proper]{} maps. (Recall that a map $f:X\to Y$ is called proper if for any compact $K\subset Y,\; f^{-1}(K)$ is compact; of course, if $X$ is compact and $f$ is continuous, then $f$ is proper). Let $\mathcal{C}$ be the category whose objects are commutative $C^\ast$-algebras and whose morphisms are [proper]{} $\ast$-homomorphisms. (A $\ast$ -homomorphism $f: A \rightarrow B$ is called proper if for any approximate identity $(e_i)$ in $A$, $f(e_i)$ is an approximate identity in $B$. See Appendix A for definitions.) Define two contravariant functors $$C_0 :\mathcal{C} \to \mathcal{S}, \quad \Omega : \mathcal{S}\to \mathcal{C},$$ as follows. For a locally compact Hausdorff space $X$, let $C_0(X)$ denote the algebra of complex valued continuous functions on $X$ that “vanish at $\infty$". This means for any $\epsilon >0$ there is a compact subset $K\subset X$ such that $\|f(x)\|< \epsilon$ for $x\notin K$: $$C_0(X)= \{f: X \rightarrow \mathbb{C}, \text{ f is continuous and $f(\infty)=0$}\}.$$ Under pointwise addition and scalar multiplication $C_0(X)$ is obviously an algebra over the field of complex numbers $\mathbb{C}$. Endowed with the sup-norm $$\\| f \\| = \\| f \\|_\infty = \sup \{ \|f(x) \| ; \; x\in X \},$$ and $\ast$-operation $$f\mapsto f^\ast,\; f^\ast (x) = \bar{f}(x),$$ one checks that $C_0(X)$ is a commutative $C^\ast$-algebra. If $f:X\to Y$ is a continuous and proper map, let $$C_0(f)=f^: C_0(Y) \longrightarrow C_0(X), \quad f^(g)=g\circ f,$$ be the pullback of $f$. It is a proper $\ast$- homomorphism of $C^\ast$-algebras. We have thus defined the functor $C_0$. To define $\Omega$, called the [functor of points]{} or the [spectrum functor]{}, let $A$ be a commutative $C^\ast$-algebra. Let $$\Omega(A) = \text{set of characters of $A$} =Hom_{\mathcal{C}}(A, \mathbb{C}),$$ where a [character]{} is simply a nonzero algebra map $A \rightarrow \mathbb{C}$. (it turns out that they are also $$-morphisms). $\Omega (A)$ is a locally compact Hausdorff space under the topology of pointwise convergence. Given a proper morphism of $C^$-algebras $f:A\to B$, let $$\Omega (f) :\Omega (B) \to \Omega (A), \quad \Omega (f)=f^,$$ where $f^\ast (\varphi )=\varphi \circ f$. It can be shown that $\Omega (f)$ is a proper and continuous map. To show that $C_0$ and $\Omega$ are equivalences of categories, quasi-inverse to each other, one shows that for any locally compact Hausdorff space $X$ and any commutative $C^\ast$-algebra $A$, there are natural isomorphisms $$X\overset{\sim}{\longrightarrow}\Omega (C_0 (X)), \quad \quad x\mapsto e_x,$$ $$A\overset{\sim}{\longrightarrow} C_0 (\Omega (A)), \quad \quad a\mapsto \hat{a}.$$ Here $e_x$ is the [evaluation at $x$]{} map defined by $e_x(f)=f(x)$, and $a\mapsto \hat{a}$ is the celebrated [Gelfand transform]{} defined by $\hat{a}(\varphi)=\varphi (a).$ The first isomorphism is much easier to establish and does not require the theory of Banach algebras. The second isomorphism is what is proved by Gelfand and Naimark in 1943 [@gelnai] using Gelfand’s theory of commutative Banach algebras. We sketch a proof of this result is Appendix A. Under the Gelfand-Naimark correspondence compact Hausdorff spaces correspond to unital $C^$-algebras. We therefore have a duality, or equivalence of categories $$\{ \mbox{\bf compact Hausdorff spaces} \} \simeq\{ \mbox{\bf commutative unital $C^\ast$-algebras} \}^{op}.$$ Based on Gelfand-Naimark theorem, we can think of the dual of the category of not necessarily commutative $C^$-algebras as the category of [noncommutative locally compact Hausdorff spaces]{}. Various operations and concepts for spaces can be paraphrased in terms of algebras of functions on spaces and can then be immediately generalized to noncommutative spaces. This is the easy part of noncommutative geometry! Here is a dictionary suggested by the Gelfand-Naimark theorem: $$\begin{matrix} \mbox{\bf space } & \mbox{\bf algebra }\\ \mbox{compact} & \mbox{unital}\\ \mbox{1-point compactification} & \mbox{unitization}\\ \mbox{Stone-Cech compactification} & \mbox{multiplier algebra}\\ \mbox{closed subspace; inclusion } & \mbox{closed ideal; quotient algebra}\\ \mbox{surjection} & \mbox{injection}\\ \mbox{injection} & \mbox{surjection}\\ \mbox{homoemorphism} & \mbox{automorphism}\\ \mbox{Borel measure} & \mbox{ positive functional}\\ \mbox{probability measure} & \mbox{state}\\ \mbox{disjoint union}& \mbox{direct sum}\\ \mbox{cartesian product} & \mbox{minimal tensor product}\\ \end{matrix}$$ Vector bundles and finite projective modules --------------------------------------------- [Swan’s Theorem]{} [@swa] states that the category of complex vector bundles on a compact Hausdorff space $X$ is equivalent to the category of finite (i.e. finitely generated) projective modules over the algebra $C(X)$ of continuous functions on $X$: $$\{\text{\bf vector bundles on X}\} \simeq \{ \text{\bf finite projective C(X)-modules}\}.$$ There are similar results for real and quaternionic vector bundles [@swa]. This result was motivated and in fact is the topological counterpart of a an analogous result, due to Serre, which characterizes algebraic vector bundles over an affine algebraic variety as finite projective modules over the coordinate ring of the variety. Swan’s theorem sometimes is called the Serre-Swan theorem. Recall that a right module $P$ over a unital algebra $A$ is called [projective]{} if there exists a right $A$-module $Q$ such that $$P\oplus Q\simeq \bigoplus_I A,$$ is a free module. Equivalently, $P$ is projective if every module surjection $P \to Q \to 0$ splits as a right $A$-module map. $P$ is called [finite]{} if there exists a surjection $A^n \to P \to 0$ for some integer $n$. We describe the Serre-Swan correspondence between vector bundles and finite projective modules. Given a vector bundle $p: E\rightarrow X$ , let $$P= \Gamma (E)=\{s: X\rightarrow E;\; ps=id_X \}$$ be the set of all continuous [global sections]{} of $E$. It is clear that under fiberwise scalar multiplication and addition, $P$ is a $C(X)$ module. If $f: E \to F$ is a bundle map, we define a module map $\Gamma (f): \Gamma (E) \to \Gamma (F)$ by $\Gamma (f)(s)(x)=f(s(x))$ for all $s\in \Gamma (E)$ and $x\in X$. We have thus defined a functor $\Gamma $, called the global section functor, from the category of vector bundles over $X$ and continuous bundle maps to the category of $C(X)$-modules and module maps. Using compactness of $X$ and a partition of unity one shows that there is a vector bundle $F$ on $X$ such that $E\oplus F\simeq X\times \mathbb{C}^n$ is a trivial bundle. Let $Q$ be the space of global sections of $F$. We have $$P \oplus Q\simeq A^n,$$ which shows that $P$ is finite projective. To show that all finite projective $C(X)$-modules arise in this way we proceed as follows. Given a finite projective $C(X)$-module $P$, let $Q$ be a $C(X)$-module such that $P \oplus Q\simeq A^n,$ for some integer $n$. Let $e:A^n\rightarrow A^n$ be the right $A$-linear map corresponding to the projection into first coordinate: $(p, q)\mapsto (p, 0)$. It is obviously an idempotent in $M_n(C(X))$. One defines a vector bundle $E$ as the image of this idempotent $e$: $$E=\{(x, v); \; e(x)v=v, \mbox{for all}\, x \in X, \,v\in \mathbb{C}^n\}\subset X\times \mathbb{C}^n.$$ Now it is easily shown that $\Gamma (E)\simeq P$. With some more work it is shown that the functor $\Gamma$ is full and faithful and hence defines an equivalence of categories. Based on the Serre-Swan theorem, one usually thinks of finite projective modules over noncommutative algebras as [noncommutative vector bundles]{}. We give a few examples starting with a commutative one.\ [Examples]{}\ 1. The [Hopf line bundle]{} on the two sphere $S^2$, also known as [magnetic monopole bundle]{}, can be defined in various ways. (It was discovered, independently, by Hopf and Dirac in 1931, motivated by very different considerations). Here is an approach that lends itself to noncommutative generalizations. Let $\sigma_1$, $\sigma_2$, $\sigma_3$, be three matrices in $M_2(\mathbb{C})$ that satisfy the [canonical anticommutation relations]{}: $$\sigma_i \sigma_j +\sigma_j \sigma_i =2\delta_{ij},$$ for all $i, j=1, 2, 3.$ Here $\delta_{ij}$ is the Kronecker symbol. A canonical choice is the so called [Pauli spin matrices]{}: $$\sigma_1=\left( \begin{matrix} 0 & 1 \\ 1 & 0 \end{matrix} \right) \; , \quad \sigma_2 = \left( \begin{matrix} 0 & i \\ -i & 0 \end{matrix} \right) \; , \quad \sigma_3 = \left( \begin{matrix} 1 & 0 \\ 0 & -1 \end{matrix} \right) \; .$$ Define a function $$F: S^2 \rightarrow M_2(\mathbb{C}), \quad F(x_1, x_2, x_3)=x_1\sigma_1 +x_2\sigma_2 +x_3\sigma_3,$$ where $x_1, x_2, x_3$ are coordinate functions on $S^2$ so that $x_1^2+ x_2^2+ x_3^2=1$. Then $F^2 (x)=I_{2\times 2}$ for all $x\in S^2$ and therefore $$e=\frac{1+F}{2}$$ is an idempotent in $M_2(C(S^2))$. It thus defines a complex vector bundle on $S^2$. We have, $$e(x_1, x_2, x_3)= \frac{1}{2} \left( \begin{matrix} 1 + x_3 & x_1 + ix_2 \\ x_1-ix_2 & 1-x_3 \end{matrix} \right).$$ Since $$rank\, F(x)= trace \,F(x)=1$$ for all $x\in S^2$, we have in fact a complex line bundle over $S^2$. It can be shown that it is the line bundle associated to the Hopf fibration $$S^1 \to S^3 \to S^2.$$ Incidentally, $e$ induces a map $f: S^2 \to P^1 (\mathbb{C} )$, where $f(x)$ is the 1-dimensional subspace defined by the image of $F(x)$, which is 1-1 and onto. Our line bundle is just the pull back of the canonical line bundle over $P^1 (\mathbb{C} )$. This example can be generalized to higher dimensional spheres. One can construct matrices $\sigma_1, \cdots, \sigma_{2n+1}$ in $M_{2^n} (\mathbb{C})$ satisfying the [Clifford algebra relations]{} [@kar] $$\sigma_i\sigma_j + \sigma_j\sigma_i = 2\delta_{ij},$$ for all $i, j=1, \cdots, 2n+1.$ Define a matrix valued function $F$ on the $2n$ dimensional sphere $S^{2n}$, $F\in M_{2^n} (C(S^{2n}))$ by $$F=\sum_{i=1}^{2n+1} x_i \sigma_i.$$ Then $F^2(x) =1$ for all $x\in S^{2n}$ and hence $e=\frac{1+F}{2}$ is an idempotent.\ 2. (Hopf line bundle on quantum spheres)\ The Podleś quantum sphere $S^2_q$ is the $\ast$-algebra generated over $\mathbb{C}$ by the elements $a, a^{\ast}$ and $b$ subject to the relations $$a a^{\ast}+q^{-4}b^2=1,\;a^{\ast}a+b^2=1,\;ab=q^{-2}ba,\;a^{\ast}b=q^2ba^{\ast}.$$ The quantum analogue of the Dirac(or Hopf) monopole line bundle over $S^2$ is given by the following idempotent in $M_2(S^2_q)$ [@bm] $$\mathbf{e}_q=\frac{1}{2}\left[\begin{array}{cc} 1+q^{-2}b & qa \\ q^{-1} a^{\ast} & 1-b \end{array}\right].$$ It can be directly checked that $\mathbf{e}_q^2=\mathbf{e}_q.$\ 3. (Projective modules on the noncommutative torus)\ Let us first recall the definition of the [smooth noncommutative torus]{} $\mathcal{A}_{\theta}$ [@ac80]. Among several possible definitions the following is the most direct one. Let $\theta \in \mathbb{R}$ be a fixed parameter and let $$\mathcal{A}_{\theta}=\{\sum_{m, n \in \mathbb{Z}}a_{mn}U^mV^n;\; (a_{mn})\in \mathcal{S}(\mathbb{Z}^2)\},$$ where $\mathcal{S}(\mathbb{Z}^2)$ is the Schwartz space of rapidly decreasing sequences $(a_{mn})\in \mathbb{C}$ indexed by $\mathbb{Z}^2$. The relation $$VU=e^{2\pi i\theta}UV,$$ defines an algebra structure on $\mathcal{A}_{\theta}$. Let $E=\mathcal{S}(\mathbb{R})$ be the Schwartz space of rapidly decreasing functions on $\mathbb{R}$. It is easily checked that the following formulas define an $\mathcal{A}_{\theta}$ module structure on $E$: $$(Uf)(x)=f(x+\theta), \quad (Vf)(x)=e^{-2 \pi i \theta}f(x).$$ It can shown that $E$ is finite and projective [@ac80]. This construction can be generalized [@ac80; @cr; @rie2]. Let $n, m$ be integers with $m>0$ and let $E_{n, m}=\mathcal{S}(\mathbb{R}\times \mathbb{Z}_m)$, where $\mathbb{Z}_m$ is the cyclic group of order $m$. The following formulas define an $\mathcal{A}_{\theta}$ module structure on $E_{n, m}$: $$\begin{aligned} (Uf)(x, j)&=&f(x+\theta -\frac{n}{m}, j-1),\\ (Vf)(x, j)&=&e^{2 \pi i (x-j\frac{n}{m})}f(x, j).\end{aligned}$$ It can be shown that when $n-m\theta \neq 0$, the module $E_{n, m}$ is finite and projective. In particular for irrational $\theta$ it is always finite and projective. For more examples of noncommutative vector bundles see [@cl]. Affine varieties and finitely generated commutative reduced algebras -------------------------------------------------------------------- In commutative algebra, [Hilbert’s Nullstellensatz]{} [@cg] states that the category of affine algebraic varieties over an algebraically closed field $\mathbb{F}$ is anti-equivalent to the category of finitely generated commutative reduced unital $\mathbb{F}$ algebras: $$\{ \text{\bf affine algebraic varieties}\} \simeq$$ $$\{\text{\bf finitely generated commutative reduced algebras}\}^{op}.$$ Recall that an [affine algebraic variety]{} (sometimes called an [algebraic set]{}) over a field $\mathbb{F}$ is a subset of an affine space $\mathbb{F}^n$ which is the set of zeros of a set of polynomials in $n$ variables over $\mathbb{F}$. A morphism between affine varieties $V \subset \mathbb{F}^n$ and $W\subset \mathbb{F}^m$ is a map $f: V \longrightarrow W$ which is the restriction of a polynomial map $\mathbb{F}^n \rightarrow \mathbb{F}^m.$ It is clear that affine varieties and morphisms between them form a category. A [reduced algebra]{} is by definition an algebra with no [nilpotent elements]{}, i.e. if $x^n=0$ for some $n$ then $x=0$. The above correspondence associates to a variety $V\subset \mathbb{F}^n$ its [coordinate ring]{} $\mathbb{F}[V]$ defined by $$\mathbb{F}[V]:=Hom_{Aff}(V, \{pt\})\simeq \mathbb{F}[x_1, \cdots, x_n]/I,$$ where $I$ is the [vanishing ideal]{} of $V$ defined by $$I=\{f\in \mathbb{F}[x_1, \cdots, x_n]; f(x)=0 \;\mbox{for all}\; x\in V\}.$$ Obviously $\mathbb{F}[V]$ is a finitely generated commutative unital reduced algebra. Moreover, Given a morphism of varieties $f: V \to W$, its pull-back defines an algebra homomorphism $f^: \mathbb{F}[W] \to \mathbb{F}[V]$. We have thus defined the contravariant [coordinate ring functor]{} from affine varieties to finitely generated reduced commutative unital algebras. Given a finitely generated commutative unital algebra $A$ with $n$ generators we can write it as $$A\simeq \mathbb{F}[x_1, \cdots, x_n]/I,$$ where the ideal $I$ is a [radical ideal]{} if and only if $A$ is a reduced algebra. Let $$V:=\{x \in \mathbb{F}^n; f(x)=0\; \mbox{for all} \; f\in I\},$$ denote the variety defined by the ideal $I$. The classical form of Nullstellensatz [@har] states that if $\mathbb{F}$ is algebraically closed and $A$ is reduced then $A$ can be recovered as the coordinate ring of the variety $V$: $$\mathbb{F}[V]\simeq A=\mathbb{F}[x_1, \cdots, x_n]/I.$$ This is the main step in showing that the coordinate ring functor is an anti-equivalence of categories. Showing that the functor is full and faithful is easier. In Section 6 we sketch a proof of this fact when $\mathbb{F}$ is the field of complex numbers. Affine schemes and commutative rings ------------------------------------ The above correspondence between finitely generated reduced commutative algebras and affine varieties in not an ideal result. One is naturally interested in larger classes of algebras, like algebras with nilpotent elements as well as algebras over fields which are not algebraically closed or algebras over arbitrary rings; this last case in particularly important in number theory. In general one wants to know what kind of geometric objects correspond to a commutative ring and how this correspondence goes. [Affine schemes]{} are exactly defined to address this question. We follow the exposition in [@har]. Let $A$ be a commutative unital ring. The [prime spectrum]{} (or simply the [spectrum]{}) of $A$ is a pair $(\mbox{Spec} A, \mathcal{O}_A)$ where $\mbox{Spec} A$ is a topological space and $\mathcal{O}_A$ is a sheaf of rings on $\mbox{Spec} A$ defined as follows. As a set $\mbox{Spec} A$ consists of all [prime ideals]{} of $A$ (an ideal $I\subset A$ is called [prime ]{} if for all $a, b$ in $A$, $ab \in A$ implies that either $a\in I$, or $b\in I$). Given an ideal $I\subset A$, let $V(I)\subset \mbox{Spec} A$ be the set of all prime ideals which contain $I$. We can define a topology on $\mbox{Spec} A$, called the [Zariski topology]{}, by declaring sets of the type $V(I)$ to be closed (this makes sense since the easily established relations $V(IJ)=V(I) \cup V(J)$ and $V(\sum I_i)= \cap V(I_i)$ show that the intersection of a family of closed sets is closed and the union of two closed sets is closed as well). One checks that $\mbox{Spec} A$ is always compact but is not necessarily Hausdorff. For each prime ideal $p\subset A$, let $A_p$ denote the [localization]{} of $A$ at $p$. For an open set $U\subset \mbox{Spec} A$, let $\mathcal{O}_A(U)$ be the set of all continuous sections $s: U \to \cup_{p\in U} A_p$. (By definition a section $s$ is called continuous if locally around any point $p\in U$ it is of the form $\frac{f}{g}$, with $g\notin p$). One checks that $\mathcal{O}_A$ is a sheaf of commutative rings on $\mbox{Spec} A$. Now $(\mbox{Spec} A, \mathcal{O}_A)$ is a so called [ringed space]{} and $A \mapsto (\mbox{Spec} A, \mathcal{O}_A)$ is functor called the [spectrum functor]{}. A unital ring homomorphism $f: A \to B$ defines a continuous map $f^: \mbox{Spec} B \to \mbox{Spec} A$ by $f^(p)= f^{-1}(p)$ for all prime ideals $p\subset B$. (note that if $I$ is a maximal ideal $f^{-1}(I)$ is not necessarily maximal. This is one of the reasons one considers, for arbitrary rings, the prime spectrum and not the maximal spectrum.) An [affine scheme]{} is a ringed space $(X, \mathcal{O})$ such that $X$ is homeomorphic to $\mbox{Spec} A$ for a commutative ring $A$ and $\mathcal{O}$ is isomorphic to $\mathcal{O}_A$. The spectrum functor defines an equivalence of categories $$\{\mbox{\bf affine schemes}\} \simeq \{\mbox{ \bf commutative rings}\}^{op} .$$ The inverse equivalence is given by the [global section functor]{} that sends an affine scheme to the ring of its global sections. In the same vein categories of modules over a ring can be identified with categories of sheaves of modules over the spectrum of the ring. Let $A$ be a commutative ring and let $M$ be an $A$-module. We define a sheaf of modules $\mathcal{M} $ over $\mbox{Spec} A$ as follows. For each prime ideal $p\subset A$, let $M_p$ denote the localization of $M$ at $p$. For any open set $U\subset \mbox{Spec} A$ let $\mathcal{M}(U)$ denote the set of continuous sections $s: U\longrightarrow \cup_p M_p$ (this means that $s$ is locally a fraction $\frac{m}{f}$ with $m\in M$ and $f\in A_p$). One can recover $M$ from $\mathcal{M}$ by showing that $M\simeq \Gamma \mathcal{M}$ is the space of global sections of $M$. The functors $M\mapsto \mathcal{M}$ and $\mathcal{M}\mapsto \Gamma \mathcal{M}$ define equivalence of categories [@har]: $$\{\mbox{\bf modules over $A$}\} \simeq \{\mbox{\bf quasi-coherent sheaves on Spec $A$}\}.$$ Compact Riemann surfaces and algebraic function fields ------------------------------------------------------ It can be shown that the category of compact Riemann surfaces is anti-equivalent to the category of algebraic function fields: $$\{ \text{\bf compact Riemann surfaces} \}\simeq \{ \text{\bf algebraic function fields}\}^{op}.$$ Recall that a [Riemann surface]{} is a complex manifold of complex dimension one. A morphism between Riemann surfaces $X$ and $Y$ is a holomorphic map $f: X \rightarrow Y$. An [algebraic function field]{} is a finite extension of the field $\mathbb{C}(x)$ of rational functions in one variable. A morphism of function fields is simply an algebra map. To a compact Riemann surface one associates the field $M(X)$ of meromorphic functions on $X$. For example the field of meromorphic functions on the Riemann sphere is the field of rational functions $\mathbb{C}(x)$. In the other direction, to a finite extension of $\mathbb{C}(x)$ one associates the compact Riemann surface of the algebraic function $p(z, w)=0.$ Here $w$ is a generator of the field over $\mathbb{C}(x)$. This correspondence is essentially due to Riemann. Sets and Boolean algebras ------------------------- Perhaps the simplest notion of space, free of any extra structure, is the concept of a set. In a sense set theory can be regarded as the geometrization of logic. There is a duality between the category of sets and the category of complete atomic Boolean algebras (see, e.g., M. Barr’s Acyclic Models, CRM Monograph Series, Vol 17, AMS publications, 2002): $$\{\text{\bf{sets}}\}\simeq \{ \text{\bf{complete atomic Boolean algebras}}\}^{op}.$$ Recall that a [Boolean algebra]{} is a unital ring $B$ in which $x^2=x$ for all $x$ in $B$. A Boolean algebra is necessarily commutative as can be easily shown. One defines an order relation on $B$ by declaring $x\leq y$ if there is an $y'$ such that $x=yy'$. It can be checked that this is in fact a partial order relation on $B$. An [Atom]{} in a Boolean algebra is an element $x$ such that $x>0$ and there is no $y$ with $0<y<x$. A Boolean algebra is [atomic]{} if every element $x$ is the supremos of all the atoms smaller than $x$. A Boolean algebra is [complete]{} if every subset has a supremum and infimum. A morphism of complete Boolean algebras is a unital ring map which preserves all infs and sups. (Of course any unital ring map between Boolean algebras preserves finite sups and infs). Now, given a set $S$ let $$B={\bf 2}^S= \{f: S \longrightarrow {\bf 2}\},$$ where ${\bf 2}:=\{0, 1\}.$ Note that $B$ is a complete atomic Boolean algebra. Any map $f: S \rightarrow T$ between sets defines a morphism of complete atomic Boolean algebras via pullback: $ f^(g):=g\circ f$, and $ S \mapsto {\bf 2}^S$ is a contravariant functor. In the opposite direction, given a Boolean algebra $B$, one defines its [spectrum]{} $\Omega (B)$ by $$\Omega (B)= Hom_{Boolean} (B, {\bf 2}),$$ where we now think of [2]{} as a Boolean algebra with two elements. It can be shown that the two functors that we have defined are anti-equivalences of categories, quasi-inverse to each other. Thus once again we have a duality between a certain category of geometric objects, namely sets, and a category of commutative algebras, namely complete atomic Boolean algebras. Noncommutative quotients ========================= In this section we recall the method of noncommutative quotients as advanced by Connes in [@acb]. This is a technique that allows one to replace “bad quotients” by nice noncommutative spaces, represented by noncommutative algebras. In some cases, like noncommutative quotients for group actions, the noncommutative quotient can be defined as a crossed product algebra. In general, however, noncommutative quotients are defined as groupoid algebras. In Section 3.1 we recall the definition of a groupoid together with its various refinements like topological, smooth and étale groupoids. In Section 3.2 we define the groupoid algebra of a groupoid and give several examples. An important concept is Morita equivalence of algebras. We treat both the purely algebraic theory as well as the notion of strong Morita equivalence for $C^$-algebras in Section 3.3. Finally noncommutative quotients are defined in Section 3.4. Groupoids ---------- A groupoid is a small category in which every morphism is an isomorphism. Let $\mathcal{G}$ be groupoid. We denote the set of objects of $\mathcal{G}$ by $\mathcal{G}^{(0)}$ and, by a small abuse of notation, the set of morphisms of $\mathcal{G}$ by $\mathcal{G}$. Every morphism has a [source]{}, has a [target]{} and has an [inverse]{}. They define maps, denoted by $s$, $t$, and $i$, respectively, $$s:\mathcal{G} \longrightarrow \mathcal{G}^{(0)}, \quad t:\mathcal{G} \longrightarrow \mathcal{G}^{(0)},$$ $$i: \mathcal{G} \longrightarrow \mathcal{G}.$$ Composition $\gamma_1 \circ \gamma_2$ of morphisms $\gamma_1 $ and $ \gamma_2$ is only defined if $ s(\gamma_1)=t(\gamma_2)$. Composition defines a map $$\circ: \mathcal{G}^{(2)}=\{(\gamma_1, \gamma_2); s(\gamma_1)= t(\gamma_2)\}\longrightarrow \mathcal{G}.$$ [Examples]{}\ 1. (Groups). Every group $G$ defines at least two groupoids in a natural way:\ 1.a Define a category $\mathcal{G}$ with one object \* and $$Hom_{\mathcal{G}}(, )=G,$$ where the composition of morphisms is simply the group multiplication. This is obviously a groupoid.\ 2.b Define a category $\mathcal{G}$ with $$obj \; \mathcal{G}=G, \quad Hom_{\mathcal{G}} (s, t)=\{g\in G; \; gsg^{-1}=t\}.$$ Again, with composition defined by group multiplication, $\mathcal{G}$ is a groupoid. 2\. (Equivalence relations) Let $\sim$ denote an equivalence relation on a set $X$. We define a groupoid $\mathcal{G}$, called the [graph of $\sim$]{}, as follows. Let $$\begin{aligned} obj \; \mathcal{G}= X, \quad Hom_{\mathcal{G}}(x, y)&=&{} \quad \mbox{if}\; x\sim y, \\ &=& \varnothing \quad \mbox{otherwise.}\end{aligned}$$ Note that the set of morphisms of $\mathcal{G}$ is identified with the graph of the relation $\sim$ in the usual sense: $$\mathcal{G}=\{ (x, y); \; x\sim y \}\subset X\times X.$$ Two extreme cases of this graph construction are particularly important. When the equivalence relation reduces to equality, i.e., $x\sim y$ iff $x=y$, we have $$\mathcal{G}= \Delta (X)=\{(x, x); \;x\in X\}.$$ On the other extreme when $x\sim y$ for all $x$ and $y$, we obtain the [groupoid of pairs]{} where $$\mathcal{G}= X\times X.$$ 3\. (Group actions). Example 1) can be generalized as follows. Let $$G \times X \longrightarrow X, \quad (g, x)\mapsto gx,$$ denote the action of a group $G$ on a set $X$. We define a groupoid $\mathcal{G}=X\rtimes G$, called the [transformation groupoid]{} of the action, as follows. Let $obj \;\mathcal{G} =X$, and $$Hom_{\mathcal{G}}(x, y)=\{g\in G; \; gx=y\}.$$ Composition of morphisms is defined via group multiplication. It is easily checked that $\mathcal{G}$ is a groupoid. Its set of morphisms can be identified as $$\mathcal{G}\simeq X \times G,$$ where the composition of morphisms is given by $$(gx, h) \circ (x, g)= (x, hg).$$ Note that Example 1.a corresponds to the action of a group on a point and example 1.b corresponds to the action of a group on itself via conjugation. As we shall see, one can not get very far with just discrete groupoids. To get really interesting examples like the groupoids associated to continuous actions of topological groups and to foliations, one needs to consider topological as well as smooth groupoids, much in the same way as one studies topological and Lie groups. A [topological groupoid]{} is a groupoid such that its set of morphisms $\mathcal{G}$, and set of objects $\mathcal{G}^{(0)}$ are topological spaces, and its composition, source, target and inversion maps are continuous. A special class of topological groupoids, called étale or r-discrete groupoids, are particularly convenient to work with. An [étale groupoid]{} is a topological groupoid such that its set of morphisms $\mathcal{G}$ is a locally compact topological space and the fibers of the target map $t:\mathcal{G}\to \mathcal{G}^{(0)}$ $$\mathcal{G}^x= t^{-1}(x), \quad x\in \mathcal{G}^{(0)},$$ are discrete. A [Lie groupoid]{} is a groupoid such that $\mathcal{G}$ and $\mathcal{G}^{(0)}$ are smooth manifolds, the inclusion $\mathcal{G}^{(0)} \to \mathcal{G}$ as well as the maps $s, t, i$ and the composition map $\circ$ are smooth, and $s$ and $t$ are submersions. This last condition will gaurantee that the domain of the composition map $\mathcal{G}^{(2)}=\{(\gamma_1, \gamma_2); s(\gamma_1)= t(\gamma_2)\}$ is a smooth manifold.\ [Examples:]{}\ 1. Let $G$ be a discrete group acting by homeomorphisms on a locally compact Hausdorff space $X$. The transformation groupoid $X\rtimes G$ is naturally an étale groupoid. If $G$ is a Lie group acting smoothly on a smooth manifold $X$, then the transformation groupoid $X\rtimes G$ is a Lie groupoid.\ 2. Let $(V, \mathcal{F})$ be a foliated manifold and let $T$ be a complete transversal for the foliation. This means that $T$ is transversal to the leaves of the foliation and each leaf has at least one intersection with $T$. One defines an (smooth) étale groupoid $\mathcal{G}$ as follows. The objects of $\mathcal{G}$ is the transversal $T$ with its smooth structure. For any two points $x$ and $y$ in $T$ let $Hom_{\mathcal{G}} (x, y)=\varnothing$ if $x$ and $y$ are not in the same leaf. When they are in the same leaf, say $L$, let $Hom_{\mathcal{G}} (x, y)$ denote the set of all continuous paths in $L$ connecting $x$ and $y$ modulo the equivalence relation defined by [holonomy]{}. It can be shown that $\mathcal{G}$ is an smooth étale groupoid (cf. [@acb] for details and many examples). Groupoid algebras ----------------- The notion of [groupoid algebra]{} of a groupoid is a generalization of the notion of [group algebra]{} (or [convolution algebra]{}) of a group and it reduces to group algebras for groupoids with one object. To define the groupoid algebra of a locally compact topological groupoid in general one needs the analogue of a Haar measure for groupoids called a Haar system. While we won’t recall its general definition here, we should mention that, unlike locally compact groups, an arbirtary locally compact groupoid need not have a Haar system [@ren]. For discrete groupoids as well as étale groupoids and Lie groupoids, however, the convolution product can be easily defined. We start by recalling the definition of the groupoid algebra of a discrete groupoid. As we shall see, in the discrete case the groupoid algebra can be easily described in terms of matrix algebras and group algebras. Let $\mathcal{G}$ be a discrete groupoid and let $$\mathbb{C} \mathcal{G}=\bigoplus_{\gamma \in \mathcal{G}} \mathbb{C} \gamma,$$ denote the vector space generated by the set of morphisms of $\mathcal{G}$ as its basis. The formula $$\begin{aligned} \gamma_1 \gamma_2 &=&\gamma_1 \circ \gamma_2, \quad \mbox{if} \; \gamma_1 \circ \gamma_2 \quad \mbox{is defined},\\ \gamma_1 \gamma_2&=&0, \quad \mbox{ otherwise,}\end{aligned}$$ defines an associative product on $\mathbb{C} \mathcal{G}$. The resulting algebra is called the [groupoid algebra]{} of the groupoid $\mathcal{G}$. Note that $\mathbb{C} \mathcal{G}$ is unital if and only if the set $\mathcal{G}^{(0)}$ of objects of $\mathcal{G}$ is finite. The unit then is given by $$1=\sum_{x \in \mathcal{G}^{0}} id_x.$$ An alternative description of the groupoid algebra $\mathbb{C} \mathcal{G}$ which is more appropriate for generalization to étale and topological groupoids is as follows. Note that $$\mathbb{C} \mathcal{G} \simeq \{ f: \mathcal{G} \to \mathbb{C}; \;f \mbox{ has finite support}\},$$ and the product is given by the [convolution product]{} $$(fg) (\gamma)=\sum_{\gamma_1 \circ \gamma_2 =\gamma}f(\gamma_1)g(\gamma_2).$$ Given an étale groupoid $ \mathcal{G}$, Let $$C_c(\mathcal{G})=\{ f: \mathcal{G} \to \mathbb{C}; \; \mbox{ f is continuous and has compact support} \}.$$ Under the above convolution product $C_c(\mathcal{G})$ is an algebra called the [convolution algebra]{} of $\mathcal{G}.$ Note that for any $\gamma$ the above sum is finite (why?) and the convolution of two functions with compact support has compact support. When the groupoid is the transformation groupoid of a group action, the groupoid algebra reduces to a [crossed product algebra]{}. We recall its general definition. Let $G$ be a discrete group, $A$ be an algebra and let $Aut (A)$ denote the group of automorphisms of $A$. An [action]{} of $G$ on $A$ is a group homomorphism $$\alpha :G\longrightarrow {\text Aut} (A).$$ Sometimes one refers to the triple $(A, G, \alpha )$ as a [noncommutative dynamical system]{}. We use the simplified notation $g(a):=\alpha (g)(a).$ The [crossed product]{} or [semidirect product]{} algebra $A\rtimes G$ is defined as follows. As a vector space $$A\rtimes G =A\otimes \mathbb{C}G.$$ Its product is defined by $$(a\otimes g)(b\otimes h)=ag(b)\otimes gh.$$ It is easily checked that endowed with the above product, $A\rtimes G$ is an associative algebra. It is unital if and only if $A$ is unital and $G$ acts by unital automorphisms. One checks that $A\rtimes G$ is the universal algebra generated by subalgebras $A$ and $\mathbb{C}G$ subject to the relation $$gag^{-1}=g(a),$$ for all $g$ in $G$ and $a$ in $A$. We need a $C^$-algebraic analogue of the above construction. In the above situation assume that $A$ is a $C^$-algebra and let $Aut (A)$ denote the group of $C^$-automorphisms of $A$. We define a pre-$C^$ norm on the algebraic crossed product $A\rtimes G$ as follows. We choose a faithful representation $$\pi : A\longrightarrow \mathcal{L}(H)$$ of $A$ on a Hilbert space $H$ and then define a faithful representation of the algebraic crossed product $A\rtimes G$ on the Hilbert space $l^2 (G, H)$ by $$(a\otimes g)(\psi)(h)= \pi(a)\psi(g^{-1}h).$$ The [reduced $C^$-crossed product]{} $A\rtimes_r G$ is by definition the completion of the algebraic crossed product with respect to the norm induced from the above faithful representation. It can be shown that this definition is independent of the choice of the faithful representation $\pi$ [@D].\ [Examples:]{}\ 1. It is not difficult to determine the general form of the groupoid algebra of a discrete groupoid. Let $$\mathcal{G}^0=\bigcup_i \mathcal{G}^0_i$$ denote the decomposition of the set of objects of $\mathcal{G}$ into its “connected components". By definition two objects $x$ and $y$ belong to the same connected component if there is a morphism $\gamma$ with source $(\gamma)=x$ and target $(\gamma)=y$. We have a direct sum decomposition of the groupoid algebra $\mathbb{C}\mathcal{G}$: $$\mathbb{C}\mathcal{G}\simeq \bigoplus_i \mathbb{C}\mathcal{G}_i.$$ Thus suffices to consider only groupoids with a connected set of objects. Choose an object $x_0\in \mathcal{G}^0$, and let $$G=Hom_{\mathcal{G}}(x_0, x_0)$$ be the [isotropy group]{} of $x_0$. Assume first that $G=\{1\}$ is the trivial group. For simplicity assume that $\mathcal{G}^0$ is a finite set with $n$ elements. In other words our groupoid $\mathcal{G}$ is the groupoid of pairs on a set of $n$ elements. Recall that $$\mathcal{G}=\{(i, j); \; i, j=1, \cdots n\}$$ with composition given by $$(l, k)\circ (j, i)= (l, i) \quad \mbox{if}\; k=j.$$ (Composition is not defined otherwise). We claim that $$\mathbb{C}\mathcal{G}\simeq M_n(\mathbb{C}).$$ Indeed it is easily checked that the map $$(i, j)\mapsto E_{i, j},$$ where $E_{i, j}$ denote the matrix units defines an algebra isomorphism. As is emphasized by Connes in the opening section of [@acb], this is in fact the way Heisenberg discovered matrices in the context of quantum mechanics [@hei]. In other words noncommutative algebras appeared first in quantum mechanics as a groupoid algebra! We recommend the reader carefully examine the arguments of [@hei] and [@acb]. In general, but still assuming that $\mathcal{G}^{(0)}$ is connected, it is easy to see that $$\mathbb{C}\mathcal{G}\simeq \mathbb{C} G\otimes M_n(\mathbb{C}),$$ where $G=Hom_{\mathcal{G}}(x_0, x_0)$ is the isotropy group of $\mathcal{G}$.\ 2. We look at groupoid algebras of certain étale groupoids.\ a. We start by an example from [@acb]: an étale groupoid defined by an equivalence relation. Let $$X=[0, 1]\times \{1\} \cup [0, 1]\times \{2\}$$ denote the disjoint union of two copies of the interval $[0, 1].$ Let $\sim$ denote the equivalence relation that identifies $(x, 1)$ in the first copy with $(x, 2) $ in the second copy for $0<x<1$. Let $\mathcal{G}$ denote the corresponding groupoid with its topology inherited from $X\times X$. It is clear that $\mathcal{G}$ is an étale groupoid. The elements of the groupoid algebra $C_c(\mathcal{G})$ can be identified as continuous matrix valued functions on $[0, 1]$ satisfying some boundary condition: $$C_c(\mathcal{G})=\{f: [0, 1]\to M_2(\mathbb{C}); \quad f(0)\; \mbox{and}\; f(1)\; \mbox{are diagonal} \}.$$ b\. Let $X$ be a locally compact Hausdorff space and assume a discrete group $G$ acts on $X$ by homemorphisms. The induced action on $A=C_c(X)$, the algebra of continuous $\mathbb{C}$-valued functions on $X$ with compact support, is defined by $$(gf)(x)=f(g^{-1}x)$$ for all $f\in A$, $g\in G$, and $x\in X$. Let $\mathcal{G}=X\rtimes G$ denote the transformation groupoid defined by the action of $G$ on $X$.\ Exercise: Show that we have an algebra isomorphism $$C_c(X\rtimes G)\simeq C_c(X)\rtimes G.$$ The groupoid $C^$-algebra $C^_r(\mathcal{G})$, though we have not defined it as such, turns out to be isomorphic with the reduced crossed product algebra $C_0(X)\rtimes_r G$.\ 3. Let $X$ be a locally compact Hausdorff space and let $\mathcal{G}$ denote the groupoid of pairs on $X$. Its groupoid $C^$-algebra $C^ \mathcal{G}$ is defined as follows. Let $\mu$ be a positive Borel measure on $X$. We define a convolution product and an $$-operation on the space of morphisms of $\mathcal{G}$, $C_c( X\times X)$, by $$\begin{aligned} fg(x, y)&=&\int_X f(x,z) g(z, y)d\mu (z),\\ f^(x, y)&=& \overline{f(y,x)}.\end{aligned}$$ These operations turn $C_c(X\times X)$ into an $$-algebra. Convolution product defines a canonical $$-representation of this algebra on $L^2(X, \mu)$ by $$C_c(X\times X)\longrightarrow \mathcal{L}(L^2(X, \mu)), \quad f\mapsto f-.$$ The integral operator associated to a continuous function with compact support is a compact operator and it can be shown that the completion of the image of this map is the space of compact operators on $L^2(X, \mu).$ Thus we have $$C^* \mathcal{G} \simeq \mathcal{K}(L^2 (X, \mu).$$ In the other extreme, for the equivalence relation defined by equality the groupoid $C^$-algebra is given by $$C^ \mathcal{G}\simeq C_0(X).$$ These two examples are continuous $C^$-analogues of our discrete example 1 above. Morita equivalence ------------------- ### Algebraic theory In this section algebra means an associative unital algebra over a commutative ground ring $k$. All modules are assumed to be unitary in the sense that the unit of the algebra acts as identity operator on the module. Let $A$ be an algebra. We denote by $\mathcal{M}_A$ the category of right $A$-modules. Algebras $A$ and $B$ are called [Morita equivalent]{} if there is an equivalence of categories $$\mathcal{M}_A\simeq \mathcal{M}_B.$$ In general there are many ways to define a functor $F: \mathcal{M}_A\to \mathcal{M}_B$. By a simple observation of Eilenberg-Watts, however, if $F$ preserves finite limits and colimits, then there exists a unique $A-B$ bimodule $X$ such that $$F(M)= M\otimes_A X, \quad \text{for all}\quad M\in \mathcal{M}_A.$$ Composition of functors obtained in this way simply correspond to the balanced tensor product of the defining bimodules. It is therefore clear that $A$ and $B$ are Morita equivalent if and only if there exists an $A-B$ bimodule $X $ and a $B-A$ bimodule $Y$ such that we have isomorphisms of bimodules $$X\otimes_B Y \simeq A, \quad Y\otimes_A X\simeq B,$$ where the $A$-bimodule structure on $A$ is defined by $a(b)c=abc$, and similarly for $B$. Such bimodules are called [invertible (or equivalence) bimodules]{}. Given an $A-B$ bimodule $X$, we define algebra homomorphisms $$A \longrightarrow End_B(X), \quad B\longrightarrow End_A(X),$$ $$a\mapsto L_a, \quad b\mapsto R_b,$$ where $L_a$ is the operator of left multiplication by $a$ and $R_b$ is the operator of right multiplication by $b$. In general it is rather hard to characterize the invertible bimodules. The following theorem is one of the main results of Morita: An $A-B$ bimodule $X$ is invertible if and only if $X$ is finite and projective both as a left $A$-module and as a right $B$-module and the natural maps $$A \rightarrow End_B(X), \quad B\rightarrow End_A(X),$$ are algebra isomorphisms. [Example]{}. Any unital algebra $A$ is Morita equivalent to the algebra $M_n(A)$ of $n\times n$ matrices over $A$. The $(A, M_n(A))$ equivalence bimodule is $X=A^n$ with obvious left $A$-action and right $M_n(A)$-action. This example can be generalized as follows. [Example]{}. Let $P$ be a finite projective left $A$-module and let $$B=End_A(P).$$ Then the algebras $A$ and $B$ are Morita equivalent. The equivalence $A-B$ bimodule is given by $X=P$ with obvious $A-B$ bimodule structure. As a special case, we obtain the following geometric example. [Example]{} Let $X$ be a compact Hausdorff space and $E$ a complex vector bundle on $X$. Then the algebras $A= C(X)$ of continuous functions on $X$ and $B= \Gamma (End (E))$ of global sections of the endomorphisms bundle of $E$ are Morita equivalent. In fact, in view of Swan’s theorem this is a special case of the last example with $P= \Gamma (E)$ the global sections of $E$. There are analogous results for real as well as quaternionic vector bundles. If $X$ happens to be a smooth manifold we can let $A$ to be the algebra of smooth functions on $X$ and $B$ be the algebra of smooth sections of $End(E)$. Given a category $\mathcal{C}$ we can consider the category $Fun (\mathcal{C})$ whose objects are functors from $\mathcal{C} \rightarrow \mathcal{C}$ and whose morphisms are natural transformations between functors. The [center]{} of a category $\mathcal{C}$ is by definition the set of natural transformations from the identity functor to itself: $$\mathcal{Z}(\mathcal{C}):=Hom_{Fun (\mathcal{C})} (Id,Id).$$ Equivalent categories obviously have isomorphic centers. Let $\mathcal{Z} (A)=\{a\in A; \,ab=ba\, \mbox{for all}\, b\in A\}$ denote the center of an algebra $A$. It is easily seen that for $\mathcal{C}=\mathcal{M}_A$ the natural map $$\mathcal{Z}(A) \rightarrow \mathcal{Z}(\mathcal{C}), \quad a\mapsto R_a,$$ where $R_a(m)=ma$ for any module $M$ and any $m\in M$, is one to one and onto. It follows that Morita equivalent algebras have isomorphic centers: $$A \overset{M.E.}{\sim}B \; \Rightarrow \; \mathcal{Z}(A)\simeq \mathcal{Z}(B).$$ In particular two commutative algebras are Morita equivalent if and only if they are isomorphic. We say that commutativity is not a [Morita invariant property]{}.\ Exercise: Let $A$ be a unital $k$-algebra. Show that there is a 1-1 correspondence between space of traces on $A$ and $M_n(A)$. Extend this fact to arbitrary Morita equivalent algebras. We will see in Section 4 that Morita equivalent algebras have isomorphic Hochschild and cyclic (co)homology groups. They have isomorphic algebraic $K$-theory as well. ### Strong Morita equivalence Extending the Morita theory to non-unital algebras and to topological algebras needs more work and is not an easy task. For (not necessarily unital) $C^$-algebras we have Rieffel’s notion of [strong Morita equivalence]{} that we recall below. For $C^$-algebras one is mostly interested in their $$-representations on a Hilbert space. Thus one must consider equivalence $A-B$ bimodules $X$ such that if $H$ is a Hilbert space and a right $A$-module, then $H\otimes_A X$ is a Hilbert space as well. This leads naturally to the concepts of Hilbert module and Hilbert bimodule that we recall below. Let $B$ be a not necessarily unital $C^$-algebra. A right [Hilbert module]{} over $B$ is a right $B$-module $X$ endowed with a $B$-valued inner product such that $X$ is complete with respect to its natural norm. More precisely, we have a sesquilinear map $$X\times X \longrightarrow B, \quad (x, y) \mapsto <x, y>,$$ such that for all $x, y$ in $X$ and $b$ in $B$ we have $$<x, y> =<y, x>^, \, <x, yb>=<x, y>b, \,\mbox{and}\, <x, x> >0 \,\mbox{for}\, x\neq 0.$$ It can be shown that $\\| x\\|:=\\| <x, x>\\|^{1/2}$ is a norm on $X$. We assume $X$ is complete with respect to this norm. Of course, for $B=\mathbb{C}$, a Hilbert $B$-module is just a Hilbert space. A very simple geometric example to keep in mind is the following. Let $M$ be a compact Hausdorff space and let $E$ be a complex vector bundle on $M$ endowed with a Hermitian inner product. One defines a Hilbert module structure on the space $X=\Gamma (E)$ of continuous sections of $E$ by $$<s, t>(m)= <s(m), t(m)>_m$$ for continuous sections $s$ and $t$ and $m\in M$. A morphism of Hilbert $B$-modules $X$ and $Y$ is a bounded $B$-module map $X\to Y$. Every bounded operator on a Hilbert space has an adjoint. This is not the case for Hilbert modules. (This is simply because, even purely algebraically, a submodule of a module need not have a complementary submodule). A bounded $B$-linear map $T: X\rightarrow X$ is called [adjointable]{} if there is a bounded $B$-linear map $T^: X\rightarrow X$ such that for all $x$ and $y$ in $X$ we have $$<Tx, y>=<x, T^y>.$$ Let $\mathcal{L}_B(X)$ denote the algebra of bounded adjointable $B$-module maps $X\rightarrow X$. It is a $C^$-algebra. Let $A$ and $B$ be $C^$-algebras. A [Hilbert $A-B$ bimodule]{} consists of a right Hilbert $B$-module $X$ and a $C^$ map $$A \longrightarrow \mathcal{L}_B(X).$$ $C^$-algebras $A$ and $B$ are called [strongly Morita equivalent]{} if there is a Hilbert $A-B$ bimodule $X$ and a Hilbert $B-A$ bimodule $Y$ such that we have isomorphisms of bimodules $$X\otimes_B Y \simeq A, \quad \quad Y\otimes_A X\simeq B.$$ (The tensor products are completions of algebraic tensor products with respect to thier natural pre-Hilbert module structures.) Two unital $C^$-algebras are strongly Morita equivalent if and only if they are Morita equivalent as algebras. Any $C^$-algebra $A$ is Morita equivalent to its [stabilization]{} $A\otimes \mathcal{K}$, where $\otimes$ is the $C^$ tensor product and $\mathcal{K}$ is the algebra of compact operators on a Hilbert space. Strongly Morita equivalent algebras have naturally isomorphic topological $K$-theory. (cf. [@gvf] for more details, and a proof of these statements or references to original sources.) Noncommutative quotients ------------------------- >From a purely set theoretic point of view, all one needs to form a quotient space $X/\sim$ is an equivalence relation $\sim $ on a set $X$. The equivalence relation however is usually obtained from a much richer structure by forgetting part of this structure. For example, $\sim$ may arise from an action of a group $G$ on $X$ where $x\sim y$ if and only if $gx=y$ for some $g$ in $G$ ([orbit equivalence]{}). Note that there may be, in general, many $g$ with this property. That is $x$ may be identifiable with $y$ in more than one way. Of course when we form the equivalence relation this extra information is lost. The key idea in dealing with bad quotients in Connes’ theory is to keep track of this extra information! We call, rather vaguely, this extra structure the [quotient data]{}. Now Connes’s dictum in forming noncommutative quotients can be summarized as follows: $$\mbox{\bf quotient data} \rightsquigarrow \mbox{\bf groupoid} \rightsquigarrow \mbox{\bf groupoid algebra},$$ where the noncommutative quotient is defined to be the groupoid algebra itself. Why is this a reasonable approach? The answer is that first of all, by a theorem of M. Rieffel (see below) when the classical quotient, defined by a group action, is a reasonable space, the algebra of continuous functions on the classical quotient is strongly Morita equivalent to the groupoid algebra. Now it is known that Morita equivalent algebras have isomorphic $K$-theory, Hochschild and cyclic (co)homology groups. Thus the topological invariants defined via noncommutative geometry are the same for the two constructions and no information is lost. For bad quotients there is no reasonable space but we think of the noncommutative algebra defined as a groupoid algebra as representing a noncommutative quotient space. Thanks to noncommutative geometry, tools like K-theory, K-homology, cyclic cohomology and the local index formula, etc., can be applied to great advantage in the study of these noncommutative spaces.\ [Example 1.]{}\ a) We start with a simple example from [@acb]. Let $X=\{a, b\}$ be a set with two elements and define an equivalence relation on $X$ that identifies $a$ and $b$, $a\sim b$: $$\overset{a}{\bullet} \leftrightsquigarrow \overset{b}{\bullet}$$ The corresponding groupoid here is the groupoid of pairs on the set $X$. By Example 1 in Section 3.3 its groupoid algebra is the algebra of 2 by 2 matrices $M_2(\mathbb{C})$. The identification is given by $$f_{aa}(a, a)+ f_{ab}(a, b)+f_{ba}(b, a)+f_{bb}(b, b)\mapsto \left( \begin{matrix} f_{aa} & f_{ab} \\ f_{ba} & f_{bb} \end{matrix} \right).$$ The algebra of functions on the classical quotient, on the other hand, is given by $$\{f: X \to \mathbb{C}; \, f(a)=f(b)\}\simeq \mathbb{C}.$$ Thus the classical quotient and the noncommutative quotient are Morita equivalent.\ $$M_2(\mathbb{C}) \xleftarrow{\text{noncommutative quotient}}\; \overset{a}{\bullet} \leftrightsquigarrow \overset{b}{\bullet} \; \xrightarrow{\text{classical quotient}} \mathbb{C}$$\ b) The above example can be generalized. For example let $X$ be a finite set with $n$ elements with the equivalence relation $x\sim y$ for all $x, y$ in $X$. The corresponding groupoid $\mathcal{G}$ is the groupoid of pairs and its groupoid algebra, representing the noncommutative quotient, is $$\mathbb{C}\mathcal{G} \simeq M_n(\mathbb{C}).$$ The algebra of functions on the classical quotient is given by $$\{f: X \to \mathbb{C}; \, f(a)=f(b)\; \mbox{for all a, b in $X$}\}\simeq \mathbb{C}.$$ Again the classical quotient is obviously Morita equivalent to the noncommutative quotient.\ c) Let $G$ be a group (not necessarily finite) acting on a finite set $X$. The algebra of functions on the classical quotient is $$C(X/G)=\{f: X \to \mathbb{C}; \, f(x)=f(gx) \, \mbox{for all}\, g \in G, x\in X\} \simeq \bigoplus_{\mathcal{O}} \mathbb{C},$$ where $\mathcal{O}$ denotes the set of orbits of $X$ under the action of $G$. The noncommutative quotient, on the other hand, is defined to be the groupoid algebra of the transformation groupoid $ \mathcal{G}=X\rtimes G$. Note that as we saw before this algebra is isomorphic to the crossed product algebra $C(X)\rtimes G$. From Section 3.3 we have, $$\mathbb{C}\mathcal{G}\simeq C(X)\rtimes G \simeq \bigoplus_{ i\in \mathcal{O}}G_i \otimes M_{n_i}(\mathbb{C}),$$ where $G_i$ is the isotropy group of the i-th orbit, and $n_i$ is the size of the i-th orbit. Comparing the classical quotient with the noncommutative quotient we see that:\ i) If the action of $G$ is free then $G_i=\{1\}$ for all orbits $i$ and therefore the two algebras are Morita equivalent: $$C(X/G)\simeq \bigoplus_{\mathcal{O}} \mathbb{C}\overset{M. E.}{\sim} \bigoplus_{ i\in \mathcal{O}} M_{n_i}(\mathbb{C})\simeq \mathbb{C}\mathcal{G}.$$ ii\) The information about the isotropy groups is not lost in the noncommutative quotient construction, while the classical quotient totally neglects the isotropy groups.\ d) Let $X$ be a locally compact Hausdorff space and consider the equivalence relation $x\sim y$ for all $x$ and $y$ in $X$. The corresponding groupoid is again the groupoid of pairs. It is a locally compact topological groupoid and its groupoid $C^$-algebra as we saw in Section 3.3 is the algebra of compact operators $\mathcal{K} (L^2(X, \mu))$. This algebra is obviously strongly Morita equivalent to the classical quotient algebra $\mathbb{C}$.\ [Example 2.]{} Let $\theta \in \mathbb{R}$ be a fixed real number. Consider the action of the group of integers $G=\mathbb{Z}$ on the unit circle $\mathbb{T}=\{z\in \mathbb{C}; \; \|z\|=1\}$ via [rotation by $\theta$]{}: $$(n, z)\mapsto e^{2\pi i n \theta}z.$$ For $\theta =\frac{p}{q}$ a rational number, the quotient space $\mathbb{T}/\mathbb{Z}$ is a circle and the classical quotient algebra $$``C(\mathbb{T}/\mathbb{Z})":=\{f \in C(\mathbb{T}); \; f(gz)=f(z), \mbox{ for all}\; g, z\} \simeq C(\mathbb{T}).$$ The noncommutative quotient $A_{\theta}=C(\mathbb{T})\rtimes_r \mathbb{Z}$, for any $\theta$, is the unital $C^$-algebra generated by two unitaries $U$ and $V$ subject to the relation $$VU=e^{2\pi i \theta}UV.$$ It can be shown that when $\theta$ is a rational number $A_{\theta}$ is isomorphic to the space of continuous sections of the endomorphism bundle $End (E)$ of a complex vector bundle $E$ over the 2-torus $\mathbb{T}^2$. Thus for $\theta$ rational, $A_{\theta}$ is Morita equivalent to $C(\mathbb{T}^2)$: $$C(\mathbb{T})\rtimes_r \mathbb{Z} \, \overset{M.E.}{\sim} \, C(\mathbb{T}^2).$$ If $\theta$ is an irrational number then each orbit is dense in $\mathbb{T}$ and the quotient space $\mathbb{T}/\mathbb{Z}$ has only two open set. It is an uncountable set with a trivial topology. In particular it is not Hausdorff. Obviously, a continuous function on the circle which is constant on each orbit is necessarily constant since orbits are dense. Therefore $$``C(\mathbb{T}/\mathbb{Z})" \simeq \mathbb{C}.$$ The noncommutative quotient $A_{\theta}$ in this case is a simple $C^$-algebra and is not Morita equivalent to $C(\mathbb{T}^2)$. Let $G$ be a discrete group acting by homeomorphisms on a locally compact Hausdorff space $X$. Recall that the action is called [free]{} if for all $g\neq e,$ we have $gx\neq x$ for all $x\in G$. The action is called [proper]{} if the map $G\times X \to X, (g, x)\mapsto gx$ is a proper map in the sense that the inverse image of a compact set is compact. One shows that when the action is free and proper the orbit space $X/G$ of a locally compact and Hausdorff space is again a locally compact and Hausdorff space. Similarly, if $X$ is a smooth manifold and the action is free and proper then there exists a unique smooth structure on $X/G$ such that the quotient map $X\to X/G$ is smooth. The following result of M. Rieffel [@rie] clarifies the relation between the classical quotients and noncommutative quotients for group actions: Assume $G$ acts freely and properly on a locally compact Hausdorff space $X$. Then we have a strong Morita equivalence between the $C^$-algebras $C_0(X/G)$ and $C_0(X)\rtimes_r G.$ Cyclic cohomology ================= Cyclic cohomology was discovered by Alain Connes in 1981 [@ac81]. One of Connes’ main motivations came from index theory on foliated spaces. The $K$-theoretic index of a transversally elliptic operator on a foliated manifold is an element of the $K$-theory group of a noncommutative algebra, called the foliation algebra of the given foliated manifold. Connes realized that to identify this class it would be desirable to have a noncommutative analogue of the Chern character with values in a, as yet unknown, cohomology theory for noncommutative algebras that would play the role of de Rham homology of smooth manifolds. Now to define a noncommutative de Rham theory for noncommutative algebras is a highly nontrivial matter. This is in sharp contrast with the situation in $K$-theory where extending the topological $K$-theory to Banach algebras is essentially a routine matter. Note that the usual algebraic formulation of de Rham theory starts with the module of Kaehler differentials and its exterior algebra which does not make sense for noncommutative algebras. Instead the answer was found by Connes by analyzing the algebraic structures hidden in [traces of products of commutators]{}. These expressions are directly defined in terms of an elliptic operator and its parametix and were shown, via an index formula, to give the index of the operator when paired with a $K$-theory class. Let us read what Connes wrote in the Oberwolfach conference notebook after his talk, summarizing his discovery and how he arrived at it [@ac81]:\ “The transverse elliptic theory for foliations requires as a preliminary step a purely algebraic work, of computing for a noncommutative algebra $\mathcal{A}$ the cohomology of the following complex: $n$-cochains are multilinear functions $\varphi (f^0, \cdots, f^n)$ of $f^0, \cdots ,f^n \in \mathcal{A}$ where $$\varphi (f^1, \cdots, f^0)=(-1)^n \varphi (f^0, \cdots, f^n)$$ and the boundary is $$\begin{aligned} b\varphi (f^0, \cdots,f^{n+1}) &=&\varphi (f^0f^1, \cdots,f^{n+1})-\varphi (f^0, f^1f^2,\cdots, f^{n+1})+\cdots \\& &+(-1)^{n+1} \varphi (f^{n+1}f^0, \cdots, f^n).\end{aligned}$$ The basic class associated to a transversally elliptic operator, for $\mathcal{A}=$ the algebra of the foliation is given by: $$\varphi (f^0, \cdots,f^n)=Trace (\varepsilon F[F, f^0][F, f^1]\cdots[F, f^n]), \quad f^i\in \mathcal{A}$$ where $$F=\left( \begin{matrix} 0& Q \\ P & 0 \end{matrix} \right), \quad \varepsilon= \left( \begin{matrix} 1& 0 \\ 0& -1 \end{matrix} \right),$$ and $Q$ is a parametrix of $P$. An operation $$S: H^n(\mathcal{A})\to H^{n+2}(\mathcal{A})$$ is constructed as well as a pairing $$K(\mathcal{A}) \times H(\mathcal{A})\to \mathbb{C}$$ where $K(\mathcal{A})$ is the algebraic $K$-theory of $A$. It gives the index of the operator from its associated class $\varphi$. Moreover $<e, \varphi>=<e, S \varphi>$ so that the important group to determine is the inductive limit $H_g=\underset{\to}{Lim} H^n(\mathcal{A})$ for the map $S$. Using the tools of homological algebra the groups $H^n(\mathcal{A}, \mathcal{A}^)$ of Hochschild cohomology with coefficients in the bimodule $\mathcal{A}^$ are easier to determine and the solution of the problem is obtained in two steps,\ 1) the construction of a map $$B: H^n(\mathcal{A}, \mathcal{A}^) \to H^{n-1}(\mathcal{A})$$ and the proof of a long exact sequence $$\cdots\to H^n(\mathcal{A}, \mathcal{A}^) \overset{B}{\to} H^{n-1}(\mathcal{A})\overset{S}{\to} H^{n+1}(\mathcal{A})\overset{I}{\to}H^{n+1}(\mathcal{A}, \mathcal{A}^)\to \cdots$$ where $I$ is the obvious map from the cohomology of the above complex to the Hochschild cohomology.\ 2) The construction of a spectral sequence with $E_2$ term given by the cohomology of the degree $-1$ differential $I\circ B$ on the Hochschild groups $H^n(\mathcal{A}, \mathcal{A}^)$ and which converges strongly to a graded group associated to the inductive limit.* This purely algebraic theory is then used. For $\mathcal{A}=C^{\infty}(V)$ one gets the de Rham homology of currents, and for the pseudo torus, i.e. the algebra of the Kronecker foliation, one finds that the Hochschild cohomology depends on the Diophantine nature of the rotation number while the above theory gives $H^0_g$ of dimension $1$, $H^1_g$ of dimension $2$, and $H^0_g$ of dimension $1$ as expected but from some remarkable cancellations". \ In a different direction, cyclic homology also appeared in the 1983 work of Tsygan [@tsy] and was used also, independently, by Loday and Quillen [@lq]. The Loday-Quillen-Tsygan theorem states that the cyclic homology of an algebra $A$ is the primitive part (in the sense of Hopf algebras) of the Lie algebra homology of the Lie algebra $gl(A)$ of stable matrices. Equivalently, the Lie algebra homology of $gl(A)$ is isomorphic with the exterior algebra over the cyclic homology of $A$ with dimension shifted by 1: $$H^{Lie}_{\bullet}(gl(A))\simeq \wedge (HC_{\bullet}(A)[-1]).$$ We won’t pursue this connection in these notes. In section 4.1 we recall basic notions of Hochschild (co)homology theory and give several computations. Theorems of Connes [@ac85] (resp. Hochschild-Kostant-Rosenberg [@hkr]) on the Hochschild cohomology of the algebra of smooth functions on a manifold (resp. algebra of regular functions on a smooth affine variety) are among the most important results of this theory. In Section 4.2 we define cyclic cohomology via Connes’ cyclic complex. The easiest approach, perhaps, to introduce the map $B$ is to introduce first a bicomplex called cyclic bicomplex. This leads to a new definition of cyclic (co)homology, a definition of the operator $B$, and a proof of the long exact sequence of Connes, relating Hochschild, and cyclic cohomology groups. A third definition of cyclic cohomology is via Connes’s $(b, B)$-bicomplex. The equivalence of these three definitions is established by explicit maps. Finally we recall Connes’ computation of the cyclic (co)homology of the algebra of smooth functions on a manifold, and Burghelea’s result on the cyclic homology of group rings. Hochschild (co)homology ----------------------- Hochschild cohomology of associative algebras was defined by G. Hochschild through an explicit complex in [@hoc]. This complex is a generalization of the standard complex for group cohomology. One of the original motivations was to give a cohomological criterion for separability of algebras as well as a classification of (simple types) of algebra extensions in terms of second cohomology. Once it was realized, by Cartan and Eilenberg [@ce], that Hochschild cohomology is an example of their newly discovered theory of derived functors, tools of homological algebra like resolutions became available. The Hochschild-Kostant-Rosenberg theorem [@hkr] and its smooth version by Connes[@ac85] identifies the Hochschild homology of the algebra of regular functions on a smooth affine variety or smooth functions on a manifold with differential forms and is among the most important results of this theory. Because of this result one usually thinks of the Hochschild homology of an algebra $A$ with coefficients in $A$ as a noncommutative analogue of differential forms on $A$. As we shall see later in this section Hochschild (co)homology is related to cyclic (co)homology through Connes’ long exact sequence. For this reason computing the Hochschild (co)homology is often the first step in computing the cyclic (co)homology of a given algebra. Let $A$ be an algebra and let $M$ be an $A$-bimodule. Thus $M$ is a left and right $A$-module and the two actions are compatible in the sense that $a(mb)=(am)b,$ for all $a,b$ in $A$ and $m$ in $M$. The [Hochschild cochain complex of $A$ with coefficients in $M$]{}, denoted $( C^{\bullet}(A, M), \delta )$, is defined as $$C^0(A, M)=M, \quad C^n(A, M)=Hom(A^{\otimes n}, M), \quad n\geq 1,$$ $$\begin{aligned} (\delta m)(a)&=&ma-am,\\ (\delta f)(a_1, \cdots, a_{n+1})&=& a_1f(a_2, \cdots, a_{n+1})+\sum_{i=1}^n (-1)^{i+1} f(a_1, \cdots, a_ia_{i+1},\cdots , a_{n+1})\\ & & + (-1)^{n+1}f(a_1, \cdots, a_{n})a_{n+1},\end{aligned}$$ where $m \in M=C^0(A, M)$, and $f \in C^n(A,M), n\geq 1$. One checks that $$\delta^2 =0.$$ The cohomology of the complex $(C^{\bullet}(A,M), \delta )$ is by definition the [Hochschild cohomology]{} of $A$ with coefficients in $M$ and will be denoted by $H^{\bullet} (A, M).$ Among all bimodules $M$ over an algebra $A$, the following two bimodules play an important role:\ 1) M=A, with bimodule structure $a(b)c=abc$, for all $a, b, c$ in $A$. The Hochschild complex $C^{\bullet} (A, A)$ is also known as the [deformation complex ]{}, or [Gerstenhaber complex]{} of $A$. It plays an important role in deformation theory of associative algebras pioneered by Gerstenhaber [@ger]. For example it is easy to see that $H^2(A, A)$ is the space of [infinitesimal deformations]{} of $A$ and $H^3(A, A)$ is the [space of obstructions]{} for deformations of $A$.\ 2) $M= A^=Hom (A, k)$ with bimodule structure defined by $$(afb)(c)=f(bca),$$ for all $a, b, c$ in $A$, and $f$ in $A^$. This bimodule is relevant to cyclic cohomology. Indeed as we shall see the Hochschild groups $H^{\bullet}(A, A^)$ and the cyclic cohomology groups $HC^{\bullet}(A)$ enter into a long exact sequence (Connes’s long sequence). Using the identification $$Hom(A^{\otimes n}, A^) \simeq Hom(A^{\otimes (n+1)}, k), \quad f\mapsto \varphi,$$ $$\varphi(a_0, a_1, \cdots ,a_n)=f(a_1, \cdots, a_n)(a_0),$$ the Hochschild differential $\delta$ is transformed into the differential $b$ given by $$\begin{aligned} b\varphi(a_0, \cdots ,a_{n+1})&=&\sum_{i=0}^n (-1)^i \varphi (a_0, \cdots a_ia_{i+1}, \cdots, a_{n+1})\\ & &+(-1)^{n+1}\varphi (a_{n+1} a_0, a_1, \cdots ,a_n).\end{aligned}$$ Thus for $n=0, 1, 2$ we have the following formulas for $b$: $$\begin{aligned} b\varphi (a_0, a_1)&=&\varphi (a_0a_1)-\varphi (a_1a_0),\\ b\varphi (a_0, a_1, a_2)&=&\varphi (a_0a_1, a_2)-\varphi (a_0, a_1a_2)+ \varphi (a_2a_0, a_1),\\ b\varphi (a_0, a_1, a_2, a_3)&=&\varphi (a_0a_1, a_2, a_3) -\varphi (a_0, a_1a_2, a_3)\\ & &+\varphi (a_0, a_1, a_2a_3) -\varphi (a_3a_0, a_1, a_2).\end{aligned}$$ We give a few examples of Hochschild cohomology in low dimensions.\ [Examples]{}\ 1. $n=0$. It is clear that $$H^0(A, M)=\{m\in M; ma=am \quad \text{for all} \quad a\in A\}.$$ In particular for $M=A^$, $$H^0(A, A^)=\{f: A \to k; \; f(ab)=f(ba)\; \mbox{ for all}\; a, b \in A\},$$ is the space of traces on $A$.\ Exercise: For $A=k[x, \frac{d}{dx}]$, the algebra of differential operators with polynomial coefficients, show that $H^0(A, A^)=0.$\ 2. $n=1$. A Hochschild [1-cocycle]{} $f \in C^1(A, M)$ is simply a [derivation]{}, i.e. a linear map $f:A \rightarrow M$ such that $$f(ab)=af(b) +f(a)b,$$ for all $a, b$ in $A$. A cocycle is a [coboundary]{} if and only if the corresponding derivation is [inner]{}, that is there exists $m$ in $M$ such that $f(a)=ma-am$ for all $a$ in $A$. Therefore $$H^1(A, M)=\frac{\text{derivations}}{\text{inner derivations}}$$ Sometimes this is called the space of [outer derivations]{} of $A$ to $M$.\ Exercise: 1) Show that any derivation on the algebra $C(X)$ of continuous functions on a compact Hausdorff space $X$ is zero. (Hint: If $f=g^2$ and $g(x)=0$ then $f'(x)=0$.)\ 2) Show that any derivation on the matrix algebra $M_n(k)$ is inner. (This was proved by Dirac in [@dir] where derivations are called [quantum differentials]{}.\ 3) Show that any derivation on the [Weyl algebra]{} $A=k[x, \frac{d}{dx}]$ is inner as well.\ 3. $n=2.$ We show, following Hochschild [@hoc], that $H^2(A, M)$ classifies [abelian extensions]{} of $A$ by $M$. Let $A$ be a unital algebra over a field $k$. By definition, an abelian extension is an exact sequence of algebras $$0\longrightarrow M\longrightarrow B \longrightarrow A\longrightarrow 0,$$ such that $B$ is unital, $M$ has trivial multiplication ($M^2=0$), and the induced $A$-bimodule structure on $M$ coincides with the original bimodule structure. Let $E(A,M)$ denote the set of isomorphism classes of such extensions. We define a natural bijection $$E(A, M) \simeq H^2(A, M)$$ as follows. Given an extension as above, let $s: A \rightarrow B$ be a linear splitting for the projection $B\rightarrow A$, and let $f:A \otimes A \rightarrow M$ be its [curvature]{} defined by, $$f(a, b)=s(ab)-s(a)s(b),$$ for all $a, b$ in $A$. One can easily check that $f$ is a Hochschild 2-cocycle and its class is independent of the choice of splitting $s$. In the other direction, given a 2-cochain $f:A \otimes A \rightarrow M$, we try to define a multiplication on $B=A\oplus M$ via $$(a,m)(a',m')= (aa', am'+ma'+f(a, a')).$$ It can be checked that this defines an associative multiplication if and only if $f$ is a 2-cocycle. The extension associated to a 2-cocycle $f$ is the extension $$0\longrightarrow M \longrightarrow A\oplus M \longrightarrow A\longrightarrow 0.$$ It can be checked that these two maps are bijective and inverse to each other. We show that Hochschild cohomology is a derived functor. Let $A^{op}$ denote the [opposite algebra]{} of $A$, where $A^{op}=A$ and the new multiplication is defined by $a.b:=ba$. There is a one to one correspondence between $A$-bimodules and left $A\otimes A^{op}$-modules defined by $$(a\otimes b^{op})m=amb.$$ Define a functor from the category of left $A\otimes A^{op}$ modules to $k$-modules by $$M\mapsto Hom_{A\otimes A^{op}}(A, M)=\{m\in M; ma=am \quad \text{for all} \quad a\in A\}=H^0(A, M).$$ To show that Hochschild cohomology is the derived functor of the functor $ Hom_{A\otimes A^{op}}(A, -)$, we introduce the [bar resolution]{} of $A$. It is defined by $$0\longleftarrow A\overset{b'}{\longleftarrow} B_1(A)\overset{b'}{\longleftarrow} B_2(A) \cdots,$$ where $B_n(A)=A\otimes A^{op}\otimes A^{\otimes n}$ is the free left $A\otimes A^{op}$ module generated by $A^{\otimes n}$. The differential $b'$ is defined by $$\begin{aligned} b'(a\otimes b\otimes a_1\otimes \cdots \otimes a_n)&=&aa_1\otimes b\otimes a_2 \cdots \otimes a_n\\ & &+\sum_{i=1}^{n-1} (-1)^i(a\otimes b\otimes a_1\otimes \cdots a_ia_{i+1}\cdots \otimes a_n)\\ & &+(-1)^{n}(a\otimes a_nb\otimes a_1\otimes \cdots \otimes a_{n-1}).\end{aligned}$$ Define the operators $s:B_n(A)\rightarrow B_{n+1}(A)$, $n\geq 0$, by $$s(a\otimes b\otimes a_1\otimes \cdots \otimes a_n)=1\otimes b\otimes a\otimes a_1\otimes \cdots \otimes a_n.$$ One checks that $$b's+sb'=id$$ which shows that $(B_{\bullet}(A), b')$ is acyclic. Thus $(B_{\bullet}(A), b')$ is a projective resolution of $A$ as an $A$-bimodule. Now, for any $A$-bimodule $M$ we have $$Hom_{A\otimes A^{op}} (B_{\bullet}(A), M)\simeq (C^{\bullet}(A, M), \delta),$$ which shows that Hochschild cohomology is a derived functor. One can therefore use resolutions to compute Hochschild cohomology groups. Here are a few exercises\ 1. Let $$A=T(V)=k\oplus V \oplus V^{\otimes 2}\oplus \cdots,$$ be the tensor algebra of a vector space $V$. Show that $$0\longleftarrow T(V) \overset{\delta}{\longleftarrow} T(V)\otimes T(V) \overset{\delta}{\longleftarrow} T(V)\otimes V \otimes T(V) \longleftarrow 0,$$ $$\delta (x\otimes y)=xy, \quad \delta (x\otimes v \otimes y)=xv \otimes y -x\otimes vy,$$ is a free resolution of $T(V)$. Conclude that $A$ has Hochschild cohomological dimension 1 in the sense that $H^n(A, M)=0$ for all $M$ and all $n\geq 2.$ Compute $H^0$ and $H^1$ [@lod].\ 2. Let $A=k[x_1, \cdots ,x_n]$ be the polynomial algebra in $n$ variables over a field $k$ of characteristic zero. Let $V$ be an $n$ dimensional vector space over $k$. Define a resolution of the form $$0\leftarrow A \leftarrow A\otimes A \leftarrow A\otimes V\otimes A \leftarrow \cdots A\otimes \wedge^i V\otimes A \cdots \leftarrow A \otimes \wedge^n V \otimes A \leftarrow 0$$ by tensoring resolutions in 1) above for one dimensional vector spaces. Conclude that for any symmetric $A$-bimodule $M$, $$H^i(A, M) \simeq M\otimes \wedge^i V, \quad i=0, 1,\cdots.$$ Before proceeding further let us recall the definition of the [Hochschild homology]{} of an algebra with coefficients in a bimodule $M$. The [Hochschild complex of $A$ with coefficients in $M$]{}, $(C_{\bullet}(A, M), \delta)$, is defined by $$C_0(A,M)=M, \quad \text{ and} \quad C_n(A,M)=M \otimes A^{\otimes n}, \; n=1, 2,\cdots$$ and the [Hochschild boundary]{} $\delta :C_n(A, M)\longrightarrow C_{n-1}(A, M)$ is defined by $$\begin{aligned} \delta (m\otimes a_1\otimes \cdots a_n)&=&m a_1\otimes a_1\cdots a_n + \sum_{i=1}^{n-1}(-1)^i m\otimes a_1\otimes a_ia_{i+1}\cdots a_n\\ & & +(-1)^n a_nm \otimes a_1\otimes \cdots a_n.\end{aligned}$$ The Hochschild homology of $A$ with coefficients in $M$ is, by definition, the homology of the complex $(C_{\bullet}(A, M), \delta)$. We denote this homology by $H_{\bullet}(A, M)$. It is clear that $$H_0(A, M)=M/[A, M],$$ where $[A, M]$ is the subspace of $M$ spanned by commutators $am-ma$ for $a$ in $A$ and $m$ in $M$. The following facts are easily established:\ 1. Hochschild homology $H_{\bullet}(A, M)$ is the derived functor of the functor $$A\otimes A^{op}-Mod \longrightarrow k-Mod, \quad M\mapsto A\otimes_{A\otimes A^{op}} M,$$ i.e. $$H_n(A, M)=Tor_n^{A\otimes A^{op}}(A, M).$$ For the proof one uses the bar resolution as before.\ 2. (Duality) Let $M^=Hom(M, k)$. It is an $A$-bimodule via $(afb)(m)=f(bma).$ One checks that the natural isomorphism $$Hom (A^{\otimes n}, M^)\simeq Hom (M\otimes A^{\otimes n}, k),\quad n=0, 1,\cdots$$ is compatible with differentials. Thus if $k$ is field of characteristic zero, we have $$H^{\bullet} (A, M^) \simeq (H_{\bullet}(A, M))^.$$ >From now on we denote by $HH^n(A)$ the Hochschild group $H^n(A, A^)$ and by $HH_n(A)$ the Hochschild group $H_n(A, A)$. For applications of Hochschild and cyclic (co)homology to noncommutative geometry, it is crucial to consider topological algebras, topological bimodules and continuous chains and cochains on them. For example while the algebraic Hochschild groups of the algebra of smooth functions on a smooth manifold are not known, its topological Hochschild (co)homology is computed by Connes as we recall below. We will give only a brief outline of the definitions and refer the reader to [@ac85; @acb] for more details. Let $A$ be a locally convex topological algebra and $M$ be a locally convex topological $A$-bimodule. Thus $A$ is a locally convex topological vector space and the multiplication map $A\times A \to A$ is continuous. Similarly $M$ is a locally convex topological vector space such that both module maps $A\times M \to M$ and $M\times A \to M$ are continuous. In the definition of continuous Hochschild homology one uses the [projective tensor product]{} $M\hat{\otimes}A^{\hat{\otimes}n}$ of locally convex spaces. The algebraic Hochschild boundary, being continuous, naturally extends to topological completions. For cohomology one should use [jointly continuous]{} multilinear maps $$\varphi : A\times \cdots \times A \to M.$$ With these provisions, the rest of the algebraic formalism remains the same and carries over to the topological set up. In using projective resolutions, one should use only those topological resolutions that admit a continuous linear splitting. This gaurantees that the comparison theorem for projective resolutions remain true in the continuous setting. We give a few examples of Hochschild (co)homology computations. In particular we shall see that group homology and Lie algebra homology are instances of Hochschild homology. We start by recalling the classical results of Connes [@ac85] and Hochschild-Kostant-Rosenberg [@hkr] on the Hochschild homology of smooth commutative algebras.\ [Example]{} (Commutative Algebras)\ Let $A$ be a commutative unital algebra over a ring $k$. We recall the definition of the [algebraic de Rham complex]{} of $A$. The module of 1-forms over $A$, denoted by $\Omega^1 A$, is defined to be a left $A$-module $\Omega^1 A$ with a universal derivation $$d: A\longrightarrow \Omega^1 A.$$ This means that any other derivation $\delta : A \rightarrow M$ into a left $A$-module $M$, uniquely factorizes through $d$. One usually defines $\Omega^1 A=I/I^2$ where $I$ is the kernel of the multiplication map $A\otimes A \rightarrow A$. Note that since $A$ is commutative this map is an algebra homomorphism. $d$ is defined by $$d(a)= a\otimes 1-1\otimes a \quad \text{mod} (I^2).$$ One defines the space of n-forms on $A$ as the n-th exterior power of the $A$-module $\Omega^1 A$: $$\Omega^n A: =\wedge^n_A \Omega^1A.$$ There is a unique extension of $d$ to a graded derivation $$d:\Omega^{\bullet}A \longrightarrow \Omega^{\bullet +1}A.$$ It satisfies the relation $d^2=0$. The [algebraic de Rham cohomology]{} of $A$ is defined to be the cohomology of the complex $(\Omega^{\bullet}A, d)$. Let $M$ be a symmetric $A$-bimodule. We compare the complex $(M\otimes_A \Omega^{\bullet}A, 0)$ with the Hochschild complex of $A$ with coefficients in $M$. Consider the [antisymmetrization map]{} $$\varepsilon_n: M\otimes_A \Omega^nA \longrightarrow M\otimes A^{\otimes n}, \quad n=0, 1, 2, \cdots,$$ $$\varepsilon_n (m\otimes da_1\wedge \cdots \wedge da_n)=\sum_{\sigma \in S_n} sgn (\sigma) m\otimes a_{\sigma (1)}\otimes \cdots \otimes a_{\sigma (n)},$$ where $S_n$ denotes the symmetric group on $n$ letters. We also have a map $$\mu_n :M\otimes A^{\otimes n} \longrightarrow M\otimes_A \Omega^n A, \quad n=0, 1, \cdots$$ $$\mu_n (m\otimes a_1\otimes \cdots a_n)= m\otimes da_1 \wedge \cdots \wedge da_n.$$ One checks that both maps are morphisms of complexes, i.e. $$\delta \circ \varepsilon_n =0, \quad \mu_n \circ \delta =0.$$ Moreover, one has $$\mu_n \circ \varepsilon_n =n! \, Id_n.$$ It follows that if $k$ is a field of characteristic zero then the antisymmetrization map $$\varepsilon_n: M\otimes_A \Omega^nA \longrightarrow H_n(A, M),$$ is an inclusion. For $M=A$ we obtain a natural inclusion $$\Omega^nA \longrightarrow HH_n(A).$$ The celebrated Hochschild-Kostant-Rosenberg theorem [@hkr] states that if $A$ is the algebra of regular functions on an smooth affine variety the above map is an isomorphism. Let $M$ be a smooth closed manifold and let $A=C^{\infty}(M)$ be the algebra of smooth complex valued functions on $M$. It is a locally convex (in fact, Frechet) topological algebra. Fixing a finite atlas on $M$, one defines a family of seminorms $$p_n(f)=sup \{\|\partial^{I}(f)\|; \; \|I\|\leq n\},$$ where the supremum is over all coordinate charts. It is easily seen that the induced topology is independent of the choice of atlas. In [@ac85], using an explicit resolution, Connes shows that the canonical map $$HH_n^{cont}(A) \to \Omega^n M, \quad f_0\otimes \cdots \otimes f_n \mapsto f_0df_1 \cdots df_n,$$ is an isomorphism. In fact the original, equivalent, formulation of Connes in [@ac85] is for continuous Hochschild cohomology $HH^n (A)$ which is shown to be isomorphic to the continuous dual of $\Omega^n M$ (space of n-dimensional [de Rham currents]{}).\ [Example]{} (Group Algebras)\ It is clear from the original definitions that group (co)homology is an example of Hochschild (co)homology. Let $G$ be a group and let $M$ be a left $G$-module over the ground ring $k$. Recall that the standard complex for computing group cohomology [@lod] is the complex $$M \overset{\delta}{\longrightarrow} C^1(G, M)\overset{\delta}{\longrightarrow} C^2(G, M) \overset{\delta}{\longrightarrow} \cdots,$$ where $$C^n(G, M)=\{f: G^n \longrightarrow M \}.$$ The differential $\delta $ is defined by $$(\delta m )(g)=gm-m,$$ $$\begin{aligned} \delta f(g_1, \cdots ,g_{n+1})&=&g_1f(g_2, \cdots, g_{n+1}) +\sum_{i=1}^n (-1)^if(g_1, \cdots g_ig_{i+1}, \cdots g_{n+1})\\ & & +(-1)^{n+1} f(g_1, g_2, \cdots, g_n).\end{aligned}$$ Let $A=kG$ denote the group algebra of the group $G$ over $k$. Then $ M$ is a $kG$-bimodule via the two actions $$g.m =g(m), \quad m.g=m,$$ for all $g$ in $G$ and $m$ in $M$. It is clear that for all $n$, $$C^n(kG, M)\simeq C^n(G, M),$$ and the two differentials are the same. It follows that the cohomology of $G$ with coefficients in $M$ coincides with the Hochschild cohomology of $kG$ with coefficients in $M$.\ [Example]{} (Enveloping Algebras).\ We show that Lie algebra (co)homology is an example of Hochschild (co)homology. Let $\mathfrak{g}$ be a Lie algebra and $M$ be a $\mathfrak{g}$-module. This simply means that we have a Lie algebra morphism $$\mathfrak{g} \longrightarrow End_k(M).$$ The [Lie algebra homology]{} of $\mathfrak{g}$ with coefficients in $M$ is the homology of the [Chevalley-Eilenberg complex]{} defined by $$M \longleftarrow M\otimes \wedge^1 \mathfrak{g} \longleftarrow M\otimes \wedge^2 \mathfrak{g} \longleftarrow M \cdots,$$ where the differential is defined by $$\delta (m\otimes X)= X(m),$$ $$\begin{aligned} \delta (m\otimes X_1 \wedge X_2\wedge \cdots \wedge X_n)&= & \sum_{i<j} (-1)^{i+j} m\otimes [X_i, X_j]\wedge X_1 \cdots \wedge \hat{X_i}\cdots \hat{X_j}\cdots \wedge X_n\\ &+ & \sum_i (-1)^iX_i(m) \otimes X_1 \wedge \cdots \hat{X_i}\wedge \cdots \wedge X_n.\end{aligned}$$ One checks that $\delta^2 =0.$ Let $U(\mathfrak{g})$ denote the enveloping algebra of $\mathfrak{g}$. Given a $\mathfrak{g}$ module $M$ we define a $U(\mathfrak{g})$-bimodule $M'=M$ with left and right $U(\mathfrak{g})$-actions defined by $$X\cdot m= X(m), \quad m\cdot X=0.$$ Define a map $$\varepsilon_n : C_n^{Lie}(\mathfrak{g}, M) \longrightarrow C_n(U(\mathfrak{g}), M'),$$ $$\varepsilon_n (m\otimes X_1\wedge \cdots \wedge X_n)=\sum_{\sigma \in S_n} sgn (\sigma) m\otimes X_{\sigma (1)}\otimes \cdots \otimes X_{\sigma (n)}.$$ One checks that $\varepsilon$ is a chain map (prove this!). We claim that it is a quasi-isomorphism, i.e. it induces isomorphism on homology. To prove this, we define a filtration on $(C_{\bullet}(U(\mathfrak{g}), M), \delta)$ using the Poincare-Birkhoff-Witt filtration on $U(\mathfrak{g})$. The associated $E^1$ term is the de Rham complex of the symmetric algebra $S(\mathfrak{g})$. The induced map is the antisymmetrization map $$\varepsilon_n : M\otimes \wedge^n \mathfrak{g} \to M\otimes S(\mathfrak{g})^{\otimes n}.$$ By Hochschild-Kostant-Rosenberg’s theorem, this map is a quasi-isomorphism hence the original map is a quasi-isomorphism.\ [Example]{} (Morita Invariance of Hochschild (Co)Homology)\ Let $A$ and $B$ be unital Morita equivalent $k$-algebras. Let $X$ be an equivalence $A-B$ bimodule and $Y$ its inverse bimodule. Let $M$ be an $A-A$ bimodule and $N=Y\otimes_A M\otimes_A X$ the corresponding $B$-bimodule. Morita invariance of Hochschild homology states that there is a natural isomorphism $$H_n(A, M) \simeq H_n(B,N),$$ for all $n\geq 0$ [@lod]. We sketch a proof of this result for the special case where $B=M_k(A)$ is the algebra of $k$ by $k$ matrices over $A$. Let $M$ be an $A$-bimodule, and let $M_k(M)$ be the space of $k$ by $k$ matrices with coefficients in $M$. It is a bimodule over $M_k(A)$. The [generalized trace map]{} is defined by $$Tr: C_n(M_k(A), M_k(M))\longrightarrow C_n(A, M),$$ $$\begin{aligned} Tr( \alpha_0 \otimes m_0 \otimes \alpha_1\otimes a_1 \otimes \cdots \otimes \alpha_n \otimes a_n ) &=&\\ tr (\alpha_0 \otimes \alpha_1 \otimes \cdots \alpha_n) m_0\otimes a_1 \otimes \cdots a_n, & &\end{aligned}$$ where $\alpha_i \in M_k(k), a_i\in A, m_0 \in M,$ and $tr:M_k(k) \longrightarrow k$ is the standard trace of matrices.\ Exercise:\ 1. Show that $Tr$ is a chain map.\ 2. Let $i: A \rightarrow M_k(A)$ be the map that sends $a$ in $A$ to the matrix with only one non-zero component in the upper left corner equal to $a$. There is a similar map $M \rightarrow M_k(M)$. Define a map $$i_: C_n(A, M) \longrightarrow C_n(M_k(A), M_k(M)),$$ $$i_ (m\otimes a_1 \otimes \cdots \otimes a_n)=i(m)\otimes i(a_1) \otimes \cdots \otimes i(a_n).$$ Show that $$Tr\circ i_* =id.$$ It is however not true that $i_* \circ Tr =id.$ There is instead a homotopy between $i_* \circ Tr$ and $id$. The homotopy is given in [@lod] and we won’t reproduce it here. As a special case of the Morita invariance theorem, we have an isomorphism of Hochschild homology groups $$HH_n(A)=HH_n(M_k(A)),$$ for all $n$ and $k$. We need to know, for example when defining the noncommutative Chern character map, that inner automorphisms act by identity on Hochschild homology and inner derivations act by zero. Let $A$ be an algebra, let $u\in A$ be an invertible element and let $a\in A$ be any element. They induce the chain maps $$\Theta: C_n(A) \to C_n(A) \quad a_0\otimes \cdots \otimes a_n \mapsto ua_0u^{-1}\otimes \cdots \otimes ua_nu^{-1},$$ $$L_a :C_n(A) \to C_n(A) \quad a_0\otimes \cdots \otimes a_n \mapsto \sum_{i=0}^na_0\otimes \cdots \otimes [a, a_i]\otimes \cdots \otimes a_n.$$ $\Theta$ induces the identity map on Hochschild homology and $L_a$ induces the zero map. The maps [@lod], $h_i: A^{\otimes n+1} \to A^{\otimes n+2}, \; i=0, \cdots, n$ $$h_i(a_0\otimes \cdots \otimes a_n)=(a_0u^{-1}\otimes ua_1u^{-1}, \cdots, u\otimes a_{i+1}\cdots \otimes a_n)$$ define a homotopy $$h=\sum_{i=0}^n (-1)^i h_i$$ betweem $id$ and $\Theta$. For the second part one checks again that the maps $h_i: A^{\otimes n+1} \to A^{\otimes n+2}, \; i=0, \cdots, n$, $$h_i(a_0\otimes \cdots \otimes a_n)=(a_0\otimes \cdots a_i\otimes a\cdots \otimes a_n),$$ define a homotopy between $L_a$ and $0$ [@lod]. Cyclic (co)homology ------------------- ### Connes’ cyclic complex Cyclic cohomology was first defined by Connes [@ac81; @ac85] through a remarkable subcomplex of the Hochschild complex called the [cyclic complex]{}. Let $k$ be a field of characteristic zero and let $(C^{\bullet}(A), b)$ denote the Hochschild complex of a $k$-algebra $A$ with coefficients in the $A$-bimodule $A^$. We have $$C^n(A)=Hom (A^{\otimes (n+1)}, k), \quad n=0, 1, \cdots,$$ $$\begin{aligned} (b f)(a_0, \cdots ,a_{n+1})&=&\sum_{i=0}^n(-1)^if(a_0, \cdots,a_ia_{i+1},\cdots ,a_{n+1})\\ & &+(-1)^{n+1}f(a_{n+1}a_0, \cdots,\cdots ,a_{n}).\end{aligned}$$ An $n$-cochain $f \in C^n(A)$ is called [cyclic]{} if $$f(a_n, a_0, \cdots ,a_{n-1})=(-1)^n f(a_0, a_1, \cdots, a_n)$$ for all $a_0, \cdots ,a_n$ in $A$. We denote the space of cyclic cochains on $A$ by $C^n_{\lambda}(A).$ The space of cyclic cochains is invariant under the action of $b$, i.e. for all $n$, $$b \,C^n_{\lambda} (A) \subset C^{n+1}_{\lambda} (A).$$ Define the operators $\lambda :C^n(A) \rightarrow C^n(A)$ and $b': C^n(A) \rightarrow C^{n+1}(A)$ by $$\begin{aligned} (\lambda f)(a_0, \cdots ,a_n)&=&(-1)^n f(a_n, a_0, \cdots ,a_{n-1}),\\ (b' f)(a_0, \cdots ,a_{n+1})&=&\sum_{i=0}^n(-1)^if(a_0, \cdots,a_ia_{i+1},\cdots ,a_{n+1}).\end{aligned}$$ One checks that $$(1-\lambda)b=b'(1-\lambda).$$ Since $$C^n_{\lambda} (A)=Ker (1-\lambda),$$ the lemma is proved. We therefore have a subcomplex of the Hochschild complex, called the [cyclic complex]{} of $A$: $$C_{\lambda}^0 (A)\overset{b}{\longrightarrow} C_{\lambda}^1 (A) \overset{b}{\longrightarrow}C^2_{\lambda}(A) \overset{b}{\longrightarrow}\cdots .$$ The cohomology of this complex is called the [cyclic cohomology]{} of $A$ and will be denoted by $HC^n (A)$, $n=0, 1, 2,\cdots.$ A cocycle for cyclic cohomology is called a [cyclic cocycle]{}. It satisfies the two conditions: $$(1-\lambda )f=0, \quad {\text and} \quad bf=0.$$ The inclusion of complexes $$(C^{\bullet}_{\lambda}(A), b) \longrightarrow (C^{\bullet}(A), b),$$ induces a map $I$ from the cyclic cohomology of $A$ to the Hochschild cohomology of $A$ with coefficients in the $A$-bimodule $A^$: $$I: HC^n(A) \longrightarrow HH^n(A), \quad n=0, 1, 2, \cdots.$$ We shall see that this map is part of a long exact sequence relating Hochschild and cyclic cohomology, called Connes’ long exact sequence. For the moment we mention that $I$ need not be injective (see example below).\ [Examples]{}\ 1. Let $ A=k$ be a field of characteristic zero. We have $$C_{\lambda}^{2n}(k)\simeq k, \quad C_{\lambda}^{2n+1}(k)=0.$$ The cyclic complex reduces to $$0\longrightarrow k\longrightarrow 0\longrightarrow k \longrightarrow \cdots.$$ It follows that for all $n\geq 0,$ $$HC^{2n}(k)=k, \quad HC^{2n+1}(k)=0.$$ Since $HH^n(k)=0$ for $n\geq 1$, we conclude that the map $I$ need not be injective and the cyclic complex is not a retraction of the Hochschild complex.\ 2. It is clear that $HC^0(A)=HH^0(A)$ is the space of traces on $A$.\ 3. Let $A=C^{\infty}(M)$ be the algebra of smooth functions on a closed smooth manifolds $M$ of dimension $n$. One checks that $$\varphi (f_0, f_1, \cdots , f_n)=\int_M f_0df_1 \cdots df_n,$$ is a cyclic $n$-cocycle on $A$. In fact the Hochschild cocycle property $b\varphi =0$ is a consequence of the graded commutativity of the algebra of differential forms and the cyclic property $(1-\lambda)\varphi =0$ follows from Stokes formula. This example can be generalized in several directions. For example, Let $V$ be an $m$-dimensional closed singular chain (a cycle) on $M$, e.g. $V$ can be a closed $m$-dimensional submanifold of $M$. Then integration on $V$ defines an $m$-dimensional cyclic cocycle on $A$: $$\varphi (f_0, f_1, \cdots , f_m)=\int_V f_0df_1 \cdots df_m.$$ We obtain a map $$H_m(M, \mathbb{C}) \longrightarrow HC^m(A), \quad m=0, 1, \cdots,$$ from singular homology (or its equivalents) to cyclic cohomology. More generally, let $C$ be a [closed $m$-dimensional de Rham current]{} on $M$. Thus $C: \Omega^m M \to \mathbb{C}$ is a continuous linear functional on $\Omega^m M$ such that $dC (\omega):= C(d\omega)=0$ for all $\omega \in \Omega^{m-1} M.$ Then one checks that the cochain $\varphi$ defined by $$\varphi (f_0, f_1, \cdots , f_m)=< C, f_0df_1 \cdots df_m>,$$ is a cyclic cocycle on $A$. A noncommutative generalization of this procedure involves the notion of a [cycle on an algebra]{} due to Connes [@ac85] that we recall now. It gives a geometric and intuitively appealing presentation for cyclic cocycles. It also leads to a definition of cup product in cyclic cohomology and the $S$ operator. Let $$\Omega =\Omega^0 \oplus \Omega^1 \oplus \Omega^1 \oplus \Omega^2 \cdots$$ be a differential graded algebra, where we assume that $\Omega^0 =A$ and the differential $d: \Omega^i \rightarrow \Omega^{i+1}$ increases the degree. $d$ is a graded derivation in the sense that $$d(\omega_1 \omega_2)= d(\omega_1)\omega_2 +(-1)^{deg (\omega_1)}\omega_1 d(\omega_2), \quad \text{and} \quad d^2=0.$$ A [closed graded trace]{} of dimension $n$ on $\Omega$ is a linear map $$\int : \Omega^n \longrightarrow k$$ such that $$\int d\omega =0, \quad \text{and} \int (\omega_1 \omega_2-(-1)^{deg (\omega_1) deg (\omega_2)} \omega_2 \omega_1)=0,$$ for all $\omega$ in $\Omega^{n-1}$, $\omega_1$ in $\Omega^i$, $\omega_2$ in $\Omega^j$ and $i+j=n.$ A triple of the form $(A, \Omega, \int)$ is called a [cycle]{} over the algebra $A$. Given a closed graded trace $\int$ on $A$, one defines a cyclic $n$-cocycle on $A$ by $$\varphi (a_0, a_1, \cdots ,a_n)=\int a_0da_1 \cdots da_n.$$ Exercise: Check that $\varphi$ is a cyclic $n$-cocycle. Conversely, one can show that any cyclic cocycle on $A$ is obtained in this way. To do this we introduce the algebra $(\Omega A, d)$ of [noncommutative differential forms]{} on $A$ as follows. $\Omega A$ is the universal (nonunital) differential graded algebra generated by $A$ as a subalgebra. We have $\Omega^0 A=A$, and $\Omega^n A$ is linearly generated over $k$ by expressions $a_0 da_1 \cdots da_n$ and $ da_1 \cdots da_n$ for $a_i \in A$ (cf. [@ac85] for details). The differential $d$ is defined by $$d(a_0 da_1 \cdots da_n)=da_0 da_1 \cdots da_n, \quad \text{and} \quad d ( da_1 \cdots da_n)=0.$$ Now it is easily checked that the relation $$\varphi (a_0, a_1, \cdots ,a_n)=\int a_0da_1 \cdots da_n,$$ defines a 1-1 correspondence $$\{\text{cyclic n-cocycles on $A$}\} \simeq \{\text{ closed graded traces on $\Omega^n A$}\}.$$ Exercise: Give a similar description for Hochschild cocycles $\varphi \in C^n(A, A^).$\ 3. (From group cocycles to cyclic cocycles)\ Let G be a discrete group and let $c(g_1, \cdots, g_n)$ be a group $n$-cocycle on $G$. Assume $c$ is [normalized]{} in the sense that $$c(g_1, \cdots , g_n)=0,$$ if $g_i=e$ for some $i$, or if $g_1g_2\cdots g_n=e$. One checks that $$\begin{aligned} \varphi_c(g_0, \cdots , g_n)&=&c(g_1, \cdots ,g_n) \quad if \quad g_0g_1\cdots g_n=e,\\ &=& 0 \quad \mbox{otherwise},\end{aligned}$$ is a cyclic n-cocycle on the group algebra $kG$ [@ac83; @acb]. In this way one obtains a map $$H^n(G, k) \longrightarrow HC^n(kG), \quad c\mapsto \varphi_c.$$ By a theorem of D. Burghelea [@bur] (see below), the cyclic cohomology group $HC^n (kG)$ decomposes over the conjugacy classes of $G$ and the component corresponding to the conjugacy class of the identity is exactly the group cohomology $H^n(G, k)$.\ 4. (From Lie algebra cocycles to cyclic cocycles)\ Let $\mathfrak{g}$ be a Lie algebra acting by derivations on an algebra $A$. This means we have a Lie algebra map $$\mathfrak{g}\rightarrow Der(A, A),$$ from $\mathfrak{g}$ into the Lie algebra of derivations on $A$. Let $\tau : A\rightarrow k$ be an [invariant trace]{} on $A$. Thus $\tau$ is a trace on $A$ and $$\tau (X(a))=0 \quad \text{ for all $X \in \mathfrak{g}, \, a \in A$}.$$ For each $n\geq 0$, define a map $$\wedge^n g \longrightarrow C^n(A), \quad c\mapsto \varphi_c$$ $$\varphi_c(a_0, a_1, \cdots a_n)=\sum_{\sigma \in S_n}sgn (\sigma)\tau (a_0 X_{\sigma (1)}(a_1)\cdots X_{\sigma (n)}(a_n)),$$ where $c=X_1 \wedge \cdots \wedge X_n.$\ Exercise: Check that 1) For any $c$, $\varphi_c$ is a Hochschild cocycle, i.e. $b\varphi_c =0$. 2) If $c$ is a Lie algebra cycle (i.e. if $\delta (c)=0$), the $\varphi_c$ is a cyclic cocycle. We therefore obtain, for each $n\geq 0$, a map $$\chi_{\tau}: H_n^{Lie}(\mathfrak{g}, k) \longrightarrow HC^n (A), \quad c\mapsto \varphi_c,$$ from the Lie algebra homology of $\mathfrak{g}$ with trivial coefficients to the cyclic cohomology of $A$ [@ac83]. In particular if $\mathfrak{g}$ is abelian then $H^{Lie}_n(\mathfrak{g})\simeq \wedge^n(\mathfrak{g})$ and we obtain a well defined map $$\wedge^n(\mathfrak{g}) \longrightarrow HC^n(A), \quad n=0, 1, \cdots.$$ Here is an example of this construction, first appeared in [@ac80]. Let $A=\mathcal{A}_{\theta}$ denote the “algebra of smooth functions" on the noncommutative torus. Let $X_1 =(1, 0), X_2=(0, 1)$. There is an action of the abelian Lie algebra $\mathbb{R}^2$ on $\mathcal{A}_{\theta}$ defined on generators of $\mathcal{A}_{\theta}$ by $$\begin{aligned} X_1(U)=U, &\quad & X_1(V)=0,\\ X_2(U)=0, & \quad & X_2(V)=V.\end{aligned}$$ The induced derivations on $\mathcal{A}_{\theta}$ are given by $$\begin{aligned} X_1(\sum a_{m, n}U^m V^n)&=& \sum ma_{m, n}U^m V^n,\\ X_2(\sum a_{m, n}U^m V^n)&=& \sum na_{m, n}U^m V^n.\end{aligned}$$ It is easily checked that the trace $\tau$ on $\mathcal{A}_{\theta}$ defined by $$\tau (\sum a_{m, n}U^mV^n)=a_{0,0},$$ is invariant under the above action of $\mathbb{R}^2$. The generators of $H^{Lie}_{\bullet}(\mathbb{R}^2, \mathbb{R})$ are: $1, X_1, X_2, X_1\wedge X_2$. We therefore obtain the following 0-dimensional, 1-dimensional and 2-dimensional cyclic cocycles on $\mathcal{A}_{\theta}$: $$\begin{aligned} \varphi_0 (a_0)&=& \tau (a_0),\\ \varphi_1(a_0, a_1)&=& \tau(a_0 X_1 (a_1)), \quad \varphi'_1 (a_0, a_1)=\tau (a_0 X_2(a_1)),\\ \varphi_2(a_0, a_1, a_2)&=&\tau (a_0( X_1(a_1)X_2(a_2)-X_2(a_1)X_1(a_2))).\end{aligned}$$ It is shown by Connes [@ac85] that these classes generate the continuous periodic cyclic cohomology of $\mathcal{A}_{\theta}$.\ 5. (Cup product and the $S$-operation on cyclic cohomology)\ Let $(A, \Omega, \int)$ be an $m$-dimensional cycle on an algebra $A$ and $(B, \Omega', \int')$ and $n$-dimensional cycle on an algebra $B$. Let $\Omega \otimes \Omega'$ denote the (graded) tensor product of the differential graded algebras $\Omega$ and $\Omega'$. By definition, we have $$(\Omega \otimes \Omega')_n =\bigoplus_{i+j=n}\Omega_i \otimes \Omega'_j,$$ $$d(\omega \otimes \omega')=(d\omega) \otimes \omega' + (-1)^{deg (\omega)}\omega \otimes (d\omega'),$$ $$\int'' \omega \otimes \omega'= \int \omega \int' \omega', \quad \text{if} \quad \deg (\omega)=m, \; \deg (\omega')=n.$$ It is easily checked that $\int'' $ is a closed graded trace of dimension $m+n$ on $\Omega \otimes \Omega'$. Using the universal property of noncommutative differential forms, applied to the identity map $A\otimes B \longrightarrow \Omega_0 \otimes \Omega_0'$, one obtains a morphism of differential graded algebras $$(\Omega (A\otimes B), d) \longrightarrow (\Omega \otimes \Omega', d).$$ We therefore obtain a closed graded trace of dimension $m+n$ on $(\Omega(A\otimes B), d).$ In [@ac85], it is shown that the resulting map, called [cup product]{} in cyclic cohomology, $$\#: HC^m(A) \otimes HC^n(B) \to HC^{m+n}(A\otimes B)$$ is well defined.\ Exercise: By following the steps in the definition of the cup product, find an “explicit” formula for $\varphi \# \psi$, when $\varphi$ is an $m$-dimensional cyclic cocycle and $\psi$ is an $n$-dimensional cyclic cocycle [@ac85]. Let $\beta$ denote the generator of $HC^2 (k)$ defined by $\beta (1, 1, 1)=1$. Using the cup product with $\beta$ we obtain the $S$-map $$S: HC^n(A) \to HC^{n+2}(A), \quad \quad \varphi \mapsto \varphi \# \beta.$$ Exercise: Find explicit formulas for $S \varphi$ when $\varphi$ is a $0$ or $1$ dimensional cyclic cocycle.\ In the next section we give a different approach to $S$ via the cyclic bicomplex. The [generalized trace map]{} $$Tr : C_n^{\lambda} (A) \to C_n^{\lambda} (M_p (A)),$$ defined by $$(Tr \varphi )(a_0\otimes m_0, \cdots, a_n\otimes m_n)= tr (m_0\cdots m_n) \varphi (a_0, \cdots, a_n),$$ can be shown to be an example of cup product as well [@ac85]. Indeed we have $$Tr (\varphi)= \varphi \# tr.$$ So far we studied the cyclic cohomology of algebras. There is a “dual” theory called [cyclic homology]{} that we introduce now. Let $A$ be an algebra and for $n\geq 0$ let $C_n(A)= A^{\otimes (n+1)}$. For each $n\geq 0$, define the operators $b: C_n(A) \longrightarrow C_{n-1}(A)$, $b': C_n(A) \longrightarrow C_{n-1}(A),$ and $ \lambda: C_n(A) \longrightarrow C_{n}(A)$ by $$\begin{aligned} b(a_0\otimes \cdots \otimes a_n) & =& \sum_{i=0}^{n-1} (-1)^i( a_0\otimes \cdots a_i a_{i+1} \otimes a_n)\\ & +& (-1)^n(a_na_0\otimes a_1\cdots \otimes a_{n-1}),\\ b'(a_0\otimes \cdots \otimes a_n) & =& \sum_{i=0}^{n-1} (-1)^i( a_0\otimes \cdots a_i a_{i+1} \otimes a_n),\\ \lambda (a_0\otimes \cdots \otimes a_n) & =& (-1)^n( a_n\otimes a_{0} \cdots \otimes a_{n-1}).\end{aligned}$$ The relation $$(1-\lambda )b'=b(1-\lambda)$$ can be easily established. Let $(C_{\bullet}(A), b)$ denote the Hochschild complex of $A$ with coefficients in the $A$-bimodule $A$ and let $$C_n^{\lambda}(A):= C_n(A)/Im (1-\lambda).$$ The relation $ (1-\lambda )b'=b(1-\lambda)$ shows that the operator $b$ is well-defined on $C_{\bullet}^{\lambda}(A)$. The quotient complex $$(C_{\bullet}^{\lambda}(A), b)$$ is called [cyclic complex]{} of $A$. Its homology, denoted by $HC_n (A), n=0, 1, \cdots$, is called the [cyclic homology]{} of $A$.\ [Example]{} For $n=0$, $HC_0(A)=A/[A, A]$ is the [commutator quotient]{} of $A$. Here $[A, A]$ denotes the subspace of $A$ generated by the commutators $ab-ba$, for $a$ and $b$ in $A$. ### Connes’ long exact sequence Our goal in this section is to establish the long exact sequence of Connes relating Hochschild and cyclic homology groups. There is a similar sequence relating Hochschild and cyclic cohomology groups. Connes’ original proof in [@ac85] is based on the notion of [cobordism of cycles]{}. This leads to an operator $B: HH^n(A) \longrightarrow HC^{n-1}(A).$ He then shows that the three operators $I: HC^n(A) \longrightarrow HH^n(A)$, $S: HC^n (A) \longrightarrow HC^{n+2}(A)$ and $B$ fit into a long exact sequence. An alternative approach is based on the following bicomplex, called the [cyclic bicomplex]{} of $A$ and denoted by $\mathcal{C}(A)$ [@ac83; @lq; @lod]: $$\begin{CD} \vdots @.\vdots @.\vdots @.\\ A^{\otimes 3}@<1-\lambda<< A^{\otimes 3} @<N<< A^{\otimes 3}@<1-\lambda<< \dots \\ @VV bV @VV-b'V @VV bV \\ A^{\otimes 2}@<1-\lambda<< A^{\otimes 2} @<N<< A^{\otimes 2}@<1-\lambda<< \dots \\ @VV bV @VV-b'V @VV bV \\ A@<1-\lambda<< A @<N<< A@<1-\lambda<< \dots \end{CD}$$ Here the operator $N: A^{\otimes (n+1)} \longrightarrow A^{\otimes (n+1)}$ is defined by $$N=1+\lambda +\lambda^2 +\cdots +\lambda^n.$$ The relations $$\begin{aligned} N(1-\lambda)=(1-\lambda)N=0, & \quad &Nb=b'N, \\ (1-\lambda )b'=b(1-\lambda),& &\end{aligned}$$ can be easily verified. Coupled with relations $b^2=0,\, b'^2=0, $ it follows that $\mathcal{C}(A)$ is a bicomplex. In particular we can consider its total complex $(Tot \mathcal{C}(A), \delta).$ We show that there is a quasi-isomorphism of complexes $$(Tot \mathcal{C}(A), \delta) \overset{ q. i.}{\sim} (C^{\lambda}(A), b).$$ To this end, for $n\geq 0$, define a map $$\begin{aligned} (Tot_n \mathcal{C}(A), \delta) &\longrightarrow & (C^{\lambda}_{n}(A), b)\\ (x_0, x_1, \cdots ,x_n)&\mapsto & [x_n],\end{aligned}$$ where $[x_n]$ denotes the class of $x_n \in A^{\otimes (n+1)} $ in $C^{\lambda}_n(A)=A^{\otimes (n+1)}/Im (1-\lambda).$ One checks that this is a morphism of complexes. Now assume that $k$ is a field of characteristic zero. We show that the rows of $\mathcal{C}(A)$, i.e. the complexes $$A^{\otimes n+1}\overset{ 1-\lambda }{\longleftarrow} A^{\otimes n+1} \overset{ N }{\longleftarrow} A^{\otimes n+1} \cdots$$ are exact for each $n\geq 0$. The relation $$Ker (1-\lambda)=Im N$$ is obvious. To show that $Ker N= Im (1-\lambda)$, define the operator $$H=\frac{1}{n+1} (1+2 \lambda +3 \lambda^2 +\cdots + (n+1)\lambda^n): A^{\otimes n+1} \longrightarrow A^{\otimes n+1}.$$ We have $$(1-\lambda) H= N+1,$$ which shows that $Ker N= Im (1-\lambda).$ It follows that the spectral sequence converging to the total homology of $\mathcal{C}(A)$ collapses and the total homology is the homology of its $E^2$-term. But the $E^2$-term is exactly the cyclic complex $(C^{\lambda}_{\bullet}(A), b)$ of $A$. Alternatively, one can simply apply a “tic-tac-toe" argument to finish the proof. We can utilize the 2-periodicity of the cyclic bicomplex $\mathcal{C}(A)$ to define a short exact sequence of complexes $$0\longrightarrow Tot' \mathcal{C}(A) \longrightarrow Tot \mathcal{C}(A) \longrightarrow Tot \mathcal{C}(A) \longrightarrow 0,$$ where $Tot' \mathcal{C}(A)$ denotes the total complex of the first two columns and $Tot \mathcal{C}(A)[2]$ is the shifted by two 2 complex. The last map is defined by truncation: $$(x_0, x_1, \cdots ,x_n) \mapsto (x_0, \cdots, x_{n-2}).$$ The kernel of this map is the total complex of the first two columns of $\mathcal{C}(A)$: $$Tot'_n \mathcal{C}(A)= A^{\otimes n}\oplus A^{\otimes (n-1)}.$$ Now when $A$ is unital the $b'$-complex is acyclic. To prove this we define an operator $s: A^{\otimes n} \longrightarrow A^{\otimes n+1}$, $$s(a_0\otimes \cdots \otimes a_{n-1})=1\otimes a_0\otimes \cdots \otimes a_{n-1}.$$ One checks that $$b's+sb'=id,$$ which shows that the $b'$-complex is acyclic. We conclude that the complex $Tot'_{\bullet} \mathcal{C}(A)$ is homotopy equivalent to the Hochschild complex $(C_{\bullet}(A), b)$. Therefore the long exact sequence associated to the above short exact sequence is of the form: $$\cdots \longrightarrow HC_n(A) \overset{S}{\longrightarrow} HC_{n-2}(A) \overset{B}{\longrightarrow} HH_{n-1}(A) \overset{I}{\longrightarrow} HC_{n-1}(A) \longrightarrow \cdots.$$ This exact sequence was first obtained by Connes in 1981 [@ac81; @ac85] in its cohomological form: $$\cdots \longleftarrow HC^n(A) \overset{S}{\longleftarrow} HC^{n-2}(A) \overset{B}{\longleftarrow} HH^{n-1}(A) \overset{I}{\longleftarrow} HC^{n-1}(A) \longleftarrow \cdots.$$ The [periodicity operator]{} $$S: HC_n(A) \rightarrow HC_{n-2}(A)$$ is induced by the truncation map $$(x_0, x_1, \cdots ,x_n) \mapsto (x_0, \cdots, x_{n-2}).$$ The operator $B$ is the [connecting homomorphism]{} of the long exact sequence. It can therefore be represented on the level of chains by the formula $$B=Ns(1-\lambda): C_n(A) \rightarrow C_{n+1}(A).$$ Exercise: Recall how the connecting homomorphism is defined for the long exact sequence associated to a short exact sequence of complexes and drive the above formula for $B$. We defined, using the cup product with the generator of $HC^2 (k)$, an operation $S: HC^n(A) \longrightarrow HC^{n+2}(A)$, for $n=0,1, \cdots $. It can be shown that this definition coincides with the cohomological form of the above definition of $S$. Typical applications of Connes’ long exact sequence involve extracting information on cyclic homology from Hochschild homology. We list some of them:\ 1. Let $f:A \rightarrow A'$ be an algebra homomorphism and suppose that the induced maps on Hochschild groups $$f_: HH_n(A) \longrightarrow HH_n(A'),$$ are isomorphisms for all $n\geq 0$. Then $$f_: HC_n(A) \longrightarrow HC_n(A')$$ is an isomorphism for all $n\geq 0.$ This simply follows by comparing the $SBI$ sequences for $A$ and $B$ and applying the “five lemma". In particular, it follows that inner automorphisms act as identity on cyclic homology, while inner derivations act like zero on cyclic homology.\ 2. (Morita invariance of cyclic homology). Let $A$ and $B$ be Morita equivalent unital algebras. The Morita invariance property of cyclic homology states that there is a natural isomorphism $$HC_n(A)\simeq HC_n (B), \quad n=0, 1, \cdots$$ For a proof of this fact in general see [@lod]. In the special case where $B=M_k(A)$ a simple proof can be given as follows. Indeed, by Morita invariance property of Hochschild homology, we know that the inclusion $i: A \rightarrow M_k(A)$ induces isomorphism on Hochschild groups and therefore on cyclic groups by 1) above.\ The bicomplex $\mathcal{C}(A)$ can be extended to the left. We obtain a bicomplex, in the upper half plane, $$\begin{CD} \vdots @.\vdots @.\vdots @.\\ \dots @<N<< A^{\otimes 3}@<1-\lambda<< A^{\otimes 3} @<N<< A^{\otimes 3}@<1-\lambda<< \dots \\ @VV bV @VV-b'V @VV bV @VV -b'V\\ \dots @<N<<A^{\otimes 2}@<1-\lambda<< A^{\otimes 2} @<N<< A^{\otimes 2}@<1-\lambda<< \dots \\ @VV bV @VV-b'V @VV bV @VV-b'V\\ \dots @<N<< A@<1-\lambda<< A @<N<< A@<1-\lambda<< \dots \\ \end{CD}$$\ The total homology (where direct products instead of direct sums is used in the definition of the total complex) of this bicomplex is by definition the [periodic cyclic homology]{} of $A$ and is denoted by $HP_{\bullet}(A)$. Note that because of the 2-periodicity of this bicomplex $HP_{\bullet}(A)$ is 2-periodic.\ Exercise: Show that the resulting homology is trivial if instead of direct products we use direct sums.\ Similarly one defines the [periodic cyclic cohomology* ]{} $HP^{\bullet}(A)$, where this time one uses direct sums for the definition of the total complex. Since cohomology and direct limits commute, these periodic groups are indeed direct limits of cyclic cohomology groups under the $S$-map: $$HP^n(A)=\varinjlim HC^{n+2k}(A).$$ One checks, using the relations $$b's+sb'= 1,\, bN=Nb', \, (1-\lambda)b=b'(1-\lambda)$$ that $$B^2=0, \quad bB+Bb=0.$$ Indeed, we have $$\begin{aligned} B^2=Ns(1-\lambda) Ns(1-\lambda)&=&0,\\ bB+Bb=b Ns(1-\lambda) +Ns(1-\lambda)b&=&\\ Nb's(1-\lambda)+Nsb'(1-\lambda)&=&\\ N(1)(1-\lambda)&=&0.\end{aligned}$$ The relations $b^2=B^2=bB+Bb=0$ suggest the following new bicomplex for unital algebras, called [Connes’ (b, B)-bicomplex]{}. It was first defined in [@ac81; @ac85]. As we shall see, it leads to a third definition of cyclic (co)homology. Let $A$ be a unital algebra. The $(b, B)$-bicomplex of $A$ is the following bicomplex: $$\begin{CD} \vdots @.\vdots @.\vdots \\ A^{\otimes 3}@<B<< A^{\otimes 2}@<B<< A \\ @VVbV @VVbV\\ A^{\otimes 2}@<B<<A\\ @VVbV\\ A \end{CD}$$ The complexes $Tot \mathcal{B}(A)$ and $Tot \mathcal{C}(A)$ are homotopy equivalent. We define explicit chain maps between these complexes and show that they are chain homotopic via explicit homotopies. Define $$\begin{aligned} I: Tot \mathcal{B}(A) &\rightarrow & Tot \mathcal{C}(A), \quad I=id +sN\\ J: Tot \mathcal{C}(A) &\rightarrow & Tot \mathcal{B}(A), \quad J=id +Ns.\end{aligned}$$ One checks that $I$ and $J$ are chain maps. Now consider the operators $$\begin{aligned} g: Tot \mathcal{B}(A) &\rightarrow & Tot \mathcal{B}(A), \quad g=B_0s^2N\\ h: Tot \mathcal{C}(A) & \rightarrow & Tot \mathcal{C}(A), \quad h=s,\end{aligned}$$ where $B_0=(1-\lambda)s.$ We have, by direct computation: $$\begin{aligned} I\circ J&=& id + h \delta +\delta h,\\ I\circ J&=& id + g \delta' +\delta' g,\\\end{aligned}$$ where $\delta$ (resp. $\delta'$) denotes the differential of $Tot \mathcal{C}(A)$ (resp. $Tot \mathcal{B}(A)$). (cf. [@kha2] for details). The operator $I$ appears in [@lq]. The operator $h$ and the second homotopy formula was first defined in [@kha2]. It is clear that the above result extends to periodic cyclic (co)homology.\ [Examples]{}\ 1. (Algebra of smooth functions). Let $M$ be a closed smooth manifold, $A=C^{\infty}(M)$ denote the algebra of smooth complex valued functions on $M$, and let $(\Omega^{\bullet} M, d)$ denote the de Rham complex of $M$. We saw that, by a theorem of Connes, the map $$\mu : C_n(A) \rightarrow \Omega^n M, \quad \mu (f_0 \otimes \cdots \otimes f_n)= \frac{1}{n!}f_0df_1 \cdots df_n,$$ induces an isomorphism between the continuous Hochschild homology of $A$ and differential forms on $M$: $$HH_n(A)\simeq \Omega^n M.$$ To compute the continuous cyclic homology of $A$, we first show that under the map $\mu$ the operator $B$ corresponds to the de Rham differential $d$. More precisely, for each integer $n\geq 0$ we have a commutative diagram: $$\begin{CD} C_{n}(A)@> \mu >> \Omega^{n} M \\ @VVBV @VVdV\\ C_{n+1}(A)@>\mu >> \Omega^{n+1} M\\ \end{CD}$$\ We have $$\begin{aligned} \mu B (f_0\otimes \cdots \otimes f_n)&=&\mu \sum_{i=0}^n (-1)^{ni}(1\otimes f_i \otimes \cdots \otimes f_{i-1}-(-1)^n f_i \otimes \cdots f_{i-1} \otimes 1)\\ &= & \frac{1}{(n+1)!}\sum_{i=0}^n (-1)^{ni}df_i \cdots df_{i-1}\\ &= &\frac{1}{(n+1)!}(n+1)df_0 \cdots df_n\\ &=& d \mu (f_0\otimes \cdots \otimes f_n).\end{aligned}$$ It follows that $\mu$ defines a morphism of bicomplexes $$\mathcal{B}(A) \longrightarrow \Omega (A),$$ Where $\Omega (A)$ is the bicomplex $$\begin{CD} \vdots @.\vdots @.\vdots \\ \Omega^2M@<d<< \Omega^1M@<d<< \Omega^0M \\ @VV0V @VV0V\\ \Omega^1M@<d<<\Omega^0M\\ @VV0V\\ \Omega^0M \end{CD}$$ Since $\mu$ induces isomorphisms on row homologies, it induces isomorphisms on total homologies as well. Thus we have [@ac85]: $$HC_n(A)\simeq \Omega^n M/Im d \oplus H^{n-2}_{dR}(M) \oplus \cdots \oplus H^k_{dR} (M),$$ where k=0 if $n$ is even and $k=1$ if $n$ is odd. Using the same map $\mu$ acting between the corresponding periodic complexes, one concludes that the periodic cyclic homology of $A$ is given by $$HP_k (A)\simeq \bigoplus_i H^{2i+k}_{dR}(M), \quad k=0, 1.$$ 2\. (Group algebras). Let $kG$ denote the group algebra of a discrete group $G$ over a field $k$ of characteristic zero. By a theorem of Burghelea [@bur], Hochschild and cyclic homology groups of $kG$ decompose over the set of conjugacy classes of $G$ where each summand is the group homology (with trivial coefficients) of a group associated to a conjugacy class. We recall this result. Let $\widehat{G}$ denote the set of conjugacy classes of $G$, $G'$ be the set of conjugacy classes of elements of finite order, and let $G''$ denote the set of conjugacy classes of elements of infinite order. For an element $g \in G$, let $C_g=\{h\in G; \; hg=gh\}$ denote the centralizer of $g$, and let $W_g= C_g/<g>$, where $<g>$ is the group generated by $g$. Note that these groups depend, up to isomorphism, only on the conjugacy class of $g$. We denote the group homology with trivial coefficients of a group $K$ by $H_(K)$. The Hochschild homology of $kG$ is given by [@bur; @acb; @lod]: $$HH_n(kG)\simeq \bigoplus_{g\in \widehat{G}}H_n(C_g).$$ There is a similar, but more complicated, decomposition for the cyclic homology of $kG$: $$HC_n(kG)\simeq \bigoplus_{g\in G'}(\bigoplus_{i\geq 0}H_{n-2i}(W_g)) \bigoplus_{g\in G''}H_n(W_g).$$ In particular, the Hochschild group has $H_n(G)$ as a direct summand, while the cyclic homology group has $\oplus_i H_{n-2i}(G)$ as a direct summand (corresponding to the conjugacy class of the identity element of $G$). Chern-Connes character ======================= Recall that the classical [Chern character]{} is a natural transformation from $K$-theory to ordinary cohomology theory with rational coefficients [@ms]. More precisely for each compact Hausdorff space $X$ we have a natural homomorphism $$Ch: K^0(X)\longrightarrow \bigoplus_{i\geq 0} H^{2i}(X, \mathbb{Q}),$$ where $K^0$ (resp. $H$) denote the $K$-theory (resp. Cech cohomology with rational coefficients). It satisfies certain axioms and these axioms completely characterize $Ch$. But we won’t recall these axioms here since they are not very useful for finding the noncommutative analogue of $Ch$. It is furthermore known that $Ch$ is a rational isomorphism in the sense that upon tensoring it with $\mathbb{Q}$ we obtain an isomorphism $$Ch_{\mathbb{Q}}: K^0(X) \otimes \mathbb{Q} \overset{\sim}{\longrightarrow} \bigoplus_{i\geq 0} H^{2i}(X, \mathbb{Q}).$$ When $X$ is a smooth manifold there is an alternative construction of $Ch$, called the [Chern-Weil construction]{}, that uses the differential geometric notions of connection and curvature on vector bundles [@ms]. It goes as follows. Let $E$ be a complex vector bundle on $X$ and let $\nabla$ be a connection on $E$. Thus $$\nabla : C^{\infty}(E) \longrightarrow C^{\infty}(E)\otimes_A \Omega^1 X$$ is a $\mathbb{C}$-linear map satisfying the Leibniz condition $$\nabla (fs)=f \nabla (s) +s\otimes df,$$ for all smooth sections $s$ of $E$ and smooth functions $f$ on $X$. Let $$\hat{\nabla} : C^{\infty}(E)\otimes_A \Omega^{\bullet}X\longrightarrow C^{\infty}(E)\otimes_A \Omega^{\bullet +1} X,$$ denote the natural extension of $\nabla$ satisfying a graded Leibniz property. It can be easily shown that the [curvature operator]{} $\hat{\nabla}^2$ is an $\Omega^{\bullet}X$-linear map. Thus it is completely determined by its restriction to $C^{\infty}( E)$. This gives us the [curvature form]{} $$R \in C^{\infty}(End (E))\otimes \Omega^2 X$$ of $\nabla$. Let $$Tr: C^{\infty}( End (E))\otimes_A \Omega^{ev} X\to \Omega^{ev} X,$$ denote the canonical trace. The Chern character of $E$ is then defined to be the class of the non-homogeneous even form $$Ch(E)= Tr (e^{R}).$$ (We have omitted the normalization factor of $\frac{1}{2\pi i}$ to be multiplied by $R$.) One shows that $Ch(E)$ is a closed form and its cohomology class is independent of the choice of connection. In [@ac80; @ac85; @acb], Connes shows that this Chern-Weil theory admits a vast generalization. For example, for an algebra $A$ and each integer $n\geq 0$ there are natural maps, called [Chern-Connes character]{} maps, $$Ch^{2n}_0 : K_0(A) \longrightarrow HC_{2n}(A),$$ $$Ch^{2n+1}_1 : K_1(A) \longrightarrow HC_{2n+1}(A),$$ compatible with $S$-operation. Alternatively the noncommutative Chern character can be defined as a pairing between cyclic cohomology groups and $K$-theory called [Chern-Connes pairing]{}: $$HC^{2n}(A) \otimes K_0(A) \longrightarrow k,$$ $$HC^{2n+1}(A) \otimes K_1(A) \longrightarrow k.$$ These pairings are shown to be compatible with the periodicity operator $S$ in the sense that $$<[\varphi], [e]>=<S[\varphi], [e]>,$$ and thus induce a pairing between periodic cyclic cohomology and $K$-theory. We start by recalling the definition of these pairings. Let $\varphi (a_0,\cdots ,a_{2n})$ be an [even cyclic cocycle]{} on an algebra $A$. For each integer $k\geq 1$, let $$\tilde{\varphi}=tr\# \varphi \in C_{\lambda}^{2n}(M_k(A))$$ denote the extension of $\varphi$ to the algebra of $k\times k$ matrices on $A$. Note that $\tilde{\varphi}$ is a cyclic cocycle as well and is given by the formula $$\tilde{\varphi} (m_0\otimes a_0, \cdots, m_{2n}\otimes a_{2n})=tr(m_0\cdots m_{2n})\varphi (a_0, \cdots, a_{2n}).$$ Let $e \in M_k(A)$ be an idempotent. Define a bilinear map by the formula $$<[\varphi], [e]>= \tilde{\varphi} (e, \cdots,e).$$ Let us first check that the value of the pairing depends only on the cyclic cohomology class of $\varphi$ in $HC^{2n}(A)$. Suffices to assume $k=1$ (why?). Let $\varphi =b\psi$ with $\psi \in C_{\lambda}^{2n-1}(A)$, be a coboundary. Then we have $$\begin{aligned} \varphi (e, \cdots,e)&=&b\psi (e, \cdots, e)\\ &=&\psi (ee, e, \cdots,e) -\psi (e, ee, \cdots, e)+ \cdots +(-1)^{2n} \psi(ee, e, \cdots, e)\\ &=&\psi (e, \cdots,e)\\ &=&0,\end{aligned}$$ where the last relation follows from the cyclic property of $\psi$. To verify that the value of $<[\varphi], [e]>$, for fixed $\varphi$, only depends on the class of $[e] \in K_0(A)$ we have to check that for $u\in GL_k(A)$ an invertible matrix, we have $<[\varphi], [e]>=<[\varphi], [ueu^{-1}]>$. Again suffices to show this for $k=1$. But this is exactly the fact, proved in the last section, that inner automorphisms act by identity on cyclic cohomology. The formulas in the [odd case]{} are as follows. Given an invertible matrix $u\in M_k(A)$ and an odd cyclic cocycle $\varphi (a_0, \cdots ,a_{2n+1})$ on $A$, we have $$<[\varphi], [u]>= \tilde{\varphi}(u^{-1}-1, u-1, \cdots, u^{-1}-1, u-1).$$ Exercise: Show that the above formula defines a pairing $K_1(A) \otimes HC^{2n+1}(A) \to k.$ There are also formulas for Chern-Connes pairings when the cyclic cocycle is in the $(b, B)$ or cyclic bicomplex; but we won’t recall them here (cf. [@acb], [@lod]). There is an alternative infinitesimal proof" of the well-definement of these pairings which works for Banach (or certain classes of topological) algebras where elements of $K_0(A)$ can be defined as smooth homotopy classes of idempotents [@acb]: Let $e_t, 0\leq t\leq 1,$ be a smooth family of idempotents in a Banach algebra $A$. There exists an smooth family $x_t, 0\leq t\leq 1$ of elements of $A$ such that $$\overset{.}{e_t}:= \frac{d}{dt} (e_t)=[x_t, e_t], \quad \text{for} \quad 0\leq t\leq 1.$$ Let $$x_t=[\overset{.}{e_t}, e_t]=\overset{.}{e_t}e_t-e_t\overset{.}{e_t}.$$ Differentiating the idempotent condition $e_t^2=e_t$ with respect to $t$ we obtain $$\frac{d}{dt}(e_t^2)=\overset{.}{e_t}e_t +e_t\overset{.}{e_t}=\overset{.}{e_t}.$$ Multiplying this last relation on the left by $e_t$ we obtain $$e_t\overset{.}{e_t}e_t=0.$$ Now we have $$[x_t, e_t]=[\overset{.}{e_t}e_t-e_t\overset{.}{e_t}, e_t]= \overset{.}{e_t}e_t +e_t\overset{.}{e_t}=\overset{.}{e_t}.$$ It follows that if $\tau: A \rightarrow \mathbb{C}$ is a trace (= a cyclic zero cocycle), then $$\frac{d}{dt} <\tau , e_t>= \frac{d}{dt} \tau (e_t)=\tau (\overset{.}{e_t})= \tau ([x_t, e_t])=0.$$ So that the value of the pairing, for a fixed $\tau$, depends only on the homotopy class of the idempotent. This shows that the pairing $$\{\mbox{ traces on $A$}\} \times K_0 (A) \longrightarrow \mathbb{C}$$ is well-defined. This is generalized in Let $\varphi (a_0, \cdots ,a_{2n})$ be a cyclic $2n$-cocycle on $A$ and let $e_t$ be a smooth family of idempotents in $A$. Then the number $$<[\varphi], [e_t]>= \varphi (e_t, \cdots , e_t)$$ is constant in $t$. Differentiating with respect to $t$ and using the above Lemma, we obtain $$\begin{aligned} \frac{d}{dt}\varphi (e_t, \cdots, e_t)&=&\varphi (\overset{.}{e_t}, \cdots, e_t) +\varphi (e_t, \overset{.}{e_t}, \cdots, e_t)\\ & & \cdots +\varphi (e_t, \cdots, e_t, \overset{.}{e_t})\\ &=& \sum_{i=0}^{2n} \varphi (e_t, \cdots, [x_t, e_t], \cdots, e_t)\\ &=&L_{x_t} \varphi (e_t, \cdots, e_t).\end{aligned}$$ We saw that inner derivations act trivially on Hochschild and cyclic cohomology. This means that for each $t$ there is a cyclic cocchain $\psi_t$ such that the [Lie derivative]{} $L_{x_t} \varphi =b\psi_t$. We then have $$\frac{d}{dt}\varphi (e_t, \cdots, e_t)= (b\psi_t)( e_t, \cdots, e_t)=0.$$ Exercise: Repeat the above proof in the odd case. The formulas for the even and odd [Chern-Connes character]{} maps $$Ch^{2n}_0 : K_0(A) \longrightarrow HC_{2n}(A),$$ $$Ch^{2n+1}_1 : K_1(A) \longrightarrow HC_{2n+1}(A),$$ are as follows. In the even case, given an idempotent $e=(e_{ij})\in M_k(A)$, we have $$Ch^{2n}_0 (e)= Tr (\underbrace{e\otimes e\cdots \otimes e}_{2n+1})=\sum_{i_0,i_1, \cdots i_{2n}} e_{i_0 i_1}\otimes e_{i_1i_2}\otimes \cdots \otimes e_{i_{2n}i_{0}}.$$ In low dimensions we have $$\begin{aligned} Ch^0_0(e)&=&\sum_{i=1}^k e_{ii},\\ Ch^2_0(e)&=&\sum_{i_0=1}^k \sum_{i_1=1}^k \sum_{i_2=1}^ke_{i_0i_1}\otimes e_{i_1i_2}\otimes e_{i_2i_0}.\end{aligned}$$ In the odd case, given an invertible matrix $u\in M_k(A)$, we have $$Ch^{2n+1}_1 ([u])=Tr (\underbrace{(u^{-1}-1)\otimes (u-1)\otimes\cdots \otimes (u^{-1}-1)\otimes (u-1)}_{2n+2}).$$\ [Examples:]{}\ 1. For $n=0$, $HC^0(A)$ is the space of traces on $A$. Therefore the Chern-Connes pairing reduces to the map $$\{\mbox{traces on $A$}\} \times K_0(A) \longrightarrow k,$$ $$<\tau, [e]>=\sum_{i=1}^n \tau (e_{ii}),$$ where $e=[e_{ij}] \in M_n(A)$ is an idempotent. The induced function on $K_0 (A)$ is called the [dimension function]{} and denoted by $dim_{\tau}$. Here is an slightly different approach to this dimension function. Let $E$ be a finite projective right $A$-module. A trace $\tau$ on $A$ induces a trace on the endomorphism algebra of $E$, $$Tr: End_A(E) \longrightarrow k$$ as follows. First assume that $E=A^n$ is a free module. Then $End_A (E)\simeq M_n(A)$ and our trace map is defined by $$Tr (a_{i, j})=\sum a_{ii}.$$ It is easy to check that the above map is a trace. In general, there is an $A$-module $F$ such that $E\oplus F\simeq A^n$ is a free module and $End_A(E)$ embeds in $M_n(A)$. One can check that the induced trace on $End_A(E)$ is independent of the choice of splitting.\ Exercise: Since $E$ is finite and projectice, we have $End_A(E)\simeq E\otimes_A E^$. The induced trace is simply the canonical pairing between $E$ and $E^$. The [dimension function]{} associated to a trace $\tau$ on $A$ is the additive map $$dim_{\tau}: K_0(A) \longrightarrow k,$$ induced by the map $$dim_{\tau}(E)=Tr (id_E),$$ for any finite projective $A$-module $E$. It is clear that if $E$ is a vector bundle on a connected topological space $X$ and $\tau (f) =f(x_0)$, where $x_0 \in X$ is a fixed point, then $dim_{\tau} (E)$ is the rank of the vector bundle $E$ and is an integer. One of the striking features of noncommutative geometry is the existence of noncommutative vector bundles with non integral dimensions. A beautiful example of this phenomenon is shown by [Rieffel’s idempotent]{} $e\in \mathcal{A}_{\theta}$ with $\tau (e)=\theta$, where $\tau$ is the canonical trace on the noncommutative torus [@acb].\ 2. Let $A=C^{\infty}(S^1)$ denote the algebra of smooth complex valued functions on the circle. One knows that $K_1(A)\simeq K^1 (S^1)\simeq \mathbb{Z}$ and $u(z)=z$ is a generator of this group. Let $$\varphi (f_0, f_1)=\int_{S^1} f_0df_1$$ denote the cyclic cocycle on $A$ representing the fundamental class of $S^1$ in de Rham homology. We have $$<[\varphi], [u]>= \varphi (u, u^{-1})=\int_{S^1} udu^{-1}=-2\pi i.$$ Alternatively the Chern character $$Ch_1^1 ([u]) =u\otimes u^{-1}\in HC_1(A)\simeq H^1_{dR} (S^1),$$ is the class of the differential form $\omega = z^{-1}dz$, representing the fundamental class of $S^1$ in de Rham cohomology.\ 3. Let $A=C^{\infty} (S^2)$ and let $e\in M_2(A)$ denote the idempotent representing the Hopf line bundle on $S^2$: $$e= \frac{1}{2}\left( \begin{matrix} 1 + x_3 & x_1 + ix_2 \\ x_1-ix_2 & 1-x_3 \end{matrix} \right).$$ Let us check that under the map $$HC_2(A)\to \Omega^2 S^2, \quad a_0\otimes a_1 \otimes a_2 \mapsto a_0da_1 da_2,$$ the Chern-Connes character of $e$ corresponds to the fundamental class of $S^2$. We have $$Ch_0^2(e)=Tr (e\otimes e\otimes e) \mapsto Tr (edede)=$$ $$\frac{1}{8}Tr \left( \begin{matrix} 1 + x_3 & x_1 + ix_2 \\ x_1-ix_2 & 1-x_3 \end{matrix} \right) \left( \begin{matrix} dx_3 & dx_1 + idx_2 \\ dx_1-idx_2 & -dx_3 \end{matrix} \right) \left( \begin{matrix} dx_3 & dx_1 + idx_2 \\ dx_1-idx_2 & -dx_3 \end{matrix} \right).$$ Performing the computation one obtains $$Ch_0^2(e)\mapsto \frac{-i}{2}(x_1dx_2 dx_3-x_2dx_1 dx_3 +x_3dx_1 dx_2).$$ One can then integrate this 2-form on the two sphere $S^2$. The result is $-2\pi i$. In particular the class of $e$ in $K_0(A)$ is non-trivial, a fact which can not be proved using just $Ch_0^0(e)=Tr(e)$.\ 4. For smooth commutative algebras, the noncommutative Chern character reduces to the classical Chern character. We verify this only in the even case. Let $M$ be a smooth closed manifold and let $E$ be a complex vector bundle on $M$. Let $e\in C^{\infty}(M, M_n(\mathbb{C})$ be an idempotent representing the vector bundle $E$. One can check that the following formula defines a connection on $E$, called the Levi-Civita or Grassmanian connection: $$\nabla (eV)=edV,$$ where $V: M \to \mathbb{C}^n$ is a smooth function and $eV$ represents an arbitrary smooth section of $E$. Computing the curvature form we obtain $$R (eV)=\hat{\nabla}^2 (eV)= ede dV= edede. eV,$$ which shows that the curvature form is the “matrix valued 2-form" $$R=edede.$$ >From $e^2=e$, one easily obtains $ede.e=0.$ This implies that $$R^n=(edede)^n=e\underbrace{dede\cdots dede}_{2n}.$$ Under the canonical map $$HC_{2n}(A) \to H^{2n}_{dR} (M), \quad a_0\otimes \cdots \otimes a_{2n}\mapsto \frac{1}{(2n)!}a_0da_1 \cdots da_{2n},$$ we have $$Ch_{n}^0(e):=Tr(e\otimes \cdots \otimes e)\mapsto \frac{1}{(2n)!}Tr (edede\cdots de).$$ The classical Chern-Weil formula for $Ch(E)$ is $$Ch (E)= Tr (e^R)=Tr (\sum_{n\geq 0} \frac{R^n}{n!}).$$ So that its $n$ th component is given by $$Tr \frac{R^n}{n!}= \frac{1}{n!}Tr ((edede)^n)= \frac{1}{n!}Tr (ede\cdots de).$$ Banach and $C^$-algebras ========================= By an [algebra]{} in this book we mean an [associative algebra]{} over a commutative unital ground ring $k$. An algebra is called [unital]{} if there is a (necessarily unique) element $1 \in A$ such that $1a = a1 = a$ for all $a\in A$. It is called [commutative]{} if $ab=ba$ for all $a,b \in A$. Now let $k=\mathbb{R}$ or $\mathbb{C}$ be the field of real or complex numbers. A [norm]{} on a real or complex algebra $A$ is a map $$\\| \; \\| : A\to \mathbb{R},$$ such that for all $a, b$ in $A$ and $\lambda$ in $k$ we have: 1. $\\| a \\| \ge 0 , \; \text{and}\; \\| a \\| = 0 \; \text{iff} \; a=0,$ 2. $\\| a+b \\| \le \\| a \\| + \\| b \\|, $ 3. $\\| \lambda a \\| = \|\lambda \| \\| a \\|,$ 4. $\\| ab \\| \le \\| a \\|\\| b \\|.$ If $A$ is unital, we assume that $\\| 1 \\| = 1$. An algebra endowed with a norm is called a [normed algebra]{}. A [Banach algebra]{} is a normed algebra which is [complete]{}. Recall that a normed vector space $A$ is called complete if any Cauchy sequence in $A$ is convergent. One of the main consequences of completeness is that absolutely convergent series are convergent, i.e. if $ \sum_{n=1}^\infty \\| a_n \\|$ is convergent, then $\sum_{n=1}^\infty a_n$ is convergent in $A$. In particular the geometric series $\sum_{n=1}^{\infty}a_n $ is convergent if $\\| a \\|<1$. From this it easily follows that the group of invertible elements in a unital Banach algebra $A$ is an open subset of $A$. An [$\ast$-algebra]{} is a complex algebra endowed with an $\ast$-operation, i.e. a map $$\ast: A\to A, \hspace{20pt} a\mapsto a^\ast,$$ which is [anti-linear]{} and [involutive]{}: 1. $(a+b)^\ast = a^\ast + b^\ast , \; (\lambda a)^\ast =\bar{\lambda} a^\ast, $ 2. $(ab)^\ast = b^\ast a ^\ast,$ 3. $(a^\ast )^\ast = a, $ for all $a, b$ in $A$ and $\lambda $ in $\mathbb{C}$. A [Banach $\ast$-algebra]{} is a complex Banach algebra endowed with an $\ast$-operation such that for all $a\in A, \\| a^\ast \\| = \\| a \\|$. In particular for all $a$ in $A$ we have $$\\| a^\ast a \\| \le \\| a^\ast \\|\\| a \\| = \\| a \\|^2.$$ A [$C^\ast$-algebra]{} is a Banach $\ast$-algebra $A$ such that for all $a \in A$, $$\\| aa^\ast \\| = \\| a \\|^2.$$ We refer to this last identity as the [$C^$-identity]{}. For reasons that will become clear later in this section, $C^\ast$-algebras occupy a very special place among all Banach algebras. This should be compared with the role played by Hilbert spaces among all Banach spaces. In fact, as we shall see there is an intimate relationship between Hilbert spaces and $C^$-algebras thanks to the GNS construction and the Gelfand-Naimark embedding theorem. A [morphism]{} of $C^$-algebras is an algebra homomorphism $$f: A \longrightarrow B$$ between $C^$-algebras $A$ and $B$ such that $f$ preserves the $$ structure, i.e. $$f(a^)=f(a)^, \quad \text{for all} \; a\in A.$$ It can be shown that morphisms of $C^-$ algebras are [contractive]{} in the sense that for all $a \in A$, $$\\| f(a)\\| \le \\| a\\|.$$ In particular they are automatically continuous. It follows from this fact that the norm of a $C^$-algebra is unique in the sense that if $(A, \\|\, \\|_1)$ and $(A, \\|\, \\|_2)$ are both $C^-$algebras then $$\\|a \\|_1 =\\|a \\|_2,$$ for all $a\in A$. Note also that a morphism $f$ of $C^-$algebras is an [isomorphism]{} if an only if $f$ is one to one and onto. Isomorphisms of $C^$-algebras are necessarily [isometric]{}. In sharp distinction from $C^$-algebras, one can have different Banach algebra norms on the same algebra. For example on the algebra of $n\times n$ matrices one can have different Banach algebra norms and only one of them is a $C^$-norm.\ [Examples:]{}\ 1. (commutative $C^\ast$-algebras). Let $X$ be a locally compact Hausdorff space. We associate to $X$ several classes of algebras of functions on $X$ which are $C^\ast$-algebras. 2. Let $$C_0 (X) = \{f:X\to \mathbb{C} ; f\mbox{ is continuous and $f$ vanishes at $\infty$} \} \; .$$ By definition, $f$ vanishes at $\infty$ if for all $ \epsilon >0$ there exists a compact subset $K\subset X$ so that $\|f(x) \| < \epsilon $ for all $x \in X\setminus K$. Under pointwise addition and scalar multiplication $C_0(X)$ is obviously an algebra over the field of complex numbers $\mathbb{C}$. Endowed with the sup-norm $$\\| f \\| = \\| f \\|_\infty = \sup \{ \|f(x) \| ; \; x\in X \},$$ and $\ast$-operation $$f\mapsto f^\ast,\; f^\ast (x) = \bar{f}(x),$$ one checks that $C_0(X)$ is a commutative $C^\ast$-algebra. It is unital if and only if $X$ is compact. If $X$ is compact, we simply write $C(X)$ instead of $C_0(X)$. By a theorem of Gelfand and Naimark, any commutative $C^$-algebra is of the type $C_0(X)$ for some locally compact Hausdorff space $X$ (see below). 3. Let $$C_b (X) = \{ f:X\to \mathbb{C} ; \; f \mbox{ is continuous and bounded} \}.$$ Then with the same operations as above, $C_b (X)$ is a unital $C^\ast$-algebra. Note that $C_0 (X) \subset C_b (X)$ is an [essential ideal]{} in $C_b (X)$. (An ideal $I$ in an algebra $A$ is an essential ideal if for all $a$ in $A$, $aI = 0 \Rightarrow a=0.) $ 4. (commutative Banach algebras). It is easy to give examples of Banach algebras which are not $C^\ast$-algebras. For an integer $ n\geq 1$, let $$C^{n} [0,1] = \{ f: [0,1] \to \mathbb{C} ; \; f \in C^n \},$$ be the space of functions with continuous $n$-th derivative. We denote the $i$-th derivative of $f$ by $f^{(i)}$. With the norm $$\\| f \\|_n = \sum_{i=0}^n \frac{\\| f^{(i)} \\|_\infty}{i!}$$ and the $\ast$-operation $f^\ast (x) = \bar{f}(x)$, one checks that $C^{n}[0, 1]$ is a Banach $\ast$-algebra. It is, however, not a $C^\ast$-algebra as one can easily show that the $C^\ast$-identity fails. Note that for all $f \in C^{n} [0,1]$, $$\\| f \\|_\infty \le \\| f \\|_n \; .$$ 5. (noncommutative $C^\ast$-algebras). By a theorem of Gelfand and Naimark recalled below, any $C^\ast$-algebra can be realized as a closed $\ast$-subalgebra of the algebra of bounded operators on a complex Hilbert space. We start with the simplest examples: the algebra of complex $n$ by $n$ matrices. 6. Let $A=M_n(\mathbb{C})$ be the algebra of $n$ by $n$ matrices over the field of complex numbers $\mathbb{C}$. With operator norm and the standard adjoint operation $T\mapsto T^$, $A$ is a $C^$-algebra (see below). A direct sum of matrix algebras $$A=M_{n_1}(\mathbb{C})\oplus M_{n_2}(\mathbb{C})\oplus \cdots \oplus M_{n_k}(\mathbb{C})$$ is a $C^$-algebra as well. It can be shown that any finite dimensional $C^$-algebra is unital and is a direct sum of matrix algebras as above [@D]. In other words, finite dimensional $C^$-algebras are semi-simple. 7. The above example can be generalized as follows. Let $H$ be a complex Hilbert space and let $ A= \mathcal{L}(H)$ denote the set of bounded linear operators $H\to H$. For a bounded operator $T$, we define $T^\ast$ to be the [adjoint]{} of $T$ defined by $$<Tx, y > = <x, T^\ast y>, \hspace{20pt} x,y \in H.$$ Under the usual algebraic operations of addition and multiplication of operators and the [operator norm]{} $$\\| T \\| = \sup \{ \\| T(x) \\|; \; \\| x \\| \le 1 \},$$ $\mathcal{L}(H)$ is a $C^\ast$-algebra. It is clear that any subalgebra $A\subset \mathcal{L}(H)$ which is [self-adjoint]{} in the sense that $$T\in A \Rightarrow T^\ast \in A,$$ and is [norm closed]{}, in the sense that $$T_n \in A,\; \\|T_n- T\\| \rightarrow 0 \Rightarrow T\in A,$$ is a $C^\ast$-algebra. 8. (group $C^{\ast}-$algebras). Let $G$ be a discrete group and let $H=l^2 G$ denote the Hilbert space of square summable functions on $G$; $$H=\{f:G\to \mathbb{C}; \sum_{g\in G} \|f(g)\|^2<\infty \}.$$ The [left regular representation]{} of $G$ is the unitary representation $\pi$ of $G$ on $H$, defined by $$(\pi g)f (h)=f(g^{-1}h).$$ It has a linear extension to an (injective) algebra homomorphism $$\pi :\mathbb{C}G \longrightarrow \mathcal{L}(H),$$ from the group algebra of $G$ to the algebra of bounded operators on $H$. Its image $\pi (\mathbb{C}G)$ is a $\ast$-subalgebra of $\mathcal{L}(H)$. The [reduced group $C^$-algebra]{} of $G$, denoted by $C^_rG$, is the norm closure of $\pi (\mathbb{C}G)$ in $\mathcal{L}(H)$. It is obviously a $C^$-algebra. There is second $C^$-algebra associated to any discrete group $G$ as follows. The (non-reduced) [group $C^$-algebra]{} of $G$ is the norm completion of the $\ast$-algebra $\mathbb{C}G$ under the norm $$\\|f\\|=\mbox{sup}\; \{\\|\pi (f)\\|; \pi \, \mbox{ is a $\ast$-representation of $\mathbb{C}G$}\},$$ where by a $\ast$-representation we mean a $\ast$-representation on a Hilbert space. Note that $\\|f\\|$ is finite since for $f=\sum_{g \in G}a_g g$ (finite sum) and any \-representation $\pi$ we have $$\\| \pi (f)\\|\leq \sum \\| \pi (a_gg)\\| \leq \sum \|a_g\| \\|\pi (g)\\| \leq \sum \|a_g\|.$$ By its very definition it is clear that there is a 1-1 correspondence between unitary representations of $G$ and $C^$ representations of $C^ G$. Since the identity map $id: (\mathbb{C}G, \\| \,\\|) \to (\mathbb{C}G, \\| \,\\|_r)$ is continuous, we obtain a surjective $C^-$algebra homomorphism $$C^G \longrightarrow C^_rG.$$ It is known that this map is an isomorphism if and only if $G$ is an amenable group [@D]. Abelian groups are amenable. We give a few examples of reduced group $C^$-algebras. Let $G$ be an abelian group and $\hat{G}=Hom (G, \mathbb{T})$ the group of characters of $G$. It is a locally compact Hausdorff space. Moreover it is easily seen that $\hat{G}$ is in fact homeomorphic with the space of characters, or the maximal ideal space, of the $C^$-algebra $C^_rG$. Thus the Gelfand transform defines an isomorphism of $C^$-algebras $$C^G \simeq C_0(\hat{G}).$$ In general one should think of the group $C^$-algebra of a group $G$ as the “algebra of functions" on the noncommutative space representing the unitary dual of $G$. Note that, by the above paragraph, this is justified in the commutative case. In the noncommutative case, the unitary dual is a badly behaved space in general but the noncommutative dual is a perfectly legitimate noncommutative space (see the unitary dual of the infinite dihedral group in [@acb] and its noncommutative replacement). Let $G$ be a finite group. Since $G$ is finite the group $C^$-algebra coincides with the group algebra of $G$. From basic representation theory we know that the group algebra $\mathbb{C}G$ decomposes as a sum of matrix algebras $$C^G\simeq \mathbb{C}G\simeq \oplus M_{n_i}(\mathbb{C}),$$ where the summation is over the set of conjugacy classes of $G$. It is generally believed that the classic paper of Gelfand and Naimark [@gelnai] is the birth place of the theory of $C^$-algebras. The following two results on the structure of $C^$-algebras are proved in this paper:\ a\) Let $A$ be a commutative $C^$-algebra and let $\Omega (A)$ denote the maximal ideal space of $A$. Then the Gelfand transform $$A \rightarrow C_0(\Omega (A)), \quad a\mapsto \hat{a},$$ is an isomorphism of $C^$-algebras.\ b) Any $C^\ast$-algebra is isomorphic to a $C^\ast$-subalgebra of the algebra $\mathcal{L}(H)$ of bounded linear operators on some Hilbert space $H$. In the remainder of this appendix we sketch the proofs of statements a) and b) above. They are based on Gelfand’s theory of commutative Banach algebras, and the Gelfand-Naimark-Segal (GNS) construction of representations of $C^-$algebras from states, respectively. Gelfand’s theory of commutative Banach algebras ----------------------------------------------- The whole theory is based on the notion of spectrum of an element of a Banach algebra and the fact that the spectrum is non-empty. The notion of spectrum can be defined for elements of an arbitrary algebra and it can be easily shown that for finitely generated complex algebras the spectrum is non-empty. As is shown in [@cg], this latter fact leads to an easy proof of Hilbert’s Nullstellensatz. This makes the proofs of the two major duality theorems remarkably similar. We use this approach in this book. Let $A$ be a unital algebra over a field $\mathbb{F}$. The [spectrum]{} of an element $a\in A$ is defined by $$\mbox{sp}(a) = \{ \lambda \in \mathbb{F} ; \; a-\lambda 1 \; \mbox{is {\it not} invertible} \} \; .$$ We should think of the spectrum as the noncommutative analogue of the set of values of a function. This is justified in Example 1 below.\ [Examples:]{}\ 1. Let $A=C(X)$ be the algebra of continuous complex valued functions on a compact space $X$. For any $f\in A$, $$\mbox{sp} (f) = \{ f(x) ;\; x\in X \},$$ is the range of $f$. 2. Let $A=M_n (\mathbb{F} )$ be the algebra of $n\times n$ matrices with coefficients in $\mathbb{F}$. For any matrix $a\in A$ $$\mbox{sp}(a) = \{ \lambda \in \mathbb{F} ; \; \det (a-\lambda 1 ) = 0 \} \; ,$$ is the set of eigenvalues of $a$. Exercise:\ 1) Show that if $a$ is nilpotent then sp $(a)=\{0\}$.\ 2) Show that $$\mbox{sp} (ab)\setminus \{0\}=\mbox{sp} (ba)\setminus \{0\}.$$ 3) Let $T: H \to H$ be a Fredholm operator on a Hilbert space $H$. Let $Q$ be an operator such that $1-PQ$ and $1-QP$ are trace class operators. Show that $$Index (T)= Tr(1-PQ)-Tr(1-QP).$$\ In general, the spectrum may be empty. We give two general results that guarantee the spectrum is non-empty. They are at the foundation of Gelfand-Naimark theorem and Hilbert’s Nullstellensatz. Part b) is in [@cg]. 1. (Gelfand) Let $A$ be a unital Banach algebra over $\mathbb{C}$. Then for any $a\in A,\; \mbox{sp}(a) \ne \emptyset . $ 2. Let $A$ be a unital algebra over $\mathbb{C}$. Assume $\dim_{\mathbb{C}} A$ is countable. Then for any $a\in A,\; \mbox{sp} (a) \ne \emptyset .$ Furthermore, an element $a$ is nilpotent if and only if sp $(a)=\{0\}.$ We sketch a proof of both statements. For a) assume the spectrum of an element $a$ is empty. Then the function $$R: \mathbb{C} \to A, \quad \lambda \mapsto (a-\lambda 1)^{-1},$$ is holomorphic (in an extended sense), non-constant, and bounded. This is easily shown to contradict the classical Liouville’s theorem from complex analysis. For b), again assume the spectrum of a is empty. Then it can be shown that the uncountable set $$\{(a-\lambda 1)^{-1}; \; \lambda \in \mathbb{C} \}$$ is a linearly independent set. But this contradicts the fact that $dim_{\mathbb{C}}A$ is countable. For the second part of b), assume sp $(a)=\{0\}.$ Since $dim_{\mathbb{C}}(A)$ is countable, any element $a\in A$ satisfies a polynomial equation. Let $$p(a)=a^k(a-\lambda_1)\cdots (a-\lambda_n)=0$$ be the minimal polynomial of $A$. Then $n=0$ since otherwise an element $a-\lambda_i$ is not invertible with $\lambda_i \neq 0$. But this contradicts our assumption that sp $(a)=\{0\}.$ The other direction is true in general and is easy. The first part of the following corollary is known as Gelfand-Mazur theorem. Let $A$ be either a unital complex Banach algebra or a unital complex algebra such that $\dim_{\mathbb{C}} A$ is countable. If $A$ is a division algebra, then $A\simeq\mathbb{C}.$ Let $A$ be an algebra. By a [character]{} of $A$ we mean a non-zero algebra homomorphism $$\varphi :A\to \mathbb{F}.$$ Note that if $A$ is unital, then $\varphi (1)=1$. We establish the link between characters and maximal ideals of $A$. For the following result $A$ is either a commutative unital complex Banach algebra, or is a commutative unital algebra with $\dim_\mathbb{C} A$ countable. The relation $I=\ker\varphi$ defines a 1-1 correspondence between the set of maximal ideals of $A$ and the set of characters of $A$. Before embarking on the proof of Gelfand-Naimark theorem, we sketch a proof of Hilbert’s Nullstellensatz, following [@cg]. Let $$A=\mathbb{C} [x_1, \cdots ,x_n]/I$$ be a finitely generated commutative reduced algebra. Recall that reduced means if $a^n=0$ for some $n$ then $a=0$ (no nilpotent elements). Equivalently the ideal $I$ is radical. Let $$V=\{ z \in \mathbb{C}^n ; \; p(z)=0, \mbox{ for all $p$ in $I$}\},$$ let $J(V)$ be the ideal of functions vanishing on $V$, and let $$\mathbb{C}[V]=\mathbb{C}[x_1, \cdots ,x_n]/J(V)$$ be the algebra of regular functions on $V$. Since $I\subset J(V)$, we have an algebra homomorphism $$\pi: A \longrightarrow \mathbb{C}[V].$$ One of the original forms of Hilbert’s Nullstellensatz states that this map is an isomorphism. It is clearly surjective. For its injectivity, let $a\in A$ and let $\pi (a)=0$, or equivalently $\pi (a)\in J(V)$. Since $a$ vanishes on all points of $V$, it follows that $a$ is in the intersection of all the maximal ideals of $A$. This shows that its spectrum sp ($A$)=$\{0\}$. By Theorem A.2 (b), it follows that $a$ is nilpotent and since $A$ is reduced, we have $a=0$. The rest of this section is devoted to sketch a proof of the Gelfand-Naimark theorem on the structure of commutative $C^$-algebras. . Let $A$ be a unital Banach algebra. It is easy to see that any character of $A$ is continuous of norm 1. To prove this, note that if this is not the case then there exists an $a\in A$, with $\\| a \\| < 1$ and $\varphi (a) = 1$. Let $ b = \sum_{n\ge 1} a^n$. Then from $a+ab = b$, we have $$\varphi (b) = \varphi (a) + \varphi (a) \varphi (b) = 1 + \varphi (b),$$ which is impossible. Therefore $\\| \varphi \\| \le 1$, and since $\varphi (1) = 1, \; \\| \varphi \\| =1$. Let $A$ be a complex Banach algebra and let $\Omega (A)$ denote the set of characters of $A$. Thus if $A$ is unital, then $\Omega (A)$= set of maximal ideals of $A$. It is clear that a pointwise limit of characters is again a character. Thus $\Omega (A)$ is a closed subset of the unit ball of the dual space $A^\ast$. Since the latter space is a compact Hausdorff space in the $\mbox{weak}^\ast$ topology, we conclude that $\Omega (A)$ is also a compact Hausdorff space. If $A$ is not unital, let $A^+ = A\oplus \mathbb{C}$ be the unitization of $A$. It is clear that $\Omega (A) = \Omega (A^+) \setminus \{ \varphi_0 \},$ where $\varphi_0$ is the trivial character $\varphi_0 (a)=0$ for all $a \in A$. Since $\Omega (A^+ )$ is compact, we conclude that $\Omega (A)$ is a locally compact Hausdorff space. We have thus proved the lemma: Let $A$ be a Banach algebra. Then $\Omega (A)$ is a locally compact Hausdorff space. $\Omega (A)$ is compact if and only if $A$ is unital. Let $f: A\to B$ be a continuous homomorphism of commutative unital Banach algebras. Define a map $\Omega (f)=f^\ast : \Omega (B) \to \Omega (A)$ by $$f^\ast (\varphi ) = \varphi \circ f.$$ It is clear that $f^\ast \varphi$ is multiplicative, linear and continuous. Thus we have defined the [spectrum functor]{} $\Omega$ from the category of commutative unital complex Banach algebras to the category of compact Hausdorff spaces. Next we define the [Gelfand transform]{}. Let $A$ be a commutative Banach algebra. The Gelfand transform is the map $\Gamma$ defined by $$\Gamma: A\to C_0 ( \Omega (A)),$$ $$a\mapsto \hat{a} \; , \; \hat{a} (\varphi ) = \varphi (a).$$ This map is obviously an algebra homomorphism. It is also clear that $\\| \Gamma \\| \le 1$, i.e. $\Gamma$ is contractive. The spectrum of an element is easily seen to be a closed and bounded subset of $\mathbb{C}$ (this follows from: $\\|1-a\\|<1 \Rightarrow a$ is invertible with inverse $a^{-1}=\sum_{n\geq 0} (1-a)^n.$) Let $$r(a)=\text{max}\; \{\|\lambda \|; \; \lambda \in \text{sp}(a)\},$$ denote the [spectral radius]{} of an element $a \in A.$ Note that $r(a) \leq \\|a\\|.$ Now Corollary 6.2 tells us that $$\mbox{sp}(a)=\{\varphi (a); \: \varphi \in \Omega (A)\}.$$ Then following result is then immediate: The Gelfand transform $A\to C_0 (\Omega (A))$ is a norm decreasing algebra homomorphism and its image separates points of $\Omega (A)$. Moreover, for all $a\in A$, $\\| \hat{a} \\|_{\infty} = r(a) \; .$ The kernel of the Gelfand transform is called the [radical]{} of the Banach algebra $A$. It consists of elements $a$ whose spectral radius $r(a)=0$, or equivalently, sp$(a)=\{ 0 \}$. Hence the radical contains all the nilpotent elements. But it may contain more. An element $a$ is called [quasi-nilpotent]{} if sp$(a)=0$. $A$ is said to be [semi-simple]{} if its radical is zero, i.e. the only quasi-nilpotent elements of $A$ is 0.\ [Example.]{} We give an example of a commutative Banach algebra for which the Gelfand transform is injective but not surjective. Let $H(D)$ be the space of continuous functions on the unit disk $D$ which are holomorphic in the interior of the disk. With the sup-norm $\\| f\\|=\\| f\\|_\infty$ it is a Banach algebra. It is, however, not a $C^\ast$-algebra (why?). Show that $\Omega (A)\simeq D$ and the Gelfand transform is an isometric embedding $H(D)\to C(D)$. We are now ready to prove the first main theorem of Gelfand and Naimark in [@gelnai]: for commutative $C^$-algebras $\Gamma$ is an isometric $$-isomorphism. We need a few simple facts about $C^$-algebras first. Let $f: A \to B$ be a morphism of $C^$-algebras. It is easily seen that sp $(f(a))\subset $ sp ($a$). Hence, using the $C^$-identity, we have $$\\|f(a)\\|^2=\\|f(a) f(a^)\\| =\\|f(aa^)\\| =r(a^a)\leq \\|a^a\\|=\\|a\\|^2.$$ Let $a\in A$ be a standpoint element ($a=a^$). Then sp ($a$) $\subset \mathbb{R}$ is real. Indeed since $e^{ia}$ is unitary its spectrum is located on the unit circle (why?). Hence for $\lambda \in sp (a)$, $e^{i\lambda}$ is located on the unit circle which shows that $\lambda$ is real. From this it follows that if $f: A\to \mathbb{C}$ is an algebra homomorphism (a character), then $f (a^)=\overline{f(a)},$ for all $a$. Let $A$ be a commutative $C^\ast$-algebra. The Gelfand transform $A\to C_0 (\Omega (A))$ is an isomorphism of $C^\ast$-algebras. We prove the unital case. The non-unital case follows with minor modifications [@D]. What we have shown so far amounts to the fact that $\Gamma$ is an isometric $$-algebra map whose image separates points of $\Omega (A)$. Thus $\Gamma (A)$ is a closed $$-subalgebra of $C(\Omega (A))$ that separates points of $\Omega (A)$. By Stone-Weierstrass theorem, $\Gamma (A)=C(\Omega (A)).$ The above theorem is one of the landmarks of Gelfand’s theory of commutative Banach algebras. While a complete classification of all commutative Banach algebras seems to be impossible, this result classifies all commutative $C^\ast$-algebras. Care must be applied in dealing with non-compact spaces and non-unital algebras. First of all it is clear that if $X$ is not compact then the pull-back $f^: C_0(Y)\to C_0(X), \quad f^(g)=g\circ f$ of a continuous map $f:X \to Y$ is well defined if and only if $f$ is a proper map. Secondly, one notes that not all $C^$-maps $C_0(Y)\to C_0(X)$ are obtained in this way. For example, for $X=(0, 1)$ an open interval and $Y$ a single point, the zero map $0: C_0(Y) \to C_0(X)$, which is always a $C^$-morphism, is not the pull-back of any proper map. It turns out that one way to single out the appropriate class of morphisms is as follows. A morphism of $C^$-algebras $\varphi: A \to B$ is called [proper]{} if for an approximate unit $(e_i)$ of $A$, $\varphi (e_i)$ is an approximate unit of $B$. Recall that [@D; @gvf] an [approximate unit]{} of a $C^$-algebra $A$ is an increasing net of positive elements $(e_i), i\in I,$ of $A$ such that for all $a \in A$, $e_ia\to a$, or equivalently, $ae_i \to a$. Now it can be shown that any proper map $C_0(Y) \to C_0(X)$ is the pull-back of a proper map $X\to Y$. Similarly, in the other direction, if $\varphi : A \to B$ is a proper morphism of $C^$-algebras, then $\Omega (\varphi): \Omega (B) \to \Omega (A)$ is a proper continuous map. We refer to [@gvf] and references therein for more details. We are now half-way through showing that the functors $C_0$ and $\Omega$ are quasi-inverse to each other. But the proof of the other half is much simpler and is left to the reader. Let $X$ be a locally compact Hausdorff space. Then the [evaluation map]{} $$X\to \Omega (C_0 (X)), \quad \; x\mapsto \varphi_x,$$ $$\varphi_x (f)=f(x),$$ is a homeomorphism. We give a few elementary applications of the Gelfand-Naimark correspondence between commutative $C^$-algebras and locally compact spaces. 1. (Idempotents and connectedness). Let $A$ be a unital commutative $C^\ast$-algebra. Then $\Omega (A)$ is disconnected iff $A$ has a non-trivial idempotent (i.e. an element $e\ne 0, 1$ such that $e^2 = e$). 2. (Ideals and closed subsets). Let $X$ be a compact Hausdorff space. The Gelfand-Naimark duality shows that there is a 1-1 correspondence $$\{ \mbox{closed subsets of } X \} \leftrightarrow \{ \mbox{closed ideals of } C(X) \},$$ where to each closed subset $Y\subset X$, we associate the ideal $$I= \{ f\in C(X) ; f(y) = 0 \quad \forall\; y\in Y \}$$ of all functions vanishing on $Y$. We have natural isomorphisms $$C_0 (X\setminus Y) \simeq I,$$ $$C_0(X)/ I \simeq C_0(Y),$$ $$C_0(X/ Y)\simeq C_0 (X\setminus Y)^+.$$ 3. (Essential ideals and compactification) Let $X$ be a locally compact Hausdorff space. Recall that a [Hausdorff compactification]{} of $X$ is a compact Hausdorff space $Y$ where $X$ is homeomorphic to a dense subset of $Y$. We consider $X$ as a subspace of $Y$. Then $X$ is open in $Y$ and its boundary $Y\setminus X$ is compact. We have an exact sequence $$0\to C_0(X) \to C(Y) \to C(Y\setminus X) \to 0,$$ where $C_0 (X)$ is an essential ideal of $C(Y)$. Conversely, show that any extension $$0\to C_0(X) \to A \to B \to 0,$$ Where $A$ is a unital $C^$-algebra and $C_0(X)$ is an essential ideal of $A$, defines a Hausdorff compactification of $X$. Thus, we have a 1-1 correspondence between Hausdorff compactifications of $X$ and (isomorphism classes of) essential extensions of $C_0(X)$. In particular, the 1-point compactification and the Stone-Cech compactifications correspond to $$0\to C_0(X) \to C_0 (X)^+ \to \mathbb{C} \to 0,$$ and $$0\to C_0(X) \to C_b (X) \to C(\beta X) \to 0$$ States and the GNS construction -------------------------------- Our goal in this section is to sketch a proof of the second main result of Gelfand and Naimark in [@gelnai] to the effect that any $C^$-algebra can be embedded in the algebra of bounded operators on a Hilbert space. The main idea of the proof is an adaptation of the idea of [left regular representation]{} of algebras to the context of $C^$-algebras, called Gelfand-Naimark-Segal (GNS) construction. A [positive linear functional]{} on a $C^$-algebra $A$ is a $\mathbb{C}$-linear map $\varphi : A \longrightarrow \mathbb{C}$ such that for all $a$ in $A$, $$\varphi (a^* a)\geq 0.$$ A [state]{} on $A$ is a positive linear functional $\varphi$ with $\\|\varphi\\|=1.$ It can be shown that if $A$ is unital then this last condition is equivalent to $\varphi (1)=1.$ If $\varphi_1$ and $\varphi_2$ are states then for any $t\in [0, 1]$, t$\varphi_1$ +(1-t) $\varphi_2$ is a state as well. Thus the set of states of $A$, denoted by $\mathcal{S}(A)$, form a convex subset of the unit ball of $A^$. The extreme points of $\mathcal{S}(A)$ are called [pure states]{}.\ [Examples:]{}\ 1. States are noncommutative analogues of probability measures. This idea is corroborated by the Riesz representation theorem: For a locally compact Hausdorff space $X$ there is a 1-1 correspondence between states on $C_0(X)$ and Borel probability measures on $X$. To a probability measure $\mu$ is associated the stats $\varphi$ defined by $$\varphi (f)=\int_X fd\mu.$$ $\varphi$ is a pure state if and only if $\mu =\delta_x$ is a Dirac measure for a point $x\in X$. 2\. Let $A=M_n(\mathbb{C})$ and $p\in A$ be a positive matrix with $tr(p)=1$. (Such matrices, and their infinite dimensional analogues, are called [density matrices]{} in quantum statistical mechanics.) Then $$\varphi (a)=tr (ap)$$ defines a state on $A$. 3\. Let $\pi : A \longrightarrow \mathcal{L}(H)$ be a [representation* ]{} of a unital $C^$-algebra $A$ on a Hilbert space $H$. This simply means that $\pi$ is a morphism of unital $C^$-algebras. Let $x\in H$ be a vector of length one . Then $$\varphi (a)=<\pi (a)x, x>$$ defines a state on $A$, called a [vector state]{}. In the following we show that, conversely, any state on $A$ is a vector state with respect to a suitable representation called the GNS representation. Let $\varphi$ be a positive linear functional on $A$. Then $$<a, b>:=\varphi (b^a)$$ is linear in the first variable and anti-linear in the second variable. It is also semi-definite in the sense that $<a, a>=\varphi (a^a)\geq 0$ for all $a$ in $A$. Thus it satisfies the [Cauchy- Schwartz]{} inequality: for all $a, b$ $$\|\varphi(b^a)\|^2\leq \varphi(a^a)\varphi(b^b).$$ Let $$N=\{ a\in A; \quad f(a^a)=0\}.$$ It is easy to see, using the above Cauchy-Schwarz inequality, that $N$ is a closed left ideal of $A$ and the following positive definite inner product is well-defined on the quotient space $A/N$: $$<x+N, y+N>:= <x, y>.$$ Let $H_{\varphi}$ denote the Hilbert space completion of $A/N$ under the above inner product. The [left regular representation]{} $A \times A \to A, (a, b)\mapsto ab$ of $A$ on itself induces a bounded linear map $A \times A/N \to A/N, (a, b+N)\mapsto ab+N$. We denote its unique extension to $H_{\varphi}$ by $$\pi_{\varphi} : A \longrightarrow \mathcal{L}(H_{\varphi}).$$ The representation $(\pi_{\varphi}, H_{\varphi})$ is called the GNS representation defined by the state $\varphi$. The state $\varphi $ can be recovered from the representation $(\pi_{\varphi}, H_{\varphi})$ as a vector state as follows. Let $x=\pi_{\varphi}(1)$. Then for all $a$ in $A$, $$\varphi (a)=<(\pi_{\varphi}a)(x), x>.$$ The representation $(\pi_{\varphi}, H_{\varphi})$ may not be faithful. It can be shown that it is irreducible if and only if $\varphi$ is a pure state [@D]. To construct a faithful representation, and hence an embedding of $A$ into the algebra of bounded operators on a Hilbert space, one first shows that there are enough pure states on $A$. The proof of the following result is based on Hahn-Banach and Krein-Milman theorems. For any selfadjoint element $a$ of $A$, there exists a pure state $\varphi$ on $A$ such that $\varphi (a)=\\|a\\|.$ Using the GNS representation associated to $\varphi$, we can then construct, for any $a \in A$, an irreducible representation $\pi$ of $A$ such that $\\|\pi (a) \\|=\|\varphi (a)\|=\\|a\\|.$ We can now prove the second theorem of Gelfand and Naimark. Every $C^$-algebra is isomorphic to a $C^$-subalgebra of the algebra of bounded operators on a Hilbert space. Let $\pi=\sum_{\varphi \in \mathcal{S}(A)} \pi_{\varphi}$ denote the direct sum of all GNS representations for all states of $A$. By the above remark $\pi$ is faithful. Idempotents and finite projective modules ========================================= Let $A$ be a unital algebra over a commutative ring $k$ and let $\mathcal{M}_A$ denote the category of right $A$-modules. We assume our modules $M$ are [unitary]{} in the sense that the unit of the algebra acts as the identity of $M$. A morphism of this category is a right $A$-module map $f: M \rightarrow N$, i.e. $f(ma)=f(m)a$, for all $a$ in $A$ and $m$ in $M$. A [free]{} module, indexed by a set $I$, is a module of the type $$M=A^I=\bigoplus_IA,$$ where the action of $A$ is by component-wise right multiplication. Equivalently, $M$ is free if and only if there are elements $m_i \in M, i\in I$, such that any $m\in M$ can be uniquely expressed as a finite sum $m=\sum_i m_i a_i$. A module $M$ is called [finite]{} (= [finitely generated]{}) if there are elements $m_1, m_2, \cdots ,m_k$ in $M$ such that every element of $m\in M$ can be expressed as $m=m_1a_1 +\cdots +m_ka_k$, for some $a_i\in A.$ Equivalently $M$ is finite if there is a surjective $A$-module map $A^k \rightarrow M$ for some integer $k$. Free modules correspond to trivial vector bundles. To obtain a more interesting class of modules we consider the class of [projective modules]{}. A module $P$ is called projective if it is a direct summand of a free module. That is there exists a module $Q$ such that $$P\oplus Q \simeq A^I.$$ A module is said to be [finite projective]{} (= [finitely generated projective]{}), if it is both finitely generated and projective. Let $P$ be an $A$-module. The following conditions on $P$ are equivalent: 1. $P$ is projective. 2. Any surjection $$M\overset{f}{\to} P \to 0,$$ splits in the category of $A$-modules. 3. For all $A$-modules $N$ and $M$ and morphisms $f,g$ with $g$ surjective in the following diagram, there exists a morphism $\tilde{f}$ such that the diagram commutes: $$\xy \xymatrix@=9ex{ & P \ar@{-->}[dl]_{\exists \tilde{f}} \ar[d]^f\\ N \ar[r]^g & M \ar[r] & 0 } \endxy$$ We say that $\tilde{f}$ is a lifting of $f$ along $g$. 4. The functor $$\mbox{Hom}_A (P,-):\mathcal{M}_A\to \mathcal{M}_k$$ is exact in the sense that for any short exact sequence of $A$-modules $$0 \to R\to S\to T \to 0,$$ the sequence of $k$-modules $$0\to \mbox{Hom}_A (P,R)\to \mbox{Hom}_A (P,S) \to \mbox{Hom}_A (P,T) \to 0$$ is exact. <!-- --> 1. Free modules are projective. 2. If $A$ is a division ring, then any $A$-module is free, hence projective. 3. $M=\mathbb{Z}/n\mathbb{Z}, \; n\ge 2$, is not projective as a $\mathbb{Z}$-module. 4. A direct sum $P=\oplus_i P_i$ of modules is projective iff each summand $P_i$ is projective. 5. (idempotents) Let $$e\in M_n (A) = \mbox{End}_A (A^n), \quad e^2 = e,$$ be an idempotent. Multiplication by $e$ defines a right $A$-module map $$P_e : A^n \to A^n,\; \quad x\mapsto ex.$$ Let $P=\mbox{Im} (P_e)$ and $Q=\mbox{Ker} (P_e)$ be the image and kernel of $P_e$. Then, using the idempotent condition $e^2=e$, we obtain a direct sum decomposition $$P\oplus Q = A^n,$$ which shows that both $P$ and $Q$ are projective modules. Moreover they are obviously finitely generated. It follows that both $P$ and $Q$ are finite projective modules. Conversely, given any finite projective module $P$, let $Q$ be a module such that $P\oplus Q \simeq A^n$ for some integer $n$. Let $e: A^n \rightarrow A^n$ be the right $A$-module map that corresponds to the projection map $$(p, q)\mapsto (p, 0).$$ Then it is easily seen that we have an isomorphism of $A$-modules $$P\simeq P_e.$$ This shows that any finite projective module is obtained from an idempotent in some matrix algebra over $A$. The idempotent $e \in M_n(A)$ associated to a finite projective $A$-module $P$ depends of course on the choice of the splitting $P\oplus Q\simeq A^n.$ Let $P\oplus Q'\simeq A^m$ be another splitting and $f\in M_m(A)$ the corresponding idempotent. Define the operators $u\in Hom_A(A^m, A^n)$, $v\in Hom_A(A^n, A^m)$ as compositions $$u: A^m \overset{\sim}{\longrightarrow} P\oplus Q\longrightarrow P\longrightarrow P\oplus Q' \overset{\sim}{\longrightarrow} A^n,$$ $$v: A^n \overset{\sim}{\longrightarrow} P\oplus Q'\longrightarrow P\longrightarrow P\oplus Q \overset{\sim}{\longrightarrow} A^m.$$ We have $$uv=e, \quad vu=f.$$ In general, two idempotents satisfying the above relations are called Murray-von Neumann equivalent. Conversely it is easily seen that Murray-von Neumann equivalent idempotents define isomorphic finite projective modules. Defining finite projective modules through idempotents is certainly very convenient since both finiteness and projectivity of the module are automatic in this case. In some cases however this is not very useful. For example, modules over quantum tori are directly defined and checking directly that they are finite and projective is difficult. In this case the following method, due to Rieffel [@rie2], is very useful. (I am grateful to Henrique Bursztyn for bringing this method to my attention). Let A and B be unital algebras over a ground ring $k$. let X be an (A,B)-bimodule endowed with $k$-bilinear pairings (“algebra-valued" inner products): $$<-,->_A : X \times X \longrightarrow A,$$ $$<-,->_B : X \times X \longrightarrow B,$$ satisfying the “associativity" condition $$<x, y>_Az = x<y, z>_B \quad \text{ for all x, y, z in $X$},$$ and the “fullness" conditions: $$<X, X>_A=A \quad \text{ and}\quad <X, X>_B=B.$$ We claim that $X$ is a finite projective left $A$-module. Let $1_B$ be the unit of $B$. By fullness of $<-,->_B$, we can find $x_i, y_i $ in $X, i=1, \cdots, k$ such that $$1_B = \sum_{i=1}^k <x_i, y_i>_B.$$ Let $e_i, i=1, \cdots, k$, be a basis for $A^k$. Define the map $$P: A^k \longrightarrow X, \quad P(e_i)=y_i.$$ We claim that this map splits and hence $X$ is finite and projective. Consider the map $$I: X \longrightarrow A^k, \quad I(x)= \sum_i <x,x_i>_Ae_i.$$ We have $$P\circ I(x)= \sum_i <x,x_i>_Ay_i = \sum_i x<x_i,y_i>_B \quad \text{(by associativity)}$$ $$= x \quad \text{(since $\sum_i<x_i,y_i>_B=1_B$)}.$$ Equivalence of categories ========================== There are at least two ways to compare categories: isomorphism and equivalence. [Isomorphism]{} of categories is a very strong requirement and is hardly useful. [Equivalence]{} of categories, on the other hand, is a much more flexible concept and is very useful. Categories $A$ and $B$ are said to be [equivalent]{} if there is a functor $F:A\to B$ and a functor $G:B\to A$, called a [quasi-inverse]{} of $F$, such that $$F\circ G \simeq 1_B,\quad G\circ F\simeq 1_A \;,$$ where $\simeq$ means isomorphism, or natural equivalence, of functors. This means, for every $X\in \mbox{obj} A, \; Y\in \mbox{obj} B,$ $$FG(Y)\sim Y, \quad \text{and} \quad GF(X)\sim X,$$ where $\sim$ denotes natural isomorphism of objects. If $F\circ G = 1_B$ and $G\circ F=1_A$ (equality of functors), then we say that categories $A$ and $B$ are [isomorphic]{}, and $F$ is an isomorphism. Categories $A$ and $B$ are said to be [antiequivalent]{} if the opposite category $A^{op}$ is equivalent to $B$. Note that a functor $F:A \to B$ is an isomorphism if and only if $F:\mbox{obj} A \to \mbox{obj} B$ is 1-1, onto and $F$ is [full and faithful]{} in the sense that for all $X,Y\in \mbox{obj} A,$ $$F:\mbox{Hom}_A (X,Y) \to \mbox{Hom}_B (F(X),F(Y))$$ is 1-1 (faithful) and onto (full). It is easy to see that an equivalence $F:A\to B$ is full and faithful, but it may not be 1-1, or onto on the class of objects. As a result an equivalence may have many quasi-inverses. The following concept clarifies the situation with objects of equivalent categories. A subcategory $A'$ of a category $A$ is called [skeletal]{} if 1) the embedding $A' \to A$ is full, i.e. $$\mbox{Hom}_{A'} (X,Y) = \mbox{Hom}_A (X,Y)$$ for all $X,Y \in \mbox{obj} A'$ and 2) for any object $X\in \mbox{obj} A$, there is a unique object $X^\prime \in \mbox{obj} A^\prime$ isomorphic to $X$. Any skeleton of $A$ is equivalent to $A$ and it is not difficult to see that two categories $A$ and $B$ are equivalent if and only if they have isomorphic skeletal subcategories $A^\prime$ and $B^\prime$. In some examples, like the Gelfand-Naimark theorem, there is a canonical choice for a quasi-inverse for a given equivalence functor $F$ ($F=C_0$ and $G=\Omega$). There are instances, however, like the Serre-Swan theorem, where there is no canonical choice for a quasi-inverse. The following proposition gives a necessary and sufficient condition for a functor $F$ to be an equivalence of categories. We leave its simple proof to the reader. A functor $F:A\to B$ is an equivalence of categories if and only if 1. $F$ is full and faithful, and 2. Any object $Y\in \mbox{obj} B$ is isomorphic to an object of the form $F(X)$, for some $X \in obj A$. [9]{} R. Akbarpour, and M. Khalkhali, Hopf algebra equivariant cyclic homology and cyclic homology of crossed product algebras. J. Reine Angew. Math. 559 (2003), 137–152. R. Akbarpour, and M. Khalkhali, Equivariant cyclic cohomology of $ H$-algebras. $K$-Theory 29 (2003), no. 4. L. Brown, R. Douglas, and P. Fillmore, Extensions of $C\sp$-algebras and $K$-homology. Ann. of Math. (2) [105]{} (1977), no. 2, 265–324. B. Blackadar, $K$-theory for operator algebras. Mathematical Sciences Research Institute Publications, 5. Springer-Verlag, New York, 1986. M. Born, W. Heisenberg, and P. Jordan, [On quantum mechanics II]{}, Zs. Phys. [35]{} (1926), 557-615 (translated and reproduced in [@vdw]). T. Brzeziński, and S. Majid, Shahn Quantum group gauge theory on quantum spaces. Comm. Math. Phys. [157]{} (1993), no. 3, 591–638 D. Burghelea, The cyclic homology of the group rings. Comment. Math. Helv. [60]{} (1985), no. 3, 354–365. H. Cartan, and S. Eilenberg, Homological algebra. With an appendix by David A. Buchsbaum. Reprint of the 1956 original. Princeton Landmarks in Mathematics. Princeton University Press, Princeton, NJ, 1999. N. Chriss, and V. Ginzburg, Representation theory and complex geometry. Birkhäuser Boston, Inc., Boston, MA, 1997 A. Connes, [ $C\sp{} $ algébres et géométrie différentielle]{}. (French) C. R. Acad. Sci. Paris Sér. A-B [290]{} (1980), no. 13, A599–A604. A. Connes, [Spectral sequence and homology of currents for operator algebras]{}. Oberwolfach Tagungsber., 41/81, Funktionalanalysis und $C^$-Algebren, 27-9/3-10, 1981. A. Connes, [Cohomologie cyclique et foncteurs ${\rm Ext}\sp n$ ]{}. (French) (Cyclic cohomology and functors ${\rm Ext}\sp n$) C. R. Acad. Sci. Paris Sér. I Math. [296]{} (1983), no. 23, 953–958. A. Connes, [Noncommutative differential geometry]{}. Inst. Hautes Études Sci. Publ. Math. No. [62]{} (1985), 257–360. A. Connes, A. Cyclic cohomology and the transverse fundamental class of a foliation. Geometric methods in operator algebras (Kyoto, 1983), 52–144, Pitman Res. Notes Math. Ser., 123, Longman Sci. Tech., Harlow, 198 A. Connes, [Noncommutative geometry]{}. Academic Press, Inc., San Diego, CA, 1994. A. Connes, Noncommutative geometry year 2000. Highlights of mathematical physics (London, 2000), 49–110, Amer. Math. Soc., Providence, RI, 2002 A. Connes, J. Cuntz, E. Guentner, N. Higson, J. Kaminker, J. Roberts, Noncommutative geometry. Lectures given at the C.I.M.E. Summer School held in Martina Franca, September 3–9, 2000. Edited by S. Doplicher and R. Longo. Lecture Notes in Mathematics, 1831. Springer-Verlag, Berlin, 2004. A. Connes and H. Moscovici, [Hopf algebras, cyclic cohomology and the transverse index theorem]{}, Comm. Math. Phys. [198]{} (1998), no. 1, 199–246. A. Connes and H. Moscovici, [Cyclic cohomology and Hopf algebra symmetry.]{} Conference Moshé Flato 1999 (Dijon). Lett. Math. Phys. [52]{} (2000), no. 1, 1–28. A. Connes and H. Moscovici, Rankin-Cohen brackets and the Hopf algebra of transverse geometry, to appear in the Moscow Mathematical Journal (volume dedicated to Pierre Cartier). A. Connes and H. Moscovici, Modular Hecke algebras and their Hopf symmetry, to appear in the Moscow Mathematical Journal (volume dedicated to Pierre Cartier). A. Connes, and G. Landi, Noncommutative manifolds, the instanton algebra and isospectral deformations. Comm. Math. Phys. [221]{} (2001), no. 1, 141–159. A. Connes, and M. Marcolli, From physics to number theory via noncommutative geometry. Part I: Quantum Statistical Mechanics of Q-lattices, math.NT/0404128. A. Connes, and M. Rieffel, Yang-Mills for noncommutative two-tori. Operator algebras and mathematical physics (Iowa City, Iowa, 1985), 237–266, Contemp. Math., [62]{}, 1987. J. Cuntz, Cyclic theory, bivariant $K$-theory and the bivariant Chern-Connes character. Cyclic homology in non-commutative geometry, 1–71, Encyclopaedia Math. Sci., 121, Springer, Berlin, 2004. J. Cuntz, and M. Khalkhali, Cyclic cohomology and noncommutative geometry. Proceedings of the workshop held in Waterloo, ON, June 14–18, 1995. Fields Institute Communications, 17. J. Cuntz, and D. Quillen,[ Cyclic homology and nonsingularity.]{} J. Amer. Math. Soc. [8]{} (1995), no. 2, 373-442. J. Cuntz, and D. Quillen,[ Algebra extensions and nonsingularity.]{} J. Amer. Math. Soc. [8]{} (1995), no. 2, 251–289. J. Cuntz, and D. Quillen, Excision in bivariant periodic cyclic cohomology. Invent. Math. [127]{} (1997), no. 1, 67–98. J. Cuntz, G. Skandalis, and B. Tsygan, Cyclic homology in non-commutative geometry. Encyclopaedia of Mathematical Sciences, 121. Operator Algebras and Non-commutative Geometry, II. Springer-Verlag, Berlin, 2004. K. Davidson, $C\sp $-algebras by example, Fields Institute Monographs, 6. American Mathematical Society, Providence, RI, 1996. P. Dirac, [The fundamental equations of quantum mechanics]{}, Proc. Roy. Soc. A [109]{} (1926), 642-653 (reproduced in [@vdw]). P. A. Fillmore, A user’s guide to operator algebras, Canadian Mathematical Society Series of Monographs and Advanced Texts. John Wiley and Sons, Inc., New York, 1996 B. Feĭgin, and B. L. Tsygan,[ Additive $K$-theory. $K$-theory, arithmetic and geometry]{} (Moscow, 1984–1986), 67–209, Lecture Notes in Math., [1289]{}, Springer, Berlin, 1987. I. M. Gelfand, and M. A. Naimark, [On the embedding of normed rings into the ring of operators in Hilbert space]{}, Mat. Sbornik [12]{} (1943), 197-213. M. Gerstenhaber, The cohomology structure of an associative ring. Ann. of Math. (2) [78]{} 1963 267–288. J. M. Gracia-Bondia, J. C. Varilly, and H. Figueroa, [Elements of Noncommutative Geometry]{}, Birkhaeuser, 2000. P. Hajac, M. Khalkhali, B. Rangipour, and Y. Sommerhäuser, Stable anti-Yetter-Drinfeld modules. C. R. Math. Acad. Sci. Paris 338 (2004), no. 8, 587–590. P. Hajac, M. Khalkhali, B. Rangipour, and Y. Sommerhäuser, Hopf-cyclic homology and cohomology with coefficients. C. R. Math. Acad. Sci. Paris 338 (2004), no. 9, 667–672. R. Hartshorne, Algebraic geometry. Graduate Texts in Mathematics, No. 52. Springer-Verlag, New York-Heidelberg, 1977. W. Heisenberg, [Quantum-theoretical re-interpretation of Kinematic and mechanical relations]{}, Zs. Phys. [33]{} (1925) 879-893 (translated and reproduced in [@vdw]). N. Higson, and J. Roe, Analytic $K$-homology. Oxford Mathematical Monographs. Oxford Science Publications. Oxford University Press, Oxford, 2000 G. Hochschild, On the cohomology groups of an associative algebra. Ann. of Math. (2) 46, (1945). 58–67. G. Hochschild, B. Kostant, and A. Rosenberg, Differential forms on regular affine algebras. Trans. Amer. Math. Soc. 102 1962 383–408. M. Karoubi, $K$-theory. An introduction. Grundlehren der Mathematischen Wissenschaften, Band 226. Springer-Verlag, Berlin-New York, 1978. M. Karoubi, Homologie cyclique et $K$-théorie. (French) \[Cyclic homology and $K$-theory\] Astérisque No. 149, 1987. M. Khalkhali, On the entire cyclic cohomology of Banach algebras. Comm. Algebra [22]{} (1994), no. 14, 5861–5874. M. Khalkhali, Algebraic connections, universal bimodules and entire cyclic cohomology. Comm. Math. Phys. [161]{} (1994), no. 3, 433–446. M. Khalkhali, and B. Rangipour, A new cyclic module for Hopf algebras. $K$-Theory 27 (2002), no. 2, 111–131. M. Khalkhali, and B. Rangipour, Invariant cyclic homology. $K$-Theory 28 (2003), no. 2, 183–205. 58B34 M. Khalkhali, and B. Rangipour, Cyclic cohomology of (extended) Hopf algebras. Noncommutative geometry and quantum groups (Warsaw, 2001), 59–89, Banach Center Publ., 61, Polish Acad. Sci., Warsaw, 2003. M. Khalkhali, and B. Rangipour, A note on cyclic duality and Hopf algebras. math.KT/0310088, to be published. J. L. Loday, [Cyclic Homology.]{} Springer-Verlag, second edition (1998). J. L. Loday, and D. Quillen, Cyclic homology and the Lie algebra homology of matrices. Comment. Math. Helv. 59 (1984), no. 4, 569–591. J. Milnor, and J. Stasheff, Characteristic classes. Annals of Mathematics Studies, No. 76. Princeton University Press, Princeton, N. J. (1974). J. Renault, A groupoid approach to $C\sp{} $-algebras. Lecture Notes in Mathematics, 793. Springer, Berlin, 1980. M. Rieffel, Morita equivalence for $C\sp{} $-algebras and $W\sp{} $-algebras. J. Pure Appl. Algebra 5 (1974), 51–96 M. Rieffel, Projective modules over higher-dimensional noncommutative tori. Canad. J. Math. [40]{} (1988), no. 2, 257–338. R. Swan, Vector bundles and projective modules. Trans. Amer. Math. Soc. [105*]{} 1962 264–277. B. Tsygan, Homology of matrix Lie algebras over rings and the Hochschild homology. (Russian) Uspekhi Mat. Nauk 38 (1983), no. 2(230), 217–218. B. Van Der Waerden, [Sources of quantum mechanics]{}, Classics of Science Vol. 5, Dover. [^1]: BIRS Workshop on Noncommutative Geometry, Banff International Research Station, Banff, Alberta, Canada, April 2003, Organized by Alain Connes, Joachim Cuntz, George Elliott, Masoud Khalkhali, and Boris Tsygan. Full report available at: www.pims.math.ca/birs.	{ "pile_set_name": "ArXiv" }
--- abstract: 'We give an analytical proof of the Poincaré-type inequalities for widths of geodesic homotopies between equivariant maps valued in Hadamard metric spaces. As an application we obtain a linear bound for the length of an element conjugating two finite lists in a group acting on an Hadamard space.' author: - \| Gerasim Kokarev\ [School of Mathematics, The University of Edinburgh]{}\ [King’s Buildings, Mayfield Road, Edinburgh EH9 3JZ, UK]{}\ [Email: [[email protected]]{}]{} title: On geodesic homotopies of controlled width and conjugacies in isometry groups --- [ 53C23, 58E20, 20F65. ]{} Introduction ============ Let $M$ and $M'$ be smooth Riemannian manifolds without boundary. For a smooth mapping $u:M\to M'$ by $E(u)$ we denote its energy $$\label{e} E(u)=\int\limits_M{\left\lVertdu(x)\right\rVert}^2d\mathit{Vol}(x),$$ where the norm of the linear operator $du(x):T_xM\to T_{u(x)}M'$ is induced by the Riemannian metrics on $M$ and $M'$. Let $u$ and $v$ be smooth homotopic mappings of $M$ to $M'$ and $H(s,\cdot)$ be a smooth homotopy between them. The [$L_2$-width]{} $W_2(H)$ of $H$ is defined as the $L_2$-norm of the function $$\label{length} \ell_H(x)=\text{ the length of the curve }s\mapsto H(s,x),\qquad x\in M.$$ A smooth homotopy $H(s,x)$ is called [geodesic]{} if for each $x\in M$ the track curve $s\mapsto H(s,x)$ is a geodesic. In [@KKS03; @KKS05] Kappeler, Kuksin, and Schroeder prove the following geometric inequality for the $L_2$-widths of geodesic homotopies when the target manifold $M'$ is non-positively curved. Let $M$ and $M'$ be compact Riemannian manifolds and suppose that $M'$ has non-positive sectional curvature. Let $\zeta$ be a homotopy class of maps of $M$ to $M'$. Then there exist constants $C_\star$ and $C$ with the following property: any smooth homotopic maps $u$ and $v\in\zeta$ can be joined by a geodesic homotopy $H$ whose $L_2$-width $W_2(H)$ is controlled by the energies of $u$ and $v$, $$\label{W1} W_2(H)\leqslant C_\star(E^{1/2}(u)+E^{1/2}(v))+C.$$ Moreover, if the sectional curvature of $M'$ is strictly negative, the constants $C_\star$ and $C$ can be chosen to be independent of the homotopy class $\zeta$. This inequality can be viewed as a version of the Poincaré inequality for mappings between manifolds. It also has an isoperimetric flavour; it says that the ‘measure’ of the cylinder induced by the homotopy is estimated in terms of the ‘measure’ of its boundary. Inequality is a key ingredient in the proof of compactness results for perturbed harmonic map equation [@KKS03; @GK1]. The latter, combined with old results of Uhlenbeck, yields Morse inequalities for harmonic maps with potential [@GK2]. The proof of Width Inequality I in [@KKS03; @KKS05] is based on an analogous inequality for maps of metric graphs; see [@KKS03 Th. 5.1]. In more detail, let $G$ be a finite graph and $u:G\to M'$ be a smooth map, that is whose restriction to every edge is smooth. The length $L(u)$ of $u$ is defined as the sum of the lengths of the images of the edges. By the $L_\infty$-width $W_\infty(H)$ of a homotopy $H$ we mean the $L_\infty$-norm of the length function $\ell_H(x)$, given by . Let $G$ be a finite graph and $M'$ be a compact manifold of non-positive sectional curvature. Let $\zeta$ be a homotopy class of maps $G\to M'$. Then there exist constants $C_\star$ and $C$ with the following property: any smooth homotopic maps $u$ and $v\in\zeta$ can be joined by a geodesic homotopy $H$ such that $$W_\infty(H)\leqslant C_\star(L(u)+L(v))+C.$$ Moreover, if the sectional curvature of $M'$ is strictly negative, the constants $C_\star$ and $C$ can be chosen to be independent of the homotopy class $\zeta$. The purpose of this note is two-fold: firstly, we generalise the width inequalities to the framework of equivariant maps valued in Hadamard spaces. This, in particular, includes width inequalities for homotopies between maps into non-compact metric target spaces. In contrast with the geometric methods in [@KKS03; @KKS05] (and also in [@BH]), we give an analytical proof of the width inequalities via harmonic map theory. Secondly, we use width inequalities for equivariant maps of trees to obtain informaion on algebraic properties of finitely generated groups $\Lambda$ acting by isometries on Hadamard spaces. More precisely, under some extra hypotheses, these groups satisfy the following property: given two finite conjugate lists of elements $(a_i)_{1\leqslant i\leqslant N}$ and $(b_i)_{1\leqslant i\leqslant N}$ in $\Lambda$ there exists $g\in\Lambda$ with $b_i=g^{-1}a_ig$ such that $${\left\lvertg\right\rvert}\leqslant C_\star\sum_{i=1}^N({\left\lverta_i\right\rvert}+{\left\lvertb_i\right\rvert})+C,$$ where ${\left\lvert\cdot\right\rvert}$ stands for the length $d(\cdot,e)$ in the word metric on $\Lambda$. If the group $\Lambda$ has a soluble word problem, then the latter estimate yields immediately the solubility of the conjugacy problem for finite lists in $\Lambda$. Statements and discussion of results ==================================== Width inequalities for equivariant maps --------------------------------------- Let $M$ be a compact Riemannian manifold without boundary; we denote by $\tilde M$ its universal cover and by $\Gamma$ the fundamental group $\pi_1(M)$. Let $(Y,d)$ be an Hadamard space; that is a complete length space of non-positive curvature in the sense of Alexandrov (see Sect. \[prem\] for a precise definition). Denote by $\rho$ a representation of $\Gamma$ in the isometry group of $Y$. Recall that a map $u:\tilde M\to Y$ is called $\rho$-[equivariant]{} if $$u(g\cdot x)=\rho(g)\cdot u(x)\quad\text{for all}\quad x\in\tilde M, \ g\in\Gamma.$$ For $\rho$-equivariant maps $u$ and $v$ the real-valued functions $d(u(x),v(x))$, where $x\in\tilde M$, are invariant with respect to the domain action and, hence, are defined on the quotient $M=\tilde M/\Gamma$. In particular, the quantity $$\label{L2d} d_2(u,v)=\left(\int_M d^2(u(x),v(x))\mathit{dVol}(x)\right)^{1/2}$$ defines a metric on the space of locally $L_2$-integrable $\rho$-equivariant maps. The latter can be also regarded as the $L_2$-width of a unique geodesic homotopy between $\rho$-equivariant maps. If $u$ is a locally Sobolev $W^{1,2}$-smooth $\rho$-equivariant map, then its energy density measure ${\left\lvertdu\right\rvert}^2\mathit{dVol}$ (see Sect. \[prem\]) is also $\Gamma$-invariant and the energy of $u$ is defined as the integral $$\label{E} E(u)=\int_M{\left\lvertdu\right\rvert}^2\mathit{dVol}.$$ Recall that the [ideal boundary]{} of $Y$ is defined as the set of equivalence classes of [asymptotic]{} geodesic rays, where two rays are asymptotic if they remain at a bounded distance from each other. Clearly, any action of $\Gamma$ by isometries on $Y$ extends to the action on the ideal boundary. \[t1\] Let $M$ be a compact Riemannian manifold without boundary and $Y$ be a locally compact Hadamard space. Let $\Gamma$ be the fundamental group of $M$ and $\rho:\Gamma\to\operatorname{Isom}(Y)$ be its representation whose image does not fix a point on the ideal boundary of $Y$. Then there exists a constant $C_\star$ such that for any $\rho$-equivariant locally $W^{1,2}$-smooth maps $u$ and $v$ the $L_2$-width of a geodesic homotopy $H$ between them satisfies the inequality $$\label{L2width} W_2(H)\leqslant C_\star(E^{1/2}(u)+E^{1/2}(v)).$$ The proof appears in Sect. \[proofs1\]. The idea is to prove first a similar inequality when one of the maps is an energy minimiser, and then to use compactness properties of the moduli space formed by such minimisers. The former is based on a compactness argument, mimicking the proof of the classical Poincaré inequality. Below we state a version of Theorem \[t1\] for equivariant maps of trees. First, we introduce more notation. Let $G$ be a finite connected graph without terminals and $\Gamma$ be its fundamental group $\pi_1(G)$. By $T$ we denote the universal covering tree of $G$; the group $\Gamma$ acts naturally on $T$ by the deck transformations. As above the symbol $\rho$ denotes a representation of $\Gamma$ in the isometry group of an Hadamard space $Y$. For a locally rectifiable $\rho$-equivariant map $u:T\to Y$, its length density measure ${\left\lvertdu\right\rvert}dt$ (see Sect. \[prem\]) is $\Gamma$-invariant and the length of $u$ is defined as the integral $$L(u)=\int_G{\left\lvertdu\right\rvert}dt.$$ \[t1g\] Let $G$ be a finite graph and $Y$ be a locally compact Hadamard space. Let $\Gamma$ be the fundamental group of $G$ and $\rho: \Gamma\to\operatorname{Isom}(Y)$ be its representation whose image does not fix a point on the ideal boundary of $Y$. Then there exists a constant $C_\star$ such that for any locally rectifiable $\rho$-equivariant maps $u$ and $v$ the $L_\infty$-width of a geodesic homotopy between them satisfies the inequality $$\label{L_infty_width} W_\infty(H)\leqslant C_\star(L(u)+L(v)).$$ Let $M'$ be a (not necessarily compact) Riemannian manifold whose sectional curvature is negative and bounded away from zero and the injectivity radius is positive. Let $\rho:\Gamma\to\pi_1(M')$ be a homomorphism whose image is neither trivial nor infinite cyclic. Then the latter does not fix a point on the ideal boundary of the universal cover of $M'$. Indeed, the group $\rho(\Gamma)$ is generated by hyperbolic elements (regarded as isometries of the universal cover), see [@KKS05 Lem. B.1], and the statement follows from the results in [@EO Sect. 6]. Thus, as a particular case, Theorem \[t1g\] contains the width inequality for homotopies between maps from $G$ to $M'$; this is the situtation considered in [@KKS05 Th. 0.1]. (Under the hypotheses on the homomorphism $\rho$, the homotopy class is neither trivial nor contains a map onto a closed curve.) The methods in [@KKS03; @KKS05] do not seem to yield an analogous $L_2$-width inequality (provided by Theorem \[t1\]) for non-compact targets when the dimension of the domain is greater than one. We proceed with width inequalities for representations in co-compact subgroups of $\operatorname{Isom}(Y)$. Recall that an action of a group $\Lambda$ on a metric space $(Y,d)$ is said to be [co-compact]{} if the quotient $Y/\Lambda$ is compact. Further, the action of $\Lambda$ is said to be [proper]{} if for each $y\in Y$ there exists $r>0$ such that the set $\{g\in\Lambda~\|~g\cdot B(y,r)\cap B(y,r)\ne\varnothing \}$ is finite. For a homomorphism $\rho:\Gamma\to\Lambda$ we denote by $Z$ below the centraliser of the image $\rho(\Gamma)$ in $\Lambda$. \[t2\] Let $M$ be a compact Riemannian manifold without boundary and $Y$ be a locally compact Hadamard space. Let $\Lambda$ be a group acting properly and co-compactly by isometries on $Y$. Denote by $\Gamma$ the fundamental group of $M$ and let $\rho$ be a homomorphism $\Gamma\to \Lambda$. Then there are constants $C_\star$ and $C$ such that for any $\rho$-equivariant locally $W^{1,2}$-smooth maps $u$ and $v$ there exists an element $h\in Z$ such that the $L_2$-width of a geodesic homotopy $H$ between $u$ and $h\cdot v$ satisfies the inequality $$W_2(H)\leqslant C_\star(E^{1/2}(u)+E^{1/2}(v))+C.$$ \[t2g\] Let $G$ be a finite graph and $Y$ be a locally compact Hadamard space. Let $\Lambda$ be a group acting properly and co-compactly by isometries on $Y$. Denote by $\Gamma$ the fundamental group of $G$ and let $\rho$ be a homomorphism $\Gamma\to\Lambda$. Then there are constants $C_\star$ and $C$ such that for any locally rectifiable $\rho$-equivariant maps $u$ and $v$ there exists an element $h\in Z$ such that the $L_\infty$-width of a geodesic homotopy $H$ between $u$ and $h\cdot v$ satisfies the inequality $$W_\infty(H)\leqslant C_\star(L(u)+L(v))+C.$$ If the homomorphism $\rho:\Gamma\to\Lambda$ in the theorems is trivial, then the second constant $C$ is equal to $\mathit{diam} (Y/\Lambda)\mathit{Vol}^{1/2}M$ and $\mathit{diam}(Y/\Lambda)$ in the $L_2$- and $L_\infty$-versions respectively. For non-trivial representations of $\Gamma$ it can be chosen to be zero. As a partial case, when the action of $\Lambda$ is free, Theorems \[t2\] and \[t2g\] above contain width inequalities for homotopies between continuous $W^{1,2}$-smooth maps valued in a compact metric space $Y/\Lambda$. The choice of an element $h\in Z$ in this setting corresponds to the choice of the homotopy between maps. Indeed, recall that the fundamental group of the space formed by continuous maps homotopic to $u:M\to Y/\Lambda$ is equal to the centraliser of the image $u_(\pi_1(M))$ in $\Lambda$. Conjugacies of finite lists in isometry groups ---------------------------------------------- Now we describe some applications of the width inequalities to geometric group theory. First, recall that a discrete subgroup $\Lambda$ in a Lie group $\mathbf G$ is called [lattice]{} if the quotient $\mathbf G/\Lambda$ carries a finite $\mathbf G$-invariant measure. Such a lattice is always finitely generated provided the group $\mathbf G$ is semi-simple and has $\operatorname{rank}\geqslant 2$; see ref. in [@LMR]. Choose a finite system of generators $(g_i)$ of $\Lambda$ and consider the word metric $d(\cdot,\cdot)$ on $\Lambda$ associated with the Cayley graph determined by the generators. Denote by ${\left\lvertg\right\rvert}$ the length $d(g,e)$, the distance between an element $g$ and the neutral element $e$. The following statements are essentially consequences of Theorems \[t1g\] and \[t2g\] and are explained in Sect. \[groups\]. \[group1\] Let $\mathbf G$ be a semi-simple Lie group of $\operatorname{rank}\geqslant 2$ all of whose simple factors are non-compact. Let $\Lambda$ be an irreducible lattice in $\mathbf G$ and $(a_i)_{1\leqslant i\leqslant N}$ be a finite list of elements in $\Lambda$ which does not fix a point on the ideal boundary of the associated symmetric space. Then for any conjugate (in $\Lambda$) list $(b_i)_{1\leqslant i\leqslant N}$ any conjugating element $g\in\Lambda$, $b_i=g^{-1}a_ig$, satisfies the inequality $${\left\lvertg\right\rvert}\leqslant C_\star\sum_{i=1}^N({\left\lverta_i\right\rvert}+{\left\lvertb_i\right\rvert})+C,$$ where the constants depend only on the conjugacy class of the lists. In particular, for such two given lists the set of conjugating elements is finite. An analogous statement holds if $\Lambda$ is an irreducible lattice in an almost simple $p$-adic algebraic Lie group of $\operatorname{rank}\geqslant 2$. In this case we consider lists which do not fix points on the ideal boundary of the associated Euclidean building. When the group $\mathbf G$ is algebraic, the hypothesis on the finite list $(a_i)$ is satisfied if, for example, the elements $a_i$’s generate a lattice (e.g., the whole group $\Lambda$) in $\mathbf G$. Indeed, by Borel’s density theorem the latter is Zariski dense in $\mathbf G$ and, hence, does not fix a point on the ideal boundary of the associated symmetric space. The estimate above yields immediately an algorithm deciding whether a given list of elements in $\Lambda$ is conjugate to the list $(a_i)$ in the theorem. This is a special case of the more general result due to Grunewald and Segal [@GS]: the conjugacy problem for finite lists in arithmetic groups is soluble. (Any irreducible lattice in a semi-simple Lie group of $\operatorname{rank}\geqslant 2$ is arithmetic, by the Margulis theorem.) However, we do not know whether the linear estimate for the length of the conjugating element holds under weaker hypotheses than in Theorem \[group1\]. We proceed with the conjugacy problem for finite lists in groups which act properly and co-compactly on Hadamard spaces by isometries. Recall that such groups are necessarily finitely presented; see [@BH99 I.8.11]. As above by ${\left\lvertg\right\rvert}$ we denote the length $d(g,e)$ in the word metric. \[group2\] Let $Y$ be a locally compact Hadamard space and $\Lambda$ be a group acting properly and co-compactly by isometries on $Y$. Then for any finite conjugate lists $(a_i)_{1\leqslant i\leqslant N}$ and $(b_i)_{1\leqslant i\leqslant N}$ of elements in $\Lambda$ there exists an element $g\in\Lambda$ with $b_i=g^{-1}a_ig$ such that $$\label{group_ineq} {\left\lvertg\right\rvert}\leqslant C_\star\sum_{i=1}^N({\left\lverta_i\right\rvert}+{\left\lvertb_i\right\rvert})+C,$$ where the constants depend only on the conjugacy class of the lists. Further, there exists an algorithm deciding whether two given finite lists of elements in $\Lambda$ are conjugate. When the list $(a_i)$ in the theorem consists of a single element, the solubility of the conjugacy problem is well-known. It is, for example, a consequence of an exponential (compare with our linear) bound for the length of the conjugating element in [@BH99 III.$\Gamma$.1.12]. In the context of decision problems it is worth noting that there are finitely presented groups in which the conjugacy problem for elements is soluble, but the conjugacy problem for finite lists is not. We refer to [@BH] for the explicit examples. Finally, mention that in [@BH] Bridson and Howie prove a closely related linear estimate for the length of the conjugating (two finite lists) element in Gromov hyperbolic groups. Preliminaries {#prem} ============= Sobolev spaces of maps to metric targets ---------------------------------------- We recall some background material on Sobolev spaces of maps valued in a metric space. The details can be found in [@KS]. Let $\Omega$ be a Riemannian domain and $(Y,d)$ be an arbitrary metric space. We suppose that $\Omega$ is endowed with a Lebesgue measure $\mathit{dVol}$ induced by the Riemannian volume. A measurable map $u:\Omega\to Y$ is called [locally $L_2$-integrable]{} if it has a seperable essential range and for which $d(u(\cdot),Q)$ is a locally $L_2$-integrable function on $\Omega$ for some $Q\in Y$ (and, hence, by the triangle inequality for any $Q\in Y$). If the domain $\Omega$ is bounded, then the function $$d_2(u,v)=\left(\int_\Omega d^2(u(x),v(x))\mathit{dVol}(x) \right)^{1/2}$$ defines a metric on the space of locally $L_2$-integrable maps. The latter is complete provided $Y$ is complete. The [approximate energy density]{} of a locally $L_2$-integrable map $u$ is defined for $\varepsilon>0$ as $$e_\varepsilon(u)(x)=\int_{S_\varepsilon(x)}\frac{d^2(u(x),u(x'))} {\varepsilon^{n+1}}\mathit{dVol}(x'),$$ where $S_\varepsilon(x)$ denotes the $\varepsilon$-sphere centred at $x$ and $n$ stands for the dimension of $\Omega$. The function $e_\varepsilon(x)$ is non-negative and locally $L_1$-integrable. The [energy]{} $E(u)$ of a locally $L_2$-integrable map $u$ is defined as $$E(u)=\sup_{0\leqslant f\leqslant 1}\left(\lim_{\varepsilon\to 0} \sup\int_\Omega fe_\varepsilon(u)\mathit{dVol}\right),$$ where the sup is taken with respect to compactly supported continuous functions which take values between $0$ and $1$. A locally $L_2$-integrable map $u$ is called [locally $W^{1,2}$-smooth]{} if for any relatively compact domain $D\subset\Omega$ the energy $E(\left.u\right\|_D)$ is finite. Due to the results of Korevaar and Schoen [@KS Sect. 1] a locally $L_2$-integrable map $u$ is locally $W^{1,2}$-smooth if and only if there exists a locally $L_1$-integrable function $e(u)$ such that the measures $e_\varepsilon(u)\mathit{dVol}$ converge weakly to the measure $e(u)\mathit{dVol}$ as $\varepsilon\to 0$. The function $e(u)$, also denoted by ${\left\lvertdu\right\rvert}^2$, is called the [energy density]{} of $u$, and the energy $E(u)$ is equal to the total mass $\int e(u)\mathit{dVol}$. Now suppose that the domain $\Omega$ is $1$-dimensional, that is an interval $I=(a,b)$. For a map $u:I\to Y$ one can also define the [ approximate length density]{} as $$l_\varepsilon(u)(t)=\frac{d(u(t),u(t+\varepsilon))+d(u(t), u(t-\varepsilon))}{\varepsilon},\qquad t\in I.$$ Then the length of $u$ is defined by the formula similar to that for the energy, $$L(u)=\sup_{0\leqslant f\leqslant 1}\left(\lim_{\varepsilon\to 0} \sup\int_I fl_\varepsilon(u)dt\right),$$ where the sup is taken with respect to compactly supported continuous functions. A map $u:I\to Y$ is called [rectifiable]{} if its length is finite. In this case there exists a [length density function]{} (or [speed function]{}) $l(u)$ such that the lenght $L(u)$ equals $\int l(u)dt$. Hadamard spaces --------------- Recall that an [Hadamard space]{} $(Y,d)$ is a complete metric space which satisfies the following two hypotheses: - [Length Space.]{} For any two points $y_0$ and $y_1\in Y$ there exists a rectifiable curve $\gamma$ from $y_0$ to $y_1$ such that $$d(y_0,y_1)=\mathit{Length}(\gamma).$$ We call such a curve $\gamma$ [geodesic]{}. - [Triangle comparison.]{} For any three points $P$, $Q$, and $R$ in $Y$ and the choices of geodesics $\gamma_{PQ}$, $\gamma_{QR}$, and $\gamma_{RP}$ connecting the respecting points denote by $\bar P$, $\bar Q$, and $\bar R$ the vertices of the (possibly degenerate) Euclidean triangle with side lengths $\ell(\gamma_{PQ})$, $\ell(\gamma_{QR})$, and $\ell(\gamma_{RP})$ respectively. Let $Q_\lambda$ be a point on the geodesic $\gamma_{QR}$ which is a fraction $\lambda$, $0\leqslant\lambda\leqslant 1$, of the distance from $Q$ to $R$; $$d(Q_\lambda,Q)=\lambda d(Q,R),\qquad d(Q_\lambda,R)=(1-\lambda)d(Q,R).$$ Denote by $\bar Q_\lambda$ an analogous point on the side $\bar Q\bar R$ of the Euclidean triangle. The triangle comparison hypothesis says that the metric distance $d(P,Q_\lambda)$ (from $Q_\lambda$ to the opposite vertex) is bounded above by the Euclidean distance ${\left\lvert\bar P-\bar Q_\lambda\right\rvert}$. This inequality can be written in the following form: $$\label{H0} d^2_{PQ_\lambda}\leqslant (1-\lambda)d^2_{PQ}+\lambda d^2_{PR}- \lambda(1-\lambda)d^2_{QR}.$$ It is a direct consequence of the property (ii) above that geodesics in an Hadamard space are unique. It is also a consequence of geodesic uniqueness that an Hadamard space has to be simply-connected [@BH99 II.1]. Examples include symmetric spaces of non-compact type and Euclidean buildings, simply-connected manifolds of non-positive sectional curvature, Hilbert spaces, simply-connected Euclidean or hyperbolic simplicial complexes satisfying certain local link conditions [@BH99 II.5.4]. Another class of examples is provided by the following proposition. \[L2maps\] Let $M$ be a compact Riemannian manifold without boundary and $(Y,d)$ be an Hadamard space. Let $\rho$ be a represenation of the fundamental group $\Gamma=\pi_1(M)$ in the group of isometries of $Y$. Then the space of $\rho$-equivariant locally $L_2$-integrable maps from $\tilde M$ to $Y$ endowed with the metric is an Hadamard space. The proof follows straightforward from the definitions: the geodesics in the new space are geodesic homotopies and the triangle comparison hypothesis follows by integration of relation . A useful consequence of the triangle comparison hypothesis is the following quadrilateral comparison property due to Reshetnyak [@Re] (we refer to [@KS Cor. 2.1.3] for a proof). Let $(Y,d)$ be an Hadamard space and $P$, $Q$, $R$, and $S$ be an ordered sequence of points in $Y$. For $0\leqslant\lambda,\mu\leqslant 1$ define $P_\lambda$ to be the point which is the fraction $\lambda$ of the way from $P$ to $S$ (on the geodesic $\gamma_{PS}$) and $Q_\mu$ to be the point which is the fraction $\mu$ of the way from $Q$ to $R$ (on the opposite geodesic $\gamma_{QR}$). Then for any $0\leqslant\alpha,t\leqslant 1$ the following inequality holds: $$\label{H1} d^2_{P_tQ_t}\leqslant (1-t)d^2_{PQ}+td^2_{RS}-t(1-t)\left[\alpha (d_{PS}-d_{QR})^2+(1-\alpha)(d_{RS}-d_{PQ})^2\right].$$ Setting $\alpha$ to be equal to zero in this inequality, we deduce the convexity of the distance between geodesics $$\label{H3} d_{P_tQ_t}\leqslant (1-t)d_{PQ}+td_{RS}.$$ This implies the following energy convexity property. Let $u$ and $v$ be locally $W^{1,2}$-smooth maps from the Riemannian domain $\Omega$ to an Hadamard space $(Y,d)$. Let $H(s,\cdot)$ be a geodesic homotopy between $u$ and $v$; the point $H(s,x)$ is the fraction $s$ of the way from $u(x)$ to $v(x)$, where $x\in\Omega$. Then for any $s$ the map $H(s,\cdot)$ is locally $W^{1,2}$-smooth and for any relatively compact domain $D\subset\Omega$ its energy satisfies the inequality $$\label{Econvex} E^{1/2}(H_s)\leqslant (1-s)E^{1/2}(u)+sE^{1/2}(v).$$ Inequality also yields the length convexity along geodesic homotopies. More precisely, let $u$ and $v$ be rectifiable paths in $(Y,d)$ and let $H(s,\cdot)$ be a geodesic homotopy between them parameterised by the arc-length as above. Then for any $s$ the map $H(s,\cdot)$ is rectifiable and its length satisfies the inequality $$\label{Lconvex} L(H_s)\leqslant (1-s)L(u)+sL(v).$$ Another consequence of the triangle comparison hypothesis is the existence of the [nearest point projection]{} $\pi:Y\to A$ onto a convex subset $A$. In more detail, if $(Y,d)$ is an Hadamard space and $A$ is its non-empty closed convex subset, then for any $y\in Y$ there exists a unique point $a\in A$ which minimises the distance $d(y,a)$ among all points in $A$; see [@KS Prop. 2.5.4]. Some properties of harmonic maps -------------------------------- Let $M$ be a compact Riemannian manifold without boundary and $(Y,d)$ be an Hadamard space. As above by $\Gamma$ we denote the fundamental group of $M$ and by $\rho:\Gamma\to\operatorname{Isom}(Y)$ its representation in the isometry group of $Y$. We consider $\rho$-equivariant locally $W^{1,2}$-smooth maps $u$ from the universal cover $\tilde M$ to $Y$. The energy density of such a map $u$ is a $\Gamma$-invariant function on $\tilde M$, which can be also regarded as a function on the quotient $M=\tilde M/\Gamma$. In particular, by the energy $E(u)$ we understand the integral $\int_Me(u)\mathit{dVol}$. We call a $\rho$-equivariant map [harmonic]{} if it minimises the energy among all $\rho$-equivariant locally $W^{1,2}$-smooth maps. The following statement is a straightforward consequence of the energy convexity, formula . We state it as a proposition for the convenience of references. \[har1\] Under the hypotheses above, let $u$ and $v$ be two $\rho$-equivariant harmonic maps and $H(s,\cdot)$ be a geodesic homotopy between them; the point $H(s,x)$ is the fraction $s$ of the way from $u(x)$ and $v(x)$, where $x\in\tilde M$. Then for each $s$ the map $H(s,\cdot)$ is also $\rho$-equivariant harmonic and the energy $E(H_s)$ does not depend on $s$. We proceed with the Lipschitz continuity of harmonic maps. The following proposition is a consequence of the result by Korevaar and Schoen [@KS Th. 2.4.6]. \[har2\] Under the hypotheses above, any $\rho$-equivariant harmonic map $u$ is Lipschitz continuous and its Lipschitz constant is bounded above by $C\cdot E^{1/2}(u)$, where the constant $C$ depends on the manifold $M$ and its metric only. Now let $G$ be a finite connected graph without terminals and $\Gamma$ be its fundamental group. By $T$ we denote the universal covering tree of $G$. Similarly to the discussion above, for a locally rectifiable $\rho$-equivariant map $u:T\to Y$ the length density function $l(u)$ is $\Gamma$-invariant and, hence, descends to the quotient $G=T/\Gamma$. In particular, by the length $L(u)$ we understand the integral $\int_G l(u)dt$. It is straightforward to see that if a map $u$ minimises the length among all locally rectifiable $\rho$-equivariant maps, then its restriction to every edge is a geodesic. If the latter has constant-speed parameterisation on every edge, then it is also harmonic and the length of every edge $u_I$ satisfies the relation $L^2(u)=E(u_I)(b-a)$, see [@EeFu Lemm. 12.5]. Conversely, if $u$ is a $\rho$-equivariant harmonic map, then its restriction to every edge is a constant-speed geodesic whose squared length is proportional to the energy as above. In particular, the length is constant on the set of $\rho$-equivariant harmonic maps, where it achieves its minimum. Proofs of the width inequalities {#proofs1} ================================ We start with the following lemma. Let $M$ be a compact Riemannian manifold without boundary and $(Y,d)$ be a locally compact Hadamard space. Let $\rho:\Gamma\to\operatorname{Isom}(Y)$ be a representation of the fundamental group $\Gamma=\pi_1(M)$. Suppose that the moduli space $\mathit{Harm}$, formed by $\rho$-equivariant harmonic maps, is non-empty and bounded in $L_2$-metric. Then there exists a positive constant $C_\star$ with the following property: for any $\rho$-equivariant locally $W^{1,2}$-smooth map $u$ there exists a harmonic map $\bar u\in\mathit{Harm}$ such that $$\label{MLineq} d_2(u,\bar u)\leqslant C_\star(E^{1/2}(u)-E^{1/2}_\star),$$ where $E_\star=E(\bar u)$ is the energy minimum among $\rho$-equivariant maps. First, note that inequality is invariant under the rescaling of the metric on the target space $Y$. Hence, it is sufficient to prove the lemma under the assumption that $$\label{scale} \text{the distance }d_2(u,\bar u)\text{ is not less than one.}$$ Suppose the contrary. Then there exists a sequence of maps $u_k$ such that for any $\bar u\in\mathit{Harm}$ $$d_2(u_k,\bar u)\geqslant k(E^{1/2}(u_k)-E^{1/2}(\bar u)).$$ For each $u_k$ choose a harmonic map $\bar u_k$ at which the infimum $$d_2(u_k,\bar u_k)=\inf\{d_2(u,v): v\in\mathit{Harm}\}$$ is attained. Such a harmonic map clearly exists: it is the value of $u_k$ under the nearest point projection onto $\mathit{Harm}$. (The lower semicontinuity of the energy [@KS Th. 1.6.1] and Prop. \[har1\] imply that $\mathit{Harm}$ is a closed convex subset in the Hadamard space of $\rho$-equivariant locally $L_2$-integrable maps.) Denote by $H^k_s$, where $s\in [0,1]$, a geodesic homotopy between $\bar u_k$ and $u_k$; we set $H^k_0=\bar u_k$ and $H^k_1=u_k$. Assuming that the parameter $s$ is proportional to the arc length, we obtain $$d_2(H^k_s,H^k_0)=s\cdot d_2(u_k,\bar u_k)\geqslant s\cdot k(E^{1/2}(u_k)-E^{1/2}(\bar u_k)).$$ Recall the energy $E^{1/2}(\cdot)$ is convex along geodesic homotopies; $$s(E^{1/2}(u_k)-E^{1/2}(\bar u_k))\geqslant E^{1/2}(H^k_s)-E^{1/2}(H^k_0).$$ Combining the last two inequalities we conclude that $$\label{c} d_2(H^k_s,H^k_0)\geqslant k(E^{1/2}(H^k_s)-E^{1/2}(H^k_0)).$$ Now choose a sequence of $s_k\in [0,1]$ such that the distance $d_2(H^k_{s_k},H^k_0)$ equals one; by the assumption this is possible. Then relation implies that the sequence $E(H^k_{s_k})$ converges to $E_\star$ as $k\to +\infty$. Since the moduli space $\mathit{Harm}$ is bounded in $L_2$-metric, the latter together with the choice of the $s_k$’s implies that the sequence $H^k_{s_k}$ is bounded in the $W^{1,2}$-sense; that is $$\label{star} d_2(H^k_{s_k},w)+E(H^k_{s_k})\leqslant C,$$ where $w$ is a fixed $\rho$-equivariant map. Now by the version of Rellich’s embedding theorem [@KS Th. 1.13] we can find a subsequence $H^k_{s_k}$ (denoted by the same symbol) which converges in $L_2$-metric and point-wise to a locally $W^{1,2}$-smooth map $\bar v$. By the lower semi-continuity of the energy [@KS Th. 1.6.1] the map $\bar v$ is energy minimising and by the point-wise convergence is $\rho$-equivariant. By the choice of the $s_k$’s we clearly have $$d_2(H^k_1,H^k_{s_k})=d_2(H^k_1,H^k_0)-d_2(H^k_{s_k},H^k_0)= d_2(u_k,\bar u_k)-1.$$ Thus, the $L_2$-distance between the maps $u_k$ and $v$ can be estimated as $$d_2(u_k,\bar v)\leqslant d_2(H^k_1,H^k_{s_k})+d_2(H^k_{s_k},\bar v) =d_2(u_k,\bar u_k)+(d_2(H^k_{s_k},\bar v)-1).$$ For sufficiently large $k$ the second term on the right-hand side is negative and we arrive at a contradiction with the choice of the harmonic maps $\bar u_k$’s. The following lemma summarises known results (essentially due to [@KS]) on the moduli space $\mathit{Harm}$, formed by $\rho$-equivariant maps. \[harm1\] Let $M$ be a compact Riemannian manifold without boundary and $Y$ be a locally compact Hadamard space. Let $\Gamma$ be the fundamental group of $M$ and $\rho:\Gamma\to\operatorname{Isom}(Y)$ be its representation whose image does not fix a point on the ideal boundary of $Y$. Then the moduli space $\mathit{Harm}$, formed by $\rho$-equivariant harmonic maps, is non-empty and compact in $C^0$-topology. Since there is no direct reference for the statement on the compactness of $\mathit{Harm}$ and to make our paper more self-contained, we give a proof now. First, we explain the existence of a $\rho$-equivariant harmonic map. By [@KS Th. 2.6.4] there exists an energy minimising sequence $\{u_i\}$ of equivariant Lipschitz continuous maps, whose Lipschitz constants are uniformly bounded. Let $\Omega$ be a fundamental domain for the action of $\Gamma$ on the universal cover $\tilde M$. We claim that under the hypotheses of the theorem the ranges $u_i(\Omega)$ are contained in a bounded subset of $Y$. Indeed, suppose the contrary. Then there exists a point $x\in\Omega$ such that the sequence $\{u_i(x)\}$ is unbounded in $Y$, i.e. $$d(u_i(x),Q)\to +\infty\quad\text{for some}\quad Q\in Y.$$ For any $g\in\Gamma$ consider the sequence $d(\rho(g)\cdot u_i(x),u_i(x))$. By the equivariance of the $u_i$’s and the uniform boundedness of their Lipschitz constants we have $$d(\rho(g)\cdot u_i(x),u_i(x))\leqslant Cd(g\cdot x,x),$$ and hence the quantities on the left hand side remain bounded as $i\to +\infty$. By the convexity of the distance between geodesics, relation , we see that the (Hausdorff) distances between the geodesic segments $\overline{Qu_i(x)}$ and $\rho(g)\cdot \overline{Qu_i(x)}$ also remain bounded as $i\to +\infty$. Since $Y$ is locally compact, we can find a subsequence of $u_i$, denoted by the same symbol, such that the segments $\overline{Qu_i(x)}$ converge on compact subsets to a geodesic ray $\sigma$ with initial point at $Q$. Then the distance between $\sigma$ and $\rho(g)\cdot\sigma$ is also bounded for any $g\in\Gamma$. This shows that $\sigma$ represents a fixed point for the action of $\rho(\Gamma)$ and leads to a contradiction. Now, since $Y$ is locally compact, the Arzela-Ascoli theorem applies and we can find a subsequence of $u_i$ converging in $C^0$-topology to an energy-minimising and, hence, harmonic map. Thus, the moduli space $\mathit{Harm}$ is non-empty. Finally, we explain the compactness of $\mathit{Harm}$. Let ${u_i}$ be a sequence of $\rho$-equivariant harmonic maps. By Prop. \[har1\] their energies coincide and Prop. \[har2\] the $u_i$’s are uniformly Lipschitz continuous. The same argument as above shows that the ranges $u_i(\Omega)$ are contained in a bounded subset of $Y$. Again by the Arzela-Ascoli theorem there exists a converging subsequence. By the lower semi-continuity of the energy the limit map is energy minimising and, hence, harmonic. Thus, the moduli space $\mathit{Harm}$ is compact in $C^0$-topology among $\rho$-equivariant harmonic maps. By Lemma \[harm1\] Main Lemma I applies: for given $\rho$-equivariant maps $u$ and $v$ we can find harmonic $\rho$-equivariant maps $\bar u$ and $\bar v$ such that $d_2(u,\bar u)$ and $d_2(v,\bar v)$ are estimated as in . By Lemma \[harm1\] the moduli space $\mathit{Harm}$ is compact and, hence, the distance between $d_2(\bar u,\bar v)$ is uniformly bounded. The $L_2$-width of a geodesic homotopy $H$ between $u$ and $v$ is the distance $d_2(u,v)$, and by the triangle inequality we have $$W_2(H)\leqslant d_2(u,\bar u)+d_2(\bar u,\bar v)+d_2(v,\bar v).$$ The second term is bounded, and the first and the last can be estimated as in ; thus, we obtain $$W_2(H)\leqslant C_\star(E^{1/2}(u)+E^{1/2}(v))+C,$$ where $C$ equals $\mathit{diam}(\mathit{Harm})-2C_\star E_\star^{1/2}$. Since, under the hypotheses of the theorem, the energy minimum $E_\star$ is positive, this inequality can be re-written in the form . Now we explain the proof of Theorem \[t1g\]; it follows essentially the same idea. First, we discuss the version of Main Lemma I. By $d_\infty(u,v)$ we denote below the maximum of the distance function between maps $u$ and $v$. Let $G$ be a finite graph and $(Y,d)$ be a locally compact Hadamard space. Let $\rho:\Gamma\to\operatorname{Isom}(Y)$ be a representation of the fundamental group $\Gamma=\pi_1(G)$. Suppose that the moduli space $\mathit{Harm}$, formed by $\rho$-equivariant harmonic maps $T\to Y$, is non-empty and compact in $C^0$-topology. Then there exists a positive constant $C_\star$ with the following property: for any continuous rectifiable $\rho$-equivariant map $u$ there exists a harmonic map $\bar u\in\mathit{Harm}$ such that $$\label{MLineq2} d_\infty(u,\bar u)\leqslant C_\star(L(u)-L_\star),$$ where $L_\star=L(\bar u)$ is the length minimum among $\rho$-equivariant maps. First, without loss of generality we may assume that the maps $u:T\to Y$ under consideration are such that their restrictions to every edge are parameterised proportionally to the arc-length. Second, as in the proof of Main Lemma I, it is sufficient to prove the lemma under the assumption that the distance $d_\infty(u,\bar u)$ is not less than one. Suppose the contrary. Then there exists a sequence of maps $u_k$ and harmonic maps $\bar u_k$ such that $$d_\infty(u_k,{\bar u}_k)\geqslant k(L(u_k)-L_\star);$$ we suppose that the ${\bar u}_k$’s minimise the distance $\{d_\infty(u_k,\bar u)$, where $u\in\mathit{Harm}\}$. Denote by $H^k_s$, where $s\in [0,1]$, a geodesic homotopy between $\bar u_k$ and $u_k$. Assuming that the parameter is proportional to the arc-length and using the convexity of the length, relation , we obtain $$d_\infty(H^k_s)\geqslant k(L(H^k_s)-L(H^k_0)).$$ Choosing a sequence $s_k\in [0,1]$ such that the left-hand side above equals to one, we conclude that $L(H^k_{s_k})$ converges to $L_\star$ as $k\to +\infty$. Since the lengths of $H^k_{s_k}$ are bounded and the edges of the $H^k_{s_k}$’s are parameterised proportionally to the arc-length, we see that the sequence of the $H^k_{s_k}$’s is equicontinuous. Further, the compactness of $\mathit{Harm}$ implies that the latter sequence is $d_\infty$-bounded. Now the Arzela-Ascoli theorem applies and there exists a subsequence converging in $d_\infty$-metric to a continuous map $\bar v$. The map $\bar v$ is clearly $\rho$-equivariant and length-minimising. Moreover, it has a constant-speed parametrisation and, hence, is harmonic. Now one gets a contradiction in the same way as in the proof of Main Lemma I. First, Lemma \[harm1\] carries over the case of $\rho$-equivariant maps of trees. In more detail, we need to start with a length minimising sequence which is uniformly Lipschitz continuous. The latter can be constructed by re-parameterising any length minimising sequence proportionally to the arc-length on every edge. The rest of the proof (of Lemma \[harm1\]) carries over without essential changes. Now we simply follow the lines in the proof of Theorem \[t1\] and use Main Lemma II instead of Main Lemma I. We proceed with the proofs of Theorems \[t2\] and \[t2g\]. First, recall some notation. Let $\Lambda$ be a group acting properly and co-compactly by isometries on $Y$. For a homomorphism $\rho:\Gamma \to\Lambda$ by $Z$ we denote the centraliser of the image $\rho(\Gamma)$ in $\Lambda$. The group $Z$ acts naturally on the space of $\rho$-equivariant maps $u:\tilde M\to Y$ and, in particular, on the moduli space $\mathit{Harm}$. \[harm2\] Under the hypotheses of Theorem \[t2\], the moduli space $\mathit{Harm}$, formed by $\rho$-equivariant harmonic maps, is non-empty and the quotient $\mathit{Harm}/Z$ is compact in $C^0$-topology. We start with the existence of a $\rho$-equivariant harmonic map. By [@KS Th. 2.6.4.] there exists an energy minimising sequence $\{u_i\}$ of equivariant Lipschitz continuous maps, whose Lipschitz constants are uniformly bounded. Let $\Omega$ and $D$ be fundamental domains for the actions of $\Gamma$ on $\tilde M$ and $\Lambda$ on $Y$ respectively. Fix a point $x_\in\Omega$. Then there exists a sequence of elements $h_i\in\Lambda$ such that the maps $h_i\cdot u_i$ send $x_$ into the closure of $D$. Since the $h_i$’s are isometries, the sequence $\{h_i\cdot u_i\}$ is also energy minimising and uniformly Lipschitz continuous. Moreover, since $\Lambda$ acts co-compactly, its fundamental domain $D$ is bounded, and the uniform Lipschitz continuity implies that the ranges $h_i\cdot u_i(\Omega)$ are contained in a bounded subset of $Y$. By the Arzela-Ascoli theorem there exists a subsequence, also denoted by $h_i\cdot u_i$, converging to a limit map $v$. Now we define a homomorphism $\varphi:\Gamma\to\Lambda$ such that the limit map $v$ is $\varphi$-equivariant. For this fix a generator $g\in\Gamma$ and consider the points $$v(g\cdot x)=\lim (h_i\cdot u_i)(g\cdot x)\quad\text{and}\quad v(x)=\lim (h_i\cdot u_i)(x),$$ where $x\in\Omega$. The triangle inequality implies that $$(h_i\rho(g)h_i^{-1})\cdot v(x)\to v(g\cdot x)\quad\text{as} \quad i\to+\infty.$$ Now, since the action of $\Lambda$ is proper, the sequence $h_i\rho(g)h_i^{-1}$ contains a constant subsequence; we denote it value by $\varphi(g)\in\Lambda$. We use the $h_i$’s of this subsequence for the same procedure for another generator in $\Gamma$. Repeating the process we define $\varphi$ on all generators. It then extends as a homomorphism $\varphi:\Gamma\to\Lambda$ and the map $v$ is $\varphi$-equivariant. As a result of this procedure, we also have a sequence $h_i\in\Lambda$ such that $$h_i\rho(g)h_i^{-1}=\varphi(g)\quad\text{for any}\quad g\in\Gamma.$$ This identity implies that the $h_i$’s can be written in the form $k\cdot\bar{h_i}$, where $\bar{h_i}\in Z$, and the element $k\in\Lambda$ conjugates $\rho$ and $\varphi$. Now, since the sequence $h_i\cdot u_i$ converges to $v$, the sequence $\bar{h_i}\cdot u_i$ converges to $k^{-1}v$. Moreover, the latter is energy minimising and is formed by $\rho$-equivariant maps. Thus, the limit map $k^{-1}v$ is a harmonic $\rho$-equivariant map and the existence is demonstrated. The compactness of $\mathit{Harm}/Z$ follows by the same argument as above with the substitution of the sequence of harmonic maps for the energy minimising sequence $\{u_i\}$. By Prop. \[har1\] the former sequence is also energy minimising, and by Prop. \[har2\] is uniformly Lipschitz continuous; the argument above yields a sequence $\bar{h_i}\in Z$ such that $\bar{h_i}\cdot u_i$ converges to a $\rho$-equivariant harmonic map. Let $\mathcal H$ be a fundamental domain for the action of $Z$ on the moduli space $\mathit{Harm}$. First, Main Lemma I holds under a weaker hypothesis than the $L_2$-boundedness of $\mathit{Harm}$. More precisely, it is sufficient to assume that the domain $\mathcal H$ is bounded in the $L_2$-metric. Indeed, since the group $Z$ acts by isometries, one can suppose that the maps $\bar u_k$’s (in the proof of Main Lemma I) belong to $\mathcal H$. The boundedness of the latter is then used to obtain the $W^{1,2}$-boundedness of the sequence $H^k_{s_k}$, relation . The rest of the proof stays unchanged. Now the combination of Lemma \[harm2\] and estimate yields the statement in the fashion similar to the proof of Theorem \[t1\]. First, Main Lemma II holds under a weaker hypothesis than the compactness of the moduli space $\mathit{Harm}$. Similarly to the above, it is sufficient to assume that a fundamental domain for the action of $Z$ on $\mathit{Harm}$ is compact. Further, Lemma \[harm2\] carries over the case of $\rho$-equivariant maps of trees; the proof follows essentially the same line of argument. The combination of this version of Lemma \[harm2\] with estimate yields the statement in the same fashion as above. Finitely generated subgroups in isometry groups {#groups} =============================================== Recall that the action of a group $\Lambda$ on a metric space $(Y,d)$ by isometries defines an orbit pseudo-metric on $\Lambda$: $$d_y(g,h)=d(g\cdot y,h\cdot y),\quad\text{where}\quad g,h\in\Lambda,$$ and $y\in Y$ is a fixed reference point. For another point $\bar y\in Y$ the pseudo-metrics $d_y$ and $d_{\bar y}$ are coarsely isometric; that is there exists a constant $C$ ($=2d(y,\bar y)$) such that $$d_{\bar y}(g,h)-C\leqslant d_y(g,h)\leqslant d_{\bar y}(g,h)+C.$$ First, we show that the $L_\infty$-width inequalities imply an estimate for the conjugating element in the orbit pseudo-metric. \[orbit1\] Let $\mathbf{G}$ be a semi-simple Lie group all of whose simple factors are non-compact. Let $\Lambda$ be an irreducible lattice in $\mathbf{G}$ and $(a_i)_{1\leqslant i\leqslant N}$ be a finite list of elements in $\Lambda$ which does not fix a point on the ideal boundary of the associated symmetric space. Then for any conjugate (in $\Lambda$) list $(b_i)_{1\leqslant i\leqslant N}$ any conjugating element $g\in\Lambda$, $b_i=g^{-1}a_ig$, satisfies the inequality $$d_y(g,e)\leqslant C_\star\sum_{i=1}^{N}(d_y(a_i,e)+d_y(b_i,e)),$$ where $y\in Y$ is a reference point, and the constant depends only on the conjugacy class of the list $(a_i)$. Let $Y$ be a symmetric space associated with the Lie group $\mathbf{G}$. Under the hypotheses on $\mathbf{G}$, the natural $\mathbf{G}$-invariant Riemannian metric on $Y$ defines a distance $d$ which makes $Y$ into an Hadamard space. Consider the bouqet of $N$ copies of a circle; denote by $\Gamma=\oplus_{i=1}^{N}\mathbf Z$ its fundamental group and by $T$ its universal cover. Define a homomorphism $\rho:\Gamma\to\Lambda$ by the rule: the generator of the $i$th copy of $\mathbf{Z}$ maps into $a_i$. For a fixed reference point $y\in Y$ consider the graph in $Y$ whose vertices are points $g\cdot y$, where $g$ is a word in the alphabet $(a_i)$. The edges are geodesic arcs; two points $g_1\cdot y$ and $g_2\cdot y$ are joined by an edge if and only if $g_1^{-1}g_2$ is an element $a_i$ or its inverse. Suppose that each edge is parameterised proportionally to the arc-length. Such a parametrisation defines a $\rho$-equivariant map $u:T\to Y$, whose length $L(u)$ is given by the sum $\sum_{i=1}^Nd(a_iy,y)$. Analogously, for a conjugate list $(b_i)_{1\leqslant i\leqslant N}$ one defines a $(g^{-1}\rho g)$-equivariant map $v:T\to Y$, where $g$ is a conjugating element. Note that the map $g\cdot v$ is $\rho$-equivariant and its length $L(g\cdot v)$ coincides with $L(v)= \sum_{i=1}^Nd(b_iy,y)$. By the hypotheses of the lemma, Theorem \[t1g\] applies and we have $$d(g\cdot y,y)\leqslant W_\infty(H)\leqslant C_\star(L(u)+L(v)),$$ where $H$ is a homotopy between $u$ and $g\cdot v$. Now the combination with the expressions for the lengths finishes the proof. The statement is a direct consequence of Lemma \[orbit1\] and the solution of Kazhdan’s conjecture in [@LMR]. The latter says that the word metric (with respect to some finite set of generators) on an irreducible lattice $\Lambda$ is quasi-isometric to the orbit metric (with respect to the action on the associated symmetric space or Euclidean building) provided $\mathbf{G}$ is semi-simple and its $\operatorname{rank}\geqslant 2$. \[orbit2\] Let $Y$ be a locally compact Hadamard space and $\Lambda$ be a group acting properly and co-compactly by isometries on $Y$. Then for any finite conjugate lists $(a_i)_{1\leqslant i\leqslant N}$ and $(b_i)_{1\leqslant i\leqslant N}$ of elements in $\Lambda$ there exists an element $g\in\Lambda$ with $b_i=g^{-1}a_ig$ such that $$d_y(g,e)\leqslant C_\star\sum_{i=1}^{N}(d_y(a_i,e)+d_y(b_i,e))+C,$$ where $y\in Y$ is a reference point, and the constants depend only on the conjugacy class of the list $(a_i)$. The proof follows the same line of argument as the proof of Lemma \[orbit1\] with the use of Theorem \[t2g\] instead of Theorem \[t1g\]. By Švarc-Milnor lemma [@BH99 I.8.19] the word and orbit metrics on $\Lambda$ are quasi-isometric. The combination of this with Lemma \[orbit2\] implies the first statement of the theorem. Further, by [@BH99 III.$\Gamma$.1.4] the word problem in $\Lambda$ is soluble. This yields the algorithm deciding the conjugacy of finite lists in the following fashion. If there exists an element conjugating two given lists, then it belongs to the finite subset of $\Lambda$ formed by elements satisfying the bound . Using the solubility of the word problem, the algorithm checks all elements from this finite set. [99]{} Bridson, M., Haefliger, A. [Metric spaces of non-positive curvature.]{} Grundlehren der Mathematischen Wissenschaften, 319. Springer-Verlag, Berlin, 1999. xxii+643 pp. Bridson, M., Howie, J. [Conjugacy of finite subsets in hyperbolic groups.]{} Internat. J. Algebra Comput. [15]{} (2005), 725–756. Eberlein, P., O’Neill, B. [Visibility manifolds.]{} Pacific J. Math. [46]{} (1973), 45–109. Eells, J., Fuglede, B. [Harmonic maps between Riemannian polyhedra.]{} Cambridge Tracts in Mathematics, [142]{}. Cambridge University Press, Cambridge, 2001. xii+296 pp. Grunewald, F., Segal, D. [Some general algorithms. I. Arithmetic groups.]{} Ann. of Math. (2) [112]{} (1980), 531–583. Kappeler, T., Kuksin, S., Schroeder, V. [ Perturbations of the harmonic map equation.]{} Commun. Contemp. Math. [5]{} (2003), 629–669. Kappeler, T., Kuksin S., Schroeder, V. [Poincaré inequalities for maps with target manifold of negative curvature.]{} Mosc. Math. J. [5]{} (2005), 399–414, 494. Kokarev G. [The property of compactness of a quasilinearly perturbed equation of harmonic mappings.]{} Sbornik: Mathematics [194]{} (2003), 1055–1068 Kokarev G. [A note on Morse inequalities for harmonic maps with potential and their applications.]{} Ann. Global Anal. Geom., to appear. Korevaar, J., Schoen, R. [Sobolev spaces and harmonic maps for metric space targets.]{} Comm. Anal. Geom. [1]{} (1993), 561–659. Lubotzky, A., Mozes, S., Raghunathan, M. S. [The word and Riemannian metrics on lattices of semisimple groups.]{} Inst. Hautes Études Sci. Publ. Math. No. 91 (2000), 5–53 (2001). Rešetnjak, Yu. [Non-expansive maps in a space of curvature no greater than $K$.]{} Sibirsk. Mat. Ž. [9*]{} (1968), 918–927.	{ "pile_set_name": "ArXiv" }
--- abstract: \| Security properties are often focused on the technological side of the system. One implicitly assumes that the users will behave in the right way to preserve the property at hand. In real life, this cannot be taken for granted. In particular, security mechanisms that are difficult and costly to use are often ignored by the users, and do not really defend the system against possible attacks. Here, we propose a graded notion of security based on the complexity of the user’s strategic behavior. More precisely, we suggest that the level to which a security property $\varphi$ is satisfied can be defined in terms of (a) the complexity of the strategy that the voter needs to execute to make $\varphi$ true, and (b) the resources that the user must employ on the way. The simpler and cheaper to obtain $\varphi$, the higher the degree of security. We demonstrate how the idea works in a case study based on an electronic voting scenario. To this end, we model the vVote implementation of the [Prêt à Voter]{}voting protocol for coercion-resistant and voter-verifiable elections. Then, we identify “natural” strategies for the voter to obtain receipt-freeness, and measure the voter’s effort that they require. We also look at how hard it is for the coercer to compromise the election through a randomization attack. Keywords: electronic voting, coercion resistance, natural strategies, multi-agent models, graded security author: - Wojciech Jamroga - Damian Kurpiewski - Vadim Malvone bibliography: - 'wojtek.bib' - 'wojtek-own.bib' title: Natural Strategic Abilities in Voting Protocols --- Introduction {#sec:intro} ============ Methodology {#sec:methodology} =========== Use Case Scenario: {#sec:usecase} =================== Models {#sec:models} ====== Strategies and Their Complexity {#sec:strats} =============================== Automated Verification of Strategies {#sec:verification} ==================================== Conclusions {#sec:conclusions} ===========	{ "pile_set_name": "ArXiv" }
--- author: - \| Piotr T. ŻYCKI and Chris DONE\ [Department of Physics, University of Durham, South Road, Durham DH1 3LE, England, U.K.]{}\ David A. SMITH\ [Department of Physics and Astronomy, University of Leicester, University Road, Leicester LE1 7RH, England, U.K.]{} title: 'On accretion flow in Soft X-ray Transients' --- 34.5pc Abstract {#abstract .unnumbered} ======== We have analyzed archival [Ginga]{} data of a number of Soft X-ray Transient sources and modelled them in an attempt to constrain the geometry and physical properties of accretion. Our models include a self-consistent description of X-ray reprocessing, linking the properties of the reflected continuum and those of the iron fluorescence K$\alpha$ line, and allowing for relativistic smearing of spectral features. We derive constraints on the inner extent of the disk and its ionization, assuming a generic geometry of a central X-ray source and an external accretion disk. Fitting our models to the data we have found that in some sources – GS 1124-68 (Nova Muscae 1991) and GS 2000+25 – the evolution during decline proceeded in qualitative agreement with recent theoretical models based on advection-dominated solutions of accretion flow. These models link the high/soft–low/hard transition with a change of geometry, most importantly, the inner radius of the optically thick disk. Quantitatively, our analysis requires serious revision of the model assumptions. A remarkably exceptional case is that of GS 2023+338, where the source behavior both during outburst and decline was quite different than the other, more “standard” ones. Introduction ============ Accretion flow around black holes in compact objects (both active galaxies and stellar sources) clearly proceeds through at least two phases: optically thick, cooler plasma producing a soft thermal component, and optically thin, hot plasma where hard X-rays/$\gamma$-rays are produced (see Mushotzky et al. 1993 and Tanaka & Lewin 1995 for reviews). The geometrical configuration of the two phases is still uncertain. Simple possibilities include pure radial and pure vertical stratification of the flow, but more complicated geometries, with partial overlap are also considered. Localized active regions above the disk, appearing due to e.g. magnetic activity, are also considered (Galeev et al. 1979; Stern et al. 1995). The commonly made assumption that the observed hard X-ray spectra are due to the inverse Compton process requires an input of soft photons from the optically thick plasma to the optically thin one. The reverse interaction must then also take place: the hard X-rays will illuminate the optically thick plasma (Guilbert & Rees 1988) and they will be reflected and reprocessed. As a result, a continuum spectral component is formed by Compton reflection, with spectral features due to emission/absorption by heavy elements (primarily iron) superimposed on it. Studying the effects of such reprocessing can be extremely useful as a tool for constraining the spatial distribution of the two phases of accreting plasma. The amplitude of the reflected component constrains the solid angle subtended by the optically thick plasma from the X-ray source. The shape of the reflected continuum below $\sim 10$ keV, and the energy and strength of iron spectral features around 7 keV give information about ionization state of the reflecting plasma. Possible broadening/smearing of spectral features can constrain the plasma velocity field if it can be distinguished from e.g. broadening due to Comptonization. Black hole Soft X-ray Transients (SXT) are particularly suited to this kind of analysis. Those are a sub-class of X-ray binaries that undergo occasional, dramatic outbursts. During the subsequent decline phase the mass accretion rate, which is thought to be the most important parameter, decreases roughly exponentially, with a typical e-folding time scale of $30-60$ days. During the decline they go through a sequence of well defined states characterized by distinct spectral and timing properties (review in Tanaka & Shibazaki 1996). Data ==== For our analysis we use archival [Ginga]{} data of a number of SXT. The data combine broad band (1 to 20–30 keV) with moderate spectral resolution (18% at 6 keV), very good statistics due to large effective area of the LAC detector ($\sim 4000\,{\rm cm^{-2}}$) and systematic uncertainties of the instrument below 1%. Background subtraction can pose a problem for bright sources, due to the contamination of the background monitors by source counts. We have developed a method to estimate the background, described in some detail in Życki et al. (1998b). Models ====== Our model of X-ray reprocessing, described in details in Życki et al. (1998b), self-consistently combines the model for Compton–reflected continuum and Monte Carlo computations of the iron K$\alpha$ line. The main model parameter is the ionization parameter, $\xi\equiv 4\pi F/n$, where $F$ is the illuminating flux and $n$ electron density. The amplitude of the component is expressed as a solid angle of the reflector, normalized to $2\pi$, $f\equiv \Omega/2\pi$. The effects of relativistic smearing are computed convolving the total reprocessed spectrum with relativistic line profiles implemented in [XSPEC]{} as the “diskline” model. The main parameter here is the inner disk radius in units of $R_{\rm g} \equiv G M/c^2$, assuming a fixed illumination emissivity (we assume $F_{\rm irr}(r) \propto r^{-3}$). Figure \[zds:fig\_reproc\] shows typical model spectra. Results ======= Low/hard state -------------- The results are shown in Table \[zds:tab\_fits\] The reprocessed component is significantly present in the spectra. Its amplitude is significantly smaller than 1, typical values being 0.3 – 0.7. It is rather weakly ionized, $\xi < 50 $ corresponding to mean ionization level of iron $<$FeXVII i.e. only M-shell electrons removed. Relativistic smearing of the reprocessed component is significant in some of the spectra, and is consistent with decreasing in time, as the sources progressed through the decline phase. Even at the beginning of decline in GS 2023+338 the smearing is less than expected from a disk extending down to the last stable orbit at $r_{\rm ms} = 6\,R_{\rm g}$, typical values being $R_{\rm in} \sim 20-30 \,R_{\rm g}$ (see Figure \[zds:fig\_HS\_LS\] and Życki et al. 1997), although $R_{\rm in}$ is compatible with $r_{\rm ms}$ in Nova Muscae (Życki et al. 1998a). sp. st. Source obs. date $\Gamma$ $f$ $\xi$ $R_{\rm in}\ (R_{\rm g})$ $\chi^2$/dof --------- --------- ----------- ------------------------- ------------------------- ----------------------------------- --------------------------- -------------- LS GS 1124 23/07/91 $1.72\pm 0.02$ $0.24^{+0.11}_{-0.08}$ $17^{+40}_{-16}$ $ 50^{+\infty}_{-35} $ 15.3/24 LS GS 2023 19/06/89 $1.72\pm 0.03$ $0.66^{+0.11}_{-0.07} $ $0^{+1} $ $ 18^{+13}_{-7}$ 23.1/23 LS GS 2000 16/12/88 $1.93^{+0.14}_{-0.08} $ $0.24^{+0.80}_{-0.21} $ $ 100^{+5\times 10^4}_{-100} $ – 20/24 HS GS 1124 18/05/91 $2.29^{+0.05}_{-0.03} $ $0.64^{+0.40}_{-0.10} $ $(3.5^{+9.5}_{-3.0})\times 10^4 $ $ 18^{+22}_{-6} $ 13.5/22 HS GS 2000 18/10/88 $2.00^{+0.26}_{-0.03} $ $0.45^{+0.45}_{-0.11} $ $(15^{+45}_{-14.6})\times 10^3$ $ 7^{+2}_{-1} $ 15/21 VHS GS 1124 11/01/91 $2.02^{+0.16}_{-0.21} $ $0.35^{+0.25}_{-0.09} $ $(10^{+10}_{-8})\times 10^3 $ $ 13^{+6}_{-3} $ 26.6/31 VHS(?) GS 2023 30/05/89 $1.70\pm 0.01 $ $0.17^{+0.16}_{-0.04} $ $(6^{+18}_{-4})\times 10^3 $ $ 6^{+2} $ 25.7/27 : Results of fitting the reprocessing model to the data[]{data-label="zds:tab_fits"} High/soft state --------------- In the high state spectra the major difference in the reprocessed component is its strong ionization (Table \[zds:tab\_fits\]). Dominant iron ions are usually He- and H-like ones. Consequently, the K$\alpha$ line and absorption edges are much stronger in the composite spectra. The amplitude $f$ is also larger, close to 1. Relativistic smearing is significant, but the values of $R_{\rm in}$ differ between different objects. The fact that, for GS 2000+25, $R_{\rm in} \approx 6$ (which is not expected given the assumed $F_{\rm irr} \propto r^{-3}$ i.e. without the boundary condition term; Shakura & Sunyaev 1973) may suggest that an additional broadening of that component (e.g. due to Comptonization) takes place. We find that the soft component of the continuum spectrum cannot be described by either pure black body or multi-temperature (disk) blackbody spectrum. Additional power law tail is required, which can be modelled as additional Comptonization of a (disk) black body. This differs from results of Ebisawa et al. (1994) who were able to fit the disk blackbody to the spectra but they were using the smeared edge model with a narrow gaussian to model the iron spectral features, rather than a reprocessing model. Very high state --------------- Nova Muscae showed an example of Very High State spectrum at the peak of its outburst (Jan–Feb 1991). On 11th Jan the soft continuum component again cannot be modelled as a (disk) blackbody, so we assume a Comptonized blackbody instead (Figure \[zds:fig\_gs\], middle panel). The presence of the reprocessed component is then again highly statistically significant although its amplitude is not well constrained from above, in particular if radial distribution of ionization is allowed for. The reflection is strongly ionized and smeared, similarly to high state spectra (Table \[zds:tab\_fits\]). Thus although the overall spectrum is dominated by the hard power law ($\Gamma\sim 2$) as in LS, the properties of both the soft component and the reprocessing are similar to HS, so the suggested histeretic behavior of SXT (Miyamoto et al. 1995) does not seem to be complete. The exceptional case of GS 2023+338 =================================== GS 2023+338 was clearly an unusual source, showing fast and chaotic flux and spectral variability, and no clear soft, thermal component even when its luminosity was very high (Tanaka & Lewin 1995). The spectrum of GS 2023+338 at the peak of its outburst was rather unusual: it can be described by an optically thick comptonization of a disk blackbody radiation of rather high temperature, $k T_0 \approx 1.4\,$keV in a rather cool plasma cloud, $k T_{\rm e}\approx 10\,$keV, $\tau \approx 6$. Comparison with spectra of Nova Muscae at a similar luminosity level (Figure \[zds:fig\_gs\]) suggests that the actual mass accretion rate in GS 2023 might have been super-Eddington. The hypothesis that the observed emission is a (quasi-thermal) disk radiation is not however easily compatible with the presence of spectral features due to highly ionized iron, since in such a geometry disk irradiation is rather ineffective. The short time-scale ($\sim$few sec) flux and spectral variability can be attributed to strong and rapidly variable photo–electric absorption. During one the episodes of such strong absorption on 30th May, the source spectrum is consistent with a presence of a soft thermal component of temperature $T\approx 1\,$keV, and a steep power law tail, $\Gamma \approx 2$. Such spectra are typically observed in high/soft states of GBH (Figure \[zds:fig\_soft\]; Życki et al. 1998c). During its decline from outburst the source showed a fairly regular, exponential decline of the luminosity ($t_0\sim 30\,$ days) with again rapid variability due to photo-electric absorption superimposed on it. The amplitude of the reprocessed component decreases with time, in rough correlation with decreasing level of relativistic smearing, i.e. the behavior is consistent with $R_{\rm in}$ increasing. However, the power law spectral index remains practically constant, in sharp contrast to the behavior of GS 1124-68, GS 2000+25 and GX 339 (Życki et al. 1998b). Discussion ========== The results presented in previous Sections clearly suggest that the geometry of accretion and the physical state of reprocessing material change as sources make transition between hard and soft states. The most significant changes to the reprocessing are those of ionization of the reflector and the amplitude of reflection. In the low/hard state the solid angle of the reprocessor from the X-ray source is smaller and the reprocessor is colder than in the high/soft state. The amount of relativistic smearing does not show clear, statistically significant changes between the transitions, but $R_{\rm in}$ is consistent with increasing as the amplitude $f$ decreases. We can compare the above results with predictions of models proposed for the spectral transitions. The observed changes are qualitatively expected in the models of Mineshige (1996) and Esin et al. (1997). These models assume predominantly radial stratification of the flow, with optically thick disk extending only outside certain transition radius. The central, hot solution derives from advection dominated accretion flow solutions although when applied to typical low state, it has radiative efficiency of $\sim 0.35$. The transition radius (which can be identified with our $R_{\rm in}$) has to decrease with increasing $\dot m$ to explain softer spectra when sources are brighter, and this also explains larger amplitude of the reprocessed component and its strong ionization in high state. The behavior of $R_{\rm in}$ we find from relativistic smearing is also consistent with that prediction. Thermal radiation emitted by the disk can by Comptonized by hot upper layer of the disk, giving rise to the observed steep power law tail. Quantitatively, however, the values of $R_{\rm in}$ we derive are rather smaller than those postulated by Esin et al. In particular, in the low state we derive $R_{\rm in} \sim 20-50 \, R_{\rm g}$, while they assume $2\times 10^4 \, R_{\rm g}$. Esin et al. (1998) argue that the smaller values are still compatible with their model but this claim has not been verified by actual data modelling. Relatively small values of $R_{\rm in}$ we derive indicate that there must exist significant overlap between the central hot X-ray source and external cool disk i.e. the hot source has to extend much further out than $R_{\rm in}$: otherwise the energy release within the source would be much too small to account for the overall energetics and spectral shape (Gierliński et al. 1997; review in Poutanen 1998). Perhaps the hot plasma forms a corona at large distances. The transition between the disk and the corona at large distances could result from e.g. the thermal instability of X-ray irradiated plasma, as considered by Witt et al. (1997). An alternative geometry was proposed by Di Matteo et al. (1998). They suggest that magnetic flares above the disk produce the hard X-ray radiation, the disk itself always extends to the last stable orbit, and the hard/low – soft/high transition is due to changes in characteristic height of the flares. In order to account for harder spectra in the low state they postulate that the flares are higher above the disk in this state. This however does not naturally explain why the reflection amplitude is smaller in the hard state unless the flares are strongly concentrated towards the inner edge of the disk, so a significant fraction of photons is lost within the central hole. An alternative explanation, that the inner disk is strongly ionized and so the iron spectral features are weak, does not seem to be supported by data analysis, at least in Cyg X-1 (Done & Życki 1998). References ========== Di Matteo T. et al. 1998, MNRAS, submitted (astro-ph/9805345) Done C., Życki P. T. 1998, MNRAS, submitted Ebisawa K. et al. 1994, PASJ 46, 375 Esin A. A., McClintock J. E., Narayan R. 1997, ApJ 489, 865 Esin A. A., Narayan R., Cui W., Grove J. E., Zhang S.-N. 1998, ApJ 505, 854 Galeev A. A., Rosner R., Vaiana G. S. 1979, ApJ 229, 318 Gierliński M. et al. 1997, MNRAS 288, 958 Guilbert P. W., Rees M. J. 1988, MNRAS 233, 475 Mineshige S. 1996, PASJ 48, 93 Miyamoto S. et al. 1995, ApJ 442, L13 Mushotzky R. F., Done C., Pounds K. A. 1993, ARA&A 31, 717 Poutanen J. 1998, in Theory of Black Hole Accretion Discs, eds. M. A. Abramowicz, G. Björnsson, J. E. Pringle (CUP, Cambridge) (astro-ph/9805025) Shakura N. I., Sunyaev R. A. 1973, A&A 24, 337 Stern B. E. et al. 1995, ApJ 449, L13 Tanaka Y., Lewin W. H. G. 1995, in X–Ray Binaries, eds. W. H. G. Lewin, J. van Paradijs, E. van den Heuvel, (CUP, Cambridge) Tanaka Y., Shibazaki N. 1996, ARA&A 34, 607 Witt H. J., Czerny B., Życki P. T. 1997, MNRAS 288, 848 Życki P. T., Done C., Smith D. A. 1997, ApJ 488, L113 Życki P. T., Done C., Smith D. A. 1998a, ApJ 496, L25 Życki P. T., Done C., Smith D. A. 1998b, MNRAS, submitted (astro-ph/9811106) Życki P. T., Done C., Smith D. A. 1998c, in preparation	{ "pile_set_name": "ArXiv" }
--- abstract: \| We give a complete characterization of mixed unit interval graphs, the intersection graphs of closed, open, and half-open unit intervals of the real line. This is a proper superclass of the well known unit interval graphs. Our result solves a problem posed by Dourado, Le, Protti, Rautenbach and Szwarcfiter (Mixed unit interval graphs, Discrete Math. 312, 3357-3363 (2012)). [Keywords:]{} unit interval graph; proper interval graph; intersection graph author: - Felix Joos title: A Characterization of Mixed Unit Interval Graphs --- Institut für Optimierung und Operations Research, Universität Ulm, Ulm, Germany\ `[email protected]` Introduction ============ A graph $G$ is an interval graph, if there is a function $I$ from the vertex set of $G$ to the set of intervals of the real line such that two vertices are adjacent if and only if their assigned intervals intersect. The function $I$ is an interval representation of $G$. Interval graphs are well known and investigated [@fishburn; @golumbic; @LekBo]. There are several different algorithms that decide, if a given graph is an interval graph. See for example [@corneiletal]. An important subclass of interval graphs are unit interval graphs. An interval graph $G$ is a unit interval graph, if there is an interval representation $I$ of $G$ such that $I$ assigns to every vertex a closed interval of unit length. This subclass is well understood and easy to characterize structurally [@roberts] as well as algorithmically [@corneil]. Frankl and Maehara [@fm] showed that it does not matter, if we assign the vertices of $G$ only to closed intervals or only to open intervals of unit length. Rautenbach and Szwarcfiter [@rs] characterized, by a finite list of forbidden induced subgraphs, all interval graphs $G$ such that there is an interval representation of $G$ that uses only open and closed unit intervals. Dourado et al. [@dlprs] gave a characterization of all diamond-free interval graphs that have an interval representation such that all vertices are assigned to unit intervals, where all kinds of unit intervals are allowed and a diamond is a complete graph on four vertices minus an edge. Furthermore, they made a conjecture concerning the general case. We prove that their conjecture is not completely correct and give a complete characterization of this class. Since the conjecture is rather technical and not given by a list of forbidden subgraphs, we refer the reader to [@dlprs] for a detailed formulation of the conjecture, but roughly speaking, they missed the class of forbidden subgraphs shown in Figure \[graphsT\]. In Section 2 we introduce all definitions and relate our result to other work. In Section 3 we state and prove our results. Preliminary Remarks =================== We only consider finite, undirected, and simple graphs. Let $G$ be a graph. We denote by $V(G)$ and $E(G)$ the vertex and edge set of $G$, respectively. If $C$ is a set of vertices, then we denote by $G[C]$ the subgraph of $G$ induced by $C$. Let $\mathcal{M}$ be a set of graphs. We say $G$ is $\mathcal{M}$-free, if for every $H\in\mathcal{M}$, the graph $H$ is not an induced subgraph of $G$. For a vertex $v\in V(G)$, let the neighborhood $N_G(v)$ of $v$ be the set of all vertices that are adjacent to $v$ and let the closed neighborhood $N_G[v]$ be defined by $N_G(v)\cup \{v\}$. Two distinct vertices $u$ and $v$ are twins (in $G$) if $N_G[u]=N_G[v]$. If $G$ contains no twins, then $G$ is twin-free. Let $\mathcal{N}$ be a family of sets. We say a graph $G$ has an $\mathcal{N}$-intersection representation, if there is a function $f:V(G)\rightarrow \mathcal{N}$ such that for any two distinct vertices $u$ and $v$ there is an edge joining $u$ and $v$ if and only if $f(u)\cap f(v)\not=\emptyset$. If there is an $\mathcal{N}$-intersection representation for $G$, then $G$ is an $\mathcal{N}$-graph. Let $x,y\in \mathbb{R}$. We denote by $$[x,y]=\{z\in \mathbb{R}: x\leq z\leq y\}$$ the closed interval, by $$(x,y)=\{z\in \mathbb{R}: x< z<y\}$$ the open interval, by $$(x,y]=\{z\in \mathbb{R}: x< z\leq y\}$$ the open-closed interval, and by $$[x,y)=\{z\in \mathbb{R}: x\leq z< y\}$$ the closed-open interval of $x$ and $y$. For an interval $A$, let $\ell(A)=\inf\{x\in \mathbb{R}:x\in A\}$ and $r(A)=\sup\{x\in \mathbb{R}:x\in A\}$. If $I$ is an interval representation of $G$ and $v\in V(G)$, then we write $\ell(v)$ and $r(v)$ instead of $\ell(I(v))$ and $r(I(v))$, respectively, if there are no ambiguities. Let $\mathcal{I}^{++}$ be the set of all closed intervals, $\mathcal{I}^{--}$ be the set of all open intervals, $\mathcal{I}^{-+}$ be the set of all open-closed intervals, $\mathcal{I}^{+-}$ be the set of all closed-open intervals, and $\mathcal{I}$ be the set of all intervals. In addition, let $\mathcal{U}^{++}$ be the set of all closed unit intervals, $\mathcal{U}^{--}$ be the set of all open unit intervals, $\mathcal{U}^{-+}$ be the set of all open-closed unit intervals, $\mathcal{U}^{+-}$ be the set of all closed-open unit intervals, and $\mathcal{U}$ be the set of all unit intervals. We call a $\mathcal{U}$-graph a mixed unit interval graph. By a result of [@dlprs] and [@rs], every interval graph is an $\mathcal{I}^{++}$-graph. With our notation unit interval graphs equals $\mathcal{U}^{++}$-graphs. An interval graph $G$ is a proper interval graph if there is an interval representation of $G$ such that $I(u)\not\subseteq I(v)$ for every distinct $u,v\in V(G)$. The next result due to Roberts characterizes unit interval graphs. \[thmroberts\] The classes of unit interval graphs, proper interval graphs, and $K_{1,3}$-free interval graphs are the same. The second result shows that several natural subclasses of mixed unit interval graphs actually coincide with the class of unit interval graphs. The classes of $\mathcal{U}^{++}$-graphs, $\mathcal{U}^{--}$-graphs, $\mathcal{U}^{+-}$-graphs, $\mathcal{U}^{-+}$-graphs, and $\mathcal{U}^{+-}\cup\mathcal{U}^{-+}$-graphs are the same. A graph $G$ is a mixed proper interval graph (respectively an almost proper interval graph) if $G$ has an interval representation $I:V(G)\rightarrow \mathcal{I}$ (respectively $I:V(G)\rightarrow \mathcal{I}^{++}\cup \mathcal{I}^{--}$) such that - there are no two distinct vertices $u$ and $v$ of $G$ with $I(u),I(v)\in \mathcal{I}^{++}$, $I(u)\subseteq I(v)$, and $I(u)\not=I(v)$, and - for every vertex $u$ of $G$ with $I(u)\notin \mathcal{I}^{++}$, there is a vertex $v$ of $G$ with $I(v)\in \mathcal{I}^{++}$, $\ell(u)=\ell(v)$, and $r(u)=r(v)$. A natural class extending the class of unit interval graphs are $\mathcal{U}^{++}\cup\mathcal{U}^{--}$-graphs. These were characterized by Rautenbach and Szwarcfiter. (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+1.5,+1.5) circle (); (,)–(+1.5,+1.5) (+1,)–(+1.5,+1.5) (+2,)–(+1.5,+1.5) (+3,)–(+1.5,+1.5); \(1) at (+1.5,-0.7) [$K_{1,4}$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+1.5,+1.5) circle (); (,)–(+1.5,+1.5) (+2,)–(+1,)–(+1.5,+1.5) (+2,)–(+1.5,+1.5) (+3,)–(+1.5,+1.5); \(1) at (+1.5,-0.7) [$K_{1,4}^$]{}; (,) circle (); (+1.5,) circle (); (+3,) circle (); (+1,+1.5) circle (); (+2,+1.5) circle (); (,)–(+1,+1.5)–(+1.5,)–(+2,+1.5)– (+3,)–(+1,+1.5)–(+2,+1.5)–(,); \(1) at (+1.5,-0.7) [$K_{2,3}^$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+1,+1.5) circle (); (+2,+1.5) circle (); (,)–(+1,+1.5)– (+1,)–(+2,+1.5)– (+2,)–(+1,+1.5)– (+2,+1.5)–(+3,); \(1) at (+1.5,-0.7) [$K_{2,4}^$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+1.5,+1.5) circle (); (+1,+1.5) circle (); (,)–(+1.5,+1.5) (+2,)–(+1,)–(+1.5,+1.5) (+2,)–(+1.5,+1.5) (+3,)–(+1.5,+1.5) (+1,)–(+1,+1.5) ; For a twin-free graph $G$, the following statements are equivalent. - $G$ is a $\{K_{1,4},K_{1,4}^,K_{2,3}^,K_{2,4}^\}$-free graph. (See Figure \[u+-graph\] for an illustration.) - $G$ is an almost proper interval graph. - $G$ is a $\mathcal{U}^{++}\cup\mathcal{U}^{--}$-graph. Note that an interval representation can assign the same interval to twins and hence the restriction to twin-free graphs does not weaken the statement but simplifies the description. The next step is to allow all different types of unit intervals. The class of $\mathcal{U}$-graphs is a proper superclass of the $\mathcal{U}^{++}\cup\mathcal{U}^{--}$-graphs, because the graph illustrated in Figure \[ugraph\] is a $\mathcal{U}$-graph, but not a $\mathcal{U}^{++}\cup\mathcal{U}^{--}$-graph (it contains a $K_{1,4}^$). Dourado et al. already made some progress in characterizing this class. \[proper=unit\] For a graph $G$, the following two statements are equivalent. - $G$ is a mixed proper interval graph. - $G$ is a mixed unit interval graph. They also characterized diamond-free mixed unit interval graphs. There is another approach by Le and Rautenbach [@lr] to understand the class of $\mathcal{U}$-graphs by restricting the ends of the unit intervals to integers. They found a infinite list of forbidden induced subgraphs, which characterize these so-called integral* $\mathcal{U}$-graphs. Results ======= In this section we state and prove our main results. We start by introducing a list of forbidden induced subgraphs. See Figures \[graphsR\], \[graphsS\], \[graphsS1\], and \[graphsT\] for illustration. Let $\mathcal{R}=\bigcup_{i=0}^{\infty}\{R_i\}$, $\mathcal{S}=\bigcup_{i=1}^{\infty}\{S_i\}$, $\mathcal{S'}= \bigcup_{i=1}^{\infty}\{S_i'\}$, and $\mathcal{T}=\bigcup_{i\geq j\geq 0}\{T_{i,j}\}$. For $k\in\mathbb{N}$ let the graph $Q_k$ arise from the graph $R_k$ by deleting two vertices of degree $1$ that have a common neighbor. We call the common neighbor of the two deleted vertices and its neighbor of degree $2$ special vertices of $Q_k$. Note that if a graph $G$ is twin-free, then the interval representation of $G$ is injective. (+0.5,) circle (); (+1,+1) circle (); (+2,+1) circle (); (+2.5,) circle (); (+1.5,) circle (); (+0.5,)–(+1.5,)–(+1,+1) (+2,+1)–(+1.5,)–(+2.5,); \(1) at (+1.5,-0.7) [$R_0$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (,)–(+1,)–(+2,)–(+3,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+2.5,+1); \(1) at (+1.5,-0.7) [$R_1$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+4,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+3.5,+1) circle (); (,)–(+1,)–(+2,)–(+3,)–(+4,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+2.5,+1)–(+3,)–(+3.5,+1); \(1) at (+2,-0.7) [$R_2$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+4.5,) circle (); (+5.5,) circle (); (+6.5,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+5,+1) circle (); (+6,+1) circle (); (+3.35,) circle (/2); (+3.75,) circle (/2); (+4.15,) circle (/2); (,)–(+3,) (+4.5,)–(+6.5,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+2.5,+1)–(+3,) (+4.5,)–(+5,+1)–(+5.5,)–(+6,+1); (+1,) – (+5.5,) node \[pos=0.5,anchor=north,yshift=-0.4cm\] [$i$ triangles]{}; \(1) at (+6,-0.7) [$R_i$]{}; (+1,) circle (); (+2,) circle (); (+3,) circle (); (+2,+1.5) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+1,)–(+2,)–(+3,) (+1,)–(+1.5,+1)–(+2,)–(+2.5,+1) –(+2,+1.5)–(+2,) (+1.5,+1)–(+2,+1.5); \(1) at (+2,-0.7) [$S_1$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+1,+1.5) circle (); (,)–(+1,)–(+2,)–(+3,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+2.5,+1) (,)–(+0.5,+1)–(+1,+1.5)–(+1.5,+1) (+1,)–(+1,+1.5); \(1) at (+1.5,-0.7) [$S_2$]{}; (-1,) circle (); (-0.5,+1) circle (); (,) circle (); (+1,) circle (); (+2,) circle (); (+3.5,) circle (); (+4.5,) circle (); (+5.5,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+4,+1) circle (); (+5,+1) circle (); (,+1.5) circle (); (+2.35,) circle (/2); (+2.75,) circle (/2); (+3.15,) circle (/2); (-1,)–(+2,) (+3.5,)–(+5.5,) (-1,)–(-0.5,+1)–(,)–(+0.5,+1)–(+1,)–(+1.5,+1)–(+2,) (+3.5,)–(+4,+1)–(+4.5,)–(+5,+1) (-0.5,+1)–(,+1.5)–(+0.5,+1) (,)–(,+1.5); (-1,) – (+4.5,) node \[pos=0.5,anchor=north,yshift=-0.4cm\] [$i$ triangles]{}; \(1) at (+4.5,-0.7) [$S_i$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+0.5,+0.5) circle (); (,)–(+1,)–(+2,)–(+3,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+2.5,+1) (,)–(+0.5,+1)–(+0.5,+0.5)–(,) (+0.5,+0.5)–(+1,); \(1) at (+1.5,-0.7) [$S_2'$]{}; (+1,) circle (); (+2,) circle (); (+3,) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+1.5,+0.5) circle (); (+1,)–(+2,)–(+3,) (+1,)–(+1.5,+1)–(+2,)–(+2.5,+1) (+1.5,+1)–(+1.5,+0.5)–(+1,) (+1.5,+0.5)–(+2,); \(1) at (+2,-0.7) [$S_1'$]{}; (+1,) circle (); (+2,) circle (); (+3,) circle (); (+4.5,) circle (); (+5.5,) circle (); (+6.5,) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+5,+1) circle (); (+6,+1) circle (); (+1.5,+0.5) circle (); (+3.35,) circle (/2); (+3.75,) circle (/2); (+4.15,) circle (/2); (+1,)–(+3,) (+4.5,)–(+6.5,) (+1,)–(+1.5,+1)–(+2,)–(+2.5,+1)–(+3,) (+4.5,)–(+5,+1)–(+5.5,)–(+6,+1) (+1.5,+1)–(+1.5,+0.5)–(+1,) (+1.5,+0.5)–(+2,); (+1,) – (+5.5,) node \[pos=0.5,anchor=north,yshift=-0.4cm\] [$i$ triangles]{}; \(1) at (+6,-0.7) [$S_i'$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+0.5,+1) circle (); (+2.5,+1) circle (); (,)–(+3,) (+1,)–(+0.5,+1)–(+2,) (+1,)–(+2.5,+1)–(+2,); \(1) at (+1.5,-0.7) [$T_{0,0}$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+4,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+3.5,+1) circle (); (,)–(+4,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+3.5,+1) –(+3,)–(+1.5,+1); \(1) at (+2,-0.7) [$T_{1,0}$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+4,) circle (); (+5,) circle (); (+6,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+4.5,+1) circle (); (+5.5,+1) circle (); (,)–(+6,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+2.5,+1)–(+3,)–(+4.5,+1) (+5.5,+1)–(+5,)–(+4.5,+1)–(+4,)–(+2.5,+1); \(1) at (+3,-0.7) [$T_{2,1}$]{}; (,) circle (); (+1,) circle (); (+2,) circle (); (+3.5,) circle (); (+4.5,) circle (); (+5.5,) circle (); (+6.5,) circle (); (+8,) circle (); (+9,) circle (); (+10,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+4,+1) circle (); (+6,+1) circle (); (+8.5,+1) circle (); (+9.5,+1) circle (); (,)–(+2,) (+3.5,)–(+6.5,) (+8,)–(+10,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,) (+8,)–(+8.5,+1)–(+9,)–(+9.5,+1) (+3.5,)–(+4,+1)–(+4.5,) (+5.5,)–(+6,+1)–(+6.5,) (+4,+1)–(+5.5,) (+6,+1)–(+4.5,); (+2.35,) circle (/2); (+2.75,) circle (/2); (+3.15,) circle (/2); (+6.85,) circle (/2); (+7.25,) circle (/2); (+7.65,) circle (/2); (+1,) – (+4.5,) node \[pos=0.5,anchor=north,yshift=-0.4cm\] [$i$ triangles]{}; (+5.5,) – (+9,) node \[pos=0.5,anchor=north,yshift=-0.4cm\] [$j$ triangles]{}; \(1) at (+5,-0.7) [$T_{i,j}$]{}; \[lemma dlprs\] Let $k\in \mathbb{N}$. (a) Every $\mathcal{U}$-representation of the claw $K_{1,3}$ arises by translation (replacing $I$ by $I+x$ for some $x\in \mathbb{R}$; that is, shifting all intervals by $x$) of the following $\mathcal{U}$-representation $I: V(K_{1,3})\rightarrow \mathcal{U}$ of $K_{1,3}$, where $I(V(K_{1,3}))$ consists of the following intervals - either $[0,1]$ or $(0,1]$, - $[1,2]$ and $(1,2)$, and - either $[2,3]$ or $[2,3)$. (b) Every injective $\mathcal{U}$-representation of $Q_k$ arises by translation and inversion (replacing $I$ by $-I$; that is, multiplying all endpoints of the intervals by $-1$) of one of the two injective $\mathcal{U}$-representations $I: V(Q_k)\rightarrow \mathcal{U}$ of $Q_k$, where $I(V(Q_k))$ consists of the following intervals - either $[0,1]$ or $(0,1]$, - $[1,2]$ and $(1,2)$, and - $[i,i+1]$ and $[i,i+1)$ for $2\leq i \leq k+1$. (c) The graphs in $\{T_{0,0}\}\cup\mathcal{R}$ are minimal forbidden subgraphs for the class of $\mathcal{U}$-graphs with respect to induced subgraphs. (d) If $G$ is a $\mathcal{U}$-graph, then every induced subgraph $H$ in $G$ that is isomorphic to $Q_k$ and every vertex $u^\in V(G)\setminus V(H)$ such that $u^$ is adjacent to exactly one of the two special vertices $x$ of $H$, the vertex $u^$ has exactly one neighbor in $V(H)$, namely $x$. \[lemma1\] If a graph $G$ is a twin-free mixed unit interval graph, then $G$ is $\{K_{2,3}^\}\cup \mathcal{R}\cup \mathcal{S}\cup \mathcal{S'}\cup \mathcal{T}$-free. Proof of Lemma \[lemma1\]: It is easy to see that $G$ is $\{K_{2,3}^\}$-free. Lemma \[lemma dlprs\] (c) shows that $G$ is $\mathcal{R}$-free and Lemma \[lemma dlprs\] (d) shows that $G$ is $\mathcal{S}$-free. Let $k\in \mathbb{N}$. Note that the graph $S_k'$ arises from the graph $Q_k$ by adding a vertex $z$ and joining it to the two special vertices of $Q_k$ and the unique common neighbor of these two vertices. For contradiction, we assume that $S_k'$ has a $\mathcal{U}$-representation $I$. By Lemma \[lemma dlprs\] (b) there are only two possibilities for the $\mathcal{U}$-representation of $Q_k$. Thus we assume that the subgraph $Q_k$ of $S_k'$ has the interval representation as described in Lemma \[lemma dlprs\] (b). In both cases we conclude $\ell(z)=k+1$ and $k+1\in I(z)$. Thus $r(z)=k+2$ and hence $I(z)\in \{[k+1,k+2],[k+1,k+2)\}$. Therefore, $G$ is not twin-free, which is a contradiction. This implies that $G$ is $\mathcal{S}'$-free. By Lemma \[lemma dlprs\] (c), $G$ is $T_{0,0}$-free. Let $C$ be a claw with vertex set $\{c,a_1,a_2,a_3\}$, where $c$ is the center vertex. Denote by $v_k$ and $w_k$ the special vertices of $Q_k$. Note that $T_{k,0}$ arises by the disjoint union of the graph $Q_k$ and $C$, identifying $v_k$ and $a_1$, and adding the edges $w_kc$ and $v_ka_2$. For contradiction, we assume that $T_{k,0}$ has a $\mathcal{U}$-representation $I$. By Lemma \[lemma dlprs\] (b), we assume without loss of generality that the induced subgraph $Q_k$ of $T_{k,0}$ is represented by exactly the intervals described in Lemma \[lemma dlprs\] (b). Thus $I(v_k)=[k+1,k+2]$ and $I(w_k)=[k+1,k+2)$, because $v_ka_2\in E(T_{k,0})$ but $w_ka_2\notin E(T_{k,0})$. Since $I(v_k)$ is not an open interval and by Lemma \[lemma dlprs\] (a), we obtain $I(c)=[k+2,k+3]$ and hence $I(w_k)\cap I(c)=\emptyset$. This is a contradiction, which implies that $G$ is $\bigcup_{i\geq 0}\{T_{i,0}\}$-free. Let $i,j\in\mathbb{N}$. Note that the graph $T_{i,j}$ arises by the disjoint union of $Q_i$ and $Q_j$ and adding three edges between the special vertices of $Q_i$ and $Q_j$. We may assume that the intervals of the subgraph $Q_i$ are exactly the intervals as described in Lemma \[lemma dlprs\] (b). Let $w_i$ (respectively $v_i$) be the vertex of $Q_i$ that has one (two) neighbor(s) in the subgraph $Q_j$; that is, $I(v_i)=[i+1,i+2]$ and $I(w_i)=[i+1,i+2)$ because $N_{T_{i,j}}(w_i)\subset N_{T_{i,j}}(v_i)$. Let $w_j$ (respectively $v_j$) be the vertex of $Q_j$ that has one (two) neighbor(s) in the subgraph $Q_i$. Since the subgraph $Q_j$ has also an interval representation as described in Lemma \[lemma dlprs\] (b) and the vertices of $Q_i\setminus\{v_i,w_i\}$ and not joined by an edge to the vertices of $Q_j\setminus\{v_j,w_j\}$, we conclude that the intervals of the vertices of $Q_j$ arise by an inversion and a translation of the interval representation as described in Lemma \[lemma dlprs\] (b). This implies that $I(v_j)=[x,x+1]$ and $I(w_j)=(x,x+1]$ for some $x\in \mathbb{R}$. Obviously, $x\in[i+1,i+2]$. If $x=i+2$, then neither $v_i$ is adjacent to $w_j$ nor $w_i$ is adjacent to $v_i$. If $x\in [i+1,i+2)$, then the intervals of $w_i$ and $w_j$ intersect, which is not possible. Therefore, $G$ is $\mathcal{T}$-free and this completes the proof. $\Box$ We proceed to our main result. \[mainthm\] A twin-free graph $G$ is a mixed unit interval graph if and only if $G$ is a $\{K_{2,3}^\}\cup\mathcal{R}\cup \mathcal{S}\cup \mathcal{S'}\cup \mathcal{T}$-free interval graph. Proof of Theorem \[mainthm\]: We use a similar approach as in [@rs]. By Lemma \[lemma1\], we know if $G$ is a twin-free mixed unit interval graph, then $G$ is a $\{K_{2,3}^\}\cup\mathcal{R}\cup \mathcal{S}\cup \mathcal{S'}\cup\mathcal{T}$-free interval graph. Let $G$ be a twin-free $\{K_{2,3}^\}\cup\mathcal{R}\cup \mathcal{S}\cup \mathcal{S'}\cup\mathcal{T}$-free interval graph. We show that $G$ is a mixed proper interval graph. By Theorem \[proper=unit\], this proves Theorem \[mainthm\]. Since $G$ is an interval graph, $G$ has an $\mathcal{I}^{++}$-representation $I$. As in [@rs] we call a pair $(u,v)$ of distinct vertices a bad pair if $I(u)\subseteq I(v)$. Let $I$ be such that the number of bad pairs is as small as possible. If $I$ has no bad pair, then we are done by Theorem \[thmroberts\]. Hence we assume that there is at least one bad pair. The strategy of the proof is as follows. Claim \[c1\] to Claim \[c6a\] collect properties of $G$ and $I$, before we modify our interval representation of $G$ to show that $G$ is a mixed proper interval graph. In Claim \[c7\] to Claim 10 we prove that our modification of the interval representation preserves all intersections and non-intersections. Claim 1 to Claim \[c3\] are similar to Claim 1 to Claim 3 in [@rs], respectively. For the sake of completeness we state the proofs here. \[c1\] If $(u,v)$ is a bad pair, then there are vertices $x$ and $y$ such that $\ell(v)\leq r(x)< \ell(u)$ and $r(u)<\ell(y)\leq r(v)$. Proof of Claim 1: For contradiction, we assume the existence of a bad pair $(w,v)$ such that there is no vertex $x$ with $\ell(v)\leq r(x)<\ell(w)$. A symmetric argument implies the existence of $y$. Let $u$ be a vertex such that $\ell(u)$ is as small as possible with respect to $I(u)\subseteq I(v)$. By our assumption there is no vertex $x$ such that $\ell(v)\leq r(x)<\ell(u)$. Let $\epsilon$ be the smallest distance between two distinct endpoints of intervals of $I$. Let $I': V(G)\rightarrow \mathcal{I}^{++}$ be such that $I'(u)=[\ell(v)- \epsilon/2,r(u)]$, $I'(v)=[\ell(v),r(v)+\epsilon/2]$, and $I'(z)=I(z)$ for $z\in V(G)\setminus \{u,v\}$. By the choice of $u$ and $\epsilon$, we conclude that $I'$ is an interval representation of $G$, but $I'$ has less bad pairs than $I$, which is a contradiction to our choice of $I$. This completes the proof. $\Box$ Let $a_1$ and $a_2$ be two distinct vertices. Claim \[c1\] implies that $\ell(a_1)\not=\ell(a_2)$ and $r(a_1)\not=r(a_2)$. Suppose $\ell(a_1)<\ell(a_2)$. If $r(a_1)=\ell(a_2)$, then let $\epsilon$ be as in the proof of Claim \[c1\] and $I': V(G)\rightarrow \mathcal{I}^{++}$ be such that $I'(a_1)=[\ell(a_1),r(a_1)+\epsilon/2]$, and $I'(z)=I(z)$ for $z\in V(G)\setminus \{a_1\}$. By the choice of $\epsilon$, we conclude that $I'$ is an interval representation of $G$ with as many bad pairs as $I$. Therefore, we assume without loss of generality that we chose $I$ such that all endpoints of the intervals of $I$ are distinct. Hence the inequalities in Claim \[c1\] are strict inequalities. \[c2\] If $(u,w)$ and $(v,w)$ are bad pairs, then $u=v$, that is, no interval contains two distinct intervals. Proof of Claim \[c2\]: For contradiction, we assume that there are distinct vertices $u'$, $v'$ and $w$ such that $(u',w)$ and $(v',w)$ are bad pairs. Let $u$ be a vertex such that $(u,w)$ is a bad pair and $\ell(u)$ is as small as possible. Let $v$ be a vertex such that $(v,w)$ is a bad pair and $r(v)$ is as large as possible. Claim \[c1\] ensures two distinct vertices $x$ and $y$ such that $\ell(w)< r(x)< \ell(u)$ and $r(v)<\ell(y)< r(w)$. If $u\not=v$ and $I(u)\cap I(v)=\emptyset$, then $G[\{w,x,u,v,y\}]$ is isomorphic to $R_0$, which is a contradiction. If $u\not=v$ and $I(u)\cap I(v)\not=\emptyset$, then in the graph $G[\{w,x,u,v,y\}]$ the vertices $u$ and $v$ are twins. Since $G$ is twin-free, $u$ and $v$ do not have the same closed neighborhood in $G$ and hence there is a vertex $z$, which is adjacent to say $u$ (by symmetry) and not to $v$. Since $I(u)\subset I(w)$, $z$ is adjacent to $w$. If $z$ is not adjacent to $x$, then $G[\{w,x,z,v,y\}]$ is isomorphic to $R_0$ and if $z$ is adjacent to $x$, then $G[\{w,x,z,u,v,y\}]$ is isomorphic to $S_1$, which is a contradiction. If $u=v$, then there is a vertex $z$ such that $(z,u)$ is a bad pair because $u'$ or $v'$ is a suitable choice. We choose $z$ such that $\ell(z)$ is minimal. Claim \[c1\] ensures the existence of a vertex $x'$ such that $\ell(u)< r(x')<\ell(z)$. Note that the choice of $u$ and $z$ guarantees $\ell(x')<\ell(w)$, so $xx'\in E(G)$. Therefore, $G[\{w,x,x',u,z,y\}]$ is isomorphic to $S_1$, which is a contradiction. This completes the proof of Claim \[c2\]. $\Box$ \[c3\] If $(u,v)$ and $(u,w)$ are bad pairs, then $v=w$, that is, no interval is contained in two distinct intervals. Proof of Claim \[c3\]: Claim \[c2\] implies that neither $(v,w)$ nor $(w,v)$ is a bad pair. Thus we may assume $\ell(w)<\ell(v)<\ell(u)$ and $r(u)<r(w)<r(v)$. By Claim \[c1\], there are vertices $x$ and $y$ such that $\ell(v)< r(x)<\ell(u)$ and $r(u)<\ell(y)< r(w)$. Now, $G[\{v,w,x,u,y\}]$ is isomorphic to $K_{2,3}^$, which is a contradiction and completes the proof of Claim \[c3\]. $\Box$ A vertex $x$ is to the left* (respectively right) of a vertex $y$ (in $I$), if $r(x)<\ell(y)$ (respectively $r(y)<\ell(x)$). Two adjacent vertices $x$ and $y$ are distinguishable by vertices to the left (respectively right) of them, if there is a vertex $z$, which is adjacent to exactly one of them and to the left (respectively right) of one of them. The vertex $z$ distinguishes $x$ and $y$. Next, we show that for a bad pair $(u,v)$ there is the structure as shown in Figure \[structurebadpair\] in $G$. We introduce a positive integer $\ell_{u,v}^{\rm max}$ that, roughly speaking, indicates how large this structure is. (,) circle (); (+1.5,) circle (); (+3,) circle (); (+5,) circle (); (+6.5,) circle (); (+8,) circle (); (+10,) circle (); (+11.5,) circle (); (+13,) circle (); (+2.25,+1.5) circle (); (+5.75,+1.5) circle (); (+6.5,+1.5) circle (); (+7.25,+1.5) circle (); (+10.75,+1.5) circle (); (,)–(+3,) (+5,)–(+8,) (+10,)–(+13,) (+1.5,)–(+2.25,+1.5)–(+3,) (+10,)–(+10.75,+1.5)–(+11.5,) (+5,)–(+5.75,+1.5)–(+6.5,)–(+7.25,+1.5)– (+8,) (+6.5,+1.5)–(+6.5,); (+3.6,) circle (/2); (+4,) circle (/2); (+4.4,) circle (/2); (+8.6,) circle (/2); (+9,) circle (/2); (+9.4,) circle (/2); (+1.5,) – (+6.45,) node \[pos=0.5,anchor=north,yshift=-1.9cm\] [$\ell_{u,v}^{\rm max}-1$ triangles]{}; (+6.55,) – (+11.5,) node \[pos=0.5,anchor=north,yshift=-1.9cm\] [$r_{u,v}^{\rm max}-1$ triangles]{}; \(1) at (+6.5,-0.5) [$v$]{}; (1) at (+6.5,+2) [$u$]{}; (1) at (+5,-0.5) [$x_{u,v}^1$]{}; (1) at (+5.75,+2) [${x_{u,v}^1}'$]{}; \(1) at (+1.5,-0.5) [$x_{u,v}^{\ell_{u,v}^{\rm max}-1}$]{}; (1) at (+2.25,+2) [${x_{u,v}^{\ell_{u,v}^{\rm max}-1}}'$]{}; (1) at (,-0.5) [$x_{u,v}^{\ell_{u,v}^{\rm max}}$]{}; \(1) at (+8,-0.5) [$y_{u,v}^1$]{}; (1) at (+7.25,+2) [${y_{u,v}^1}'$]{}; \(1) at (+11.5,-0.5) [$y_{u,v}^{r_{u,v}^{\rm max}-1}$]{}; (1) at (+10.75,+2) [${y_{u,v}^{r_{u,v}^{\rm max}-1}}'$]{}; (1) at (+13,-0.5) [$y_{u,v}^{r_{u,v}^{\rm max}}$]{}; \(1) at (+6.5,-1.3) [$X_{u,v}^0$]{}; (1) at (+5.75,+3.2) [$X_{u,v}^1$]{}; (1) at (+2.5,+3.2) [$X_{u,v}^{\ell_{u,v}^{\rm max}-1}$]{}; (1) at (,+0.8) [$X_{u,v}^{\ell_{u,v}^{\rm max}}$]{}; (1) at (+7.25,+3.2) [$Y_{u,v}^1$]{}; (1) at (+11,+3.2) [$Y_{u,v}^{r_{u,v}^{\rm max}-1}$]{}; (1) at (+13,+0.8) [$Y_{u,v}^{r_{u,v}^{\rm max}}$]{}; (+5.2,+0.9) ellipse (1 and 2); (+1.85,+0.9) ellipse (1 and 2); (+7.8,+0.9) ellipse (1 and 2); (+11.0,+0.9) ellipse (1 and 2); (-0.2,-0.3) circle (0.7); (+6.5,-0.3) circle (0.6); (+13,-0.3) circle (0.7); (+5.5,)–(+7.5,) (+6,+0.4)–(+7,+0.4) (+5.5,+0.2)–(+5.5,-0.2) (+7.5,+0.2)–(+7.5,-0.2) (+6,+0.2)–(+6,+0.6) (+7,+0.2)–(+7,+0.6) (+4.8,-0.4)–(+5.7,-0.4) (+4.3,-0.8)–(+5.6,-0.8) (+4.8,-0.6)–(+4.8,-0.2) (+5.7,-0.6)–(+5.7,-0.2) (+4.3,-1)–(+4.3,-0.6) (+5.6,-1)–(+5.6,-0.6) (+3.7,+0.4)–(+4.6,+0.4) (+3.2,)–(+4.5,) (+3.7,+0.6)–(+3.7,+0.2) (+4.6,+0.6)–(+4.6,+0.2) (+3.2,+0.2)–(+3.2,-0.2) (+4.5,+0.2)–(+4.5,-0.2) (+2.5,-0.4)–(+3.4,-0.4) (+2,-0.8)–(+3.3,-0.8) (+2.5,-0.6)–(+2.5,-0.2) (+3.4,-0.6)–(+3.4,-0.2) (+2,-1)–(+2,-0.6) (+3.3,-1)–(+3.3,-0.6) (+7.3,-0.4)–(+8.2,-0.4) (+7.4,-0.8)–(+8.7,-0.8) (+7.3,-0.6)–(+7.3,-0.2) (+8.2,-0.6)–(+8.2,-0.2) (+7.4,-1)–(+7.4,-0.6) (+8.7,-1)–(+8.7,-0.6) ; \(1) at (+6.5,-0.3) [$v$]{}; (1) at (+6.5,+0.7) [$u$]{}; \(1) at (+5.1,-1.2) [$X_{u,v}^1$]{}; (1) at (+4.1,+0.9) [$X_{u,v}^2$]{}; (1) at (+2.8,-1.2) [$X_{u,v}^3$]{}; (1) at (+8,-1.2) [$Y_{u,v}^1$]{}; (+1.3,-0.1) circle (/3); (+0.9,-0.1) circle (/3); (+0.5,-0.1) circle (/3); (+9.4,-0.1) circle (/3); (+9.8,-0.1) circle (/3); (+10.2,-0.1) circle (/3); For a bad pair $(u,v)$ let $v=X_{u,v}^0$ and let $X_{u,v}^1$ be the set of vertices that are adjacent to $v$ and to the left of $u$. Let $y_{u,v}$ be a vertex to the right of $u$ and adjacent to $v$. Claim \[c1\] guarantees $\|X_{u,v}^1\|\geq 1$ and the existence of $y_{u,v}$. If $\|X_{u,v}^1\|=1$, then let $\ell_{u,v}^{\rm max}=1$ and we stop here. Suppose $\|X_{u,v}^1\|\geq 2$. Since $G$ is $R_0$-free, $X_{u,v}^1$ is a clique and since $G$ is $S_1'$-free, we conclude $\|X_{u,v}^1\|=2$. Let $\{x,x'\}= X_{u,v}^1$ such that $r(x)< r(x')$. For contradiction, we assume that there is a vertex $z$ to the right of $x$ that distinguishes $x$ and $x'$. We conclude $\ell(v)<\ell(z)$. By Claim \[c2\], $r(v)< r(z)$. This implies that $(u,z)$ is a bad pair, which contradicts Claim \[c3\]. Thus $z$ does not exist. In addition $(x,x')$ is not a bad pair, otherwise Claim \[c1\] guarantees a vertex $z$ such that $r(x)<\ell(z)<r(x')$, which is a contradiction. Thus $\ell(x)<\ell(x')<r(x)< r(x')$. Let $x_{u,v}^1=x$ and ${x_{u,v}^1}'=x'$. Note that $N_G({x^1_{u,v}}')\subset N_G(x^1_{u,v})$. Let $X_{u,v}^2=N_G(x^1_{u,v})\setminus N_G({x^1_{u,v}}')$. Note that all vertices in $X_{u,v}^2$ are to the left of ${x^1_{u,v}}'$. Since $G$ is twin-free, $\|X_{u,v}^2\|\geq 1$. If $\|X_{u,v}^2\|=1$, then let $\ell_{u,v}^{\rm max}=2$ and we stop here. Suppose $\|X_{u,v}^2\|\geq 2$. Since $G$ is $R_1$-free, $X_{u,v}^2$ is a clique and since $G$ is $S_2'$-free, we conclude $\|X_{u,v}^2\|=2$. Let $\{x,x'\}= X_{u,v}^2$ such that $r(x)< r(x')$. For contradiction, we assume that there is a vertex $z$ to the right of $x$ that distinguishes $x$ and $x'$. Since $z\notin X_{u,v}^2$, we conclude $\ell({x_{u,v}^1}')< r(z)$. If $r(z)<\ell(v)$, then $G[\{z,x,x',{x_{u,v}^1},{x_{u,v}^1}',v,u,y_{u,v}\}]$ is isomorphic to $S_2$, which is a contradiction. Thus $\ell(v)< r(z)$. If $r(z)< \ell(u)$, then $\|X_{u,v}^1\|=3$, which is a contradiction. Thus $\ell(u)< r(z)$. If $r(u)< r(z)$, then $(u,v)$ and $(u,z)$ are bad pairs, which is a contradiction to Claim \[c3\]. Thus $\ell(u)< r(z)<r(u)$. Now $G[\{z,x',{x_{u,v}^1}',v,u,y_{u,v}\}]$ is isomorphic to $T_{0,0}$, which is the final contradiction. Note that $(x,x')$ is not a bad pair, otherwise Claim 1 guarantees a vertex $z$ such that $r(x)<\ell(z)< r(x')$, which is a contradiction. Thus $\ell(x)<\ell(x')<r(x)< r(x')$. Let $x_{u,v}^2=x$ and ${x_{u,v}^2}'=x'$. Note that $N_G({x^2_{u,v}}')\subset N_G(x^2_{u,v})$. Let $X_{u,v}^3=N_G(x^2_{u,v})\setminus N_G({x^2_{u,v}}')$. Note that all vertices in $X_{u,v}^3$ are to the left of ${x^2_{u,v}}'$. We assume that for $k\geq 3$, $i\in[k-1]$ and $j\in[k]$ - we defined $X_{u,v}^j$, - $\|X_{u,v}^i\|=2$ holds, - we defined $x_{u,v}^i$ and ${x_{u,v}^i}'$, - $\ell(x_{u,v}^i)<\ell({x_{u,v}^i}')<r(x_{u,v}^i)<r({x_{u,v}^i}')$ holds, - the vertices in $X_{u,v}^{i+1}$ are to the left of ${x^{i}_{u,v}}'$, and - the vertices in $X_{u,v}^i$ are not distinguishable to the right. If $\|X_{u,v}^k\|=1$, then let $\ell_{u,v}^{\rm max}=k$ and we stop here. Suppose $\|X_{u,v}^k\|\geq 2$. Since $G$ is $R_{k-1}$-free, $X_{u,v}^k$ is a clique and since $G$ is $S_k'$-free, we obtain $\|X_{u,v}^k\|=2$. Let $\{x,x'\}= X_{u,v}^k$ such that $r(x)< r(x')$. For contradiction, we assume that there is a vertex $z$ to the right of $x$ that distinguishes $x$ and $x'$. Since $z\notin X_{u,v}^k$, we conclude $\ell({x_{u,v}^{k-1}}')< r(z)$. If $r(z)<\ell(x_{u,v}^{k-2})$, then $G[\{z,x,x',v,u,y_{u,v}\}\cup \bigcup_{i=1}^{k-1}X_{u,v}^i]$ is isomorphic to $S_k$, which is a contradiction. Thus $\ell(x_{u,v}^{k-2})< r(z)$. If $r(z)<\ell({x_{u,v}^{k-2}}')$, then $\|X_{u,v}^{k-1}\|=3$, which is a contradiction. Thus $\ell({x_{u,v}^{k-2}}')< r(z)$. If $r(z)<\ell(x_{u,v}^{k-3})$, then $G[\{z,x',{x_{u,v}^{k-1}}',v,u,y_{u,v}\}\cup \bigcup_{i=1}^{k-2}X_{u,v}^i]$ is isomorphic to $T_{k-3,0}$, which is a contradiction. Thus $\ell(x_{u,v}^{k-3})< r(z)$. If $r(z)< r(x_{u,v}^{k-2})$, then $\|X_{u,v}^{k-2}\|=3$, which is a contradiction. Thus $r(x_{u,v}^{k-2})<r(z)$ and hence $({x_{u,v}^{k-1}}',z)$ and $(x_{u,v}^{k-2},z)$ are bad pairs, which is a contradiction to Claim \[c2\]. Thus $x,x'$ are not distinguishable to the right. We obtain that $(x,x')$ is not a bad pair, otherwise Claim 1 guarantees a vertex $z$ such that $r(x)<\ell(z)< r(x')$, which is a contradiction. Thus $\ell(x)<\ell(x')<r(x)< r(x')$. Let $x_{u,v}^k=x$ and ${x_{u,v}^k}'=x'$. Note that $N_G({x^k_{u,v}}')\subset N_G(x^k_{u,v})$. Let $X_{u,v}^{k+1}=N_G(x^k_{u,v})\setminus N_G({x^k_{u,v}}')$. Note that all vertices in $X_{u,v}^{k+1}$ are to the left of ${x^k_{u,v}}'$. By induction this leads to the following properties. \[c4\] If $(u,v)$ is a bad pair, $k\in [\ell_{u,v}^{\rm max}-1]$, then the following holds: (a) $\|X_{u,v}^k\|= 2$. (b) The vertices in $X_{u,v}^k$ are not distinguishable by vertices to the right of them. (c) We have $\ell(x_{u,v}^i)<\ell({x_{u,v}^i}')<r(x_{u,v}^i)<r({x_{u,v}^i}')$, that is $(x^k_{u,v},{x^k_{u,v}}')$ and $({x^k_{u,v}}',x^k_{u,v})$ are not bad pairs. Note that $\ell_{u,v}^{\rm max}$ is the smallest integer $k$ such that $\|X_{u,v}^{k-1}\|= 2$ and $\|X_{u,v}^k\|=1$. \[c5\] If $(u,v)$ is a bad pair and $k\in [\ell_{u,v}^{\rm max}-1]$, then the following holds. (a) ${x_{u,v}^k}'$ is not contained in a bad pair. (b) There is no vertex $z\in V(G)$ such that $(x_{u,v}^k,z)$ is a bad pair. Proof of Claim \[c5\]: (a): For contradiction, we assume that there is a vertex $z\in V(G)$ such that $({x_{u,v}^k}',z)$ is a bad pair. Trivially $z\notin \{\{v,y,u\} \cup \bigcup_{i=1}^{\ell_{u,v}^{\rm max}} X_{u,v}^i \}$. We have $r({x_{u,v}^k}')<r(z)$ and $\ell(z)<\ell({x_{u,v}^k}')$. In addition $\ell(x_{u,v}^k)<\ell(z)$, otherwise $(x_{u,v}^k,z)$ is also a bad pair, which contradicts Claim \[c2\]. Claim \[c1\] implies the existence of a vertex $a$, such that $\ell(z)< r(a)< \ell({x_{u,v}^k}')$. Let $k=1$. If $r(z)<\ell(u)$, then $z\in X_{u,v}^1$, which is a contradiction to $\|X_{u,v}^1\|=2$. Thus $\ell(u)< r(z)$. If $r(z)< r(u)$, then $G[\{a,z,{x_{u,v}^k}',u,v,y\}]$ is isomorphic to $T_{0,0}$, which is a contradiction. Thus $r(u)<r(z)$ and now $(u,z)$ is a bad pair, which is a contradiction to Claim \[c2\]. Let $k\geq2$. If $r(z)<\ell({x_{u,v}^{k-1}}')$, then $z\in X_{u,v}^k$, which is a contradiction to $\|X_{u,v}^k\|=2$. Thus $\ell({x_{u,v}^{k-1}}')< r(z)$. If $r(z)< \ell(x_{u,v}^{k-2})$, then $G[\{a,z,{x_{u,v}^{k}}',v,u,y\}\cup \bigcup_{i=1}^{k-1}X_{u,v}^i]$ is isomorphic to $T_{k-1,0}$. Thus $\ell(x_{u,v}^{k-2})< r(z)$. If $r(z)< r(x_{u,v}^{k-1})$, then $z\in X_{u,v}^{k-1}$, which is a contradiction to $\|X_{u,v}^{k-1}\|=2$. Thus $r(x_{u,v}^{k-1})<r(z)$, but now $(x_{u,v}^{k-1},z)$ is also a bad pair, which is a contradiction to Claim \[c2\] and completes this part of the proof. For contradiction, we assume that there is a vertex $z\in V(G)$ such that $(z,{x_{u,v}^k}')$ is a bad pair. By Claim \[c1\], $\ell({x_{u,v}^k}')<\ell(z)$ and $r(z)<r({x_{u,v}^k}')$. By Claim \[c3\], $r(x_{u,v}^k)<r(z)$. Let $y_z$ be the vertex guaranteed by Claim \[c1\] such that $r(z)<\ell(y_z)$, but this contradicts Claim \[c4\] (b). (b): For contradiction, we assume the existence of a vertex $z\in V(G)$ such that $(x_{u,v}^k,z)$ is a bad pair. Trivially $z\not= {x_{u,v}^k}'$. If $r(z)< r({x_{u,v}^k}')$, then this contradicts Claim \[c4\] (a), that is $\|X^k_{u,v}\|=2$ and if $r({x_{u,v}^k}')<r(z)$, then $({x_{u,v}^k}',z)$ is also a bad pair and this contradicts Claim \[c2\]. This completes the proof of Claim \[c5\]. $\Box$ For a bad pair $(u,v)$ define $Y_{u,v}^k$ as $X_{u,v}^k$ by interchanging in the definition right by left. Let $r_{u,v}^{\rm max}$ be the smallest integer $k$ such that $\|Y_{u,v}^{k-1}\|= 2$ and $\|Y_{u,v}^{k}\|= 1$. By symmetry, one can prove a “y”-version of Claim \[c4\], Claim \[c5\] and Claim \[c6a\] (a) and (b). Let $\{y_{u,v}^k,{y_{u,v}^k}'\}=Y_{u,v}^k$ such that $N_G({y^k_{u,v}}')\subset N_G(y^k_{u,v})$ for $k\leq r_{u,v}^{\rm max}-1$. \[c6a\] Let $(u,v)$ and $(w,z)$ be bad pairs and $k\in[\ell_{u,v}^{\rm max}]$. (a) If $X_{u,v}^{k}\cap X_{w,z}^{\tilde{k}}\not=\emptyset$, then $x_{u,v}^{k-1}=x_{w,z}^{\tilde{k}-1}$ for $\tilde{k}\in [\ell_{w,z}^{\rm max}]$. (b) If $X_{u,v}^{k}\cap X_{w,z}^{\tilde{k}}\not=\emptyset$, then $X_{u,v}^{k}=X_{w,z}^{\tilde{k}}$ for $\tilde{k}\in [\ell_{w,z}^{\rm max}]$. (c) If $X_{u,v}^{k}\cap Y_{w,z}^{\tilde{k}}\not=\emptyset$, then $X_{u,v}^{k}\cap Y_{w,z}^{\tilde{k}}=x_{u,v}^{k}= y_{w,z}^{\tilde{k}}$ for $\tilde{k}\in [r_{w,z}^{\rm max}]$ Proof of Claim \[c6a\]: (a): For contradiction we assume $x_{u,v}^{k-1}\not=x_{w,z}^{\tilde{k}-1}$. Without loss of generality we assume $\ell(x_{u,v}^{k-1})<\ell(x_{w,z}^{\tilde{k}-1})$. Note that $x_{w,z}^{\tilde{k}-1}$ is adjacent to the vertices in $X_{u,v}^{k}\cap X_{w,z}^{\tilde{k}}$. Since the vertices in $X_{u,v}^{k}$ are not distinguishable to the right, we conclude $\ell(x_{w,z}^{\tilde{k}-1})<r(x_{u,v}^{k})$. First, we suppose $k=1$. Thus $v=x_{u,v}^{k-1}$. If $r(x_{w,z}^{\tilde{k}-1})< r(v)$, then $(x_{w,z}^{\tilde{k}-1},v)$ is a bad pair and this contradicts Claim \[c2\] and if $r(x_{w,z}^{\tilde{k}-1})> r(v)$, then $(u,x_{w,z}^{\tilde{k}-1})$ is a bad pair and this contradicts Claim \[c3\]. Now we suppose $k\geq 2$. If $r({x_{u,v}^{k-1}}')< r(x_{w,z}^{\tilde{k}-1})$, then $({x_{u,v}^{k-1}}',x_{w,z}^{\tilde{k}-1})$ is a bad pair, which contradicts Claim \[c5\] (a). Thus $r(x_{w,z}^{\tilde{k}-1})< r({x_{u,v}^{k-1}}')$. If $r({x_{u,v}^{k-1}})<r(x_{w,z}^{\tilde{k}-1})$, then $x_{w,z}^{\tilde{k}-1}\in X_{u,v}^{k-1}$, which implies $\|X_{u,v}^{k-1}\|=3$ and hence contradicts Claim \[c4\] (a). Thus $r(x_{w,z}^{\tilde{k}-1})<r({x_{u,v}^{k-1}})$. Therefore, $(x_{w,z}^{\tilde{k}-1},x_{u,v}^{k-1})$ is a bad pair. Claim \[c1\] implies the existence of a vertex $a$ which is to the left of $x_{w,z}^{\tilde{k}-1}$ and adjacent to $x_{u,v}^{k-1}$. Thus $a\in X_{u,v}^{k}$. However, $r(a)<r(x_{u,v}^{k})$, which contradicts Claim \[c4\] (c). This is the final contradiction and this completes the proof of Claim \[c6a\] (a). (b): If $\|X_{u,v}^{k}\|=\|X_{w,z}^{\tilde{k}}\|=1$, then there is nothing to show. Thus we assume, $\|X_{u,v}^{k}\|=2$. Note that by Claim \[c6a\] (a), $x_{u,v}^{k-1}=x_{w,z}^{\tilde{k}-1}$. If ${x_{u,v}^{k}}'\in X_{w,z}^{\tilde{k}}$, then ${x_{u,v}^{k}}\in X_{w,z}^{\tilde{k}}$ and we are done. Thus we assume ${x_{u,v}^{k}}'\notin X_{w,z}^{\tilde{k}}$. Since $X_{u,v}^{k}\cap X_{w,z}^{\tilde{k}}\not=\emptyset$, we conclude ${x_{u,v}^{k}}\in X_{w,z}^{\tilde{k}}$. Hence $w$ or ${x_{w,z}^{\tilde{k}-1}}'$ distinguishes the vertices in $X_{u,v}^{k}$ to the right of them, which is a contradiction to Claim \[c4\] (b). This completes the proof. (c): If $\|X_{u,v}^{k}\|=\|Y_{w,z}^{\tilde{k}}\|=1$, then there is nothing to show. Thus we assume by symmetry $\|Y_{w,z}^{\tilde{k}}\|=2$. First, we assume for contradiction ${y_{w,z}^{\tilde{k}}}'=x_{u,v}^{k}$. Note that $\ell(y_{w,z}^{\tilde{k}-1})<\ell({y_{w,z}^{\tilde{k}}}')$ and $r(y_{w,z}^{\tilde{k}-1})<r({y_{w,z}^{\tilde{k}}}')$. Suppose $\|X_{u,v}^{k}\|=1$. If $\ell(x_{u,v}^{k-1})<r(y_{w,z}^{\tilde{k}-1})$, then $y_{w,z}^{\tilde{k}-1}\in X_{u,v}^k$, which is a contradiction to $\|X_{u,v}^k\|=1$. Thus $r(y_{w,z}^{\tilde{k}-1})<\ell(x_{u,v}^{k-1})$. Note that $\ell({y_{w,z}^{\tilde{k}}}')<\ell({y_{w,z}^{\tilde{k}}})<r({y_{w,z}^{\tilde{k}-1}})$ and $r({y_{w,z}^{\tilde{k}}}')<r({y_{w,z}^{\tilde{k}}})$. Suppose $k=1$. If $r(y_{w,z}^{\tilde{k}})<\ell(u)$, then $y_{w,z}^{\tilde{k}}\in X_{u,v}^k$, which is a contradiction to $\|X_{u,v}^k\|=1$. If $\ell(u)<r(y_{w,z}^{\tilde{k}})<r(u)$, then $G[\{x_{w,z}^1,w,z,u,v,y_{u,v}^1\}\cup \bigcup_{i=1}^{\tilde{k}}Y_{w,z}^i]$ is isomorphic to $T_{\tilde{k},0}$, which is a contradiction. If $r(u)<r(y_{w,z}^{\tilde{k}})$, then $(u,y_{w,z}^{\tilde{k}})$ is a bad pair, which is a contradiction to Claim \[c3\]. Now we suppose $k\geq 2$. If $r(y_{w,z}^{\tilde{k}})<\ell({x_{u,v}^{k-1}}')$, then $y_{w,z}^{\tilde{k}}\in X_{u,v}^k$, which is a contradiction to $\|X_{u,v}^k\|=1$. If $\ell({x_{u,v}^{k-1}}')<r(y_{w,z}^{\tilde{k}})<\ell(x_{u,v}^{k-2})$, then $G[\{x_{w,z}^1,w,z,u,v,y_{u,v}^1\} \cup \bigcup_{i=1}^{\tilde{k}}Y_{w,z}^i\cup \bigcup_{i=1}^{k-1}X_{u,v}^i]$ is isomorphic to $T_{\tilde{k},k-1}$, which is a contradiction. If $\ell(x_{u,v}^{k-2})<r(y_{w,z}^{\tilde{k}})<\ell({x_{u,v}^{k-1}}')$, then $y_{w,z}^{\tilde{k}}\in X_{u,v}^{k-1}$ and hence $\|X_{u,v}^{k-1}\|=3$, which is a contradiction to Claim \[c4\] (a). If $\ell({x_{u,v}^{k-1}}')<r(y_{w,z}^{\tilde{k}})$, then $({x_{u,v}^{k-1}}',y_{w,z}^{\tilde{k}})$ is a bad pair, which is a contradiction to Claim \[c5\] (a). This shows $\|X_{u,v}^{k}\|\not=1$ and thus we suppose $\|X_{u,v}^{k}\|=2$. If $\ell(x_{u,v}^{k-1})<r(y_{w,z}^{\tilde{k}-1})$, then $y_{w,z}^{\tilde{k}-1}\in X_{u,v}^k$, which is a contradiction to $\|X_{u,v}^k\|=2$. Thus $r(y_{w,z}^{\tilde{k}-1})<\ell(x_{u,v}^{k-1})$. Note that $\ell({y_{w,z}^{\tilde{k}}}')<\ell({y_{w,z}^{\tilde{k}}})<r({y_{w,z}^{\tilde{k}-1}})$ and $r({y_{w,z}^{\tilde{k}}}')<r({y_{w,z}^{\tilde{k}}})$. If $\ell({x_{u,v}^k}')<r(y_{w,z}^{\tilde{k}-1})$, then ${x_{u,v}^k}'=y_{w,z}^{\tilde{k}}$. Thus $\{{x_{u,v}^k}',x_{u,v}^k\}=Y_{w,z}^{\tilde{k}}$. By Claim \[c4\] (b), these vertices are not distinguishable to the right and to the left. Thus they are twins, which is a contradiction. Thus $r(y_{w,z}^{\tilde{k}-1})<\ell({x_{u,v}^k}')$. Note that $\ell({y_{w,z}^{\tilde{k}}}')<\ell({y_{w,z}^{\tilde{k}}})<r({y_{w,z}^{\tilde{k}-1}})$. If $r(y_{w,z}^{\tilde{k}})<r({x_{u,v}^k}')$, then $y_{w,z}^{\tilde{k}}\in X_{u,v}^k$, which is a contradiction to $\| X_{u,v}^k\|=2$ and if $r({x_{u,v}^k}')<r(y_{w,z}^{\tilde{k}})$, then $({x_{u,v}^k}',y_{w,z}^{\tilde{k}})$ is a bad pair, which is a contradiction to Claim \[c5\] (a). This shows ${y_{w,z}^{\tilde{k}}}'\not=x_{u,v}^{k}$. A totally symmetric argumentation shows $y_{w,z}^{\tilde{k}}\not={x_{u,v}^{k}}'$. To complete the proof, we show that ${y_{w,z}^{\tilde{k}}}'\not={x_{u,v}^{k}}'$. For contradiction, we assume ${y_{w,z}^{\tilde{k}}}'={x_{u,v}^{k}}'$. If $\ell(x_{u,v}^{k-1})<r(y_{w,z}^{\tilde{k}-1})$, then $x_{u,v}^{k-1}=y_{w,z}^{\tilde{k}}$. Thus $G[\{x_{w,z}^1,w,z,u,v,y_{u,v}^1\}\cup \bigcup_{i=1}^{\tilde{k}}Y_{w,z}^i\cup \bigcup_{i=1}^{k-1}X_{u,v}^i]$ is isomorphic to $R_{k+\tilde{k}-1}$, which is a contradiction. Hence we assume $r(y_{w,z}^{\tilde{k}-1})<\ell(x_{u,v}^{k-1})$. If $\ell(x_{u,v}^k)<\ell(y_{w,z}^{\tilde{k}-1})$, then $(y_{w,z}^{\tilde{k}-1},x_{u,v}^k)$ is a bad pair, which is a contradiction to the “y”-version of Claim \[c5\] (b). Hence we assume $\ell(y_{w,z}^{\tilde{k}-1})<\ell(x_{u,v}^k)$. If $x_{u,v}^k\in Y_{w,z}^{\tilde{k}}$, then this is a contradiction to the “y”-version of Claim \[c4\] (a), because $\ell(x_{u,v}^k)<\ell({y_{w,z}^{\tilde{k}}}')$. Suppose $\tilde{k}=1$. Since $x_{u,v}^k\notin Y_{w,z}^{\tilde{k}}$, we conclude $x_{u,v}^kw\in E(G)$. If $\ell(w)<\ell(x_{u,v}^k)$, then $G[\{x_{w,z}^1,w,z,u,v,y_{u,v}^1\}\cup \bigcup_{i=1}^{k}X_{u,v}^i ]$ is isomorphic to $T_{k,0}$, which is a contradiction. If $\ell(x_{u,v}^k)<\ell(w)$, then $(w,x_{u,v}^k)$ is a bad pair, which is a contradiction to Claim \[c3\]. Hence we suppose $\tilde{k}\geq2$. Note that $\ell(x^k_{u,v})<r({y_{w,z}^{\tilde{k}-1}}')$. If $r(y_{w,z}^{\tilde{k}-2})<\ell(x_{u,v}^k)$, then $G[\{x_{w,z}^1,w,z,u,v,y_{u,v}^1\}\cup \bigcup_{i=1}^{\tilde{k}-1}Y_{w,z}^i\cup \bigcup_{i=1}^{k}X_{u,v}^i]$ is isomorphic to $T_{\tilde{k}-1,k}$. If $\ell({y_{w,z}^{\tilde{k}-1}}')<\ell(x_{u,v}^k)<r(y_{w,z}^{\tilde{k}-2})$, then $x_{u,v}^k \in Y_{w,z}^{\tilde{k}-1}$, which is a contradiction to the “y”-version of Claim \[c4\] (a). If $\ell(x_{u,v}^k)<\ell({y_{w,z}^{\tilde{k}-1}}')$, then $({y_{w,z}^{\tilde{k}-1}}',x_{u,v}^k)$ is a bad pair, which is a contradiction to the “y”-version of Claim \[c5\] (a). This completes the proof of Claim \[c6a\]. $\Box$ Next, we define step by step new interval representations of $G$ as follows. First we shorten the intervals of $X_{u,v}^{k}$ for every bad pair $(u,v)$ and $k\in [\ell_{u,v}^{\rm max}]$. Let $I':V(G)\rightarrow \mathcal{I}^{++}$ be such that $I'(x)=[\ell(x),\ell(x_{u,v}^{k-1})]$ if $x\in X_{u,v}^k$ for some bad pair $(u,v)$ and $I'(x)=I(x)$ otherwise. By Claim \[c6a\] (a), $I'$ is well-defined; that is, if $x\in X_{u,v}^k\cap X_{w,z}^{\tilde{k}}$, then $\ell(x_{u,v}^{k-1})=\ell(x_{w,z}^{\tilde{k}-1})$. Let $\ell'(x)$ and $r'(x)$ be the left and right endpoint of the interval $I'(x)$ for $x\in V(G)$, respectively. \[c7\] $I'$ is an interval representation of $G$. Proof of Claim \[c7\]: Trivially, if two intervals do not intersect in $I$, then they do not intersect in $I'$. For contradiction, we assume that there are two vertices $a,b\in V(G)$ such that $I(a)\cap I(b)\not= \emptyset$ and $I'(a)\cap I'(b)= \emptyset$. At least one interval is shortened by changing the interval representation. Say $a\in X_{u,v}^k$ for some bad pair $(u,v)$ and $k\in [\ell_{u,v}^{\rm max}]$. Hence $b\not=x_{u,v}^{k-1}$ and $\ell(x_{u,v}^{k-1})<\ell(b)$ and by Claim \[c4\] (b), $\ell(b)< r({x_{u,v}^{k}})$. We conclude that $(b,x_{u,v}^{k-1})$ is not a bad pair, otherwise Claim \[c1\] implies the existence of a vertex $z\in X_{u,v}^{k}$ to the left of $b$, but $z\notin\{x_{u,v}^{k},{x_{u,v}^{k}}'\}$, which is a contradiction to Claim \[c4\] (a). Thus $r(x_{u,v}^{k-1})<r(b)$. If $k=1$, then $(u,b)$ is also a bad pair, which is a contradiction to Claim \[c3\]. Thus $k\geq 2$. Since $\ell(b)< r({x_{u,v}^{k}})$, we obtain $\ell(b)<\ell({x_{u,v}^{k-1}}')$. Since $({x_{u,v}^{k-1}}',b)$ is not a bad pair by Claim \[c5\] (a), $r(b)<r({x_{u,v}^{k-1}}')$. Thus $b\in X_{u,v}^{k-1}$, which is a contradiction to $\|X_{u,v}^{k-1}\|=2$. $\Box$ \[c8\] The change of the interval representation of $G$ from $I$ to $I'$ creates no new bad pair $(a,b)$ such that $\{a,b\}\not= X_{u,v}^k$ for some $k\in [\ell_{u,v}^{\rm max}]$ and some bad pair $(u,v)$. Proof of Claim \[c8\]: For contradiction, we assume that $(a,b)$ is a new bad pair and $\{a,b\}\not= X_{u,v}^k$. Since $(a,b)$ is a new bad pair, $I'(a)$ is a proper subset of $I(a)$. Thus let $a\in X_{u,v}^k$ and $b\notin X_{u,v}^k$. If $a\in X_{u,v}^k$ and $\|X_{u,v}^k\|=2$, then $\ell(b)<\ell({x_{u,v}^k}')$ and $r'(a)=\ell(x_{u,v}^{k-1}) < r(b)<r({x_{u,v}^k}')$, because of Claim \[c5\] (a). Thus $b\in X_{u,v}^k$, which is a contradiction. If $a\in X_{u,v}^k$ and $\|X_{u,v}^k\|=1$, then $\ell(b)<\ell({x_{u,v}^k})$ and $r'(a)=\ell(x_{u,v}^{k-1}) < r(b)<r({x_{u,v}^k})$. Thus $b\in X_{u,v}^k$, which is the final contradiction. $\Box$ In a second step, we shorten the intervals of $Y_{u,v}^{i}$ for every bad pair $(u,v)$ and $i\in [r_{u,v}^{\rm max}]$. Let $I'':V(G)\rightarrow \mathcal{I}^{++}$ be such that $I''(y)=[r'(y_{u,v}^{k-1}),r'(y)]$ if $y\in Y_{u,v}^k$ for some bad pair $(u,v)$ and $I''(y)=I'(y)$ otherwise. Note that bad pairs are only referred to the interval representation $I$. Let $\ell''(x)$ and $r''(x)$ be the left and right endpoints of the interval $I''(x)$ for $x\in V(G)$, respectively. \[c9\] $I''$ is an interval representation of $G$. Proof of Claim \[c9\]: Again, two intervals do not intersect in $I''$ if they do not intersect in $I'$ (and in $I$). For contradiction, we assume that there are two vertices $a,b\in V(G)$ such that $I(a)\cap I(b)\not= \emptyset$ and $I''(a)\cap I''(b)= \emptyset$. Again, at least one interval is shortened by the change of the interval representation. Say $a\in Y_{u,v}^k$ for some bad pair $(u,v)$ and $k\in [r_{u,v}^{\rm max}]$. Suppose $a\in X_{w,z}^{\tilde{k}}$ for some bad pair $(w,z)$ and $\tilde{k}\in [\ell_{w,z}^{\rm max}-1]$. By Claim \[c6a\] (c), we have $a=x_{w,z}^{\tilde{k}}=y_{u,v}^k$. If $y_{u,v}^{k-1}=x_{w,z}^{\tilde{k}+1}$, then we did not change the interval of $a$. Thus we assume $y_{u,v}^{k-1}\not=x_{w,z}^{\tilde{k}+1}$. Now $\ell(y_{u,v}^{k})<r(b)<r(y_{u,v}^{k-1})$. The rest of the proof is similar to a symmetric version of the proof of Claim \[c7\]. If $a\notin X_{\tilde{u},\tilde{v}}^{\tilde{k}}$, then $r(b)<r(y_{u,v}^{k-1})$ and $\ell({y_{u,v}^{k}}')<r(b)$, if ${y_{u,v}^{k}}'$ exists, otherwise $\ell({y_{u,v}^{k}})<r(b)$. If $\ell(y_{u,v}^{k-1})<\ell(b)$, then by Claim \[c8\], $(b,y_{u,v}^{k-1})$ is a bad pair and by Claim \[c5\], $I(b)=I'(b)$. Thus Claim \[c1\] implies the existence of a vertex, which contradicts the “y”-version of Claim \[c4\] (a) and (b) and hence we suppose $\ell(b)\leq \ell(y_{u,v}^{k-1})$. Thus $k\geq 2$, otherwise $(u',b)$ is a bad pair, which contradicts Claim \[c3\]. If $\ell(b)\leq \ell({y_{u,v}^{k-1}}')$, then $({y_{u,v}^{k-1}}',b)$ is a bad pair, which contradicts the “y”-version of Claim \[c5\] (a). Therefore, $\ell({y_{u,v}^{k-1}}')< \ell(b)$, which implies $b\in Y_{u,v}^{k-1}$, but $b\notin \{{y_{u,v}^{k-1}},{y_{u,v}^{k-1}}'\}$, which contradicts the “y”-version of Claim \[c4\] (a). $\Box$ \[c10\] The change of the interval representation of $G$ from $I$ to $I''$ creates no new bad pair $(a,b)$ such that $\{a,b\}\not= X_{u,v}^k$ for some $k\in [\ell_{u,v}^{\rm max}]$ or $\{a,b\}\not= Y_{u,v}^i$ for some $i\in [r_{u,v}^{\rm max}]$ and some bad pair $(u,v)$. Proof of Claim 10: For contradiction, we assume that $(a,b)$ is a new bad pair and $Y_{u,v}^i\not=\{a,b\}\not= X_{u,v}^k$. Thus $a\in X_{u,v}^k$ or $a\in Y_{u,v}^i$ and $b\notin X_{u,v}^k$ or $b\notin Y_{u,v}^i$, respectively. If $a\in X_{u,v}^k$ and $\|X_{u,v}^k\|=2$, then $\ell(b)<\ell({x_{u,v}^k}')$ and $\ell(x_{u,v}^{k-1}) < r(b)<r({x_{u,v}^k}')$. Thus $b\in X_{u,v}^k$, which is a contradiction. If $a\in X_{u,v}^k$ and $\|X_{u,v}^k\|=1$, then $\ell(b)<\ell({x_{u,v}^k})$ and $\ell(x_{u,v}^{k-1}) < r(b)<r({x_{u,v}^k})$. Thus $b\in X_{u,v}^k$, which is a contradiction. If $a\in Y_{u,v}^i$ the proof is almost exactly the same. $\Box$ Now we are in a position to blow up some intervals to open or half-open intervals to get a mixed proper interval graph. Let $I^: V(G)\rightarrow \mathcal{I}$ be such that $$\begin{aligned} I^(x)= \left\{ \begin{array}{rl} (\ell(v),r(v)), & \text{if } (x,v) \text{ is a bad pair},\\ (\ell''(x_{u,v}^k),r''(x_{u,v}^k)], & \text{if } x={x_{u,v}^k}' \text{ for some bad pair } (u,v) \text{ and } k\in [\ell_{u,v}^{\rm max}-1],\\ \left[\ell''(y_{u,v}^i),r''(y_{u,v}^i)\right), & \text{if } x={y_{u,v}^i}' \text{ for some bad pair } (u,v) \text{ and } i\in [r_{u,v}^{\rm max}-1], \text{ and}\\ \left[\ell''(x),r''(x)\right], & \text{else.} \end{array} \right.\end{aligned}$$ Note that $I^$ is well-defined by Claim \[c5\] and Claim \[c6a\]; that is, the four cases in the definition of $I^$ induces a partition of the vertex set of $G$. Moreover, the interval representation $I^$ defines a mixed proper interval graph. As a final step, we prove that $I''$ and $I^$ define the same graph. Since we make every interval bigger, we show that for every two vertices $a,b$ such that $I''(a)\cap I''(b)=\emptyset$, we still have $I^(a)\cap I^(b)=\emptyset$. For contradiction, we assume the opposite. Let $a,b$ be two vertices such that $I''(a)\cap I''(b)=\emptyset$ and $I^(a)\cap I^(b)\not=\emptyset$. It follows by our approach and definition of our interval representation $I''$, that both $a$ and $b$ are blown up intervals. First we suppose $a$ and $b$ are intervals that are blown up to open intervals, that is, there are distinct vertices $\tilde{a}$ and $\tilde{b}$ such that $(a,\tilde{a})$ and $(b,\tilde{b})$ are bad pairs. Furthermore, the intervals of $\tilde{a}$ and $\tilde{b}$ intersect not only in one point. By Claim \[c2\] and \[c3\], we assume without loss of generality, that $\ell''(\tilde{a})<\ell''(\tilde{b})<r''(\tilde{a})<r''(\tilde{b})$. Therefore, by the construction of $I''$, we obtain $a$ is adjacent to $\tilde{b}$ and $\tilde{a}$ is adjacent to $b$, and in addition they intersect in one point, respectively. Now, $G[\{x_{a,\tilde{a}}^1,a,\tilde{a},b,\tilde{b},y_{b,\tilde{b}}^1\}]$ is isomorphic to $T_{0,0}$, which is a contradiction. Now we suppose $a$ is blown up to an open interval and $b$ is blown up to an open-closed interval (the case closed-open is exactly symmetric). Let $\tilde{a}$ be the vertex such that $(a,\tilde{a})$ is a bad pair. Let $\tilde{b},u,v\in V(G)$ and $k\in \mathbb{N}$ such that $\{b,\tilde{b}\}=X_{u,v}^k$. We suppose $\tilde{a}\not=\tilde{b}$. We conclude $\ell''(\tilde{a})<\ell''(\tilde{b})<r''(\tilde{a})<r''(\tilde{b})$. As above, we conclude $a$ is adjacent to $\tilde{b}$ and $\tilde{a}$ is adjacent to $b$, and in addition they intersect in one point, respectively. Thus $G[\{x_{a,\tilde{a}}^1,a,\tilde{a},v,u,y_{u,v}^1\}\cup \bigcup_{i=1}^k X_{u,v}^i]$ induces a $T_{k,0}$, which is a contradiction. Now we suppose $\tilde{a}=\tilde{b}$. We conclude that $G[\{x_{a,\tilde{a}}^1,a,v,u,y_{u,v}^1\}\cup \bigcup_{i=1}^{k} X_{u,v}^i]$ is isomorphic to $R_k$, which is a contradiction. It is easy to see that $a$ and $b$ cannot be both blown up to closed-open or both open-closed intervals, because $G$ is $R_k$-free for $k\geq 0$ and the definition of $I''$. Therefore, we consider finally the case that $a$ is blown up to a closed-open and $b$ to an open-closed interval. Let $\tilde{a},\tilde{b},u,v,w,z \in V(G)$ and $k,\tilde{k}\in \mathbb{N}$ such that $\{a,\tilde{a}\}=Y_{u,v}^k$ and $\{b,\tilde{b}\}=X_{w,z}^{\tilde{k}}$. First we suppose $\tilde{a}\not=\tilde{b}$. Again, we obtain $\ell''(\tilde{a})<\ell''(\tilde{b})<r''(\tilde{a})<r''(\tilde{b})$ and $a$ is adjacent to $\tilde{b}$ and $\tilde{a}$ is adjacent to $b$, and furthermore they intersect in one point, respectively. Thus $G[\{x_{u,v}^1,u,v,w,z,y_{w,z}^1\}\cup \bigcup_{i=1}^k Y_{u,v}^i\cup \bigcup_{i=1}^{\tilde{k}} X_{w,z}^i]$ is isomorphic to $T_{k,\tilde{k}}$. Next we suppose $\tilde{a}=\tilde{b}$ and hence $G[\{x_{u,v}^1,u,v,w,z,y_{w,z}^1\}\cup \bigcup_{i=1}^k Y_{u,v}^i\cup \bigcup_{i=1}^{\tilde{k}} X_{w,z}^i]$ is isomorphic to $R_{k+\tilde{k}}$. This is the final contradiction and completes the proof of Theorem \[mainthm\]. $\Box$ (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (+1,+1.5) circle (); (,)–(+1,)–(+2,)–(+3,) (+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+2.5,+1) (,)–(+0.5,+1)–(+1,+1.5)–(+1.5,+1) (+1,)–(+1,+1.5); (+2,) arc(-20:85:1.16); \(1) at (+1.5,-0.7) [$S_2''$]{}; (-1,) circle (); (-0.5,+1) circle (); (,) circle (); (+1,) circle (); (+2,) circle (); (+3,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (,+1.5) circle (); (-1,)–(+3,) (-1,)–(-0.5,+1)–(,)–(+0.5,+1)–(+1,)–(+1.5,+1)–(+2,)–(+2.5,+1) (-0.5,+1)–(,+1.5)–(+0.5,+1) (,)–(,+1.5); (+1,) arc(-20:85:1.16); \(1) at (+1,-0.7) [$S_3''$]{}; (-1,) circle (); (,) circle (); (+1,) circle (); (+2,) circle (); (+3.5,) circle (); (+4.5,) circle (); (+5.5,) circle (); (-0.5,+1) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+4,+1) circle (); (+5,+1) circle (); (,+1.5) circle (); (-1,)–(+2,) (+3.5,)–(+5.5,) (-1,)–(-0.5,+1)–(,)–(+0.5,+1)–(+1,)–(+1.5,+1)–(+2,) (+3.5,)–(+4,+1)–(+4.5,)–(+5,+1) (-0.5,+1)–(,+1.5)–(+0.5,+1) (,)–(,+1.5); (+1,) arc(-20:85:1.16); \(1) at (+5,-0.7) [$S_i''$]{}; (+2.35,) circle (/2); (+2.75,) circle (/2); (+3.15,) circle (/2); (-1,) – (+4.5,) node \[pos=0.5,anchor=north,yshift=-0.4cm\] [$i$ triangles]{}; (+1,) circle (); (+2,) circle (); (+0.5,+1) circle (); (+1.5,+1) circle (); (+2.5,+1) circle (); (,) circle (); (+1,)–(+2,) (+1,)–(+0.5,+1)–(+2,) (+1,)–(+2.5,+1)–(+2,) (+1,)–(+1.5,+1)–(+2,) (+1,)–(,)–(+0.5,+1); \(1) at (+1.5,-0.7) [$G_1$]{}; In Theorem \[mainthm\] we only consider twin-free $\mathcal{U}$-graphs to reduce the number of case distinctions in the proof. In Corollary \[maincoro\] we resolve this technical condition. See Figure \[graphsS”\] and \[graphG1\] for illustration. Let $\mathcal{S''}=\bigcup_{i=2}^{\infty}\{S_i''\}$. \[maincoro\] A graph $G$ is a mixed unit interval graph if and only if $G$ is a $\{G_1\}\cup\mathcal{R}\cup \mathcal{S}\cup \mathcal{S''}\cup \mathcal{T}$-free interval graph. Proof of Corollary \[maincoro\]: We first show that $ \{G_1\}\cup\mathcal{R}\cup \mathcal{S}\cup \mathcal{S''}\cup \mathcal{T}$ is the set of all twin-free graphs that contain all graphs of $\{K_{2,3}^\}\cup\mathcal{R}\cup \mathcal{S}\cup \mathcal{S'}\cup \mathcal{T}$ and are minimal with subject to induced subgraphs. We leave it as an exercise to show that $G_1$ is the only minimal twin-free and $R_0$-free graph that contains $K_{2,3}^$. Since all graphs in $\mathcal{R}\cup \mathcal{S}\cup \mathcal{T}$ are twin-free graphs, there is nothing to show. Let now $G\in \mathcal{S'}$, that is $G=S_k'$ for some $k\in \mathbb{N}$. With the notation as in the proof of Theorem \[mainthm\], $G$ can be interpreted as a bad pair $(u,v)$ together with $\{y_{u,v}^1\}\cup\bigcup_{i=1}^k X_{u,v}^i$ such that $\|X_{u,v}^i\|=2$ if $i<k$ and $\|X_{u,v}^k\|=3$. Note that Claim \[c4\] (b) of Theorem \[mainthm\] is still true even if $G$ is not $\mathcal{S}'$-free. Therefore, we know that the vertices in $X_{u,v}^i$ cannot be distinguished by vertices from the right. Thus the vertices that distinguish the vertices in $X_{u,v}^k$ are only adjacent to $X_{u,v}^k$. Clearly, there are at least two of them, say $a,b$. Without loss of generality $a$ and $b$ they do not have the same neighborhood on $X_{u,v}^k$. We conclude either $N_{G[X_{u,v}^k]}(a)\subset N_{G[X_{u,v}^k]}(b)$ or $N_{G[X_{u,v}^k]}(b)\subset N_{G[X_{u,v}^k]}(a)$. We assume the first possibility. Since $0<\|N_{G[X_{u,v}^k]}(x)\cap X_{u,v}^k\|<3$ for $x\in \{a,b\}$, it follows $\|N_{G[X_{u,v}^k]}(a)\cap X_{u,v}^k\|=1$ and $\|N_{G[X_{u,v}^k]}(b)\cap X_{u,v}^k\|=2$. Since $G$ is $R_k$-free, $a$ and $b$ are adjacent. Now $G[\bigcup_{i=1}^k X_{u,v}^i\cup \{a,b,u,v,y_{u,v}^1\}]$ is isomorphic to $S_{k+1}''$. This completes this part of the proof. Let $G$ be an interval graph. The relation $\sim$, where $u\sim v$ if and only if $u$ and $v$ are twins, defines an equivalence relation on $V(G)$. Let $U\subseteq V(G)$ such that there is exactly one vertex of every equivalence class in $U$. Therefore, $G[U]$ is a twin-free graph. Furthermore, $G$ contains an induced subgraph in $\{G_1\}\cup\mathcal{R}\cup \mathcal{S}\cup \mathcal{S''}\cup \mathcal{T}$ if and only if $G[U]$ contains an induced subgraph in $\{K_{2,3}^\}\cup\mathcal{R}\cup \mathcal{S}\cup \mathcal{S'}\cup \mathcal{T}$. In addition, $G[U]$ is a twin-free $\mathcal{U}$-graph if and only if $G$ is a $\mathcal{U}$-graph. By Theorem \[mainthm\] this completes the proof. $\Box$ [1]{} D.G. Corneil, A simple 3-sweep LBFS algorithm for the recognition of unit interval graphs, Discrete Appl. Math. 138, 371-379 (2004). D.G. Corneil, S. Olariu and L. Stewart, The ultimate interval graph recognition algorithm?, Proceedings of the [N]{}inth [A]{}nnual [ACM]{}-[SIAM]{} [S]{}ymposium on [D]{}iscrete [A]{}lgorithms ([S]{}an [F]{}rancisco, [CA]{}, 1998), 175-180. M.C. Dourado, V.B. Le, F. Protti, D. Rautenbach and J.L. Szwarcfiter, Mixed unit interval graphs, Discrete Math. 312, 3357-3363 (2012). P.C. Fishburn. Interval Orders and Interval Graphs. John Wiley & Sons (1985). P. Frankl and H. Maehara, Open interval-graphs versus closed interval-graphs, Discrete Math. 63, 97-100 (1987). M.C. Golumbic, Algorithmic Graph Theory and Perfect Graphs, vol. 57. Annals of Discrete Mathematics, Amsterdam, The Netherlands, 2004. V.B. Le and D. Rautenbach, Integral Mixed Unit Interval Graphs, Lecture Notes in Computer Science 7434, 495-506 (2012). C.G. Lekkerkerker and J.C. Boland, Representation of a finite graph by a set of intervals on the real line, Fund. Math., 51, 45-64 (1962). D. Rautenbach and J.L. Szwarcfiter, Unit Interval Graphs of Open and Closed Intervals, J. Graph Theory 72*(4), 418-429 (2013). F.S. Roberts, Indifference graphs, in F. Harary (Ed.), Proof Techniques in Graph Theory, Academic Press, pp.139-146 (1969).	{ "pile_set_name": "ArXiv" }

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.